diff options
135 files changed, 9431 insertions, 773 deletions
diff --git a/.ci/scripts/merge/apply-patches-by-label.py b/.ci/scripts/merge/apply-patches-by-label.py index 43ed74d7f..b2e430ac7 100644 --- a/.ci/scripts/merge/apply-patches-by-label.py +++ b/.ci/scripts/merge/apply-patches-by-label.py @@ -1,7 +1,7 @@ # Download all pull requests as patches that match a specific label # Usage: python download-patches-by-label.py <Label to Match> <Root Path Folder to DL to> -import requests, sys, json, urllib3.request, shutil, subprocess, os +import requests, sys, json, urllib3.request, shutil, subprocess, os, traceback tagline = sys.argv[2] @@ -33,4 +33,5 @@ try: for i in range(1,30): do_page(i) except: + traceback.print_exc(file=sys.stdout) sys.exit(-1) diff --git a/CMakeLists.txt b/CMakeLists.txt index eb403205c..6044e311a 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -166,7 +166,7 @@ macro(yuzu_find_packages) # Capitalization matters here. We need the naming to match the generated paths from Conan set(REQUIRED_LIBS # Cmake Pkg Prefix Version Conan Pkg - "Catch2 2.13 catch2/2.13.0" + "Catch2 2.13.7 catch2/2.13.7" "fmt 8.0 fmt/8.0.0" "lz4 1.8 lz4/1.9.2" "nlohmann_json 3.8 nlohmann_json/3.8.0" @@ -600,6 +600,7 @@ if (YUZU_USE_BUNDLED_FFMPEG) ${LIBVA_LIBRARIES}) set(FFmpeg_HWACCEL_FLAGS --enable-hwaccel=h264_vaapi + --enable-hwaccel=vp8_vaapi --enable-hwaccel=vp9_vaapi --enable-libdrm) list(APPEND FFmpeg_HWACCEL_INCLUDE_DIRS @@ -620,6 +621,7 @@ if (YUZU_USE_BUNDLED_FFMPEG) --enable-ffnvcodec --enable-nvdec --enable-hwaccel=h264_nvdec + --enable-hwaccel=vp8_nvdec --enable-hwaccel=vp9_nvdec --extra-cflags=-I${CUDA_INCLUDE_DIRS} ) @@ -670,6 +672,7 @@ if (YUZU_USE_BUNDLED_FFMPEG) --disable-postproc --disable-swresample --enable-decoder=h264 + --enable-decoder=vp8 --enable-decoder=vp9 --cc="${CMAKE_C_COMPILER}" --cxx="${CMAKE_CXX_COMPILER}" diff --git a/externals/FidelityFX-FSR/ffx-fsr/ffx_a.h b/externals/FidelityFX-FSR/ffx-fsr/ffx_a.h new file mode 100644 index 000000000..d04bff55c --- /dev/null +++ b/externals/FidelityFX-FSR/ffx-fsr/ffx_a.h @@ -0,0 +1,2656 @@ +//============================================================================================================================== +// +// [A] SHADER PORTABILITY 1.20210629 +// +//============================================================================================================================== +// FidelityFX Super Resolution Sample +// +// Copyright (c) 2021 Advanced Micro Devices, Inc. All rights reserved. +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files(the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and / or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions : +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. +//------------------------------------------------------------------------------------------------------------------------------ +// MIT LICENSE +// =========== +// Copyright (c) 2014 Michal Drobot (for concepts used in "FLOAT APPROXIMATIONS"). +// ----------- +// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation +// files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, +// modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the +// Software is furnished to do so, subject to the following conditions: +// ----------- +// The above copyright notice and this permission notice shall be included in all copies or substantial portions of the +// Software. +// ----------- +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE +// WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR +// COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, +// ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. +//------------------------------------------------------------------------------------------------------------------------------ +// ABOUT +// ===== +// Common central point for high-level shading language and C portability for various shader headers. +//------------------------------------------------------------------------------------------------------------------------------ +// DEFINES +// ======= +// A_CPU ..... Include the CPU related code. +// A_GPU ..... Include the GPU related code. +// A_GLSL .... Using GLSL. +// A_HLSL .... Using HLSL. +// A_HLSL_6_2 Using HLSL 6.2 with new 'uint16_t' and related types (requires '-enable-16bit-types'). +// A_NO_16_BIT_CAST Don't use instructions that are not availabe in SPIR-V (needed for running A_HLSL_6_2 on Vulkan) +// A_GCC ..... Using a GCC compatible compiler (else assume MSVC compatible compiler by default). +// ======= +// A_BYTE .... Support 8-bit integer. +// A_HALF .... Support 16-bit integer and floating point. +// A_LONG .... Support 64-bit integer. +// A_DUBL .... Support 64-bit floating point. +// ======= +// A_WAVE .... Support wave-wide operations. +//------------------------------------------------------------------------------------------------------------------------------ +// To get #include "ffx_a.h" working in GLSL use '#extension GL_GOOGLE_include_directive:require'. +//------------------------------------------------------------------------------------------------------------------------------ +// SIMPLIFIED TYPE SYSTEM +// ====================== +// - All ints will be unsigned with exception of when signed is required. +// - Type naming simplified and shortened "A<type><#components>", +// - H = 16-bit float (half) +// - F = 32-bit float (float) +// - D = 64-bit float (double) +// - P = 1-bit integer (predicate, not using bool because 'B' is used for byte) +// - B = 8-bit integer (byte) +// - W = 16-bit integer (word) +// - U = 32-bit integer (unsigned) +// - L = 64-bit integer (long) +// - Using "AS<type><#components>" for signed when required. +//------------------------------------------------------------------------------------------------------------------------------ +// TODO +// ==== +// - Make sure 'ALerp*(a,b,m)' does 'b*m+(-a*m+a)' (2 ops). +//------------------------------------------------------------------------------------------------------------------------------ +// CHANGE LOG +// ========== +// 20200914 - Expanded wave ops and prx code. +// 20200713 - Added [ZOL] section, fixed serious bugs in sRGB and Rec.709 color conversion code, etc. +//============================================================================================================================== +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// COMMON +//============================================================================================================================== +#define A_2PI 6.28318530718 +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// +// CPU +// +// +//============================================================================================================================== +#ifdef A_CPU + // Supporting user defined overrides. + #ifndef A_RESTRICT + #define A_RESTRICT __restrict + #endif +//------------------------------------------------------------------------------------------------------------------------------ + #ifndef A_STATIC + #define A_STATIC static + #endif +//------------------------------------------------------------------------------------------------------------------------------ + // Same types across CPU and GPU. + // Predicate uses 32-bit integer (C friendly bool). + typedef uint32_t AP1; + typedef float AF1; + typedef double AD1; + typedef uint8_t AB1; + typedef uint16_t AW1; + typedef uint32_t AU1; + typedef uint64_t AL1; + typedef int8_t ASB1; + typedef int16_t ASW1; + typedef int32_t ASU1; + typedef int64_t ASL1; +//------------------------------------------------------------------------------------------------------------------------------ + #define AD1_(a) ((AD1)(a)) + #define AF1_(a) ((AF1)(a)) + #define AL1_(a) ((AL1)(a)) + #define AU1_(a) ((AU1)(a)) +//------------------------------------------------------------------------------------------------------------------------------ + #define ASL1_(a) ((ASL1)(a)) + #define ASU1_(a) ((ASU1)(a)) +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC AU1 AU1_AF1(AF1 a){union{AF1 f;AU1 u;}bits;bits.f=a;return bits.u;} +//------------------------------------------------------------------------------------------------------------------------------ + #define A_TRUE 1 + #define A_FALSE 0 +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// CPU/GPU PORTING +// +//------------------------------------------------------------------------------------------------------------------------------ +// Get CPU and GPU to share all setup code, without duplicate code paths. +// This uses a lower-case prefix for special vector constructs. +// - In C restrict pointers are used. +// - In the shading language, in/inout/out arguments are used. +// This depends on the ability to access a vector value in both languages via array syntax (aka color[2]). +//============================================================================================================================== +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// VECTOR ARGUMENT/RETURN/INITIALIZATION PORTABILITY +//============================================================================================================================== + #define retAD2 AD1 *A_RESTRICT + #define retAD3 AD1 *A_RESTRICT + #define retAD4 AD1 *A_RESTRICT + #define retAF2 AF1 *A_RESTRICT + #define retAF3 AF1 *A_RESTRICT + #define retAF4 AF1 *A_RESTRICT + #define retAL2 AL1 *A_RESTRICT + #define retAL3 AL1 *A_RESTRICT + #define retAL4 AL1 *A_RESTRICT + #define retAU2 AU1 *A_RESTRICT + #define retAU3 AU1 *A_RESTRICT + #define retAU4 AU1 *A_RESTRICT +//------------------------------------------------------------------------------------------------------------------------------ + #define inAD2 AD1 *A_RESTRICT + #define inAD3 AD1 *A_RESTRICT + #define inAD4 AD1 *A_RESTRICT + #define inAF2 AF1 *A_RESTRICT + #define inAF3 AF1 *A_RESTRICT + #define inAF4 AF1 *A_RESTRICT + #define inAL2 AL1 *A_RESTRICT + #define inAL3 AL1 *A_RESTRICT + #define inAL4 AL1 *A_RESTRICT + #define inAU2 AU1 *A_RESTRICT + #define inAU3 AU1 *A_RESTRICT + #define inAU4 AU1 *A_RESTRICT +//------------------------------------------------------------------------------------------------------------------------------ + #define inoutAD2 AD1 *A_RESTRICT + #define inoutAD3 AD1 *A_RESTRICT + #define inoutAD4 AD1 *A_RESTRICT + #define inoutAF2 AF1 *A_RESTRICT + #define inoutAF3 AF1 *A_RESTRICT + #define inoutAF4 AF1 *A_RESTRICT + #define inoutAL2 AL1 *A_RESTRICT + #define inoutAL3 AL1 *A_RESTRICT + #define inoutAL4 AL1 *A_RESTRICT + #define inoutAU2 AU1 *A_RESTRICT + #define inoutAU3 AU1 *A_RESTRICT + #define inoutAU4 AU1 *A_RESTRICT +//------------------------------------------------------------------------------------------------------------------------------ + #define outAD2 AD1 *A_RESTRICT + #define outAD3 AD1 *A_RESTRICT + #define outAD4 AD1 *A_RESTRICT + #define outAF2 AF1 *A_RESTRICT + #define outAF3 AF1 *A_RESTRICT + #define outAF4 AF1 *A_RESTRICT + #define outAL2 AL1 *A_RESTRICT + #define outAL3 AL1 *A_RESTRICT + #define outAL4 AL1 *A_RESTRICT + #define outAU2 AU1 *A_RESTRICT + #define outAU3 AU1 *A_RESTRICT + #define outAU4 AU1 *A_RESTRICT +//------------------------------------------------------------------------------------------------------------------------------ + #define varAD2(x) AD1 x[2] + #define varAD3(x) AD1 x[3] + #define varAD4(x) AD1 x[4] + #define varAF2(x) AF1 x[2] + #define varAF3(x) AF1 x[3] + #define varAF4(x) AF1 x[4] + #define varAL2(x) AL1 x[2] + #define varAL3(x) AL1 x[3] + #define varAL4(x) AL1 x[4] + #define varAU2(x) AU1 x[2] + #define varAU3(x) AU1 x[3] + #define varAU4(x) AU1 x[4] +//------------------------------------------------------------------------------------------------------------------------------ + #define initAD2(x,y) {x,y} + #define initAD3(x,y,z) {x,y,z} + #define initAD4(x,y,z,w) {x,y,z,w} + #define initAF2(x,y) {x,y} + #define initAF3(x,y,z) {x,y,z} + #define initAF4(x,y,z,w) {x,y,z,w} + #define initAL2(x,y) {x,y} + #define initAL3(x,y,z) {x,y,z} + #define initAL4(x,y,z,w) {x,y,z,w} + #define initAU2(x,y) {x,y} + #define initAU3(x,y,z) {x,y,z} + #define initAU4(x,y,z,w) {x,y,z,w} +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// SCALAR RETURN OPS +//------------------------------------------------------------------------------------------------------------------------------ +// TODO +// ==== +// - Replace transcendentals with manual versions. +//============================================================================================================================== + #ifdef A_GCC + A_STATIC AD1 AAbsD1(AD1 a){return __builtin_fabs(a);} + A_STATIC AF1 AAbsF1(AF1 a){return __builtin_fabsf(a);} + A_STATIC AU1 AAbsSU1(AU1 a){return AU1_(__builtin_abs(ASU1_(a)));} + A_STATIC AL1 AAbsSL1(AL1 a){return AL1_(__builtin_llabs(ASL1_(a)));} + #else + A_STATIC AD1 AAbsD1(AD1 a){return fabs(a);} + A_STATIC AF1 AAbsF1(AF1 a){return fabsf(a);} + A_STATIC AU1 AAbsSU1(AU1 a){return AU1_(abs(ASU1_(a)));} + A_STATIC AL1 AAbsSL1(AL1 a){return AL1_(labs((long)ASL1_(a)));} + #endif +//------------------------------------------------------------------------------------------------------------------------------ + #ifdef A_GCC + A_STATIC AD1 ACosD1(AD1 a){return __builtin_cos(a);} + A_STATIC AF1 ACosF1(AF1 a){return __builtin_cosf(a);} + #else + A_STATIC AD1 ACosD1(AD1 a){return cos(a);} + A_STATIC AF1 ACosF1(AF1 a){return cosf(a);} + #endif +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC AD1 ADotD2(inAD2 a,inAD2 b){return a[0]*b[0]+a[1]*b[1];} + A_STATIC AD1 ADotD3(inAD3 a,inAD3 b){return a[0]*b[0]+a[1]*b[1]+a[2]*b[2];} + A_STATIC AD1 ADotD4(inAD4 a,inAD4 b){return a[0]*b[0]+a[1]*b[1]+a[2]*b[2]+a[3]*b[3];} + A_STATIC AF1 ADotF2(inAF2 a,inAF2 b){return a[0]*b[0]+a[1]*b[1];} + A_STATIC AF1 ADotF3(inAF3 a,inAF3 b){return a[0]*b[0]+a[1]*b[1]+a[2]*b[2];} + A_STATIC AF1 ADotF4(inAF4 a,inAF4 b){return a[0]*b[0]+a[1]*b[1]+a[2]*b[2]+a[3]*b[3];} +//------------------------------------------------------------------------------------------------------------------------------ + #ifdef A_GCC + A_STATIC AD1 AExp2D1(AD1 a){return __builtin_exp2(a);} + A_STATIC AF1 AExp2F1(AF1 a){return __builtin_exp2f(a);} + #else + A_STATIC AD1 AExp2D1(AD1 a){return exp2(a);} + A_STATIC AF1 AExp2F1(AF1 a){return exp2f(a);} + #endif +//------------------------------------------------------------------------------------------------------------------------------ + #ifdef A_GCC + A_STATIC AD1 AFloorD1(AD1 a){return __builtin_floor(a);} + A_STATIC AF1 AFloorF1(AF1 a){return __builtin_floorf(a);} + #else + A_STATIC AD1 AFloorD1(AD1 a){return floor(a);} + A_STATIC AF1 AFloorF1(AF1 a){return floorf(a);} + #endif +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC AD1 ALerpD1(AD1 a,AD1 b,AD1 c){return b*c+(-a*c+a);} + A_STATIC AF1 ALerpF1(AF1 a,AF1 b,AF1 c){return b*c+(-a*c+a);} +//------------------------------------------------------------------------------------------------------------------------------ + #ifdef A_GCC + A_STATIC AD1 ALog2D1(AD1 a){return __builtin_log2(a);} + A_STATIC AF1 ALog2F1(AF1 a){return __builtin_log2f(a);} + #else + A_STATIC AD1 ALog2D1(AD1 a){return log2(a);} + A_STATIC AF1 ALog2F1(AF1 a){return log2f(a);} + #endif +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC AD1 AMaxD1(AD1 a,AD1 b){return a>b?a:b;} + A_STATIC AF1 AMaxF1(AF1 a,AF1 b){return a>b?a:b;} + A_STATIC AL1 AMaxL1(AL1 a,AL1 b){return a>b?a:b;} + A_STATIC AU1 AMaxU1(AU1 a,AU1 b){return a>b?a:b;} +//------------------------------------------------------------------------------------------------------------------------------ + // These follow the convention that A integer types don't have signage, until they are operated on. + A_STATIC AL1 AMaxSL1(AL1 a,AL1 b){return (ASL1_(a)>ASL1_(b))?a:b;} + A_STATIC AU1 AMaxSU1(AU1 a,AU1 b){return (ASU1_(a)>ASU1_(b))?a:b;} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC AD1 AMinD1(AD1 a,AD1 b){return a<b?a:b;} + A_STATIC AF1 AMinF1(AF1 a,AF1 b){return a<b?a:b;} + A_STATIC AL1 AMinL1(AL1 a,AL1 b){return a<b?a:b;} + A_STATIC AU1 AMinU1(AU1 a,AU1 b){return a<b?a:b;} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC AL1 AMinSL1(AL1 a,AL1 b){return (ASL1_(a)<ASL1_(b))?a:b;} + A_STATIC AU1 AMinSU1(AU1 a,AU1 b){return (ASU1_(a)<ASU1_(b))?a:b;} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC AD1 ARcpD1(AD1 a){return 1.0/a;} + A_STATIC AF1 ARcpF1(AF1 a){return 1.0f/a;} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC AL1 AShrSL1(AL1 a,AL1 b){return AL1_(ASL1_(a)>>ASL1_(b));} + A_STATIC AU1 AShrSU1(AU1 a,AU1 b){return AU1_(ASU1_(a)>>ASU1_(b));} +//------------------------------------------------------------------------------------------------------------------------------ + #ifdef A_GCC + A_STATIC AD1 ASinD1(AD1 a){return __builtin_sin(a);} + A_STATIC AF1 ASinF1(AF1 a){return __builtin_sinf(a);} + #else + A_STATIC AD1 ASinD1(AD1 a){return sin(a);} + A_STATIC AF1 ASinF1(AF1 a){return sinf(a);} + #endif +//------------------------------------------------------------------------------------------------------------------------------ + #ifdef A_GCC + A_STATIC AD1 ASqrtD1(AD1 a){return __builtin_sqrt(a);} + A_STATIC AF1 ASqrtF1(AF1 a){return __builtin_sqrtf(a);} + #else + A_STATIC AD1 ASqrtD1(AD1 a){return sqrt(a);} + A_STATIC AF1 ASqrtF1(AF1 a){return sqrtf(a);} + #endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// SCALAR RETURN OPS - DEPENDENT +//============================================================================================================================== + A_STATIC AD1 AClampD1(AD1 x,AD1 n,AD1 m){return AMaxD1(n,AMinD1(x,m));} + A_STATIC AF1 AClampF1(AF1 x,AF1 n,AF1 m){return AMaxF1(n,AMinF1(x,m));} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC AD1 AFractD1(AD1 a){return a-AFloorD1(a);} + A_STATIC AF1 AFractF1(AF1 a){return a-AFloorF1(a);} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC AD1 APowD1(AD1 a,AD1 b){return AExp2D1(b*ALog2D1(a));} + A_STATIC AF1 APowF1(AF1 a,AF1 b){return AExp2F1(b*ALog2F1(a));} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC AD1 ARsqD1(AD1 a){return ARcpD1(ASqrtD1(a));} + A_STATIC AF1 ARsqF1(AF1 a){return ARcpF1(ASqrtF1(a));} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC AD1 ASatD1(AD1 a){return AMinD1(1.0,AMaxD1(0.0,a));} + A_STATIC AF1 ASatF1(AF1 a){return AMinF1(1.0f,AMaxF1(0.0f,a));} +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// VECTOR OPS +//------------------------------------------------------------------------------------------------------------------------------ +// These are added as needed for production or prototyping, so not necessarily a complete set. +// They follow a convention of taking in a destination and also returning the destination value to increase utility. +//============================================================================================================================== + A_STATIC retAD2 opAAbsD2(outAD2 d,inAD2 a){d[0]=AAbsD1(a[0]);d[1]=AAbsD1(a[1]);return d;} + A_STATIC retAD3 opAAbsD3(outAD3 d,inAD3 a){d[0]=AAbsD1(a[0]);d[1]=AAbsD1(a[1]);d[2]=AAbsD1(a[2]);return d;} + A_STATIC retAD4 opAAbsD4(outAD4 d,inAD4 a){d[0]=AAbsD1(a[0]);d[1]=AAbsD1(a[1]);d[2]=AAbsD1(a[2]);d[3]=AAbsD1(a[3]);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC retAF2 opAAbsF2(outAF2 d,inAF2 a){d[0]=AAbsF1(a[0]);d[1]=AAbsF1(a[1]);return d;} + A_STATIC retAF3 opAAbsF3(outAF3 d,inAF3 a){d[0]=AAbsF1(a[0]);d[1]=AAbsF1(a[1]);d[2]=AAbsF1(a[2]);return d;} + A_STATIC retAF4 opAAbsF4(outAF4 d,inAF4 a){d[0]=AAbsF1(a[0]);d[1]=AAbsF1(a[1]);d[2]=AAbsF1(a[2]);d[3]=AAbsF1(a[3]);return d;} +//============================================================================================================================== + A_STATIC retAD2 opAAddD2(outAD2 d,inAD2 a,inAD2 b){d[0]=a[0]+b[0];d[1]=a[1]+b[1];return d;} + A_STATIC retAD3 opAAddD3(outAD3 d,inAD3 a,inAD3 b){d[0]=a[0]+b[0];d[1]=a[1]+b[1];d[2]=a[2]+b[2];return d;} + A_STATIC retAD4 opAAddD4(outAD4 d,inAD4 a,inAD4 b){d[0]=a[0]+b[0];d[1]=a[1]+b[1];d[2]=a[2]+b[2];d[3]=a[3]+b[3];return d;} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC retAF2 opAAddF2(outAF2 d,inAF2 a,inAF2 b){d[0]=a[0]+b[0];d[1]=a[1]+b[1];return d;} + A_STATIC retAF3 opAAddF3(outAF3 d,inAF3 a,inAF3 b){d[0]=a[0]+b[0];d[1]=a[1]+b[1];d[2]=a[2]+b[2];return d;} + A_STATIC retAF4 opAAddF4(outAF4 d,inAF4 a,inAF4 b){d[0]=a[0]+b[0];d[1]=a[1]+b[1];d[2]=a[2]+b[2];d[3]=a[3]+b[3];return d;} +//============================================================================================================================== + A_STATIC retAD2 opAAddOneD2(outAD2 d,inAD2 a,AD1 b){d[0]=a[0]+b;d[1]=a[1]+b;return d;} + A_STATIC retAD3 opAAddOneD3(outAD3 d,inAD3 a,AD1 b){d[0]=a[0]+b;d[1]=a[1]+b;d[2]=a[2]+b;return d;} + A_STATIC retAD4 opAAddOneD4(outAD4 d,inAD4 a,AD1 b){d[0]=a[0]+b;d[1]=a[1]+b;d[2]=a[2]+b;d[3]=a[3]+b;return d;} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC retAF2 opAAddOneF2(outAF2 d,inAF2 a,AF1 b){d[0]=a[0]+b;d[1]=a[1]+b;return d;} + A_STATIC retAF3 opAAddOneF3(outAF3 d,inAF3 a,AF1 b){d[0]=a[0]+b;d[1]=a[1]+b;d[2]=a[2]+b;return d;} + A_STATIC retAF4 opAAddOneF4(outAF4 d,inAF4 a,AF1 b){d[0]=a[0]+b;d[1]=a[1]+b;d[2]=a[2]+b;d[3]=a[3]+b;return d;} +//============================================================================================================================== + A_STATIC retAD2 opACpyD2(outAD2 d,inAD2 a){d[0]=a[0];d[1]=a[1];return d;} + A_STATIC retAD3 opACpyD3(outAD3 d,inAD3 a){d[0]=a[0];d[1]=a[1];d[2]=a[2];return d;} + A_STATIC retAD4 opACpyD4(outAD4 d,inAD4 a){d[0]=a[0];d[1]=a[1];d[2]=a[2];d[3]=a[3];return d;} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC retAF2 opACpyF2(outAF2 d,inAF2 a){d[0]=a[0];d[1]=a[1];return d;} + A_STATIC retAF3 opACpyF3(outAF3 d,inAF3 a){d[0]=a[0];d[1]=a[1];d[2]=a[2];return d;} + A_STATIC retAF4 opACpyF4(outAF4 d,inAF4 a){d[0]=a[0];d[1]=a[1];d[2]=a[2];d[3]=a[3];return d;} +//============================================================================================================================== + A_STATIC retAD2 opALerpD2(outAD2 d,inAD2 a,inAD2 b,inAD2 c){d[0]=ALerpD1(a[0],b[0],c[0]);d[1]=ALerpD1(a[1],b[1],c[1]);return d;} + A_STATIC retAD3 opALerpD3(outAD3 d,inAD3 a,inAD3 b,inAD3 c){d[0]=ALerpD1(a[0],b[0],c[0]);d[1]=ALerpD1(a[1],b[1],c[1]);d[2]=ALerpD1(a[2],b[2],c[2]);return d;} + A_STATIC retAD4 opALerpD4(outAD4 d,inAD4 a,inAD4 b,inAD4 c){d[0]=ALerpD1(a[0],b[0],c[0]);d[1]=ALerpD1(a[1],b[1],c[1]);d[2]=ALerpD1(a[2],b[2],c[2]);d[3]=ALerpD1(a[3],b[3],c[3]);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC retAF2 opALerpF2(outAF2 d,inAF2 a,inAF2 b,inAF2 c){d[0]=ALerpF1(a[0],b[0],c[0]);d[1]=ALerpF1(a[1],b[1],c[1]);return d;} + A_STATIC retAF3 opALerpF3(outAF3 d,inAF3 a,inAF3 b,inAF3 c){d[0]=ALerpF1(a[0],b[0],c[0]);d[1]=ALerpF1(a[1],b[1],c[1]);d[2]=ALerpF1(a[2],b[2],c[2]);return d;} + A_STATIC retAF4 opALerpF4(outAF4 d,inAF4 a,inAF4 b,inAF4 c){d[0]=ALerpF1(a[0],b[0],c[0]);d[1]=ALerpF1(a[1],b[1],c[1]);d[2]=ALerpF1(a[2],b[2],c[2]);d[3]=ALerpF1(a[3],b[3],c[3]);return d;} +//============================================================================================================================== + A_STATIC retAD2 opALerpOneD2(outAD2 d,inAD2 a,inAD2 b,AD1 c){d[0]=ALerpD1(a[0],b[0],c);d[1]=ALerpD1(a[1],b[1],c);return d;} + A_STATIC retAD3 opALerpOneD3(outAD3 d,inAD3 a,inAD3 b,AD1 c){d[0]=ALerpD1(a[0],b[0],c);d[1]=ALerpD1(a[1],b[1],c);d[2]=ALerpD1(a[2],b[2],c);return d;} + A_STATIC retAD4 opALerpOneD4(outAD4 d,inAD4 a,inAD4 b,AD1 c){d[0]=ALerpD1(a[0],b[0],c);d[1]=ALerpD1(a[1],b[1],c);d[2]=ALerpD1(a[2],b[2],c);d[3]=ALerpD1(a[3],b[3],c);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC retAF2 opALerpOneF2(outAF2 d,inAF2 a,inAF2 b,AF1 c){d[0]=ALerpF1(a[0],b[0],c);d[1]=ALerpF1(a[1],b[1],c);return d;} + A_STATIC retAF3 opALerpOneF3(outAF3 d,inAF3 a,inAF3 b,AF1 c){d[0]=ALerpF1(a[0],b[0],c);d[1]=ALerpF1(a[1],b[1],c);d[2]=ALerpF1(a[2],b[2],c);return d;} + A_STATIC retAF4 opALerpOneF4(outAF4 d,inAF4 a,inAF4 b,AF1 c){d[0]=ALerpF1(a[0],b[0],c);d[1]=ALerpF1(a[1],b[1],c);d[2]=ALerpF1(a[2],b[2],c);d[3]=ALerpF1(a[3],b[3],c);return d;} +//============================================================================================================================== + A_STATIC retAD2 opAMaxD2(outAD2 d,inAD2 a,inAD2 b){d[0]=AMaxD1(a[0],b[0]);d[1]=AMaxD1(a[1],b[1]);return d;} + A_STATIC retAD3 opAMaxD3(outAD3 d,inAD3 a,inAD3 b){d[0]=AMaxD1(a[0],b[0]);d[1]=AMaxD1(a[1],b[1]);d[2]=AMaxD1(a[2],b[2]);return d;} + A_STATIC retAD4 opAMaxD4(outAD4 d,inAD4 a,inAD4 b){d[0]=AMaxD1(a[0],b[0]);d[1]=AMaxD1(a[1],b[1]);d[2]=AMaxD1(a[2],b[2]);d[3]=AMaxD1(a[3],b[3]);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC retAF2 opAMaxF2(outAF2 d,inAF2 a,inAF2 b){d[0]=AMaxF1(a[0],b[0]);d[1]=AMaxF1(a[1],b[1]);return d;} + A_STATIC retAF3 opAMaxF3(outAF3 d,inAF3 a,inAF3 b){d[0]=AMaxF1(a[0],b[0]);d[1]=AMaxF1(a[1],b[1]);d[2]=AMaxF1(a[2],b[2]);return d;} + A_STATIC retAF4 opAMaxF4(outAF4 d,inAF4 a,inAF4 b){d[0]=AMaxF1(a[0],b[0]);d[1]=AMaxF1(a[1],b[1]);d[2]=AMaxF1(a[2],b[2]);d[3]=AMaxF1(a[3],b[3]);return d;} +//============================================================================================================================== + A_STATIC retAD2 opAMinD2(outAD2 d,inAD2 a,inAD2 b){d[0]=AMinD1(a[0],b[0]);d[1]=AMinD1(a[1],b[1]);return d;} + A_STATIC retAD3 opAMinD3(outAD3 d,inAD3 a,inAD3 b){d[0]=AMinD1(a[0],b[0]);d[1]=AMinD1(a[1],b[1]);d[2]=AMinD1(a[2],b[2]);return d;} + A_STATIC retAD4 opAMinD4(outAD4 d,inAD4 a,inAD4 b){d[0]=AMinD1(a[0],b[0]);d[1]=AMinD1(a[1],b[1]);d[2]=AMinD1(a[2],b[2]);d[3]=AMinD1(a[3],b[3]);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC retAF2 opAMinF2(outAF2 d,inAF2 a,inAF2 b){d[0]=AMinF1(a[0],b[0]);d[1]=AMinF1(a[1],b[1]);return d;} + A_STATIC retAF3 opAMinF3(outAF3 d,inAF3 a,inAF3 b){d[0]=AMinF1(a[0],b[0]);d[1]=AMinF1(a[1],b[1]);d[2]=AMinF1(a[2],b[2]);return d;} + A_STATIC retAF4 opAMinF4(outAF4 d,inAF4 a,inAF4 b){d[0]=AMinF1(a[0],b[0]);d[1]=AMinF1(a[1],b[1]);d[2]=AMinF1(a[2],b[2]);d[3]=AMinF1(a[3],b[3]);return d;} +//============================================================================================================================== + A_STATIC retAD2 opAMulD2(outAD2 d,inAD2 a,inAD2 b){d[0]=a[0]*b[0];d[1]=a[1]*b[1];return d;} + A_STATIC retAD3 opAMulD3(outAD3 d,inAD3 a,inAD3 b){d[0]=a[0]*b[0];d[1]=a[1]*b[1];d[2]=a[2]*b[2];return d;} + A_STATIC retAD4 opAMulD4(outAD4 d,inAD4 a,inAD4 b){d[0]=a[0]*b[0];d[1]=a[1]*b[1];d[2]=a[2]*b[2];d[3]=a[3]*b[3];return d;} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC retAF2 opAMulF2(outAF2 d,inAF2 a,inAF2 b){d[0]=a[0]*b[0];d[1]=a[1]*b[1];return d;} + A_STATIC retAF3 opAMulF3(outAF3 d,inAF3 a,inAF3 b){d[0]=a[0]*b[0];d[1]=a[1]*b[1];d[2]=a[2]*b[2];return d;} + A_STATIC retAF4 opAMulF4(outAF4 d,inAF4 a,inAF4 b){d[0]=a[0]*b[0];d[1]=a[1]*b[1];d[2]=a[2]*b[2];d[3]=a[3]*b[3];return d;} +//============================================================================================================================== + A_STATIC retAD2 opAMulOneD2(outAD2 d,inAD2 a,AD1 b){d[0]=a[0]*b;d[1]=a[1]*b;return d;} + A_STATIC retAD3 opAMulOneD3(outAD3 d,inAD3 a,AD1 b){d[0]=a[0]*b;d[1]=a[1]*b;d[2]=a[2]*b;return d;} + A_STATIC retAD4 opAMulOneD4(outAD4 d,inAD4 a,AD1 b){d[0]=a[0]*b;d[1]=a[1]*b;d[2]=a[2]*b;d[3]=a[3]*b;return d;} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC retAF2 opAMulOneF2(outAF2 d,inAF2 a,AF1 b){d[0]=a[0]*b;d[1]=a[1]*b;return d;} + A_STATIC retAF3 opAMulOneF3(outAF3 d,inAF3 a,AF1 b){d[0]=a[0]*b;d[1]=a[1]*b;d[2]=a[2]*b;return d;} + A_STATIC retAF4 opAMulOneF4(outAF4 d,inAF4 a,AF1 b){d[0]=a[0]*b;d[1]=a[1]*b;d[2]=a[2]*b;d[3]=a[3]*b;return d;} +//============================================================================================================================== + A_STATIC retAD2 opANegD2(outAD2 d,inAD2 a){d[0]=-a[0];d[1]=-a[1];return d;} + A_STATIC retAD3 opANegD3(outAD3 d,inAD3 a){d[0]=-a[0];d[1]=-a[1];d[2]=-a[2];return d;} + A_STATIC retAD4 opANegD4(outAD4 d,inAD4 a){d[0]=-a[0];d[1]=-a[1];d[2]=-a[2];d[3]=-a[3];return d;} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC retAF2 opANegF2(outAF2 d,inAF2 a){d[0]=-a[0];d[1]=-a[1];return d;} + A_STATIC retAF3 opANegF3(outAF3 d,inAF3 a){d[0]=-a[0];d[1]=-a[1];d[2]=-a[2];return d;} + A_STATIC retAF4 opANegF4(outAF4 d,inAF4 a){d[0]=-a[0];d[1]=-a[1];d[2]=-a[2];d[3]=-a[3];return d;} +//============================================================================================================================== + A_STATIC retAD2 opARcpD2(outAD2 d,inAD2 a){d[0]=ARcpD1(a[0]);d[1]=ARcpD1(a[1]);return d;} + A_STATIC retAD3 opARcpD3(outAD3 d,inAD3 a){d[0]=ARcpD1(a[0]);d[1]=ARcpD1(a[1]);d[2]=ARcpD1(a[2]);return d;} + A_STATIC retAD4 opARcpD4(outAD4 d,inAD4 a){d[0]=ARcpD1(a[0]);d[1]=ARcpD1(a[1]);d[2]=ARcpD1(a[2]);d[3]=ARcpD1(a[3]);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + A_STATIC retAF2 opARcpF2(outAF2 d,inAF2 a){d[0]=ARcpF1(a[0]);d[1]=ARcpF1(a[1]);return d;} + A_STATIC retAF3 opARcpF3(outAF3 d,inAF3 a){d[0]=ARcpF1(a[0]);d[1]=ARcpF1(a[1]);d[2]=ARcpF1(a[2]);return d;} + A_STATIC retAF4 opARcpF4(outAF4 d,inAF4 a){d[0]=ARcpF1(a[0]);d[1]=ARcpF1(a[1]);d[2]=ARcpF1(a[2]);d[3]=ARcpF1(a[3]);return d;} +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// HALF FLOAT PACKING +//============================================================================================================================== + // Convert float to half (in lower 16-bits of output). + // Same fast technique as documented here: ftp://ftp.fox-toolkit.org/pub/fasthalffloatconversion.pdf + // Supports denormals. + // Conversion rules are to make computations possibly "safer" on the GPU, + // -INF & -NaN -> -65504 + // +INF & +NaN -> +65504 + A_STATIC AU1 AU1_AH1_AF1(AF1 f){ + static AW1 base[512]={ + 0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000, + 0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000, + 0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000, + 0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000, + 0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000, + 0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000, + 0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0001,0x0002,0x0004,0x0008,0x0010,0x0020,0x0040,0x0080,0x0100, + 0x0200,0x0400,0x0800,0x0c00,0x1000,0x1400,0x1800,0x1c00,0x2000,0x2400,0x2800,0x2c00,0x3000,0x3400,0x3800,0x3c00, + 0x4000,0x4400,0x4800,0x4c00,0x5000,0x5400,0x5800,0x5c00,0x6000,0x6400,0x6800,0x6c00,0x7000,0x7400,0x7800,0x7bff, + 0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff, + 0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff, + 0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff, + 0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff, + 0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff, + 0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff, + 0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff, + 0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000, + 0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000, + 0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000, + 0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000, + 0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000, + 0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000, + 0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8001,0x8002,0x8004,0x8008,0x8010,0x8020,0x8040,0x8080,0x8100, + 0x8200,0x8400,0x8800,0x8c00,0x9000,0x9400,0x9800,0x9c00,0xa000,0xa400,0xa800,0xac00,0xb000,0xb400,0xb800,0xbc00, + 0xc000,0xc400,0xc800,0xcc00,0xd000,0xd400,0xd800,0xdc00,0xe000,0xe400,0xe800,0xec00,0xf000,0xf400,0xf800,0xfbff, + 0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff, + 0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff, + 0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff, + 0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff, + 0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff, + 0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff, + 0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff}; + static AB1 shift[512]={ + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x17,0x16,0x15,0x14,0x13,0x12,0x11,0x10,0x0f, + 0x0e,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d, + 0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x17,0x16,0x15,0x14,0x13,0x12,0x11,0x10,0x0f, + 0x0e,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d, + 0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18, + 0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18}; + union{AF1 f;AU1 u;}bits;bits.f=f;AU1 u=bits.u;AU1 i=u>>23;return (AU1)(base[i])+((u&0x7fffff)>>shift[i]);} +//------------------------------------------------------------------------------------------------------------------------------ + // Used to output packed constant. + A_STATIC AU1 AU1_AH2_AF2(inAF2 a){return AU1_AH1_AF1(a[0])+(AU1_AH1_AF1(a[1])<<16);} +#endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// +// GLSL +// +// +//============================================================================================================================== +#if defined(A_GLSL) && defined(A_GPU) + #ifndef A_SKIP_EXT + #ifdef A_HALF + #extension GL_EXT_shader_16bit_storage:require + #extension GL_EXT_shader_explicit_arithmetic_types:require + #endif +//------------------------------------------------------------------------------------------------------------------------------ + #ifdef A_LONG + #extension GL_ARB_gpu_shader_int64:require + #extension GL_NV_shader_atomic_int64:require + #endif +//------------------------------------------------------------------------------------------------------------------------------ + #ifdef A_WAVE + #extension GL_KHR_shader_subgroup_arithmetic:require + #extension GL_KHR_shader_subgroup_ballot:require + #extension GL_KHR_shader_subgroup_quad:require + #extension GL_KHR_shader_subgroup_shuffle:require + #endif + #endif +//============================================================================================================================== + #define AP1 bool + #define AP2 bvec2 + #define AP3 bvec3 + #define AP4 bvec4 +//------------------------------------------------------------------------------------------------------------------------------ + #define AF1 float + #define AF2 vec2 + #define AF3 vec3 + #define AF4 vec4 +//------------------------------------------------------------------------------------------------------------------------------ + #define AU1 uint + #define AU2 uvec2 + #define AU3 uvec3 + #define AU4 uvec4 +//------------------------------------------------------------------------------------------------------------------------------ + #define ASU1 int + #define ASU2 ivec2 + #define ASU3 ivec3 + #define ASU4 ivec4 +//============================================================================================================================== + #define AF1_AU1(x) uintBitsToFloat(AU1(x)) + #define AF2_AU2(x) uintBitsToFloat(AU2(x)) + #define AF3_AU3(x) uintBitsToFloat(AU3(x)) + #define AF4_AU4(x) uintBitsToFloat(AU4(x)) +//------------------------------------------------------------------------------------------------------------------------------ + #define AU1_AF1(x) floatBitsToUint(AF1(x)) + #define AU2_AF2(x) floatBitsToUint(AF2(x)) + #define AU3_AF3(x) floatBitsToUint(AF3(x)) + #define AU4_AF4(x) floatBitsToUint(AF4(x)) +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AU1_AH1_AF1_x(AF1 a){return packHalf2x16(AF2(a,0.0));} + #define AU1_AH1_AF1(a) AU1_AH1_AF1_x(AF1(a)) +//------------------------------------------------------------------------------------------------------------------------------ + #define AU1_AH2_AF2 packHalf2x16 + #define AU1_AW2Unorm_AF2 packUnorm2x16 + #define AU1_AB4Unorm_AF4 packUnorm4x8 +//------------------------------------------------------------------------------------------------------------------------------ + #define AF2_AH2_AU1 unpackHalf2x16 + #define AF2_AW2Unorm_AU1 unpackUnorm2x16 + #define AF4_AB4Unorm_AU1 unpackUnorm4x8 +//============================================================================================================================== + AF1 AF1_x(AF1 a){return AF1(a);} + AF2 AF2_x(AF1 a){return AF2(a,a);} + AF3 AF3_x(AF1 a){return AF3(a,a,a);} + AF4 AF4_x(AF1 a){return AF4(a,a,a,a);} + #define AF1_(a) AF1_x(AF1(a)) + #define AF2_(a) AF2_x(AF1(a)) + #define AF3_(a) AF3_x(AF1(a)) + #define AF4_(a) AF4_x(AF1(a)) +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AU1_x(AU1 a){return AU1(a);} + AU2 AU2_x(AU1 a){return AU2(a,a);} + AU3 AU3_x(AU1 a){return AU3(a,a,a);} + AU4 AU4_x(AU1 a){return AU4(a,a,a,a);} + #define AU1_(a) AU1_x(AU1(a)) + #define AU2_(a) AU2_x(AU1(a)) + #define AU3_(a) AU3_x(AU1(a)) + #define AU4_(a) AU4_x(AU1(a)) +//============================================================================================================================== + AU1 AAbsSU1(AU1 a){return AU1(abs(ASU1(a)));} + AU2 AAbsSU2(AU2 a){return AU2(abs(ASU2(a)));} + AU3 AAbsSU3(AU3 a){return AU3(abs(ASU3(a)));} + AU4 AAbsSU4(AU4 a){return AU4(abs(ASU4(a)));} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 ABfe(AU1 src,AU1 off,AU1 bits){return bitfieldExtract(src,ASU1(off),ASU1(bits));} + AU1 ABfi(AU1 src,AU1 ins,AU1 mask){return (ins&mask)|(src&(~mask));} + // Proxy for V_BFI_B32 where the 'mask' is set as 'bits', 'mask=(1<<bits)-1', and 'bits' needs to be an immediate. + AU1 ABfiM(AU1 src,AU1 ins,AU1 bits){return bitfieldInsert(src,ins,0,ASU1(bits));} +//------------------------------------------------------------------------------------------------------------------------------ + // V_MED3_F32. + AF1 AClampF1(AF1 x,AF1 n,AF1 m){return clamp(x,n,m);} + AF2 AClampF2(AF2 x,AF2 n,AF2 m){return clamp(x,n,m);} + AF3 AClampF3(AF3 x,AF3 n,AF3 m){return clamp(x,n,m);} + AF4 AClampF4(AF4 x,AF4 n,AF4 m){return clamp(x,n,m);} +//------------------------------------------------------------------------------------------------------------------------------ + // V_FRACT_F32 (note DX frac() is different). + AF1 AFractF1(AF1 x){return fract(x);} + AF2 AFractF2(AF2 x){return fract(x);} + AF3 AFractF3(AF3 x){return fract(x);} + AF4 AFractF4(AF4 x){return fract(x);} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 ALerpF1(AF1 x,AF1 y,AF1 a){return mix(x,y,a);} + AF2 ALerpF2(AF2 x,AF2 y,AF2 a){return mix(x,y,a);} + AF3 ALerpF3(AF3 x,AF3 y,AF3 a){return mix(x,y,a);} + AF4 ALerpF4(AF4 x,AF4 y,AF4 a){return mix(x,y,a);} +//------------------------------------------------------------------------------------------------------------------------------ + // V_MAX3_F32. + AF1 AMax3F1(AF1 x,AF1 y,AF1 z){return max(x,max(y,z));} + AF2 AMax3F2(AF2 x,AF2 y,AF2 z){return max(x,max(y,z));} + AF3 AMax3F3(AF3 x,AF3 y,AF3 z){return max(x,max(y,z));} + AF4 AMax3F4(AF4 x,AF4 y,AF4 z){return max(x,max(y,z));} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AMax3SU1(AU1 x,AU1 y,AU1 z){return AU1(max(ASU1(x),max(ASU1(y),ASU1(z))));} + AU2 AMax3SU2(AU2 x,AU2 y,AU2 z){return AU2(max(ASU2(x),max(ASU2(y),ASU2(z))));} + AU3 AMax3SU3(AU3 x,AU3 y,AU3 z){return AU3(max(ASU3(x),max(ASU3(y),ASU3(z))));} + AU4 AMax3SU4(AU4 x,AU4 y,AU4 z){return AU4(max(ASU4(x),max(ASU4(y),ASU4(z))));} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AMax3U1(AU1 x,AU1 y,AU1 z){return max(x,max(y,z));} + AU2 AMax3U2(AU2 x,AU2 y,AU2 z){return max(x,max(y,z));} + AU3 AMax3U3(AU3 x,AU3 y,AU3 z){return max(x,max(y,z));} + AU4 AMax3U4(AU4 x,AU4 y,AU4 z){return max(x,max(y,z));} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AMaxSU1(AU1 a,AU1 b){return AU1(max(ASU1(a),ASU1(b)));} + AU2 AMaxSU2(AU2 a,AU2 b){return AU2(max(ASU2(a),ASU2(b)));} + AU3 AMaxSU3(AU3 a,AU3 b){return AU3(max(ASU3(a),ASU3(b)));} + AU4 AMaxSU4(AU4 a,AU4 b){return AU4(max(ASU4(a),ASU4(b)));} +//------------------------------------------------------------------------------------------------------------------------------ + // Clamp has an easier pattern match for med3 when some ordering is known. + // V_MED3_F32. + AF1 AMed3F1(AF1 x,AF1 y,AF1 z){return max(min(x,y),min(max(x,y),z));} + AF2 AMed3F2(AF2 x,AF2 y,AF2 z){return max(min(x,y),min(max(x,y),z));} + AF3 AMed3F3(AF3 x,AF3 y,AF3 z){return max(min(x,y),min(max(x,y),z));} + AF4 AMed3F4(AF4 x,AF4 y,AF4 z){return max(min(x,y),min(max(x,y),z));} +//------------------------------------------------------------------------------------------------------------------------------ + // V_MIN3_F32. + AF1 AMin3F1(AF1 x,AF1 y,AF1 z){return min(x,min(y,z));} + AF2 AMin3F2(AF2 x,AF2 y,AF2 z){return min(x,min(y,z));} + AF3 AMin3F3(AF3 x,AF3 y,AF3 z){return min(x,min(y,z));} + AF4 AMin3F4(AF4 x,AF4 y,AF4 z){return min(x,min(y,z));} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AMin3SU1(AU1 x,AU1 y,AU1 z){return AU1(min(ASU1(x),min(ASU1(y),ASU1(z))));} + AU2 AMin3SU2(AU2 x,AU2 y,AU2 z){return AU2(min(ASU2(x),min(ASU2(y),ASU2(z))));} + AU3 AMin3SU3(AU3 x,AU3 y,AU3 z){return AU3(min(ASU3(x),min(ASU3(y),ASU3(z))));} + AU4 AMin3SU4(AU4 x,AU4 y,AU4 z){return AU4(min(ASU4(x),min(ASU4(y),ASU4(z))));} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AMin3U1(AU1 x,AU1 y,AU1 z){return min(x,min(y,z));} + AU2 AMin3U2(AU2 x,AU2 y,AU2 z){return min(x,min(y,z));} + AU3 AMin3U3(AU3 x,AU3 y,AU3 z){return min(x,min(y,z));} + AU4 AMin3U4(AU4 x,AU4 y,AU4 z){return min(x,min(y,z));} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AMinSU1(AU1 a,AU1 b){return AU1(min(ASU1(a),ASU1(b)));} + AU2 AMinSU2(AU2 a,AU2 b){return AU2(min(ASU2(a),ASU2(b)));} + AU3 AMinSU3(AU3 a,AU3 b){return AU3(min(ASU3(a),ASU3(b)));} + AU4 AMinSU4(AU4 a,AU4 b){return AU4(min(ASU4(a),ASU4(b)));} +//------------------------------------------------------------------------------------------------------------------------------ + // Normalized trig. Valid input domain is {-256 to +256}. No GLSL compiler intrinsic exists to map to this currently. + // V_COS_F32. + AF1 ANCosF1(AF1 x){return cos(x*AF1_(A_2PI));} + AF2 ANCosF2(AF2 x){return cos(x*AF2_(A_2PI));} + AF3 ANCosF3(AF3 x){return cos(x*AF3_(A_2PI));} + AF4 ANCosF4(AF4 x){return cos(x*AF4_(A_2PI));} +//------------------------------------------------------------------------------------------------------------------------------ + // Normalized trig. Valid input domain is {-256 to +256}. No GLSL compiler intrinsic exists to map to this currently. + // V_SIN_F32. + AF1 ANSinF1(AF1 x){return sin(x*AF1_(A_2PI));} + AF2 ANSinF2(AF2 x){return sin(x*AF2_(A_2PI));} + AF3 ANSinF3(AF3 x){return sin(x*AF3_(A_2PI));} + AF4 ANSinF4(AF4 x){return sin(x*AF4_(A_2PI));} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 ARcpF1(AF1 x){return AF1_(1.0)/x;} + AF2 ARcpF2(AF2 x){return AF2_(1.0)/x;} + AF3 ARcpF3(AF3 x){return AF3_(1.0)/x;} + AF4 ARcpF4(AF4 x){return AF4_(1.0)/x;} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 ARsqF1(AF1 x){return AF1_(1.0)/sqrt(x);} + AF2 ARsqF2(AF2 x){return AF2_(1.0)/sqrt(x);} + AF3 ARsqF3(AF3 x){return AF3_(1.0)/sqrt(x);} + AF4 ARsqF4(AF4 x){return AF4_(1.0)/sqrt(x);} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 ASatF1(AF1 x){return clamp(x,AF1_(0.0),AF1_(1.0));} + AF2 ASatF2(AF2 x){return clamp(x,AF2_(0.0),AF2_(1.0));} + AF3 ASatF3(AF3 x){return clamp(x,AF3_(0.0),AF3_(1.0));} + AF4 ASatF4(AF4 x){return clamp(x,AF4_(0.0),AF4_(1.0));} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AShrSU1(AU1 a,AU1 b){return AU1(ASU1(a)>>ASU1(b));} + AU2 AShrSU2(AU2 a,AU2 b){return AU2(ASU2(a)>>ASU2(b));} + AU3 AShrSU3(AU3 a,AU3 b){return AU3(ASU3(a)>>ASU3(b));} + AU4 AShrSU4(AU4 a,AU4 b){return AU4(ASU4(a)>>ASU4(b));} +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// GLSL BYTE +//============================================================================================================================== + #ifdef A_BYTE + #define AB1 uint8_t + #define AB2 u8vec2 + #define AB3 u8vec3 + #define AB4 u8vec4 +//------------------------------------------------------------------------------------------------------------------------------ + #define ASB1 int8_t + #define ASB2 i8vec2 + #define ASB3 i8vec3 + #define ASB4 i8vec4 +//------------------------------------------------------------------------------------------------------------------------------ + AB1 AB1_x(AB1 a){return AB1(a);} + AB2 AB2_x(AB1 a){return AB2(a,a);} + AB3 AB3_x(AB1 a){return AB3(a,a,a);} + AB4 AB4_x(AB1 a){return AB4(a,a,a,a);} + #define AB1_(a) AB1_x(AB1(a)) + #define AB2_(a) AB2_x(AB1(a)) + #define AB3_(a) AB3_x(AB1(a)) + #define AB4_(a) AB4_x(AB1(a)) + #endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// GLSL HALF +//============================================================================================================================== + #ifdef A_HALF + #define AH1 float16_t + #define AH2 f16vec2 + #define AH3 f16vec3 + #define AH4 f16vec4 +//------------------------------------------------------------------------------------------------------------------------------ + #define AW1 uint16_t + #define AW2 u16vec2 + #define AW3 u16vec3 + #define AW4 u16vec4 +//------------------------------------------------------------------------------------------------------------------------------ + #define ASW1 int16_t + #define ASW2 i16vec2 + #define ASW3 i16vec3 + #define ASW4 i16vec4 +//============================================================================================================================== + #define AH2_AU1(x) unpackFloat2x16(AU1(x)) + AH4 AH4_AU2_x(AU2 x){return AH4(unpackFloat2x16(x.x),unpackFloat2x16(x.y));} + #define AH4_AU2(x) AH4_AU2_x(AU2(x)) + #define AW2_AU1(x) unpackUint2x16(AU1(x)) + #define AW4_AU2(x) unpackUint4x16(pack64(AU2(x))) +//------------------------------------------------------------------------------------------------------------------------------ + #define AU1_AH2(x) packFloat2x16(AH2(x)) + AU2 AU2_AH4_x(AH4 x){return AU2(packFloat2x16(x.xy),packFloat2x16(x.zw));} + #define AU2_AH4(x) AU2_AH4_x(AH4(x)) + #define AU1_AW2(x) packUint2x16(AW2(x)) + #define AU2_AW4(x) unpack32(packUint4x16(AW4(x))) +//============================================================================================================================== + #define AW1_AH1(x) halfBitsToUint16(AH1(x)) + #define AW2_AH2(x) halfBitsToUint16(AH2(x)) + #define AW3_AH3(x) halfBitsToUint16(AH3(x)) + #define AW4_AH4(x) halfBitsToUint16(AH4(x)) +//------------------------------------------------------------------------------------------------------------------------------ + #define AH1_AW1(x) uint16BitsToHalf(AW1(x)) + #define AH2_AW2(x) uint16BitsToHalf(AW2(x)) + #define AH3_AW3(x) uint16BitsToHalf(AW3(x)) + #define AH4_AW4(x) uint16BitsToHalf(AW4(x)) +//============================================================================================================================== + AH1 AH1_x(AH1 a){return AH1(a);} + AH2 AH2_x(AH1 a){return AH2(a,a);} + AH3 AH3_x(AH1 a){return AH3(a,a,a);} + AH4 AH4_x(AH1 a){return AH4(a,a,a,a);} + #define AH1_(a) AH1_x(AH1(a)) + #define AH2_(a) AH2_x(AH1(a)) + #define AH3_(a) AH3_x(AH1(a)) + #define AH4_(a) AH4_x(AH1(a)) +//------------------------------------------------------------------------------------------------------------------------------ + AW1 AW1_x(AW1 a){return AW1(a);} + AW2 AW2_x(AW1 a){return AW2(a,a);} + AW3 AW3_x(AW1 a){return AW3(a,a,a);} + AW4 AW4_x(AW1 a){return AW4(a,a,a,a);} + #define AW1_(a) AW1_x(AW1(a)) + #define AW2_(a) AW2_x(AW1(a)) + #define AW3_(a) AW3_x(AW1(a)) + #define AW4_(a) AW4_x(AW1(a)) +//============================================================================================================================== + AW1 AAbsSW1(AW1 a){return AW1(abs(ASW1(a)));} + AW2 AAbsSW2(AW2 a){return AW2(abs(ASW2(a)));} + AW3 AAbsSW3(AW3 a){return AW3(abs(ASW3(a)));} + AW4 AAbsSW4(AW4 a){return AW4(abs(ASW4(a)));} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AClampH1(AH1 x,AH1 n,AH1 m){return clamp(x,n,m);} + AH2 AClampH2(AH2 x,AH2 n,AH2 m){return clamp(x,n,m);} + AH3 AClampH3(AH3 x,AH3 n,AH3 m){return clamp(x,n,m);} + AH4 AClampH4(AH4 x,AH4 n,AH4 m){return clamp(x,n,m);} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AFractH1(AH1 x){return fract(x);} + AH2 AFractH2(AH2 x){return fract(x);} + AH3 AFractH3(AH3 x){return fract(x);} + AH4 AFractH4(AH4 x){return fract(x);} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 ALerpH1(AH1 x,AH1 y,AH1 a){return mix(x,y,a);} + AH2 ALerpH2(AH2 x,AH2 y,AH2 a){return mix(x,y,a);} + AH3 ALerpH3(AH3 x,AH3 y,AH3 a){return mix(x,y,a);} + AH4 ALerpH4(AH4 x,AH4 y,AH4 a){return mix(x,y,a);} +//------------------------------------------------------------------------------------------------------------------------------ + // No packed version of max3. + AH1 AMax3H1(AH1 x,AH1 y,AH1 z){return max(x,max(y,z));} + AH2 AMax3H2(AH2 x,AH2 y,AH2 z){return max(x,max(y,z));} + AH3 AMax3H3(AH3 x,AH3 y,AH3 z){return max(x,max(y,z));} + AH4 AMax3H4(AH4 x,AH4 y,AH4 z){return max(x,max(y,z));} +//------------------------------------------------------------------------------------------------------------------------------ + AW1 AMaxSW1(AW1 a,AW1 b){return AW1(max(ASU1(a),ASU1(b)));} + AW2 AMaxSW2(AW2 a,AW2 b){return AW2(max(ASU2(a),ASU2(b)));} + AW3 AMaxSW3(AW3 a,AW3 b){return AW3(max(ASU3(a),ASU3(b)));} + AW4 AMaxSW4(AW4 a,AW4 b){return AW4(max(ASU4(a),ASU4(b)));} +//------------------------------------------------------------------------------------------------------------------------------ + // No packed version of min3. + AH1 AMin3H1(AH1 x,AH1 y,AH1 z){return min(x,min(y,z));} + AH2 AMin3H2(AH2 x,AH2 y,AH2 z){return min(x,min(y,z));} + AH3 AMin3H3(AH3 x,AH3 y,AH3 z){return min(x,min(y,z));} + AH4 AMin3H4(AH4 x,AH4 y,AH4 z){return min(x,min(y,z));} +//------------------------------------------------------------------------------------------------------------------------------ + AW1 AMinSW1(AW1 a,AW1 b){return AW1(min(ASU1(a),ASU1(b)));} + AW2 AMinSW2(AW2 a,AW2 b){return AW2(min(ASU2(a),ASU2(b)));} + AW3 AMinSW3(AW3 a,AW3 b){return AW3(min(ASU3(a),ASU3(b)));} + AW4 AMinSW4(AW4 a,AW4 b){return AW4(min(ASU4(a),ASU4(b)));} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 ARcpH1(AH1 x){return AH1_(1.0)/x;} + AH2 ARcpH2(AH2 x){return AH2_(1.0)/x;} + AH3 ARcpH3(AH3 x){return AH3_(1.0)/x;} + AH4 ARcpH4(AH4 x){return AH4_(1.0)/x;} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 ARsqH1(AH1 x){return AH1_(1.0)/sqrt(x);} + AH2 ARsqH2(AH2 x){return AH2_(1.0)/sqrt(x);} + AH3 ARsqH3(AH3 x){return AH3_(1.0)/sqrt(x);} + AH4 ARsqH4(AH4 x){return AH4_(1.0)/sqrt(x);} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 ASatH1(AH1 x){return clamp(x,AH1_(0.0),AH1_(1.0));} + AH2 ASatH2(AH2 x){return clamp(x,AH2_(0.0),AH2_(1.0));} + AH3 ASatH3(AH3 x){return clamp(x,AH3_(0.0),AH3_(1.0));} + AH4 ASatH4(AH4 x){return clamp(x,AH4_(0.0),AH4_(1.0));} +//------------------------------------------------------------------------------------------------------------------------------ + AW1 AShrSW1(AW1 a,AW1 b){return AW1(ASW1(a)>>ASW1(b));} + AW2 AShrSW2(AW2 a,AW2 b){return AW2(ASW2(a)>>ASW2(b));} + AW3 AShrSW3(AW3 a,AW3 b){return AW3(ASW3(a)>>ASW3(b));} + AW4 AShrSW4(AW4 a,AW4 b){return AW4(ASW4(a)>>ASW4(b));} + #endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// GLSL DOUBLE +//============================================================================================================================== + #ifdef A_DUBL + #define AD1 double + #define AD2 dvec2 + #define AD3 dvec3 + #define AD4 dvec4 +//------------------------------------------------------------------------------------------------------------------------------ + AD1 AD1_x(AD1 a){return AD1(a);} + AD2 AD2_x(AD1 a){return AD2(a,a);} + AD3 AD3_x(AD1 a){return AD3(a,a,a);} + AD4 AD4_x(AD1 a){return AD4(a,a,a,a);} + #define AD1_(a) AD1_x(AD1(a)) + #define AD2_(a) AD2_x(AD1(a)) + #define AD3_(a) AD3_x(AD1(a)) + #define AD4_(a) AD4_x(AD1(a)) +//============================================================================================================================== + AD1 AFractD1(AD1 x){return fract(x);} + AD2 AFractD2(AD2 x){return fract(x);} + AD3 AFractD3(AD3 x){return fract(x);} + AD4 AFractD4(AD4 x){return fract(x);} +//------------------------------------------------------------------------------------------------------------------------------ + AD1 ALerpD1(AD1 x,AD1 y,AD1 a){return mix(x,y,a);} + AD2 ALerpD2(AD2 x,AD2 y,AD2 a){return mix(x,y,a);} + AD3 ALerpD3(AD3 x,AD3 y,AD3 a){return mix(x,y,a);} + AD4 ALerpD4(AD4 x,AD4 y,AD4 a){return mix(x,y,a);} +//------------------------------------------------------------------------------------------------------------------------------ + AD1 ARcpD1(AD1 x){return AD1_(1.0)/x;} + AD2 ARcpD2(AD2 x){return AD2_(1.0)/x;} + AD3 ARcpD3(AD3 x){return AD3_(1.0)/x;} + AD4 ARcpD4(AD4 x){return AD4_(1.0)/x;} +//------------------------------------------------------------------------------------------------------------------------------ + AD1 ARsqD1(AD1 x){return AD1_(1.0)/sqrt(x);} + AD2 ARsqD2(AD2 x){return AD2_(1.0)/sqrt(x);} + AD3 ARsqD3(AD3 x){return AD3_(1.0)/sqrt(x);} + AD4 ARsqD4(AD4 x){return AD4_(1.0)/sqrt(x);} +//------------------------------------------------------------------------------------------------------------------------------ + AD1 ASatD1(AD1 x){return clamp(x,AD1_(0.0),AD1_(1.0));} + AD2 ASatD2(AD2 x){return clamp(x,AD2_(0.0),AD2_(1.0));} + AD3 ASatD3(AD3 x){return clamp(x,AD3_(0.0),AD3_(1.0));} + AD4 ASatD4(AD4 x){return clamp(x,AD4_(0.0),AD4_(1.0));} + #endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// GLSL LONG +//============================================================================================================================== + #ifdef A_LONG + #define AL1 uint64_t + #define AL2 u64vec2 + #define AL3 u64vec3 + #define AL4 u64vec4 +//------------------------------------------------------------------------------------------------------------------------------ + #define ASL1 int64_t + #define ASL2 i64vec2 + #define ASL3 i64vec3 + #define ASL4 i64vec4 +//------------------------------------------------------------------------------------------------------------------------------ + #define AL1_AU2(x) packUint2x32(AU2(x)) + #define AU2_AL1(x) unpackUint2x32(AL1(x)) +//------------------------------------------------------------------------------------------------------------------------------ + AL1 AL1_x(AL1 a){return AL1(a);} + AL2 AL2_x(AL1 a){return AL2(a,a);} + AL3 AL3_x(AL1 a){return AL3(a,a,a);} + AL4 AL4_x(AL1 a){return AL4(a,a,a,a);} + #define AL1_(a) AL1_x(AL1(a)) + #define AL2_(a) AL2_x(AL1(a)) + #define AL3_(a) AL3_x(AL1(a)) + #define AL4_(a) AL4_x(AL1(a)) +//============================================================================================================================== + AL1 AAbsSL1(AL1 a){return AL1(abs(ASL1(a)));} + AL2 AAbsSL2(AL2 a){return AL2(abs(ASL2(a)));} + AL3 AAbsSL3(AL3 a){return AL3(abs(ASL3(a)));} + AL4 AAbsSL4(AL4 a){return AL4(abs(ASL4(a)));} +//------------------------------------------------------------------------------------------------------------------------------ + AL1 AMaxSL1(AL1 a,AL1 b){return AL1(max(ASU1(a),ASU1(b)));} + AL2 AMaxSL2(AL2 a,AL2 b){return AL2(max(ASU2(a),ASU2(b)));} + AL3 AMaxSL3(AL3 a,AL3 b){return AL3(max(ASU3(a),ASU3(b)));} + AL4 AMaxSL4(AL4 a,AL4 b){return AL4(max(ASU4(a),ASU4(b)));} +//------------------------------------------------------------------------------------------------------------------------------ + AL1 AMinSL1(AL1 a,AL1 b){return AL1(min(ASU1(a),ASU1(b)));} + AL2 AMinSL2(AL2 a,AL2 b){return AL2(min(ASU2(a),ASU2(b)));} + AL3 AMinSL3(AL3 a,AL3 b){return AL3(min(ASU3(a),ASU3(b)));} + AL4 AMinSL4(AL4 a,AL4 b){return AL4(min(ASU4(a),ASU4(b)));} + #endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// WAVE OPERATIONS +//============================================================================================================================== + #ifdef A_WAVE + // Where 'x' must be a compile time literal. + AF1 AWaveXorF1(AF1 v,AU1 x){return subgroupShuffleXor(v,x);} + AF2 AWaveXorF2(AF2 v,AU1 x){return subgroupShuffleXor(v,x);} + AF3 AWaveXorF3(AF3 v,AU1 x){return subgroupShuffleXor(v,x);} + AF4 AWaveXorF4(AF4 v,AU1 x){return subgroupShuffleXor(v,x);} + AU1 AWaveXorU1(AU1 v,AU1 x){return subgroupShuffleXor(v,x);} + AU2 AWaveXorU2(AU2 v,AU1 x){return subgroupShuffleXor(v,x);} + AU3 AWaveXorU3(AU3 v,AU1 x){return subgroupShuffleXor(v,x);} + AU4 AWaveXorU4(AU4 v,AU1 x){return subgroupShuffleXor(v,x);} +//------------------------------------------------------------------------------------------------------------------------------ + #ifdef A_HALF + AH2 AWaveXorH2(AH2 v,AU1 x){return AH2_AU1(subgroupShuffleXor(AU1_AH2(v),x));} + AH4 AWaveXorH4(AH4 v,AU1 x){return AH4_AU2(subgroupShuffleXor(AU2_AH4(v),x));} + AW2 AWaveXorW2(AW2 v,AU1 x){return AW2_AU1(subgroupShuffleXor(AU1_AW2(v),x));} + AW4 AWaveXorW4(AW4 v,AU1 x){return AW4_AU2(subgroupShuffleXor(AU2_AW4(v),x));} + #endif + #endif +//============================================================================================================================== +#endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// +// HLSL +// +// +//============================================================================================================================== +#if defined(A_HLSL) && defined(A_GPU) + #ifdef A_HLSL_6_2 + #define AP1 bool + #define AP2 bool2 + #define AP3 bool3 + #define AP4 bool4 +//------------------------------------------------------------------------------------------------------------------------------ + #define AF1 float32_t + #define AF2 float32_t2 + #define AF3 float32_t3 + #define AF4 float32_t4 +//------------------------------------------------------------------------------------------------------------------------------ + #define AU1 uint32_t + #define AU2 uint32_t2 + #define AU3 uint32_t3 + #define AU4 uint32_t4 +//------------------------------------------------------------------------------------------------------------------------------ + #define ASU1 int32_t + #define ASU2 int32_t2 + #define ASU3 int32_t3 + #define ASU4 int32_t4 + #else + #define AP1 bool + #define AP2 bool2 + #define AP3 bool3 + #define AP4 bool4 +//------------------------------------------------------------------------------------------------------------------------------ + #define AF1 float + #define AF2 float2 + #define AF3 float3 + #define AF4 float4 +//------------------------------------------------------------------------------------------------------------------------------ + #define AU1 uint + #define AU2 uint2 + #define AU3 uint3 + #define AU4 uint4 +//------------------------------------------------------------------------------------------------------------------------------ + #define ASU1 int + #define ASU2 int2 + #define ASU3 int3 + #define ASU4 int4 + #endif +//============================================================================================================================== + #define AF1_AU1(x) asfloat(AU1(x)) + #define AF2_AU2(x) asfloat(AU2(x)) + #define AF3_AU3(x) asfloat(AU3(x)) + #define AF4_AU4(x) asfloat(AU4(x)) +//------------------------------------------------------------------------------------------------------------------------------ + #define AU1_AF1(x) asuint(AF1(x)) + #define AU2_AF2(x) asuint(AF2(x)) + #define AU3_AF3(x) asuint(AF3(x)) + #define AU4_AF4(x) asuint(AF4(x)) +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AU1_AH1_AF1_x(AF1 a){return f32tof16(a);} + #define AU1_AH1_AF1(a) AU1_AH1_AF1_x(AF1(a)) +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AU1_AH2_AF2_x(AF2 a){return f32tof16(a.x)|(f32tof16(a.y)<<16);} + #define AU1_AH2_AF2(a) AU1_AH2_AF2_x(AF2(a)) + #define AU1_AB4Unorm_AF4(x) D3DCOLORtoUBYTE4(AF4(x)) +//------------------------------------------------------------------------------------------------------------------------------ + AF2 AF2_AH2_AU1_x(AU1 x){return AF2(f16tof32(x&0xFFFF),f16tof32(x>>16));} + #define AF2_AH2_AU1(x) AF2_AH2_AU1_x(AU1(x)) +//============================================================================================================================== + AF1 AF1_x(AF1 a){return AF1(a);} + AF2 AF2_x(AF1 a){return AF2(a,a);} + AF3 AF3_x(AF1 a){return AF3(a,a,a);} + AF4 AF4_x(AF1 a){return AF4(a,a,a,a);} + #define AF1_(a) AF1_x(AF1(a)) + #define AF2_(a) AF2_x(AF1(a)) + #define AF3_(a) AF3_x(AF1(a)) + #define AF4_(a) AF4_x(AF1(a)) +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AU1_x(AU1 a){return AU1(a);} + AU2 AU2_x(AU1 a){return AU2(a,a);} + AU3 AU3_x(AU1 a){return AU3(a,a,a);} + AU4 AU4_x(AU1 a){return AU4(a,a,a,a);} + #define AU1_(a) AU1_x(AU1(a)) + #define AU2_(a) AU2_x(AU1(a)) + #define AU3_(a) AU3_x(AU1(a)) + #define AU4_(a) AU4_x(AU1(a)) +//============================================================================================================================== + AU1 AAbsSU1(AU1 a){return AU1(abs(ASU1(a)));} + AU2 AAbsSU2(AU2 a){return AU2(abs(ASU2(a)));} + AU3 AAbsSU3(AU3 a){return AU3(abs(ASU3(a)));} + AU4 AAbsSU4(AU4 a){return AU4(abs(ASU4(a)));} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 ABfe(AU1 src,AU1 off,AU1 bits){AU1 mask=(1u<<bits)-1;return (src>>off)&mask;} + AU1 ABfi(AU1 src,AU1 ins,AU1 mask){return (ins&mask)|(src&(~mask));} + AU1 ABfiM(AU1 src,AU1 ins,AU1 bits){AU1 mask=(1u<<bits)-1;return (ins&mask)|(src&(~mask));} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AClampF1(AF1 x,AF1 n,AF1 m){return max(n,min(x,m));} + AF2 AClampF2(AF2 x,AF2 n,AF2 m){return max(n,min(x,m));} + AF3 AClampF3(AF3 x,AF3 n,AF3 m){return max(n,min(x,m));} + AF4 AClampF4(AF4 x,AF4 n,AF4 m){return max(n,min(x,m));} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AFractF1(AF1 x){return x-floor(x);} + AF2 AFractF2(AF2 x){return x-floor(x);} + AF3 AFractF3(AF3 x){return x-floor(x);} + AF4 AFractF4(AF4 x){return x-floor(x);} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 ALerpF1(AF1 x,AF1 y,AF1 a){return lerp(x,y,a);} + AF2 ALerpF2(AF2 x,AF2 y,AF2 a){return lerp(x,y,a);} + AF3 ALerpF3(AF3 x,AF3 y,AF3 a){return lerp(x,y,a);} + AF4 ALerpF4(AF4 x,AF4 y,AF4 a){return lerp(x,y,a);} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AMax3F1(AF1 x,AF1 y,AF1 z){return max(x,max(y,z));} + AF2 AMax3F2(AF2 x,AF2 y,AF2 z){return max(x,max(y,z));} + AF3 AMax3F3(AF3 x,AF3 y,AF3 z){return max(x,max(y,z));} + AF4 AMax3F4(AF4 x,AF4 y,AF4 z){return max(x,max(y,z));} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AMax3SU1(AU1 x,AU1 y,AU1 z){return AU1(max(ASU1(x),max(ASU1(y),ASU1(z))));} + AU2 AMax3SU2(AU2 x,AU2 y,AU2 z){return AU2(max(ASU2(x),max(ASU2(y),ASU2(z))));} + AU3 AMax3SU3(AU3 x,AU3 y,AU3 z){return AU3(max(ASU3(x),max(ASU3(y),ASU3(z))));} + AU4 AMax3SU4(AU4 x,AU4 y,AU4 z){return AU4(max(ASU4(x),max(ASU4(y),ASU4(z))));} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AMax3U1(AU1 x,AU1 y,AU1 z){return max(x,max(y,z));} + AU2 AMax3U2(AU2 x,AU2 y,AU2 z){return max(x,max(y,z));} + AU3 AMax3U3(AU3 x,AU3 y,AU3 z){return max(x,max(y,z));} + AU4 AMax3U4(AU4 x,AU4 y,AU4 z){return max(x,max(y,z));} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AMaxSU1(AU1 a,AU1 b){return AU1(max(ASU1(a),ASU1(b)));} + AU2 AMaxSU2(AU2 a,AU2 b){return AU2(max(ASU2(a),ASU2(b)));} + AU3 AMaxSU3(AU3 a,AU3 b){return AU3(max(ASU3(a),ASU3(b)));} + AU4 AMaxSU4(AU4 a,AU4 b){return AU4(max(ASU4(a),ASU4(b)));} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AMed3F1(AF1 x,AF1 y,AF1 z){return max(min(x,y),min(max(x,y),z));} + AF2 AMed3F2(AF2 x,AF2 y,AF2 z){return max(min(x,y),min(max(x,y),z));} + AF3 AMed3F3(AF3 x,AF3 y,AF3 z){return max(min(x,y),min(max(x,y),z));} + AF4 AMed3F4(AF4 x,AF4 y,AF4 z){return max(min(x,y),min(max(x,y),z));} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AMin3F1(AF1 x,AF1 y,AF1 z){return min(x,min(y,z));} + AF2 AMin3F2(AF2 x,AF2 y,AF2 z){return min(x,min(y,z));} + AF3 AMin3F3(AF3 x,AF3 y,AF3 z){return min(x,min(y,z));} + AF4 AMin3F4(AF4 x,AF4 y,AF4 z){return min(x,min(y,z));} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AMin3SU1(AU1 x,AU1 y,AU1 z){return AU1(min(ASU1(x),min(ASU1(y),ASU1(z))));} + AU2 AMin3SU2(AU2 x,AU2 y,AU2 z){return AU2(min(ASU2(x),min(ASU2(y),ASU2(z))));} + AU3 AMin3SU3(AU3 x,AU3 y,AU3 z){return AU3(min(ASU3(x),min(ASU3(y),ASU3(z))));} + AU4 AMin3SU4(AU4 x,AU4 y,AU4 z){return AU4(min(ASU4(x),min(ASU4(y),ASU4(z))));} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AMin3U1(AU1 x,AU1 y,AU1 z){return min(x,min(y,z));} + AU2 AMin3U2(AU2 x,AU2 y,AU2 z){return min(x,min(y,z));} + AU3 AMin3U3(AU3 x,AU3 y,AU3 z){return min(x,min(y,z));} + AU4 AMin3U4(AU4 x,AU4 y,AU4 z){return min(x,min(y,z));} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AMinSU1(AU1 a,AU1 b){return AU1(min(ASU1(a),ASU1(b)));} + AU2 AMinSU2(AU2 a,AU2 b){return AU2(min(ASU2(a),ASU2(b)));} + AU3 AMinSU3(AU3 a,AU3 b){return AU3(min(ASU3(a),ASU3(b)));} + AU4 AMinSU4(AU4 a,AU4 b){return AU4(min(ASU4(a),ASU4(b)));} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 ANCosF1(AF1 x){return cos(x*AF1_(A_2PI));} + AF2 ANCosF2(AF2 x){return cos(x*AF2_(A_2PI));} + AF3 ANCosF3(AF3 x){return cos(x*AF3_(A_2PI));} + AF4 ANCosF4(AF4 x){return cos(x*AF4_(A_2PI));} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 ANSinF1(AF1 x){return sin(x*AF1_(A_2PI));} + AF2 ANSinF2(AF2 x){return sin(x*AF2_(A_2PI));} + AF3 ANSinF3(AF3 x){return sin(x*AF3_(A_2PI));} + AF4 ANSinF4(AF4 x){return sin(x*AF4_(A_2PI));} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 ARcpF1(AF1 x){return rcp(x);} + AF2 ARcpF2(AF2 x){return rcp(x);} + AF3 ARcpF3(AF3 x){return rcp(x);} + AF4 ARcpF4(AF4 x){return rcp(x);} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 ARsqF1(AF1 x){return rsqrt(x);} + AF2 ARsqF2(AF2 x){return rsqrt(x);} + AF3 ARsqF3(AF3 x){return rsqrt(x);} + AF4 ARsqF4(AF4 x){return rsqrt(x);} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 ASatF1(AF1 x){return saturate(x);} + AF2 ASatF2(AF2 x){return saturate(x);} + AF3 ASatF3(AF3 x){return saturate(x);} + AF4 ASatF4(AF4 x){return saturate(x);} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AShrSU1(AU1 a,AU1 b){return AU1(ASU1(a)>>ASU1(b));} + AU2 AShrSU2(AU2 a,AU2 b){return AU2(ASU2(a)>>ASU2(b));} + AU3 AShrSU3(AU3 a,AU3 b){return AU3(ASU3(a)>>ASU3(b));} + AU4 AShrSU4(AU4 a,AU4 b){return AU4(ASU4(a)>>ASU4(b));} +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// HLSL BYTE +//============================================================================================================================== + #ifdef A_BYTE + #endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// HLSL HALF +//============================================================================================================================== + #ifdef A_HALF + #ifdef A_HLSL_6_2 + #define AH1 float16_t + #define AH2 float16_t2 + #define AH3 float16_t3 + #define AH4 float16_t4 +//------------------------------------------------------------------------------------------------------------------------------ + #define AW1 uint16_t + #define AW2 uint16_t2 + #define AW3 uint16_t3 + #define AW4 uint16_t4 +//------------------------------------------------------------------------------------------------------------------------------ + #define ASW1 int16_t + #define ASW2 int16_t2 + #define ASW3 int16_t3 + #define ASW4 int16_t4 + #else + #define AH1 min16float + #define AH2 min16float2 + #define AH3 min16float3 + #define AH4 min16float4 +//------------------------------------------------------------------------------------------------------------------------------ + #define AW1 min16uint + #define AW2 min16uint2 + #define AW3 min16uint3 + #define AW4 min16uint4 +//------------------------------------------------------------------------------------------------------------------------------ + #define ASW1 min16int + #define ASW2 min16int2 + #define ASW3 min16int3 + #define ASW4 min16int4 + #endif +//============================================================================================================================== + // Need to use manual unpack to get optimal execution (don't use packed types in buffers directly). + // Unpack requires this pattern: https://gpuopen.com/first-steps-implementing-fp16/ + AH2 AH2_AU1_x(AU1 x){AF2 t=f16tof32(AU2(x&0xFFFF,x>>16));return AH2(t);} + AH4 AH4_AU2_x(AU2 x){return AH4(AH2_AU1_x(x.x),AH2_AU1_x(x.y));} + AW2 AW2_AU1_x(AU1 x){AU2 t=AU2(x&0xFFFF,x>>16);return AW2(t);} + AW4 AW4_AU2_x(AU2 x){return AW4(AW2_AU1_x(x.x),AW2_AU1_x(x.y));} + #define AH2_AU1(x) AH2_AU1_x(AU1(x)) + #define AH4_AU2(x) AH4_AU2_x(AU2(x)) + #define AW2_AU1(x) AW2_AU1_x(AU1(x)) + #define AW4_AU2(x) AW4_AU2_x(AU2(x)) +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AU1_AH2_x(AH2 x){return f32tof16(x.x)+(f32tof16(x.y)<<16);} + AU2 AU2_AH4_x(AH4 x){return AU2(AU1_AH2_x(x.xy),AU1_AH2_x(x.zw));} + AU1 AU1_AW2_x(AW2 x){return AU1(x.x)+(AU1(x.y)<<16);} + AU2 AU2_AW4_x(AW4 x){return AU2(AU1_AW2_x(x.xy),AU1_AW2_x(x.zw));} + #define AU1_AH2(x) AU1_AH2_x(AH2(x)) + #define AU2_AH4(x) AU2_AH4_x(AH4(x)) + #define AU1_AW2(x) AU1_AW2_x(AW2(x)) + #define AU2_AW4(x) AU2_AW4_x(AW4(x)) +//============================================================================================================================== + #if defined(A_HLSL_6_2) && !defined(A_NO_16_BIT_CAST) + #define AW1_AH1(x) asuint16(x) + #define AW2_AH2(x) asuint16(x) + #define AW3_AH3(x) asuint16(x) + #define AW4_AH4(x) asuint16(x) + #else + #define AW1_AH1(a) AW1(f32tof16(AF1(a))) + #define AW2_AH2(a) AW2(AW1_AH1((a).x),AW1_AH1((a).y)) + #define AW3_AH3(a) AW3(AW1_AH1((a).x),AW1_AH1((a).y),AW1_AH1((a).z)) + #define AW4_AH4(a) AW4(AW1_AH1((a).x),AW1_AH1((a).y),AW1_AH1((a).z),AW1_AH1((a).w)) + #endif +//------------------------------------------------------------------------------------------------------------------------------ + #if defined(A_HLSL_6_2) && !defined(A_NO_16_BIT_CAST) + #define AH1_AW1(x) asfloat16(x) + #define AH2_AW2(x) asfloat16(x) + #define AH3_AW3(x) asfloat16(x) + #define AH4_AW4(x) asfloat16(x) + #else + #define AH1_AW1(a) AH1(f16tof32(AU1(a))) + #define AH2_AW2(a) AH2(AH1_AW1((a).x),AH1_AW1((a).y)) + #define AH3_AW3(a) AH3(AH1_AW1((a).x),AH1_AW1((a).y),AH1_AW1((a).z)) + #define AH4_AW4(a) AH4(AH1_AW1((a).x),AH1_AW1((a).y),AH1_AW1((a).z),AH1_AW1((a).w)) + #endif +//============================================================================================================================== + AH1 AH1_x(AH1 a){return AH1(a);} + AH2 AH2_x(AH1 a){return AH2(a,a);} + AH3 AH3_x(AH1 a){return AH3(a,a,a);} + AH4 AH4_x(AH1 a){return AH4(a,a,a,a);} + #define AH1_(a) AH1_x(AH1(a)) + #define AH2_(a) AH2_x(AH1(a)) + #define AH3_(a) AH3_x(AH1(a)) + #define AH4_(a) AH4_x(AH1(a)) +//------------------------------------------------------------------------------------------------------------------------------ + AW1 AW1_x(AW1 a){return AW1(a);} + AW2 AW2_x(AW1 a){return AW2(a,a);} + AW3 AW3_x(AW1 a){return AW3(a,a,a);} + AW4 AW4_x(AW1 a){return AW4(a,a,a,a);} + #define AW1_(a) AW1_x(AW1(a)) + #define AW2_(a) AW2_x(AW1(a)) + #define AW3_(a) AW3_x(AW1(a)) + #define AW4_(a) AW4_x(AW1(a)) +//============================================================================================================================== + AW1 AAbsSW1(AW1 a){return AW1(abs(ASW1(a)));} + AW2 AAbsSW2(AW2 a){return AW2(abs(ASW2(a)));} + AW3 AAbsSW3(AW3 a){return AW3(abs(ASW3(a)));} + AW4 AAbsSW4(AW4 a){return AW4(abs(ASW4(a)));} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AClampH1(AH1 x,AH1 n,AH1 m){return max(n,min(x,m));} + AH2 AClampH2(AH2 x,AH2 n,AH2 m){return max(n,min(x,m));} + AH3 AClampH3(AH3 x,AH3 n,AH3 m){return max(n,min(x,m));} + AH4 AClampH4(AH4 x,AH4 n,AH4 m){return max(n,min(x,m));} +//------------------------------------------------------------------------------------------------------------------------------ + // V_FRACT_F16 (note DX frac() is different). + AH1 AFractH1(AH1 x){return x-floor(x);} + AH2 AFractH2(AH2 x){return x-floor(x);} + AH3 AFractH3(AH3 x){return x-floor(x);} + AH4 AFractH4(AH4 x){return x-floor(x);} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 ALerpH1(AH1 x,AH1 y,AH1 a){return lerp(x,y,a);} + AH2 ALerpH2(AH2 x,AH2 y,AH2 a){return lerp(x,y,a);} + AH3 ALerpH3(AH3 x,AH3 y,AH3 a){return lerp(x,y,a);} + AH4 ALerpH4(AH4 x,AH4 y,AH4 a){return lerp(x,y,a);} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AMax3H1(AH1 x,AH1 y,AH1 z){return max(x,max(y,z));} + AH2 AMax3H2(AH2 x,AH2 y,AH2 z){return max(x,max(y,z));} + AH3 AMax3H3(AH3 x,AH3 y,AH3 z){return max(x,max(y,z));} + AH4 AMax3H4(AH4 x,AH4 y,AH4 z){return max(x,max(y,z));} +//------------------------------------------------------------------------------------------------------------------------------ + AW1 AMaxSW1(AW1 a,AW1 b){return AW1(max(ASU1(a),ASU1(b)));} + AW2 AMaxSW2(AW2 a,AW2 b){return AW2(max(ASU2(a),ASU2(b)));} + AW3 AMaxSW3(AW3 a,AW3 b){return AW3(max(ASU3(a),ASU3(b)));} + AW4 AMaxSW4(AW4 a,AW4 b){return AW4(max(ASU4(a),ASU4(b)));} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AMin3H1(AH1 x,AH1 y,AH1 z){return min(x,min(y,z));} + AH2 AMin3H2(AH2 x,AH2 y,AH2 z){return min(x,min(y,z));} + AH3 AMin3H3(AH3 x,AH3 y,AH3 z){return min(x,min(y,z));} + AH4 AMin3H4(AH4 x,AH4 y,AH4 z){return min(x,min(y,z));} +//------------------------------------------------------------------------------------------------------------------------------ + AW1 AMinSW1(AW1 a,AW1 b){return AW1(min(ASU1(a),ASU1(b)));} + AW2 AMinSW2(AW2 a,AW2 b){return AW2(min(ASU2(a),ASU2(b)));} + AW3 AMinSW3(AW3 a,AW3 b){return AW3(min(ASU3(a),ASU3(b)));} + AW4 AMinSW4(AW4 a,AW4 b){return AW4(min(ASU4(a),ASU4(b)));} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 ARcpH1(AH1 x){return rcp(x);} + AH2 ARcpH2(AH2 x){return rcp(x);} + AH3 ARcpH3(AH3 x){return rcp(x);} + AH4 ARcpH4(AH4 x){return rcp(x);} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 ARsqH1(AH1 x){return rsqrt(x);} + AH2 ARsqH2(AH2 x){return rsqrt(x);} + AH3 ARsqH3(AH3 x){return rsqrt(x);} + AH4 ARsqH4(AH4 x){return rsqrt(x);} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 ASatH1(AH1 x){return saturate(x);} + AH2 ASatH2(AH2 x){return saturate(x);} + AH3 ASatH3(AH3 x){return saturate(x);} + AH4 ASatH4(AH4 x){return saturate(x);} +//------------------------------------------------------------------------------------------------------------------------------ + AW1 AShrSW1(AW1 a,AW1 b){return AW1(ASW1(a)>>ASW1(b));} + AW2 AShrSW2(AW2 a,AW2 b){return AW2(ASW2(a)>>ASW2(b));} + AW3 AShrSW3(AW3 a,AW3 b){return AW3(ASW3(a)>>ASW3(b));} + AW4 AShrSW4(AW4 a,AW4 b){return AW4(ASW4(a)>>ASW4(b));} + #endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// HLSL DOUBLE +//============================================================================================================================== + #ifdef A_DUBL + #ifdef A_HLSL_6_2 + #define AD1 float64_t + #define AD2 float64_t2 + #define AD3 float64_t3 + #define AD4 float64_t4 + #else + #define AD1 double + #define AD2 double2 + #define AD3 double3 + #define AD4 double4 + #endif +//------------------------------------------------------------------------------------------------------------------------------ + AD1 AD1_x(AD1 a){return AD1(a);} + AD2 AD2_x(AD1 a){return AD2(a,a);} + AD3 AD3_x(AD1 a){return AD3(a,a,a);} + AD4 AD4_x(AD1 a){return AD4(a,a,a,a);} + #define AD1_(a) AD1_x(AD1(a)) + #define AD2_(a) AD2_x(AD1(a)) + #define AD3_(a) AD3_x(AD1(a)) + #define AD4_(a) AD4_x(AD1(a)) +//============================================================================================================================== + AD1 AFractD1(AD1 a){return a-floor(a);} + AD2 AFractD2(AD2 a){return a-floor(a);} + AD3 AFractD3(AD3 a){return a-floor(a);} + AD4 AFractD4(AD4 a){return a-floor(a);} +//------------------------------------------------------------------------------------------------------------------------------ + AD1 ALerpD1(AD1 x,AD1 y,AD1 a){return lerp(x,y,a);} + AD2 ALerpD2(AD2 x,AD2 y,AD2 a){return lerp(x,y,a);} + AD3 ALerpD3(AD3 x,AD3 y,AD3 a){return lerp(x,y,a);} + AD4 ALerpD4(AD4 x,AD4 y,AD4 a){return lerp(x,y,a);} +//------------------------------------------------------------------------------------------------------------------------------ + AD1 ARcpD1(AD1 x){return rcp(x);} + AD2 ARcpD2(AD2 x){return rcp(x);} + AD3 ARcpD3(AD3 x){return rcp(x);} + AD4 ARcpD4(AD4 x){return rcp(x);} +//------------------------------------------------------------------------------------------------------------------------------ + AD1 ARsqD1(AD1 x){return rsqrt(x);} + AD2 ARsqD2(AD2 x){return rsqrt(x);} + AD3 ARsqD3(AD3 x){return rsqrt(x);} + AD4 ARsqD4(AD4 x){return rsqrt(x);} +//------------------------------------------------------------------------------------------------------------------------------ + AD1 ASatD1(AD1 x){return saturate(x);} + AD2 ASatD2(AD2 x){return saturate(x);} + AD3 ASatD3(AD3 x){return saturate(x);} + AD4 ASatD4(AD4 x){return saturate(x);} + #endif +//============================================================================================================================== +// HLSL WAVE +//============================================================================================================================== + #ifdef A_WAVE + // Where 'x' must be a compile time literal. + AF1 AWaveXorF1(AF1 v,AU1 x){return WaveReadLaneAt(v,WaveGetLaneIndex()^x);} + AF2 AWaveXorF2(AF2 v,AU1 x){return WaveReadLaneAt(v,WaveGetLaneIndex()^x);} + AF3 AWaveXorF3(AF3 v,AU1 x){return WaveReadLaneAt(v,WaveGetLaneIndex()^x);} + AF4 AWaveXorF4(AF4 v,AU1 x){return WaveReadLaneAt(v,WaveGetLaneIndex()^x);} + AU1 AWaveXorU1(AU1 v,AU1 x){return WaveReadLaneAt(v,WaveGetLaneIndex()^x);} + AU2 AWaveXorU1(AU2 v,AU1 x){return WaveReadLaneAt(v,WaveGetLaneIndex()^x);} + AU3 AWaveXorU1(AU3 v,AU1 x){return WaveReadLaneAt(v,WaveGetLaneIndex()^x);} + AU4 AWaveXorU1(AU4 v,AU1 x){return WaveReadLaneAt(v,WaveGetLaneIndex()^x);} +//------------------------------------------------------------------------------------------------------------------------------ + #ifdef A_HALF + AH2 AWaveXorH2(AH2 v,AU1 x){return AH2_AU1(WaveReadLaneAt(AU1_AH2(v),WaveGetLaneIndex()^x));} + AH4 AWaveXorH4(AH4 v,AU1 x){return AH4_AU2(WaveReadLaneAt(AU2_AH4(v),WaveGetLaneIndex()^x));} + AW2 AWaveXorW2(AW2 v,AU1 x){return AW2_AU1(WaveReadLaneAt(AU1_AW2(v),WaveGetLaneIndex()^x));} + AW4 AWaveXorW4(AW4 v,AU1 x){return AW4_AU1(WaveReadLaneAt(AU1_AW4(v),WaveGetLaneIndex()^x));} + #endif + #endif +//============================================================================================================================== +#endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// +// GPU COMMON +// +// +//============================================================================================================================== +#ifdef A_GPU + // Negative and positive infinity. + #define A_INFP_F AF1_AU1(0x7f800000u) + #define A_INFN_F AF1_AU1(0xff800000u) +//------------------------------------------------------------------------------------------------------------------------------ + // Copy sign from 's' to positive 'd'. + AF1 ACpySgnF1(AF1 d,AF1 s){return AF1_AU1(AU1_AF1(d)|(AU1_AF1(s)&AU1_(0x80000000u)));} + AF2 ACpySgnF2(AF2 d,AF2 s){return AF2_AU2(AU2_AF2(d)|(AU2_AF2(s)&AU2_(0x80000000u)));} + AF3 ACpySgnF3(AF3 d,AF3 s){return AF3_AU3(AU3_AF3(d)|(AU3_AF3(s)&AU3_(0x80000000u)));} + AF4 ACpySgnF4(AF4 d,AF4 s){return AF4_AU4(AU4_AF4(d)|(AU4_AF4(s)&AU4_(0x80000000u)));} +//------------------------------------------------------------------------------------------------------------------------------ + // Single operation to return (useful to create a mask to use in lerp for branch free logic), + // m=NaN := 0 + // m>=0 := 0 + // m<0 := 1 + // Uses the following useful floating point logic, + // saturate(+a*(-INF)==-INF) := 0 + // saturate( 0*(-INF)== NaN) := 0 + // saturate(-a*(-INF)==+INF) := 1 + AF1 ASignedF1(AF1 m){return ASatF1(m*AF1_(A_INFN_F));} + AF2 ASignedF2(AF2 m){return ASatF2(m*AF2_(A_INFN_F));} + AF3 ASignedF3(AF3 m){return ASatF3(m*AF3_(A_INFN_F));} + AF4 ASignedF4(AF4 m){return ASatF4(m*AF4_(A_INFN_F));} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AGtZeroF1(AF1 m){return ASatF1(m*AF1_(A_INFP_F));} + AF2 AGtZeroF2(AF2 m){return ASatF2(m*AF2_(A_INFP_F));} + AF3 AGtZeroF3(AF3 m){return ASatF3(m*AF3_(A_INFP_F));} + AF4 AGtZeroF4(AF4 m){return ASatF4(m*AF4_(A_INFP_F));} +//============================================================================================================================== + #ifdef A_HALF + #ifdef A_HLSL_6_2 + #define A_INFP_H AH1_AW1((uint16_t)0x7c00u) + #define A_INFN_H AH1_AW1((uint16_t)0xfc00u) + #else + #define A_INFP_H AH1_AW1(0x7c00u) + #define A_INFN_H AH1_AW1(0xfc00u) + #endif + +//------------------------------------------------------------------------------------------------------------------------------ + AH1 ACpySgnH1(AH1 d,AH1 s){return AH1_AW1(AW1_AH1(d)|(AW1_AH1(s)&AW1_(0x8000u)));} + AH2 ACpySgnH2(AH2 d,AH2 s){return AH2_AW2(AW2_AH2(d)|(AW2_AH2(s)&AW2_(0x8000u)));} + AH3 ACpySgnH3(AH3 d,AH3 s){return AH3_AW3(AW3_AH3(d)|(AW3_AH3(s)&AW3_(0x8000u)));} + AH4 ACpySgnH4(AH4 d,AH4 s){return AH4_AW4(AW4_AH4(d)|(AW4_AH4(s)&AW4_(0x8000u)));} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 ASignedH1(AH1 m){return ASatH1(m*AH1_(A_INFN_H));} + AH2 ASignedH2(AH2 m){return ASatH2(m*AH2_(A_INFN_H));} + AH3 ASignedH3(AH3 m){return ASatH3(m*AH3_(A_INFN_H));} + AH4 ASignedH4(AH4 m){return ASatH4(m*AH4_(A_INFN_H));} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AGtZeroH1(AH1 m){return ASatH1(m*AH1_(A_INFP_H));} + AH2 AGtZeroH2(AH2 m){return ASatH2(m*AH2_(A_INFP_H));} + AH3 AGtZeroH3(AH3 m){return ASatH3(m*AH3_(A_INFP_H));} + AH4 AGtZeroH4(AH4 m){return ASatH4(m*AH4_(A_INFP_H));} + #endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// [FIS] FLOAT INTEGER SORTABLE +//------------------------------------------------------------------------------------------------------------------------------ +// Float to integer sortable. +// - If sign bit=0, flip the sign bit (positives). +// - If sign bit=1, flip all bits (negatives). +// Integer sortable to float. +// - If sign bit=1, flip the sign bit (positives). +// - If sign bit=0, flip all bits (negatives). +// Has nice side effects. +// - Larger integers are more positive values. +// - Float zero is mapped to center of integers (so clear to integer zero is a nice default for atomic max usage). +// Burns 3 ops for conversion {shift,or,xor}. +//============================================================================================================================== + AU1 AFisToU1(AU1 x){return x^(( AShrSU1(x,AU1_(31)))|AU1_(0x80000000));} + AU1 AFisFromU1(AU1 x){return x^((~AShrSU1(x,AU1_(31)))|AU1_(0x80000000));} +//------------------------------------------------------------------------------------------------------------------------------ + // Just adjust high 16-bit value (useful when upper part of 32-bit word is a 16-bit float value). + AU1 AFisToHiU1(AU1 x){return x^(( AShrSU1(x,AU1_(15)))|AU1_(0x80000000));} + AU1 AFisFromHiU1(AU1 x){return x^((~AShrSU1(x,AU1_(15)))|AU1_(0x80000000));} +//------------------------------------------------------------------------------------------------------------------------------ + #ifdef A_HALF + AW1 AFisToW1(AW1 x){return x^(( AShrSW1(x,AW1_(15)))|AW1_(0x8000));} + AW1 AFisFromW1(AW1 x){return x^((~AShrSW1(x,AW1_(15)))|AW1_(0x8000));} +//------------------------------------------------------------------------------------------------------------------------------ + AW2 AFisToW2(AW2 x){return x^(( AShrSW2(x,AW2_(15)))|AW2_(0x8000));} + AW2 AFisFromW2(AW2 x){return x^((~AShrSW2(x,AW2_(15)))|AW2_(0x8000));} + #endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// [PERM] V_PERM_B32 +//------------------------------------------------------------------------------------------------------------------------------ +// Support for V_PERM_B32 started in the 3rd generation of GCN. +//------------------------------------------------------------------------------------------------------------------------------ +// yyyyxxxx - The 'i' input. +// 76543210 +// ======== +// HGFEDCBA - Naming on permutation. +//------------------------------------------------------------------------------------------------------------------------------ +// TODO +// ==== +// - Make sure compiler optimizes this. +//============================================================================================================================== + #ifdef A_HALF + AU1 APerm0E0A(AU2 i){return((i.x )&0xffu)|((i.y<<16)&0xff0000u);} + AU1 APerm0F0B(AU2 i){return((i.x>> 8)&0xffu)|((i.y<< 8)&0xff0000u);} + AU1 APerm0G0C(AU2 i){return((i.x>>16)&0xffu)|((i.y )&0xff0000u);} + AU1 APerm0H0D(AU2 i){return((i.x>>24)&0xffu)|((i.y>> 8)&0xff0000u);} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 APermHGFA(AU2 i){return((i.x )&0x000000ffu)|(i.y&0xffffff00u);} + AU1 APermHGFC(AU2 i){return((i.x>>16)&0x000000ffu)|(i.y&0xffffff00u);} + AU1 APermHGAE(AU2 i){return((i.x<< 8)&0x0000ff00u)|(i.y&0xffff00ffu);} + AU1 APermHGCE(AU2 i){return((i.x>> 8)&0x0000ff00u)|(i.y&0xffff00ffu);} + AU1 APermHAFE(AU2 i){return((i.x<<16)&0x00ff0000u)|(i.y&0xff00ffffu);} + AU1 APermHCFE(AU2 i){return((i.x )&0x00ff0000u)|(i.y&0xff00ffffu);} + AU1 APermAGFE(AU2 i){return((i.x<<24)&0xff000000u)|(i.y&0x00ffffffu);} + AU1 APermCGFE(AU2 i){return((i.x<< 8)&0xff000000u)|(i.y&0x00ffffffu);} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 APermGCEA(AU2 i){return((i.x)&0x00ff00ffu)|((i.y<<8)&0xff00ff00u);} + AU1 APermGECA(AU2 i){return(((i.x)&0xffu)|((i.x>>8)&0xff00u)|((i.y<<16)&0xff0000u)|((i.y<<8)&0xff000000u));} + #endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// [BUC] BYTE UNSIGNED CONVERSION +//------------------------------------------------------------------------------------------------------------------------------ +// Designed to use the optimal conversion, enables the scaling to possibly be factored into other computation. +// Works on a range of {0 to A_BUC_<32,16>}, for <32-bit, and 16-bit> respectively. +//------------------------------------------------------------------------------------------------------------------------------ +// OPCODE NOTES +// ============ +// GCN does not do UNORM or SNORM for bytes in opcodes. +// - V_CVT_F32_UBYTE{0,1,2,3} - Unsigned byte to float. +// - V_CVT_PKACC_U8_F32 - Float to unsigned byte (does bit-field insert into 32-bit integer). +// V_PERM_B32 does byte packing with ability to zero fill bytes as well. +// - Can pull out byte values from two sources, and zero fill upper 8-bits of packed hi and lo. +//------------------------------------------------------------------------------------------------------------------------------ +// BYTE : FLOAT - ABuc{0,1,2,3}{To,From}U1() - Designed for V_CVT_F32_UBYTE* and V_CVT_PKACCUM_U8_F32 ops. +// ==== ===== +// 0 : 0 +// 1 : 1 +// ... +// 255 : 255 +// : 256 (just outside the encoding range) +//------------------------------------------------------------------------------------------------------------------------------ +// BYTE : FLOAT - ABuc{0,1,2,3}{To,From}U2() - Designed for 16-bit denormal tricks and V_PERM_B32. +// ==== ===== +// 0 : 0 +// 1 : 1/512 +// 2 : 1/256 +// ... +// 64 : 1/8 +// 128 : 1/4 +// 255 : 255/512 +// : 1/2 (just outside the encoding range) +//------------------------------------------------------------------------------------------------------------------------------ +// OPTIMAL IMPLEMENTATIONS ON AMD ARCHITECTURES +// ============================================ +// r=ABuc0FromU1(i) +// V_CVT_F32_UBYTE0 r,i +// -------------------------------------------- +// r=ABuc0ToU1(d,i) +// V_CVT_PKACCUM_U8_F32 r,i,0,d +// -------------------------------------------- +// d=ABuc0FromU2(i) +// Where 'k0' is an SGPR with 0x0E0A +// Where 'k1' is an SGPR with {32768.0} packed into the lower 16-bits +// V_PERM_B32 d,i.x,i.y,k0 +// V_PK_FMA_F16 d,d,k1.x,0 +// -------------------------------------------- +// r=ABuc0ToU2(d,i) +// Where 'k0' is an SGPR with {1.0/32768.0} packed into the lower 16-bits +// Where 'k1' is an SGPR with 0x???? +// Where 'k2' is an SGPR with 0x???? +// V_PK_FMA_F16 i,i,k0.x,0 +// V_PERM_B32 r.x,i,i,k1 +// V_PERM_B32 r.y,i,i,k2 +//============================================================================================================================== + // Peak range for 32-bit and 16-bit operations. + #define A_BUC_32 (255.0) + #define A_BUC_16 (255.0/512.0) +//============================================================================================================================== + #if 1 + // Designed to be one V_CVT_PKACCUM_U8_F32. + // The extra min is required to pattern match to V_CVT_PKACCUM_U8_F32. + AU1 ABuc0ToU1(AU1 d,AF1 i){return (d&0xffffff00u)|((min(AU1(i),255u) )&(0x000000ffu));} + AU1 ABuc1ToU1(AU1 d,AF1 i){return (d&0xffff00ffu)|((min(AU1(i),255u)<< 8)&(0x0000ff00u));} + AU1 ABuc2ToU1(AU1 d,AF1 i){return (d&0xff00ffffu)|((min(AU1(i),255u)<<16)&(0x00ff0000u));} + AU1 ABuc3ToU1(AU1 d,AF1 i){return (d&0x00ffffffu)|((min(AU1(i),255u)<<24)&(0xff000000u));} +//------------------------------------------------------------------------------------------------------------------------------ + // Designed to be one V_CVT_F32_UBYTE*. + AF1 ABuc0FromU1(AU1 i){return AF1((i )&255u);} + AF1 ABuc1FromU1(AU1 i){return AF1((i>> 8)&255u);} + AF1 ABuc2FromU1(AU1 i){return AF1((i>>16)&255u);} + AF1 ABuc3FromU1(AU1 i){return AF1((i>>24)&255u);} + #endif +//============================================================================================================================== + #ifdef A_HALF + // Takes {x0,x1} and {y0,y1} and builds {{x0,y0},{x1,y1}}. + AW2 ABuc01ToW2(AH2 x,AH2 y){x*=AH2_(1.0/32768.0);y*=AH2_(1.0/32768.0); + return AW2_AU1(APermGCEA(AU2(AU1_AW2(AW2_AH2(x)),AU1_AW2(AW2_AH2(y)))));} +//------------------------------------------------------------------------------------------------------------------------------ + // Designed for 3 ops to do SOA to AOS and conversion. + AU2 ABuc0ToU2(AU2 d,AH2 i){AU1 b=AU1_AW2(AW2_AH2(i*AH2_(1.0/32768.0))); + return AU2(APermHGFA(AU2(d.x,b)),APermHGFC(AU2(d.y,b)));} + AU2 ABuc1ToU2(AU2 d,AH2 i){AU1 b=AU1_AW2(AW2_AH2(i*AH2_(1.0/32768.0))); + return AU2(APermHGAE(AU2(d.x,b)),APermHGCE(AU2(d.y,b)));} + AU2 ABuc2ToU2(AU2 d,AH2 i){AU1 b=AU1_AW2(AW2_AH2(i*AH2_(1.0/32768.0))); + return AU2(APermHAFE(AU2(d.x,b)),APermHCFE(AU2(d.y,b)));} + AU2 ABuc3ToU2(AU2 d,AH2 i){AU1 b=AU1_AW2(AW2_AH2(i*AH2_(1.0/32768.0))); + return AU2(APermAGFE(AU2(d.x,b)),APermCGFE(AU2(d.y,b)));} +//------------------------------------------------------------------------------------------------------------------------------ + // Designed for 2 ops to do both AOS to SOA, and conversion. + AH2 ABuc0FromU2(AU2 i){return AH2_AW2(AW2_AU1(APerm0E0A(i)))*AH2_(32768.0);} + AH2 ABuc1FromU2(AU2 i){return AH2_AW2(AW2_AU1(APerm0F0B(i)))*AH2_(32768.0);} + AH2 ABuc2FromU2(AU2 i){return AH2_AW2(AW2_AU1(APerm0G0C(i)))*AH2_(32768.0);} + AH2 ABuc3FromU2(AU2 i){return AH2_AW2(AW2_AU1(APerm0H0D(i)))*AH2_(32768.0);} + #endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// [BSC] BYTE SIGNED CONVERSION +//------------------------------------------------------------------------------------------------------------------------------ +// Similar to [BUC]. +// Works on a range of {-/+ A_BSC_<32,16>}, for <32-bit, and 16-bit> respectively. +//------------------------------------------------------------------------------------------------------------------------------ +// ENCODING (without zero-based encoding) +// ======== +// 0 = unused (can be used to mean something else) +// 1 = lowest value +// 128 = exact zero center (zero based encoding +// 255 = highest value +//------------------------------------------------------------------------------------------------------------------------------ +// Zero-based [Zb] flips the MSB bit of the byte (making 128 "exact zero" actually zero). +// This is useful if there is a desire for cleared values to decode as zero. +//------------------------------------------------------------------------------------------------------------------------------ +// BYTE : FLOAT - ABsc{0,1,2,3}{To,From}U2() - Designed for 16-bit denormal tricks and V_PERM_B32. +// ==== ===== +// 0 : -127/512 (unused) +// 1 : -126/512 +// 2 : -125/512 +// ... +// 128 : 0 +// ... +// 255 : 127/512 +// : 1/4 (just outside the encoding range) +//============================================================================================================================== + // Peak range for 32-bit and 16-bit operations. + #define A_BSC_32 (127.0) + #define A_BSC_16 (127.0/512.0) +//============================================================================================================================== + #if 1 + AU1 ABsc0ToU1(AU1 d,AF1 i){return (d&0xffffff00u)|((min(AU1(i+128.0),255u) )&(0x000000ffu));} + AU1 ABsc1ToU1(AU1 d,AF1 i){return (d&0xffff00ffu)|((min(AU1(i+128.0),255u)<< 8)&(0x0000ff00u));} + AU1 ABsc2ToU1(AU1 d,AF1 i){return (d&0xff00ffffu)|((min(AU1(i+128.0),255u)<<16)&(0x00ff0000u));} + AU1 ABsc3ToU1(AU1 d,AF1 i){return (d&0x00ffffffu)|((min(AU1(i+128.0),255u)<<24)&(0xff000000u));} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 ABsc0ToZbU1(AU1 d,AF1 i){return ((d&0xffffff00u)|((min(AU1(trunc(i)+128.0),255u) )&(0x000000ffu)))^0x00000080u;} + AU1 ABsc1ToZbU1(AU1 d,AF1 i){return ((d&0xffff00ffu)|((min(AU1(trunc(i)+128.0),255u)<< 8)&(0x0000ff00u)))^0x00008000u;} + AU1 ABsc2ToZbU1(AU1 d,AF1 i){return ((d&0xff00ffffu)|((min(AU1(trunc(i)+128.0),255u)<<16)&(0x00ff0000u)))^0x00800000u;} + AU1 ABsc3ToZbU1(AU1 d,AF1 i){return ((d&0x00ffffffu)|((min(AU1(trunc(i)+128.0),255u)<<24)&(0xff000000u)))^0x80000000u;} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 ABsc0FromU1(AU1 i){return AF1((i )&255u)-128.0;} + AF1 ABsc1FromU1(AU1 i){return AF1((i>> 8)&255u)-128.0;} + AF1 ABsc2FromU1(AU1 i){return AF1((i>>16)&255u)-128.0;} + AF1 ABsc3FromU1(AU1 i){return AF1((i>>24)&255u)-128.0;} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 ABsc0FromZbU1(AU1 i){return AF1(((i )&255u)^0x80u)-128.0;} + AF1 ABsc1FromZbU1(AU1 i){return AF1(((i>> 8)&255u)^0x80u)-128.0;} + AF1 ABsc2FromZbU1(AU1 i){return AF1(((i>>16)&255u)^0x80u)-128.0;} + AF1 ABsc3FromZbU1(AU1 i){return AF1(((i>>24)&255u)^0x80u)-128.0;} + #endif +//============================================================================================================================== + #ifdef A_HALF + // Takes {x0,x1} and {y0,y1} and builds {{x0,y0},{x1,y1}}. + AW2 ABsc01ToW2(AH2 x,AH2 y){x=x*AH2_(1.0/32768.0)+AH2_(0.25/32768.0);y=y*AH2_(1.0/32768.0)+AH2_(0.25/32768.0); + return AW2_AU1(APermGCEA(AU2(AU1_AW2(AW2_AH2(x)),AU1_AW2(AW2_AH2(y)))));} +//------------------------------------------------------------------------------------------------------------------------------ + AU2 ABsc0ToU2(AU2 d,AH2 i){AU1 b=AU1_AW2(AW2_AH2(i*AH2_(1.0/32768.0)+AH2_(0.25/32768.0))); + return AU2(APermHGFA(AU2(d.x,b)),APermHGFC(AU2(d.y,b)));} + AU2 ABsc1ToU2(AU2 d,AH2 i){AU1 b=AU1_AW2(AW2_AH2(i*AH2_(1.0/32768.0)+AH2_(0.25/32768.0))); + return AU2(APermHGAE(AU2(d.x,b)),APermHGCE(AU2(d.y,b)));} + AU2 ABsc2ToU2(AU2 d,AH2 i){AU1 b=AU1_AW2(AW2_AH2(i*AH2_(1.0/32768.0)+AH2_(0.25/32768.0))); + return AU2(APermHAFE(AU2(d.x,b)),APermHCFE(AU2(d.y,b)));} + AU2 ABsc3ToU2(AU2 d,AH2 i){AU1 b=AU1_AW2(AW2_AH2(i*AH2_(1.0/32768.0)+AH2_(0.25/32768.0))); + return AU2(APermAGFE(AU2(d.x,b)),APermCGFE(AU2(d.y,b)));} +//------------------------------------------------------------------------------------------------------------------------------ + AU2 ABsc0ToZbU2(AU2 d,AH2 i){AU1 b=AU1_AW2(AW2_AH2(i*AH2_(1.0/32768.0)+AH2_(0.25/32768.0)))^0x00800080u; + return AU2(APermHGFA(AU2(d.x,b)),APermHGFC(AU2(d.y,b)));} + AU2 ABsc1ToZbU2(AU2 d,AH2 i){AU1 b=AU1_AW2(AW2_AH2(i*AH2_(1.0/32768.0)+AH2_(0.25/32768.0)))^0x00800080u; + return AU2(APermHGAE(AU2(d.x,b)),APermHGCE(AU2(d.y,b)));} + AU2 ABsc2ToZbU2(AU2 d,AH2 i){AU1 b=AU1_AW2(AW2_AH2(i*AH2_(1.0/32768.0)+AH2_(0.25/32768.0)))^0x00800080u; + return AU2(APermHAFE(AU2(d.x,b)),APermHCFE(AU2(d.y,b)));} + AU2 ABsc3ToZbU2(AU2 d,AH2 i){AU1 b=AU1_AW2(AW2_AH2(i*AH2_(1.0/32768.0)+AH2_(0.25/32768.0)))^0x00800080u; + return AU2(APermAGFE(AU2(d.x,b)),APermCGFE(AU2(d.y,b)));} +//------------------------------------------------------------------------------------------------------------------------------ + AH2 ABsc0FromU2(AU2 i){return AH2_AW2(AW2_AU1(APerm0E0A(i)))*AH2_(32768.0)-AH2_(0.25);} + AH2 ABsc1FromU2(AU2 i){return AH2_AW2(AW2_AU1(APerm0F0B(i)))*AH2_(32768.0)-AH2_(0.25);} + AH2 ABsc2FromU2(AU2 i){return AH2_AW2(AW2_AU1(APerm0G0C(i)))*AH2_(32768.0)-AH2_(0.25);} + AH2 ABsc3FromU2(AU2 i){return AH2_AW2(AW2_AU1(APerm0H0D(i)))*AH2_(32768.0)-AH2_(0.25);} +//------------------------------------------------------------------------------------------------------------------------------ + AH2 ABsc0FromZbU2(AU2 i){return AH2_AW2(AW2_AU1(APerm0E0A(i)^0x00800080u))*AH2_(32768.0)-AH2_(0.25);} + AH2 ABsc1FromZbU2(AU2 i){return AH2_AW2(AW2_AU1(APerm0F0B(i)^0x00800080u))*AH2_(32768.0)-AH2_(0.25);} + AH2 ABsc2FromZbU2(AU2 i){return AH2_AW2(AW2_AU1(APerm0G0C(i)^0x00800080u))*AH2_(32768.0)-AH2_(0.25);} + AH2 ABsc3FromZbU2(AU2 i){return AH2_AW2(AW2_AU1(APerm0H0D(i)^0x00800080u))*AH2_(32768.0)-AH2_(0.25);} + #endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// HALF APPROXIMATIONS +//------------------------------------------------------------------------------------------------------------------------------ +// These support only positive inputs. +// Did not see value yet in specialization for range. +// Using quick testing, ended up mostly getting the same "best" approximation for various ranges. +// With hardware that can co-execute transcendentals, the value in approximations could be less than expected. +// However from a latency perspective, if execution of a transcendental is 4 clk, with no packed support, -> 8 clk total. +// And co-execution would require a compiler interleaving a lot of independent work for packed usage. +//------------------------------------------------------------------------------------------------------------------------------ +// The one Newton Raphson iteration form of rsq() was skipped (requires 6 ops total). +// Same with sqrt(), as this could be x*rsq() (7 ops). +//============================================================================================================================== + #ifdef A_HALF + // Minimize squared error across full positive range, 2 ops. + // The 0x1de2 based approximation maps {0 to 1} input maps to < 1 output. + AH1 APrxLoSqrtH1(AH1 a){return AH1_AW1((AW1_AH1(a)>>AW1_(1))+AW1_(0x1de2));} + AH2 APrxLoSqrtH2(AH2 a){return AH2_AW2((AW2_AH2(a)>>AW2_(1))+AW2_(0x1de2));} + AH3 APrxLoSqrtH3(AH3 a){return AH3_AW3((AW3_AH3(a)>>AW3_(1))+AW3_(0x1de2));} + AH4 APrxLoSqrtH4(AH4 a){return AH4_AW4((AW4_AH4(a)>>AW4_(1))+AW4_(0x1de2));} +//------------------------------------------------------------------------------------------------------------------------------ + // Lower precision estimation, 1 op. + // Minimize squared error across {smallest normal to 16384.0}. + AH1 APrxLoRcpH1(AH1 a){return AH1_AW1(AW1_(0x7784)-AW1_AH1(a));} + AH2 APrxLoRcpH2(AH2 a){return AH2_AW2(AW2_(0x7784)-AW2_AH2(a));} + AH3 APrxLoRcpH3(AH3 a){return AH3_AW3(AW3_(0x7784)-AW3_AH3(a));} + AH4 APrxLoRcpH4(AH4 a){return AH4_AW4(AW4_(0x7784)-AW4_AH4(a));} +//------------------------------------------------------------------------------------------------------------------------------ + // Medium precision estimation, one Newton Raphson iteration, 3 ops. + AH1 APrxMedRcpH1(AH1 a){AH1 b=AH1_AW1(AW1_(0x778d)-AW1_AH1(a));return b*(-b*a+AH1_(2.0));} + AH2 APrxMedRcpH2(AH2 a){AH2 b=AH2_AW2(AW2_(0x778d)-AW2_AH2(a));return b*(-b*a+AH2_(2.0));} + AH3 APrxMedRcpH3(AH3 a){AH3 b=AH3_AW3(AW3_(0x778d)-AW3_AH3(a));return b*(-b*a+AH3_(2.0));} + AH4 APrxMedRcpH4(AH4 a){AH4 b=AH4_AW4(AW4_(0x778d)-AW4_AH4(a));return b*(-b*a+AH4_(2.0));} +//------------------------------------------------------------------------------------------------------------------------------ + // Minimize squared error across {smallest normal to 16384.0}, 2 ops. + AH1 APrxLoRsqH1(AH1 a){return AH1_AW1(AW1_(0x59a3)-(AW1_AH1(a)>>AW1_(1)));} + AH2 APrxLoRsqH2(AH2 a){return AH2_AW2(AW2_(0x59a3)-(AW2_AH2(a)>>AW2_(1)));} + AH3 APrxLoRsqH3(AH3 a){return AH3_AW3(AW3_(0x59a3)-(AW3_AH3(a)>>AW3_(1)));} + AH4 APrxLoRsqH4(AH4 a){return AH4_AW4(AW4_(0x59a3)-(AW4_AH4(a)>>AW4_(1)));} + #endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// FLOAT APPROXIMATIONS +//------------------------------------------------------------------------------------------------------------------------------ +// Michal Drobot has an excellent presentation on these: "Low Level Optimizations For GCN", +// - Idea dates back to SGI, then to Quake 3, etc. +// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +// - sqrt(x)=rsqrt(x)*x +// - rcp(x)=rsqrt(x)*rsqrt(x) for positive x +// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +//------------------------------------------------------------------------------------------------------------------------------ +// These below are from perhaps less complete searching for optimal. +// Used FP16 normal range for testing with +4096 32-bit step size for sampling error. +// So these match up well with the half approximations. +//============================================================================================================================== + AF1 APrxLoSqrtF1(AF1 a){return AF1_AU1((AU1_AF1(a)>>AU1_(1))+AU1_(0x1fbc4639));} + AF1 APrxLoRcpF1(AF1 a){return AF1_AU1(AU1_(0x7ef07ebb)-AU1_AF1(a));} + AF1 APrxMedRcpF1(AF1 a){AF1 b=AF1_AU1(AU1_(0x7ef19fff)-AU1_AF1(a));return b*(-b*a+AF1_(2.0));} + AF1 APrxLoRsqF1(AF1 a){return AF1_AU1(AU1_(0x5f347d74)-(AU1_AF1(a)>>AU1_(1)));} +//------------------------------------------------------------------------------------------------------------------------------ + AF2 APrxLoSqrtF2(AF2 a){return AF2_AU2((AU2_AF2(a)>>AU2_(1))+AU2_(0x1fbc4639));} + AF2 APrxLoRcpF2(AF2 a){return AF2_AU2(AU2_(0x7ef07ebb)-AU2_AF2(a));} + AF2 APrxMedRcpF2(AF2 a){AF2 b=AF2_AU2(AU2_(0x7ef19fff)-AU2_AF2(a));return b*(-b*a+AF2_(2.0));} + AF2 APrxLoRsqF2(AF2 a){return AF2_AU2(AU2_(0x5f347d74)-(AU2_AF2(a)>>AU2_(1)));} +//------------------------------------------------------------------------------------------------------------------------------ + AF3 APrxLoSqrtF3(AF3 a){return AF3_AU3((AU3_AF3(a)>>AU3_(1))+AU3_(0x1fbc4639));} + AF3 APrxLoRcpF3(AF3 a){return AF3_AU3(AU3_(0x7ef07ebb)-AU3_AF3(a));} + AF3 APrxMedRcpF3(AF3 a){AF3 b=AF3_AU3(AU3_(0x7ef19fff)-AU3_AF3(a));return b*(-b*a+AF3_(2.0));} + AF3 APrxLoRsqF3(AF3 a){return AF3_AU3(AU3_(0x5f347d74)-(AU3_AF3(a)>>AU3_(1)));} +//------------------------------------------------------------------------------------------------------------------------------ + AF4 APrxLoSqrtF4(AF4 a){return AF4_AU4((AU4_AF4(a)>>AU4_(1))+AU4_(0x1fbc4639));} + AF4 APrxLoRcpF4(AF4 a){return AF4_AU4(AU4_(0x7ef07ebb)-AU4_AF4(a));} + AF4 APrxMedRcpF4(AF4 a){AF4 b=AF4_AU4(AU4_(0x7ef19fff)-AU4_AF4(a));return b*(-b*a+AF4_(2.0));} + AF4 APrxLoRsqF4(AF4 a){return AF4_AU4(AU4_(0x5f347d74)-(AU4_AF4(a)>>AU4_(1)));} +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// PQ APPROXIMATIONS +//------------------------------------------------------------------------------------------------------------------------------ +// PQ is very close to x^(1/8). The functions below Use the fast float approximation method to do +// PQ<~>Gamma2 (4th power and fast 4th root) and PQ<~>Linear (8th power and fast 8th root). Maximum error is ~0.2%. +//============================================================================================================================== +// Helpers + AF1 Quart(AF1 a) { a = a * a; return a * a;} + AF1 Oct(AF1 a) { a = a * a; a = a * a; return a * a; } + AF2 Quart(AF2 a) { a = a * a; return a * a; } + AF2 Oct(AF2 a) { a = a * a; a = a * a; return a * a; } + AF3 Quart(AF3 a) { a = a * a; return a * a; } + AF3 Oct(AF3 a) { a = a * a; a = a * a; return a * a; } + AF4 Quart(AF4 a) { a = a * a; return a * a; } + AF4 Oct(AF4 a) { a = a * a; a = a * a; return a * a; } + //------------------------------------------------------------------------------------------------------------------------------ + AF1 APrxPQToGamma2(AF1 a) { return Quart(a); } + AF1 APrxPQToLinear(AF1 a) { return Oct(a); } + AF1 APrxLoGamma2ToPQ(AF1 a) { return AF1_AU1((AU1_AF1(a) >> AU1_(2)) + AU1_(0x2F9A4E46)); } + AF1 APrxMedGamma2ToPQ(AF1 a) { AF1 b = AF1_AU1((AU1_AF1(a) >> AU1_(2)) + AU1_(0x2F9A4E46)); AF1 b4 = Quart(b); return b - b * (b4 - a) / (AF1_(4.0) * b4); } + AF1 APrxHighGamma2ToPQ(AF1 a) { return sqrt(sqrt(a)); } + AF1 APrxLoLinearToPQ(AF1 a) { return AF1_AU1((AU1_AF1(a) >> AU1_(3)) + AU1_(0x378D8723)); } + AF1 APrxMedLinearToPQ(AF1 a) { AF1 b = AF1_AU1((AU1_AF1(a) >> AU1_(3)) + AU1_(0x378D8723)); AF1 b8 = Oct(b); return b - b * (b8 - a) / (AF1_(8.0) * b8); } + AF1 APrxHighLinearToPQ(AF1 a) { return sqrt(sqrt(sqrt(a))); } + //------------------------------------------------------------------------------------------------------------------------------ + AF2 APrxPQToGamma2(AF2 a) { return Quart(a); } + AF2 APrxPQToLinear(AF2 a) { return Oct(a); } + AF2 APrxLoGamma2ToPQ(AF2 a) { return AF2_AU2((AU2_AF2(a) >> AU2_(2)) + AU2_(0x2F9A4E46)); } + AF2 APrxMedGamma2ToPQ(AF2 a) { AF2 b = AF2_AU2((AU2_AF2(a) >> AU2_(2)) + AU2_(0x2F9A4E46)); AF2 b4 = Quart(b); return b - b * (b4 - a) / (AF1_(4.0) * b4); } + AF2 APrxHighGamma2ToPQ(AF2 a) { return sqrt(sqrt(a)); } + AF2 APrxLoLinearToPQ(AF2 a) { return AF2_AU2((AU2_AF2(a) >> AU2_(3)) + AU2_(0x378D8723)); } + AF2 APrxMedLinearToPQ(AF2 a) { AF2 b = AF2_AU2((AU2_AF2(a) >> AU2_(3)) + AU2_(0x378D8723)); AF2 b8 = Oct(b); return b - b * (b8 - a) / (AF1_(8.0) * b8); } + AF2 APrxHighLinearToPQ(AF2 a) { return sqrt(sqrt(sqrt(a))); } + //------------------------------------------------------------------------------------------------------------------------------ + AF3 APrxPQToGamma2(AF3 a) { return Quart(a); } + AF3 APrxPQToLinear(AF3 a) { return Oct(a); } + AF3 APrxLoGamma2ToPQ(AF3 a) { return AF3_AU3((AU3_AF3(a) >> AU3_(2)) + AU3_(0x2F9A4E46)); } + AF3 APrxMedGamma2ToPQ(AF3 a) { AF3 b = AF3_AU3((AU3_AF3(a) >> AU3_(2)) + AU3_(0x2F9A4E46)); AF3 b4 = Quart(b); return b - b * (b4 - a) / (AF1_(4.0) * b4); } + AF3 APrxHighGamma2ToPQ(AF3 a) { return sqrt(sqrt(a)); } + AF3 APrxLoLinearToPQ(AF3 a) { return AF3_AU3((AU3_AF3(a) >> AU3_(3)) + AU3_(0x378D8723)); } + AF3 APrxMedLinearToPQ(AF3 a) { AF3 b = AF3_AU3((AU3_AF3(a) >> AU3_(3)) + AU3_(0x378D8723)); AF3 b8 = Oct(b); return b - b * (b8 - a) / (AF1_(8.0) * b8); } + AF3 APrxHighLinearToPQ(AF3 a) { return sqrt(sqrt(sqrt(a))); } + //------------------------------------------------------------------------------------------------------------------------------ + AF4 APrxPQToGamma2(AF4 a) { return Quart(a); } + AF4 APrxPQToLinear(AF4 a) { return Oct(a); } + AF4 APrxLoGamma2ToPQ(AF4 a) { return AF4_AU4((AU4_AF4(a) >> AU4_(2)) + AU4_(0x2F9A4E46)); } + AF4 APrxMedGamma2ToPQ(AF4 a) { AF4 b = AF4_AU4((AU4_AF4(a) >> AU4_(2)) + AU4_(0x2F9A4E46)); AF4 b4 = Quart(b); return b - b * (b4 - a) / (AF1_(4.0) * b4); } + AF4 APrxHighGamma2ToPQ(AF4 a) { return sqrt(sqrt(a)); } + AF4 APrxLoLinearToPQ(AF4 a) { return AF4_AU4((AU4_AF4(a) >> AU4_(3)) + AU4_(0x378D8723)); } + AF4 APrxMedLinearToPQ(AF4 a) { AF4 b = AF4_AU4((AU4_AF4(a) >> AU4_(3)) + AU4_(0x378D8723)); AF4 b8 = Oct(b); return b - b * (b8 - a) / (AF1_(8.0) * b8); } + AF4 APrxHighLinearToPQ(AF4 a) { return sqrt(sqrt(sqrt(a))); } +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// PARABOLIC SIN & COS +//------------------------------------------------------------------------------------------------------------------------------ +// Approximate answers to transcendental questions. +//------------------------------------------------------------------------------------------------------------------------------ +//============================================================================================================================== + #if 1 + // Valid input range is {-1 to 1} representing {0 to 2 pi}. + // Output range is {-1/4 to 1/4} representing {-1 to 1}. + AF1 APSinF1(AF1 x){return x*abs(x)-x;} // MAD. + AF2 APSinF2(AF2 x){return x*abs(x)-x;} + AF1 APCosF1(AF1 x){x=AFractF1(x*AF1_(0.5)+AF1_(0.75));x=x*AF1_(2.0)-AF1_(1.0);return APSinF1(x);} // 3x MAD, FRACT + AF2 APCosF2(AF2 x){x=AFractF2(x*AF2_(0.5)+AF2_(0.75));x=x*AF2_(2.0)-AF2_(1.0);return APSinF2(x);} + AF2 APSinCosF1(AF1 x){AF1 y=AFractF1(x*AF1_(0.5)+AF1_(0.75));y=y*AF1_(2.0)-AF1_(1.0);return APSinF2(AF2(x,y));} + #endif +//------------------------------------------------------------------------------------------------------------------------------ + #ifdef A_HALF + // For a packed {sin,cos} pair, + // - Native takes 16 clocks and 4 issue slots (no packed transcendentals). + // - Parabolic takes 8 clocks and 8 issue slots (only fract is non-packed). + AH1 APSinH1(AH1 x){return x*abs(x)-x;} + AH2 APSinH2(AH2 x){return x*abs(x)-x;} // AND,FMA + AH1 APCosH1(AH1 x){x=AFractH1(x*AH1_(0.5)+AH1_(0.75));x=x*AH1_(2.0)-AH1_(1.0);return APSinH1(x);} + AH2 APCosH2(AH2 x){x=AFractH2(x*AH2_(0.5)+AH2_(0.75));x=x*AH2_(2.0)-AH2_(1.0);return APSinH2(x);} // 3x FMA, 2xFRACT, AND + AH2 APSinCosH1(AH1 x){AH1 y=AFractH1(x*AH1_(0.5)+AH1_(0.75));y=y*AH1_(2.0)-AH1_(1.0);return APSinH2(AH2(x,y));} + #endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// [ZOL] ZERO ONE LOGIC +//------------------------------------------------------------------------------------------------------------------------------ +// Conditional free logic designed for easy 16-bit packing, and backwards porting to 32-bit. +//------------------------------------------------------------------------------------------------------------------------------ +// 0 := false +// 1 := true +//------------------------------------------------------------------------------------------------------------------------------ +// AndNot(x,y) -> !(x&y) .... One op. +// AndOr(x,y,z) -> (x&y)|z ... One op. +// GtZero(x) -> x>0.0 ..... One op. +// Sel(x,y,z) -> x?y:z ..... Two ops, has no precision loss. +// Signed(x) -> x<0.0 ..... One op. +// ZeroPass(x,y) -> x?0:y ..... Two ops, 'y' is a pass through safe for aliasing as integer. +//------------------------------------------------------------------------------------------------------------------------------ +// OPTIMIZATION NOTES +// ================== +// - On Vega to use 2 constants in a packed op, pass in as one AW2 or one AH2 'k.xy' and use as 'k.xx' and 'k.yy'. +// For example 'a.xy*k.xx+k.yy'. +//============================================================================================================================== + #if 1 + AU1 AZolAndU1(AU1 x,AU1 y){return min(x,y);} + AU2 AZolAndU2(AU2 x,AU2 y){return min(x,y);} + AU3 AZolAndU3(AU3 x,AU3 y){return min(x,y);} + AU4 AZolAndU4(AU4 x,AU4 y){return min(x,y);} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AZolNotU1(AU1 x){return x^AU1_(1);} + AU2 AZolNotU2(AU2 x){return x^AU2_(1);} + AU3 AZolNotU3(AU3 x){return x^AU3_(1);} + AU4 AZolNotU4(AU4 x){return x^AU4_(1);} +//------------------------------------------------------------------------------------------------------------------------------ + AU1 AZolOrU1(AU1 x,AU1 y){return max(x,y);} + AU2 AZolOrU2(AU2 x,AU2 y){return max(x,y);} + AU3 AZolOrU3(AU3 x,AU3 y){return max(x,y);} + AU4 AZolOrU4(AU4 x,AU4 y){return max(x,y);} +//============================================================================================================================== + AU1 AZolF1ToU1(AF1 x){return AU1(x);} + AU2 AZolF2ToU2(AF2 x){return AU2(x);} + AU3 AZolF3ToU3(AF3 x){return AU3(x);} + AU4 AZolF4ToU4(AF4 x){return AU4(x);} +//------------------------------------------------------------------------------------------------------------------------------ + // 2 ops, denormals don't work in 32-bit on PC (and if they are enabled, OMOD is disabled). + AU1 AZolNotF1ToU1(AF1 x){return AU1(AF1_(1.0)-x);} + AU2 AZolNotF2ToU2(AF2 x){return AU2(AF2_(1.0)-x);} + AU3 AZolNotF3ToU3(AF3 x){return AU3(AF3_(1.0)-x);} + AU4 AZolNotF4ToU4(AF4 x){return AU4(AF4_(1.0)-x);} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AZolU1ToF1(AU1 x){return AF1(x);} + AF2 AZolU2ToF2(AU2 x){return AF2(x);} + AF3 AZolU3ToF3(AU3 x){return AF3(x);} + AF4 AZolU4ToF4(AU4 x){return AF4(x);} +//============================================================================================================================== + AF1 AZolAndF1(AF1 x,AF1 y){return min(x,y);} + AF2 AZolAndF2(AF2 x,AF2 y){return min(x,y);} + AF3 AZolAndF3(AF3 x,AF3 y){return min(x,y);} + AF4 AZolAndF4(AF4 x,AF4 y){return min(x,y);} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 ASolAndNotF1(AF1 x,AF1 y){return (-x)*y+AF1_(1.0);} + AF2 ASolAndNotF2(AF2 x,AF2 y){return (-x)*y+AF2_(1.0);} + AF3 ASolAndNotF3(AF3 x,AF3 y){return (-x)*y+AF3_(1.0);} + AF4 ASolAndNotF4(AF4 x,AF4 y){return (-x)*y+AF4_(1.0);} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AZolAndOrF1(AF1 x,AF1 y,AF1 z){return ASatF1(x*y+z);} + AF2 AZolAndOrF2(AF2 x,AF2 y,AF2 z){return ASatF2(x*y+z);} + AF3 AZolAndOrF3(AF3 x,AF3 y,AF3 z){return ASatF3(x*y+z);} + AF4 AZolAndOrF4(AF4 x,AF4 y,AF4 z){return ASatF4(x*y+z);} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AZolGtZeroF1(AF1 x){return ASatF1(x*AF1_(A_INFP_F));} + AF2 AZolGtZeroF2(AF2 x){return ASatF2(x*AF2_(A_INFP_F));} + AF3 AZolGtZeroF3(AF3 x){return ASatF3(x*AF3_(A_INFP_F));} + AF4 AZolGtZeroF4(AF4 x){return ASatF4(x*AF4_(A_INFP_F));} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AZolNotF1(AF1 x){return AF1_(1.0)-x;} + AF2 AZolNotF2(AF2 x){return AF2_(1.0)-x;} + AF3 AZolNotF3(AF3 x){return AF3_(1.0)-x;} + AF4 AZolNotF4(AF4 x){return AF4_(1.0)-x;} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AZolOrF1(AF1 x,AF1 y){return max(x,y);} + AF2 AZolOrF2(AF2 x,AF2 y){return max(x,y);} + AF3 AZolOrF3(AF3 x,AF3 y){return max(x,y);} + AF4 AZolOrF4(AF4 x,AF4 y){return max(x,y);} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AZolSelF1(AF1 x,AF1 y,AF1 z){AF1 r=(-x)*z+z;return x*y+r;} + AF2 AZolSelF2(AF2 x,AF2 y,AF2 z){AF2 r=(-x)*z+z;return x*y+r;} + AF3 AZolSelF3(AF3 x,AF3 y,AF3 z){AF3 r=(-x)*z+z;return x*y+r;} + AF4 AZolSelF4(AF4 x,AF4 y,AF4 z){AF4 r=(-x)*z+z;return x*y+r;} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AZolSignedF1(AF1 x){return ASatF1(x*AF1_(A_INFN_F));} + AF2 AZolSignedF2(AF2 x){return ASatF2(x*AF2_(A_INFN_F));} + AF3 AZolSignedF3(AF3 x){return ASatF3(x*AF3_(A_INFN_F));} + AF4 AZolSignedF4(AF4 x){return ASatF4(x*AF4_(A_INFN_F));} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AZolZeroPassF1(AF1 x,AF1 y){return AF1_AU1((AU1_AF1(x)!=AU1_(0))?AU1_(0):AU1_AF1(y));} + AF2 AZolZeroPassF2(AF2 x,AF2 y){return AF2_AU2((AU2_AF2(x)!=AU2_(0))?AU2_(0):AU2_AF2(y));} + AF3 AZolZeroPassF3(AF3 x,AF3 y){return AF3_AU3((AU3_AF3(x)!=AU3_(0))?AU3_(0):AU3_AF3(y));} + AF4 AZolZeroPassF4(AF4 x,AF4 y){return AF4_AU4((AU4_AF4(x)!=AU4_(0))?AU4_(0):AU4_AF4(y));} + #endif +//============================================================================================================================== + #ifdef A_HALF + AW1 AZolAndW1(AW1 x,AW1 y){return min(x,y);} + AW2 AZolAndW2(AW2 x,AW2 y){return min(x,y);} + AW3 AZolAndW3(AW3 x,AW3 y){return min(x,y);} + AW4 AZolAndW4(AW4 x,AW4 y){return min(x,y);} +//------------------------------------------------------------------------------------------------------------------------------ + AW1 AZolNotW1(AW1 x){return x^AW1_(1);} + AW2 AZolNotW2(AW2 x){return x^AW2_(1);} + AW3 AZolNotW3(AW3 x){return x^AW3_(1);} + AW4 AZolNotW4(AW4 x){return x^AW4_(1);} +//------------------------------------------------------------------------------------------------------------------------------ + AW1 AZolOrW1(AW1 x,AW1 y){return max(x,y);} + AW2 AZolOrW2(AW2 x,AW2 y){return max(x,y);} + AW3 AZolOrW3(AW3 x,AW3 y){return max(x,y);} + AW4 AZolOrW4(AW4 x,AW4 y){return max(x,y);} +//============================================================================================================================== + // Uses denormal trick. + AW1 AZolH1ToW1(AH1 x){return AW1_AH1(x*AH1_AW1(AW1_(1)));} + AW2 AZolH2ToW2(AH2 x){return AW2_AH2(x*AH2_AW2(AW2_(1)));} + AW3 AZolH3ToW3(AH3 x){return AW3_AH3(x*AH3_AW3(AW3_(1)));} + AW4 AZolH4ToW4(AH4 x){return AW4_AH4(x*AH4_AW4(AW4_(1)));} +//------------------------------------------------------------------------------------------------------------------------------ + // AMD arch lacks a packed conversion opcode. + AH1 AZolW1ToH1(AW1 x){return AH1_AW1(x*AW1_AH1(AH1_(1.0)));} + AH2 AZolW2ToH2(AW2 x){return AH2_AW2(x*AW2_AH2(AH2_(1.0)));} + AH3 AZolW1ToH3(AW3 x){return AH3_AW3(x*AW3_AH3(AH3_(1.0)));} + AH4 AZolW2ToH4(AW4 x){return AH4_AW4(x*AW4_AH4(AH4_(1.0)));} +//============================================================================================================================== + AH1 AZolAndH1(AH1 x,AH1 y){return min(x,y);} + AH2 AZolAndH2(AH2 x,AH2 y){return min(x,y);} + AH3 AZolAndH3(AH3 x,AH3 y){return min(x,y);} + AH4 AZolAndH4(AH4 x,AH4 y){return min(x,y);} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 ASolAndNotH1(AH1 x,AH1 y){return (-x)*y+AH1_(1.0);} + AH2 ASolAndNotH2(AH2 x,AH2 y){return (-x)*y+AH2_(1.0);} + AH3 ASolAndNotH3(AH3 x,AH3 y){return (-x)*y+AH3_(1.0);} + AH4 ASolAndNotH4(AH4 x,AH4 y){return (-x)*y+AH4_(1.0);} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AZolAndOrH1(AH1 x,AH1 y,AH1 z){return ASatH1(x*y+z);} + AH2 AZolAndOrH2(AH2 x,AH2 y,AH2 z){return ASatH2(x*y+z);} + AH3 AZolAndOrH3(AH3 x,AH3 y,AH3 z){return ASatH3(x*y+z);} + AH4 AZolAndOrH4(AH4 x,AH4 y,AH4 z){return ASatH4(x*y+z);} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AZolGtZeroH1(AH1 x){return ASatH1(x*AH1_(A_INFP_H));} + AH2 AZolGtZeroH2(AH2 x){return ASatH2(x*AH2_(A_INFP_H));} + AH3 AZolGtZeroH3(AH3 x){return ASatH3(x*AH3_(A_INFP_H));} + AH4 AZolGtZeroH4(AH4 x){return ASatH4(x*AH4_(A_INFP_H));} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AZolNotH1(AH1 x){return AH1_(1.0)-x;} + AH2 AZolNotH2(AH2 x){return AH2_(1.0)-x;} + AH3 AZolNotH3(AH3 x){return AH3_(1.0)-x;} + AH4 AZolNotH4(AH4 x){return AH4_(1.0)-x;} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AZolOrH1(AH1 x,AH1 y){return max(x,y);} + AH2 AZolOrH2(AH2 x,AH2 y){return max(x,y);} + AH3 AZolOrH3(AH3 x,AH3 y){return max(x,y);} + AH4 AZolOrH4(AH4 x,AH4 y){return max(x,y);} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AZolSelH1(AH1 x,AH1 y,AH1 z){AH1 r=(-x)*z+z;return x*y+r;} + AH2 AZolSelH2(AH2 x,AH2 y,AH2 z){AH2 r=(-x)*z+z;return x*y+r;} + AH3 AZolSelH3(AH3 x,AH3 y,AH3 z){AH3 r=(-x)*z+z;return x*y+r;} + AH4 AZolSelH4(AH4 x,AH4 y,AH4 z){AH4 r=(-x)*z+z;return x*y+r;} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AZolSignedH1(AH1 x){return ASatH1(x*AH1_(A_INFN_H));} + AH2 AZolSignedH2(AH2 x){return ASatH2(x*AH2_(A_INFN_H));} + AH3 AZolSignedH3(AH3 x){return ASatH3(x*AH3_(A_INFN_H));} + AH4 AZolSignedH4(AH4 x){return ASatH4(x*AH4_(A_INFN_H));} + #endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// COLOR CONVERSIONS +//------------------------------------------------------------------------------------------------------------------------------ +// These are all linear to/from some other space (where 'linear' has been shortened out of the function name). +// So 'ToGamma' is 'LinearToGamma', and 'FromGamma' is 'LinearFromGamma'. +// These are branch free implementations. +// The AToSrgbF1() function is useful for stores for compute shaders for GPUs without hardware linear->sRGB store conversion. +//------------------------------------------------------------------------------------------------------------------------------ +// TRANSFER FUNCTIONS +// ================== +// 709 ..... Rec709 used for some HDTVs +// Gamma ... Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native +// Pq ...... PQ native for HDR10 +// Srgb .... The sRGB output, typical of PC displays, useful for 10-bit output, or storing to 8-bit UNORM without SRGB type +// Two ..... Gamma 2.0, fastest conversion (useful for intermediate pass approximations) +// Three ... Gamma 3.0, less fast, but good for HDR. +//------------------------------------------------------------------------------------------------------------------------------ +// KEEPING TO SPEC +// =============== +// Both Rec.709 and sRGB have a linear segment which as spec'ed would intersect the curved segment 2 times. +// (a.) For 8-bit sRGB, steps {0 to 10.3} are in the linear region (4% of the encoding range). +// (b.) For 8-bit 709, steps {0 to 20.7} are in the linear region (8% of the encoding range). +// Also there is a slight step in the transition regions. +// Precision of the coefficients in the spec being the likely cause. +// Main usage case of the sRGB code is to do the linear->sRGB converstion in a compute shader before store. +// This is to work around lack of hardware (typically only ROP does the conversion for free). +// To "correct" the linear segment, would be to introduce error, because hardware decode of sRGB->linear is fixed (and free). +// So this header keeps with the spec. +// For linear->sRGB transforms, the linear segment in some respects reduces error, because rounding in that region is linear. +// Rounding in the curved region in hardware (and fast software code) introduces error due to rounding in non-linear. +//------------------------------------------------------------------------------------------------------------------------------ +// FOR PQ +// ====== +// Both input and output is {0.0-1.0}, and where output 1.0 represents 10000.0 cd/m^2. +// All constants are only specified to FP32 precision. +// External PQ source reference, +// - https://github.com/ampas/aces-dev/blob/master/transforms/ctl/utilities/ACESlib.Utilities_Color.a1.0.1.ctl +//------------------------------------------------------------------------------------------------------------------------------ +// PACKED VERSIONS +// =============== +// These are the A*H2() functions. +// There is no PQ functions as FP16 seemed to not have enough precision for the conversion. +// The remaining functions are "good enough" for 8-bit, and maybe 10-bit if not concerned about a few 1-bit errors. +// Precision is lowest in the 709 conversion, higher in sRGB, higher still in Two and Gamma (when using 2.2 at least). +//------------------------------------------------------------------------------------------------------------------------------ +// NOTES +// ===== +// Could be faster for PQ conversions to be in ALU or a texture lookup depending on usage case. +//============================================================================================================================== + #if 1 + AF1 ATo709F1(AF1 c){AF3 j=AF3(0.018*4.5,4.5,0.45);AF2 k=AF2(1.099,-0.099); + return clamp(j.x ,c*j.y ,pow(c,j.z )*k.x +k.y );} + AF2 ATo709F2(AF2 c){AF3 j=AF3(0.018*4.5,4.5,0.45);AF2 k=AF2(1.099,-0.099); + return clamp(j.xx ,c*j.yy ,pow(c,j.zz )*k.xx +k.yy );} + AF3 ATo709F3(AF3 c){AF3 j=AF3(0.018*4.5,4.5,0.45);AF2 k=AF2(1.099,-0.099); + return clamp(j.xxx,c*j.yyy,pow(c,j.zzz)*k.xxx+k.yyy);} +//------------------------------------------------------------------------------------------------------------------------------ + // Note 'rcpX' is '1/x', where the 'x' is what would be used in AFromGamma(). + AF1 AToGammaF1(AF1 c,AF1 rcpX){return pow(c,AF1_(rcpX));} + AF2 AToGammaF2(AF2 c,AF1 rcpX){return pow(c,AF2_(rcpX));} + AF3 AToGammaF3(AF3 c,AF1 rcpX){return pow(c,AF3_(rcpX));} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AToPqF1(AF1 x){AF1 p=pow(x,AF1_(0.159302)); + return pow((AF1_(0.835938)+AF1_(18.8516)*p)/(AF1_(1.0)+AF1_(18.6875)*p),AF1_(78.8438));} + AF2 AToPqF1(AF2 x){AF2 p=pow(x,AF2_(0.159302)); + return pow((AF2_(0.835938)+AF2_(18.8516)*p)/(AF2_(1.0)+AF2_(18.6875)*p),AF2_(78.8438));} + AF3 AToPqF1(AF3 x){AF3 p=pow(x,AF3_(0.159302)); + return pow((AF3_(0.835938)+AF3_(18.8516)*p)/(AF3_(1.0)+AF3_(18.6875)*p),AF3_(78.8438));} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AToSrgbF1(AF1 c){AF3 j=AF3(0.0031308*12.92,12.92,1.0/2.4);AF2 k=AF2(1.055,-0.055); + return clamp(j.x ,c*j.y ,pow(c,j.z )*k.x +k.y );} + AF2 AToSrgbF2(AF2 c){AF3 j=AF3(0.0031308*12.92,12.92,1.0/2.4);AF2 k=AF2(1.055,-0.055); + return clamp(j.xx ,c*j.yy ,pow(c,j.zz )*k.xx +k.yy );} + AF3 AToSrgbF3(AF3 c){AF3 j=AF3(0.0031308*12.92,12.92,1.0/2.4);AF2 k=AF2(1.055,-0.055); + return clamp(j.xxx,c*j.yyy,pow(c,j.zzz)*k.xxx+k.yyy);} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AToTwoF1(AF1 c){return sqrt(c);} + AF2 AToTwoF2(AF2 c){return sqrt(c);} + AF3 AToTwoF3(AF3 c){return sqrt(c);} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AToThreeF1(AF1 c){return pow(c,AF1_(1.0/3.0));} + AF2 AToThreeF2(AF2 c){return pow(c,AF2_(1.0/3.0));} + AF3 AToThreeF3(AF3 c){return pow(c,AF3_(1.0/3.0));} + #endif +//============================================================================================================================== + #if 1 + // Unfortunately median won't work here. + AF1 AFrom709F1(AF1 c){AF3 j=AF3(0.081/4.5,1.0/4.5,1.0/0.45);AF2 k=AF2(1.0/1.099,0.099/1.099); + return AZolSelF1(AZolSignedF1(c-j.x ),c*j.y ,pow(c*k.x +k.y ,j.z ));} + AF2 AFrom709F2(AF2 c){AF3 j=AF3(0.081/4.5,1.0/4.5,1.0/0.45);AF2 k=AF2(1.0/1.099,0.099/1.099); + return AZolSelF2(AZolSignedF2(c-j.xx ),c*j.yy ,pow(c*k.xx +k.yy ,j.zz ));} + AF3 AFrom709F3(AF3 c){AF3 j=AF3(0.081/4.5,1.0/4.5,1.0/0.45);AF2 k=AF2(1.0/1.099,0.099/1.099); + return AZolSelF3(AZolSignedF3(c-j.xxx),c*j.yyy,pow(c*k.xxx+k.yyy,j.zzz));} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AFromGammaF1(AF1 c,AF1 x){return pow(c,AF1_(x));} + AF2 AFromGammaF2(AF2 c,AF1 x){return pow(c,AF2_(x));} + AF3 AFromGammaF3(AF3 c,AF1 x){return pow(c,AF3_(x));} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AFromPqF1(AF1 x){AF1 p=pow(x,AF1_(0.0126833)); + return pow(ASatF1(p-AF1_(0.835938))/(AF1_(18.8516)-AF1_(18.6875)*p),AF1_(6.27739));} + AF2 AFromPqF1(AF2 x){AF2 p=pow(x,AF2_(0.0126833)); + return pow(ASatF2(p-AF2_(0.835938))/(AF2_(18.8516)-AF2_(18.6875)*p),AF2_(6.27739));} + AF3 AFromPqF1(AF3 x){AF3 p=pow(x,AF3_(0.0126833)); + return pow(ASatF3(p-AF3_(0.835938))/(AF3_(18.8516)-AF3_(18.6875)*p),AF3_(6.27739));} +//------------------------------------------------------------------------------------------------------------------------------ + // Unfortunately median won't work here. + AF1 AFromSrgbF1(AF1 c){AF3 j=AF3(0.04045/12.92,1.0/12.92,2.4);AF2 k=AF2(1.0/1.055,0.055/1.055); + return AZolSelF1(AZolSignedF1(c-j.x ),c*j.y ,pow(c*k.x +k.y ,j.z ));} + AF2 AFromSrgbF2(AF2 c){AF3 j=AF3(0.04045/12.92,1.0/12.92,2.4);AF2 k=AF2(1.0/1.055,0.055/1.055); + return AZolSelF2(AZolSignedF2(c-j.xx ),c*j.yy ,pow(c*k.xx +k.yy ,j.zz ));} + AF3 AFromSrgbF3(AF3 c){AF3 j=AF3(0.04045/12.92,1.0/12.92,2.4);AF2 k=AF2(1.0/1.055,0.055/1.055); + return AZolSelF3(AZolSignedF3(c-j.xxx),c*j.yyy,pow(c*k.xxx+k.yyy,j.zzz));} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AFromTwoF1(AF1 c){return c*c;} + AF2 AFromTwoF2(AF2 c){return c*c;} + AF3 AFromTwoF3(AF3 c){return c*c;} +//------------------------------------------------------------------------------------------------------------------------------ + AF1 AFromThreeF1(AF1 c){return c*c*c;} + AF2 AFromThreeF2(AF2 c){return c*c*c;} + AF3 AFromThreeF3(AF3 c){return c*c*c;} + #endif +//============================================================================================================================== + #ifdef A_HALF + AH1 ATo709H1(AH1 c){AH3 j=AH3(0.018*4.5,4.5,0.45);AH2 k=AH2(1.099,-0.099); + return clamp(j.x ,c*j.y ,pow(c,j.z )*k.x +k.y );} + AH2 ATo709H2(AH2 c){AH3 j=AH3(0.018*4.5,4.5,0.45);AH2 k=AH2(1.099,-0.099); + return clamp(j.xx ,c*j.yy ,pow(c,j.zz )*k.xx +k.yy );} + AH3 ATo709H3(AH3 c){AH3 j=AH3(0.018*4.5,4.5,0.45);AH2 k=AH2(1.099,-0.099); + return clamp(j.xxx,c*j.yyy,pow(c,j.zzz)*k.xxx+k.yyy);} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AToGammaH1(AH1 c,AH1 rcpX){return pow(c,AH1_(rcpX));} + AH2 AToGammaH2(AH2 c,AH1 rcpX){return pow(c,AH2_(rcpX));} + AH3 AToGammaH3(AH3 c,AH1 rcpX){return pow(c,AH3_(rcpX));} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AToSrgbH1(AH1 c){AH3 j=AH3(0.0031308*12.92,12.92,1.0/2.4);AH2 k=AH2(1.055,-0.055); + return clamp(j.x ,c*j.y ,pow(c,j.z )*k.x +k.y );} + AH2 AToSrgbH2(AH2 c){AH3 j=AH3(0.0031308*12.92,12.92,1.0/2.4);AH2 k=AH2(1.055,-0.055); + return clamp(j.xx ,c*j.yy ,pow(c,j.zz )*k.xx +k.yy );} + AH3 AToSrgbH3(AH3 c){AH3 j=AH3(0.0031308*12.92,12.92,1.0/2.4);AH2 k=AH2(1.055,-0.055); + return clamp(j.xxx,c*j.yyy,pow(c,j.zzz)*k.xxx+k.yyy);} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AToTwoH1(AH1 c){return sqrt(c);} + AH2 AToTwoH2(AH2 c){return sqrt(c);} + AH3 AToTwoH3(AH3 c){return sqrt(c);} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AToThreeF1(AH1 c){return pow(c,AH1_(1.0/3.0));} + AH2 AToThreeF2(AH2 c){return pow(c,AH2_(1.0/3.0));} + AH3 AToThreeF3(AH3 c){return pow(c,AH3_(1.0/3.0));} + #endif +//============================================================================================================================== + #ifdef A_HALF + AH1 AFrom709H1(AH1 c){AH3 j=AH3(0.081/4.5,1.0/4.5,1.0/0.45);AH2 k=AH2(1.0/1.099,0.099/1.099); + return AZolSelH1(AZolSignedH1(c-j.x ),c*j.y ,pow(c*k.x +k.y ,j.z ));} + AH2 AFrom709H2(AH2 c){AH3 j=AH3(0.081/4.5,1.0/4.5,1.0/0.45);AH2 k=AH2(1.0/1.099,0.099/1.099); + return AZolSelH2(AZolSignedH2(c-j.xx ),c*j.yy ,pow(c*k.xx +k.yy ,j.zz ));} + AH3 AFrom709H3(AH3 c){AH3 j=AH3(0.081/4.5,1.0/4.5,1.0/0.45);AH2 k=AH2(1.0/1.099,0.099/1.099); + return AZolSelH3(AZolSignedH3(c-j.xxx),c*j.yyy,pow(c*k.xxx+k.yyy,j.zzz));} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AFromGammaH1(AH1 c,AH1 x){return pow(c,AH1_(x));} + AH2 AFromGammaH2(AH2 c,AH1 x){return pow(c,AH2_(x));} + AH3 AFromGammaH3(AH3 c,AH1 x){return pow(c,AH3_(x));} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AHromSrgbF1(AH1 c){AH3 j=AH3(0.04045/12.92,1.0/12.92,2.4);AH2 k=AH2(1.0/1.055,0.055/1.055); + return AZolSelH1(AZolSignedH1(c-j.x ),c*j.y ,pow(c*k.x +k.y ,j.z ));} + AH2 AHromSrgbF2(AH2 c){AH3 j=AH3(0.04045/12.92,1.0/12.92,2.4);AH2 k=AH2(1.0/1.055,0.055/1.055); + return AZolSelH2(AZolSignedH2(c-j.xx ),c*j.yy ,pow(c*k.xx +k.yy ,j.zz ));} + AH3 AHromSrgbF3(AH3 c){AH3 j=AH3(0.04045/12.92,1.0/12.92,2.4);AH2 k=AH2(1.0/1.055,0.055/1.055); + return AZolSelH3(AZolSignedH3(c-j.xxx),c*j.yyy,pow(c*k.xxx+k.yyy,j.zzz));} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AFromTwoH1(AH1 c){return c*c;} + AH2 AFromTwoH2(AH2 c){return c*c;} + AH3 AFromTwoH3(AH3 c){return c*c;} +//------------------------------------------------------------------------------------------------------------------------------ + AH1 AFromThreeH1(AH1 c){return c*c*c;} + AH2 AFromThreeH2(AH2 c){return c*c*c;} + AH3 AFromThreeH3(AH3 c){return c*c*c;} + #endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// CS REMAP +//============================================================================================================================== + // Simple remap 64x1 to 8x8 with rotated 2x2 pixel quads in quad linear. + // 543210 + // ====== + // ..xxx. + // yy...y + AU2 ARmp8x8(AU1 a){return AU2(ABfe(a,1u,3u),ABfiM(ABfe(a,3u,3u),a,1u));} +//============================================================================================================================== + // More complex remap 64x1 to 8x8 which is necessary for 2D wave reductions. + // 543210 + // ====== + // .xx..x + // y..yy. + // Details, + // LANE TO 8x8 MAPPING + // =================== + // 00 01 08 09 10 11 18 19 + // 02 03 0a 0b 12 13 1a 1b + // 04 05 0c 0d 14 15 1c 1d + // 06 07 0e 0f 16 17 1e 1f + // 20 21 28 29 30 31 38 39 + // 22 23 2a 2b 32 33 3a 3b + // 24 25 2c 2d 34 35 3c 3d + // 26 27 2e 2f 36 37 3e 3f + AU2 ARmpRed8x8(AU1 a){return AU2(ABfiM(ABfe(a,2u,3u),a,1u),ABfiM(ABfe(a,3u,3u),ABfe(a,1u,2u),2u));} +//============================================================================================================================== + #ifdef A_HALF + AW2 ARmp8x8H(AU1 a){return AW2(ABfe(a,1u,3u),ABfiM(ABfe(a,3u,3u),a,1u));} + AW2 ARmpRed8x8H(AU1 a){return AW2(ABfiM(ABfe(a,2u,3u),a,1u),ABfiM(ABfe(a,3u,3u),ABfe(a,1u,2u),2u));} + #endif +#endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// REFERENCE +// +//------------------------------------------------------------------------------------------------------------------------------ +// IEEE FLOAT RULES +// ================ +// - saturate(NaN)=0, saturate(-INF)=0, saturate(+INF)=1 +// - {+/-}0 * {+/-}INF = NaN +// - -INF + (+INF) = NaN +// - {+/-}0 / {+/-}0 = NaN +// - {+/-}INF / {+/-}INF = NaN +// - a<(-0) := sqrt(a) = NaN (a=-0.0 won't NaN) +// - 0 == -0 +// - 4/0 = +INF +// - 4/-0 = -INF +// - 4+INF = +INF +// - 4-INF = -INF +// - 4*(+INF) = +INF +// - 4*(-INF) = -INF +// - -4*(+INF) = -INF +// - sqrt(+INF) = +INF +//------------------------------------------------------------------------------------------------------------------------------ +// FP16 ENCODING +// ============= +// fedcba9876543210 +// ---------------- +// ......mmmmmmmmmm 10-bit mantissa (encodes 11-bit 0.5 to 1.0 except for denormals) +// .eeeee.......... 5-bit exponent +// .00000.......... denormals +// .00001.......... -14 exponent +// .11110.......... 15 exponent +// .111110000000000 infinity +// .11111nnnnnnnnnn NaN with n!=0 +// s............... sign +//------------------------------------------------------------------------------------------------------------------------------ +// FP16/INT16 ALIASING DENORMAL +// ============================ +// 11-bit unsigned integers alias with half float denormal/normal values, +// 1 = 2^(-24) = 1/16777216 ....................... first denormal value +// 2 = 2^(-23) +// ... +// 1023 = 2^(-14)*(1-2^(-10)) = 2^(-14)*(1-1/1024) ... last denormal value +// 1024 = 2^(-14) = 1/16384 .......................... first normal value that still maps to integers +// 2047 .............................................. last normal value that still maps to integers +// Scaling limits, +// 2^15 = 32768 ...................................... largest power of 2 scaling +// Largest pow2 conversion mapping is at *32768, +// 1 : 2^(-9) = 1/512 +// 2 : 1/256 +// 4 : 1/128 +// 8 : 1/64 +// 16 : 1/32 +// 32 : 1/16 +// 64 : 1/8 +// 128 : 1/4 +// 256 : 1/2 +// 512 : 1 +// 1024 : 2 +// 2047 : a little less than 4 +//============================================================================================================================== +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// +// GPU/CPU PORTABILITY +// +// +//------------------------------------------------------------------------------------------------------------------------------ +// This is the GPU implementation. +// See the CPU implementation for docs. +//============================================================================================================================== +#ifdef A_GPU + #define A_TRUE true + #define A_FALSE false + #define A_STATIC +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// VECTOR ARGUMENT/RETURN/INITIALIZATION PORTABILITY +//============================================================================================================================== + #define retAD2 AD2 + #define retAD3 AD3 + #define retAD4 AD4 + #define retAF2 AF2 + #define retAF3 AF3 + #define retAF4 AF4 + #define retAL2 AL2 + #define retAL3 AL3 + #define retAL4 AL4 + #define retAU2 AU2 + #define retAU3 AU3 + #define retAU4 AU4 +//------------------------------------------------------------------------------------------------------------------------------ + #define inAD2 in AD2 + #define inAD3 in AD3 + #define inAD4 in AD4 + #define inAF2 in AF2 + #define inAF3 in AF3 + #define inAF4 in AF4 + #define inAL2 in AL2 + #define inAL3 in AL3 + #define inAL4 in AL4 + #define inAU2 in AU2 + #define inAU3 in AU3 + #define inAU4 in AU4 +//------------------------------------------------------------------------------------------------------------------------------ + #define inoutAD2 inout AD2 + #define inoutAD3 inout AD3 + #define inoutAD4 inout AD4 + #define inoutAF2 inout AF2 + #define inoutAF3 inout AF3 + #define inoutAF4 inout AF4 + #define inoutAL2 inout AL2 + #define inoutAL3 inout AL3 + #define inoutAL4 inout AL4 + #define inoutAU2 inout AU2 + #define inoutAU3 inout AU3 + #define inoutAU4 inout AU4 +//------------------------------------------------------------------------------------------------------------------------------ + #define outAD2 out AD2 + #define outAD3 out AD3 + #define outAD4 out AD4 + #define outAF2 out AF2 + #define outAF3 out AF3 + #define outAF4 out AF4 + #define outAL2 out AL2 + #define outAL3 out AL3 + #define outAL4 out AL4 + #define outAU2 out AU2 + #define outAU3 out AU3 + #define outAU4 out AU4 +//------------------------------------------------------------------------------------------------------------------------------ + #define varAD2(x) AD2 x + #define varAD3(x) AD3 x + #define varAD4(x) AD4 x + #define varAF2(x) AF2 x + #define varAF3(x) AF3 x + #define varAF4(x) AF4 x + #define varAL2(x) AL2 x + #define varAL3(x) AL3 x + #define varAL4(x) AL4 x + #define varAU2(x) AU2 x + #define varAU3(x) AU3 x + #define varAU4(x) AU4 x +//------------------------------------------------------------------------------------------------------------------------------ + #define initAD2(x,y) AD2(x,y) + #define initAD3(x,y,z) AD3(x,y,z) + #define initAD4(x,y,z,w) AD4(x,y,z,w) + #define initAF2(x,y) AF2(x,y) + #define initAF3(x,y,z) AF3(x,y,z) + #define initAF4(x,y,z,w) AF4(x,y,z,w) + #define initAL2(x,y) AL2(x,y) + #define initAL3(x,y,z) AL3(x,y,z) + #define initAL4(x,y,z,w) AL4(x,y,z,w) + #define initAU2(x,y) AU2(x,y) + #define initAU3(x,y,z) AU3(x,y,z) + #define initAU4(x,y,z,w) AU4(x,y,z,w) +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// SCALAR RETURN OPS +//============================================================================================================================== + #define AAbsD1(a) abs(AD1(a)) + #define AAbsF1(a) abs(AF1(a)) +//------------------------------------------------------------------------------------------------------------------------------ + #define ACosD1(a) cos(AD1(a)) + #define ACosF1(a) cos(AF1(a)) +//------------------------------------------------------------------------------------------------------------------------------ + #define ADotD2(a,b) dot(AD2(a),AD2(b)) + #define ADotD3(a,b) dot(AD3(a),AD3(b)) + #define ADotD4(a,b) dot(AD4(a),AD4(b)) + #define ADotF2(a,b) dot(AF2(a),AF2(b)) + #define ADotF3(a,b) dot(AF3(a),AF3(b)) + #define ADotF4(a,b) dot(AF4(a),AF4(b)) +//------------------------------------------------------------------------------------------------------------------------------ + #define AExp2D1(a) exp2(AD1(a)) + #define AExp2F1(a) exp2(AF1(a)) +//------------------------------------------------------------------------------------------------------------------------------ + #define AFloorD1(a) floor(AD1(a)) + #define AFloorF1(a) floor(AF1(a)) +//------------------------------------------------------------------------------------------------------------------------------ + #define ALog2D1(a) log2(AD1(a)) + #define ALog2F1(a) log2(AF1(a)) +//------------------------------------------------------------------------------------------------------------------------------ + #define AMaxD1(a,b) max(a,b) + #define AMaxF1(a,b) max(a,b) + #define AMaxL1(a,b) max(a,b) + #define AMaxU1(a,b) max(a,b) +//------------------------------------------------------------------------------------------------------------------------------ + #define AMinD1(a,b) min(a,b) + #define AMinF1(a,b) min(a,b) + #define AMinL1(a,b) min(a,b) + #define AMinU1(a,b) min(a,b) +//------------------------------------------------------------------------------------------------------------------------------ + #define ASinD1(a) sin(AD1(a)) + #define ASinF1(a) sin(AF1(a)) +//------------------------------------------------------------------------------------------------------------------------------ + #define ASqrtD1(a) sqrt(AD1(a)) + #define ASqrtF1(a) sqrt(AF1(a)) +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// SCALAR RETURN OPS - DEPENDENT +//============================================================================================================================== + #define APowD1(a,b) pow(AD1(a),AF1(b)) + #define APowF1(a,b) pow(AF1(a),AF1(b)) +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// VECTOR OPS +//------------------------------------------------------------------------------------------------------------------------------ +// These are added as needed for production or prototyping, so not necessarily a complete set. +// They follow a convention of taking in a destination and also returning the destination value to increase utility. +//============================================================================================================================== + #ifdef A_DUBL + AD2 opAAbsD2(outAD2 d,inAD2 a){d=abs(a);return d;} + AD3 opAAbsD3(outAD3 d,inAD3 a){d=abs(a);return d;} + AD4 opAAbsD4(outAD4 d,inAD4 a){d=abs(a);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AD2 opAAddD2(outAD2 d,inAD2 a,inAD2 b){d=a+b;return d;} + AD3 opAAddD3(outAD3 d,inAD3 a,inAD3 b){d=a+b;return d;} + AD4 opAAddD4(outAD4 d,inAD4 a,inAD4 b){d=a+b;return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AD2 opAAddOneD2(outAD2 d,inAD2 a,AD1 b){d=a+AD2_(b);return d;} + AD3 opAAddOneD3(outAD3 d,inAD3 a,AD1 b){d=a+AD3_(b);return d;} + AD4 opAAddOneD4(outAD4 d,inAD4 a,AD1 b){d=a+AD4_(b);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AD2 opACpyD2(outAD2 d,inAD2 a){d=a;return d;} + AD3 opACpyD3(outAD3 d,inAD3 a){d=a;return d;} + AD4 opACpyD4(outAD4 d,inAD4 a){d=a;return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AD2 opALerpD2(outAD2 d,inAD2 a,inAD2 b,inAD2 c){d=ALerpD2(a,b,c);return d;} + AD3 opALerpD3(outAD3 d,inAD3 a,inAD3 b,inAD3 c){d=ALerpD3(a,b,c);return d;} + AD4 opALerpD4(outAD4 d,inAD4 a,inAD4 b,inAD4 c){d=ALerpD4(a,b,c);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AD2 opALerpOneD2(outAD2 d,inAD2 a,inAD2 b,AD1 c){d=ALerpD2(a,b,AD2_(c));return d;} + AD3 opALerpOneD3(outAD3 d,inAD3 a,inAD3 b,AD1 c){d=ALerpD3(a,b,AD3_(c));return d;} + AD4 opALerpOneD4(outAD4 d,inAD4 a,inAD4 b,AD1 c){d=ALerpD4(a,b,AD4_(c));return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AD2 opAMaxD2(outAD2 d,inAD2 a,inAD2 b){d=max(a,b);return d;} + AD3 opAMaxD3(outAD3 d,inAD3 a,inAD3 b){d=max(a,b);return d;} + AD4 opAMaxD4(outAD4 d,inAD4 a,inAD4 b){d=max(a,b);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AD2 opAMinD2(outAD2 d,inAD2 a,inAD2 b){d=min(a,b);return d;} + AD3 opAMinD3(outAD3 d,inAD3 a,inAD3 b){d=min(a,b);return d;} + AD4 opAMinD4(outAD4 d,inAD4 a,inAD4 b){d=min(a,b);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AD2 opAMulD2(outAD2 d,inAD2 a,inAD2 b){d=a*b;return d;} + AD3 opAMulD3(outAD3 d,inAD3 a,inAD3 b){d=a*b;return d;} + AD4 opAMulD4(outAD4 d,inAD4 a,inAD4 b){d=a*b;return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AD2 opAMulOneD2(outAD2 d,inAD2 a,AD1 b){d=a*AD2_(b);return d;} + AD3 opAMulOneD3(outAD3 d,inAD3 a,AD1 b){d=a*AD3_(b);return d;} + AD4 opAMulOneD4(outAD4 d,inAD4 a,AD1 b){d=a*AD4_(b);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AD2 opANegD2(outAD2 d,inAD2 a){d=-a;return d;} + AD3 opANegD3(outAD3 d,inAD3 a){d=-a;return d;} + AD4 opANegD4(outAD4 d,inAD4 a){d=-a;return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AD2 opARcpD2(outAD2 d,inAD2 a){d=ARcpD2(a);return d;} + AD3 opARcpD3(outAD3 d,inAD3 a){d=ARcpD3(a);return d;} + AD4 opARcpD4(outAD4 d,inAD4 a){d=ARcpD4(a);return d;} + #endif +//============================================================================================================================== + AF2 opAAbsF2(outAF2 d,inAF2 a){d=abs(a);return d;} + AF3 opAAbsF3(outAF3 d,inAF3 a){d=abs(a);return d;} + AF4 opAAbsF4(outAF4 d,inAF4 a){d=abs(a);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AF2 opAAddF2(outAF2 d,inAF2 a,inAF2 b){d=a+b;return d;} + AF3 opAAddF3(outAF3 d,inAF3 a,inAF3 b){d=a+b;return d;} + AF4 opAAddF4(outAF4 d,inAF4 a,inAF4 b){d=a+b;return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AF2 opAAddOneF2(outAF2 d,inAF2 a,AF1 b){d=a+AF2_(b);return d;} + AF3 opAAddOneF3(outAF3 d,inAF3 a,AF1 b){d=a+AF3_(b);return d;} + AF4 opAAddOneF4(outAF4 d,inAF4 a,AF1 b){d=a+AF4_(b);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AF2 opACpyF2(outAF2 d,inAF2 a){d=a;return d;} + AF3 opACpyF3(outAF3 d,inAF3 a){d=a;return d;} + AF4 opACpyF4(outAF4 d,inAF4 a){d=a;return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AF2 opALerpF2(outAF2 d,inAF2 a,inAF2 b,inAF2 c){d=ALerpF2(a,b,c);return d;} + AF3 opALerpF3(outAF3 d,inAF3 a,inAF3 b,inAF3 c){d=ALerpF3(a,b,c);return d;} + AF4 opALerpF4(outAF4 d,inAF4 a,inAF4 b,inAF4 c){d=ALerpF4(a,b,c);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AF2 opALerpOneF2(outAF2 d,inAF2 a,inAF2 b,AF1 c){d=ALerpF2(a,b,AF2_(c));return d;} + AF3 opALerpOneF3(outAF3 d,inAF3 a,inAF3 b,AF1 c){d=ALerpF3(a,b,AF3_(c));return d;} + AF4 opALerpOneF4(outAF4 d,inAF4 a,inAF4 b,AF1 c){d=ALerpF4(a,b,AF4_(c));return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AF2 opAMaxF2(outAF2 d,inAF2 a,inAF2 b){d=max(a,b);return d;} + AF3 opAMaxF3(outAF3 d,inAF3 a,inAF3 b){d=max(a,b);return d;} + AF4 opAMaxF4(outAF4 d,inAF4 a,inAF4 b){d=max(a,b);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AF2 opAMinF2(outAF2 d,inAF2 a,inAF2 b){d=min(a,b);return d;} + AF3 opAMinF3(outAF3 d,inAF3 a,inAF3 b){d=min(a,b);return d;} + AF4 opAMinF4(outAF4 d,inAF4 a,inAF4 b){d=min(a,b);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AF2 opAMulF2(outAF2 d,inAF2 a,inAF2 b){d=a*b;return d;} + AF3 opAMulF3(outAF3 d,inAF3 a,inAF3 b){d=a*b;return d;} + AF4 opAMulF4(outAF4 d,inAF4 a,inAF4 b){d=a*b;return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AF2 opAMulOneF2(outAF2 d,inAF2 a,AF1 b){d=a*AF2_(b);return d;} + AF3 opAMulOneF3(outAF3 d,inAF3 a,AF1 b){d=a*AF3_(b);return d;} + AF4 opAMulOneF4(outAF4 d,inAF4 a,AF1 b){d=a*AF4_(b);return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AF2 opANegF2(outAF2 d,inAF2 a){d=-a;return d;} + AF3 opANegF3(outAF3 d,inAF3 a){d=-a;return d;} + AF4 opANegF4(outAF4 d,inAF4 a){d=-a;return d;} +//------------------------------------------------------------------------------------------------------------------------------ + AF2 opARcpF2(outAF2 d,inAF2 a){d=ARcpF2(a);return d;} + AF3 opARcpF3(outAF3 d,inAF3 a){d=ARcpF3(a);return d;} + AF4 opARcpF4(outAF4 d,inAF4 a){d=ARcpF4(a);return d;} +#endif diff --git a/externals/FidelityFX-FSR/ffx-fsr/ffx_fsr1.h b/externals/FidelityFX-FSR/ffx-fsr/ffx_fsr1.h new file mode 100644 index 000000000..15ecfde5c --- /dev/null +++ b/externals/FidelityFX-FSR/ffx-fsr/ffx_fsr1.h @@ -0,0 +1,1199 @@ +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// +// AMD FidelityFX SUPER RESOLUTION [FSR 1] ::: SPATIAL SCALING & EXTRAS - v1.20210629 +// +// +//------------------------------------------------------------------------------------------------------------------------------ +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//------------------------------------------------------------------------------------------------------------------------------ +// FidelityFX Super Resolution Sample +// +// Copyright (c) 2021 Advanced Micro Devices, Inc. All rights reserved. +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files(the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and / or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions : +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. +//------------------------------------------------------------------------------------------------------------------------------ +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//------------------------------------------------------------------------------------------------------------------------------ +// ABOUT +// ===== +// FSR is a collection of algorithms relating to generating a higher resolution image. +// This specific header focuses on single-image non-temporal image scaling, and related tools. +// +// The core functions are EASU and RCAS: +// [EASU] Edge Adaptive Spatial Upsampling ....... 1x to 4x area range spatial scaling, clamped adaptive elliptical filter. +// [RCAS] Robust Contrast Adaptive Sharpening .... A non-scaling variation on CAS. +// RCAS needs to be applied after EASU as a separate pass. +// +// Optional utility functions are: +// [LFGA] Linear Film Grain Applicator ........... Tool to apply film grain after scaling. +// [SRTM] Simple Reversible Tone-Mapper .......... Linear HDR {0 to FP16_MAX} to {0 to 1} and back. +// [TEPD] Temporal Energy Preserving Dither ...... Temporally energy preserving dithered {0 to 1} linear to gamma 2.0 conversion. +// See each individual sub-section for inline documentation. +//------------------------------------------------------------------------------------------------------------------------------ +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//------------------------------------------------------------------------------------------------------------------------------ +// FUNCTION PERMUTATIONS +// ===================== +// *F() ..... Single item computation with 32-bit. +// *H() ..... Single item computation with 16-bit, with packing (aka two 16-bit ops in parallel) when possible. +// *Hx2() ... Processing two items in parallel with 16-bit, easier packing. +// Not all interfaces in this file have a *Hx2() form. +//============================================================================================================================== +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// FSR - [EASU] EDGE ADAPTIVE SPATIAL UPSAMPLING +// +//------------------------------------------------------------------------------------------------------------------------------ +// EASU provides a high quality spatial-only scaling at relatively low cost. +// Meaning EASU is appropiate for laptops and other low-end GPUs. +// Quality from 1x to 4x area scaling is good. +//------------------------------------------------------------------------------------------------------------------------------ +// The scalar uses a modified fast approximation to the standard lanczos(size=2) kernel. +// EASU runs in a single pass, so it applies a directionally and anisotropically adaptive radial lanczos. +// This is also kept as simple as possible to have minimum runtime. +//------------------------------------------------------------------------------------------------------------------------------ +// The lanzcos filter has negative lobes, so by itself it will introduce ringing. +// To remove all ringing, the algorithm uses the nearest 2x2 input texels as a neighborhood, +// and limits output to the minimum and maximum of that neighborhood. +//------------------------------------------------------------------------------------------------------------------------------ +// Input image requirements: +// +// Color needs to be encoded as 3 channel[red, green, blue](e.g.XYZ not supported) +// Each channel needs to be in the range[0, 1] +// Any color primaries are supported +// Display / tonemapping curve needs to be as if presenting to sRGB display or similar(e.g.Gamma 2.0) +// There should be no banding in the input +// There should be no high amplitude noise in the input +// There should be no noise in the input that is not at input pixel granularity +// For performance purposes, use 32bpp formats +//------------------------------------------------------------------------------------------------------------------------------ +// Best to apply EASU at the end of the frame after tonemapping +// but before film grain or composite of the UI. +//------------------------------------------------------------------------------------------------------------------------------ +// Example of including this header for D3D HLSL : +// +// #define A_GPU 1 +// #define A_HLSL 1 +// #define A_HALF 1 +// #include "ffx_a.h" +// #define FSR_EASU_H 1 +// #define FSR_RCAS_H 1 +// //declare input callbacks +// #include "ffx_fsr1.h" +// +// Example of including this header for Vulkan GLSL : +// +// #define A_GPU 1 +// #define A_GLSL 1 +// #define A_HALF 1 +// #include "ffx_a.h" +// #define FSR_EASU_H 1 +// #define FSR_RCAS_H 1 +// //declare input callbacks +// #include "ffx_fsr1.h" +// +// Example of including this header for Vulkan HLSL : +// +// #define A_GPU 1 +// #define A_HLSL 1 +// #define A_HLSL_6_2 1 +// #define A_NO_16_BIT_CAST 1 +// #define A_HALF 1 +// #include "ffx_a.h" +// #define FSR_EASU_H 1 +// #define FSR_RCAS_H 1 +// //declare input callbacks +// #include "ffx_fsr1.h" +// +// Example of declaring the required input callbacks for GLSL : +// The callbacks need to gather4 for each color channel using the specified texture coordinate 'p'. +// EASU uses gather4 to reduce position computation logic and for free Arrays of Structures to Structures of Arrays conversion. +// +// AH4 FsrEasuRH(AF2 p){return AH4(textureGather(sampler2D(tex,sam),p,0));} +// AH4 FsrEasuGH(AF2 p){return AH4(textureGather(sampler2D(tex,sam),p,1));} +// AH4 FsrEasuBH(AF2 p){return AH4(textureGather(sampler2D(tex,sam),p,2));} +// ... +// The FsrEasuCon function needs to be called from the CPU or GPU to set up constants. +// The difference in viewport and input image size is there to support Dynamic Resolution Scaling. +// To use FsrEasuCon() on the CPU, define A_CPU before including ffx_a and ffx_fsr1. +// Including a GPU example here, the 'con0' through 'con3' values would be stored out to a constant buffer. +// AU4 con0,con1,con2,con3; +// FsrEasuCon(con0,con1,con2,con3, +// 1920.0,1080.0, // Viewport size (top left aligned) in the input image which is to be scaled. +// 3840.0,2160.0, // The size of the input image. +// 2560.0,1440.0); // The output resolution. +//============================================================================================================================== +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// CONSTANT SETUP +//============================================================================================================================== +// Call to setup required constant values (works on CPU or GPU). +A_STATIC void FsrEasuCon( +outAU4 con0, +outAU4 con1, +outAU4 con2, +outAU4 con3, +// This the rendered image resolution being upscaled +AF1 inputViewportInPixelsX, +AF1 inputViewportInPixelsY, +// This is the resolution of the resource containing the input image (useful for dynamic resolution) +AF1 inputSizeInPixelsX, +AF1 inputSizeInPixelsY, +// This is the display resolution which the input image gets upscaled to +AF1 outputSizeInPixelsX, +AF1 outputSizeInPixelsY){ + // Output integer position to a pixel position in viewport. + con0[0]=AU1_AF1(inputViewportInPixelsX*ARcpF1(outputSizeInPixelsX)); + con0[1]=AU1_AF1(inputViewportInPixelsY*ARcpF1(outputSizeInPixelsY)); + con0[2]=AU1_AF1(AF1_(0.5)*inputViewportInPixelsX*ARcpF1(outputSizeInPixelsX)-AF1_(0.5)); + con0[3]=AU1_AF1(AF1_(0.5)*inputViewportInPixelsY*ARcpF1(outputSizeInPixelsY)-AF1_(0.5)); + // Viewport pixel position to normalized image space. + // This is used to get upper-left of 'F' tap. + con1[0]=AU1_AF1(ARcpF1(inputSizeInPixelsX)); + con1[1]=AU1_AF1(ARcpF1(inputSizeInPixelsY)); + // Centers of gather4, first offset from upper-left of 'F'. + // +---+---+ + // | | | + // +--(0)--+ + // | b | c | + // +---F---+---+---+ + // | e | f | g | h | + // +--(1)--+--(2)--+ + // | i | j | k | l | + // +---+---+---+---+ + // | n | o | + // +--(3)--+ + // | | | + // +---+---+ + con1[2]=AU1_AF1(AF1_( 1.0)*ARcpF1(inputSizeInPixelsX)); + con1[3]=AU1_AF1(AF1_(-1.0)*ARcpF1(inputSizeInPixelsY)); + // These are from (0) instead of 'F'. + con2[0]=AU1_AF1(AF1_(-1.0)*ARcpF1(inputSizeInPixelsX)); + con2[1]=AU1_AF1(AF1_( 2.0)*ARcpF1(inputSizeInPixelsY)); + con2[2]=AU1_AF1(AF1_( 1.0)*ARcpF1(inputSizeInPixelsX)); + con2[3]=AU1_AF1(AF1_( 2.0)*ARcpF1(inputSizeInPixelsY)); + con3[0]=AU1_AF1(AF1_( 0.0)*ARcpF1(inputSizeInPixelsX)); + con3[1]=AU1_AF1(AF1_( 4.0)*ARcpF1(inputSizeInPixelsY)); + con3[2]=con3[3]=0;} + +//If the an offset into the input image resource +A_STATIC void FsrEasuConOffset( + outAU4 con0, + outAU4 con1, + outAU4 con2, + outAU4 con3, + // This the rendered image resolution being upscaled + AF1 inputViewportInPixelsX, + AF1 inputViewportInPixelsY, + // This is the resolution of the resource containing the input image (useful for dynamic resolution) + AF1 inputSizeInPixelsX, + AF1 inputSizeInPixelsY, + // This is the display resolution which the input image gets upscaled to + AF1 outputSizeInPixelsX, + AF1 outputSizeInPixelsY, + // This is the input image offset into the resource containing it (useful for dynamic resolution) + AF1 inputOffsetInPixelsX, + AF1 inputOffsetInPixelsY) { + FsrEasuCon(con0, con1, con2, con3, inputViewportInPixelsX, inputViewportInPixelsY, inputSizeInPixelsX, inputSizeInPixelsY, outputSizeInPixelsX, outputSizeInPixelsY); + con0[2] = AU1_AF1(AF1_(0.5) * inputViewportInPixelsX * ARcpF1(outputSizeInPixelsX) - AF1_(0.5) + inputOffsetInPixelsX); + con0[3] = AU1_AF1(AF1_(0.5) * inputViewportInPixelsY * ARcpF1(outputSizeInPixelsY) - AF1_(0.5) + inputOffsetInPixelsY); +} +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// NON-PACKED 32-BIT VERSION +//============================================================================================================================== +#if defined(A_GPU)&&defined(FSR_EASU_F) + // Input callback prototypes, need to be implemented by calling shader + AF4 FsrEasuRF(AF2 p); + AF4 FsrEasuGF(AF2 p); + AF4 FsrEasuBF(AF2 p); +//------------------------------------------------------------------------------------------------------------------------------ + // Filtering for a given tap for the scalar. + void FsrEasuTapF( + inout AF3 aC, // Accumulated color, with negative lobe. + inout AF1 aW, // Accumulated weight. + AF2 off, // Pixel offset from resolve position to tap. + AF2 dir, // Gradient direction. + AF2 len, // Length. + AF1 lob, // Negative lobe strength. + AF1 clp, // Clipping point. + AF3 c){ // Tap color. + // Rotate offset by direction. + AF2 v; + v.x=(off.x*( dir.x))+(off.y*dir.y); + v.y=(off.x*(-dir.y))+(off.y*dir.x); + // Anisotropy. + v*=len; + // Compute distance^2. + AF1 d2=v.x*v.x+v.y*v.y; + // Limit to the window as at corner, 2 taps can easily be outside. + d2=min(d2,clp); + // Approximation of lancos2 without sin() or rcp(), or sqrt() to get x. + // (25/16 * (2/5 * x^2 - 1)^2 - (25/16 - 1)) * (1/4 * x^2 - 1)^2 + // |_______________________________________| |_______________| + // base window + // The general form of the 'base' is, + // (a*(b*x^2-1)^2-(a-1)) + // Where 'a=1/(2*b-b^2)' and 'b' moves around the negative lobe. + AF1 wB=AF1_(2.0/5.0)*d2+AF1_(-1.0); + AF1 wA=lob*d2+AF1_(-1.0); + wB*=wB; + wA*=wA; + wB=AF1_(25.0/16.0)*wB+AF1_(-(25.0/16.0-1.0)); + AF1 w=wB*wA; + // Do weighted average. + aC+=c*w;aW+=w;} +//------------------------------------------------------------------------------------------------------------------------------ + // Accumulate direction and length. + void FsrEasuSetF( + inout AF2 dir, + inout AF1 len, + AF2 pp, + AP1 biS,AP1 biT,AP1 biU,AP1 biV, + AF1 lA,AF1 lB,AF1 lC,AF1 lD,AF1 lE){ + // Compute bilinear weight, branches factor out as predicates are compiler time immediates. + // s t + // u v + AF1 w = AF1_(0.0); + if(biS)w=(AF1_(1.0)-pp.x)*(AF1_(1.0)-pp.y); + if(biT)w= pp.x *(AF1_(1.0)-pp.y); + if(biU)w=(AF1_(1.0)-pp.x)* pp.y ; + if(biV)w= pp.x * pp.y ; + // Direction is the '+' diff. + // a + // b c d + // e + // Then takes magnitude from abs average of both sides of 'c'. + // Length converts gradient reversal to 0, smoothly to non-reversal at 1, shaped, then adding horz and vert terms. + AF1 dc=lD-lC; + AF1 cb=lC-lB; + AF1 lenX=max(abs(dc),abs(cb)); + lenX=APrxLoRcpF1(lenX); + AF1 dirX=lD-lB; + dir.x+=dirX*w; + lenX=ASatF1(abs(dirX)*lenX); + lenX*=lenX; + len+=lenX*w; + // Repeat for the y axis. + AF1 ec=lE-lC; + AF1 ca=lC-lA; + AF1 lenY=max(abs(ec),abs(ca)); + lenY=APrxLoRcpF1(lenY); + AF1 dirY=lE-lA; + dir.y+=dirY*w; + lenY=ASatF1(abs(dirY)*lenY); + lenY*=lenY; + len+=lenY*w;} +//------------------------------------------------------------------------------------------------------------------------------ + void FsrEasuF( + out AF3 pix, + AU2 ip, // Integer pixel position in output. + AU4 con0, // Constants generated by FsrEasuCon(). + AU4 con1, + AU4 con2, + AU4 con3){ +//------------------------------------------------------------------------------------------------------------------------------ + // Get position of 'f'. + AF2 pp=AF2(ip)*AF2_AU2(con0.xy)+AF2_AU2(con0.zw); + AF2 fp=floor(pp); + pp-=fp; +//------------------------------------------------------------------------------------------------------------------------------ + // 12-tap kernel. + // b c + // e f g h + // i j k l + // n o + // Gather 4 ordering. + // a b + // r g + // For packed FP16, need either {rg} or {ab} so using the following setup for gather in all versions, + // a b <- unused (z) + // r g + // a b a b + // r g r g + // a b + // r g <- unused (z) + // Allowing dead-code removal to remove the 'z's. + AF2 p0=fp*AF2_AU2(con1.xy)+AF2_AU2(con1.zw); + // These are from p0 to avoid pulling two constants on pre-Navi hardware. + AF2 p1=p0+AF2_AU2(con2.xy); + AF2 p2=p0+AF2_AU2(con2.zw); + AF2 p3=p0+AF2_AU2(con3.xy); + AF4 bczzR=FsrEasuRF(p0); + AF4 bczzG=FsrEasuGF(p0); + AF4 bczzB=FsrEasuBF(p0); + AF4 ijfeR=FsrEasuRF(p1); + AF4 ijfeG=FsrEasuGF(p1); + AF4 ijfeB=FsrEasuBF(p1); + AF4 klhgR=FsrEasuRF(p2); + AF4 klhgG=FsrEasuGF(p2); + AF4 klhgB=FsrEasuBF(p2); + AF4 zzonR=FsrEasuRF(p3); + AF4 zzonG=FsrEasuGF(p3); + AF4 zzonB=FsrEasuBF(p3); +//------------------------------------------------------------------------------------------------------------------------------ + // Simplest multi-channel approximate luma possible (luma times 2, in 2 FMA/MAD). + AF4 bczzL=bczzB*AF4_(0.5)+(bczzR*AF4_(0.5)+bczzG); + AF4 ijfeL=ijfeB*AF4_(0.5)+(ijfeR*AF4_(0.5)+ijfeG); + AF4 klhgL=klhgB*AF4_(0.5)+(klhgR*AF4_(0.5)+klhgG); + AF4 zzonL=zzonB*AF4_(0.5)+(zzonR*AF4_(0.5)+zzonG); + // Rename. + AF1 bL=bczzL.x; + AF1 cL=bczzL.y; + AF1 iL=ijfeL.x; + AF1 jL=ijfeL.y; + AF1 fL=ijfeL.z; + AF1 eL=ijfeL.w; + AF1 kL=klhgL.x; + AF1 lL=klhgL.y; + AF1 hL=klhgL.z; + AF1 gL=klhgL.w; + AF1 oL=zzonL.z; + AF1 nL=zzonL.w; + // Accumulate for bilinear interpolation. + AF2 dir=AF2_(0.0); + AF1 len=AF1_(0.0); + FsrEasuSetF(dir,len,pp,true, false,false,false,bL,eL,fL,gL,jL); + FsrEasuSetF(dir,len,pp,false,true ,false,false,cL,fL,gL,hL,kL); + FsrEasuSetF(dir,len,pp,false,false,true ,false,fL,iL,jL,kL,nL); + FsrEasuSetF(dir,len,pp,false,false,false,true ,gL,jL,kL,lL,oL); +//------------------------------------------------------------------------------------------------------------------------------ + // Normalize with approximation, and cleanup close to zero. + AF2 dir2=dir*dir; + AF1 dirR=dir2.x+dir2.y; + AP1 zro=dirR<AF1_(1.0/32768.0); + dirR=APrxLoRsqF1(dirR); + dirR=zro?AF1_(1.0):dirR; + dir.x=zro?AF1_(1.0):dir.x; + dir*=AF2_(dirR); + // Transform from {0 to 2} to {0 to 1} range, and shape with square. + len=len*AF1_(0.5); + len*=len; + // Stretch kernel {1.0 vert|horz, to sqrt(2.0) on diagonal}. + AF1 stretch=(dir.x*dir.x+dir.y*dir.y)*APrxLoRcpF1(max(abs(dir.x),abs(dir.y))); + // Anisotropic length after rotation, + // x := 1.0 lerp to 'stretch' on edges + // y := 1.0 lerp to 2x on edges + AF2 len2=AF2(AF1_(1.0)+(stretch-AF1_(1.0))*len,AF1_(1.0)+AF1_(-0.5)*len); + // Based on the amount of 'edge', + // the window shifts from +/-{sqrt(2.0) to slightly beyond 2.0}. + AF1 lob=AF1_(0.5)+AF1_((1.0/4.0-0.04)-0.5)*len; + // Set distance^2 clipping point to the end of the adjustable window. + AF1 clp=APrxLoRcpF1(lob); +//------------------------------------------------------------------------------------------------------------------------------ + // Accumulation mixed with min/max of 4 nearest. + // b c + // e f g h + // i j k l + // n o + AF3 min4=min(AMin3F3(AF3(ijfeR.z,ijfeG.z,ijfeB.z),AF3(klhgR.w,klhgG.w,klhgB.w),AF3(ijfeR.y,ijfeG.y,ijfeB.y)), + AF3(klhgR.x,klhgG.x,klhgB.x)); + AF3 max4=max(AMax3F3(AF3(ijfeR.z,ijfeG.z,ijfeB.z),AF3(klhgR.w,klhgG.w,klhgB.w),AF3(ijfeR.y,ijfeG.y,ijfeB.y)), + AF3(klhgR.x,klhgG.x,klhgB.x)); + // Accumulation. + AF3 aC=AF3_(0.0); + AF1 aW=AF1_(0.0); + FsrEasuTapF(aC,aW,AF2( 0.0,-1.0)-pp,dir,len2,lob,clp,AF3(bczzR.x,bczzG.x,bczzB.x)); // b + FsrEasuTapF(aC,aW,AF2( 1.0,-1.0)-pp,dir,len2,lob,clp,AF3(bczzR.y,bczzG.y,bczzB.y)); // c + FsrEasuTapF(aC,aW,AF2(-1.0, 1.0)-pp,dir,len2,lob,clp,AF3(ijfeR.x,ijfeG.x,ijfeB.x)); // i + FsrEasuTapF(aC,aW,AF2( 0.0, 1.0)-pp,dir,len2,lob,clp,AF3(ijfeR.y,ijfeG.y,ijfeB.y)); // j + FsrEasuTapF(aC,aW,AF2( 0.0, 0.0)-pp,dir,len2,lob,clp,AF3(ijfeR.z,ijfeG.z,ijfeB.z)); // f + FsrEasuTapF(aC,aW,AF2(-1.0, 0.0)-pp,dir,len2,lob,clp,AF3(ijfeR.w,ijfeG.w,ijfeB.w)); // e + FsrEasuTapF(aC,aW,AF2( 1.0, 1.0)-pp,dir,len2,lob,clp,AF3(klhgR.x,klhgG.x,klhgB.x)); // k + FsrEasuTapF(aC,aW,AF2( 2.0, 1.0)-pp,dir,len2,lob,clp,AF3(klhgR.y,klhgG.y,klhgB.y)); // l + FsrEasuTapF(aC,aW,AF2( 2.0, 0.0)-pp,dir,len2,lob,clp,AF3(klhgR.z,klhgG.z,klhgB.z)); // h + FsrEasuTapF(aC,aW,AF2( 1.0, 0.0)-pp,dir,len2,lob,clp,AF3(klhgR.w,klhgG.w,klhgB.w)); // g + FsrEasuTapF(aC,aW,AF2( 1.0, 2.0)-pp,dir,len2,lob,clp,AF3(zzonR.z,zzonG.z,zzonB.z)); // o + FsrEasuTapF(aC,aW,AF2( 0.0, 2.0)-pp,dir,len2,lob,clp,AF3(zzonR.w,zzonG.w,zzonB.w)); // n +//------------------------------------------------------------------------------------------------------------------------------ + // Normalize and dering. + pix=min(max4,max(min4,aC*AF3_(ARcpF1(aW))));} +#endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// PACKED 16-BIT VERSION +//============================================================================================================================== +#if defined(A_GPU)&&defined(A_HALF)&&defined(FSR_EASU_H) +// Input callback prototypes, need to be implemented by calling shader + AH4 FsrEasuRH(AF2 p); + AH4 FsrEasuGH(AF2 p); + AH4 FsrEasuBH(AF2 p); +//------------------------------------------------------------------------------------------------------------------------------ + // This runs 2 taps in parallel. + void FsrEasuTapH( + inout AH2 aCR,inout AH2 aCG,inout AH2 aCB, + inout AH2 aW, + AH2 offX,AH2 offY, + AH2 dir, + AH2 len, + AH1 lob, + AH1 clp, + AH2 cR,AH2 cG,AH2 cB){ + AH2 vX,vY; + vX=offX* dir.xx +offY*dir.yy; + vY=offX*(-dir.yy)+offY*dir.xx; + vX*=len.x;vY*=len.y; + AH2 d2=vX*vX+vY*vY; + d2=min(d2,AH2_(clp)); + AH2 wB=AH2_(2.0/5.0)*d2+AH2_(-1.0); + AH2 wA=AH2_(lob)*d2+AH2_(-1.0); + wB*=wB; + wA*=wA; + wB=AH2_(25.0/16.0)*wB+AH2_(-(25.0/16.0-1.0)); + AH2 w=wB*wA; + aCR+=cR*w;aCG+=cG*w;aCB+=cB*w;aW+=w;} +//------------------------------------------------------------------------------------------------------------------------------ + // This runs 2 taps in parallel. + void FsrEasuSetH( + inout AH2 dirPX,inout AH2 dirPY, + inout AH2 lenP, + AH2 pp, + AP1 biST,AP1 biUV, + AH2 lA,AH2 lB,AH2 lC,AH2 lD,AH2 lE){ + AH2 w = AH2_(0.0); + if(biST)w=(AH2(1.0,0.0)+AH2(-pp.x,pp.x))*AH2_(AH1_(1.0)-pp.y); + if(biUV)w=(AH2(1.0,0.0)+AH2(-pp.x,pp.x))*AH2_( pp.y); + // ABS is not free in the packed FP16 path. + AH2 dc=lD-lC; + AH2 cb=lC-lB; + AH2 lenX=max(abs(dc),abs(cb)); + lenX=ARcpH2(lenX); + AH2 dirX=lD-lB; + dirPX+=dirX*w; + lenX=ASatH2(abs(dirX)*lenX); + lenX*=lenX; + lenP+=lenX*w; + AH2 ec=lE-lC; + AH2 ca=lC-lA; + AH2 lenY=max(abs(ec),abs(ca)); + lenY=ARcpH2(lenY); + AH2 dirY=lE-lA; + dirPY+=dirY*w; + lenY=ASatH2(abs(dirY)*lenY); + lenY*=lenY; + lenP+=lenY*w;} +//------------------------------------------------------------------------------------------------------------------------------ + void FsrEasuH( + out AH3 pix, + AU2 ip, + AU4 con0, + AU4 con1, + AU4 con2, + AU4 con3){ +//------------------------------------------------------------------------------------------------------------------------------ + AF2 pp=AF2(ip)*AF2_AU2(con0.xy)+AF2_AU2(con0.zw); + AF2 fp=floor(pp); + pp-=fp; + AH2 ppp=AH2(pp); +//------------------------------------------------------------------------------------------------------------------------------ + AF2 p0=fp*AF2_AU2(con1.xy)+AF2_AU2(con1.zw); + AF2 p1=p0+AF2_AU2(con2.xy); + AF2 p2=p0+AF2_AU2(con2.zw); + AF2 p3=p0+AF2_AU2(con3.xy); + AH4 bczzR=FsrEasuRH(p0); + AH4 bczzG=FsrEasuGH(p0); + AH4 bczzB=FsrEasuBH(p0); + AH4 ijfeR=FsrEasuRH(p1); + AH4 ijfeG=FsrEasuGH(p1); + AH4 ijfeB=FsrEasuBH(p1); + AH4 klhgR=FsrEasuRH(p2); + AH4 klhgG=FsrEasuGH(p2); + AH4 klhgB=FsrEasuBH(p2); + AH4 zzonR=FsrEasuRH(p3); + AH4 zzonG=FsrEasuGH(p3); + AH4 zzonB=FsrEasuBH(p3); +//------------------------------------------------------------------------------------------------------------------------------ + AH4 bczzL=bczzB*AH4_(0.5)+(bczzR*AH4_(0.5)+bczzG); + AH4 ijfeL=ijfeB*AH4_(0.5)+(ijfeR*AH4_(0.5)+ijfeG); + AH4 klhgL=klhgB*AH4_(0.5)+(klhgR*AH4_(0.5)+klhgG); + AH4 zzonL=zzonB*AH4_(0.5)+(zzonR*AH4_(0.5)+zzonG); + AH1 bL=bczzL.x; + AH1 cL=bczzL.y; + AH1 iL=ijfeL.x; + AH1 jL=ijfeL.y; + AH1 fL=ijfeL.z; + AH1 eL=ijfeL.w; + AH1 kL=klhgL.x; + AH1 lL=klhgL.y; + AH1 hL=klhgL.z; + AH1 gL=klhgL.w; + AH1 oL=zzonL.z; + AH1 nL=zzonL.w; + // This part is different, accumulating 2 taps in parallel. + AH2 dirPX=AH2_(0.0); + AH2 dirPY=AH2_(0.0); + AH2 lenP=AH2_(0.0); + FsrEasuSetH(dirPX,dirPY,lenP,ppp,true, false,AH2(bL,cL),AH2(eL,fL),AH2(fL,gL),AH2(gL,hL),AH2(jL,kL)); + FsrEasuSetH(dirPX,dirPY,lenP,ppp,false,true ,AH2(fL,gL),AH2(iL,jL),AH2(jL,kL),AH2(kL,lL),AH2(nL,oL)); + AH2 dir=AH2(dirPX.r+dirPX.g,dirPY.r+dirPY.g); + AH1 len=lenP.r+lenP.g; +//------------------------------------------------------------------------------------------------------------------------------ + AH2 dir2=dir*dir; + AH1 dirR=dir2.x+dir2.y; + AP1 zro=dirR<AH1_(1.0/32768.0); + dirR=APrxLoRsqH1(dirR); + dirR=zro?AH1_(1.0):dirR; + dir.x=zro?AH1_(1.0):dir.x; + dir*=AH2_(dirR); + len=len*AH1_(0.5); + len*=len; + AH1 stretch=(dir.x*dir.x+dir.y*dir.y)*APrxLoRcpH1(max(abs(dir.x),abs(dir.y))); + AH2 len2=AH2(AH1_(1.0)+(stretch-AH1_(1.0))*len,AH1_(1.0)+AH1_(-0.5)*len); + AH1 lob=AH1_(0.5)+AH1_((1.0/4.0-0.04)-0.5)*len; + AH1 clp=APrxLoRcpH1(lob); +//------------------------------------------------------------------------------------------------------------------------------ + // FP16 is different, using packed trick to do min and max in same operation. + AH2 bothR=max(max(AH2(-ijfeR.z,ijfeR.z),AH2(-klhgR.w,klhgR.w)),max(AH2(-ijfeR.y,ijfeR.y),AH2(-klhgR.x,klhgR.x))); + AH2 bothG=max(max(AH2(-ijfeG.z,ijfeG.z),AH2(-klhgG.w,klhgG.w)),max(AH2(-ijfeG.y,ijfeG.y),AH2(-klhgG.x,klhgG.x))); + AH2 bothB=max(max(AH2(-ijfeB.z,ijfeB.z),AH2(-klhgB.w,klhgB.w)),max(AH2(-ijfeB.y,ijfeB.y),AH2(-klhgB.x,klhgB.x))); + // This part is different for FP16, working pairs of taps at a time. + AH2 pR=AH2_(0.0); + AH2 pG=AH2_(0.0); + AH2 pB=AH2_(0.0); + AH2 pW=AH2_(0.0); + FsrEasuTapH(pR,pG,pB,pW,AH2( 0.0, 1.0)-ppp.xx,AH2(-1.0,-1.0)-ppp.yy,dir,len2,lob,clp,bczzR.xy,bczzG.xy,bczzB.xy); + FsrEasuTapH(pR,pG,pB,pW,AH2(-1.0, 0.0)-ppp.xx,AH2( 1.0, 1.0)-ppp.yy,dir,len2,lob,clp,ijfeR.xy,ijfeG.xy,ijfeB.xy); + FsrEasuTapH(pR,pG,pB,pW,AH2( 0.0,-1.0)-ppp.xx,AH2( 0.0, 0.0)-ppp.yy,dir,len2,lob,clp,ijfeR.zw,ijfeG.zw,ijfeB.zw); + FsrEasuTapH(pR,pG,pB,pW,AH2( 1.0, 2.0)-ppp.xx,AH2( 1.0, 1.0)-ppp.yy,dir,len2,lob,clp,klhgR.xy,klhgG.xy,klhgB.xy); + FsrEasuTapH(pR,pG,pB,pW,AH2( 2.0, 1.0)-ppp.xx,AH2( 0.0, 0.0)-ppp.yy,dir,len2,lob,clp,klhgR.zw,klhgG.zw,klhgB.zw); + FsrEasuTapH(pR,pG,pB,pW,AH2( 1.0, 0.0)-ppp.xx,AH2( 2.0, 2.0)-ppp.yy,dir,len2,lob,clp,zzonR.zw,zzonG.zw,zzonB.zw); + AH3 aC=AH3(pR.x+pR.y,pG.x+pG.y,pB.x+pB.y); + AH1 aW=pW.x+pW.y; +//------------------------------------------------------------------------------------------------------------------------------ + // Slightly different for FP16 version due to combined min and max. + pix=min(AH3(bothR.y,bothG.y,bothB.y),max(-AH3(bothR.x,bothG.x,bothB.x),aC*AH3_(ARcpH1(aW))));} +#endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// FSR - [RCAS] ROBUST CONTRAST ADAPTIVE SHARPENING +// +//------------------------------------------------------------------------------------------------------------------------------ +// CAS uses a simplified mechanism to convert local contrast into a variable amount of sharpness. +// RCAS uses a more exact mechanism, solving for the maximum local sharpness possible before clipping. +// RCAS also has a built in process to limit sharpening of what it detects as possible noise. +// RCAS sharper does not support scaling, as it should be applied after EASU scaling. +// Pass EASU output straight into RCAS, no color conversions necessary. +//------------------------------------------------------------------------------------------------------------------------------ +// RCAS is based on the following logic. +// RCAS uses a 5 tap filter in a cross pattern (same as CAS), +// w n +// w 1 w for taps w m e +// w s +// Where 'w' is the negative lobe weight. +// output = (w*(n+e+w+s)+m)/(4*w+1) +// RCAS solves for 'w' by seeing where the signal might clip out of the {0 to 1} input range, +// 0 == (w*(n+e+w+s)+m)/(4*w+1) -> w = -m/(n+e+w+s) +// 1 == (w*(n+e+w+s)+m)/(4*w+1) -> w = (1-m)/(n+e+w+s-4*1) +// Then chooses the 'w' which results in no clipping, limits 'w', and multiplies by the 'sharp' amount. +// This solution above has issues with MSAA input as the steps along the gradient cause edge detection issues. +// So RCAS uses 4x the maximum and 4x the minimum (depending on equation)in place of the individual taps. +// As well as switching from 'm' to either the minimum or maximum (depending on side), to help in energy conservation. +// This stabilizes RCAS. +// RCAS does a simple highpass which is normalized against the local contrast then shaped, +// 0.25 +// 0.25 -1 0.25 +// 0.25 +// This is used as a noise detection filter, to reduce the effect of RCAS on grain, and focus on real edges. +// +// GLSL example for the required callbacks : +// +// AH4 FsrRcasLoadH(ASW2 p){return AH4(imageLoad(imgSrc,ASU2(p)));} +// void FsrRcasInputH(inout AH1 r,inout AH1 g,inout AH1 b) +// { +// //do any simple input color conversions here or leave empty if none needed +// } +// +// FsrRcasCon need to be called from the CPU or GPU to set up constants. +// Including a GPU example here, the 'con' value would be stored out to a constant buffer. +// +// AU4 con; +// FsrRcasCon(con, +// 0.0); // The scale is {0.0 := maximum sharpness, to N>0, where N is the number of stops (halving) of the reduction of sharpness}. +// --------------- +// RCAS sharpening supports a CAS-like pass-through alpha via, +// #define FSR_RCAS_PASSTHROUGH_ALPHA 1 +// RCAS also supports a define to enable a more expensive path to avoid some sharpening of noise. +// Would suggest it is better to apply film grain after RCAS sharpening (and after scaling) instead of using this define, +// #define FSR_RCAS_DENOISE 1 +//============================================================================================================================== +// This is set at the limit of providing unnatural results for sharpening. +#define FSR_RCAS_LIMIT (0.25-(1.0/16.0)) +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// CONSTANT SETUP +//============================================================================================================================== +// Call to setup required constant values (works on CPU or GPU). +A_STATIC void FsrRcasCon( +outAU4 con, +// The scale is {0.0 := maximum, to N>0, where N is the number of stops (halving) of the reduction of sharpness}. +AF1 sharpness){ + // Transform from stops to linear value. + sharpness=AExp2F1(-sharpness); + varAF2(hSharp)=initAF2(sharpness,sharpness); + con[0]=AU1_AF1(sharpness); + con[1]=AU1_AH2_AF2(hSharp); + con[2]=0; + con[3]=0;} +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// NON-PACKED 32-BIT VERSION +//============================================================================================================================== +#if defined(A_GPU)&&defined(FSR_RCAS_F) + // Input callback prototypes that need to be implemented by calling shader + AF4 FsrRcasLoadF(ASU2 p); + void FsrRcasInputF(inout AF1 r,inout AF1 g,inout AF1 b); +//------------------------------------------------------------------------------------------------------------------------------ + void FsrRcasF( + out AF1 pixR, // Output values, non-vector so port between RcasFilter() and RcasFilterH() is easy. + out AF1 pixG, + out AF1 pixB, + #ifdef FSR_RCAS_PASSTHROUGH_ALPHA + out AF1 pixA, + #endif + AU2 ip, // Integer pixel position in output. + AU4 con){ // Constant generated by RcasSetup(). + // Algorithm uses minimal 3x3 pixel neighborhood. + // b + // d e f + // h + ASU2 sp=ASU2(ip); + AF3 b=FsrRcasLoadF(sp+ASU2( 0,-1)).rgb; + AF3 d=FsrRcasLoadF(sp+ASU2(-1, 0)).rgb; + #ifdef FSR_RCAS_PASSTHROUGH_ALPHA + AF4 ee=FsrRcasLoadF(sp); + AF3 e=ee.rgb;pixA=ee.a; + #else + AF3 e=FsrRcasLoadF(sp).rgb; + #endif + AF3 f=FsrRcasLoadF(sp+ASU2( 1, 0)).rgb; + AF3 h=FsrRcasLoadF(sp+ASU2( 0, 1)).rgb; + // Rename (32-bit) or regroup (16-bit). + AF1 bR=b.r; + AF1 bG=b.g; + AF1 bB=b.b; + AF1 dR=d.r; + AF1 dG=d.g; + AF1 dB=d.b; + AF1 eR=e.r; + AF1 eG=e.g; + AF1 eB=e.b; + AF1 fR=f.r; + AF1 fG=f.g; + AF1 fB=f.b; + AF1 hR=h.r; + AF1 hG=h.g; + AF1 hB=h.b; + // Run optional input transform. + FsrRcasInputF(bR,bG,bB); + FsrRcasInputF(dR,dG,dB); + FsrRcasInputF(eR,eG,eB); + FsrRcasInputF(fR,fG,fB); + FsrRcasInputF(hR,hG,hB); + // Luma times 2. + AF1 bL=bB*AF1_(0.5)+(bR*AF1_(0.5)+bG); + AF1 dL=dB*AF1_(0.5)+(dR*AF1_(0.5)+dG); + AF1 eL=eB*AF1_(0.5)+(eR*AF1_(0.5)+eG); + AF1 fL=fB*AF1_(0.5)+(fR*AF1_(0.5)+fG); + AF1 hL=hB*AF1_(0.5)+(hR*AF1_(0.5)+hG); + // Noise detection. + AF1 nz=AF1_(0.25)*bL+AF1_(0.25)*dL+AF1_(0.25)*fL+AF1_(0.25)*hL-eL; + nz=ASatF1(abs(nz)*APrxMedRcpF1(AMax3F1(AMax3F1(bL,dL,eL),fL,hL)-AMin3F1(AMin3F1(bL,dL,eL),fL,hL))); + nz=AF1_(-0.5)*nz+AF1_(1.0); + // Min and max of ring. + AF1 mn4R=min(AMin3F1(bR,dR,fR),hR); + AF1 mn4G=min(AMin3F1(bG,dG,fG),hG); + AF1 mn4B=min(AMin3F1(bB,dB,fB),hB); + AF1 mx4R=max(AMax3F1(bR,dR,fR),hR); + AF1 mx4G=max(AMax3F1(bG,dG,fG),hG); + AF1 mx4B=max(AMax3F1(bB,dB,fB),hB); + // Immediate constants for peak range. + AF2 peakC=AF2(1.0,-1.0*4.0); + // Limiters, these need to be high precision RCPs. + AF1 hitMinR=mn4R*ARcpF1(AF1_(4.0)*mx4R); + AF1 hitMinG=mn4G*ARcpF1(AF1_(4.0)*mx4G); + AF1 hitMinB=mn4B*ARcpF1(AF1_(4.0)*mx4B); + AF1 hitMaxR=(peakC.x-mx4R)*ARcpF1(AF1_(4.0)*mn4R+peakC.y); + AF1 hitMaxG=(peakC.x-mx4G)*ARcpF1(AF1_(4.0)*mn4G+peakC.y); + AF1 hitMaxB=(peakC.x-mx4B)*ARcpF1(AF1_(4.0)*mn4B+peakC.y); + AF1 lobeR=max(-hitMinR,hitMaxR); + AF1 lobeG=max(-hitMinG,hitMaxG); + AF1 lobeB=max(-hitMinB,hitMaxB); + AF1 lobe=max(AF1_(-FSR_RCAS_LIMIT),min(AMax3F1(lobeR,lobeG,lobeB),AF1_(0.0)))*AF1_AU1(con.x); + // Apply noise removal. + #ifdef FSR_RCAS_DENOISE + lobe*=nz; + #endif + // Resolve, which needs the medium precision rcp approximation to avoid visible tonality changes. + AF1 rcpL=APrxMedRcpF1(AF1_(4.0)*lobe+AF1_(1.0)); + pixR=(lobe*bR+lobe*dR+lobe*hR+lobe*fR+eR)*rcpL; + pixG=(lobe*bG+lobe*dG+lobe*hG+lobe*fG+eG)*rcpL; + pixB=(lobe*bB+lobe*dB+lobe*hB+lobe*fB+eB)*rcpL; + return;} +#endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// NON-PACKED 16-BIT VERSION +//============================================================================================================================== +#if defined(A_GPU)&&defined(A_HALF)&&defined(FSR_RCAS_H) + // Input callback prototypes that need to be implemented by calling shader + AH4 FsrRcasLoadH(ASW2 p); + void FsrRcasInputH(inout AH1 r,inout AH1 g,inout AH1 b); +//------------------------------------------------------------------------------------------------------------------------------ + void FsrRcasH( + out AH1 pixR, // Output values, non-vector so port between RcasFilter() and RcasFilterH() is easy. + out AH1 pixG, + out AH1 pixB, + #ifdef FSR_RCAS_PASSTHROUGH_ALPHA + out AH1 pixA, + #endif + AU2 ip, // Integer pixel position in output. + AU4 con){ // Constant generated by RcasSetup(). + // Sharpening algorithm uses minimal 3x3 pixel neighborhood. + // b + // d e f + // h + ASW2 sp=ASW2(ip); + AH3 b=FsrRcasLoadH(sp+ASW2( 0,-1)).rgb; + AH3 d=FsrRcasLoadH(sp+ASW2(-1, 0)).rgb; + #ifdef FSR_RCAS_PASSTHROUGH_ALPHA + AH4 ee=FsrRcasLoadH(sp); + AH3 e=ee.rgb;pixA=ee.a; + #else + AH3 e=FsrRcasLoadH(sp).rgb; + #endif + AH3 f=FsrRcasLoadH(sp+ASW2( 1, 0)).rgb; + AH3 h=FsrRcasLoadH(sp+ASW2( 0, 1)).rgb; + // Rename (32-bit) or regroup (16-bit). + AH1 bR=b.r; + AH1 bG=b.g; + AH1 bB=b.b; + AH1 dR=d.r; + AH1 dG=d.g; + AH1 dB=d.b; + AH1 eR=e.r; + AH1 eG=e.g; + AH1 eB=e.b; + AH1 fR=f.r; + AH1 fG=f.g; + AH1 fB=f.b; + AH1 hR=h.r; + AH1 hG=h.g; + AH1 hB=h.b; + // Run optional input transform. + FsrRcasInputH(bR,bG,bB); + FsrRcasInputH(dR,dG,dB); + FsrRcasInputH(eR,eG,eB); + FsrRcasInputH(fR,fG,fB); + FsrRcasInputH(hR,hG,hB); + // Luma times 2. + AH1 bL=bB*AH1_(0.5)+(bR*AH1_(0.5)+bG); + AH1 dL=dB*AH1_(0.5)+(dR*AH1_(0.5)+dG); + AH1 eL=eB*AH1_(0.5)+(eR*AH1_(0.5)+eG); + AH1 fL=fB*AH1_(0.5)+(fR*AH1_(0.5)+fG); + AH1 hL=hB*AH1_(0.5)+(hR*AH1_(0.5)+hG); + // Noise detection. + AH1 nz=AH1_(0.25)*bL+AH1_(0.25)*dL+AH1_(0.25)*fL+AH1_(0.25)*hL-eL; + nz=ASatH1(abs(nz)*APrxMedRcpH1(AMax3H1(AMax3H1(bL,dL,eL),fL,hL)-AMin3H1(AMin3H1(bL,dL,eL),fL,hL))); + nz=AH1_(-0.5)*nz+AH1_(1.0); + // Min and max of ring. + AH1 mn4R=min(AMin3H1(bR,dR,fR),hR); + AH1 mn4G=min(AMin3H1(bG,dG,fG),hG); + AH1 mn4B=min(AMin3H1(bB,dB,fB),hB); + AH1 mx4R=max(AMax3H1(bR,dR,fR),hR); + AH1 mx4G=max(AMax3H1(bG,dG,fG),hG); + AH1 mx4B=max(AMax3H1(bB,dB,fB),hB); + // Immediate constants for peak range. + AH2 peakC=AH2(1.0,-1.0*4.0); + // Limiters, these need to be high precision RCPs. + AH1 hitMinR=mn4R*ARcpH1(AH1_(4.0)*mx4R); + AH1 hitMinG=mn4G*ARcpH1(AH1_(4.0)*mx4G); + AH1 hitMinB=mn4B*ARcpH1(AH1_(4.0)*mx4B); + AH1 hitMaxR=(peakC.x-mx4R)*ARcpH1(AH1_(4.0)*mn4R+peakC.y); + AH1 hitMaxG=(peakC.x-mx4G)*ARcpH1(AH1_(4.0)*mn4G+peakC.y); + AH1 hitMaxB=(peakC.x-mx4B)*ARcpH1(AH1_(4.0)*mn4B+peakC.y); + AH1 lobeR=max(-hitMinR,hitMaxR); + AH1 lobeG=max(-hitMinG,hitMaxG); + AH1 lobeB=max(-hitMinB,hitMaxB); + AH1 lobe=max(AH1_(-FSR_RCAS_LIMIT),min(AMax3H1(lobeR,lobeG,lobeB),AH1_(0.0)))*AH2_AU1(con.y).x; + // Apply noise removal. + #ifdef FSR_RCAS_DENOISE + lobe*=nz; + #endif + // Resolve, which needs the medium precision rcp approximation to avoid visible tonality changes. + AH1 rcpL=APrxMedRcpH1(AH1_(4.0)*lobe+AH1_(1.0)); + pixR=(lobe*bR+lobe*dR+lobe*hR+lobe*fR+eR)*rcpL; + pixG=(lobe*bG+lobe*dG+lobe*hG+lobe*fG+eG)*rcpL; + pixB=(lobe*bB+lobe*dB+lobe*hB+lobe*fB+eB)*rcpL;} +#endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// PACKED 16-BIT VERSION +//============================================================================================================================== +#if defined(A_GPU)&&defined(A_HALF)&&defined(FSR_RCAS_HX2) + // Input callback prototypes that need to be implemented by the calling shader + AH4 FsrRcasLoadHx2(ASW2 p); + void FsrRcasInputHx2(inout AH2 r,inout AH2 g,inout AH2 b); +//------------------------------------------------------------------------------------------------------------------------------ + // Can be used to convert from packed Structures of Arrays to Arrays of Structures for store. + void FsrRcasDepackHx2(out AH4 pix0,out AH4 pix1,AH2 pixR,AH2 pixG,AH2 pixB){ + #ifdef A_HLSL + // Invoke a slower path for DX only, since it won't allow uninitialized values. + pix0.a=pix1.a=0.0; + #endif + pix0.rgb=AH3(pixR.x,pixG.x,pixB.x); + pix1.rgb=AH3(pixR.y,pixG.y,pixB.y);} +//------------------------------------------------------------------------------------------------------------------------------ + void FsrRcasHx2( + // Output values are for 2 8x8 tiles in a 16x8 region. + // pix<R,G,B>.x = left 8x8 tile + // pix<R,G,B>.y = right 8x8 tile + // This enables later processing to easily be packed as well. + out AH2 pixR, + out AH2 pixG, + out AH2 pixB, + #ifdef FSR_RCAS_PASSTHROUGH_ALPHA + out AH2 pixA, + #endif + AU2 ip, // Integer pixel position in output. + AU4 con){ // Constant generated by RcasSetup(). + // No scaling algorithm uses minimal 3x3 pixel neighborhood. + ASW2 sp0=ASW2(ip); + AH3 b0=FsrRcasLoadHx2(sp0+ASW2( 0,-1)).rgb; + AH3 d0=FsrRcasLoadHx2(sp0+ASW2(-1, 0)).rgb; + #ifdef FSR_RCAS_PASSTHROUGH_ALPHA + AH4 ee0=FsrRcasLoadHx2(sp0); + AH3 e0=ee0.rgb;pixA.r=ee0.a; + #else + AH3 e0=FsrRcasLoadHx2(sp0).rgb; + #endif + AH3 f0=FsrRcasLoadHx2(sp0+ASW2( 1, 0)).rgb; + AH3 h0=FsrRcasLoadHx2(sp0+ASW2( 0, 1)).rgb; + ASW2 sp1=sp0+ASW2(8,0); + AH3 b1=FsrRcasLoadHx2(sp1+ASW2( 0,-1)).rgb; + AH3 d1=FsrRcasLoadHx2(sp1+ASW2(-1, 0)).rgb; + #ifdef FSR_RCAS_PASSTHROUGH_ALPHA + AH4 ee1=FsrRcasLoadHx2(sp1); + AH3 e1=ee1.rgb;pixA.g=ee1.a; + #else + AH3 e1=FsrRcasLoadHx2(sp1).rgb; + #endif + AH3 f1=FsrRcasLoadHx2(sp1+ASW2( 1, 0)).rgb; + AH3 h1=FsrRcasLoadHx2(sp1+ASW2( 0, 1)).rgb; + // Arrays of Structures to Structures of Arrays conversion. + AH2 bR=AH2(b0.r,b1.r); + AH2 bG=AH2(b0.g,b1.g); + AH2 bB=AH2(b0.b,b1.b); + AH2 dR=AH2(d0.r,d1.r); + AH2 dG=AH2(d0.g,d1.g); + AH2 dB=AH2(d0.b,d1.b); + AH2 eR=AH2(e0.r,e1.r); + AH2 eG=AH2(e0.g,e1.g); + AH2 eB=AH2(e0.b,e1.b); + AH2 fR=AH2(f0.r,f1.r); + AH2 fG=AH2(f0.g,f1.g); + AH2 fB=AH2(f0.b,f1.b); + AH2 hR=AH2(h0.r,h1.r); + AH2 hG=AH2(h0.g,h1.g); + AH2 hB=AH2(h0.b,h1.b); + // Run optional input transform. + FsrRcasInputHx2(bR,bG,bB); + FsrRcasInputHx2(dR,dG,dB); + FsrRcasInputHx2(eR,eG,eB); + FsrRcasInputHx2(fR,fG,fB); + FsrRcasInputHx2(hR,hG,hB); + // Luma times 2. + AH2 bL=bB*AH2_(0.5)+(bR*AH2_(0.5)+bG); + AH2 dL=dB*AH2_(0.5)+(dR*AH2_(0.5)+dG); + AH2 eL=eB*AH2_(0.5)+(eR*AH2_(0.5)+eG); + AH2 fL=fB*AH2_(0.5)+(fR*AH2_(0.5)+fG); + AH2 hL=hB*AH2_(0.5)+(hR*AH2_(0.5)+hG); + // Noise detection. + AH2 nz=AH2_(0.25)*bL+AH2_(0.25)*dL+AH2_(0.25)*fL+AH2_(0.25)*hL-eL; + nz=ASatH2(abs(nz)*APrxMedRcpH2(AMax3H2(AMax3H2(bL,dL,eL),fL,hL)-AMin3H2(AMin3H2(bL,dL,eL),fL,hL))); + nz=AH2_(-0.5)*nz+AH2_(1.0); + // Min and max of ring. + AH2 mn4R=min(AMin3H2(bR,dR,fR),hR); + AH2 mn4G=min(AMin3H2(bG,dG,fG),hG); + AH2 mn4B=min(AMin3H2(bB,dB,fB),hB); + AH2 mx4R=max(AMax3H2(bR,dR,fR),hR); + AH2 mx4G=max(AMax3H2(bG,dG,fG),hG); + AH2 mx4B=max(AMax3H2(bB,dB,fB),hB); + // Immediate constants for peak range. + AH2 peakC=AH2(1.0,-1.0*4.0); + // Limiters, these need to be high precision RCPs. + AH2 hitMinR=mn4R*ARcpH2(AH2_(4.0)*mx4R); + AH2 hitMinG=mn4G*ARcpH2(AH2_(4.0)*mx4G); + AH2 hitMinB=mn4B*ARcpH2(AH2_(4.0)*mx4B); + AH2 hitMaxR=(peakC.x-mx4R)*ARcpH2(AH2_(4.0)*mn4R+peakC.y); + AH2 hitMaxG=(peakC.x-mx4G)*ARcpH2(AH2_(4.0)*mn4G+peakC.y); + AH2 hitMaxB=(peakC.x-mx4B)*ARcpH2(AH2_(4.0)*mn4B+peakC.y); + AH2 lobeR=max(-hitMinR,hitMaxR); + AH2 lobeG=max(-hitMinG,hitMaxG); + AH2 lobeB=max(-hitMinB,hitMaxB); + AH2 lobe=max(AH2_(-FSR_RCAS_LIMIT),min(AMax3H2(lobeR,lobeG,lobeB),AH2_(0.0)))*AH2_(AH2_AU1(con.y).x); + // Apply noise removal. + #ifdef FSR_RCAS_DENOISE + lobe*=nz; + #endif + // Resolve, which needs the medium precision rcp approximation to avoid visible tonality changes. + AH2 rcpL=APrxMedRcpH2(AH2_(4.0)*lobe+AH2_(1.0)); + pixR=(lobe*bR+lobe*dR+lobe*hR+lobe*fR+eR)*rcpL; + pixG=(lobe*bG+lobe*dG+lobe*hG+lobe*fG+eG)*rcpL; + pixB=(lobe*bB+lobe*dB+lobe*hB+lobe*fB+eB)*rcpL;} +#endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// FSR - [LFGA] LINEAR FILM GRAIN APPLICATOR +// +//------------------------------------------------------------------------------------------------------------------------------ +// Adding output-resolution film grain after scaling is a good way to mask both rendering and scaling artifacts. +// Suggest using tiled blue noise as film grain input, with peak noise frequency set for a specific look and feel. +// The 'Lfga*()' functions provide a convenient way to introduce grain. +// These functions limit grain based on distance to signal limits. +// This is done so that the grain is temporally energy preserving, and thus won't modify image tonality. +// Grain application should be done in a linear colorspace. +// The grain should be temporally changing, but have a temporal sum per pixel that adds to zero (non-biased). +//------------------------------------------------------------------------------------------------------------------------------ +// Usage, +// FsrLfga*( +// color, // In/out linear colorspace color {0 to 1} ranged. +// grain, // Per pixel grain texture value {-0.5 to 0.5} ranged, input is 3-channel to support colored grain. +// amount); // Amount of grain (0 to 1} ranged. +//------------------------------------------------------------------------------------------------------------------------------ +// Example if grain texture is monochrome: 'FsrLfgaF(color,AF3_(grain),amount)' +//============================================================================================================================== +#if defined(A_GPU) + // Maximum grain is the minimum distance to the signal limit. + void FsrLfgaF(inout AF3 c,AF3 t,AF1 a){c+=(t*AF3_(a))*min(AF3_(1.0)-c,c);} +#endif +//============================================================================================================================== +#if defined(A_GPU)&&defined(A_HALF) + // Half precision version (slower). + void FsrLfgaH(inout AH3 c,AH3 t,AH1 a){c+=(t*AH3_(a))*min(AH3_(1.0)-c,c);} +//------------------------------------------------------------------------------------------------------------------------------ + // Packed half precision version (faster). + void FsrLfgaHx2(inout AH2 cR,inout AH2 cG,inout AH2 cB,AH2 tR,AH2 tG,AH2 tB,AH1 a){ + cR+=(tR*AH2_(a))*min(AH2_(1.0)-cR,cR);cG+=(tG*AH2_(a))*min(AH2_(1.0)-cG,cG);cB+=(tB*AH2_(a))*min(AH2_(1.0)-cB,cB);} +#endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// FSR - [SRTM] SIMPLE REVERSIBLE TONE-MAPPER +// +//------------------------------------------------------------------------------------------------------------------------------ +// This provides a way to take linear HDR color {0 to FP16_MAX} and convert it into a temporary {0 to 1} ranged post-tonemapped linear. +// The tonemapper preserves RGB ratio, which helps maintain HDR color bleed during filtering. +//------------------------------------------------------------------------------------------------------------------------------ +// Reversible tonemapper usage, +// FsrSrtm*(color); // {0 to FP16_MAX} converted to {0 to 1}. +// FsrSrtmInv*(color); // {0 to 1} converted into {0 to 32768, output peak safe for FP16}. +//============================================================================================================================== +#if defined(A_GPU) + void FsrSrtmF(inout AF3 c){c*=AF3_(ARcpF1(AMax3F1(c.r,c.g,c.b)+AF1_(1.0)));} + // The extra max solves the c=1.0 case (which is a /0). + void FsrSrtmInvF(inout AF3 c){c*=AF3_(ARcpF1(max(AF1_(1.0/32768.0),AF1_(1.0)-AMax3F1(c.r,c.g,c.b))));} +#endif +//============================================================================================================================== +#if defined(A_GPU)&&defined(A_HALF) + void FsrSrtmH(inout AH3 c){c*=AH3_(ARcpH1(AMax3H1(c.r,c.g,c.b)+AH1_(1.0)));} + void FsrSrtmInvH(inout AH3 c){c*=AH3_(ARcpH1(max(AH1_(1.0/32768.0),AH1_(1.0)-AMax3H1(c.r,c.g,c.b))));} +//------------------------------------------------------------------------------------------------------------------------------ + void FsrSrtmHx2(inout AH2 cR,inout AH2 cG,inout AH2 cB){ + AH2 rcp=ARcpH2(AMax3H2(cR,cG,cB)+AH2_(1.0));cR*=rcp;cG*=rcp;cB*=rcp;} + void FsrSrtmInvHx2(inout AH2 cR,inout AH2 cG,inout AH2 cB){ + AH2 rcp=ARcpH2(max(AH2_(1.0/32768.0),AH2_(1.0)-AMax3H2(cR,cG,cB)));cR*=rcp;cG*=rcp;cB*=rcp;} +#endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// FSR - [TEPD] TEMPORAL ENERGY PRESERVING DITHER +// +//------------------------------------------------------------------------------------------------------------------------------ +// Temporally energy preserving dithered {0 to 1} linear to gamma 2.0 conversion. +// Gamma 2.0 is used so that the conversion back to linear is just to square the color. +// The conversion comes in 8-bit and 10-bit modes, designed for output to 8-bit UNORM or 10:10:10:2 respectively. +// Given good non-biased temporal blue noise as dither input, +// the output dither will temporally conserve energy. +// This is done by choosing the linear nearest step point instead of perceptual nearest. +// See code below for details. +//------------------------------------------------------------------------------------------------------------------------------ +// DX SPEC RULES FOR FLOAT->UNORM 8-BIT CONVERSION +// =============================================== +// - Output is 'uint(floor(saturate(n)*255.0+0.5))'. +// - Thus rounding is to nearest. +// - NaN gets converted to zero. +// - INF is clamped to {0.0 to 1.0}. +//============================================================================================================================== +#if defined(A_GPU) + // Hand tuned integer position to dither value, with more values than simple checkerboard. + // Only 32-bit has enough precision for this compddation. + // Output is {0 to <1}. + AF1 FsrTepdDitF(AU2 p,AU1 f){ + AF1 x=AF1_(p.x+f); + AF1 y=AF1_(p.y); + // The 1.61803 golden ratio. + AF1 a=AF1_((1.0+sqrt(5.0))/2.0); + // Number designed to provide a good visual pattern. + AF1 b=AF1_(1.0/3.69); + x=x*a+(y*b); + return AFractF1(x);} +//------------------------------------------------------------------------------------------------------------------------------ + // This version is 8-bit gamma 2.0. + // The 'c' input is {0 to 1}. + // Output is {0 to 1} ready for image store. + void FsrTepdC8F(inout AF3 c,AF1 dit){ + AF3 n=sqrt(c); + n=floor(n*AF3_(255.0))*AF3_(1.0/255.0); + AF3 a=n*n; + AF3 b=n+AF3_(1.0/255.0);b=b*b; + // Ratio of 'a' to 'b' required to produce 'c'. + // APrxLoRcpF1() won't work here (at least for very high dynamic ranges). + // APrxMedRcpF1() is an IADD,FMA,MUL. + AF3 r=(c-b)*APrxMedRcpF3(a-b); + // Use the ratio as a cutoff to choose 'a' or 'b'. + // AGtZeroF1() is a MUL. + c=ASatF3(n+AGtZeroF3(AF3_(dit)-r)*AF3_(1.0/255.0));} +//------------------------------------------------------------------------------------------------------------------------------ + // This version is 10-bit gamma 2.0. + // The 'c' input is {0 to 1}. + // Output is {0 to 1} ready for image store. + void FsrTepdC10F(inout AF3 c,AF1 dit){ + AF3 n=sqrt(c); + n=floor(n*AF3_(1023.0))*AF3_(1.0/1023.0); + AF3 a=n*n; + AF3 b=n+AF3_(1.0/1023.0);b=b*b; + AF3 r=(c-b)*APrxMedRcpF3(a-b); + c=ASatF3(n+AGtZeroF3(AF3_(dit)-r)*AF3_(1.0/1023.0));} +#endif +//============================================================================================================================== +#if defined(A_GPU)&&defined(A_HALF) + AH1 FsrTepdDitH(AU2 p,AU1 f){ + AF1 x=AF1_(p.x+f); + AF1 y=AF1_(p.y); + AF1 a=AF1_((1.0+sqrt(5.0))/2.0); + AF1 b=AF1_(1.0/3.69); + x=x*a+(y*b); + return AH1(AFractF1(x));} +//------------------------------------------------------------------------------------------------------------------------------ + void FsrTepdC8H(inout AH3 c,AH1 dit){ + AH3 n=sqrt(c); + n=floor(n*AH3_(255.0))*AH3_(1.0/255.0); + AH3 a=n*n; + AH3 b=n+AH3_(1.0/255.0);b=b*b; + AH3 r=(c-b)*APrxMedRcpH3(a-b); + c=ASatH3(n+AGtZeroH3(AH3_(dit)-r)*AH3_(1.0/255.0));} +//------------------------------------------------------------------------------------------------------------------------------ + void FsrTepdC10H(inout AH3 c,AH1 dit){ + AH3 n=sqrt(c); + n=floor(n*AH3_(1023.0))*AH3_(1.0/1023.0); + AH3 a=n*n; + AH3 b=n+AH3_(1.0/1023.0);b=b*b; + AH3 r=(c-b)*APrxMedRcpH3(a-b); + c=ASatH3(n+AGtZeroH3(AH3_(dit)-r)*AH3_(1.0/1023.0));} +//============================================================================================================================== + // This computes dither for positions 'p' and 'p+{8,0}'. + AH2 FsrTepdDitHx2(AU2 p,AU1 f){ + AF2 x; + x.x=AF1_(p.x+f); + x.y=x.x+AF1_(8.0); + AF1 y=AF1_(p.y); + AF1 a=AF1_((1.0+sqrt(5.0))/2.0); + AF1 b=AF1_(1.0/3.69); + x=x*AF2_(a)+AF2_(y*b); + return AH2(AFractF2(x));} +//------------------------------------------------------------------------------------------------------------------------------ + void FsrTepdC8Hx2(inout AH2 cR,inout AH2 cG,inout AH2 cB,AH2 dit){ + AH2 nR=sqrt(cR); + AH2 nG=sqrt(cG); + AH2 nB=sqrt(cB); + nR=floor(nR*AH2_(255.0))*AH2_(1.0/255.0); + nG=floor(nG*AH2_(255.0))*AH2_(1.0/255.0); + nB=floor(nB*AH2_(255.0))*AH2_(1.0/255.0); + AH2 aR=nR*nR; + AH2 aG=nG*nG; + AH2 aB=nB*nB; + AH2 bR=nR+AH2_(1.0/255.0);bR=bR*bR; + AH2 bG=nG+AH2_(1.0/255.0);bG=bG*bG; + AH2 bB=nB+AH2_(1.0/255.0);bB=bB*bB; + AH2 rR=(cR-bR)*APrxMedRcpH2(aR-bR); + AH2 rG=(cG-bG)*APrxMedRcpH2(aG-bG); + AH2 rB=(cB-bB)*APrxMedRcpH2(aB-bB); + cR=ASatH2(nR+AGtZeroH2(dit-rR)*AH2_(1.0/255.0)); + cG=ASatH2(nG+AGtZeroH2(dit-rG)*AH2_(1.0/255.0)); + cB=ASatH2(nB+AGtZeroH2(dit-rB)*AH2_(1.0/255.0));} +//------------------------------------------------------------------------------------------------------------------------------ + void FsrTepdC10Hx2(inout AH2 cR,inout AH2 cG,inout AH2 cB,AH2 dit){ + AH2 nR=sqrt(cR); + AH2 nG=sqrt(cG); + AH2 nB=sqrt(cB); + nR=floor(nR*AH2_(1023.0))*AH2_(1.0/1023.0); + nG=floor(nG*AH2_(1023.0))*AH2_(1.0/1023.0); + nB=floor(nB*AH2_(1023.0))*AH2_(1.0/1023.0); + AH2 aR=nR*nR; + AH2 aG=nG*nG; + AH2 aB=nB*nB; + AH2 bR=nR+AH2_(1.0/1023.0);bR=bR*bR; + AH2 bG=nG+AH2_(1.0/1023.0);bG=bG*bG; + AH2 bB=nB+AH2_(1.0/1023.0);bB=bB*bB; + AH2 rR=(cR-bR)*APrxMedRcpH2(aR-bR); + AH2 rG=(cG-bG)*APrxMedRcpH2(aG-bG); + AH2 rB=(cB-bB)*APrxMedRcpH2(aB-bB); + cR=ASatH2(nR+AGtZeroH2(dit-rR)*AH2_(1.0/1023.0)); + cG=ASatH2(nG+AGtZeroH2(dit-rG)*AH2_(1.0/1023.0)); + cB=ASatH2(nB+AGtZeroH2(dit-rB)*AH2_(1.0/1023.0));} +#endif diff --git a/externals/FidelityFX-FSR/license.txt b/externals/FidelityFX-FSR/license.txt new file mode 100644 index 000000000..324cba594 --- /dev/null +++ b/externals/FidelityFX-FSR/license.txt @@ -0,0 +1,19 @@ +Copyright (c) 2021 Advanced Micro Devices, Inc. All rights reserved. + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in +all copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +THE SOFTWARE. diff --git a/src/common/logging/backend.cpp b/src/common/logging/backend.cpp index 0e85a9c1d..c51c05b28 100644 --- a/src/common/logging/backend.cpp +++ b/src/common/logging/backend.cpp @@ -6,6 +6,7 @@ #include <chrono> #include <climits> #include <exception> +#include <stop_token> #include <thread> #include <vector> @@ -186,6 +187,10 @@ public: initialization_in_progress_suppress_logging = false; } + static void Start() { + instance->StartBackendThread(); + } + Impl(const Impl&) = delete; Impl& operator=(const Impl&) = delete; @@ -201,7 +206,7 @@ public: } void PushEntry(Class log_class, Level log_level, const char* filename, unsigned int line_num, - const char* function, std::string message) { + const char* function, std::string&& message) { if (!filter.CheckMessage(log_class, log_level)) return; const Entry& entry = @@ -211,40 +216,41 @@ public: private: Impl(const std::filesystem::path& file_backend_filename, const Filter& filter_) - : filter{filter_}, file_backend{file_backend_filename}, backend_thread{std::thread([this] { - Common::SetCurrentThreadName("yuzu:Log"); - Entry entry; - const auto write_logs = [this, &entry]() { - ForEachBackend([&entry](Backend& backend) { backend.Write(entry); }); - }; - while (true) { - entry = message_queue.PopWait(); - if (entry.final_entry) { - break; - } - write_logs(); - } - // Drain the logging queue. Only writes out up to MAX_LOGS_TO_WRITE to prevent a - // case where a system is repeatedly spamming logs even on close. - int max_logs_to_write = filter.IsDebug() ? INT_MAX : 100; - while (max_logs_to_write-- && message_queue.Pop(entry)) { - write_logs(); - } - })} {} + : filter{filter_}, file_backend{file_backend_filename} {} ~Impl() { StopBackendThread(); } + void StartBackendThread() { + backend_thread = std::thread([this] { + Common::SetCurrentThreadName("yuzu:Log"); + Entry entry; + const auto write_logs = [this, &entry]() { + ForEachBackend([&entry](Backend& backend) { backend.Write(entry); }); + }; + while (!stop.stop_requested()) { + entry = message_queue.PopWait(stop.get_token()); + if (entry.filename != nullptr) { + write_logs(); + } + } + // Drain the logging queue. Only writes out up to MAX_LOGS_TO_WRITE to prevent a + // case where a system is repeatedly spamming logs even on close. + int max_logs_to_write = filter.IsDebug() ? INT_MAX : 100; + while (max_logs_to_write-- && message_queue.Pop(entry)) { + write_logs(); + } + }); + } + void StopBackendThread() { - Entry stop_entry{}; - stop_entry.final_entry = true; - message_queue.Push(stop_entry); + stop.request_stop(); backend_thread.join(); } Entry CreateEntry(Class log_class, Level log_level, const char* filename, unsigned int line_nr, - const char* function, std::string message) const { + const char* function, std::string&& message) const { using std::chrono::duration_cast; using std::chrono::microseconds; using std::chrono::steady_clock; @@ -257,7 +263,6 @@ private: .line_num = line_nr, .function = function, .message = std::move(message), - .final_entry = false, }; } @@ -278,8 +283,9 @@ private: ColorConsoleBackend color_console_backend{}; FileBackend file_backend; + std::stop_source stop; std::thread backend_thread; - MPSCQueue<Entry> message_queue{}; + MPSCQueue<Entry, true> message_queue{}; std::chrono::steady_clock::time_point time_origin{std::chrono::steady_clock::now()}; }; } // namespace @@ -288,6 +294,10 @@ void Initialize() { Impl::Initialize(); } +void Start() { + Impl::Start(); +} + void DisableLoggingInTests() { initialization_in_progress_suppress_logging = true; } diff --git a/src/common/logging/backend.h b/src/common/logging/backend.h index cb7839ee9..bf785f402 100644 --- a/src/common/logging/backend.h +++ b/src/common/logging/backend.h @@ -14,6 +14,8 @@ class Filter; /// Initializes the logging system. This should be the first thing called in main. void Initialize(); +void Start(); + void DisableLoggingInTests(); /** diff --git a/src/common/logging/log_entry.h b/src/common/logging/log_entry.h index dd6f44841..b28570071 100644 --- a/src/common/logging/log_entry.h +++ b/src/common/logging/log_entry.h @@ -22,7 +22,6 @@ struct Entry { unsigned int line_num = 0; std::string function; std::string message; - bool final_entry = false; }; } // namespace Common::Log diff --git a/src/common/math_util.h b/src/common/math_util.h index 4c38d8040..510c4e56d 100644 --- a/src/common/math_util.h +++ b/src/common/math_util.h @@ -48,8 +48,8 @@ struct Rectangle { } [[nodiscard]] Rectangle<T> Scale(const float s) const { - return Rectangle{left, top, static_cast<T>(left + GetWidth() * s), - static_cast<T>(top + GetHeight() * s)}; + return Rectangle{left, top, static_cast<T>(static_cast<float>(left + GetWidth()) * s), + static_cast<T>(static_cast<float>(top + GetHeight()) * s)}; } }; diff --git a/src/common/settings.cpp b/src/common/settings.cpp index 9dd5e3efb..3bcaa072f 100644 --- a/src/common/settings.cpp +++ b/src/common/settings.cpp @@ -47,7 +47,9 @@ void LogSettings() { log_setting("System_TimeZoneIndex", values.time_zone_index.GetValue()); log_setting("Core_UseMultiCore", values.use_multi_core.GetValue()); log_setting("CPU_Accuracy", values.cpu_accuracy.GetValue()); - log_setting("Renderer_UseResolutionFactor", values.resolution_factor.GetValue()); + log_setting("Renderer_UseResolutionScaling", values.resolution_setup.GetValue()); + log_setting("Renderer_ScalingFilter", values.scaling_filter.GetValue()); + log_setting("Renderer_AntiAliasing", values.anti_aliasing.GetValue()); log_setting("Renderer_UseSpeedLimit", values.use_speed_limit.GetValue()); log_setting("Renderer_SpeedLimit", values.speed_limit.GetValue()); log_setting("Renderer_UseDiskShaderCache", values.use_disk_shader_cache.GetValue()); @@ -105,6 +107,55 @@ float Volume() { return values.volume.GetValue() / 100.0f; } +void UpdateRescalingInfo() { + const auto setup = values.resolution_setup.GetValue(); + auto& info = values.resolution_info; + info.downscale = false; + switch (setup) { + case ResolutionSetup::Res1_2X: + info.up_scale = 1; + info.down_shift = 1; + info.downscale = true; + break; + case ResolutionSetup::Res3_4X: + info.up_scale = 3; + info.down_shift = 2; + info.downscale = true; + break; + case ResolutionSetup::Res1X: + info.up_scale = 1; + info.down_shift = 0; + break; + case ResolutionSetup::Res2X: + info.up_scale = 2; + info.down_shift = 0; + break; + case ResolutionSetup::Res3X: + info.up_scale = 3; + info.down_shift = 0; + break; + case ResolutionSetup::Res4X: + info.up_scale = 4; + info.down_shift = 0; + break; + case ResolutionSetup::Res5X: + info.up_scale = 5; + info.down_shift = 0; + break; + case ResolutionSetup::Res6X: + info.up_scale = 6; + info.down_shift = 0; + break; + default: + UNREACHABLE(); + info.up_scale = 1; + info.down_shift = 0; + } + info.up_factor = static_cast<f32>(info.up_scale) / (1U << info.down_shift); + info.down_factor = static_cast<f32>(1U << info.down_shift) / info.up_scale; + info.active = info.up_scale != 1 || info.down_shift != 0; +} + void RestoreGlobalState(bool is_powered_on) { // If a game is running, DO NOT restore the global settings state if (is_powered_on) { diff --git a/src/common/settings.h b/src/common/settings.h index 9ff4cf85d..42f8b4a7d 100644 --- a/src/common/settings.h +++ b/src/common/settings.h @@ -52,6 +52,56 @@ enum class NvdecEmulation : u32 { GPU = 2, }; +enum class ResolutionSetup : u32 { + Res1_2X = 0, + Res3_4X = 1, + Res1X = 2, + Res2X = 3, + Res3X = 4, + Res4X = 5, + Res5X = 6, + Res6X = 7, +}; + +enum class ScalingFilter : u32 { + NearestNeighbor = 0, + Bilinear = 1, + Bicubic = 2, + Gaussian = 3, + ScaleForce = 4, + Fsr = 5, + LastFilter = Fsr, +}; + +enum class AntiAliasing : u32 { + None = 0, + Fxaa = 1, + LastAA = Fxaa, +}; + +struct ResolutionScalingInfo { + u32 up_scale{1}; + u32 down_shift{0}; + f32 up_factor{1.0f}; + f32 down_factor{1.0f}; + bool active{}; + bool downscale{}; + + s32 ScaleUp(s32 value) const { + if (value == 0) { + return 0; + } + return std::max((value * static_cast<s32>(up_scale)) >> static_cast<s32>(down_shift), 1); + } + + u32 ScaleUp(u32 value) const { + if (value == 0U) { + return 0U; + } + return std::max((value * up_scale) >> down_shift, 1U); + } +}; + /** The BasicSetting class is a simple resource manager. It defines a label and default value * alongside the actual value of the setting for simpler and less-error prone use with frontend * configurations. Setting a default value and label is required, though subclasses may deviate from @@ -451,7 +501,10 @@ struct Values { "disable_shader_loop_safety_checks"}; Setting<int> vulkan_device{0, "vulkan_device"}; - Setting<u16> resolution_factor{1, "resolution_factor"}; + ResolutionScalingInfo resolution_info{}; + Setting<ResolutionSetup> resolution_setup{ResolutionSetup::Res1X, "resolution_setup"}; + Setting<ScalingFilter> scaling_filter{ScalingFilter::Bilinear, "scaling_filter"}; + Setting<AntiAliasing> anti_aliasing{AntiAliasing::None, "anti_aliasing"}; // *nix platforms may have issues with the borderless windowed fullscreen mode. // Default to exclusive fullscreen on these platforms for now. RangedSetting<FullscreenMode> fullscreen_mode{ @@ -462,7 +515,7 @@ struct Values { #endif FullscreenMode::Borderless, FullscreenMode::Exclusive, "fullscreen_mode"}; RangedSetting<int> aspect_ratio{0, 0, 3, "aspect_ratio"}; - RangedSetting<int> max_anisotropy{0, 0, 4, "max_anisotropy"}; + RangedSetting<int> max_anisotropy{0, 0, 5, "max_anisotropy"}; Setting<bool> use_speed_limit{true, "use_speed_limit"}; RangedSetting<u16> speed_limit{100, 0, 9999, "speed_limit"}; Setting<bool> use_disk_shader_cache{true, "use_disk_shader_cache"}; @@ -595,6 +648,8 @@ std::string GetTimeZoneString(); void LogSettings(); +void UpdateRescalingInfo(); + // Restore the global state of all applicable settings in the Values struct void RestoreGlobalState(bool is_powered_on); diff --git a/src/core/frontend/framebuffer_layout.cpp b/src/core/frontend/framebuffer_layout.cpp index 0832463d6..4b58b672a 100644 --- a/src/core/frontend/framebuffer_layout.cpp +++ b/src/core/frontend/framebuffer_layout.cpp @@ -44,16 +44,13 @@ FramebufferLayout DefaultFrameLayout(u32 width, u32 height) { return res; } -FramebufferLayout FrameLayoutFromResolutionScale(u32 res_scale) { - u32 width, height; +FramebufferLayout FrameLayoutFromResolutionScale(f32 res_scale) { + const bool is_docked = Settings::values.use_docked_mode.GetValue(); + const u32 screen_width = is_docked ? ScreenDocked::Width : ScreenUndocked::Width; + const u32 screen_height = is_docked ? ScreenDocked::Height : ScreenUndocked::Height; - if (Settings::values.use_docked_mode.GetValue()) { - width = ScreenDocked::Width * res_scale; - height = ScreenDocked::Height * res_scale; - } else { - width = ScreenUndocked::Width * res_scale; - height = ScreenUndocked::Height * res_scale; - } + const u32 width = static_cast<u32>(static_cast<f32>(screen_width) * res_scale); + const u32 height = static_cast<u32>(static_cast<f32>(screen_height) * res_scale); return DefaultFrameLayout(width, height); } diff --git a/src/core/frontend/framebuffer_layout.h b/src/core/frontend/framebuffer_layout.h index e2e3bbbb3..2e36c0163 100644 --- a/src/core/frontend/framebuffer_layout.h +++ b/src/core/frontend/framebuffer_layout.h @@ -60,7 +60,7 @@ FramebufferLayout DefaultFrameLayout(u32 width, u32 height); * Convenience method to get frame layout by resolution scale * @param res_scale resolution scale factor */ -FramebufferLayout FrameLayoutFromResolutionScale(u32 res_scale); +FramebufferLayout FrameLayoutFromResolutionScale(f32 res_scale); /** * Convenience method to determine emulation aspect ratio diff --git a/src/core/hle/service/am/am.cpp b/src/core/hle/service/am/am.cpp index 50c2ace93..aee8d4f93 100644 --- a/src/core/hle/service/am/am.cpp +++ b/src/core/hle/service/am/am.cpp @@ -797,15 +797,11 @@ void ICommonStateGetter::GetDefaultDisplayResolution(Kernel::HLERequestContext& rb.Push(ResultSuccess); if (Settings::values.use_docked_mode.GetValue()) { - rb.Push(static_cast<u32>(Service::VI::DisplayResolution::DockedWidth) * - static_cast<u32>(Settings::values.resolution_factor.GetValue())); - rb.Push(static_cast<u32>(Service::VI::DisplayResolution::DockedHeight) * - static_cast<u32>(Settings::values.resolution_factor.GetValue())); + rb.Push(static_cast<u32>(Service::VI::DisplayResolution::DockedWidth)); + rb.Push(static_cast<u32>(Service::VI::DisplayResolution::DockedHeight)); } else { - rb.Push(static_cast<u32>(Service::VI::DisplayResolution::UndockedWidth) * - static_cast<u32>(Settings::values.resolution_factor.GetValue())); - rb.Push(static_cast<u32>(Service::VI::DisplayResolution::UndockedHeight) * - static_cast<u32>(Settings::values.resolution_factor.GetValue())); + rb.Push(static_cast<u32>(Service::VI::DisplayResolution::UndockedWidth)); + rb.Push(static_cast<u32>(Service::VI::DisplayResolution::UndockedHeight)); } } diff --git a/src/core/hle/service/vi/vi.cpp b/src/core/hle/service/vi/vi.cpp index 63d5242c4..75ee3e5e4 100644 --- a/src/core/hle/service/vi/vi.cpp +++ b/src/core/hle/service/vi/vi.cpp @@ -541,11 +541,8 @@ private: switch (transaction) { case TransactionId::Connect: { IGBPConnectRequestParcel request{ctx.ReadBuffer()}; - IGBPConnectResponseParcel response{ - static_cast<u32>(static_cast<u32>(DisplayResolution::UndockedWidth) * - Settings::values.resolution_factor.GetValue()), - static_cast<u32>(static_cast<u32>(DisplayResolution::UndockedHeight) * - Settings::values.resolution_factor.GetValue())}; + IGBPConnectResponseParcel response{static_cast<u32>(DisplayResolution::UndockedWidth), + static_cast<u32>(DisplayResolution::UndockedHeight)}; buffer_queue.Connect(); @@ -775,15 +772,11 @@ private: rb.Push(ResultSuccess); if (Settings::values.use_docked_mode.GetValue()) { - rb.Push(static_cast<u32>(Service::VI::DisplayResolution::DockedWidth) * - static_cast<u32>(Settings::values.resolution_factor.GetValue())); - rb.Push(static_cast<u32>(Service::VI::DisplayResolution::DockedHeight) * - static_cast<u32>(Settings::values.resolution_factor.GetValue())); + rb.Push(static_cast<u32>(Service::VI::DisplayResolution::DockedWidth)); + rb.Push(static_cast<u32>(Service::VI::DisplayResolution::DockedHeight)); } else { - rb.Push(static_cast<u32>(Service::VI::DisplayResolution::UndockedWidth) * - static_cast<u32>(Settings::values.resolution_factor.GetValue())); - rb.Push(static_cast<u32>(Service::VI::DisplayResolution::UndockedHeight) * - static_cast<u32>(Settings::values.resolution_factor.GetValue())); + rb.Push(static_cast<u32>(Service::VI::DisplayResolution::UndockedWidth)); + rb.Push(static_cast<u32>(Service::VI::DisplayResolution::UndockedHeight)); } rb.PushRaw<float>(60.0f); // This wouldn't seem to be correct for 30 fps games. @@ -1063,10 +1056,8 @@ private: // This only returns the fixed values of 1280x720 and makes no distinguishing // between docked and undocked dimensions. We take the liberty of applying // the resolution scaling factor here. - rb.Push(static_cast<u64>(DisplayResolution::UndockedWidth) * - static_cast<u32>(Settings::values.resolution_factor.GetValue())); - rb.Push(static_cast<u64>(DisplayResolution::UndockedHeight) * - static_cast<u32>(Settings::values.resolution_factor.GetValue())); + rb.Push(static_cast<u64>(DisplayResolution::UndockedWidth)); + rb.Push(static_cast<u64>(DisplayResolution::UndockedHeight)); } void SetLayerScalingMode(Kernel::HLERequestContext& ctx) { @@ -1099,8 +1090,6 @@ private: LOG_WARNING(Service_VI, "(STUBBED) called"); DisplayInfo display_info; - display_info.width *= static_cast<u64>(Settings::values.resolution_factor.GetValue()); - display_info.height *= static_cast<u64>(Settings::values.resolution_factor.GetValue()); ctx.WriteBuffer(&display_info, sizeof(DisplayInfo)); IPC::ResponseBuilder rb{ctx, 4}; rb.Push(ResultSuccess); diff --git a/src/core/telemetry_session.cpp b/src/core/telemetry_session.cpp index 191475f71..654db0b52 100644 --- a/src/core/telemetry_session.cpp +++ b/src/core/telemetry_session.cpp @@ -229,8 +229,6 @@ void TelemetrySession::AddInitialInfo(Loader::AppLoader& app_loader, AddField(field_type, "Core_UseMultiCore", Settings::values.use_multi_core.GetValue()); AddField(field_type, "Renderer_Backend", TranslateRenderer(Settings::values.renderer_backend.GetValue())); - AddField(field_type, "Renderer_ResolutionFactor", - Settings::values.resolution_factor.GetValue()); AddField(field_type, "Renderer_UseSpeedLimit", Settings::values.use_speed_limit.GetValue()); AddField(field_type, "Renderer_SpeedLimit", Settings::values.speed_limit.GetValue()); AddField(field_type, "Renderer_UseDiskShaderCache", diff --git a/src/shader_recompiler/CMakeLists.txt b/src/shader_recompiler/CMakeLists.txt index b5b7e5e83..bc3df80c8 100644 --- a/src/shader_recompiler/CMakeLists.txt +++ b/src/shader_recompiler/CMakeLists.txt @@ -221,6 +221,7 @@ add_library(shader_recompiler STATIC ir_opt/lower_fp16_to_fp32.cpp ir_opt/lower_int64_to_int32.cpp ir_opt/passes.h + ir_opt/rescaling_pass.cpp ir_opt/ssa_rewrite_pass.cpp ir_opt/texture_pass.cpp ir_opt/verification_pass.cpp diff --git a/src/shader_recompiler/backend/bindings.h b/src/shader_recompiler/backend/bindings.h index 35503000c..669702553 100644 --- a/src/shader_recompiler/backend/bindings.h +++ b/src/shader_recompiler/backend/bindings.h @@ -14,6 +14,8 @@ struct Bindings { u32 storage_buffer{}; u32 texture{}; u32 image{}; + u32 texture_scaling_index{}; + u32 image_scaling_index{}; }; } // namespace Shader::Backend diff --git a/src/shader_recompiler/backend/glasm/emit_context.cpp b/src/shader_recompiler/backend/glasm/emit_context.cpp index 069c019ad..8fd459dfe 100644 --- a/src/shader_recompiler/backend/glasm/emit_context.cpp +++ b/src/shader_recompiler/backend/glasm/emit_context.cpp @@ -6,6 +6,7 @@ #include "shader_recompiler/backend/bindings.h" #include "shader_recompiler/backend/glasm/emit_context.h" +#include "shader_recompiler/backend/glasm/emit_glasm.h" #include "shader_recompiler/frontend/ir/program.h" #include "shader_recompiler/profile.h" #include "shader_recompiler/runtime_info.h" @@ -55,7 +56,8 @@ EmitContext::EmitContext(IR::Program& program, Bindings& bindings, const Profile } if (!runtime_info.glasm_use_storage_buffers) { if (const size_t num = info.storage_buffers_descriptors.size(); num > 0) { - Add("PARAM c[{}]={{program.local[0..{}]}};", num, num - 1); + const size_t index{num + PROGRAM_LOCAL_PARAMETER_STORAGE_BUFFER_BASE}; + Add("PARAM c[{}]={{program.local[0..{}]}};", index, index - 1); } } stage = program.stage; diff --git a/src/shader_recompiler/backend/glasm/emit_glasm.cpp b/src/shader_recompiler/backend/glasm/emit_glasm.cpp index 4ce1c4f54..004658546 100644 --- a/src/shader_recompiler/backend/glasm/emit_glasm.cpp +++ b/src/shader_recompiler/backend/glasm/emit_glasm.cpp @@ -448,6 +448,9 @@ std::string EmitGLASM(const Profile& profile, const RuntimeInfo& runtime_info, I header += fmt::format("SHARED_MEMORY {};", program.shared_memory_size); header += fmt::format("SHARED shared_mem[]={{program.sharedmem}};"); } + if (program.info.uses_rescaling_uniform) { + header += "PARAM scaling[1]={program.local[0..0]};"; + } header += "TEMP "; for (size_t index = 0; index < ctx.reg_alloc.NumUsedRegisters(); ++index) { header += fmt::format("R{},", index); diff --git a/src/shader_recompiler/backend/glasm/emit_glasm.h b/src/shader_recompiler/backend/glasm/emit_glasm.h index bcb55f062..292655acb 100644 --- a/src/shader_recompiler/backend/glasm/emit_glasm.h +++ b/src/shader_recompiler/backend/glasm/emit_glasm.h @@ -13,6 +13,8 @@ namespace Shader::Backend::GLASM { +constexpr u32 PROGRAM_LOCAL_PARAMETER_STORAGE_BUFFER_BASE = 1; + [[nodiscard]] std::string EmitGLASM(const Profile& profile, const RuntimeInfo& runtime_info, IR::Program& program, Bindings& bindings); diff --git a/src/shader_recompiler/backend/glasm/emit_glasm_image.cpp b/src/shader_recompiler/backend/glasm/emit_glasm_image.cpp index 09e3a9b82..d325d31c7 100644 --- a/src/shader_recompiler/backend/glasm/emit_glasm_image.cpp +++ b/src/shader_recompiler/backend/glasm/emit_glasm_image.cpp @@ -608,6 +608,24 @@ void EmitImageWrite(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, Re ctx.Add("STOREIM.{} {},{},{},{};", format, image, color, coord, type); } +void EmitIsTextureScaled(EmitContext& ctx, IR::Inst& inst, const IR::Value& index) { + if (!index.IsImmediate()) { + throw NotImplementedException("Non-constant texture rescaling"); + } + ctx.Add("AND.U RC.x,scaling[0].x,{};" + "SNE.S {},RC.x,0;", + 1u << index.U32(), ctx.reg_alloc.Define(inst)); +} + +void EmitIsImageScaled(EmitContext& ctx, IR::Inst& inst, const IR::Value& index) { + if (!index.IsImmediate()) { + throw NotImplementedException("Non-constant texture rescaling"); + } + ctx.Add("AND.U RC.x,scaling[0].y,{};" + "SNE.S {},RC.x,0;", + 1u << index.U32(), ctx.reg_alloc.Define(inst)); +} + void EmitImageAtomicIAdd32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, Register coord, ScalarU32 value) { ImageAtomic(ctx, inst, index, coord, value, "ADD.U32"); diff --git a/src/shader_recompiler/backend/glasm/emit_glasm_instructions.h b/src/shader_recompiler/backend/glasm/emit_glasm_instructions.h index 12afda43b..1f343bff5 100644 --- a/src/shader_recompiler/backend/glasm/emit_glasm_instructions.h +++ b/src/shader_recompiler/backend/glasm/emit_glasm_instructions.h @@ -72,6 +72,7 @@ void EmitInvocationId(EmitContext& ctx, IR::Inst& inst); void EmitSampleId(EmitContext& ctx, IR::Inst& inst); void EmitIsHelperInvocation(EmitContext& ctx, IR::Inst& inst); void EmitYDirection(EmitContext& ctx, IR::Inst& inst); +void EmitResolutionDownFactor(EmitContext& ctx, IR::Inst& inst); void EmitLoadLocal(EmitContext& ctx, IR::Inst& inst, ScalarU32 word_offset); void EmitWriteLocal(EmitContext& ctx, ScalarU32 word_offset, ScalarU32 value); void EmitUndefU1(EmitContext& ctx, IR::Inst& inst); @@ -303,6 +304,8 @@ void EmitIAdd64(EmitContext& ctx, IR::Inst& inst, Register a, Register b); void EmitISub32(EmitContext& ctx, IR::Inst& inst, ScalarS32 a, ScalarS32 b); void EmitISub64(EmitContext& ctx, IR::Inst& inst, Register a, Register b); void EmitIMul32(EmitContext& ctx, IR::Inst& inst, ScalarS32 a, ScalarS32 b); +void EmitSDiv32(EmitContext& ctx, IR::Inst& inst, ScalarS32 a, ScalarS32 b); +void EmitUDiv32(EmitContext& ctx, IR::Inst& inst, ScalarU32 a, ScalarU32 b); void EmitINeg32(EmitContext& ctx, IR::Inst& inst, ScalarS32 value); void EmitINeg64(EmitContext& ctx, IR::Inst& inst, Register value); void EmitIAbs32(EmitContext& ctx, IR::Inst& inst, ScalarS32 value); @@ -553,6 +556,8 @@ void EmitImageGradient(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, void EmitImageRead(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, Register coord); void EmitImageWrite(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, Register coord, Register color); +void EmitIsTextureScaled(EmitContext& ctx, IR::Inst& inst, const IR::Value& index); +void EmitIsImageScaled(EmitContext& ctx, IR::Inst& inst, const IR::Value& index); void EmitBindlessImageAtomicIAdd32(EmitContext&); void EmitBindlessImageAtomicSMin32(EmitContext&); void EmitBindlessImageAtomicUMin32(EmitContext&); diff --git a/src/shader_recompiler/backend/glasm/emit_glasm_integer.cpp b/src/shader_recompiler/backend/glasm/emit_glasm_integer.cpp index f55c26b76..8aa494a4d 100644 --- a/src/shader_recompiler/backend/glasm/emit_glasm_integer.cpp +++ b/src/shader_recompiler/backend/glasm/emit_glasm_integer.cpp @@ -90,6 +90,14 @@ void EmitIMul32(EmitContext& ctx, IR::Inst& inst, ScalarS32 a, ScalarS32 b) { ctx.Add("MUL.S {}.x,{},{};", inst, a, b); } +void EmitSDiv32(EmitContext& ctx, IR::Inst& inst, ScalarS32 a, ScalarS32 b) { + ctx.Add("DIV.S {}.x,{},{};", inst, a, b); +} + +void EmitUDiv32(EmitContext& ctx, IR::Inst& inst, ScalarU32 a, ScalarU32 b) { + ctx.Add("DIV.U {}.x,{},{};", inst, a, b); +} + void EmitINeg32(EmitContext& ctx, IR::Inst& inst, ScalarS32 value) { if (value.type != Type::Register && static_cast<s32>(value.imm_u32) < 0) { ctx.Add("MOV.S {},{};", inst, -static_cast<s32>(value.imm_u32)); diff --git a/src/shader_recompiler/backend/glasm/emit_glasm_not_implemented.cpp b/src/shader_recompiler/backend/glasm/emit_glasm_not_implemented.cpp index e537f6073..681aeda8d 100644 --- a/src/shader_recompiler/backend/glasm/emit_glasm_not_implemented.cpp +++ b/src/shader_recompiler/backend/glasm/emit_glasm_not_implemented.cpp @@ -210,6 +210,10 @@ void EmitYDirection(EmitContext& ctx, IR::Inst& inst) { ctx.Add("MOV.F {}.x,y_direction[0].w;", inst); } +void EmitResolutionDownFactor(EmitContext& ctx, IR::Inst& inst) { + ctx.Add("MOV.F {}.x,scaling[0].z;", inst); +} + void EmitUndefU1(EmitContext& ctx, IR::Inst& inst) { ctx.Add("MOV.S {}.x,0;", inst); } diff --git a/src/shader_recompiler/backend/glsl/emit_context.cpp b/src/shader_recompiler/backend/glsl/emit_context.cpp index 4e6f2c0fe..97bd59302 100644 --- a/src/shader_recompiler/backend/glsl/emit_context.cpp +++ b/src/shader_recompiler/backend/glsl/emit_context.cpp @@ -393,6 +393,9 @@ EmitContext::EmitContext(IR::Program& program, Bindings& bindings, const Profile DefineGenericOutput(index, program.invocations); } } + if (info.uses_rescaling_uniform) { + header += "layout(location=0) uniform vec4 scaling;"; + } DefineConstantBuffers(bindings); DefineStorageBuffers(bindings); SetupImages(bindings); diff --git a/src/shader_recompiler/backend/glsl/emit_glsl_context_get_set.cpp b/src/shader_recompiler/backend/glsl/emit_glsl_context_get_set.cpp index 170db269a..4c26f3829 100644 --- a/src/shader_recompiler/backend/glsl/emit_glsl_context_get_set.cpp +++ b/src/shader_recompiler/backend/glsl/emit_glsl_context_get_set.cpp @@ -445,6 +445,10 @@ void EmitYDirection(EmitContext& ctx, IR::Inst& inst) { ctx.AddF32("{}=gl_FrontMaterial.ambient.a;", inst); } +void EmitResolutionDownFactor(EmitContext& ctx, IR::Inst& inst) { + ctx.AddF32("{}=scaling.z;", inst); +} + void EmitLoadLocal(EmitContext& ctx, IR::Inst& inst, std::string_view word_offset) { ctx.AddU32("{}=lmem[{}];", inst, word_offset); } diff --git a/src/shader_recompiler/backend/glsl/emit_glsl_image.cpp b/src/shader_recompiler/backend/glsl/emit_glsl_image.cpp index 447eb8e0a..2f78d0267 100644 --- a/src/shader_recompiler/backend/glsl/emit_glsl_image.cpp +++ b/src/shader_recompiler/backend/glsl/emit_glsl_image.cpp @@ -612,6 +612,22 @@ void EmitImageAtomicExchange32(EmitContext& ctx, IR::Inst& inst, const IR::Value value); } +void EmitIsTextureScaled(EmitContext& ctx, IR::Inst& inst, const IR::Value& index) { + if (!index.IsImmediate()) { + throw NotImplementedException("Non-constant texture rescaling"); + } + const u32 image_index{index.U32()}; + ctx.AddU1("{}=(ftou(scaling.x)&{})!=0;", inst, 1u << image_index); +} + +void EmitIsImageScaled(EmitContext& ctx, IR::Inst& inst, const IR::Value& index) { + if (!index.IsImmediate()) { + throw NotImplementedException("Non-constant texture rescaling"); + } + const u32 image_index{index.U32()}; + ctx.AddU1("{}=(ftou(scaling.y)&{})!=0;", inst, 1u << image_index); +} + void EmitBindlessImageSampleImplicitLod(EmitContext&) { NotImplemented(); } diff --git a/src/shader_recompiler/backend/glsl/emit_glsl_instructions.h b/src/shader_recompiler/backend/glsl/emit_glsl_instructions.h index 5936d086f..f86502e4c 100644 --- a/src/shader_recompiler/backend/glsl/emit_glsl_instructions.h +++ b/src/shader_recompiler/backend/glsl/emit_glsl_instructions.h @@ -85,6 +85,7 @@ void EmitInvocationId(EmitContext& ctx, IR::Inst& inst); void EmitSampleId(EmitContext& ctx, IR::Inst& inst); void EmitIsHelperInvocation(EmitContext& ctx, IR::Inst& inst); void EmitYDirection(EmitContext& ctx, IR::Inst& inst); +void EmitResolutionDownFactor(EmitContext& ctx, IR::Inst& inst); void EmitLoadLocal(EmitContext& ctx, IR::Inst& inst, std::string_view word_offset); void EmitWriteLocal(EmitContext& ctx, std::string_view word_offset, std::string_view value); void EmitUndefU1(EmitContext& ctx, IR::Inst& inst); @@ -362,6 +363,8 @@ void EmitIAdd64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::strin void EmitISub32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b); void EmitISub64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b); void EmitIMul32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b); +void EmitSDiv32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b); +void EmitUDiv32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b); void EmitINeg32(EmitContext& ctx, IR::Inst& inst, std::string_view value); void EmitINeg64(EmitContext& ctx, IR::Inst& inst, std::string_view value); void EmitIAbs32(EmitContext& ctx, IR::Inst& inst, std::string_view value); @@ -627,6 +630,8 @@ void EmitImageRead(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, std::string_view coords); void EmitImageWrite(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, std::string_view coords, std::string_view color); +void EmitIsTextureScaled(EmitContext& ctx, IR::Inst& inst, const IR::Value& index); +void EmitIsImageScaled(EmitContext& ctx, IR::Inst& inst, const IR::Value& index); void EmitBindlessImageAtomicIAdd32(EmitContext&); void EmitBindlessImageAtomicSMin32(EmitContext&); void EmitBindlessImageAtomicUMin32(EmitContext&); diff --git a/src/shader_recompiler/backend/glsl/emit_glsl_integer.cpp b/src/shader_recompiler/backend/glsl/emit_glsl_integer.cpp index 38419f88f..88c1d4c5e 100644 --- a/src/shader_recompiler/backend/glsl/emit_glsl_integer.cpp +++ b/src/shader_recompiler/backend/glsl/emit_glsl_integer.cpp @@ -78,6 +78,14 @@ void EmitIMul32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::strin ctx.AddU32("{}=uint({}*{});", inst, a, b); } +void EmitSDiv32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) { + ctx.AddU32("{}=uint(int({})/int({}));", inst, a, b); +} + +void EmitUDiv32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) { + ctx.AddU32("{}={}/{};", inst, a, b); +} + void EmitINeg32(EmitContext& ctx, IR::Inst& inst, std::string_view value) { ctx.AddU32("{}=uint(-({}));", inst, value); } diff --git a/src/shader_recompiler/backend/spirv/emit_context.cpp b/src/shader_recompiler/backend/spirv/emit_context.cpp index 2885e6799..723455462 100644 --- a/src/shader_recompiler/backend/spirv/emit_context.cpp +++ b/src/shader_recompiler/backend/spirv/emit_context.cpp @@ -7,11 +7,14 @@ #include <climits> #include <string_view> +#include <boost/container/static_vector.hpp> + #include <fmt/format.h> #include "common/common_types.h" #include "common/div_ceil.h" #include "shader_recompiler/backend/spirv/emit_context.h" +#include "shader_recompiler/backend/spirv/emit_spirv.h" namespace Shader::Backend::SPIRV { namespace { @@ -430,15 +433,33 @@ Id DescType(EmitContext& ctx, Id sampled_type, Id pointer_type, u32 count) { } } -size_t FindNextUnusedLocation(const std::bitset<IR::NUM_GENERICS>& used_locations, - size_t start_offset) { +size_t FindAndSetNextUnusedLocation(std::bitset<IR::NUM_GENERICS>& used_locations, + size_t& start_offset) { for (size_t location = start_offset; location < used_locations.size(); ++location) { if (!used_locations.test(location)) { + start_offset = location; + used_locations.set(location); return location; } } throw RuntimeError("Unable to get an unused location for legacy attribute"); } + +Id DefineLegacyInput(EmitContext& ctx, std::bitset<IR::NUM_GENERICS>& used_locations, + size_t& start_offset) { + const Id id{DefineInput(ctx, ctx.F32[4], true)}; + const size_t location = FindAndSetNextUnusedLocation(used_locations, start_offset); + ctx.Decorate(id, spv::Decoration::Location, location); + return id; +} + +Id DefineLegacyOutput(EmitContext& ctx, std::bitset<IR::NUM_GENERICS>& used_locations, + size_t& start_offset, std::optional<u32> invocations) { + const Id id{DefineOutput(ctx, ctx.F32[4], invocations)}; + const size_t location = FindAndSetNextUnusedLocation(used_locations, start_offset); + ctx.Decorate(id, spv::Decoration::Location, location); + return id; +} } // Anonymous namespace void VectorTypes::Define(Sirit::Module& sirit_ctx, Id base_type, std::string_view name) { @@ -456,8 +477,9 @@ void VectorTypes::Define(Sirit::Module& sirit_ctx, Id base_type, std::string_vie EmitContext::EmitContext(const Profile& profile_, const RuntimeInfo& runtime_info_, IR::Program& program, Bindings& bindings) - : Sirit::Module(profile_.supported_spirv), profile{profile_}, - runtime_info{runtime_info_}, stage{program.stage} { + : Sirit::Module(profile_.supported_spirv), profile{profile_}, runtime_info{runtime_info_}, + stage{program.stage}, texture_rescaling_index{bindings.texture_scaling_index}, + image_rescaling_index{bindings.image_scaling_index} { const bool is_unified{profile.unified_descriptor_binding}; u32& uniform_binding{is_unified ? bindings.unified : bindings.uniform_buffer}; u32& storage_binding{is_unified ? bindings.unified : bindings.storage_buffer}; @@ -474,10 +496,11 @@ EmitContext::EmitContext(const Profile& profile_, const RuntimeInfo& runtime_inf DefineStorageBuffers(program.info, storage_binding); DefineTextureBuffers(program.info, texture_binding); DefineImageBuffers(program.info, image_binding); - DefineTextures(program.info, texture_binding); - DefineImages(program.info, image_binding); + DefineTextures(program.info, texture_binding, bindings.texture_scaling_index); + DefineImages(program.info, image_binding, bindings.image_scaling_index); DefineAttributeMemAccess(program.info); DefineGlobalMemoryFunctions(program.info); + DefineRescalingInput(program.info); } EmitContext::~EmitContext() = default; @@ -520,6 +543,64 @@ Id EmitContext::BitOffset16(const IR::Value& offset) { return OpBitwiseAnd(U32[1], OpShiftLeftLogical(U32[1], Def(offset), Const(3u)), Const(16u)); } +Id EmitContext::InputLegacyAttribute(IR::Attribute attribute) { + if (attribute >= IR::Attribute::ColorFrontDiffuseR && + attribute <= IR::Attribute::ColorFrontDiffuseA) { + return input_front_color; + } + if (attribute >= IR::Attribute::ColorFrontSpecularR && + attribute <= IR::Attribute::ColorFrontSpecularA) { + return input_front_secondary_color; + } + if (attribute >= IR::Attribute::ColorBackDiffuseR && + attribute <= IR::Attribute::ColorBackDiffuseA) { + return input_back_color; + } + if (attribute >= IR::Attribute::ColorBackSpecularR && + attribute <= IR::Attribute::ColorBackSpecularA) { + return input_back_secondary_color; + } + if (attribute == IR::Attribute::FogCoordinate) { + return input_fog_frag_coord; + } + if (attribute >= IR::Attribute::FixedFncTexture0S && + attribute <= IR::Attribute::FixedFncTexture9Q) { + u32 index = + (static_cast<u32>(attribute) - static_cast<u32>(IR::Attribute::FixedFncTexture0S)) / 4; + return input_fixed_fnc_textures[index]; + } + throw InvalidArgument("Attribute is not legacy attribute {}", attribute); +} + +Id EmitContext::OutputLegacyAttribute(IR::Attribute attribute) { + if (attribute >= IR::Attribute::ColorFrontDiffuseR && + attribute <= IR::Attribute::ColorFrontDiffuseA) { + return output_front_color; + } + if (attribute >= IR::Attribute::ColorFrontSpecularR && + attribute <= IR::Attribute::ColorFrontSpecularA) { + return output_front_secondary_color; + } + if (attribute >= IR::Attribute::ColorBackDiffuseR && + attribute <= IR::Attribute::ColorBackDiffuseA) { + return output_back_color; + } + if (attribute >= IR::Attribute::ColorBackSpecularR && + attribute <= IR::Attribute::ColorBackSpecularA) { + return output_back_secondary_color; + } + if (attribute == IR::Attribute::FogCoordinate) { + return output_fog_frag_coord; + } + if (attribute >= IR::Attribute::FixedFncTexture0S && + attribute <= IR::Attribute::FixedFncTexture9Q) { + u32 index = + (static_cast<u32>(attribute) - static_cast<u32>(IR::Attribute::FixedFncTexture0S)) / 4; + return output_fixed_fnc_textures[index]; + } + throw InvalidArgument("Attribute is not legacy attribute {}", attribute); +} + void EmitContext::DefineCommonTypes(const Info& info) { void_id = TypeVoid(); @@ -920,6 +1001,73 @@ void EmitContext::DefineGlobalMemoryFunctions(const Info& info) { define(&StorageDefinitions::U32x4, storage_types.U32x4, U32[4], sizeof(u32[4])); } +void EmitContext::DefineRescalingInput(const Info& info) { + if (!info.uses_rescaling_uniform) { + return; + } + if (profile.unified_descriptor_binding) { + DefineRescalingInputPushConstant(); + } else { + DefineRescalingInputUniformConstant(); + } +} + +void EmitContext::DefineRescalingInputPushConstant() { + boost::container::static_vector<Id, 3> members{}; + u32 member_index{0}; + + rescaling_textures_type = TypeArray(U32[1], Const(4u)); + Decorate(rescaling_textures_type, spv::Decoration::ArrayStride, 4u); + members.push_back(rescaling_textures_type); + rescaling_textures_member_index = member_index++; + + rescaling_images_type = TypeArray(U32[1], Const(NUM_IMAGE_SCALING_WORDS)); + Decorate(rescaling_images_type, spv::Decoration::ArrayStride, 4u); + members.push_back(rescaling_images_type); + rescaling_images_member_index = member_index++; + + if (stage != Stage::Compute) { + members.push_back(F32[1]); + rescaling_downfactor_member_index = member_index++; + } + const Id push_constant_struct{TypeStruct(std::span(members.data(), members.size()))}; + Decorate(push_constant_struct, spv::Decoration::Block); + Name(push_constant_struct, "ResolutionInfo"); + + MemberDecorate(push_constant_struct, rescaling_textures_member_index, spv::Decoration::Offset, + static_cast<u32>(offsetof(RescalingLayout, rescaling_textures))); + MemberName(push_constant_struct, rescaling_textures_member_index, "rescaling_textures"); + + MemberDecorate(push_constant_struct, rescaling_images_member_index, spv::Decoration::Offset, + static_cast<u32>(offsetof(RescalingLayout, rescaling_images))); + MemberName(push_constant_struct, rescaling_images_member_index, "rescaling_images"); + + if (stage != Stage::Compute) { + MemberDecorate(push_constant_struct, rescaling_downfactor_member_index, + spv::Decoration::Offset, + static_cast<u32>(offsetof(RescalingLayout, down_factor))); + MemberName(push_constant_struct, rescaling_downfactor_member_index, "down_factor"); + } + const Id pointer_type{TypePointer(spv::StorageClass::PushConstant, push_constant_struct)}; + rescaling_push_constants = AddGlobalVariable(pointer_type, spv::StorageClass::PushConstant); + Name(rescaling_push_constants, "rescaling_push_constants"); + + if (profile.supported_spirv >= 0x00010400) { + interfaces.push_back(rescaling_push_constants); + } +} + +void EmitContext::DefineRescalingInputUniformConstant() { + const Id pointer_type{TypePointer(spv::StorageClass::UniformConstant, F32[4])}; + rescaling_uniform_constant = + AddGlobalVariable(pointer_type, spv::StorageClass::UniformConstant); + Decorate(rescaling_uniform_constant, spv::Decoration::Location, 0u); + + if (profile.supported_spirv >= 0x00010400) { + interfaces.push_back(rescaling_uniform_constant); + } +} + void EmitContext::DefineConstantBuffers(const Info& info, u32& binding) { if (info.constant_buffer_descriptors.empty()) { return; @@ -1108,7 +1256,7 @@ void EmitContext::DefineImageBuffers(const Info& info, u32& binding) { } } -void EmitContext::DefineTextures(const Info& info, u32& binding) { +void EmitContext::DefineTextures(const Info& info, u32& binding, u32& scaling_index) { textures.reserve(info.texture_descriptors.size()); for (const TextureDescriptor& desc : info.texture_descriptors) { const Id image_type{ImageType(*this, desc)}; @@ -1130,13 +1278,14 @@ void EmitContext::DefineTextures(const Info& info, u32& binding) { interfaces.push_back(id); } ++binding; + ++scaling_index; } if (info.uses_atomic_image_u32) { image_u32 = TypePointer(spv::StorageClass::Image, U32[1]); } } -void EmitContext::DefineImages(const Info& info, u32& binding) { +void EmitContext::DefineImages(const Info& info, u32& binding, u32& scaling_index) { images.reserve(info.image_descriptors.size()); for (const ImageDescriptor& desc : info.image_descriptors) { if (desc.count != 1) { @@ -1157,6 +1306,7 @@ void EmitContext::DefineImages(const Info& info, u32& binding) { interfaces.push_back(id); } ++binding; + ++scaling_index; } } @@ -1279,22 +1429,26 @@ void EmitContext::DefineInputs(const IR::Program& program) { } size_t previous_unused_location = 0; if (loads.AnyComponent(IR::Attribute::ColorFrontDiffuseR)) { - const size_t location = FindNextUnusedLocation(used_locations, previous_unused_location); - previous_unused_location = location; - used_locations.set(location); - const Id id{DefineInput(*this, F32[4], true)}; - Decorate(id, spv::Decoration::Location, location); - input_front_color = id; + input_front_color = DefineLegacyInput(*this, used_locations, previous_unused_location); + } + if (loads.AnyComponent(IR::Attribute::ColorFrontSpecularR)) { + input_front_secondary_color = + DefineLegacyInput(*this, used_locations, previous_unused_location); + } + if (loads.AnyComponent(IR::Attribute::ColorBackDiffuseR)) { + input_back_color = DefineLegacyInput(*this, used_locations, previous_unused_location); + } + if (loads.AnyComponent(IR::Attribute::ColorBackSpecularR)) { + input_back_secondary_color = + DefineLegacyInput(*this, used_locations, previous_unused_location); + } + if (loads.AnyComponent(IR::Attribute::FogCoordinate)) { + input_fog_frag_coord = DefineLegacyInput(*this, used_locations, previous_unused_location); } for (size_t index = 0; index < NUM_FIXEDFNCTEXTURE; ++index) { if (loads.AnyComponent(IR::Attribute::FixedFncTexture0S + index * 4)) { - const size_t location = - FindNextUnusedLocation(used_locations, previous_unused_location); - previous_unused_location = location; - used_locations.set(location); - const Id id{DefineInput(*this, F32[4], true)}; - Decorate(id, spv::Decoration::Location, location); - input_fixed_fnc_textures[index] = id; + input_fixed_fnc_textures[index] = + DefineLegacyInput(*this, used_locations, previous_unused_location); } } if (stage == Stage::TessellationEval) { @@ -1356,22 +1510,29 @@ void EmitContext::DefineOutputs(const IR::Program& program) { } size_t previous_unused_location = 0; if (info.stores.AnyComponent(IR::Attribute::ColorFrontDiffuseR)) { - const size_t location = FindNextUnusedLocation(used_locations, previous_unused_location); - previous_unused_location = location; - used_locations.set(location); - const Id id{DefineOutput(*this, F32[4], invocations)}; - Decorate(id, spv::Decoration::Location, static_cast<u32>(location)); - output_front_color = id; + output_front_color = + DefineLegacyOutput(*this, used_locations, previous_unused_location, invocations); + } + if (info.stores.AnyComponent(IR::Attribute::ColorFrontSpecularR)) { + output_front_secondary_color = + DefineLegacyOutput(*this, used_locations, previous_unused_location, invocations); + } + if (info.stores.AnyComponent(IR::Attribute::ColorBackDiffuseR)) { + output_back_color = + DefineLegacyOutput(*this, used_locations, previous_unused_location, invocations); + } + if (info.stores.AnyComponent(IR::Attribute::ColorBackSpecularR)) { + output_back_secondary_color = + DefineLegacyOutput(*this, used_locations, previous_unused_location, invocations); + } + if (info.stores.AnyComponent(IR::Attribute::FogCoordinate)) { + output_fog_frag_coord = + DefineLegacyOutput(*this, used_locations, previous_unused_location, invocations); } for (size_t index = 0; index < NUM_FIXEDFNCTEXTURE; ++index) { if (info.stores.AnyComponent(IR::Attribute::FixedFncTexture0S + index * 4)) { - const size_t location = - FindNextUnusedLocation(used_locations, previous_unused_location); - previous_unused_location = location; - used_locations.set(location); - const Id id{DefineOutput(*this, F32[4], invocations)}; - Decorate(id, spv::Decoration::Location, location); - output_fixed_fnc_textures[index] = id; + output_fixed_fnc_textures[index] = + DefineLegacyOutput(*this, used_locations, previous_unused_location, invocations); } } switch (stage) { diff --git a/src/shader_recompiler/backend/spirv/emit_context.h b/src/shader_recompiler/backend/spirv/emit_context.h index 847d0c0e6..63f8185d9 100644 --- a/src/shader_recompiler/backend/spirv/emit_context.h +++ b/src/shader_recompiler/backend/spirv/emit_context.h @@ -113,6 +113,9 @@ public: [[nodiscard]] Id BitOffset8(const IR::Value& offset); [[nodiscard]] Id BitOffset16(const IR::Value& offset); + Id InputLegacyAttribute(IR::Attribute attribute); + Id OutputLegacyAttribute(IR::Attribute attribute); + Id Const(u32 value) { return Constant(U32[1], value); } @@ -235,6 +238,16 @@ public: Id indexed_load_func{}; Id indexed_store_func{}; + Id rescaling_uniform_constant{}; + Id rescaling_push_constants{}; + Id rescaling_textures_type{}; + Id rescaling_images_type{}; + u32 rescaling_textures_member_index{}; + u32 rescaling_images_member_index{}; + u32 rescaling_downfactor_member_index{}; + u32 texture_rescaling_index{}; + u32 image_rescaling_index{}; + Id local_memory{}; Id shared_memory_u8{}; @@ -269,12 +282,20 @@ public: Id input_position{}; Id input_front_color{}; + Id input_front_secondary_color{}; + Id input_back_color{}; + Id input_back_secondary_color{}; + Id input_fog_frag_coord{}; std::array<Id, 10> input_fixed_fnc_textures{}; std::array<Id, 32> input_generics{}; Id output_point_size{}; Id output_position{}; Id output_front_color{}; + Id output_front_secondary_color{}; + Id output_back_color{}; + Id output_back_secondary_color{}; + Id output_fog_frag_coord{}; std::array<Id, 10> output_fixed_fnc_textures{}; std::array<std::array<GenericElementInfo, 4>, 32> output_generics{}; @@ -299,10 +320,13 @@ private: void DefineStorageBuffers(const Info& info, u32& binding); void DefineTextureBuffers(const Info& info, u32& binding); void DefineImageBuffers(const Info& info, u32& binding); - void DefineTextures(const Info& info, u32& binding); - void DefineImages(const Info& info, u32& binding); + void DefineTextures(const Info& info, u32& binding, u32& scaling_index); + void DefineImages(const Info& info, u32& binding, u32& scaling_index); void DefineAttributeMemAccess(const Info& info); void DefineGlobalMemoryFunctions(const Info& info); + void DefineRescalingInput(const Info& info); + void DefineRescalingInputPushConstant(); + void DefineRescalingInputUniformConstant(); void DefineInputs(const IR::Program& program); void DefineOutputs(const IR::Program& program); diff --git a/src/shader_recompiler/backend/spirv/emit_spirv.h b/src/shader_recompiler/backend/spirv/emit_spirv.h index db0c935fe..4b25534ce 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv.h +++ b/src/shader_recompiler/backend/spirv/emit_spirv.h @@ -16,6 +16,19 @@ namespace Shader::Backend::SPIRV { +constexpr u32 NUM_TEXTURE_SCALING_WORDS = 4; +constexpr u32 NUM_IMAGE_SCALING_WORDS = 2; +constexpr u32 NUM_TEXTURE_AND_IMAGE_SCALING_WORDS = + NUM_TEXTURE_SCALING_WORDS + NUM_IMAGE_SCALING_WORDS; + +struct RescalingLayout { + alignas(16) std::array<u32, NUM_TEXTURE_SCALING_WORDS> rescaling_textures; + alignas(16) std::array<u32, NUM_IMAGE_SCALING_WORDS> rescaling_images; + alignas(16) u32 down_factor; +}; +constexpr u32 RESCALING_LAYOUT_WORDS_OFFSET = offsetof(RescalingLayout, rescaling_textures); +constexpr u32 RESCALING_LAYOUT_DOWN_FACTOR_OFFSET = offsetof(RescalingLayout, down_factor); + [[nodiscard]] std::vector<u32> EmitSPIRV(const Profile& profile, const RuntimeInfo& runtime_info, IR::Program& program, Bindings& bindings); diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp index 6f60c6574..bac683ae1 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp @@ -43,23 +43,12 @@ Id AttrPointer(EmitContext& ctx, Id pointer_type, Id vertex, Id base, Args&&... } } -bool IsFixedFncTexture(IR::Attribute attribute) { - return attribute >= IR::Attribute::FixedFncTexture0S && - attribute <= IR::Attribute::FixedFncTexture9Q; -} - -u32 FixedFncTextureAttributeIndex(IR::Attribute attribute) { - if (!IsFixedFncTexture(attribute)) { - throw InvalidArgument("Attribute {} is not a FixedFncTexture", attribute); - } - return (static_cast<u32>(attribute) - static_cast<u32>(IR::Attribute::FixedFncTexture0S)) / 4u; -} - -u32 FixedFncTextureAttributeElement(IR::Attribute attribute) { - if (!IsFixedFncTexture(attribute)) { - throw InvalidArgument("Attribute {} is not a FixedFncTexture", attribute); - } - return static_cast<u32>(attribute) % 4u; +bool IsLegacyAttribute(IR::Attribute attribute) { + return (attribute >= IR::Attribute::ColorFrontDiffuseR && + attribute <= IR::Attribute::ColorBackSpecularA) || + attribute == IR::Attribute::FogCoordinate || + (attribute >= IR::Attribute::FixedFncTexture0S && + attribute <= IR::Attribute::FixedFncTexture9Q); } template <typename... Args> @@ -93,12 +82,16 @@ std::optional<OutAttr> OutputAttrPointer(EmitContext& ctx, IR::Attribute attr) { return OutputAccessChain(ctx, ctx.output_f32, info.id, index_id); } } - if (IsFixedFncTexture(attr)) { - const u32 index{FixedFncTextureAttributeIndex(attr)}; - const u32 element{FixedFncTextureAttributeElement(attr)}; - const Id element_id{ctx.Const(element)}; - return OutputAccessChain(ctx, ctx.output_f32, ctx.output_fixed_fnc_textures[index], - element_id); + if (IsLegacyAttribute(attr)) { + if (attr == IR::Attribute::FogCoordinate) { + return OutputAccessChain(ctx, ctx.output_f32, ctx.OutputLegacyAttribute(attr), + ctx.Const(0u)); + } else { + const u32 element{static_cast<u32>(attr) % 4}; + const Id element_id{ctx.Const(element)}; + return OutputAccessChain(ctx, ctx.output_f32, ctx.OutputLegacyAttribute(attr), + element_id); + } } switch (attr) { case IR::Attribute::PointSize: @@ -111,14 +104,6 @@ std::optional<OutAttr> OutputAttrPointer(EmitContext& ctx, IR::Attribute attr) { const Id element_id{ctx.Const(element)}; return OutputAccessChain(ctx, ctx.output_f32, ctx.output_position, element_id); } - case IR::Attribute::ColorFrontDiffuseR: - case IR::Attribute::ColorFrontDiffuseG: - case IR::Attribute::ColorFrontDiffuseB: - case IR::Attribute::ColorFrontDiffuseA: { - const u32 element{static_cast<u32>(attr) % 4}; - const Id element_id{ctx.Const(element)}; - return OutputAccessChain(ctx, ctx.output_f32, ctx.output_front_color, element_id); - } case IR::Attribute::ClipDistance0: case IR::Attribute::ClipDistance1: case IR::Attribute::ClipDistance2: @@ -341,11 +326,17 @@ Id EmitGetAttribute(EmitContext& ctx, IR::Attribute attr, Id vertex) { const Id value{ctx.OpLoad(type->id, pointer)}; return type->needs_cast ? ctx.OpBitcast(ctx.F32[1], value) : value; } - if (IsFixedFncTexture(attr)) { - const u32 index{FixedFncTextureAttributeIndex(attr)}; - const Id attr_id{ctx.input_fixed_fnc_textures[index]}; - const Id attr_ptr{AttrPointer(ctx, ctx.input_f32, vertex, attr_id, ctx.Const(element))}; - return ctx.OpLoad(ctx.F32[1], attr_ptr); + if (IsLegacyAttribute(attr)) { + if (attr == IR::Attribute::FogCoordinate) { + const Id attr_ptr{AttrPointer(ctx, ctx.input_f32, vertex, + ctx.InputLegacyAttribute(attr), ctx.Const(0u))}; + return ctx.OpLoad(ctx.F32[1], attr_ptr); + } else { + const Id element_id{ctx.Const(element)}; + const Id attr_ptr{AttrPointer(ctx, ctx.input_f32, vertex, + ctx.InputLegacyAttribute(attr), element_id)}; + return ctx.OpLoad(ctx.F32[1], attr_ptr); + } } switch (attr) { case IR::Attribute::PrimitiveId: @@ -356,13 +347,6 @@ Id EmitGetAttribute(EmitContext& ctx, IR::Attribute attr, Id vertex) { case IR::Attribute::PositionW: return ctx.OpLoad(ctx.F32[1], AttrPointer(ctx, ctx.input_f32, vertex, ctx.input_position, ctx.Const(element))); - case IR::Attribute::ColorFrontDiffuseR: - case IR::Attribute::ColorFrontDiffuseG: - case IR::Attribute::ColorFrontDiffuseB: - case IR::Attribute::ColorFrontDiffuseA: { - return ctx.OpLoad(ctx.F32[1], AttrPointer(ctx, ctx.input_f32, vertex, ctx.input_front_color, - ctx.Const(element))); - } case IR::Attribute::InstanceId: if (ctx.profile.support_vertex_instance_id) { return ctx.OpBitcast(ctx.F32[1], ctx.OpLoad(ctx.U32[1], ctx.instance_id)); @@ -542,6 +526,18 @@ Id EmitYDirection(EmitContext& ctx) { return ctx.Const(ctx.runtime_info.y_negate ? -1.0f : 1.0f); } +Id EmitResolutionDownFactor(EmitContext& ctx) { + if (ctx.profile.unified_descriptor_binding) { + const Id pointer_type{ctx.TypePointer(spv::StorageClass::PushConstant, ctx.F32[1])}; + const Id index{ctx.Const(ctx.rescaling_downfactor_member_index)}; + const Id pointer{ctx.OpAccessChain(pointer_type, ctx.rescaling_push_constants, index)}; + return ctx.OpLoad(ctx.F32[1], pointer); + } else { + const Id composite{ctx.OpLoad(ctx.F32[4], ctx.rescaling_uniform_constant)}; + return ctx.OpCompositeExtract(ctx.F32[1], composite, 2u); + } +} + Id EmitLoadLocal(EmitContext& ctx, Id word_offset) { const Id pointer{ctx.OpAccessChain(ctx.private_u32, ctx.local_memory, word_offset)}; return ctx.OpLoad(ctx.U32[1], pointer); diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp index 1d5364309..4d168a96d 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp @@ -224,6 +224,36 @@ Id Emit(MethodPtrType sparse_ptr, MethodPtrType non_sparse_ptr, EmitContext& ctx Decorate(ctx, inst, sample); return ctx.OpCompositeExtract(result_type, sample, 1U); } + +Id IsScaled(EmitContext& ctx, const IR::Value& index, Id member_index, u32 base_index) { + const Id push_constant_u32{ctx.TypePointer(spv::StorageClass::PushConstant, ctx.U32[1])}; + Id bit{}; + if (index.IsImmediate()) { + // Use BitwiseAnd instead of BitfieldExtract for better codegen on Nvidia OpenGL. + // LOP32I.NZ is used to set the predicate rather than BFE+ISETP. + const u32 index_value{index.U32() + base_index}; + const Id word_index{ctx.Const(index_value / 32)}; + const Id bit_index_mask{ctx.Const(1u << (index_value % 32))}; + const Id pointer{ctx.OpAccessChain(push_constant_u32, ctx.rescaling_push_constants, + member_index, word_index)}; + const Id word{ctx.OpLoad(ctx.U32[1], pointer)}; + bit = ctx.OpBitwiseAnd(ctx.U32[1], word, bit_index_mask); + } else { + Id index_value{ctx.Def(index)}; + if (base_index != 0) { + index_value = ctx.OpIAdd(ctx.U32[1], index_value, ctx.Const(base_index)); + } + const Id bit_index{ctx.OpBitwiseAnd(ctx.U32[1], index_value, ctx.Const(31u))}; + bit = ctx.OpBitFieldUExtract(ctx.U32[1], index_value, bit_index, ctx.Const(1u)); + } + return ctx.OpINotEqual(ctx.U1, bit, ctx.u32_zero_value); +} + +Id BitTest(EmitContext& ctx, Id mask, Id bit) { + const Id shifted{ctx.OpShiftRightLogical(ctx.U32[1], mask, bit)}; + const Id bit_value{ctx.OpBitwiseAnd(ctx.U32[1], shifted, ctx.Const(1u))}; + return ctx.OpINotEqual(ctx.U1, bit_value, ctx.u32_zero_value); +} } // Anonymous namespace Id EmitBindlessImageSampleImplicitLod(EmitContext&) { @@ -470,4 +500,28 @@ void EmitImageWrite(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id ctx.OpImageWrite(Image(ctx, index, info), coords, color); } +Id EmitIsTextureScaled(EmitContext& ctx, const IR::Value& index) { + if (ctx.profile.unified_descriptor_binding) { + const Id member_index{ctx.Const(ctx.rescaling_textures_member_index)}; + return IsScaled(ctx, index, member_index, ctx.texture_rescaling_index); + } else { + const Id composite{ctx.OpLoad(ctx.F32[4], ctx.rescaling_uniform_constant)}; + const Id mask_f32{ctx.OpCompositeExtract(ctx.F32[1], composite, 0u)}; + const Id mask{ctx.OpBitcast(ctx.U32[1], mask_f32)}; + return BitTest(ctx, mask, ctx.Def(index)); + } +} + +Id EmitIsImageScaled(EmitContext& ctx, const IR::Value& index) { + if (ctx.profile.unified_descriptor_binding) { + const Id member_index{ctx.Const(ctx.rescaling_images_member_index)}; + return IsScaled(ctx, index, member_index, ctx.image_rescaling_index); + } else { + const Id composite{ctx.OpLoad(ctx.F32[4], ctx.rescaling_uniform_constant)}; + const Id mask_f32{ctx.OpCompositeExtract(ctx.F32[1], composite, 1u)}; + const Id mask{ctx.OpBitcast(ctx.U32[1], mask_f32)}; + return BitTest(ctx, mask, ctx.Def(index)); + } +} + } // namespace Shader::Backend::SPIRV diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_instructions.h b/src/shader_recompiler/backend/spirv/emit_spirv_instructions.h index c9db1c164..6cd22dd3e 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_instructions.h +++ b/src/shader_recompiler/backend/spirv/emit_spirv_instructions.h @@ -75,6 +75,7 @@ Id EmitInvocationId(EmitContext& ctx); Id EmitSampleId(EmitContext& ctx); Id EmitIsHelperInvocation(EmitContext& ctx); Id EmitYDirection(EmitContext& ctx); +Id EmitResolutionDownFactor(EmitContext& ctx); Id EmitLoadLocal(EmitContext& ctx, Id word_offset); void EmitWriteLocal(EmitContext& ctx, Id word_offset, Id value); Id EmitUndefU1(EmitContext& ctx); @@ -283,6 +284,8 @@ Id EmitIAdd64(EmitContext& ctx, Id a, Id b); Id EmitISub32(EmitContext& ctx, Id a, Id b); Id EmitISub64(EmitContext& ctx, Id a, Id b); Id EmitIMul32(EmitContext& ctx, Id a, Id b); +Id EmitSDiv32(EmitContext& ctx, Id a, Id b); +Id EmitUDiv32(EmitContext& ctx, Id a, Id b); Id EmitINeg32(EmitContext& ctx, Id value); Id EmitINeg64(EmitContext& ctx, Id value); Id EmitIAbs32(EmitContext& ctx, Id value); @@ -510,6 +513,8 @@ Id EmitImageGradient(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, I Id derivates, Id offset, Id lod_clamp); Id EmitImageRead(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords); void EmitImageWrite(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords, Id color); +Id EmitIsTextureScaled(EmitContext& ctx, const IR::Value& index); +Id EmitIsImageScaled(EmitContext& ctx, const IR::Value& index); Id EmitBindlessImageAtomicIAdd32(EmitContext&); Id EmitBindlessImageAtomicSMin32(EmitContext&); Id EmitBindlessImageAtomicUMin32(EmitContext&); diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_integer.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_integer.cpp index 3501d7495..50277eec3 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_integer.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_integer.cpp @@ -72,6 +72,14 @@ Id EmitIMul32(EmitContext& ctx, Id a, Id b) { return ctx.OpIMul(ctx.U32[1], a, b); } +Id EmitSDiv32(EmitContext& ctx, Id a, Id b) { + return ctx.OpSDiv(ctx.U32[1], a, b); +} + +Id EmitUDiv32(EmitContext& ctx, Id a, Id b) { + return ctx.OpUDiv(ctx.U32[1], a, b); +} + Id EmitINeg32(EmitContext& ctx, Id value) { return ctx.OpSNegate(ctx.U32[1], value); } diff --git a/src/shader_recompiler/frontend/ir/basic_block.cpp b/src/shader_recompiler/frontend/ir/basic_block.cpp index 7c08b25ce..974efa4a0 100644 --- a/src/shader_recompiler/frontend/ir/basic_block.cpp +++ b/src/shader_recompiler/frontend/ir/basic_block.cpp @@ -22,6 +22,11 @@ void Block::AppendNewInst(Opcode op, std::initializer_list<Value> args) { PrependNewInst(end(), op, args); } +Block::iterator Block::PrependNewInst(iterator insertion_point, const Inst& base_inst) { + Inst* const inst{inst_pool->Create(base_inst)}; + return instructions.insert(insertion_point, *inst); +} + Block::iterator Block::PrependNewInst(iterator insertion_point, Opcode op, std::initializer_list<Value> args, u32 flags) { Inst* const inst{inst_pool->Create(op, flags)}; diff --git a/src/shader_recompiler/frontend/ir/basic_block.h b/src/shader_recompiler/frontend/ir/basic_block.h index 9ce1ed07e..fbfe98266 100644 --- a/src/shader_recompiler/frontend/ir/basic_block.h +++ b/src/shader_recompiler/frontend/ir/basic_block.h @@ -40,6 +40,9 @@ public: /// Appends a new instruction to the end of this basic block. void AppendNewInst(Opcode op, std::initializer_list<Value> args); + /// Prepends a copy of an instruction to this basic block before the insertion point. + iterator PrependNewInst(iterator insertion_point, const Inst& base_inst); + /// Prepends a new instruction to this basic block before the insertion point. iterator PrependNewInst(iterator insertion_point, Opcode op, std::initializer_list<Value> args = {}, u32 flags = 0); diff --git a/src/shader_recompiler/frontend/ir/ir_emitter.cpp b/src/shader_recompiler/frontend/ir/ir_emitter.cpp index 13159a68d..356f889ac 100644 --- a/src/shader_recompiler/frontend/ir/ir_emitter.cpp +++ b/src/shader_recompiler/frontend/ir/ir_emitter.cpp @@ -375,6 +375,10 @@ F32 IREmitter::YDirection() { return Inst<F32>(Opcode::YDirection); } +F32 IREmitter::ResolutionDownFactor() { + return Inst<F32>(Opcode::ResolutionDownFactor); +} + U32 IREmitter::LaneId() { return Inst<U32>(Opcode::LaneId); } @@ -1141,6 +1145,10 @@ U32 IREmitter::IMul(const U32& a, const U32& b) { return Inst<U32>(Opcode::IMul32, a, b); } +U32 IREmitter::IDiv(const U32& a, const U32& b, bool is_signed) { + return Inst<U32>(is_signed ? Opcode::SDiv32 : Opcode::UDiv32, a, b); +} + U32U64 IREmitter::INeg(const U32U64& value) { switch (value.Type()) { case Type::U32: @@ -1938,6 +1946,14 @@ Value IREmitter::ImageAtomicExchange(const Value& handle, const Value& coords, c return Inst(op, Flags{info}, handle, coords, value); } +U1 IREmitter::IsTextureScaled(const U32& index) { + return Inst<U1>(Opcode::IsTextureScaled, index); +} + +U1 IREmitter::IsImageScaled(const U32& index) { + return Inst<U1>(Opcode::IsImageScaled, index); +} + U1 IREmitter::VoteAll(const U1& value) { return Inst<U1>(Opcode::VoteAll, value); } diff --git a/src/shader_recompiler/frontend/ir/ir_emitter.h b/src/shader_recompiler/frontend/ir/ir_emitter.h index 1b89ca5a0..13eefa88b 100644 --- a/src/shader_recompiler/frontend/ir/ir_emitter.h +++ b/src/shader_recompiler/frontend/ir/ir_emitter.h @@ -102,6 +102,8 @@ public: [[nodiscard]] U1 IsHelperInvocation(); [[nodiscard]] F32 YDirection(); + [[nodiscard]] F32 ResolutionDownFactor(); + [[nodiscard]] U32 LaneId(); [[nodiscard]] U32 LoadGlobalU8(const U64& address); @@ -207,6 +209,7 @@ public: [[nodiscard]] U32U64 IAdd(const U32U64& a, const U32U64& b); [[nodiscard]] U32U64 ISub(const U32U64& a, const U32U64& b); [[nodiscard]] U32 IMul(const U32& a, const U32& b); + [[nodiscard]] U32 IDiv(const U32& a, const U32& b, bool is_signed = false); [[nodiscard]] U32U64 INeg(const U32U64& value); [[nodiscard]] U32 IAbs(const U32& value); [[nodiscard]] U32U64 ShiftLeftLogical(const U32U64& base, const U32& shift); @@ -356,6 +359,10 @@ public: TextureInstInfo info); [[nodiscard]] Value ImageAtomicExchange(const Value& handle, const Value& coords, const Value& value, TextureInstInfo info); + + [[nodiscard]] U1 IsTextureScaled(const U32& index); + [[nodiscard]] U1 IsImageScaled(const U32& index); + [[nodiscard]] U1 VoteAll(const U1& value); [[nodiscard]] U1 VoteAny(const U1& value); [[nodiscard]] U1 VoteEqual(const U1& value); diff --git a/src/shader_recompiler/frontend/ir/microinstruction.cpp b/src/shader_recompiler/frontend/ir/microinstruction.cpp index 30b470bdd..97e2bf6af 100644 --- a/src/shader_recompiler/frontend/ir/microinstruction.cpp +++ b/src/shader_recompiler/frontend/ir/microinstruction.cpp @@ -47,6 +47,17 @@ Inst::Inst(IR::Opcode op_, u32 flags_) noexcept : op{op_}, flags{flags_} { } } +Inst::Inst(const Inst& base) : op{base.op}, flags{base.flags} { + if (base.op == Opcode::Phi) { + throw NotImplementedException("Copying phi node"); + } + std::construct_at(&args); + const size_t num_args{base.NumArgs()}; + for (size_t index = 0; index < num_args; ++index) { + SetArg(index, base.Arg(index)); + } +} + Inst::~Inst() { if (op == Opcode::Phi) { std::destroy_at(&phi_args); diff --git a/src/shader_recompiler/frontend/ir/opcodes.inc b/src/shader_recompiler/frontend/ir/opcodes.inc index d91098c80..6929919df 100644 --- a/src/shader_recompiler/frontend/ir/opcodes.inc +++ b/src/shader_recompiler/frontend/ir/opcodes.inc @@ -62,6 +62,7 @@ OPCODE(InvocationId, U32, OPCODE(SampleId, U32, ) OPCODE(IsHelperInvocation, U1, ) OPCODE(YDirection, F32, ) +OPCODE(ResolutionDownFactor, F32, ) // Undefined OPCODE(UndefU1, U1, ) @@ -286,6 +287,8 @@ OPCODE(IAdd64, U64, U64, OPCODE(ISub32, U32, U32, U32, ) OPCODE(ISub64, U64, U64, U64, ) OPCODE(IMul32, U32, U32, U32, ) +OPCODE(SDiv32, U32, U32, U32, ) +OPCODE(UDiv32, U32, U32, U32, ) OPCODE(INeg32, U32, U32, ) OPCODE(INeg64, U64, U64, ) OPCODE(IAbs32, U32, U32, ) @@ -490,6 +493,9 @@ OPCODE(ImageGradient, F32x4, Opaq OPCODE(ImageRead, U32x4, Opaque, Opaque, ) OPCODE(ImageWrite, Void, Opaque, Opaque, U32x4, ) +OPCODE(IsTextureScaled, U1, U32, ) +OPCODE(IsImageScaled, U1, U32, ) + // Atomic Image operations OPCODE(BindlessImageAtomicIAdd32, U32, U32, Opaque, U32, ) diff --git a/src/shader_recompiler/frontend/ir/value.h b/src/shader_recompiler/frontend/ir/value.h index 6c9ef6bdd..947579852 100644 --- a/src/shader_recompiler/frontend/ir/value.h +++ b/src/shader_recompiler/frontend/ir/value.h @@ -116,10 +116,10 @@ public: class Inst : public boost::intrusive::list_base_hook<> { public: explicit Inst(IR::Opcode op_, u32 flags_) noexcept; + explicit Inst(const Inst& base); ~Inst(); Inst& operator=(const Inst&) = delete; - Inst(const Inst&) = delete; Inst& operator=(Inst&&) = delete; Inst(Inst&&) = delete; diff --git a/src/shader_recompiler/frontend/maxwell/translate_program.cpp b/src/shader_recompiler/frontend/maxwell/translate_program.cpp index 2fc542f0e..267ebe4af 100644 --- a/src/shader_recompiler/frontend/maxwell/translate_program.cpp +++ b/src/shader_recompiler/frontend/maxwell/translate_program.cpp @@ -179,6 +179,10 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo Optimization::TexturePass(env, program); Optimization::ConstantPropagationPass(program); + + if (Settings::values.resolution_info.active) { + Optimization::RescalingPass(program); + } Optimization::DeadCodeEliminationPass(program); if (Settings::values.renderer_debug) { Optimization::VerificationPass(program); diff --git a/src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp b/src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp index f69e1c9cc..1e476d83d 100644 --- a/src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp +++ b/src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp @@ -430,6 +430,11 @@ void VisitUsages(Info& info, IR::Inst& inst) { case IR::Opcode::IsHelperInvocation: info.uses_is_helper_invocation = true; break; + case IR::Opcode::ResolutionDownFactor: + case IR::Opcode::IsTextureScaled: + case IR::Opcode::IsImageScaled: + info.uses_rescaling_uniform = true; + break; case IR::Opcode::LaneId: info.uses_subgroup_invocation_id = true; break; diff --git a/src/shader_recompiler/ir_opt/passes.h b/src/shader_recompiler/ir_opt/passes.h index 2f89b1ea0..f877c7ba0 100644 --- a/src/shader_recompiler/ir_opt/passes.h +++ b/src/shader_recompiler/ir_opt/passes.h @@ -19,6 +19,7 @@ void GlobalMemoryToStorageBufferPass(IR::Program& program); void IdentityRemovalPass(IR::Program& program); void LowerFp16ToFp32(IR::Program& program); void LowerInt64ToInt32(IR::Program& program); +void RescalingPass(IR::Program& program); void SsaRewritePass(IR::Program& program); void TexturePass(Environment& env, IR::Program& program); void VerificationPass(const IR::Program& program); diff --git a/src/shader_recompiler/ir_opt/rescaling_pass.cpp b/src/shader_recompiler/ir_opt/rescaling_pass.cpp new file mode 100644 index 000000000..c28500dd1 --- /dev/null +++ b/src/shader_recompiler/ir_opt/rescaling_pass.cpp @@ -0,0 +1,327 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include "common/alignment.h" +#include "common/settings.h" +#include "shader_recompiler/environment.h" +#include "shader_recompiler/frontend/ir/ir_emitter.h" +#include "shader_recompiler/frontend/ir/modifiers.h" +#include "shader_recompiler/frontend/ir/program.h" +#include "shader_recompiler/frontend/ir/value.h" +#include "shader_recompiler/ir_opt/passes.h" +#include "shader_recompiler/shader_info.h" + +namespace Shader::Optimization { +namespace { +[[nodiscard]] bool IsTextureTypeRescalable(TextureType type) { + switch (type) { + case TextureType::Color2D: + case TextureType::ColorArray2D: + return true; + case TextureType::Color1D: + case TextureType::ColorArray1D: + case TextureType::Color3D: + case TextureType::ColorCube: + case TextureType::ColorArrayCube: + case TextureType::Buffer: + break; + } + return false; +} + +void VisitMark(IR::Block& block, IR::Inst& inst) { + switch (inst.GetOpcode()) { + case IR::Opcode::ShuffleIndex: + case IR::Opcode::ShuffleUp: + case IR::Opcode::ShuffleDown: + case IR::Opcode::ShuffleButterfly: { + const IR::Value shfl_arg{inst.Arg(0)}; + if (shfl_arg.IsImmediate()) { + break; + } + const IR::Inst* const arg_inst{shfl_arg.InstRecursive()}; + if (arg_inst->GetOpcode() != IR::Opcode::BitCastU32F32) { + break; + } + const IR::Value bitcast_arg{arg_inst->Arg(0)}; + if (bitcast_arg.IsImmediate()) { + break; + } + IR::Inst* const bitcast_inst{bitcast_arg.InstRecursive()}; + bool must_patch_outside = false; + if (bitcast_inst->GetOpcode() == IR::Opcode::GetAttribute) { + const IR::Attribute attr{bitcast_inst->Arg(0).Attribute()}; + switch (attr) { + case IR::Attribute::PositionX: + case IR::Attribute::PositionY: + bitcast_inst->SetFlags<u32>(0xDEADBEEF); + must_patch_outside = true; + break; + default: + break; + } + } + if (must_patch_outside) { + const auto it{IR::Block::InstructionList::s_iterator_to(inst)}; + IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)}; + const IR::F32 new_inst{&*block.PrependNewInst(it, inst)}; + const IR::F32 up_factor{ir.FPRecip(ir.ResolutionDownFactor())}; + const IR::Value converted{ir.FPMul(new_inst, up_factor)}; + inst.ReplaceUsesWith(converted); + } + break; + } + + default: + break; + } +} + +void PatchFragCoord(IR::Block& block, IR::Inst& inst) { + IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)}; + const IR::F32 down_factor{ir.ResolutionDownFactor()}; + const IR::F32 frag_coord{ir.GetAttribute(inst.Arg(0).Attribute())}; + const IR::F32 downscaled_frag_coord{ir.FPMul(frag_coord, down_factor)}; + inst.ReplaceUsesWith(downscaled_frag_coord); +} + +void PatchPointSize(IR::Block& block, IR::Inst& inst) { + IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)}; + const IR::F32 point_value{inst.Arg(1)}; + const IR::F32 up_factor{ir.FPRecip(ir.ResolutionDownFactor())}; + const IR::F32 upscaled_point_value{ir.FPMul(point_value, up_factor)}; + inst.SetArg(1, upscaled_point_value); +} + +[[nodiscard]] IR::U32 Scale(IR::IREmitter& ir, const IR::U1& is_scaled, const IR::U32& value) { + IR::U32 scaled_value{value}; + if (const u32 up_scale = Settings::values.resolution_info.up_scale; up_scale != 1) { + scaled_value = ir.IMul(scaled_value, ir.Imm32(up_scale)); + } + if (const u32 down_shift = Settings::values.resolution_info.down_shift; down_shift != 0) { + scaled_value = ir.ShiftRightArithmetic(scaled_value, ir.Imm32(down_shift)); + } + return IR::U32{ir.Select(is_scaled, scaled_value, value)}; +} + +[[nodiscard]] IR::U32 SubScale(IR::IREmitter& ir, const IR::U1& is_scaled, const IR::U32& value, + const IR::Attribute attrib) { + const IR::F32 up_factor{ir.Imm32(Settings::values.resolution_info.up_factor)}; + const IR::F32 base{ir.FPMul(ir.ConvertUToF(32, 32, value), up_factor)}; + const IR::F32 frag_coord{ir.GetAttribute(attrib)}; + const IR::F32 down_factor{ir.Imm32(Settings::values.resolution_info.down_factor)}; + const IR::F32 floor{ir.FPMul(up_factor, ir.FPFloor(ir.FPMul(frag_coord, down_factor)))}; + const IR::F16F32F64 deviation{ir.FPAdd(base, ir.FPAdd(frag_coord, ir.FPNeg(floor)))}; + return IR::U32{ir.Select(is_scaled, ir.ConvertFToU(32, deviation), value)}; +} + +[[nodiscard]] IR::U32 DownScale(IR::IREmitter& ir, const IR::U1& is_scaled, const IR::U32& value) { + IR::U32 scaled_value{value}; + if (const u32 down_shift = Settings::values.resolution_info.down_shift; down_shift != 0) { + scaled_value = ir.ShiftLeftLogical(scaled_value, ir.Imm32(down_shift)); + } + if (const u32 up_scale = Settings::values.resolution_info.up_scale; up_scale != 1) { + scaled_value = ir.IDiv(scaled_value, ir.Imm32(up_scale)); + } + return IR::U32{ir.Select(is_scaled, scaled_value, value)}; +} + +void PatchImageQueryDimensions(IR::Block& block, IR::Inst& inst) { + const auto it{IR::Block::InstructionList::s_iterator_to(inst)}; + IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)}; + const auto info{inst.Flags<IR::TextureInstInfo>()}; + const IR::U1 is_scaled{ir.IsTextureScaled(ir.Imm32(info.descriptor_index))}; + switch (info.type) { + case TextureType::Color2D: + case TextureType::ColorArray2D: { + const IR::Value new_inst{&*block.PrependNewInst(it, inst)}; + const IR::U32 width{DownScale(ir, is_scaled, IR::U32{ir.CompositeExtract(new_inst, 0)})}; + const IR::U32 height{DownScale(ir, is_scaled, IR::U32{ir.CompositeExtract(new_inst, 1)})}; + const IR::Value replacement{ir.CompositeConstruct( + width, height, ir.CompositeExtract(new_inst, 2), ir.CompositeExtract(new_inst, 3))}; + inst.ReplaceUsesWith(replacement); + break; + } + case TextureType::Color1D: + case TextureType::ColorArray1D: + case TextureType::Color3D: + case TextureType::ColorCube: + case TextureType::ColorArrayCube: + case TextureType::Buffer: + // Nothing to patch here + break; + } +} + +void ScaleIntegerComposite(IR::IREmitter& ir, IR::Inst& inst, const IR::U1& is_scaled, + size_t index) { + const IR::Value composite{inst.Arg(index)}; + if (composite.IsEmpty()) { + return; + } + const auto info{inst.Flags<IR::TextureInstInfo>()}; + const IR::U32 x{Scale(ir, is_scaled, IR::U32{ir.CompositeExtract(composite, 0)})}; + const IR::U32 y{Scale(ir, is_scaled, IR::U32{ir.CompositeExtract(composite, 1)})}; + switch (info.type) { + case TextureType::Color2D: + inst.SetArg(index, ir.CompositeConstruct(x, y)); + break; + case TextureType::ColorArray2D: { + const IR::U32 z{ir.CompositeExtract(composite, 2)}; + inst.SetArg(index, ir.CompositeConstruct(x, y, z)); + break; + } + case TextureType::Color1D: + case TextureType::ColorArray1D: + case TextureType::Color3D: + case TextureType::ColorCube: + case TextureType::ColorArrayCube: + case TextureType::Buffer: + // Nothing to patch here + break; + } +} + +void SubScaleCoord(IR::IREmitter& ir, IR::Inst& inst, const IR::U1& is_scaled) { + const auto info{inst.Flags<IR::TextureInstInfo>()}; + const IR::Value coord{inst.Arg(1)}; + const IR::U32 coord_x{ir.CompositeExtract(coord, 0)}; + const IR::U32 coord_y{ir.CompositeExtract(coord, 1)}; + + const IR::U32 scaled_x{SubScale(ir, is_scaled, coord_x, IR::Attribute::PositionX)}; + const IR::U32 scaled_y{SubScale(ir, is_scaled, coord_y, IR::Attribute::PositionY)}; + switch (info.type) { + case TextureType::Color2D: + inst.SetArg(1, ir.CompositeConstruct(scaled_x, scaled_y)); + break; + case TextureType::ColorArray2D: { + const IR::U32 z{ir.CompositeExtract(coord, 2)}; + inst.SetArg(1, ir.CompositeConstruct(scaled_x, scaled_y, z)); + break; + } + case TextureType::Color1D: + case TextureType::ColorArray1D: + case TextureType::Color3D: + case TextureType::ColorCube: + case TextureType::ColorArrayCube: + case TextureType::Buffer: + // Nothing to patch here + break; + } +} + +void SubScaleImageFetch(IR::Block& block, IR::Inst& inst) { + IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)}; + const auto info{inst.Flags<IR::TextureInstInfo>()}; + if (!IsTextureTypeRescalable(info.type)) { + return; + } + const IR::U1 is_scaled{ir.IsTextureScaled(ir.Imm32(info.descriptor_index))}; + SubScaleCoord(ir, inst, is_scaled); + // Scale ImageFetch offset + ScaleIntegerComposite(ir, inst, is_scaled, 2); +} + +void SubScaleImageRead(IR::Block& block, IR::Inst& inst) { + IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)}; + const auto info{inst.Flags<IR::TextureInstInfo>()}; + if (!IsTextureTypeRescalable(info.type)) { + return; + } + const IR::U1 is_scaled{ir.IsImageScaled(ir.Imm32(info.descriptor_index))}; + SubScaleCoord(ir, inst, is_scaled); +} + +void PatchImageFetch(IR::Block& block, IR::Inst& inst) { + IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)}; + const auto info{inst.Flags<IR::TextureInstInfo>()}; + if (!IsTextureTypeRescalable(info.type)) { + return; + } + const IR::U1 is_scaled{ir.IsTextureScaled(ir.Imm32(info.descriptor_index))}; + ScaleIntegerComposite(ir, inst, is_scaled, 1); + // Scale ImageFetch offset + ScaleIntegerComposite(ir, inst, is_scaled, 2); +} + +void PatchImageRead(IR::Block& block, IR::Inst& inst) { + IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)}; + const auto info{inst.Flags<IR::TextureInstInfo>()}; + if (!IsTextureTypeRescalable(info.type)) { + return; + } + const IR::U1 is_scaled{ir.IsImageScaled(ir.Imm32(info.descriptor_index))}; + ScaleIntegerComposite(ir, inst, is_scaled, 1); +} + +void Visit(const IR::Program& program, IR::Block& block, IR::Inst& inst) { + const bool is_fragment_shader{program.stage == Stage::Fragment}; + switch (inst.GetOpcode()) { + case IR::Opcode::GetAttribute: { + const IR::Attribute attr{inst.Arg(0).Attribute()}; + switch (attr) { + case IR::Attribute::PositionX: + case IR::Attribute::PositionY: + if (is_fragment_shader && inst.Flags<u32>() != 0xDEADBEEF) { + PatchFragCoord(block, inst); + } + break; + default: + break; + } + break; + } + case IR::Opcode::SetAttribute: { + const IR::Attribute attr{inst.Arg(0).Attribute()}; + switch (attr) { + case IR::Attribute::PointSize: + if (inst.Flags<u32>() != 0xDEADBEEF) { + PatchPointSize(block, inst); + } + break; + default: + break; + } + break; + } + case IR::Opcode::ImageQueryDimensions: + PatchImageQueryDimensions(block, inst); + break; + case IR::Opcode::ImageFetch: + if (is_fragment_shader) { + SubScaleImageFetch(block, inst); + } else { + PatchImageFetch(block, inst); + } + break; + case IR::Opcode::ImageRead: + if (is_fragment_shader) { + SubScaleImageRead(block, inst); + } else { + PatchImageRead(block, inst); + } + break; + default: + break; + } +} +} // Anonymous namespace + +void RescalingPass(IR::Program& program) { + const bool is_fragment_shader{program.stage == Stage::Fragment}; + if (is_fragment_shader) { + for (IR::Block* const block : program.post_order_blocks) { + for (IR::Inst& inst : block->Instructions()) { + VisitMark(*block, inst); + } + } + } + for (IR::Block* const block : program.post_order_blocks) { + for (IR::Inst& inst : block->Instructions()) { + Visit(program, *block, inst); + } + } +} + +} // namespace Shader::Optimization diff --git a/src/shader_recompiler/shader_info.h b/src/shader_recompiler/shader_info.h index 4ef4dbd40..9f375c30e 100644 --- a/src/shader_recompiler/shader_info.h +++ b/src/shader_recompiler/shader_info.h @@ -172,6 +172,7 @@ struct Info { bool uses_global_memory{}; bool uses_atomic_image_u32{}; bool uses_shadow_lod{}; + bool uses_rescaling_uniform{}; IR::Type used_constant_buffer_types{}; IR::Type used_storage_buffer_types{}; @@ -190,4 +191,13 @@ struct Info { ImageDescriptors image_descriptors; }; +template <typename Descriptors> +u32 NumDescriptors(const Descriptors& descriptors) { + u32 num{}; + for (const auto& desc : descriptors) { + num += desc.count; + } + return num; +} + } // namespace Shader diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index 269db21a5..91a30fef7 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -15,6 +15,8 @@ add_library(video_core STATIC command_classes/codecs/codec.h command_classes/codecs/h264.cpp command_classes/codecs/h264.h + command_classes/codecs/vp8.cpp + command_classes/codecs/vp8.h command_classes/codecs/vp9.cpp command_classes/codecs/vp9.h command_classes/codecs/vp9_types.h @@ -130,6 +132,8 @@ add_library(video_core STATIC renderer_vulkan/vk_descriptor_pool.h renderer_vulkan/vk_fence_manager.cpp renderer_vulkan/vk_fence_manager.h + renderer_vulkan/vk_fsr.cpp + renderer_vulkan/vk_fsr.h renderer_vulkan/vk_graphics_pipeline.cpp renderer_vulkan/vk_graphics_pipeline.h renderer_vulkan/vk_master_semaphore.cpp diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index d350c9b36..43bed63ac 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -853,12 +853,14 @@ void BufferCache<P>::CommitAsyncFlushesHigh() { } if constexpr (USE_MEMORY_MAPS) { auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes); + runtime.PreCopyBarrier(); for (auto& [copy, buffer_id] : downloads) { // Have in mind the staging buffer offset for the copy copy.dst_offset += download_staging.offset; const std::array copies{copy}; - runtime.CopyBuffer(download_staging.buffer, slot_buffers[buffer_id], copies); + runtime.CopyBuffer(download_staging.buffer, slot_buffers[buffer_id], copies, false); } + runtime.PostCopyBarrier(); runtime.Finish(); for (const auto& [copy, buffer_id] : downloads) { const Buffer& buffer = slot_buffers[buffer_id]; diff --git a/src/video_core/command_classes/codecs/codec.cpp b/src/video_core/command_classes/codecs/codec.cpp index 61966cbfe..916277811 100644 --- a/src/video_core/command_classes/codecs/codec.cpp +++ b/src/video_core/command_classes/codecs/codec.cpp @@ -8,6 +8,7 @@ #include "common/settings.h" #include "video_core/command_classes/codecs/codec.h" #include "video_core/command_classes/codecs/h264.h" +#include "video_core/command_classes/codecs/vp8.h" #include "video_core/command_classes/codecs/vp9.h" #include "video_core/gpu.h" #include "video_core/memory_manager.h" @@ -46,6 +47,7 @@ void AVFrameDeleter(AVFrame* ptr) { Codec::Codec(GPU& gpu_, const NvdecCommon::NvdecRegisters& regs) : gpu(gpu_), state{regs}, h264_decoder(std::make_unique<Decoder::H264>(gpu)), + vp8_decoder(std::make_unique<Decoder::VP8>(gpu)), vp9_decoder(std::make_unique<Decoder::VP9>(gpu)) {} Codec::~Codec() { @@ -135,7 +137,9 @@ void Codec::Initialize() { switch (current_codec) { case NvdecCommon::VideoCodec::H264: return AV_CODEC_ID_H264; - case NvdecCommon::VideoCodec::Vp9: + case NvdecCommon::VideoCodec::VP8: + return AV_CODEC_ID_VP8; + case NvdecCommon::VideoCodec::VP9: return AV_CODEC_ID_VP9; default: UNIMPLEMENTED_MSG("Unknown codec {}", current_codec); @@ -176,19 +180,27 @@ void Codec::Decode() { return; } bool vp9_hidden_frame = false; - std::vector<u8> frame_data; - if (current_codec == NvdecCommon::VideoCodec::H264) { - frame_data = h264_decoder->ComposeFrameHeader(state, is_first_frame); - } else if (current_codec == NvdecCommon::VideoCodec::Vp9) { - frame_data = vp9_decoder->ComposeFrameHeader(state); - vp9_hidden_frame = vp9_decoder->WasFrameHidden(); - } + const auto& frame_data = [&]() { + switch (current_codec) { + case Tegra::NvdecCommon::VideoCodec::H264: + return h264_decoder->ComposeFrame(state, is_first_frame); + case Tegra::NvdecCommon::VideoCodec::VP8: + return vp8_decoder->ComposeFrame(state); + case Tegra::NvdecCommon::VideoCodec::VP9: + vp9_decoder->ComposeFrame(state); + vp9_hidden_frame = vp9_decoder->WasFrameHidden(); + return vp9_decoder->GetFrameBytes(); + default: + UNREACHABLE(); + return std::vector<u8>{}; + } + }(); AVPacketPtr packet{av_packet_alloc(), AVPacketDeleter}; if (!packet) { LOG_ERROR(Service_NVDRV, "av_packet_alloc failed"); return; } - packet->data = frame_data.data(); + packet->data = const_cast<u8*>(frame_data.data()); packet->size = static_cast<s32>(frame_data.size()); if (const int res = avcodec_send_packet(av_codec_ctx, packet.get()); res != 0) { LOG_DEBUG(Service_NVDRV, "avcodec_send_packet error {}", res); @@ -252,11 +264,11 @@ std::string_view Codec::GetCurrentCodecName() const { return "None"; case NvdecCommon::VideoCodec::H264: return "H264"; - case NvdecCommon::VideoCodec::Vp8: + case NvdecCommon::VideoCodec::VP8: return "VP8"; case NvdecCommon::VideoCodec::H265: return "H265"; - case NvdecCommon::VideoCodec::Vp9: + case NvdecCommon::VideoCodec::VP9: return "VP9"; default: return "Unknown"; diff --git a/src/video_core/command_classes/codecs/codec.h b/src/video_core/command_classes/codecs/codec.h index f9a80886f..13ed88382 100644 --- a/src/video_core/command_classes/codecs/codec.h +++ b/src/video_core/command_classes/codecs/codec.h @@ -29,6 +29,7 @@ using AVFramePtr = std::unique_ptr<AVFrame, decltype(&AVFrameDeleter)>; namespace Decoder { class H264; +class VP8; class VP9; } // namespace Decoder @@ -72,6 +73,7 @@ private: GPU& gpu; const NvdecCommon::NvdecRegisters& state; std::unique_ptr<Decoder::H264> h264_decoder; + std::unique_ptr<Decoder::VP8> vp8_decoder; std::unique_ptr<Decoder::VP9> vp9_decoder; std::queue<AVFramePtr> av_frames{}; diff --git a/src/video_core/command_classes/codecs/h264.cpp b/src/video_core/command_classes/codecs/h264.cpp index 5519c4705..84f1fa938 100644 --- a/src/video_core/command_classes/codecs/h264.cpp +++ b/src/video_core/command_classes/codecs/h264.cpp @@ -45,8 +45,8 @@ H264::H264(GPU& gpu_) : gpu(gpu_) {} H264::~H264() = default; -const std::vector<u8>& H264::ComposeFrameHeader(const NvdecCommon::NvdecRegisters& state, - bool is_first_frame) { +const std::vector<u8>& H264::ComposeFrame(const NvdecCommon::NvdecRegisters& state, + bool is_first_frame) { H264DecoderContext context; gpu.MemoryManager().ReadBlock(state.picture_info_offset, &context, sizeof(H264DecoderContext)); diff --git a/src/video_core/command_classes/codecs/h264.h b/src/video_core/command_classes/codecs/h264.h index bfe84a472..1899d8e7f 100644 --- a/src/video_core/command_classes/codecs/h264.h +++ b/src/video_core/command_classes/codecs/h264.h @@ -75,9 +75,9 @@ public: explicit H264(GPU& gpu); ~H264(); - /// Compose the H264 header of the frame for FFmpeg decoding - [[nodiscard]] const std::vector<u8>& ComposeFrameHeader( - const NvdecCommon::NvdecRegisters& state, bool is_first_frame = false); + /// Compose the H264 frame for FFmpeg decoding + [[nodiscard]] const std::vector<u8>& ComposeFrame(const NvdecCommon::NvdecRegisters& state, + bool is_first_frame = false); private: std::vector<u8> frame; diff --git a/src/video_core/command_classes/codecs/vp8.cpp b/src/video_core/command_classes/codecs/vp8.cpp new file mode 100644 index 000000000..32ad0ec16 --- /dev/null +++ b/src/video_core/command_classes/codecs/vp8.cpp @@ -0,0 +1,55 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <array> +#include <vector> + +#include "video_core/command_classes/codecs/vp8.h" +#include "video_core/gpu.h" +#include "video_core/memory_manager.h" + +namespace Tegra::Decoder { +VP8::VP8(GPU& gpu_) : gpu(gpu_) {} + +VP8::~VP8() = default; + +const std::vector<u8>& VP8::ComposeFrame(const NvdecCommon::NvdecRegisters& state) { + VP8PictureInfo info; + gpu.MemoryManager().ReadBlock(state.picture_info_offset, &info, sizeof(VP8PictureInfo)); + + const bool is_key_frame = info.key_frame == 1u; + const auto bitstream_size = static_cast<size_t>(info.vld_buffer_size); + const size_t header_size = is_key_frame ? 10u : 3u; + frame.resize(header_size + bitstream_size); + + // Based on page 30 of the VP8 specification. + // https://datatracker.ietf.org/doc/rfc6386/ + frame[0] = is_key_frame ? 0u : 1u; // 1-bit frame type (0: keyframe, 1: interframes). + frame[0] |= static_cast<u8>((info.version & 7u) << 1u); // 3-bit version number + frame[0] |= static_cast<u8>(1u << 4u); // 1-bit show_frame flag + + // The next 19-bits are the first partition size + frame[0] |= static_cast<u8>((info.first_part_size & 7u) << 5u); + frame[1] = static_cast<u8>((info.first_part_size & 0x7f8u) >> 3u); + frame[2] = static_cast<u8>((info.first_part_size & 0x7f800u) >> 11u); + + if (is_key_frame) { + frame[3] = 0x9du; + frame[4] = 0x01u; + frame[5] = 0x2au; + // TODO(ameerj): Horizontal/Vertical Scale + // 16 bits: (2 bits Horizontal Scale << 14) | Width (14 bits) + frame[6] = static_cast<u8>(info.frame_width & 0xff); + frame[7] = static_cast<u8>(((info.frame_width >> 8) & 0x3f)); + // 16 bits:(2 bits Vertical Scale << 14) | Height (14 bits) + frame[8] = static_cast<u8>(info.frame_height & 0xff); + frame[9] = static_cast<u8>(((info.frame_height >> 8) & 0x3f)); + } + const u64 bitstream_offset = state.frame_bitstream_offset; + gpu.MemoryManager().ReadBlock(bitstream_offset, frame.data() + header_size, bitstream_size); + + return frame; +} + +} // namespace Tegra::Decoder diff --git a/src/video_core/command_classes/codecs/vp8.h b/src/video_core/command_classes/codecs/vp8.h new file mode 100644 index 000000000..41fc7b403 --- /dev/null +++ b/src/video_core/command_classes/codecs/vp8.h @@ -0,0 +1,74 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <array> +#include <vector> + +#include "common/common_funcs.h" +#include "common/common_types.h" +#include "video_core/command_classes/nvdec_common.h" + +namespace Tegra { +class GPU; +namespace Decoder { + +class VP8 { +public: + explicit VP8(GPU& gpu); + ~VP8(); + + /// Compose the VP8 frame for FFmpeg decoding + [[nodiscard]] const std::vector<u8>& ComposeFrame(const NvdecCommon::NvdecRegisters& state); + +private: + std::vector<u8> frame; + GPU& gpu; + + struct VP8PictureInfo { + INSERT_PADDING_WORDS_NOINIT(14); + u16 frame_width; // actual frame width + u16 frame_height; // actual frame height + u8 key_frame; + u8 version; + union { + u8 raw; + BitField<0, 2, u8> tile_format; + BitField<2, 3, u8> gob_height; + BitField<5, 3, u8> reserverd_surface_format; + }; + u8 error_conceal_on; // 1: error conceal on; 0: off + u32 first_part_size; // the size of first partition(frame header and mb header partition) + u32 hist_buffer_size; // in units of 256 + u32 vld_buffer_size; // in units of 1 + // Current frame buffers + std::array<u32, 2> frame_stride; // [y_c] + u32 luma_top_offset; // offset of luma top field in units of 256 + u32 luma_bot_offset; // offset of luma bottom field in units of 256 + u32 luma_frame_offset; // offset of luma frame in units of 256 + u32 chroma_top_offset; // offset of chroma top field in units of 256 + u32 chroma_bot_offset; // offset of chroma bottom field in units of 256 + u32 chroma_frame_offset; // offset of chroma frame in units of 256 + + INSERT_PADDING_BYTES_NOINIT(0x1c); // NvdecDisplayParams + + // Decode picture buffer related + s8 current_output_memory_layout; + // output NV12/NV24 setting. index 0: golden; 1: altref; 2: last + std::array<s8, 3> output_memory_layout; + + u8 segmentation_feature_data_update; + INSERT_PADDING_BYTES_NOINIT(3); + + // ucode return result + u32 result_value; + std::array<u32, 8> partition_offset; + INSERT_PADDING_WORDS_NOINIT(3); + }; + static_assert(sizeof(VP8PictureInfo) == 0xc0, "PictureInfo is an invalid size"); +}; + +} // namespace Decoder +} // namespace Tegra diff --git a/src/video_core/command_classes/codecs/vp9.cpp b/src/video_core/command_classes/codecs/vp9.cpp index d7e749485..2c00181fa 100644 --- a/src/video_core/command_classes/codecs/vp9.cpp +++ b/src/video_core/command_classes/codecs/vp9.cpp @@ -770,7 +770,7 @@ VpxBitStreamWriter VP9::ComposeUncompressedHeader() { return uncomp_writer; } -const std::vector<u8>& VP9::ComposeFrameHeader(const NvdecCommon::NvdecRegisters& state) { +void VP9::ComposeFrame(const NvdecCommon::NvdecRegisters& state) { std::vector<u8> bitstream; { Vp9FrameContainer curr_frame = GetCurrentFrame(state); @@ -792,7 +792,6 @@ const std::vector<u8>& VP9::ComposeFrameHeader(const NvdecCommon::NvdecRegisters frame.begin() + uncompressed_header.size()); std::copy(bitstream.begin(), bitstream.end(), frame.begin() + uncompressed_header.size() + compressed_header.size()); - return frame; } VpxRangeEncoder::VpxRangeEncoder() { diff --git a/src/video_core/command_classes/codecs/vp9.h b/src/video_core/command_classes/codecs/vp9.h index e6e9fc17e..2e735c792 100644 --- a/src/video_core/command_classes/codecs/vp9.h +++ b/src/video_core/command_classes/codecs/vp9.h @@ -116,16 +116,20 @@ public: VP9(VP9&&) = default; VP9& operator=(VP9&&) = delete; - /// Composes the VP9 frame from the GPU state information. Based on the official VP9 spec - /// documentation - [[nodiscard]] const std::vector<u8>& ComposeFrameHeader( - const NvdecCommon::NvdecRegisters& state); + /// Composes the VP9 frame from the GPU state information. + /// Based on the official VP9 spec documentation + void ComposeFrame(const NvdecCommon::NvdecRegisters& state); /// Returns true if the most recent frame was a hidden frame. [[nodiscard]] bool WasFrameHidden() const { return !current_frame_info.show_frame; } + /// Returns a const reference to the composed frame data. + [[nodiscard]] const std::vector<u8>& GetFrameBytes() const { + return frame; + } + private: /// Generates compressed header probability updates in the bitstream writer template <typename T, std::size_t N> diff --git a/src/video_core/command_classes/nvdec.cpp b/src/video_core/command_classes/nvdec.cpp index b5c55f14a..9aaf5247e 100644 --- a/src/video_core/command_classes/nvdec.cpp +++ b/src/video_core/command_classes/nvdec.cpp @@ -35,7 +35,8 @@ AVFramePtr Nvdec::GetFrame() { void Nvdec::Execute() { switch (codec->GetCurrentCodec()) { case NvdecCommon::VideoCodec::H264: - case NvdecCommon::VideoCodec::Vp9: + case NvdecCommon::VideoCodec::VP8: + case NvdecCommon::VideoCodec::VP9: codec->Decode(); break; default: diff --git a/src/video_core/command_classes/nvdec_common.h b/src/video_core/command_classes/nvdec_common.h index 6a24e00a0..8a35c44a1 100644 --- a/src/video_core/command_classes/nvdec_common.h +++ b/src/video_core/command_classes/nvdec_common.h @@ -13,9 +13,9 @@ namespace Tegra::NvdecCommon { enum class VideoCodec : u64 { None = 0x0, H264 = 0x3, - Vp8 = 0x5, + VP8 = 0x5, H265 = 0x7, - Vp9 = 0x9, + VP9 = 0x9, }; // NVDEC should use a 32-bit address space, but is mapped to 64-bit, @@ -50,7 +50,10 @@ struct NvdecRegisters { u64 h264_last_surface_chroma_offset; ///< 0x0858 std::array<u64, 17> surface_luma_offset; ///< 0x0860 std::array<u64, 17> surface_chroma_offset; ///< 0x08E8 - INSERT_PADDING_WORDS_NOINIT(132); ///< 0x0970 + INSERT_PADDING_WORDS_NOINIT(68); ///< 0x0970 + u64 vp8_prob_data_offset; ///< 0x0A80 + u64 vp8_header_partition_buf_offset; ///< 0x0A88 + INSERT_PADDING_WORDS_NOINIT(60); ///< 0x0A90 u64 vp9_entropy_probs_offset; ///< 0x0B80 u64 vp9_backward_updates_offset; ///< 0x0B88 u64 vp9_last_frame_segmap_offset; ///< 0x0B90 @@ -81,6 +84,8 @@ ASSERT_REG_POSITION(h264_last_surface_luma_offset, 0x10A); ASSERT_REG_POSITION(h264_last_surface_chroma_offset, 0x10B); ASSERT_REG_POSITION(surface_luma_offset, 0x10C); ASSERT_REG_POSITION(surface_chroma_offset, 0x11D); +ASSERT_REG_POSITION(vp8_prob_data_offset, 0x150); +ASSERT_REG_POSITION(vp8_header_partition_buf_offset, 0x151); ASSERT_REG_POSITION(vp9_entropy_probs_offset, 0x170); ASSERT_REG_POSITION(vp9_backward_updates_offset, 0x171); ASSERT_REG_POSITION(vp9_last_frame_segmap_offset, 0x172); diff --git a/src/video_core/dirty_flags.h b/src/video_core/dirty_flags.h index f0d545f90..d63ad5a35 100644 --- a/src/video_core/dirty_flags.h +++ b/src/video_core/dirty_flags.h @@ -29,6 +29,8 @@ enum : u8 { ColorBuffer6, ColorBuffer7, ZetaBuffer, + RescaleViewports, + RescaleScissors, VertexBuffers, VertexBuffer0, diff --git a/src/video_core/host_shaders/CMakeLists.txt b/src/video_core/host_shaders/CMakeLists.txt index 20d748c12..d779a967a 100644 --- a/src/video_core/host_shaders/CMakeLists.txt +++ b/src/video_core/host_shaders/CMakeLists.txt @@ -1,3 +1,11 @@ +set(FIDELITYFX_INCLUDE_DIR ${CMAKE_SOURCE_DIR}/externals/FidelityFX-FSR/ffx-fsr) + +set(GLSL_INCLUDES + fidelityfx_fsr.comp + ${FIDELITYFX_INCLUDE_DIR}/ffx_a.h + ${FIDELITYFX_INCLUDE_DIR}/ffx_fsr1.h +) + set(SHADER_FILES astc_decoder.comp block_linear_unswizzle_2d.comp @@ -5,14 +13,25 @@ set(SHADER_FILES convert_depth_to_float.frag convert_float_to_depth.frag full_screen_triangle.vert + fxaa.frag + fxaa.vert opengl_copy_bc4.comp opengl_present.frag opengl_present.vert + opengl_present_scaleforce.frag pitch_unswizzle.comp + present_bicubic.frag + present_gaussian.frag vulkan_blit_color_float.frag vulkan_blit_depth_stencil.frag + vulkan_fidelityfx_fsr_easu_fp16.comp + vulkan_fidelityfx_fsr_easu_fp32.comp + vulkan_fidelityfx_fsr_rcas_fp16.comp + vulkan_fidelityfx_fsr_rcas_fp32.comp vulkan_present.frag vulkan_present.vert + vulkan_present_scaleforce_fp16.frag + vulkan_present_scaleforce_fp32.frag vulkan_quad_indexed.comp vulkan_uint8.comp ) @@ -76,7 +95,7 @@ foreach(FILENAME IN ITEMS ${SHADER_FILES}) OUTPUT ${SPIRV_HEADER_FILE} COMMAND - ${GLSLANGVALIDATOR} -V ${QUIET_FLAG} ${GLSL_FLAGS} --variable-name ${SPIRV_VARIABLE_NAME} -o ${SPIRV_HEADER_FILE} ${SOURCE_FILE} + ${GLSLANGVALIDATOR} -V ${QUIET_FLAG} -I"${FIDELITYFX_INCLUDE_DIR}" ${GLSL_FLAGS} --variable-name ${SPIRV_VARIABLE_NAME} -o ${SPIRV_HEADER_FILE} ${SOURCE_FILE} MAIN_DEPENDENCY ${SOURCE_FILE} ) @@ -84,9 +103,12 @@ foreach(FILENAME IN ITEMS ${SHADER_FILES}) endif() endforeach() +set(SHADER_SOURCES ${SHADER_FILES}) +list(APPEND SHADER_SOURCES ${GLSL_INCLUDES}) + add_custom_target(host_shaders DEPENDS ${SHADER_HEADERS} SOURCES - ${SHADER_FILES} + ${SHADER_SOURCES} ) diff --git a/src/video_core/host_shaders/fidelityfx_fsr.comp b/src/video_core/host_shaders/fidelityfx_fsr.comp new file mode 100644 index 000000000..6b97f789d --- /dev/null +++ b/src/video_core/host_shaders/fidelityfx_fsr.comp @@ -0,0 +1,116 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +//!#version 460 core +#extension GL_ARB_separate_shader_objects : enable +#extension GL_ARB_shading_language_420pack : enable +#extension GL_GOOGLE_include_directive : enable +#extension GL_EXT_shader_explicit_arithmetic_types : require + +// FidelityFX Super Resolution Sample +// +// Copyright (c) 2021 Advanced Micro Devices, Inc. All rights reserved. +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files(the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and / or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions : +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +layout( push_constant ) uniform constants { + uvec4 Const0; + uvec4 Const1; + uvec4 Const2; + uvec4 Const3; +}; + +layout(set=0,binding=0) uniform sampler2D InputTexture; +layout(set=0,binding=1,rgba16f) uniform image2D OutputTexture; + +#define A_GPU 1 +#define A_GLSL 1 + +#ifndef YUZU_USE_FP16 + #include "ffx_a.h" + + #if USE_EASU + #define FSR_EASU_F 1 + AF4 FsrEasuRF(AF2 p) { AF4 res = textureGather(InputTexture, p, 0); return res; } + AF4 FsrEasuGF(AF2 p) { AF4 res = textureGather(InputTexture, p, 1); return res; } + AF4 FsrEasuBF(AF2 p) { AF4 res = textureGather(InputTexture, p, 2); return res; } + #endif + #if USE_RCAS + #define FSR_RCAS_F 1 + AF4 FsrRcasLoadF(ASU2 p) { return texelFetch(InputTexture, ASU2(p), 0); } + void FsrRcasInputF(inout AF1 r, inout AF1 g, inout AF1 b) {} + #endif +#else + #define A_HALF + #include "ffx_a.h" + + #if USE_EASU + #define FSR_EASU_H 1 + AH4 FsrEasuRH(AF2 p) { AH4 res = AH4(textureGather(InputTexture, p, 0)); return res; } + AH4 FsrEasuGH(AF2 p) { AH4 res = AH4(textureGather(InputTexture, p, 1)); return res; } + AH4 FsrEasuBH(AF2 p) { AH4 res = AH4(textureGather(InputTexture, p, 2)); return res; } + #endif + #if USE_RCAS + #define FSR_RCAS_H 1 + AH4 FsrRcasLoadH(ASW2 p) { return AH4(texelFetch(InputTexture, ASU2(p), 0)); } + void FsrRcasInputH(inout AH1 r,inout AH1 g,inout AH1 b){} + #endif +#endif + +#include "ffx_fsr1.h" + +void CurrFilter(AU2 pos) { +#if USE_BILINEAR + AF2 pp = (AF2(pos) * AF2_AU2(Const0.xy) + AF2_AU2(Const0.zw)) * AF2_AU2(Const1.xy) + AF2(0.5, -0.5) * AF2_AU2(Const1.zw); + imageStore(OutputTexture, ASU2(pos), textureLod(InputTexture, pp, 0.0)); +#endif +#if USE_EASU + #ifndef YUZU_USE_FP16 + AF3 c; + FsrEasuF(c, pos, Const0, Const1, Const2, Const3); + imageStore(OutputTexture, ASU2(pos), AF4(c, 1)); + #else + AH3 c; + FsrEasuH(c, pos, Const0, Const1, Const2, Const3); + imageStore(OutputTexture, ASU2(pos), AH4(c, 1)); + #endif +#endif +#if USE_RCAS + #ifndef YUZU_USE_FP16 + AF3 c; + FsrRcasF(c.r, c.g, c.b, pos, Const0); + imageStore(OutputTexture, ASU2(pos), AF4(c, 1)); + #else + AH3 c; + FsrRcasH(c.r, c.g, c.b, pos, Const0); + imageStore(OutputTexture, ASU2(pos), AH4(c, 1)); + #endif +#endif +} + +layout(local_size_x=64) in; +void main() { + // Do remapping of local xy in workgroup for a more PS-like swizzle pattern. + AU2 gxy = ARmp8x8(gl_LocalInvocationID.x) + AU2(gl_WorkGroupID.x << 4u, gl_WorkGroupID.y << 4u); + CurrFilter(gxy); + gxy.x += 8u; + CurrFilter(gxy); + gxy.y += 8u; + CurrFilter(gxy); + gxy.x -= 8u; + CurrFilter(gxy); +} diff --git a/src/video_core/host_shaders/fxaa.frag b/src/video_core/host_shaders/fxaa.frag new file mode 100644 index 000000000..02f4068d1 --- /dev/null +++ b/src/video_core/host_shaders/fxaa.frag @@ -0,0 +1,76 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +// Source code is adapted from +// https://www.geeks3d.com/20110405/fxaa-fast-approximate-anti-aliasing-demo-glsl-opengl-test-radeon-geforce/3/ + +#version 460 + +#ifdef VULKAN + +#define BINDING_COLOR_TEXTURE 1 + +#else // ^^^ Vulkan ^^^ // vvv OpenGL vvv + +#define BINDING_COLOR_TEXTURE 0 + +#endif + +layout (location = 0) in vec4 posPos; + +layout (location = 0) out vec4 frag_color; + +layout (binding = BINDING_COLOR_TEXTURE) uniform sampler2D input_texture; + +const float FXAA_SPAN_MAX = 8.0; +const float FXAA_REDUCE_MUL = 1.0 / 8.0; +const float FXAA_REDUCE_MIN = 1.0 / 128.0; + +#define FxaaTexLod0(t, p) textureLod(t, p, 0.0) +#define FxaaTexOff(t, p, o) textureLodOffset(t, p, 0.0, o) + +vec3 FxaaPixelShader(vec4 posPos, sampler2D tex) { + + vec3 rgbNW = FxaaTexLod0(tex, posPos.zw).xyz; + vec3 rgbNE = FxaaTexOff(tex, posPos.zw, ivec2(1,0)).xyz; + vec3 rgbSW = FxaaTexOff(tex, posPos.zw, ivec2(0,1)).xyz; + vec3 rgbSE = FxaaTexOff(tex, posPos.zw, ivec2(1,1)).xyz; + vec3 rgbM = FxaaTexLod0(tex, posPos.xy).xyz; +/*---------------------------------------------------------*/ + vec3 luma = vec3(0.299, 0.587, 0.114); + float lumaNW = dot(rgbNW, luma); + float lumaNE = dot(rgbNE, luma); + float lumaSW = dot(rgbSW, luma); + float lumaSE = dot(rgbSE, luma); + float lumaM = dot(rgbM, luma); +/*---------------------------------------------------------*/ + float lumaMin = min(lumaM, min(min(lumaNW, lumaNE), min(lumaSW, lumaSE))); + float lumaMax = max(lumaM, max(max(lumaNW, lumaNE), max(lumaSW, lumaSE))); +/*---------------------------------------------------------*/ + vec2 dir; + dir.x = -((lumaNW + lumaNE) - (lumaSW + lumaSE)); + dir.y = ((lumaNW + lumaSW) - (lumaNE + lumaSE)); +/*---------------------------------------------------------*/ + float dirReduce = max( + (lumaNW + lumaNE + lumaSW + lumaSE) * (0.25 * FXAA_REDUCE_MUL), + FXAA_REDUCE_MIN); + float rcpDirMin = 1.0/(min(abs(dir.x), abs(dir.y)) + dirReduce); + dir = min(vec2( FXAA_SPAN_MAX, FXAA_SPAN_MAX), + max(vec2(-FXAA_SPAN_MAX, -FXAA_SPAN_MAX), + dir * rcpDirMin)) / textureSize(tex, 0); +/*--------------------------------------------------------*/ + vec3 rgbA = (1.0 / 2.0) * ( + FxaaTexLod0(tex, posPos.xy + dir * (1.0 / 3.0 - 0.5)).xyz + + FxaaTexLod0(tex, posPos.xy + dir * (2.0 / 3.0 - 0.5)).xyz); + vec3 rgbB = rgbA * (1.0 / 2.0) + (1.0 / 4.0) * ( + FxaaTexLod0(tex, posPos.xy + dir * (0.0 / 3.0 - 0.5)).xyz + + FxaaTexLod0(tex, posPos.xy + dir * (3.0 / 3.0 - 0.5)).xyz); + float lumaB = dot(rgbB, luma); + if((lumaB < lumaMin) || (lumaB > lumaMax)) return rgbA; + return rgbB; +} + +void main() { + frag_color = vec4(FxaaPixelShader(posPos, input_texture), 1.0); +} diff --git a/src/video_core/host_shaders/fxaa.vert b/src/video_core/host_shaders/fxaa.vert new file mode 100644 index 000000000..ac20c04e9 --- /dev/null +++ b/src/video_core/host_shaders/fxaa.vert @@ -0,0 +1,38 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#version 460 + +out gl_PerVertex { + vec4 gl_Position; +}; + +const vec2 vertices[4] = + vec2[4](vec2(-1.0, 1.0), vec2(1.0, 1.0), vec2(-1.0, -1.0), vec2(1.0, -1.0)); + +layout (location = 0) out vec4 posPos; + +#ifdef VULKAN + +#define BINDING_COLOR_TEXTURE 0 +#define VERTEX_ID gl_VertexIndex + +#else // ^^^ Vulkan ^^^ // vvv OpenGL vvv + +#define BINDING_COLOR_TEXTURE 0 +#define VERTEX_ID gl_VertexID + +#endif + +layout (binding = BINDING_COLOR_TEXTURE) uniform sampler2D input_texture; + +const float FXAA_SUBPIX_SHIFT = 0; + +void main() { + vec2 vertex = vertices[VERTEX_ID]; + gl_Position = vec4(vertex, 0.0, 1.0); + vec2 vert_tex_coord = (vertex + 1.0) / 2.0; + posPos.xy = vert_tex_coord; + posPos.zw = vert_tex_coord - (0.5 + FXAA_SUBPIX_SHIFT) / textureSize(input_texture, 0); +} diff --git a/src/video_core/host_shaders/opengl_present_scaleforce.frag b/src/video_core/host_shaders/opengl_present_scaleforce.frag new file mode 100644 index 000000000..71ff9e1e3 --- /dev/null +++ b/src/video_core/host_shaders/opengl_present_scaleforce.frag @@ -0,0 +1,130 @@ +// MIT License +// +// Copyright (c) 2020 BreadFish64 +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in all +// copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +// SOFTWARE. + +// Adapted from https://github.com/BreadFish64/ScaleFish/tree/master/scaleforce + +//! #version 460 + +#extension GL_ARB_separate_shader_objects : enable + +#ifdef YUZU_USE_FP16 + +#extension GL_AMD_gpu_shader_half_float : enable +#extension GL_NV_gpu_shader5 : enable + +#define lfloat float16_t +#define lvec2 f16vec2 +#define lvec3 f16vec3 +#define lvec4 f16vec4 + +#else + +#define lfloat float +#define lvec2 vec2 +#define lvec3 vec3 +#define lvec4 vec4 + +#endif + +#ifdef VULKAN + +#define BINDING_COLOR_TEXTURE 1 + +#else // ^^^ Vulkan ^^^ // vvv OpenGL vvv + +#define BINDING_COLOR_TEXTURE 0 + +#endif + +layout (location = 0) in vec2 tex_coord; + +layout (location = 0) out vec4 frag_color; + +layout (binding = BINDING_COLOR_TEXTURE) uniform sampler2D input_texture; + +const bool ignore_alpha = true; + +lfloat ColorDist1(lvec4 a, lvec4 b) { + // https://en.wikipedia.org/wiki/YCbCr#ITU-R_BT.2020_conversion + const lvec3 K = lvec3(0.2627, 0.6780, 0.0593); + const lfloat scaleB = lfloat(0.5) / (lfloat(1.0) - K.b); + const lfloat scaleR = lfloat(0.5) / (lfloat(1.0) - K.r); + lvec4 diff = a - b; + lfloat Y = dot(diff.rgb, K); + lfloat Cb = scaleB * (diff.b - Y); + lfloat Cr = scaleR * (diff.r - Y); + lvec3 YCbCr = lvec3(Y, Cb, Cr); + lfloat d = length(YCbCr); + if (ignore_alpha) { + return d; + } + return sqrt(a.a * b.a * d * d + diff.a * diff.a); +} + +lvec4 ColorDist(lvec4 ref, lvec4 A, lvec4 B, lvec4 C, lvec4 D) { + return lvec4( + ColorDist1(ref, A), + ColorDist1(ref, B), + ColorDist1(ref, C), + ColorDist1(ref, D) + ); +} + +vec4 Scaleforce(sampler2D tex, vec2 tex_coord) { + lvec4 bl = lvec4(textureOffset(tex, tex_coord, ivec2(-1, -1))); + lvec4 bc = lvec4(textureOffset(tex, tex_coord, ivec2(0, -1))); + lvec4 br = lvec4(textureOffset(tex, tex_coord, ivec2(1, -1))); + lvec4 cl = lvec4(textureOffset(tex, tex_coord, ivec2(-1, 0))); + lvec4 cc = lvec4(texture(tex, tex_coord)); + lvec4 cr = lvec4(textureOffset(tex, tex_coord, ivec2(1, 0))); + lvec4 tl = lvec4(textureOffset(tex, tex_coord, ivec2(-1, 1))); + lvec4 tc = lvec4(textureOffset(tex, tex_coord, ivec2(0, 1))); + lvec4 tr = lvec4(textureOffset(tex, tex_coord, ivec2(1, 1))); + + lvec4 offset_tl = ColorDist(cc, tl, tc, tr, cr); + lvec4 offset_br = ColorDist(cc, br, bc, bl, cl); + + // Calculate how different cc is from the texels around it + const lfloat plus_weight = lfloat(1.5); + const lfloat cross_weight = lfloat(1.5); + lfloat total_dist = dot(offset_tl + offset_br, lvec4(cross_weight, plus_weight, cross_weight, plus_weight)); + + if (total_dist == lfloat(0.0)) { + return cc; + } else { + // Add together all the distances with direction taken into account + lvec4 tmp = offset_tl - offset_br; + lvec2 total_offset = tmp.wy * plus_weight + (tmp.zz + lvec2(-tmp.x, tmp.x)) * cross_weight; + + // When the image has thin points, they tend to split apart. + // This is because the texels all around are different and total_offset reaches into clear areas. + // This works pretty well to keep the offset in bounds for these cases. + lfloat clamp_val = length(total_offset) / total_dist; + vec2 final_offset = vec2(clamp(total_offset, -clamp_val, clamp_val)) / textureSize(tex, 0); + + return texture(tex, tex_coord - final_offset); + } +} + +void main() { + frag_color = Scaleforce(input_texture, tex_coord); +} diff --git a/src/video_core/host_shaders/present_bicubic.frag b/src/video_core/host_shaders/present_bicubic.frag new file mode 100644 index 000000000..902b70c2b --- /dev/null +++ b/src/video_core/host_shaders/present_bicubic.frag @@ -0,0 +1,67 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#version 460 core + +#ifdef VULKAN + +#define BINDING_COLOR_TEXTURE 1 + +#else // ^^^ Vulkan ^^^ // vvv OpenGL vvv + +#define BINDING_COLOR_TEXTURE 0 + +#endif + + +layout (location = 0) in vec2 frag_tex_coord; + +layout (location = 0) out vec4 color; + +layout (binding = BINDING_COLOR_TEXTURE) uniform sampler2D color_texture; + +vec4 cubic(float v) { + vec4 n = vec4(1.0, 2.0, 3.0, 4.0) - v; + vec4 s = n * n * n; + float x = s.x; + float y = s.y - 4.0 * s.x; + float z = s.z - 4.0 * s.y + 6.0 * s.x; + float w = 6.0 - x - y - z; + return vec4(x, y, z, w) * (1.0 / 6.0); +} + +vec4 textureBicubic( sampler2D textureSampler, vec2 texCoords ) { + + vec2 texSize = textureSize(textureSampler, 0); + vec2 invTexSize = 1.0 / texSize; + + texCoords = texCoords * texSize - 0.5; + + vec2 fxy = fract(texCoords); + texCoords -= fxy; + + vec4 xcubic = cubic(fxy.x); + vec4 ycubic = cubic(fxy.y); + + vec4 c = texCoords.xxyy + vec2(-0.5, +1.5).xyxy; + + vec4 s = vec4(xcubic.xz + xcubic.yw, ycubic.xz + ycubic.yw); + vec4 offset = c + vec4(xcubic.yw, ycubic.yw) / s; + + offset *= invTexSize.xxyy; + + vec4 sample0 = texture(textureSampler, offset.xz); + vec4 sample1 = texture(textureSampler, offset.yz); + vec4 sample2 = texture(textureSampler, offset.xw); + vec4 sample3 = texture(textureSampler, offset.yw); + + float sx = s.x / (s.x + s.y); + float sy = s.z / (s.z + s.w); + + return mix(mix(sample3, sample2, sx), mix(sample1, sample0, sx), sy); +} + +void main() { + color = vec4(textureBicubic(color_texture, frag_tex_coord).rgb, 1.0f); +} diff --git a/src/video_core/host_shaders/present_gaussian.frag b/src/video_core/host_shaders/present_gaussian.frag new file mode 100644 index 000000000..66fed3238 --- /dev/null +++ b/src/video_core/host_shaders/present_gaussian.frag @@ -0,0 +1,70 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +// Code adapted from the following sources: +// - https://learnopengl.com/Advanced-Lighting/Bloom +// - https://www.rastergrid.com/blog/2010/09/efficient-gaussian-blur-with-linear-sampling/ + +#version 460 core + +#ifdef VULKAN + +#define BINDING_COLOR_TEXTURE 1 + +#else // ^^^ Vulkan ^^^ // vvv OpenGL vvv + +#define BINDING_COLOR_TEXTURE 0 + +#endif + +layout(location = 0) in vec2 frag_tex_coord; + +layout(location = 0) out vec4 color; + +layout(binding = BINDING_COLOR_TEXTURE) uniform sampler2D color_texture; + +const float offset[3] = float[](0.0, 1.3846153846, 3.2307692308); +const float weight[3] = float[](0.2270270270, 0.3162162162, 0.0702702703); + +vec4 blurVertical(sampler2D textureSampler, vec2 coord, vec2 norm) { + vec4 result = vec4(0.0f); + for (int i = 1; i < 3; i++) { + result += texture(textureSampler, vec2(coord) + (vec2(0.0, offset[i]) * norm)) * weight[i]; + result += texture(textureSampler, vec2(coord) - (vec2(0.0, offset[i]) * norm)) * weight[i]; + } + return result; +} + +vec4 blurHorizontal(sampler2D textureSampler, vec2 coord, vec2 norm) { + vec4 result = vec4(0.0f); + for (int i = 1; i < 3; i++) { + result += texture(textureSampler, vec2(coord) + (vec2(offset[i], 0.0) * norm)) * weight[i]; + result += texture(textureSampler, vec2(coord) - (vec2(offset[i], 0.0) * norm)) * weight[i]; + } + return result; +} + +vec4 blurDiagonal(sampler2D textureSampler, vec2 coord, vec2 norm) { + vec4 result = vec4(0.0f); + for (int i = 1; i < 3; i++) { + result += + texture(textureSampler, vec2(coord) + (vec2(offset[i], offset[i]) * norm)) * weight[i]; + result += + texture(textureSampler, vec2(coord) - (vec2(offset[i], offset[i]) * norm)) * weight[i]; + } + return result; +} + +void main() { + vec3 base = texture(color_texture, vec2(frag_tex_coord)).rgb * weight[0]; + vec2 tex_offset = 1.0f / textureSize(color_texture, 0); + + // TODO(Blinkhawk): This code can be optimized through shader group instructions. + vec3 horizontal = blurHorizontal(color_texture, frag_tex_coord, tex_offset).rgb; + vec3 vertical = blurVertical(color_texture, frag_tex_coord, tex_offset).rgb; + vec3 diagonalA = blurDiagonal(color_texture, frag_tex_coord, tex_offset).rgb; + vec3 diagonalB = blurDiagonal(color_texture, frag_tex_coord, tex_offset * vec2(1.0, -1.0)).rgb; + vec3 combination = mix(mix(horizontal, vertical, 0.5f), mix(diagonalA, diagonalB, 0.5f), 0.5f); + color = vec4(combination + base, 1.0f); +} diff --git a/src/video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp16.comp b/src/video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp16.comp new file mode 100644 index 000000000..1c96a7905 --- /dev/null +++ b/src/video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp16.comp @@ -0,0 +1,11 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#version 460 core +#extension GL_GOOGLE_include_directive : enable + +#define YUZU_USE_FP16 +#define USE_EASU 1 + +#include "fidelityfx_fsr.comp" diff --git a/src/video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp32.comp b/src/video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp32.comp new file mode 100644 index 000000000..f4daff739 --- /dev/null +++ b/src/video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp32.comp @@ -0,0 +1,10 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#version 460 core +#extension GL_GOOGLE_include_directive : enable + +#define USE_EASU 1 + +#include "fidelityfx_fsr.comp" diff --git a/src/video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp16.comp b/src/video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp16.comp new file mode 100644 index 000000000..6b6796dd1 --- /dev/null +++ b/src/video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp16.comp @@ -0,0 +1,11 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#version 460 core +#extension GL_GOOGLE_include_directive : enable + +#define YUZU_USE_FP16 +#define USE_RCAS 1 + +#include "fidelityfx_fsr.comp" diff --git a/src/video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp32.comp b/src/video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp32.comp new file mode 100644 index 000000000..f785eebf3 --- /dev/null +++ b/src/video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp32.comp @@ -0,0 +1,10 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#version 460 core +#extension GL_GOOGLE_include_directive : enable + +#define USE_RCAS 1 + +#include "fidelityfx_fsr.comp" diff --git a/src/video_core/host_shaders/vulkan_present_scaleforce_fp16.frag b/src/video_core/host_shaders/vulkan_present_scaleforce_fp16.frag new file mode 100644 index 000000000..924c03060 --- /dev/null +++ b/src/video_core/host_shaders/vulkan_present_scaleforce_fp16.frag @@ -0,0 +1,7 @@ +#version 460 + +#extension GL_GOOGLE_include_directive : enable + +#define YUZU_USE_FP16 + +#include "opengl_present_scaleforce.frag" diff --git a/src/video_core/host_shaders/vulkan_present_scaleforce_fp32.frag b/src/video_core/host_shaders/vulkan_present_scaleforce_fp32.frag new file mode 100644 index 000000000..a594b83ca --- /dev/null +++ b/src/video_core/host_shaders/vulkan_present_scaleforce_fp32.frag @@ -0,0 +1,5 @@ +#version 460 + +#extension GL_GOOGLE_include_directive : enable + +#include "opengl_present_scaleforce.frag" diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp index 187a28e4d..d4dd10bb6 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp @@ -5,6 +5,7 @@ #include <algorithm> #include <span> +#include "shader_recompiler/backend/glasm/emit_glasm.h" #include "video_core/buffer_cache/buffer_cache.h" #include "video_core/renderer_opengl/gl_buffer_cache.h" #include "video_core/renderer_opengl/gl_device.h" @@ -229,8 +230,10 @@ void BufferCacheRuntime::BindStorageBuffer(size_t stage, u32 binding_index, Buff .padding = 0, }; buffer.MakeResident(is_written ? GL_READ_WRITE : GL_READ_ONLY); - glProgramLocalParametersI4uivNV(PROGRAM_LUT[stage], binding_index, 1, - reinterpret_cast<const GLuint*>(&ssbo)); + glProgramLocalParametersI4uivNV( + PROGRAM_LUT[stage], + Shader::Backend::GLASM::PROGRAM_LOCAL_PARAMETER_STORAGE_BUFFER_BASE + binding_index, 1, + reinterpret_cast<const GLuint*>(&ssbo)); } } @@ -250,8 +253,10 @@ void BufferCacheRuntime::BindComputeStorageBuffer(u32 binding_index, Buffer& buf .padding = 0, }; buffer.MakeResident(is_written ? GL_READ_WRITE : GL_READ_ONLY); - glProgramLocalParametersI4uivNV(GL_COMPUTE_PROGRAM_NV, binding_index, 1, - reinterpret_cast<const GLuint*>(&ssbo)); + glProgramLocalParametersI4uivNV( + GL_COMPUTE_PROGRAM_NV, + Shader::Backend::GLASM::PROGRAM_LOCAL_PARAMETER_STORAGE_BUFFER_BASE + binding_index, 1, + reinterpret_cast<const GLuint*>(&ssbo)); } } diff --git a/src/video_core/renderer_opengl/gl_compute_pipeline.cpp b/src/video_core/renderer_opengl/gl_compute_pipeline.cpp index aa1cc592f..5c1f21c65 100644 --- a/src/video_core/renderer_opengl/gl_compute_pipeline.cpp +++ b/src/video_core/renderer_opengl/gl_compute_pipeline.cpp @@ -19,15 +19,6 @@ using VideoCommon::ImageId; constexpr u32 MAX_TEXTURES = 64; constexpr u32 MAX_IMAGES = 16; -template <typename Range> -u32 AccumulateCount(const Range& range) { - u32 num{}; - for (const auto& desc : range) { - num += desc.count; - } - return num; -} - size_t ComputePipelineKey::Hash() const noexcept { return static_cast<size_t>( Common::CityHash64(reinterpret_cast<const char*>(this), sizeof *this)); @@ -58,17 +49,17 @@ ComputePipeline::ComputePipeline(const Device& device, TextureCache& texture_cac std::copy_n(info.constant_buffer_used_sizes.begin(), uniform_buffer_sizes.size(), uniform_buffer_sizes.begin()); - num_texture_buffers = AccumulateCount(info.texture_buffer_descriptors); - num_image_buffers = AccumulateCount(info.image_buffer_descriptors); + num_texture_buffers = Shader::NumDescriptors(info.texture_buffer_descriptors); + num_image_buffers = Shader::NumDescriptors(info.image_buffer_descriptors); - const u32 num_textures{num_texture_buffers + AccumulateCount(info.texture_descriptors)}; + const u32 num_textures{num_texture_buffers + Shader::NumDescriptors(info.texture_descriptors)}; ASSERT(num_textures <= MAX_TEXTURES); - const u32 num_images{num_image_buffers + AccumulateCount(info.image_descriptors)}; + const u32 num_images{num_image_buffers + Shader::NumDescriptors(info.image_descriptors)}; ASSERT(num_images <= MAX_IMAGES); const bool is_glasm{assembly_program.handle != 0}; - const u32 num_storage_buffers{AccumulateCount(info.storage_buffers_descriptors)}; + const u32 num_storage_buffers{Shader::NumDescriptors(info.storage_buffers_descriptors)}; use_storage_buffers = !is_glasm || num_storage_buffers < device.GetMaxGLASMStorageBufferBlocks(); writes_global_memory = !use_storage_buffers && @@ -88,8 +79,7 @@ void ComputePipeline::Configure() { } texture_cache.SynchronizeComputeDescriptors(); - std::array<ImageViewId, MAX_TEXTURES + MAX_IMAGES> image_view_ids; - boost::container::static_vector<u32, MAX_TEXTURES + MAX_IMAGES> image_view_indices; + boost::container::static_vector<VideoCommon::ImageViewInOut, MAX_TEXTURES + MAX_IMAGES> views; std::array<GLuint, MAX_TEXTURES> samplers; std::array<GLuint, MAX_TEXTURES> textures; std::array<GLuint, MAX_IMAGES> images; @@ -119,33 +109,39 @@ void ComputePipeline::Configure() { } return TexturePair(gpu_memory.Read<u32>(addr), via_header_index); }}; - const auto add_image{[&](const auto& desc) { + const auto add_image{[&](const auto& desc, bool blacklist) { for (u32 index = 0; index < desc.count; ++index) { const auto handle{read_handle(desc, index)}; - image_view_indices.push_back(handle.first); + views.push_back({ + .index = handle.first, + .blacklist = blacklist, + .id = {}, + }); } }}; for (const auto& desc : info.texture_buffer_descriptors) { for (u32 index = 0; index < desc.count; ++index) { const auto handle{read_handle(desc, index)}; - image_view_indices.push_back(handle.first); + views.push_back({handle.first}); samplers[sampler_binding++] = 0; } } - std::ranges::for_each(info.image_buffer_descriptors, add_image); + for (const auto& desc : info.image_buffer_descriptors) { + add_image(desc, false); + } for (const auto& desc : info.texture_descriptors) { for (u32 index = 0; index < desc.count; ++index) { const auto handle{read_handle(desc, index)}; - image_view_indices.push_back(handle.first); + views.push_back({handle.first}); Sampler* const sampler = texture_cache.GetComputeSampler(handle.second); samplers[sampler_binding++] = sampler->Handle(); } } - std::ranges::for_each(info.image_descriptors, add_image); - - const std::span indices_span(image_view_indices.data(), image_view_indices.size()); - texture_cache.FillComputeImageViews(indices_span, image_view_ids); + for (const auto& desc : info.image_descriptors) { + add_image(desc, desc.is_written); + } + texture_cache.FillComputeImageViews(std::span(views.data(), views.size())); if (assembly_program.handle != 0) { program_manager.BindComputeAssemblyProgram(assembly_program.handle); @@ -161,7 +157,7 @@ void ComputePipeline::Configure() { if constexpr (is_image) { is_written = desc.is_written; } - ImageView& image_view{texture_cache.GetImageView(image_view_ids[texbuf_index])}; + ImageView& image_view{texture_cache.GetImageView(views[texbuf_index].id)}; buffer_cache.BindComputeTextureBuffer(texbuf_index, image_view.GpuAddr(), image_view.BufferSize(), image_view.format, is_written, is_image); @@ -177,23 +173,45 @@ void ComputePipeline::Configure() { buffer_cache.runtime.SetImagePointers(textures.data(), images.data()); buffer_cache.BindHostComputeBuffers(); - const ImageId* views_it{image_view_ids.data() + num_texture_buffers + num_image_buffers}; + const VideoCommon::ImageViewInOut* views_it{views.data() + num_texture_buffers + + num_image_buffers}; texture_binding += num_texture_buffers; image_binding += num_image_buffers; + u32 texture_scaling_mask{}; for (const auto& desc : info.texture_descriptors) { for (u32 index = 0; index < desc.count; ++index) { - ImageView& image_view{texture_cache.GetImageView(*(views_it++))}; - textures[texture_binding++] = image_view.Handle(desc.type); + ImageView& image_view{texture_cache.GetImageView((views_it++)->id)}; + textures[texture_binding] = image_view.Handle(desc.type); + if (texture_cache.IsRescaling(image_view)) { + texture_scaling_mask |= 1u << texture_binding; + } + ++texture_binding; } } + u32 image_scaling_mask{}; for (const auto& desc : info.image_descriptors) { for (u32 index = 0; index < desc.count; ++index) { - ImageView& image_view{texture_cache.GetImageView(*(views_it++))}; + ImageView& image_view{texture_cache.GetImageView((views_it++)->id)}; if (desc.is_written) { texture_cache.MarkModification(image_view.image_id); } - images[image_binding++] = image_view.StorageView(desc.type, desc.format); + images[image_binding] = image_view.StorageView(desc.type, desc.format); + if (texture_cache.IsRescaling(image_view)) { + image_scaling_mask |= 1u << image_binding; + } + ++image_binding; + } + } + if (info.uses_rescaling_uniform) { + const f32 float_texture_scaling_mask{Common::BitCast<f32>(texture_scaling_mask)}; + const f32 float_image_scaling_mask{Common::BitCast<f32>(image_scaling_mask)}; + if (assembly_program.handle != 0) { + glProgramLocalParameter4fARB(GL_COMPUTE_PROGRAM_NV, 0, float_texture_scaling_mask, + float_image_scaling_mask, 0.0f, 0.0f); + } else { + glProgramUniform4f(source_program.handle, 0, float_texture_scaling_mask, + float_image_scaling_mask, 0.0f, 0.0f); } } if (texture_binding != 0) { diff --git a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp index bccb37a58..f8495896c 100644 --- a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp +++ b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp @@ -15,7 +15,7 @@ #include "video_core/renderer_opengl/gl_shader_util.h" #include "video_core/renderer_opengl/gl_state_tracker.h" #include "video_core/shader_notify.h" -#include "video_core/texture_cache/texture_cache_base.h" +#include "video_core/texture_cache/texture_cache.h" #if defined(_MSC_VER) && defined(NDEBUG) #define LAMBDA_FORCEINLINE [[msvc::forceinline]] @@ -27,6 +27,7 @@ namespace OpenGL { namespace { using Shader::ImageBufferDescriptor; using Shader::ImageDescriptor; +using Shader::NumDescriptors; using Shader::TextureBufferDescriptor; using Shader::TextureDescriptor; using Tegra::Texture::TexturePair; @@ -35,15 +36,6 @@ using VideoCommon::ImageId; constexpr u32 MAX_TEXTURES = 64; constexpr u32 MAX_IMAGES = 8; -template <typename Range> -u32 AccumulateCount(const Range& range) { - u32 num{}; - for (const auto& desc : range) { - num += desc.count; - } - return num; -} - GLenum Stage(size_t stage_index) { switch (stage_index) { case 0: @@ -204,23 +196,23 @@ GraphicsPipeline::GraphicsPipeline( base_uniform_bindings[stage + 1] = base_uniform_bindings[stage]; base_storage_bindings[stage + 1] = base_storage_bindings[stage]; - base_uniform_bindings[stage + 1] += AccumulateCount(info.constant_buffer_descriptors); - base_storage_bindings[stage + 1] += AccumulateCount(info.storage_buffers_descriptors); + base_uniform_bindings[stage + 1] += NumDescriptors(info.constant_buffer_descriptors); + base_storage_bindings[stage + 1] += NumDescriptors(info.storage_buffers_descriptors); } enabled_uniform_buffer_masks[stage] = info.constant_buffer_mask; std::ranges::copy(info.constant_buffer_used_sizes, uniform_buffer_sizes[stage].begin()); - const u32 num_tex_buffer_bindings{AccumulateCount(info.texture_buffer_descriptors)}; + const u32 num_tex_buffer_bindings{NumDescriptors(info.texture_buffer_descriptors)}; num_texture_buffers[stage] += num_tex_buffer_bindings; num_textures += num_tex_buffer_bindings; - const u32 num_img_buffers_bindings{AccumulateCount(info.image_buffer_descriptors)}; + const u32 num_img_buffers_bindings{NumDescriptors(info.image_buffer_descriptors)}; num_image_buffers[stage] += num_img_buffers_bindings; num_images += num_img_buffers_bindings; - num_textures += AccumulateCount(info.texture_descriptors); - num_images += AccumulateCount(info.image_descriptors); - num_storage_buffers += AccumulateCount(info.storage_buffers_descriptors); + num_textures += NumDescriptors(info.texture_descriptors); + num_images += NumDescriptors(info.image_descriptors); + num_storage_buffers += NumDescriptors(info.storage_buffers_descriptors); writes_global_memory |= std::ranges::any_of( info.storage_buffers_descriptors, [](const auto& desc) { return desc.is_written; }); @@ -288,10 +280,9 @@ GraphicsPipeline::GraphicsPipeline( template <typename Spec> void GraphicsPipeline::ConfigureImpl(bool is_indexed) { - std::array<ImageId, MAX_TEXTURES + MAX_IMAGES> image_view_ids; - std::array<u32, MAX_TEXTURES + MAX_IMAGES> image_view_indices; + std::array<VideoCommon::ImageViewInOut, MAX_TEXTURES + MAX_IMAGES> views; std::array<GLuint, MAX_TEXTURES> samplers; - size_t image_view_index{}; + size_t views_index{}; GLsizei sampler_binding{}; texture_cache.SynchronizeGraphicsDescriptors(); @@ -336,30 +327,34 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { } return TexturePair(gpu_memory.Read<u32>(addr), via_header_index); }}; - const auto add_image{[&](const auto& desc) { + const auto add_image{[&](const auto& desc, bool blacklist) LAMBDA_FORCEINLINE { for (u32 index = 0; index < desc.count; ++index) { const auto handle{read_handle(desc, index)}; - image_view_indices[image_view_index++] = handle.first; + views[views_index++] = { + .index = handle.first, + .blacklist = blacklist, + .id = {}, + }; } }}; if constexpr (Spec::has_texture_buffers) { for (const auto& desc : info.texture_buffer_descriptors) { for (u32 index = 0; index < desc.count; ++index) { const auto handle{read_handle(desc, index)}; - image_view_indices[image_view_index++] = handle.first; + views[views_index++] = {handle.first}; samplers[sampler_binding++] = 0; } } } if constexpr (Spec::has_image_buffers) { for (const auto& desc : info.image_buffer_descriptors) { - add_image(desc); + add_image(desc, false); } } for (const auto& desc : info.texture_descriptors) { for (u32 index = 0; index < desc.count; ++index) { const auto handle{read_handle(desc, index)}; - image_view_indices[image_view_index++] = handle.first; + views[views_index++] = {handle.first}; Sampler* const sampler{texture_cache.GetGraphicsSampler(handle.second)}; samplers[sampler_binding++] = sampler->Handle(); @@ -367,7 +362,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { } if constexpr (Spec::has_images) { for (const auto& desc : info.image_descriptors) { - add_image(desc); + add_image(desc, desc.is_written); } } }}; @@ -386,13 +381,12 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { if constexpr (Spec::enabled_stages[4]) { config_stage(4); } - const std::span indices_span(image_view_indices.data(), image_view_index); - texture_cache.FillGraphicsImageViews(indices_span, image_view_ids); + texture_cache.FillGraphicsImageViews<Spec::has_images>(std::span(views.data(), views_index)); texture_cache.UpdateRenderTargets(false); state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle()); - ImageId* texture_buffer_index{image_view_ids.data()}; + VideoCommon::ImageViewInOut* texture_buffer_it{views.data()}; const auto bind_stage_info{[&](size_t stage) LAMBDA_FORCEINLINE { size_t index{}; const auto add_buffer{[&](const auto& desc) { @@ -402,12 +396,12 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { if constexpr (is_image) { is_written = desc.is_written; } - ImageView& image_view{texture_cache.GetImageView(*texture_buffer_index)}; + ImageView& image_view{texture_cache.GetImageView(texture_buffer_it->id)}; buffer_cache.BindGraphicsTextureBuffer(stage, index, image_view.GpuAddr(), image_view.BufferSize(), image_view.format, is_written, is_image); ++index; - ++texture_buffer_index; + ++texture_buffer_it; } }}; const Shader::Info& info{stage_infos[stage]}; @@ -423,13 +417,9 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { add_buffer(desc); } } - for (const auto& desc : info.texture_descriptors) { - texture_buffer_index += desc.count; - } + texture_buffer_it += Shader::NumDescriptors(info.texture_descriptors); if constexpr (Spec::has_images) { - for (const auto& desc : info.image_descriptors) { - texture_buffer_index += desc.count; - } + texture_buffer_it += Shader::NumDescriptors(info.image_descriptors); } }}; if constexpr (Spec::enabled_stages[0]) { @@ -453,12 +443,13 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { if (!is_built.load(std::memory_order::relaxed)) { WaitForBuild(); } - if (assembly_programs[0].handle != 0) { + const bool use_assembly{assembly_programs[0].handle != 0}; + if (use_assembly) { program_manager.BindAssemblyPrograms(assembly_programs, enabled_stages_mask); } else { program_manager.BindSourcePrograms(source_programs); } - const ImageId* views_it{image_view_ids.data()}; + const VideoCommon::ImageViewInOut* views_it{views.data()}; GLsizei texture_binding = 0; GLsizei image_binding = 0; std::array<GLuint, MAX_TEXTURES> textures; @@ -473,20 +464,49 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { views_it += num_texture_buffers[stage]; views_it += num_image_buffers[stage]; + u32 texture_scaling_mask{}; + u32 image_scaling_mask{}; + u32 stage_texture_binding{}; + u32 stage_image_binding{}; + const auto& info{stage_infos[stage]}; for (const auto& desc : info.texture_descriptors) { for (u32 index = 0; index < desc.count; ++index) { - ImageView& image_view{texture_cache.GetImageView(*(views_it++))}; - textures[texture_binding++] = image_view.Handle(desc.type); + ImageView& image_view{texture_cache.GetImageView((views_it++)->id)}; + textures[texture_binding] = image_view.Handle(desc.type); + if (texture_cache.IsRescaling(image_view)) { + texture_scaling_mask |= 1u << stage_texture_binding; + } + ++texture_binding; + ++stage_texture_binding; } } for (const auto& desc : info.image_descriptors) { for (u32 index = 0; index < desc.count; ++index) { - ImageView& image_view{texture_cache.GetImageView(*(views_it++))}; + ImageView& image_view{texture_cache.GetImageView((views_it++)->id)}; if (desc.is_written) { texture_cache.MarkModification(image_view.image_id); } - images[image_binding++] = image_view.StorageView(desc.type, desc.format); + images[image_binding] = image_view.StorageView(desc.type, desc.format); + if (texture_cache.IsRescaling(image_view)) { + image_scaling_mask |= 1u << stage_image_binding; + } + ++image_binding; + ++stage_image_binding; + } + } + if (info.uses_rescaling_uniform) { + const f32 float_texture_scaling_mask{Common::BitCast<f32>(texture_scaling_mask)}; + const f32 float_image_scaling_mask{Common::BitCast<f32>(image_scaling_mask)}; + const bool is_rescaling{texture_cache.IsRescaling()}; + const f32 config_down_factor{Settings::values.resolution_info.down_factor}; + const f32 down_factor{is_rescaling ? config_down_factor : 1.0f}; + if (use_assembly) { + glProgramLocalParameter4fARB(AssemblyStage(stage), 0, float_texture_scaling_mask, + float_image_scaling_mask, down_factor, 0.0f); + } else { + glProgramUniform4f(source_programs[stage].handle, 0, float_texture_scaling_mask, + float_image_scaling_mask, down_factor, 0.0f); } } }}; diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index a6d9f7c43..9b516c64f 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -184,6 +184,10 @@ void RasterizerOpenGL::Clear() { SyncRasterizeEnable(); SyncStencilTestState(); + std::scoped_lock lock{texture_cache.mutex}; + texture_cache.UpdateRenderTargets(true); + state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle()); + SyncViewport(); if (regs.clear_flags.scissor) { SyncScissorTest(); } else { @@ -192,10 +196,6 @@ void RasterizerOpenGL::Clear() { } UNIMPLEMENTED_IF(regs.clear_flags.viewport); - std::scoped_lock lock{texture_cache.mutex}; - texture_cache.UpdateRenderTargets(true); - state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle()); - if (use_color) { glClearBufferfv(GL_COLOR, regs.clear_buffers.RT, regs.clear_color); } @@ -214,8 +214,6 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { query_cache.UpdateCounters(); - SyncState(); - GraphicsPipeline* const pipeline{shader_cache.CurrentGraphicsPipeline()}; if (!pipeline) { return; @@ -223,6 +221,8 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; pipeline->Configure(is_indexed); + SyncState(); + const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(maxwell3d.regs.draw.topology); BeginTransformFeedback(pipeline, primitive_mode); @@ -533,7 +533,8 @@ void RasterizerOpenGL::SyncViewport() { auto& flags = maxwell3d.dirty.flags; const auto& regs = maxwell3d.regs; - const bool dirty_viewport = flags[Dirty::Viewports]; + const bool rescale_viewports = flags[VideoCommon::Dirty::RescaleViewports]; + const bool dirty_viewport = flags[Dirty::Viewports] || rescale_viewports; const bool dirty_clip_control = flags[Dirty::ClipControl]; if (dirty_clip_control || flags[Dirty::FrontFace]) { @@ -553,8 +554,7 @@ void RasterizerOpenGL::SyncViewport() { } glFrontFace(mode); } - - if (dirty_viewport || flags[Dirty::ClipControl]) { + if (dirty_viewport || dirty_clip_control) { flags[Dirty::ClipControl] = false; bool flip_y = false; @@ -570,37 +570,58 @@ void RasterizerOpenGL::SyncViewport() { state_tracker.ClipControl(origin, depth); state_tracker.SetYNegate(regs.screen_y_control.y_negate != 0); } + const bool is_rescaling{texture_cache.IsRescaling()}; + const float scale = is_rescaling ? Settings::values.resolution_info.up_factor : 1.0f; + const auto conv = [scale](float value) -> GLfloat { + float new_value = value * scale; + if (scale < 1.0f) { + const bool sign = std::signbit(value); + new_value = std::round(std::abs(new_value)); + new_value = sign ? -new_value : new_value; + } + return static_cast<GLfloat>(new_value); + }; if (dirty_viewport) { flags[Dirty::Viewports] = false; - const bool force = flags[Dirty::ViewportTransform]; + const bool force = flags[Dirty::ViewportTransform] || rescale_viewports; flags[Dirty::ViewportTransform] = false; + flags[VideoCommon::Dirty::RescaleViewports] = false; - for (std::size_t i = 0; i < Maxwell::NumViewports; ++i) { - if (!force && !flags[Dirty::Viewport0 + i]) { + for (size_t index = 0; index < Maxwell::NumViewports; ++index) { + if (!force && !flags[Dirty::Viewport0 + index]) { continue; } - flags[Dirty::Viewport0 + i] = false; + flags[Dirty::Viewport0 + index] = false; - const auto& src = regs.viewport_transform[i]; - const Common::Rectangle<f32> rect{src.GetRect()}; - glViewportIndexedf(static_cast<GLuint>(i), rect.left, rect.bottom, rect.GetWidth(), - rect.GetHeight()); + const auto& src = regs.viewport_transform[index]; + GLfloat x = conv(src.translate_x - src.scale_x); + GLfloat y = conv(src.translate_y - src.scale_y); + GLfloat width = conv(src.scale_x * 2.0f); + GLfloat height = conv(src.scale_y * 2.0f); + + if (height < 0) { + y += height; + height = -height; + } + glViewportIndexedf(static_cast<GLuint>(index), x, y, width != 0.0f ? width : 1.0f, + height != 0.0f ? height : 1.0f); const GLdouble reduce_z = regs.depth_mode == Maxwell::DepthMode::MinusOneToOne; const GLdouble near_depth = src.translate_z - src.scale_z * reduce_z; const GLdouble far_depth = src.translate_z + src.scale_z; if (device.HasDepthBufferFloat()) { - glDepthRangeIndexeddNV(static_cast<GLuint>(i), near_depth, far_depth); + glDepthRangeIndexeddNV(static_cast<GLuint>(index), near_depth, far_depth); } else { - glDepthRangeIndexed(static_cast<GLuint>(i), near_depth, far_depth); + glDepthRangeIndexed(static_cast<GLuint>(index), near_depth, far_depth); } if (!GLAD_GL_NV_viewport_swizzle) { continue; } - glViewportSwizzleNV(static_cast<GLuint>(i), MaxwellToGL::ViewportSwizzle(src.swizzle.x), + glViewportSwizzleNV(static_cast<GLuint>(index), + MaxwellToGL::ViewportSwizzle(src.swizzle.x), MaxwellToGL::ViewportSwizzle(src.swizzle.y), MaxwellToGL::ViewportSwizzle(src.swizzle.z), MaxwellToGL::ViewportSwizzle(src.swizzle.w)); @@ -903,14 +924,34 @@ void RasterizerOpenGL::SyncLogicOpState() { void RasterizerOpenGL::SyncScissorTest() { auto& flags = maxwell3d.dirty.flags; - if (!flags[Dirty::Scissors]) { + if (!flags[Dirty::Scissors] && !flags[VideoCommon::Dirty::RescaleScissors]) { return; } flags[Dirty::Scissors] = false; + const bool force = flags[VideoCommon::Dirty::RescaleScissors]; + flags[VideoCommon::Dirty::RescaleScissors] = false; + const auto& regs = maxwell3d.regs; + + const auto& resolution = Settings::values.resolution_info; + const bool is_rescaling{texture_cache.IsRescaling()}; + const u32 up_scale = is_rescaling ? resolution.up_scale : 1U; + const u32 down_shift = is_rescaling ? resolution.down_shift : 0U; + const auto scale_up = [up_scale, down_shift](u32 value) -> u32 { + if (value == 0) { + return 0U; + } + const u32 upset = value * up_scale; + u32 acumm{}; + if ((up_scale >> down_shift) == 0) { + acumm = upset % 2; + } + const u32 converted_value = upset >> down_shift; + return std::max<u32>(converted_value + acumm, 1U); + }; for (std::size_t index = 0; index < Maxwell::NumViewports; ++index) { - if (!flags[Dirty::Scissor0 + index]) { + if (!force && !flags[Dirty::Scissor0 + index]) { continue; } flags[Dirty::Scissor0 + index] = false; @@ -918,8 +959,8 @@ void RasterizerOpenGL::SyncScissorTest() { const auto& src = regs.scissor_test[index]; if (src.enable) { glEnablei(GL_SCISSOR_TEST, static_cast<GLuint>(index)); - glScissorIndexed(static_cast<GLuint>(index), src.min_x, src.min_y, - src.max_x - src.min_x, src.max_y - src.min_y); + glScissorIndexed(static_cast<GLuint>(index), scale_up(src.min_x), scale_up(src.min_y), + scale_up(src.max_x - src.min_x), scale_up(src.max_y - src.min_y)); } else { glDisablei(GL_SCISSOR_TEST, static_cast<GLuint>(index)); } @@ -935,8 +976,9 @@ void RasterizerOpenGL::SyncPointState() { oglEnable(GL_POINT_SPRITE, maxwell3d.regs.point_sprite_enable); oglEnable(GL_PROGRAM_POINT_SIZE, maxwell3d.regs.vp_point_size.enable); - - glPointSize(std::max(1.0f, maxwell3d.regs.point_size)); + const bool is_rescaling{texture_cache.IsRescaling()}; + const float scale = is_rescaling ? Settings::values.resolution_info.up_factor : 1.0f; + glPointSize(std::max(1.0f, maxwell3d.regs.point_size * scale)); } void RasterizerOpenGL::SyncLineState() { diff --git a/src/video_core/renderer_opengl/gl_resource_manager.cpp b/src/video_core/renderer_opengl/gl_resource_manager.cpp index 8695c29e3..5e7101d28 100644 --- a/src/video_core/renderer_opengl/gl_resource_manager.cpp +++ b/src/video_core/renderer_opengl/gl_resource_manager.cpp @@ -166,7 +166,12 @@ void OGLFramebuffer::Create() { return; MICROPROFILE_SCOPE(OpenGL_ResourceCreation); + // Bind to READ_FRAMEBUFFER to stop Nvidia's driver from creating an EXT_framebuffer instead of + // a core framebuffer. EXT framebuffer attachments have to match in size and can be shared + // across contexts. yuzu doesn't share framebuffers across contexts and we need attachments with + // mismatching size, this is why core framebuffers are preferred. glGenFramebuffers(1, &handle); + glBindFramebuffer(GL_READ_FRAMEBUFFER, handle); } void OGLFramebuffer::Release() { diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index 02682bd76..42ef67628 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -426,16 +426,14 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline( // Normal path programs[index] = TranslateProgram(pools.inst, pools.block, env, cfg, host_info); - for (const auto& desc : programs[index].info.storage_buffers_descriptors) { - total_storage_buffers += desc.count; - } + total_storage_buffers += + Shader::NumDescriptors(programs[index].info.storage_buffers_descriptors); } else { // VertexB path when VertexA is present. auto& program_va{programs[0]}; auto program_vb{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)}; - for (const auto& desc : program_vb.info.storage_buffers_descriptors) { - total_storage_buffers += desc.count; - } + total_storage_buffers += + Shader::NumDescriptors(program_vb.info.storage_buffers_descriptors); programs[index] = MergeDualVertexPrograms(program_va, program_vb, env); } } @@ -510,10 +508,7 @@ std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline( Shader::Maxwell::Flow::CFG cfg{env, pools.flow_block, env.StartAddress()}; auto program{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)}; - u32 num_storage_buffers{}; - for (const auto& desc : program.info.storage_buffers_descriptors) { - num_storage_buffers += desc.count; - } + const u32 num_storage_buffers{Shader::NumDescriptors(program.info.storage_buffers_descriptors)}; Shader::RuntimeInfo info; info.glasm_use_storage_buffers = num_storage_buffers <= device.GetMaxGLASMStorageBufferBlocks(); diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp index 8c3ca3d82..2f7d98d8b 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp @@ -9,8 +9,8 @@ #include <glad/glad.h> +#include "common/literals.h" #include "common/settings.h" - #include "video_core/renderer_opengl/gl_device.h" #include "video_core/renderer_opengl/gl_shader_manager.h" #include "video_core/renderer_opengl/gl_state_tracker.h" @@ -42,6 +42,7 @@ using VideoCore::Surface::IsPixelFormatSRGB; using VideoCore::Surface::MaxPixelFormat; using VideoCore::Surface::PixelFormat; using VideoCore::Surface::SurfaceType; +using namespace Common::Literals; struct CopyOrigin { GLint level; @@ -316,6 +317,52 @@ void AttachTexture(GLuint fbo, GLenum attachment, const ImageView* image_view) { } } +OGLTexture MakeImage(const VideoCommon::ImageInfo& info, GLenum gl_internal_format) { + const GLenum target = ImageTarget(info); + const GLsizei width = info.size.width; + const GLsizei height = info.size.height; + const GLsizei depth = info.size.depth; + const int max_host_mip_levels = std::bit_width(info.size.width); + const GLsizei num_levels = std::min(info.resources.levels, max_host_mip_levels); + const GLsizei num_layers = info.resources.layers; + const GLsizei num_samples = info.num_samples; + + GLuint handle = 0; + OGLTexture texture; + if (target != GL_TEXTURE_BUFFER) { + texture.Create(target); + handle = texture.handle; + } + switch (target) { + case GL_TEXTURE_1D_ARRAY: + glTextureStorage2D(handle, num_levels, gl_internal_format, width, num_layers); + break; + case GL_TEXTURE_2D_ARRAY: + glTextureStorage3D(handle, num_levels, gl_internal_format, width, height, num_layers); + break; + case GL_TEXTURE_2D_MULTISAMPLE_ARRAY: { + // TODO: Where should 'fixedsamplelocations' come from? + const auto [samples_x, samples_y] = SamplesLog2(info.num_samples); + glTextureStorage3DMultisample(handle, num_samples, gl_internal_format, width >> samples_x, + height >> samples_y, num_layers, GL_FALSE); + break; + } + case GL_TEXTURE_RECTANGLE: + glTextureStorage2D(handle, num_levels, gl_internal_format, width, height); + break; + case GL_TEXTURE_3D: + glTextureStorage3D(handle, num_levels, gl_internal_format, width, height, depth); + break; + case GL_TEXTURE_BUFFER: + UNREACHABLE(); + break; + default: + UNREACHABLE_MSG("Invalid target=0x{:x}", target); + break; + } + return texture; +} + [[nodiscard]] bool IsPixelFormatBGR(PixelFormat format) { switch (format) { case PixelFormat::B5G6R5_UNORM: @@ -359,7 +406,8 @@ ImageBufferMap::~ImageBufferMap() { TextureCacheRuntime::TextureCacheRuntime(const Device& device_, ProgramManager& program_manager, StateTracker& state_tracker_) - : device{device_}, state_tracker{state_tracker_}, util_shaders(program_manager) { + : device{device_}, state_tracker{state_tracker_}, + util_shaders(program_manager), resolution{Settings::values.resolution_info} { static constexpr std::array TARGETS{GL_TEXTURE_1D_ARRAY, GL_TEXTURE_2D_ARRAY, GL_TEXTURE_3D}; for (size_t i = 0; i < TARGETS.size(); ++i) { const GLenum target = TARGETS[i]; @@ -426,6 +474,13 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, ProgramManager& set_view(Shader::TextureType::ColorArray1D, null_image_1d_array.handle); set_view(Shader::TextureType::ColorArray2D, null_image_view_2d_array.handle); set_view(Shader::TextureType::ColorArrayCube, null_image_cube_array.handle); + + if (resolution.active) { + for (size_t i = 0; i < rescale_draw_fbos.size(); ++i) { + rescale_draw_fbos[i].Create(); + rescale_read_fbos[i].Create(); + } + } } TextureCacheRuntime::~TextureCacheRuntime() = default; @@ -442,6 +497,15 @@ ImageBufferMap TextureCacheRuntime::DownloadStagingBuffer(size_t size) { return download_buffers.RequestMap(size, false); } +u64 TextureCacheRuntime::GetDeviceLocalMemory() const { + if (GLAD_GL_NVX_gpu_memory_info) { + GLint cur_avail_mem_kb = 0; + glGetIntegerv(GL_GPU_MEMORY_INFO_CURRENT_AVAILABLE_VIDMEM_NVX, &cur_avail_mem_kb); + return static_cast<u64>(cur_avail_mem_kb) * 1_KiB; + } + return 2_GiB; // Return minimum requirements +} + void TextureCacheRuntime::CopyImage(Image& dst_image, Image& src_image, std::span<const ImageCopy> copies) { const GLuint dst_name = dst_image.Handle(); @@ -605,13 +669,13 @@ std::optional<size_t> TextureCacheRuntime::StagingBuffers::FindBuffer(size_t req return found; } -Image::Image(TextureCacheRuntime& runtime, const VideoCommon::ImageInfo& info_, GPUVAddr gpu_addr_, +Image::Image(TextureCacheRuntime& runtime_, const VideoCommon::ImageInfo& info_, GPUVAddr gpu_addr_, VAddr cpu_addr_) - : VideoCommon::ImageBase(info_, gpu_addr_, cpu_addr_) { - if (CanBeAccelerated(runtime, info)) { + : VideoCommon::ImageBase(info_, gpu_addr_, cpu_addr_), runtime{&runtime_} { + if (CanBeAccelerated(*runtime, info)) { flags |= ImageFlagBits::AcceleratedUpload; } - if (IsConverted(runtime.device, info.format, info.type)) { + if (IsConverted(runtime->device, info.format, info.type)) { flags |= ImageFlagBits::Converted; gl_internal_format = IsPixelFormatSRGB(info.format) ? GL_SRGB8_ALPHA8 : GL_RGBA8; gl_format = GL_RGBA; @@ -622,58 +686,25 @@ Image::Image(TextureCacheRuntime& runtime, const VideoCommon::ImageInfo& info_, gl_format = tuple.format; gl_type = tuple.type; } - const GLenum target = ImageTarget(info); - const GLsizei width = info.size.width; - const GLsizei height = info.size.height; - const GLsizei depth = info.size.depth; - const int max_host_mip_levels = std::bit_width(info.size.width); - const GLsizei num_levels = std::min(info.resources.levels, max_host_mip_levels); - const GLsizei num_layers = info.resources.layers; - const GLsizei num_samples = info.num_samples; - - GLuint handle = 0; - if (target != GL_TEXTURE_BUFFER) { - texture.Create(target); - handle = texture.handle; - } - switch (target) { - case GL_TEXTURE_1D_ARRAY: - glTextureStorage2D(handle, num_levels, gl_internal_format, width, num_layers); - break; - case GL_TEXTURE_2D_ARRAY: - glTextureStorage3D(handle, num_levels, gl_internal_format, width, height, num_layers); - break; - case GL_TEXTURE_2D_MULTISAMPLE_ARRAY: { - // TODO: Where should 'fixedsamplelocations' come from? - const auto [samples_x, samples_y] = SamplesLog2(info.num_samples); - glTextureStorage3DMultisample(handle, num_samples, gl_internal_format, width >> samples_x, - height >> samples_y, num_layers, GL_FALSE); - break; - } - case GL_TEXTURE_RECTANGLE: - glTextureStorage2D(handle, num_levels, gl_internal_format, width, height); - break; - case GL_TEXTURE_3D: - glTextureStorage3D(handle, num_levels, gl_internal_format, width, height, depth); - break; - case GL_TEXTURE_BUFFER: - UNREACHABLE(); - break; - default: - UNREACHABLE_MSG("Invalid target=0x{:x}", target); - break; - } - if (runtime.device.HasDebuggingToolAttached()) { + texture = MakeImage(info, gl_internal_format); + current_texture = texture.handle; + if (runtime->device.HasDebuggingToolAttached()) { const std::string name = VideoCommon::Name(*this); - glObjectLabel(target == GL_TEXTURE_BUFFER ? GL_BUFFER : GL_TEXTURE, handle, - static_cast<GLsizei>(name.size()), name.data()); + glObjectLabel(ImageTarget(info) == GL_TEXTURE_BUFFER ? GL_BUFFER : GL_TEXTURE, + texture.handle, static_cast<GLsizei>(name.size()), name.data()); } } +Image::Image(const VideoCommon::NullImageParams& params) : VideoCommon::ImageBase{params} {} + Image::~Image() = default; void Image::UploadMemory(const ImageBufferMap& map, std::span<const VideoCommon::BufferImageCopy> copies) { + const bool is_rescaled = True(flags & ImageFlagBits::Rescaled); + if (is_rescaled) { + ScaleDown(true); + } glBindBuffer(GL_PIXEL_UNPACK_BUFFER, map.buffer); glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, map.offset, unswizzled_size_bytes); @@ -693,12 +724,18 @@ void Image::UploadMemory(const ImageBufferMap& map, } CopyBufferToImage(copy, map.offset); } + if (is_rescaled) { + ScaleUp(); + } } void Image::DownloadMemory(ImageBufferMap& map, std::span<const VideoCommon::BufferImageCopy> copies) { + const bool is_rescaled = True(flags & ImageFlagBits::Rescaled); + if (is_rescaled) { + ScaleDown(); + } glMemoryBarrier(GL_PIXEL_BUFFER_BARRIER_BIT); // TODO: Move this to its own API - glBindBuffer(GL_PIXEL_PACK_BUFFER, map.buffer); glPixelStorei(GL_PACK_ALIGNMENT, 1); @@ -716,6 +753,9 @@ void Image::DownloadMemory(ImageBufferMap& map, } CopyImageToBuffer(copy, map.offset); } + if (is_rescaled) { + ScaleUp(true); + } } GLuint Image::StorageHandle() noexcept { @@ -741,11 +781,11 @@ GLuint Image::StorageHandle() noexcept { return store_view.handle; } store_view.Create(); - glTextureView(store_view.handle, ImageTarget(info), texture.handle, GL_RGBA8, 0, + glTextureView(store_view.handle, ImageTarget(info), current_texture, GL_RGBA8, 0, info.resources.levels, 0, info.resources.layers); return store_view.handle; default: - return texture.handle; + return current_texture; } } @@ -849,6 +889,140 @@ void Image::CopyImageToBuffer(const VideoCommon::BufferImageCopy& copy, size_t b } } +void Image::Scale(bool up_scale) { + const auto format_type = GetFormatType(info.format); + const GLenum attachment = [format_type] { + switch (format_type) { + case SurfaceType::ColorTexture: + return GL_COLOR_ATTACHMENT0; + case SurfaceType::Depth: + return GL_DEPTH_ATTACHMENT; + case SurfaceType::DepthStencil: + return GL_DEPTH_STENCIL_ATTACHMENT; + default: + UNREACHABLE(); + return GL_COLOR_ATTACHMENT0; + } + }(); + const GLenum mask = [format_type] { + switch (format_type) { + case SurfaceType::ColorTexture: + return GL_COLOR_BUFFER_BIT; + case SurfaceType::Depth: + return GL_DEPTH_BUFFER_BIT; + case SurfaceType::DepthStencil: + return GL_STENCIL_BUFFER_BIT | GL_DEPTH_BUFFER_BIT; + default: + UNREACHABLE(); + return GL_COLOR_BUFFER_BIT; + } + }(); + const size_t fbo_index = [format_type] { + switch (format_type) { + case SurfaceType::ColorTexture: + return 0; + case SurfaceType::Depth: + return 1; + case SurfaceType::DepthStencil: + return 2; + default: + UNREACHABLE(); + return 0; + } + }(); + const bool is_2d = info.type == ImageType::e2D; + const bool is_color{(mask & GL_COLOR_BUFFER_BIT) != 0}; + // Integer formats must use NEAREST filter + const bool linear_color_format{is_color && !IsPixelFormatInteger(info.format)}; + const GLenum filter = linear_color_format ? GL_LINEAR : GL_NEAREST; + + const auto& resolution = runtime->resolution; + const u32 scaled_width = resolution.ScaleUp(info.size.width); + const u32 scaled_height = is_2d ? resolution.ScaleUp(info.size.height) : info.size.height; + const u32 original_width = info.size.width; + const u32 original_height = info.size.height; + + if (!upscaled_backup.handle) { + auto dst_info = info; + dst_info.size.width = scaled_width; + dst_info.size.height = scaled_height; + upscaled_backup = MakeImage(dst_info, gl_internal_format); + } + const u32 src_width = up_scale ? original_width : scaled_width; + const u32 src_height = up_scale ? original_height : scaled_height; + const u32 dst_width = up_scale ? scaled_width : original_width; + const u32 dst_height = up_scale ? scaled_height : original_height; + const auto src_handle = up_scale ? texture.handle : upscaled_backup.handle; + const auto dst_handle = up_scale ? upscaled_backup.handle : texture.handle; + + // TODO (ameerj): Investigate other GL states that affect blitting. + glDisablei(GL_SCISSOR_TEST, 0); + glViewportIndexedf(0, 0.0f, 0.0f, static_cast<GLfloat>(dst_width), + static_cast<GLfloat>(dst_height)); + + const GLuint read_fbo = runtime->rescale_read_fbos[fbo_index].handle; + const GLuint draw_fbo = runtime->rescale_draw_fbos[fbo_index].handle; + for (s32 layer = 0; layer < info.resources.layers; ++layer) { + for (s32 level = 0; level < info.resources.levels; ++level) { + const u32 src_level_width = std::max(1u, src_width >> level); + const u32 src_level_height = std::max(1u, src_height >> level); + const u32 dst_level_width = std::max(1u, dst_width >> level); + const u32 dst_level_height = std::max(1u, dst_height >> level); + + glNamedFramebufferTextureLayer(read_fbo, attachment, src_handle, level, layer); + glNamedFramebufferTextureLayer(draw_fbo, attachment, dst_handle, level, layer); + + glBlitNamedFramebuffer(read_fbo, draw_fbo, 0, 0, src_level_width, src_level_height, 0, + 0, dst_level_width, dst_level_height, mask, filter); + } + } + current_texture = dst_handle; + auto& state_tracker = runtime->GetStateTracker(); + state_tracker.NotifyViewport0(); + state_tracker.NotifyScissor0(); +} + +bool Image::ScaleUp(bool ignore) { + if (True(flags & ImageFlagBits::Rescaled)) { + return false; + } + if (gl_format == 0 && gl_type == 0) { + // compressed textures + return false; + } + if (info.type == ImageType::Linear) { + UNREACHABLE(); + return false; + } + flags |= ImageFlagBits::Rescaled; + if (!runtime->resolution.active) { + return false; + } + has_scaled = true; + if (ignore) { + current_texture = upscaled_backup.handle; + return true; + } + Scale(true); + return true; +} + +bool Image::ScaleDown(bool ignore) { + if (False(flags & ImageFlagBits::Rescaled)) { + return false; + } + flags &= ~ImageFlagBits::Rescaled; + if (!runtime->resolution.active) { + return false; + } + if (ignore) { + current_texture = texture.handle; + return true; + } + Scale(false); + return true; +} + ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewInfo& info, ImageId image_id_, Image& image) : VideoCommon::ImageViewBase{info, image.info, image_id_}, views{runtime.null_image_views} { @@ -862,7 +1036,7 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI flat_range = info.range; set_object_label = device.HasDebuggingToolAttached(); is_render_target = info.IsRenderTarget(); - original_texture = image.texture.handle; + original_texture = image.Handle(); num_samples = image.info.num_samples; if (!is_render_target) { swizzle[0] = info.x_source; @@ -950,9 +1124,11 @@ ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, const VideoCommon::ImageViewInfo& view_info) : VideoCommon::ImageViewBase{info, view_info} {} -ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::NullImageParams& params) +ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::NullImageViewParams& params) : VideoCommon::ImageViewBase{params}, views{runtime.null_image_views} {} +ImageView::~ImageView() = default; + GLuint ImageView::StorageView(Shader::TextureType texture_type, Shader::ImageFormat image_format) { if (image_format == Shader::ImageFormat::Typeless) { return Handle(texture_type); @@ -1037,7 +1213,8 @@ Sampler::Sampler(TextureCacheRuntime& runtime, const TSCEntry& config) { glSamplerParameterfv(handle, GL_TEXTURE_BORDER_COLOR, config.BorderColor().data()); if (GLAD_GL_ARB_texture_filter_anisotropic || GLAD_GL_EXT_texture_filter_anisotropic) { - glSamplerParameterf(handle, GL_TEXTURE_MAX_ANISOTROPY, config.MaxAnisotropy()); + const f32 max_anisotropy = std::clamp(config.MaxAnisotropy(), 1.0f, 16.0f); + glSamplerParameterf(handle, GL_TEXTURE_MAX_ANISOTROPY, max_anisotropy); } else { LOG_WARNING(Render_OpenGL, "GL_ARB_texture_filter_anisotropic is required"); } @@ -1056,13 +1233,8 @@ Sampler::Sampler(TextureCacheRuntime& runtime, const TSCEntry& config) { Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM_RT> color_buffers, ImageView* depth_buffer, const VideoCommon::RenderTargets& key) { - // Bind to READ_FRAMEBUFFER to stop Nvidia's driver from creating an EXT_framebuffer instead of - // a core framebuffer. EXT framebuffer attachments have to match in size and can be shared - // across contexts. yuzu doesn't share framebuffers across contexts and we need attachments with - // mismatching size, this is why core framebuffers are preferred. - GLuint handle; - glGenFramebuffers(1, &handle); - glBindFramebuffer(GL_READ_FRAMEBUFFER, handle); + framebuffer.Create(); + GLuint handle = framebuffer.handle; GLsizei num_buffers = 0; std::array<GLenum, NUM_RT> gl_draw_buffers; @@ -1110,31 +1282,31 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM const std::string name = VideoCommon::Name(key); glObjectLabel(GL_FRAMEBUFFER, handle, static_cast<GLsizei>(name.size()), name.data()); } - framebuffer.handle = handle; } +Framebuffer::~Framebuffer() = default; + void BGRCopyPass::CopyBGR(Image& dst_image, Image& src_image, std::span<const VideoCommon::ImageCopy> copies) { static constexpr VideoCommon::Offset3D zero_offset{0, 0, 0}; - const u32 requested_pbo_size = - std::max(src_image.unswizzled_size_bytes, dst_image.unswizzled_size_bytes); - - if (bgr_pbo_size < requested_pbo_size) { - bgr_pbo.Create(); - bgr_pbo_size = requested_pbo_size; - glNamedBufferData(bgr_pbo.handle, bgr_pbo_size, nullptr, GL_STREAM_COPY); - } + const u32 img_bpp = BytesPerBlock(src_image.info.format); for (const ImageCopy& copy : copies) { ASSERT(copy.src_offset == zero_offset); ASSERT(copy.dst_offset == zero_offset); - + const u32 num_src_layers = static_cast<u32>(copy.src_subresource.num_layers); + const u32 copy_size = copy.extent.width * copy.extent.height * num_src_layers * img_bpp; + if (bgr_pbo_size < copy_size) { + bgr_pbo.Create(); + bgr_pbo_size = copy_size; + glNamedBufferData(bgr_pbo.handle, bgr_pbo_size, nullptr, GL_STREAM_COPY); + } // Copy from source to PBO glPixelStorei(GL_PACK_ALIGNMENT, 1); glPixelStorei(GL_PACK_ROW_LENGTH, copy.extent.width); glBindBuffer(GL_PIXEL_PACK_BUFFER, bgr_pbo.handle); glGetTextureSubImage(src_image.Handle(), 0, 0, 0, 0, copy.extent.width, copy.extent.height, - copy.src_subresource.num_layers, src_image.GlFormat(), - src_image.GlType(), static_cast<GLsizei>(bgr_pbo_size), nullptr); + num_src_layers, src_image.GlFormat(), src_image.GlType(), + static_cast<GLsizei>(bgr_pbo_size), nullptr); // Copy from PBO to destination in desired GL format glPixelStorei(GL_UNPACK_ALIGNMENT, 1); diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h index 1ca2c90be..1bb762568 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.h +++ b/src/video_core/renderer_opengl/gl_texture_cache.h @@ -15,6 +15,10 @@ #include "video_core/texture_cache/image_view_base.h" #include "video_core/texture_cache/texture_cache_base.h" +namespace Settings { +struct ResolutionScalingInfo; +} + namespace OpenGL { class Device; @@ -78,9 +82,11 @@ public: ImageBufferMap DownloadStagingBuffer(size_t size); + u64 GetDeviceLocalMemory() const; + void CopyImage(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies); - void ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view) { + void ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view, bool rescaled) { UNIMPLEMENTED(); } @@ -110,6 +116,12 @@ public: bool HasNativeASTC() const noexcept; + void TickFrame() {} + + StateTracker& GetStateTracker() { + return state_tracker; + } + private: struct StagingBuffers { explicit StagingBuffers(GLenum storage_flags_, GLenum map_flags_); @@ -149,6 +161,10 @@ private: OGLTextureView null_image_view_cube; std::array<GLuint, Shader::NUM_TEXTURE_TYPES> null_image_views{}; + + std::array<OGLFramebuffer, 3> rescale_draw_fbos; + std::array<OGLFramebuffer, 3> rescale_read_fbos; + const Settings::ResolutionScalingInfo& resolution; }; class Image : public VideoCommon::ImageBase { @@ -157,6 +173,7 @@ class Image : public VideoCommon::ImageBase { public: explicit Image(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, GPUVAddr gpu_addr, VAddr cpu_addr); + explicit Image(const VideoCommon::NullImageParams&); ~Image(); @@ -174,7 +191,7 @@ public: GLuint StorageHandle() noexcept; GLuint Handle() const noexcept { - return texture.handle; + return current_texture; } GLuint GlFormat() const noexcept { @@ -185,16 +202,25 @@ public: return gl_type; } + bool ScaleUp(bool ignore = false); + + bool ScaleDown(bool ignore = false); + private: void CopyBufferToImage(const VideoCommon::BufferImageCopy& copy, size_t buffer_offset); void CopyImageToBuffer(const VideoCommon::BufferImageCopy& copy, size_t buffer_offset); + void Scale(bool up_scale); + OGLTexture texture; + OGLTexture upscaled_backup; OGLTextureView store_view; GLenum gl_internal_format = GL_NONE; GLenum gl_format = GL_NONE; GLenum gl_type = GL_NONE; + TextureCacheRuntime* runtime{}; + GLuint current_texture{}; }; class ImageView : public VideoCommon::ImageViewBase { @@ -206,7 +232,15 @@ public: const VideoCommon::ImageViewInfo&, GPUVAddr); explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, const VideoCommon::ImageViewInfo& view_info); - explicit ImageView(TextureCacheRuntime&, const VideoCommon::NullImageParams&); + explicit ImageView(TextureCacheRuntime&, const VideoCommon::NullImageViewParams&); + + ~ImageView(); + + ImageView(const ImageView&) = delete; + ImageView& operator=(const ImageView&) = delete; + + ImageView(ImageView&&) = default; + ImageView& operator=(ImageView&&) = default; [[nodiscard]] GLuint StorageView(Shader::TextureType texture_type, Shader::ImageFormat image_format); @@ -276,6 +310,14 @@ public: explicit Framebuffer(TextureCacheRuntime&, std::span<ImageView*, NUM_RT> color_buffers, ImageView* depth_buffer, const VideoCommon::RenderTargets& key); + ~Framebuffer(); + + Framebuffer(const Framebuffer&) = delete; + Framebuffer& operator=(const Framebuffer&) = delete; + + Framebuffer(Framebuffer&&) = default; + Framebuffer& operator=(Framebuffer&&) = default; + [[nodiscard]] GLuint Handle() const noexcept { return framebuffer.handle; } @@ -293,7 +335,7 @@ struct TextureCacheParams { static constexpr bool ENABLE_VALIDATION = true; static constexpr bool FRAMEBUFFER_BLITS = true; static constexpr bool HAS_EMULATED_COPIES = true; - static constexpr bool HAS_DEVICE_MEMORY_INFO = false; + static constexpr bool HAS_DEVICE_MEMORY_INFO = true; using Runtime = OpenGL::TextureCacheRuntime; using Image = OpenGL::Image; diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp index 7d7cba69c..28daacd82 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.cpp +++ b/src/video_core/renderer_opengl/renderer_opengl.cpp @@ -21,8 +21,13 @@ #include "core/memory.h" #include "core/perf_stats.h" #include "core/telemetry_session.h" +#include "video_core/host_shaders/fxaa_frag.h" +#include "video_core/host_shaders/fxaa_vert.h" #include "video_core/host_shaders/opengl_present_frag.h" +#include "video_core/host_shaders/opengl_present_scaleforce_frag.h" #include "video_core/host_shaders/opengl_present_vert.h" +#include "video_core/host_shaders/present_bicubic_frag.h" +#include "video_core/host_shaders/present_gaussian_frag.h" #include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_shader_manager.h" #include "video_core/renderer_opengl/gl_shader_util.h" @@ -208,7 +213,9 @@ void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuf framebuffer_crop_rect = framebuffer.crop_rect; const VAddr framebuffer_addr{framebuffer.address + framebuffer.offset}; - if (rasterizer.AccelerateDisplay(framebuffer, framebuffer_addr, framebuffer.stride)) { + screen_info.was_accelerated = + rasterizer.AccelerateDisplay(framebuffer, framebuffer_addr, framebuffer.stride); + if (screen_info.was_accelerated) { return; } @@ -251,12 +258,25 @@ void RendererOpenGL::LoadColorToActiveGLTexture(u8 color_r, u8 color_g, u8 color void RendererOpenGL::InitOpenGLObjects() { // Create shader programs + fxaa_vertex = CreateProgram(HostShaders::FXAA_VERT, GL_VERTEX_SHADER); + fxaa_fragment = CreateProgram(HostShaders::FXAA_FRAG, GL_FRAGMENT_SHADER); present_vertex = CreateProgram(HostShaders::OPENGL_PRESENT_VERT, GL_VERTEX_SHADER); - present_fragment = CreateProgram(HostShaders::OPENGL_PRESENT_FRAG, GL_FRAGMENT_SHADER); + present_bilinear_fragment = CreateProgram(HostShaders::OPENGL_PRESENT_FRAG, GL_FRAGMENT_SHADER); + present_bicubic_fragment = CreateProgram(HostShaders::PRESENT_BICUBIC_FRAG, GL_FRAGMENT_SHADER); + present_gaussian_fragment = + CreateProgram(HostShaders::PRESENT_GAUSSIAN_FRAG, GL_FRAGMENT_SHADER); + present_scaleforce_fragment = + CreateProgram(fmt::format("#version 460\n{}", HostShaders::OPENGL_PRESENT_SCALEFORCE_FRAG), + GL_FRAGMENT_SHADER); // Generate presentation sampler present_sampler.Create(); glSamplerParameteri(present_sampler.handle, GL_TEXTURE_MIN_FILTER, GL_LINEAR); + glSamplerParameteri(present_sampler.handle, GL_TEXTURE_MAG_FILTER, GL_LINEAR); + + present_sampler_nn.Create(); + glSamplerParameteri(present_sampler_nn.handle, GL_TEXTURE_MIN_FILTER, GL_NEAREST); + glSamplerParameteri(present_sampler_nn.handle, GL_TEXTURE_MAG_FILTER, GL_NEAREST); // Generate VBO handle for drawing vertex_buffer.Create(); @@ -274,6 +294,8 @@ void RendererOpenGL::InitOpenGLObjects() { // Clear screen to black LoadColorToActiveGLTexture(0, 0, 0, 0, screen_info.texture); + + fxaa_framebuffer.Create(); } void RendererOpenGL::AddTelemetryFields() { @@ -325,18 +347,130 @@ void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture, texture.resource.Release(); texture.resource.Create(GL_TEXTURE_2D); glTextureStorage2D(texture.resource.handle, 1, internal_format, texture.width, texture.height); + fxaa_texture.Release(); + fxaa_texture.Create(GL_TEXTURE_2D); + glTextureStorage2D(fxaa_texture.handle, 1, GL_RGBA16F, + Settings::values.resolution_info.ScaleUp(screen_info.texture.width), + Settings::values.resolution_info.ScaleUp(screen_info.texture.height)); + glNamedFramebufferTexture(fxaa_framebuffer.handle, GL_COLOR_ATTACHMENT0, fxaa_texture.handle, + 0); } void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { + // TODO: Signal state tracker about these changes + state_tracker.NotifyScreenDrawVertexArray(); + state_tracker.NotifyPolygonModes(); + state_tracker.NotifyViewport0(); + state_tracker.NotifyScissor0(); + state_tracker.NotifyColorMask(0); + state_tracker.NotifyBlend0(); + state_tracker.NotifyFramebuffer(); + state_tracker.NotifyFrontFace(); + state_tracker.NotifyCullTest(); + state_tracker.NotifyDepthTest(); + state_tracker.NotifyStencilTest(); + state_tracker.NotifyPolygonOffset(); + state_tracker.NotifyRasterizeEnable(); + state_tracker.NotifyFramebufferSRGB(); + state_tracker.NotifyLogicOp(); + state_tracker.NotifyClipControl(); + state_tracker.NotifyAlphaTest(); + + state_tracker.ClipControl(GL_LOWER_LEFT, GL_ZERO_TO_ONE); + // Update background color before drawing glClearColor(Settings::values.bg_red.GetValue() / 255.0f, Settings::values.bg_green.GetValue() / 255.0f, Settings::values.bg_blue.GetValue() / 255.0f, 1.0f); + glEnable(GL_CULL_FACE); + glDisable(GL_COLOR_LOGIC_OP); + glDisable(GL_DEPTH_TEST); + glDisable(GL_STENCIL_TEST); + glDisable(GL_POLYGON_OFFSET_FILL); + glDisable(GL_RASTERIZER_DISCARD); + glDisable(GL_ALPHA_TEST); + glDisablei(GL_BLEND, 0); + glPolygonMode(GL_FRONT_AND_BACK, GL_FILL); + glCullFace(GL_BACK); + glFrontFace(GL_CW); + glColorMaski(0, GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE); + + glBindTextureUnit(0, screen_info.display_texture); + + if (Settings::values.anti_aliasing.GetValue() == Settings::AntiAliasing::Fxaa) { + program_manager.BindPresentPrograms(fxaa_vertex.handle, fxaa_fragment.handle); + + glEnablei(GL_SCISSOR_TEST, 0); + auto viewport_width = screen_info.texture.width; + auto scissor_width = framebuffer_crop_rect.GetWidth(); + if (scissor_width <= 0) { + scissor_width = viewport_width; + } + auto viewport_height = screen_info.texture.height; + auto scissor_height = framebuffer_crop_rect.GetHeight(); + if (scissor_height <= 0) { + scissor_height = viewport_height; + } + if (screen_info.was_accelerated) { + viewport_width = Settings::values.resolution_info.ScaleUp(viewport_width); + scissor_width = Settings::values.resolution_info.ScaleUp(scissor_width); + viewport_height = Settings::values.resolution_info.ScaleUp(viewport_height); + scissor_height = Settings::values.resolution_info.ScaleUp(scissor_height); + } + glScissorIndexed(0, 0, 0, scissor_width, scissor_height); + glViewportIndexedf(0, 0.0f, 0.0f, static_cast<GLfloat>(viewport_width), + static_cast<GLfloat>(viewport_height)); + glDepthRangeIndexed(0, 0.0, 0.0); + + glBindSampler(0, present_sampler.handle); + GLint old_read_fb; + GLint old_draw_fb; + glGetIntegerv(GL_READ_FRAMEBUFFER_BINDING, &old_read_fb); + glGetIntegerv(GL_DRAW_FRAMEBUFFER_BINDING, &old_draw_fb); + glBindFramebuffer(GL_DRAW_FRAMEBUFFER, fxaa_framebuffer.handle); + + glDrawArrays(GL_TRIANGLE_STRIP, 0, 4); + + glBindFramebuffer(GL_READ_FRAMEBUFFER, old_read_fb); + glBindFramebuffer(GL_DRAW_FRAMEBUFFER, old_draw_fb); + + glBindTextureUnit(0, fxaa_texture.handle); + } + // Set projection matrix const std::array ortho_matrix = MakeOrthographicMatrix(static_cast<float>(layout.width), static_cast<float>(layout.height)); - program_manager.BindPresentPrograms(present_vertex.handle, present_fragment.handle); + + GLuint fragment_handle; + const auto filter = Settings::values.scaling_filter.GetValue(); + switch (filter) { + case Settings::ScalingFilter::NearestNeighbor: + fragment_handle = present_bilinear_fragment.handle; + break; + case Settings::ScalingFilter::Bilinear: + fragment_handle = present_bilinear_fragment.handle; + break; + case Settings::ScalingFilter::Bicubic: + fragment_handle = present_bicubic_fragment.handle; + break; + case Settings::ScalingFilter::Gaussian: + fragment_handle = present_gaussian_fragment.handle; + break; + case Settings::ScalingFilter::ScaleForce: + fragment_handle = present_scaleforce_fragment.handle; + break; + case Settings::ScalingFilter::Fsr: + LOG_WARNING( + Render_OpenGL, + "FidelityFX FSR Super Sampling is not supported in OpenGL, changing to ScaleForce"); + fragment_handle = present_scaleforce_fragment.handle; + break; + default: + fragment_handle = present_bilinear_fragment.handle; + break; + } + program_manager.BindPresentPrograms(present_vertex.handle, fragment_handle); glProgramUniformMatrix3x2fv(present_vertex.handle, ModelViewMatrixLocation, 1, GL_FALSE, ortho_matrix.data()); @@ -370,6 +504,11 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { scale_v = static_cast<f32>(framebuffer_crop_rect.GetHeight()) / static_cast<f32>(screen_info.texture.height); } + if (Settings::values.anti_aliasing.GetValue() == Settings::AntiAliasing::Fxaa && + !screen_info.was_accelerated) { + scale_u /= Settings::values.resolution_info.up_factor; + scale_v /= Settings::values.resolution_info.up_factor; + } const auto& screen = layout.screen; const std::array vertices = { @@ -380,47 +519,14 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { }; glNamedBufferSubData(vertex_buffer.handle, 0, sizeof(vertices), std::data(vertices)); - // TODO: Signal state tracker about these changes - state_tracker.NotifyScreenDrawVertexArray(); - state_tracker.NotifyPolygonModes(); - state_tracker.NotifyViewport0(); - state_tracker.NotifyScissor0(); - state_tracker.NotifyColorMask(0); - state_tracker.NotifyBlend0(); - state_tracker.NotifyFramebuffer(); - state_tracker.NotifyFrontFace(); - state_tracker.NotifyCullTest(); - state_tracker.NotifyDepthTest(); - state_tracker.NotifyStencilTest(); - state_tracker.NotifyPolygonOffset(); - state_tracker.NotifyRasterizeEnable(); - state_tracker.NotifyFramebufferSRGB(); - state_tracker.NotifyLogicOp(); - state_tracker.NotifyClipControl(); - state_tracker.NotifyAlphaTest(); - - state_tracker.ClipControl(GL_LOWER_LEFT, GL_ZERO_TO_ONE); - glEnable(GL_CULL_FACE); if (screen_info.display_srgb) { glEnable(GL_FRAMEBUFFER_SRGB); } else { glDisable(GL_FRAMEBUFFER_SRGB); } - glDisable(GL_COLOR_LOGIC_OP); - glDisable(GL_DEPTH_TEST); - glDisable(GL_STENCIL_TEST); - glDisable(GL_POLYGON_OFFSET_FILL); - glDisable(GL_RASTERIZER_DISCARD); - glDisable(GL_ALPHA_TEST); - glDisablei(GL_BLEND, 0); glDisablei(GL_SCISSOR_TEST, 0); - glPolygonMode(GL_FRONT_AND_BACK, GL_FILL); - glCullFace(GL_BACK); - glFrontFace(GL_CW); - glColorMaski(0, GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE); glViewportIndexedf(0, 0.0f, 0.0f, static_cast<GLfloat>(layout.width), static_cast<GLfloat>(layout.height)); - glDepthRangeIndexed(0, 0.0, 0.0); glEnableVertexAttribArray(PositionLocation); glEnableVertexAttribArray(TexCoordLocation); @@ -440,8 +546,11 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { glBindVertexBuffer(0, vertex_buffer.handle, 0, sizeof(ScreenRectVertex)); } - glBindTextureUnit(0, screen_info.display_texture); - glBindSampler(0, present_sampler.handle); + if (Settings::values.scaling_filter.GetValue() != Settings::ScalingFilter::NearestNeighbor) { + glBindSampler(0, present_sampler.handle); + } else { + glBindSampler(0, present_sampler_nn.handle); + } glClear(GL_COLOR_BUFFER_BIT); glDrawArrays(GL_TRIANGLE_STRIP, 0, 4); diff --git a/src/video_core/renderer_opengl/renderer_opengl.h b/src/video_core/renderer_opengl/renderer_opengl.h index d455f572f..cda333cad 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.h +++ b/src/video_core/renderer_opengl/renderer_opengl.h @@ -50,6 +50,7 @@ struct TextureInfo { /// Structure used for storing information about the display target for the Switch screen struct ScreenInfo { GLuint display_texture{}; + bool was_accelerated = false; bool display_srgb{}; const Common::Rectangle<float> display_texcoords{0.0f, 0.0f, 1.0f, 1.0f}; TextureInfo texture; @@ -109,9 +110,15 @@ private: // OpenGL object IDs OGLSampler present_sampler; + OGLSampler present_sampler_nn; OGLBuffer vertex_buffer; + OGLProgram fxaa_vertex; + OGLProgram fxaa_fragment; OGLProgram present_vertex; - OGLProgram present_fragment; + OGLProgram present_bilinear_fragment; + OGLProgram present_bicubic_fragment; + OGLProgram present_gaussian_fragment; + OGLProgram present_scaleforce_fragment; OGLFramebuffer screenshot_framebuffer; // GPU address of the vertex buffer @@ -119,6 +126,8 @@ private: /// Display information for Switch screen ScreenInfo screen_info; + OGLTexture fxaa_texture; + OGLFramebuffer fxaa_framebuffer; /// OpenGL framebuffer data std::vector<u8> gl_framebuffer_data; diff --git a/src/video_core/renderer_vulkan/blit_image.cpp b/src/video_core/renderer_vulkan/blit_image.cpp index 6c1b2f063..b3884a4f5 100644 --- a/src/video_core/renderer_vulkan/blit_image.cpp +++ b/src/video_core/renderer_vulkan/blit_image.cpp @@ -363,7 +363,7 @@ BlitImageHelper::BlitImageHelper(const Device& device_, VKScheduler& scheduler_, BlitImageHelper::~BlitImageHelper() = default; -void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, const ImageView& src_image_view, +void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, VkImageView src_view, const Region2D& dst_region, const Region2D& src_region, Tegra::Engines::Fermi2D::Filter filter, Tegra::Engines::Fermi2D::Operation operation) { @@ -373,9 +373,8 @@ void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, const ImageV .operation = operation, }; const VkPipelineLayout layout = *one_texture_pipeline_layout; - const VkImageView src_view = src_image_view.Handle(Shader::TextureType::Color2D); const VkSampler sampler = is_linear ? *linear_sampler : *nearest_sampler; - const VkPipeline pipeline = FindOrEmplacePipeline(key); + const VkPipeline pipeline = FindOrEmplaceColorPipeline(key); scheduler.RequestRenderpass(dst_framebuffer); scheduler.Record([this, dst_region, src_region, pipeline, layout, sampler, src_view](vk::CommandBuffer cmdbuf) { @@ -398,10 +397,13 @@ void BlitImageHelper::BlitDepthStencil(const Framebuffer* dst_framebuffer, Tegra::Engines::Fermi2D::Operation operation) { ASSERT(filter == Tegra::Engines::Fermi2D::Filter::Point); ASSERT(operation == Tegra::Engines::Fermi2D::Operation::SrcCopy); - + const BlitImagePipelineKey key{ + .renderpass = dst_framebuffer->RenderPass(), + .operation = operation, + }; const VkPipelineLayout layout = *two_textures_pipeline_layout; const VkSampler sampler = *nearest_sampler; - const VkPipeline pipeline = BlitDepthStencilPipeline(dst_framebuffer->RenderPass()); + const VkPipeline pipeline = FindOrEmplaceDepthStencilPipeline(key); scheduler.RequestRenderpass(dst_framebuffer); scheduler.Record([dst_region, src_region, pipeline, layout, sampler, src_depth_view, src_stencil_view, this](vk::CommandBuffer cmdbuf) { @@ -419,40 +421,45 @@ void BlitImageHelper::BlitDepthStencil(const Framebuffer* dst_framebuffer, } void BlitImageHelper::ConvertD32ToR32(const Framebuffer* dst_framebuffer, - const ImageView& src_image_view) { + const ImageView& src_image_view, u32 up_scale, + u32 down_shift) { ConvertDepthToColorPipeline(convert_d32_to_r32_pipeline, dst_framebuffer->RenderPass()); - Convert(*convert_d32_to_r32_pipeline, dst_framebuffer, src_image_view); + Convert(*convert_d32_to_r32_pipeline, dst_framebuffer, src_image_view, up_scale, down_shift); } void BlitImageHelper::ConvertR32ToD32(const Framebuffer* dst_framebuffer, - const ImageView& src_image_view) { + const ImageView& src_image_view, u32 up_scale, + u32 down_shift) { ConvertColorToDepthPipeline(convert_r32_to_d32_pipeline, dst_framebuffer->RenderPass()); - Convert(*convert_r32_to_d32_pipeline, dst_framebuffer, src_image_view); + Convert(*convert_r32_to_d32_pipeline, dst_framebuffer, src_image_view, up_scale, down_shift); } void BlitImageHelper::ConvertD16ToR16(const Framebuffer* dst_framebuffer, - const ImageView& src_image_view) { + const ImageView& src_image_view, u32 up_scale, + u32 down_shift) { ConvertDepthToColorPipeline(convert_d16_to_r16_pipeline, dst_framebuffer->RenderPass()); - Convert(*convert_d16_to_r16_pipeline, dst_framebuffer, src_image_view); + Convert(*convert_d16_to_r16_pipeline, dst_framebuffer, src_image_view, up_scale, down_shift); } void BlitImageHelper::ConvertR16ToD16(const Framebuffer* dst_framebuffer, - const ImageView& src_image_view) { + const ImageView& src_image_view, u32 up_scale, + u32 down_shift) { ConvertColorToDepthPipeline(convert_r16_to_d16_pipeline, dst_framebuffer->RenderPass()); - Convert(*convert_r16_to_d16_pipeline, dst_framebuffer, src_image_view); + Convert(*convert_r16_to_d16_pipeline, dst_framebuffer, src_image_view, up_scale, down_shift); } void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer, - const ImageView& src_image_view) { + const ImageView& src_image_view, u32 up_scale, u32 down_shift) { const VkPipelineLayout layout = *one_texture_pipeline_layout; const VkImageView src_view = src_image_view.Handle(Shader::TextureType::Color2D); const VkSampler sampler = *nearest_sampler; const VkExtent2D extent{ - .width = src_image_view.size.width, - .height = src_image_view.size.height, + .width = std::max((src_image_view.size.width * up_scale) >> down_shift, 1U), + .height = std::max((src_image_view.size.height * up_scale) >> down_shift, 1U), }; scheduler.RequestRenderpass(dst_framebuffer); - scheduler.Record([pipeline, layout, sampler, src_view, extent, this](vk::CommandBuffer cmdbuf) { + scheduler.Record([pipeline, layout, sampler, src_view, extent, up_scale, down_shift, + this](vk::CommandBuffer cmdbuf) { const VkOffset2D offset{ .x = 0, .y = 0, @@ -488,7 +495,7 @@ void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_frameb scheduler.InvalidateState(); } -VkPipeline BlitImageHelper::FindOrEmplacePipeline(const BlitImagePipelineKey& key) { +VkPipeline BlitImageHelper::FindOrEmplaceColorPipeline(const BlitImagePipelineKey& key) { const auto it = std::ranges::find(blit_color_keys, key); if (it != blit_color_keys.end()) { return *blit_color_pipelines[std::distance(blit_color_keys.begin(), it)]; @@ -542,12 +549,14 @@ VkPipeline BlitImageHelper::FindOrEmplacePipeline(const BlitImagePipelineKey& ke return *blit_color_pipelines.back(); } -VkPipeline BlitImageHelper::BlitDepthStencilPipeline(VkRenderPass renderpass) { - if (blit_depth_stencil_pipeline) { - return *blit_depth_stencil_pipeline; +VkPipeline BlitImageHelper::FindOrEmplaceDepthStencilPipeline(const BlitImagePipelineKey& key) { + const auto it = std::ranges::find(blit_depth_stencil_keys, key); + if (it != blit_depth_stencil_keys.end()) { + return *blit_depth_stencil_pipelines[std::distance(blit_depth_stencil_keys.begin(), it)]; } + blit_depth_stencil_keys.push_back(key); const std::array stages = MakeStages(*full_screen_vert, *blit_depth_stencil_frag); - blit_depth_stencil_pipeline = device.GetLogical().CreateGraphicsPipeline({ + blit_depth_stencil_pipelines.push_back(device.GetLogical().CreateGraphicsPipeline({ .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, .pNext = nullptr, .flags = 0, @@ -560,15 +569,15 @@ VkPipeline BlitImageHelper::BlitDepthStencilPipeline(VkRenderPass renderpass) { .pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO, .pMultisampleState = &PIPELINE_MULTISAMPLE_STATE_CREATE_INFO, .pDepthStencilState = &PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO, - .pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_EMPTY_CREATE_INFO, + .pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_GENERIC_CREATE_INFO, .pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO, .layout = *two_textures_pipeline_layout, - .renderPass = renderpass, + .renderPass = key.renderpass, .subpass = 0, .basePipelineHandle = VK_NULL_HANDLE, .basePipelineIndex = 0, - }); - return *blit_depth_stencil_pipeline; + })); + return *blit_depth_stencil_pipelines.back(); } void BlitImageHelper::ConvertDepthToColorPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass) { diff --git a/src/video_core/renderer_vulkan/blit_image.h b/src/video_core/renderer_vulkan/blit_image.h index 33ee095c1..d77f76678 100644 --- a/src/video_core/renderer_vulkan/blit_image.h +++ b/src/video_core/renderer_vulkan/blit_image.h @@ -34,7 +34,7 @@ public: StateTracker& state_tracker, DescriptorPool& descriptor_pool); ~BlitImageHelper(); - void BlitColor(const Framebuffer* dst_framebuffer, const ImageView& src_image_view, + void BlitColor(const Framebuffer* dst_framebuffer, VkImageView src_image_view, const Region2D& dst_region, const Region2D& src_region, Tegra::Engines::Fermi2D::Filter filter, Tegra::Engines::Fermi2D::Operation operation); @@ -44,21 +44,25 @@ public: const Region2D& src_region, Tegra::Engines::Fermi2D::Filter filter, Tegra::Engines::Fermi2D::Operation operation); - void ConvertD32ToR32(const Framebuffer* dst_framebuffer, const ImageView& src_image_view); + void ConvertD32ToR32(const Framebuffer* dst_framebuffer, const ImageView& src_image_view, + u32 up_scale, u32 down_shift); - void ConvertR32ToD32(const Framebuffer* dst_framebuffer, const ImageView& src_image_view); + void ConvertR32ToD32(const Framebuffer* dst_framebuffer, const ImageView& src_image_view, + u32 up_scale, u32 down_shift); - void ConvertD16ToR16(const Framebuffer* dst_framebuffer, const ImageView& src_image_view); + void ConvertD16ToR16(const Framebuffer* dst_framebuffer, const ImageView& src_image_view, + u32 up_scale, u32 down_shift); - void ConvertR16ToD16(const Framebuffer* dst_framebuffer, const ImageView& src_image_view); + void ConvertR16ToD16(const Framebuffer* dst_framebuffer, const ImageView& src_image_view, + u32 up_scale, u32 down_shift); private: void Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer, - const ImageView& src_image_view); + const ImageView& src_image_view, u32 up_scale, u32 down_shift); - [[nodiscard]] VkPipeline FindOrEmplacePipeline(const BlitImagePipelineKey& key); + [[nodiscard]] VkPipeline FindOrEmplaceColorPipeline(const BlitImagePipelineKey& key); - [[nodiscard]] VkPipeline BlitDepthStencilPipeline(VkRenderPass renderpass); + [[nodiscard]] VkPipeline FindOrEmplaceDepthStencilPipeline(const BlitImagePipelineKey& key); void ConvertDepthToColorPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass); @@ -84,7 +88,8 @@ private: std::vector<BlitImagePipelineKey> blit_color_keys; std::vector<vk::Pipeline> blit_color_pipelines; - vk::Pipeline blit_depth_stencil_pipeline; + std::vector<BlitImagePipelineKey> blit_depth_stencil_keys; + std::vector<vk::Pipeline> blit_depth_stencil_pipelines; vk::Pipeline convert_d32_to_r32_pipeline; vk::Pipeline convert_r32_to_d32_pipeline; vk::Pipeline convert_d16_to_r16_pipeline; diff --git a/src/video_core/renderer_vulkan/pipeline_helper.h b/src/video_core/renderer_vulkan/pipeline_helper.h index 4847db6b6..11c160570 100644 --- a/src/video_core/renderer_vulkan/pipeline_helper.h +++ b/src/video_core/renderer_vulkan/pipeline_helper.h @@ -10,6 +10,7 @@ #include "common/assert.h" #include "common/common_types.h" +#include "shader_recompiler/backend/spirv/emit_spirv.h" #include "shader_recompiler/shader_info.h" #include "video_core/renderer_vulkan/vk_texture_cache.h" #include "video_core/renderer_vulkan/vk_update_descriptor.h" @@ -20,6 +21,8 @@ namespace Vulkan { +using Shader::Backend::SPIRV::NUM_TEXTURE_AND_IMAGE_SCALING_WORDS; + class DescriptorLayoutBuilder { public: DescriptorLayoutBuilder(const Device& device_) : device{&device_} {} @@ -68,18 +71,28 @@ public: } vk::PipelineLayout CreatePipelineLayout(VkDescriptorSetLayout descriptor_set_layout) const { + using Shader::Backend::SPIRV::RescalingLayout; + const u32 size_offset = is_compute ? sizeof(RescalingLayout::down_factor) : 0u; + const VkPushConstantRange range{ + .stageFlags = static_cast<VkShaderStageFlags>( + is_compute ? VK_SHADER_STAGE_COMPUTE_BIT : VK_SHADER_STAGE_ALL_GRAPHICS), + .offset = 0, + .size = static_cast<u32>(sizeof(RescalingLayout)) - size_offset, + }; return device->GetLogical().CreatePipelineLayout({ .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, .pNext = nullptr, .flags = 0, .setLayoutCount = descriptor_set_layout ? 1U : 0U, .pSetLayouts = bindings.empty() ? nullptr : &descriptor_set_layout, - .pushConstantRangeCount = 0, - .pPushConstantRanges = nullptr, + .pushConstantRangeCount = 1, + .pPushConstantRanges = &range, }); } void Add(const Shader::Info& info, VkShaderStageFlags stage) { + is_compute |= (stage & VK_SHADER_STAGE_COMPUTE_BIT) != 0; + Add(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, stage, info.constant_buffer_descriptors); Add(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, stage, info.storage_buffers_descriptors); Add(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, stage, info.texture_buffer_descriptors); @@ -115,6 +128,7 @@ private: } const Device* device{}; + bool is_compute{}; boost::container::small_vector<VkDescriptorSetLayoutBinding, 32> bindings; boost::container::small_vector<VkDescriptorUpdateTemplateEntryKHR, 32> entries; u32 binding{}; @@ -122,31 +136,68 @@ private: size_t offset{}; }; -inline void PushImageDescriptors(const Shader::Info& info, const VkSampler*& samplers, - const ImageId*& image_view_ids, TextureCache& texture_cache, - VKUpdateDescriptorQueue& update_descriptor_queue) { - for (const auto& desc : info.texture_buffer_descriptors) { - image_view_ids += desc.count; +class RescalingPushConstant { +public: + explicit RescalingPushConstant() noexcept {} + + void PushTexture(bool is_rescaled) noexcept { + *texture_ptr |= is_rescaled ? texture_bit : 0u; + texture_bit <<= 1u; + if (texture_bit == 0u) { + texture_bit = 1u; + ++texture_ptr; + } } - for (const auto& desc : info.image_buffer_descriptors) { - image_view_ids += desc.count; + + void PushImage(bool is_rescaled) noexcept { + *image_ptr |= is_rescaled ? image_bit : 0u; + image_bit <<= 1u; + if (image_bit == 0u) { + image_bit = 1u; + ++image_ptr; + } } + + const std::array<u32, NUM_TEXTURE_AND_IMAGE_SCALING_WORDS>& Data() const noexcept { + return words; + } + +private: + std::array<u32, NUM_TEXTURE_AND_IMAGE_SCALING_WORDS> words{}; + u32* texture_ptr{words.data()}; + u32* image_ptr{words.data() + Shader::Backend::SPIRV::NUM_TEXTURE_SCALING_WORDS}; + u32 texture_bit{1u}; + u32 image_bit{1u}; +}; + +inline void PushImageDescriptors(TextureCache& texture_cache, + VKUpdateDescriptorQueue& update_descriptor_queue, + const Shader::Info& info, RescalingPushConstant& rescaling, + const VkSampler*& samplers, + const VideoCommon::ImageViewInOut*& views) { + const u32 num_texture_buffers = Shader::NumDescriptors(info.texture_buffer_descriptors); + const u32 num_image_buffers = Shader::NumDescriptors(info.image_buffer_descriptors); + views += num_texture_buffers; + views += num_image_buffers; for (const auto& desc : info.texture_descriptors) { for (u32 index = 0; index < desc.count; ++index) { + const VideoCommon::ImageViewId image_view_id{(views++)->id}; const VkSampler sampler{*(samplers++)}; - ImageView& image_view{texture_cache.GetImageView(*(image_view_ids++))}; + ImageView& image_view{texture_cache.GetImageView(image_view_id)}; const VkImageView vk_image_view{image_view.Handle(desc.type)}; update_descriptor_queue.AddSampledImage(vk_image_view, sampler); + rescaling.PushTexture(texture_cache.IsRescaling(image_view)); } } for (const auto& desc : info.image_descriptors) { for (u32 index = 0; index < desc.count; ++index) { - ImageView& image_view{texture_cache.GetImageView(*(image_view_ids++))}; + ImageView& image_view{texture_cache.GetImageView((views++)->id)}; if (desc.is_written) { texture_cache.MarkModification(image_view.image_id); } const VkImageView vk_image_view{image_view.StorageView(desc.type, desc.format)}; update_descriptor_queue.AddImage(vk_image_view); + rescaling.PushImage(texture_cache.IsRescaling(image_view)); } } } diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.cpp b/src/video_core/renderer_vulkan/vk_blit_screen.cpp index 888bc7392..1e447e621 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.cpp +++ b/src/video_core/renderer_vulkan/vk_blit_screen.cpp @@ -12,14 +12,22 @@ #include "common/assert.h" #include "common/common_types.h" #include "common/math_util.h" +#include "common/settings.h" #include "core/core.h" #include "core/frontend/emu_window.h" #include "core/memory.h" #include "video_core/gpu.h" +#include "video_core/host_shaders/fxaa_frag_spv.h" +#include "video_core/host_shaders/fxaa_vert_spv.h" +#include "video_core/host_shaders/present_bicubic_frag_spv.h" +#include "video_core/host_shaders/present_gaussian_frag_spv.h" #include "video_core/host_shaders/vulkan_present_frag_spv.h" +#include "video_core/host_shaders/vulkan_present_scaleforce_fp16_frag_spv.h" +#include "video_core/host_shaders/vulkan_present_scaleforce_fp32_frag_spv.h" #include "video_core/host_shaders/vulkan_present_vert_spv.h" #include "video_core/renderer_vulkan/renderer_vulkan.h" #include "video_core/renderer_vulkan/vk_blit_screen.h" +#include "video_core/renderer_vulkan/vk_fsr.h" #include "video_core/renderer_vulkan/vk_master_semaphore.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_shader_util.h" @@ -144,8 +152,8 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, scheduler.Wait(resource_ticks[image_index]); resource_ticks[image_index] = scheduler.CurrentTick(); - UpdateDescriptorSet(image_index, - use_accelerated ? screen_info.image_view : *raw_image_views[image_index]); + VkImageView source_image_view = + use_accelerated ? screen_info.image_view : *raw_image_views[image_index]; BufferData data; SetUniformData(data, layout); @@ -222,9 +230,134 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, read_barrier); cmdbuf.CopyBufferToImage(*buffer, image, VK_IMAGE_LAYOUT_GENERAL, copy); cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, - VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, write_barrier); + VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | + VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, + 0, write_barrier); }); } + + const auto anti_alias_pass = Settings::values.anti_aliasing.GetValue(); + if (use_accelerated && anti_alias_pass != Settings::AntiAliasing::None) { + UpdateAADescriptorSet(image_index, source_image_view, false); + const u32 up_scale = Settings::values.resolution_info.up_scale; + const u32 down_shift = Settings::values.resolution_info.down_shift; + VkExtent2D size{ + .width = (up_scale * framebuffer.width) >> down_shift, + .height = (up_scale * framebuffer.height) >> down_shift, + }; + scheduler.Record([this, image_index, size, anti_alias_pass](vk::CommandBuffer cmdbuf) { + const VkImageMemoryBarrier base_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = 0, + .dstAccessMask = 0, + .oldLayout = VK_IMAGE_LAYOUT_GENERAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = {}, + .subresourceRange = + { + .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .baseMipLevel = 0, + .levelCount = 1, + .baseArrayLayer = 0, + .layerCount = 1, + }, + }; + + { + VkImageMemoryBarrier fsr_write_barrier = base_barrier; + fsr_write_barrier.image = *aa_image; + fsr_write_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED; + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, fsr_write_barrier); + } + + const f32 bg_red = Settings::values.bg_red.GetValue() / 255.0f; + const f32 bg_green = Settings::values.bg_green.GetValue() / 255.0f; + const f32 bg_blue = Settings::values.bg_blue.GetValue() / 255.0f; + const VkClearValue clear_color{ + .color = {.float32 = {bg_red, bg_green, bg_blue, 1.0f}}, + }; + const VkRenderPassBeginInfo renderpass_bi{ + .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO, + .pNext = nullptr, + .renderPass = *aa_renderpass, + .framebuffer = *aa_framebuffer, + .renderArea = + { + .offset = {0, 0}, + .extent = size, + }, + .clearValueCount = 1, + .pClearValues = &clear_color, + }; + const VkViewport viewport{ + .x = 0.0f, + .y = 0.0f, + .width = static_cast<float>(size.width), + .height = static_cast<float>(size.height), + .minDepth = 0.0f, + .maxDepth = 1.0f, + }; + const VkRect2D scissor{ + .offset = {0, 0}, + .extent = size, + }; + cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE); + switch (anti_alias_pass) { + case Settings::AntiAliasing::Fxaa: + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *aa_pipeline); + break; + default: + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *aa_pipeline); + break; + } + cmdbuf.SetViewport(0, viewport); + cmdbuf.SetScissor(0, scissor); + + cmdbuf.BindVertexBuffer(0, *buffer, offsetof(BufferData, vertices)); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, *aa_pipeline_layout, 0, + aa_descriptor_sets[image_index], {}); + cmdbuf.Draw(4, 1, 0, 0); + cmdbuf.EndRenderPass(); + + { + VkImageMemoryBarrier blit_read_barrier = base_barrier; + blit_read_barrier.image = *aa_image; + blit_read_barrier.srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT; + blit_read_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT; + + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, + VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, blit_read_barrier); + } + }); + source_image_view = *aa_image_view; + } + + if (fsr) { + auto crop_rect = framebuffer.crop_rect; + if (crop_rect.GetWidth() == 0) { + crop_rect.right = framebuffer.width; + } + if (crop_rect.GetHeight() == 0) { + crop_rect.bottom = framebuffer.height; + } + crop_rect = crop_rect.Scale(Settings::values.resolution_info.up_factor); + VkExtent2D fsr_input_size{ + .width = Settings::values.resolution_info.ScaleUp(framebuffer.width), + .height = Settings::values.resolution_info.ScaleUp(framebuffer.height), + }; + VkImageView fsr_image_view = + fsr->Draw(scheduler, image_index, source_image_view, fsr_input_size, crop_rect); + UpdateDescriptorSet(image_index, fsr_image_view, true); + } else { + const bool is_nn = + Settings::values.scaling_filter.GetValue() == Settings::ScalingFilter::NearestNeighbor; + UpdateDescriptorSet(image_index, source_image_view, is_nn); + } + scheduler.Record( [this, host_framebuffer, image_index, size = render_area](vk::CommandBuffer cmdbuf) { const f32 bg_red = Settings::values.bg_red.GetValue() / 255.0f; @@ -258,8 +391,28 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, .offset = {0, 0}, .extent = size, }; + const auto filter = Settings::values.scaling_filter.GetValue(); cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE); - cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline); + switch (filter) { + case Settings::ScalingFilter::NearestNeighbor: + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *bilinear_pipeline); + break; + case Settings::ScalingFilter::Bilinear: + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *bilinear_pipeline); + break; + case Settings::ScalingFilter::Bicubic: + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *bicubic_pipeline); + break; + case Settings::ScalingFilter::Gaussian: + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *gaussian_pipeline); + break; + case Settings::ScalingFilter::ScaleForce: + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *scaleforce_pipeline); + break; + default: + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *bilinear_pipeline); + break; + } cmdbuf.SetViewport(0, viewport); cmdbuf.SetScissor(0, scissor); @@ -281,11 +434,16 @@ VkSemaphore VKBlitScreen::DrawToSwapchain(const Tegra::FramebufferConfig& frameb } vk::Framebuffer VKBlitScreen::CreateFramebuffer(const VkImageView& image_view, VkExtent2D extent) { + return CreateFramebuffer(image_view, extent, renderpass); +} + +vk::Framebuffer VKBlitScreen::CreateFramebuffer(const VkImageView& image_view, VkExtent2D extent, + vk::RenderPass& rd) { return device.GetLogical().CreateFramebuffer(VkFramebufferCreateInfo{ .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO, .pNext = nullptr, .flags = 0, - .renderPass = *renderpass, + .renderPass = *rd, .attachmentCount = 1, .pAttachments = &image_view, .width = extent.width, @@ -308,9 +466,21 @@ void VKBlitScreen::CreateDynamicResources() { CreateRenderPass(); CreateFramebuffers(); CreateGraphicsPipeline(); + fsr.reset(); + if (Settings::values.scaling_filter.GetValue() == Settings::ScalingFilter::Fsr) { + CreateFSR(); + } } void VKBlitScreen::RefreshResources(const Tegra::FramebufferConfig& framebuffer) { + if (Settings::values.scaling_filter.GetValue() == Settings::ScalingFilter::Fsr) { + if (!fsr) { + CreateFSR(); + } + } else { + fsr.reset(); + } + if (framebuffer.width == raw_width && framebuffer.height == raw_height && !raw_images.empty()) { return; } @@ -324,7 +494,16 @@ void VKBlitScreen::RefreshResources(const Tegra::FramebufferConfig& framebuffer) void VKBlitScreen::CreateShaders() { vertex_shader = BuildShader(device, VULKAN_PRESENT_VERT_SPV); - fragment_shader = BuildShader(device, VULKAN_PRESENT_FRAG_SPV); + fxaa_vertex_shader = BuildShader(device, FXAA_VERT_SPV); + fxaa_fragment_shader = BuildShader(device, FXAA_FRAG_SPV); + bilinear_fragment_shader = BuildShader(device, VULKAN_PRESENT_FRAG_SPV); + bicubic_fragment_shader = BuildShader(device, PRESENT_BICUBIC_FRAG_SPV); + gaussian_fragment_shader = BuildShader(device, PRESENT_GAUSSIAN_FRAG_SPV); + if (device.IsFloat16Supported()) { + scaleforce_fragment_shader = BuildShader(device, VULKAN_PRESENT_SCALEFORCE_FP16_FRAG_SPV); + } else { + scaleforce_fragment_shader = BuildShader(device, VULKAN_PRESENT_SCALEFORCE_FP32_FRAG_SPV); + } } void VKBlitScreen::CreateSemaphores() { @@ -344,6 +523,13 @@ void VKBlitScreen::CreateDescriptorPool() { }, }}; + const std::array<VkDescriptorPoolSize, 1> pool_sizes_aa{{ + { + .type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .descriptorCount = static_cast<u32>(image_count * 2), + }, + }}; + const VkDescriptorPoolCreateInfo ci{ .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO, .pNext = nullptr, @@ -353,19 +539,33 @@ void VKBlitScreen::CreateDescriptorPool() { .pPoolSizes = pool_sizes.data(), }; descriptor_pool = device.GetLogical().CreateDescriptorPool(ci); + + const VkDescriptorPoolCreateInfo ci_aa{ + .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO, + .pNext = nullptr, + .flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT, + .maxSets = static_cast<u32>(image_count), + .poolSizeCount = static_cast<u32>(pool_sizes_aa.size()), + .pPoolSizes = pool_sizes_aa.data(), + }; + aa_descriptor_pool = device.GetLogical().CreateDescriptorPool(ci_aa); } void VKBlitScreen::CreateRenderPass() { + renderpass = CreateRenderPassImpl(swapchain.GetImageViewFormat()); +} + +vk::RenderPass VKBlitScreen::CreateRenderPassImpl(VkFormat format, bool is_present) { const VkAttachmentDescription color_attachment{ .flags = 0, - .format = swapchain.GetImageViewFormat(), + .format = format, .samples = VK_SAMPLE_COUNT_1_BIT, .loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR, .storeOp = VK_ATTACHMENT_STORE_OP_STORE, .stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE, .stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE, .initialLayout = VK_IMAGE_LAYOUT_UNDEFINED, - .finalLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR, + .finalLayout = is_present ? VK_IMAGE_LAYOUT_PRESENT_SRC_KHR : VK_IMAGE_LAYOUT_GENERAL, }; const VkAttachmentReference color_attachment_ref{ @@ -408,7 +608,7 @@ void VKBlitScreen::CreateRenderPass() { .pDependencies = &dependency, }; - renderpass = device.GetLogical().CreateRenderPass(renderpass_ci); + return device.GetLogical().CreateRenderPass(renderpass_ci); } void VKBlitScreen::CreateDescriptorSetLayout() { @@ -429,6 +629,23 @@ void VKBlitScreen::CreateDescriptorSetLayout() { }, }}; + const std::array<VkDescriptorSetLayoutBinding, 2> layout_bindings_aa{{ + { + .binding = 0, + .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .descriptorCount = 1, + .stageFlags = VK_SHADER_STAGE_VERTEX_BIT, + .pImmutableSamplers = nullptr, + }, + { + .binding = 1, + .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .descriptorCount = 1, + .stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT, + .pImmutableSamplers = nullptr, + }, + }}; + const VkDescriptorSetLayoutCreateInfo ci{ .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO, .pNext = nullptr, @@ -437,11 +654,21 @@ void VKBlitScreen::CreateDescriptorSetLayout() { .pBindings = layout_bindings.data(), }; + const VkDescriptorSetLayoutCreateInfo ci_aa{ + .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .bindingCount = static_cast<u32>(layout_bindings_aa.size()), + .pBindings = layout_bindings_aa.data(), + }; + descriptor_set_layout = device.GetLogical().CreateDescriptorSetLayout(ci); + aa_descriptor_set_layout = device.GetLogical().CreateDescriptorSetLayout(ci_aa); } void VKBlitScreen::CreateDescriptorSets() { const std::vector layouts(image_count, *descriptor_set_layout); + const std::vector layouts_aa(image_count, *aa_descriptor_set_layout); const VkDescriptorSetAllocateInfo ai{ .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO, @@ -451,7 +678,16 @@ void VKBlitScreen::CreateDescriptorSets() { .pSetLayouts = layouts.data(), }; + const VkDescriptorSetAllocateInfo ai_aa{ + .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO, + .pNext = nullptr, + .descriptorPool = *aa_descriptor_pool, + .descriptorSetCount = static_cast<u32>(image_count), + .pSetLayouts = layouts_aa.data(), + }; + descriptor_sets = descriptor_pool.Allocate(ai); + aa_descriptor_sets = aa_descriptor_pool.Allocate(ai_aa); } void VKBlitScreen::CreatePipelineLayout() { @@ -464,11 +700,63 @@ void VKBlitScreen::CreatePipelineLayout() { .pushConstantRangeCount = 0, .pPushConstantRanges = nullptr, }; + const VkPipelineLayoutCreateInfo ci_aa{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .setLayoutCount = 1, + .pSetLayouts = aa_descriptor_set_layout.address(), + .pushConstantRangeCount = 0, + .pPushConstantRanges = nullptr, + }; pipeline_layout = device.GetLogical().CreatePipelineLayout(ci); + aa_pipeline_layout = device.GetLogical().CreatePipelineLayout(ci_aa); } void VKBlitScreen::CreateGraphicsPipeline() { - const std::array<VkPipelineShaderStageCreateInfo, 2> shader_stages{{ + const std::array<VkPipelineShaderStageCreateInfo, 2> bilinear_shader_stages{{ + { + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_VERTEX_BIT, + .module = *vertex_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }, + { + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_FRAGMENT_BIT, + .module = *bilinear_fragment_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }, + }}; + + const std::array<VkPipelineShaderStageCreateInfo, 2> bicubic_shader_stages{{ + { + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_VERTEX_BIT, + .module = *vertex_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }, + { + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_FRAGMENT_BIT, + .module = *bicubic_fragment_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }, + }}; + + const std::array<VkPipelineShaderStageCreateInfo, 2> gaussian_shader_stages{{ { .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, .pNext = nullptr, @@ -483,7 +771,28 @@ void VKBlitScreen::CreateGraphicsPipeline() { .pNext = nullptr, .flags = 0, .stage = VK_SHADER_STAGE_FRAGMENT_BIT, - .module = *fragment_shader, + .module = *gaussian_fragment_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }, + }}; + + const std::array<VkPipelineShaderStageCreateInfo, 2> scaleforce_shader_stages{{ + { + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_VERTEX_BIT, + .module = *vertex_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }, + { + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_FRAGMENT_BIT, + .module = *scaleforce_fragment_shader, .pName = "main", .pSpecializationInfo = nullptr, }, @@ -583,12 +892,12 @@ void VKBlitScreen::CreateGraphicsPipeline() { .pDynamicStates = dynamic_states.data(), }; - const VkGraphicsPipelineCreateInfo pipeline_ci{ + const VkGraphicsPipelineCreateInfo bilinear_pipeline_ci{ .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, .pNext = nullptr, .flags = 0, - .stageCount = static_cast<u32>(shader_stages.size()), - .pStages = shader_stages.data(), + .stageCount = static_cast<u32>(bilinear_shader_stages.size()), + .pStages = bilinear_shader_stages.data(), .pVertexInputState = &vertex_input_ci, .pInputAssemblyState = &input_assembly_ci, .pTessellationState = nullptr, @@ -605,7 +914,76 @@ void VKBlitScreen::CreateGraphicsPipeline() { .basePipelineIndex = 0, }; - pipeline = device.GetLogical().CreateGraphicsPipeline(pipeline_ci); + const VkGraphicsPipelineCreateInfo bicubic_pipeline_ci{ + .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stageCount = static_cast<u32>(bicubic_shader_stages.size()), + .pStages = bicubic_shader_stages.data(), + .pVertexInputState = &vertex_input_ci, + .pInputAssemblyState = &input_assembly_ci, + .pTessellationState = nullptr, + .pViewportState = &viewport_state_ci, + .pRasterizationState = &rasterization_ci, + .pMultisampleState = &multisampling_ci, + .pDepthStencilState = nullptr, + .pColorBlendState = &color_blend_ci, + .pDynamicState = &dynamic_state_ci, + .layout = *pipeline_layout, + .renderPass = *renderpass, + .subpass = 0, + .basePipelineHandle = 0, + .basePipelineIndex = 0, + }; + + const VkGraphicsPipelineCreateInfo gaussian_pipeline_ci{ + .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stageCount = static_cast<u32>(gaussian_shader_stages.size()), + .pStages = gaussian_shader_stages.data(), + .pVertexInputState = &vertex_input_ci, + .pInputAssemblyState = &input_assembly_ci, + .pTessellationState = nullptr, + .pViewportState = &viewport_state_ci, + .pRasterizationState = &rasterization_ci, + .pMultisampleState = &multisampling_ci, + .pDepthStencilState = nullptr, + .pColorBlendState = &color_blend_ci, + .pDynamicState = &dynamic_state_ci, + .layout = *pipeline_layout, + .renderPass = *renderpass, + .subpass = 0, + .basePipelineHandle = 0, + .basePipelineIndex = 0, + }; + + const VkGraphicsPipelineCreateInfo scaleforce_pipeline_ci{ + .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stageCount = static_cast<u32>(scaleforce_shader_stages.size()), + .pStages = scaleforce_shader_stages.data(), + .pVertexInputState = &vertex_input_ci, + .pInputAssemblyState = &input_assembly_ci, + .pTessellationState = nullptr, + .pViewportState = &viewport_state_ci, + .pRasterizationState = &rasterization_ci, + .pMultisampleState = &multisampling_ci, + .pDepthStencilState = nullptr, + .pColorBlendState = &color_blend_ci, + .pDynamicState = &dynamic_state_ci, + .layout = *pipeline_layout, + .renderPass = *renderpass, + .subpass = 0, + .basePipelineHandle = 0, + .basePipelineIndex = 0, + }; + + bilinear_pipeline = device.GetLogical().CreateGraphicsPipeline(bilinear_pipeline_ci); + bicubic_pipeline = device.GetLogical().CreateGraphicsPipeline(bicubic_pipeline_ci); + gaussian_pipeline = device.GetLogical().CreateGraphicsPipeline(gaussian_pipeline_ci); + scaleforce_pipeline = device.GetLogical().CreateGraphicsPipeline(scaleforce_pipeline_ci); } void VKBlitScreen::CreateSampler() { @@ -614,8 +992,29 @@ void VKBlitScreen::CreateSampler() { .pNext = nullptr, .flags = 0, .magFilter = VK_FILTER_LINEAR, + .minFilter = VK_FILTER_LINEAR, + .mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST, + .addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER, + .addressModeV = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER, + .addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER, + .mipLodBias = 0.0f, + .anisotropyEnable = VK_FALSE, + .maxAnisotropy = 0.0f, + .compareEnable = VK_FALSE, + .compareOp = VK_COMPARE_OP_NEVER, + .minLod = 0.0f, + .maxLod = 0.0f, + .borderColor = VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK, + .unnormalizedCoordinates = VK_FALSE, + }; + + const VkSamplerCreateInfo ci_nn{ + .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .magFilter = VK_FILTER_NEAREST, .minFilter = VK_FILTER_NEAREST, - .mipmapMode = VK_SAMPLER_MIPMAP_MODE_LINEAR, + .mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST, .addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER, .addressModeV = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER, .addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER, @@ -631,6 +1030,7 @@ void VKBlitScreen::CreateSampler() { }; sampler = device.GetLogical().CreateSampler(ci); + nn_sampler = device.GetLogical().CreateSampler(ci_nn); } void VKBlitScreen::CreateFramebuffers() { @@ -639,7 +1039,7 @@ void VKBlitScreen::CreateFramebuffers() { for (std::size_t i = 0; i < image_count; ++i) { const VkImageView image_view{swapchain.GetImageViewIndex(i)}; - framebuffers[i] = CreateFramebuffer(image_view, size); + framebuffers[i] = CreateFramebuffer(image_view, size, renderpass); } } @@ -649,6 +1049,11 @@ void VKBlitScreen::ReleaseRawImages() { } raw_images.clear(); raw_buffer_commits.clear(); + + aa_image_view.reset(); + aa_image.reset(); + aa_commit = MemoryCommit{}; + buffer.reset(); buffer_commit = MemoryCommit{}; } @@ -675,8 +1080,11 @@ void VKBlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) raw_image_views.resize(image_count); raw_buffer_commits.resize(image_count); - for (size_t i = 0; i < image_count; ++i) { - raw_images[i] = device.GetLogical().CreateImage(VkImageCreateInfo{ + const auto create_image = [&](bool used_on_framebuffer = false, u32 up_scale = 1, + u32 down_shift = 0) { + u32 extra_usages = used_on_framebuffer ? VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT + : VK_IMAGE_USAGE_TRANSFER_DST_BIT; + return device.GetLogical().CreateImage(VkImageCreateInfo{ .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, .pNext = nullptr, .flags = 0, @@ -684,26 +1092,30 @@ void VKBlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) .format = GetFormat(framebuffer), .extent = { - .width = framebuffer.width, - .height = framebuffer.height, + .width = (up_scale * framebuffer.width) >> down_shift, + .height = (up_scale * framebuffer.height) >> down_shift, .depth = 1, }, .mipLevels = 1, .arrayLayers = 1, .samples = VK_SAMPLE_COUNT_1_BIT, - .tiling = VK_IMAGE_TILING_LINEAR, - .usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT, + .tiling = used_on_framebuffer ? VK_IMAGE_TILING_OPTIMAL : VK_IMAGE_TILING_LINEAR, + .usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT | extra_usages, .sharingMode = VK_SHARING_MODE_EXCLUSIVE, .queueFamilyIndexCount = 0, .pQueueFamilyIndices = nullptr, .initialLayout = VK_IMAGE_LAYOUT_UNDEFINED, }); - raw_buffer_commits[i] = memory_allocator.Commit(raw_images[i], MemoryUsage::DeviceLocal); - raw_image_views[i] = device.GetLogical().CreateImageView(VkImageViewCreateInfo{ + }; + const auto create_commit = [&](vk::Image& image) { + return memory_allocator.Commit(image, MemoryUsage::DeviceLocal); + }; + const auto create_image_view = [&](vk::Image& image) { + return device.GetLogical().CreateImageView(VkImageViewCreateInfo{ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, .pNext = nullptr, .flags = 0, - .image = *raw_images[i], + .image = *image, .viewType = VK_IMAGE_VIEW_TYPE_2D, .format = GetFormat(framebuffer), .components = @@ -722,10 +1134,211 @@ void VKBlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) .layerCount = 1, }, }); + }; + + for (size_t i = 0; i < image_count; ++i) { + raw_images[i] = create_image(); + raw_buffer_commits[i] = create_commit(raw_images[i]); + raw_image_views[i] = create_image_view(raw_images[i]); } + + // AA Resources + const u32 up_scale = Settings::values.resolution_info.up_scale; + const u32 down_shift = Settings::values.resolution_info.down_shift; + aa_image = create_image(true, up_scale, down_shift); + aa_commit = create_commit(aa_image); + aa_image_view = create_image_view(aa_image); + VkExtent2D size{ + .width = (up_scale * framebuffer.width) >> down_shift, + .height = (up_scale * framebuffer.height) >> down_shift, + }; + if (aa_renderpass) { + aa_framebuffer = CreateFramebuffer(*aa_image_view, size, aa_renderpass); + return; + } + aa_renderpass = CreateRenderPassImpl(GetFormat(framebuffer), false); + aa_framebuffer = CreateFramebuffer(*aa_image_view, size, aa_renderpass); + + const std::array<VkPipelineShaderStageCreateInfo, 2> fxaa_shader_stages{{ + { + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_VERTEX_BIT, + .module = *fxaa_vertex_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }, + { + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_FRAGMENT_BIT, + .module = *fxaa_fragment_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }, + }}; + + const auto vertex_binding_description = ScreenRectVertex::GetDescription(); + const auto vertex_attrs_description = ScreenRectVertex::GetAttributes(); + + const VkPipelineVertexInputStateCreateInfo vertex_input_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .vertexBindingDescriptionCount = 1, + .pVertexBindingDescriptions = &vertex_binding_description, + .vertexAttributeDescriptionCount = u32{vertex_attrs_description.size()}, + .pVertexAttributeDescriptions = vertex_attrs_description.data(), + }; + + const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP, + .primitiveRestartEnable = VK_FALSE, + }; + + const VkPipelineViewportStateCreateInfo viewport_state_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .viewportCount = 1, + .pViewports = nullptr, + .scissorCount = 1, + .pScissors = nullptr, + }; + + const VkPipelineRasterizationStateCreateInfo rasterization_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .depthClampEnable = VK_FALSE, + .rasterizerDiscardEnable = VK_FALSE, + .polygonMode = VK_POLYGON_MODE_FILL, + .cullMode = VK_CULL_MODE_NONE, + .frontFace = VK_FRONT_FACE_CLOCKWISE, + .depthBiasEnable = VK_FALSE, + .depthBiasConstantFactor = 0.0f, + .depthBiasClamp = 0.0f, + .depthBiasSlopeFactor = 0.0f, + .lineWidth = 1.0f, + }; + + const VkPipelineMultisampleStateCreateInfo multisampling_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .rasterizationSamples = VK_SAMPLE_COUNT_1_BIT, + .sampleShadingEnable = VK_FALSE, + .minSampleShading = 0.0f, + .pSampleMask = nullptr, + .alphaToCoverageEnable = VK_FALSE, + .alphaToOneEnable = VK_FALSE, + }; + + const VkPipelineColorBlendAttachmentState color_blend_attachment{ + .blendEnable = VK_FALSE, + .srcColorBlendFactor = VK_BLEND_FACTOR_ZERO, + .dstColorBlendFactor = VK_BLEND_FACTOR_ZERO, + .colorBlendOp = VK_BLEND_OP_ADD, + .srcAlphaBlendFactor = VK_BLEND_FACTOR_ZERO, + .dstAlphaBlendFactor = VK_BLEND_FACTOR_ZERO, + .alphaBlendOp = VK_BLEND_OP_ADD, + .colorWriteMask = VK_COLOR_COMPONENT_R_BIT | VK_COLOR_COMPONENT_G_BIT | + VK_COLOR_COMPONENT_B_BIT | VK_COLOR_COMPONENT_A_BIT, + }; + + const VkPipelineColorBlendStateCreateInfo color_blend_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .logicOpEnable = VK_FALSE, + .logicOp = VK_LOGIC_OP_COPY, + .attachmentCount = 1, + .pAttachments = &color_blend_attachment, + .blendConstants = {0.0f, 0.0f, 0.0f, 0.0f}, + }; + + static constexpr std::array dynamic_states{ + VK_DYNAMIC_STATE_VIEWPORT, + VK_DYNAMIC_STATE_SCISSOR, + }; + const VkPipelineDynamicStateCreateInfo dynamic_state_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .dynamicStateCount = static_cast<u32>(dynamic_states.size()), + .pDynamicStates = dynamic_states.data(), + }; + + const VkGraphicsPipelineCreateInfo fxaa_pipeline_ci{ + .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stageCount = static_cast<u32>(fxaa_shader_stages.size()), + .pStages = fxaa_shader_stages.data(), + .pVertexInputState = &vertex_input_ci, + .pInputAssemblyState = &input_assembly_ci, + .pTessellationState = nullptr, + .pViewportState = &viewport_state_ci, + .pRasterizationState = &rasterization_ci, + .pMultisampleState = &multisampling_ci, + .pDepthStencilState = nullptr, + .pColorBlendState = &color_blend_ci, + .pDynamicState = &dynamic_state_ci, + .layout = *aa_pipeline_layout, + .renderPass = *aa_renderpass, + .subpass = 0, + .basePipelineHandle = 0, + .basePipelineIndex = 0, + }; + + // AA + aa_pipeline = device.GetLogical().CreateGraphicsPipeline(fxaa_pipeline_ci); } -void VKBlitScreen::UpdateDescriptorSet(std::size_t image_index, VkImageView image_view) const { +void VKBlitScreen::UpdateAADescriptorSet(std::size_t image_index, VkImageView image_view, + bool nn) const { + const VkDescriptorImageInfo image_info{ + .sampler = nn ? *nn_sampler : *sampler, + .imageView = image_view, + .imageLayout = VK_IMAGE_LAYOUT_GENERAL, + }; + + const VkWriteDescriptorSet sampler_write{ + .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET, + .pNext = nullptr, + .dstSet = aa_descriptor_sets[image_index], + .dstBinding = 0, + .dstArrayElement = 0, + .descriptorCount = 1, + .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .pImageInfo = &image_info, + .pBufferInfo = nullptr, + .pTexelBufferView = nullptr, + }; + + const VkWriteDescriptorSet sampler_write_2{ + .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET, + .pNext = nullptr, + .dstSet = aa_descriptor_sets[image_index], + .dstBinding = 1, + .dstArrayElement = 0, + .descriptorCount = 1, + .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .pImageInfo = &image_info, + .pBufferInfo = nullptr, + .pTexelBufferView = nullptr, + }; + + device.GetLogical().UpdateDescriptorSets(std::array{sampler_write, sampler_write_2}, {}); +} + +void VKBlitScreen::UpdateDescriptorSet(std::size_t image_index, VkImageView image_view, + bool nn) const { const VkDescriptorBufferInfo buffer_info{ .buffer = *buffer, .offset = offsetof(BufferData, uniform), @@ -746,7 +1359,7 @@ void VKBlitScreen::UpdateDescriptorSet(std::size_t image_index, VkImageView imag }; const VkDescriptorImageInfo image_info{ - .sampler = *sampler, + .sampler = nn ? *nn_sampler : *sampler, .imageView = image_view, .imageLayout = VK_IMAGE_LAYOUT_GENERAL, }; @@ -798,17 +1411,19 @@ void VKBlitScreen::SetVertexData(BufferData& data, const Tegra::FramebufferConfi UNIMPLEMENTED_IF(framebuffer_crop_rect.top != 0); UNIMPLEMENTED_IF(framebuffer_crop_rect.left != 0); - // Scale the output by the crop width/height. This is commonly used with 1280x720 rendering - // (e.g. handheld mode) on a 1920x1080 framebuffer. f32 scale_u = 1.0f; f32 scale_v = 1.0f; - if (framebuffer_crop_rect.GetWidth() > 0) { - scale_u = static_cast<f32>(framebuffer_crop_rect.GetWidth()) / - static_cast<f32>(screen_info.width); - } - if (framebuffer_crop_rect.GetHeight() > 0) { - scale_v = static_cast<f32>(framebuffer_crop_rect.GetHeight()) / - static_cast<f32>(screen_info.height); + // Scale the output by the crop width/height. This is commonly used with 1280x720 rendering + // (e.g. handheld mode) on a 1920x1080 framebuffer. + if (!fsr) { + if (framebuffer_crop_rect.GetWidth() > 0) { + scale_u = static_cast<f32>(framebuffer_crop_rect.GetWidth()) / + static_cast<f32>(screen_info.width); + } + if (framebuffer_crop_rect.GetHeight() > 0) { + scale_v = static_cast<f32>(framebuffer_crop_rect.GetHeight()) / + static_cast<f32>(screen_info.height); + } } const auto& screen = layout.screen; @@ -822,6 +1437,15 @@ void VKBlitScreen::SetVertexData(BufferData& data, const Tegra::FramebufferConfi data.vertices[3] = ScreenRectVertex(x + w, y + h, texcoords.bottom * scale_u, right * scale_v); } +void VKBlitScreen::CreateFSR() { + const auto& layout = render_window.GetFramebufferLayout(); + const VkExtent2D fsr_size{ + .width = layout.screen.GetWidth(), + .height = layout.screen.GetHeight(), + }; + fsr = std::make_unique<FSR>(device, memory_allocator, image_count, fsr_size); +} + u64 VKBlitScreen::CalculateBufferSize(const Tegra::FramebufferConfig& framebuffer) const { return sizeof(BufferData) + GetSizeInBytes(framebuffer) * image_count; } diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.h b/src/video_core/renderer_vulkan/vk_blit_screen.h index 430bcfbca..bbca71af3 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.h +++ b/src/video_core/renderer_vulkan/vk_blit_screen.h @@ -34,6 +34,7 @@ namespace Vulkan { struct ScreenInfo; class Device; +class FSR; class RasterizerVulkan; class VKScheduler; class VKSwapchain; @@ -66,6 +67,9 @@ public: [[nodiscard]] vk::Framebuffer CreateFramebuffer(const VkImageView& image_view, VkExtent2D extent); + [[nodiscard]] vk::Framebuffer CreateFramebuffer(const VkImageView& image_view, + VkExtent2D extent, vk::RenderPass& rd); + private: struct BufferData; @@ -74,6 +78,7 @@ private: void CreateSemaphores(); void CreateDescriptorPool(); void CreateRenderPass(); + vk::RenderPass CreateRenderPassImpl(VkFormat, bool is_present = true); void CreateDescriptorSetLayout(); void CreateDescriptorSets(); void CreatePipelineLayout(); @@ -88,11 +93,14 @@ private: void CreateStagingBuffer(const Tegra::FramebufferConfig& framebuffer); void CreateRawImages(const Tegra::FramebufferConfig& framebuffer); - void UpdateDescriptorSet(std::size_t image_index, VkImageView image_view) const; + void UpdateDescriptorSet(std::size_t image_index, VkImageView image_view, bool nn) const; + void UpdateAADescriptorSet(std::size_t image_index, VkImageView image_view, bool nn) const; void SetUniformData(BufferData& data, const Layout::FramebufferLayout layout) const; void SetVertexData(BufferData& data, const Tegra::FramebufferConfig& framebuffer, const Layout::FramebufferLayout layout) const; + void CreateFSR(); + u64 CalculateBufferSize(const Tegra::FramebufferConfig& framebuffer) const; u64 GetRawImageOffset(const Tegra::FramebufferConfig& framebuffer, std::size_t image_index) const; @@ -107,14 +115,24 @@ private: const VKScreenInfo& screen_info; vk::ShaderModule vertex_shader; - vk::ShaderModule fragment_shader; + vk::ShaderModule fxaa_vertex_shader; + vk::ShaderModule fxaa_fragment_shader; + vk::ShaderModule bilinear_fragment_shader; + vk::ShaderModule bicubic_fragment_shader; + vk::ShaderModule gaussian_fragment_shader; + vk::ShaderModule scaleforce_fragment_shader; vk::DescriptorPool descriptor_pool; vk::DescriptorSetLayout descriptor_set_layout; vk::PipelineLayout pipeline_layout; - vk::Pipeline pipeline; + vk::Pipeline nearest_neightbor_pipeline; + vk::Pipeline bilinear_pipeline; + vk::Pipeline bicubic_pipeline; + vk::Pipeline gaussian_pipeline; + vk::Pipeline scaleforce_pipeline; vk::RenderPass renderpass; std::vector<vk::Framebuffer> framebuffers; vk::DescriptorSets descriptor_sets; + vk::Sampler nn_sampler; vk::Sampler sampler; vk::Buffer buffer; @@ -126,8 +144,22 @@ private: std::vector<vk::Image> raw_images; std::vector<vk::ImageView> raw_image_views; std::vector<MemoryCommit> raw_buffer_commits; + + vk::DescriptorPool aa_descriptor_pool; + vk::DescriptorSetLayout aa_descriptor_set_layout; + vk::PipelineLayout aa_pipeline_layout; + vk::Pipeline aa_pipeline; + vk::RenderPass aa_renderpass; + vk::Framebuffer aa_framebuffer; + vk::DescriptorSets aa_descriptor_sets; + vk::Image aa_image; + vk::ImageView aa_image_view; + MemoryCommit aa_commit; + u32 raw_width = 0; u32 raw_height = 0; + + std::unique_ptr<FSR> fsr; }; } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index 8ac58bc2f..5ffd93499 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -146,7 +146,7 @@ void BufferCacheRuntime::Finish() { } void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer, - std::span<const VideoCommon::BufferCopy> copies) { + std::span<const VideoCommon::BufferCopy> copies, bool barrier) { static constexpr VkMemoryBarrier READ_BARRIER{ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, .pNext = nullptr, @@ -163,10 +163,42 @@ void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer, boost::container::small_vector<VkBufferCopy, 3> vk_copies(copies.size()); std::ranges::transform(copies, vk_copies.begin(), MakeBufferCopy); scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([src_buffer, dst_buffer, vk_copies](vk::CommandBuffer cmdbuf) { + scheduler.Record([src_buffer, dst_buffer, vk_copies, barrier](vk::CommandBuffer cmdbuf) { + if (barrier) { + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + VK_PIPELINE_STAGE_TRANSFER_BIT, 0, READ_BARRIER); + } + cmdbuf.CopyBuffer(src_buffer, dst_buffer, vk_copies); + if (barrier) { + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, + VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, WRITE_BARRIER); + } + }); +} + +void BufferCacheRuntime::PreCopyBarrier() { + static constexpr VkMemoryBarrier READ_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, + .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + }; + scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.Record([](vk::CommandBuffer cmdbuf) { cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, READ_BARRIER); - cmdbuf.CopyBuffer(src_buffer, dst_buffer, vk_copies); + }); +} + +void BufferCacheRuntime::PostCopyBarrier() { + static constexpr VkMemoryBarrier WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + }; + scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.Record([](vk::CommandBuffer cmdbuf) { cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, WRITE_BARRIER); }); diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h index c27402ff0..1ee0d8420 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.h +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h @@ -69,8 +69,12 @@ public: [[nodiscard]] StagingBufferRef DownloadStagingBuffer(size_t size); + void PreCopyBarrier(); + void CopyBuffer(VkBuffer src_buffer, VkBuffer dst_buffer, - std::span<const VideoCommon::BufferCopy> copies); + std::span<const VideoCommon::BufferCopy> copies, bool barrier = true); + + void PostCopyBarrier(); void ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t size, u32 value); diff --git a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp index 44faf626a..de36bcdb7 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp @@ -22,6 +22,7 @@ namespace Vulkan { using Shader::ImageBufferDescriptor; +using Shader::Backend::SPIRV::RESCALING_LAYOUT_WORDS_OFFSET; using Tegra::Texture::TexturePair; ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descriptor_pool, @@ -108,8 +109,7 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, texture_cache.SynchronizeComputeDescriptors(); static constexpr size_t max_elements = 64; - std::array<ImageId, max_elements> image_view_ids; - boost::container::static_vector<u32, max_elements> image_view_indices; + boost::container::static_vector<VideoCommon::ImageViewInOut, max_elements> views; boost::container::static_vector<VkSampler, max_elements> samplers; const auto& qmd{kepler_compute.launch_description}; @@ -134,30 +134,37 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, } return TexturePair(gpu_memory.Read<u32>(addr), via_header_index); }}; - const auto add_image{[&](const auto& desc) { + const auto add_image{[&](const auto& desc, bool blacklist) { for (u32 index = 0; index < desc.count; ++index) { const auto handle{read_handle(desc, index)}; - image_view_indices.push_back(handle.first); + views.push_back({ + .index = handle.first, + .blacklist = blacklist, + .id = {}, + }); } }}; - std::ranges::for_each(info.texture_buffer_descriptors, add_image); - std::ranges::for_each(info.image_buffer_descriptors, add_image); + for (const auto& desc : info.texture_buffer_descriptors) { + add_image(desc, false); + } + for (const auto& desc : info.image_buffer_descriptors) { + add_image(desc, false); + } for (const auto& desc : info.texture_descriptors) { for (u32 index = 0; index < desc.count; ++index) { const auto handle{read_handle(desc, index)}; - image_view_indices.push_back(handle.first); + views.push_back({handle.first}); Sampler* const sampler = texture_cache.GetComputeSampler(handle.second); samplers.push_back(sampler->Handle()); } } - std::ranges::for_each(info.image_descriptors, add_image); - - const std::span indices_span(image_view_indices.data(), image_view_indices.size()); - texture_cache.FillComputeImageViews(indices_span, image_view_ids); + for (const auto& desc : info.image_descriptors) { + add_image(desc, desc.is_written); + } + texture_cache.FillComputeImageViews(std::span(views.data(), views.size())); buffer_cache.UnbindComputeTextureBuffers(); - ImageId* texture_buffer_ids{image_view_ids.data()}; size_t index{}; const auto add_buffer{[&](const auto& desc) { constexpr bool is_image = std::is_same_v<decltype(desc), const ImageBufferDescriptor&>; @@ -166,11 +173,10 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, if constexpr (is_image) { is_written = desc.is_written; } - ImageView& image_view = texture_cache.GetImageView(*texture_buffer_ids); + ImageView& image_view = texture_cache.GetImageView(views[index].id); buffer_cache.BindComputeTextureBuffer(index, image_view.GpuAddr(), image_view.BufferSize(), image_view.format, is_written, is_image); - ++texture_buffer_ids; ++index; } }}; @@ -180,9 +186,11 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, buffer_cache.UpdateComputeBuffers(); buffer_cache.BindHostComputeBuffers(); + RescalingPushConstant rescaling; const VkSampler* samplers_it{samplers.data()}; - const ImageId* views_it{image_view_ids.data()}; - PushImageDescriptors(info, samplers_it, views_it, texture_cache, update_descriptor_queue); + const VideoCommon::ImageViewInOut* views_it{views.data()}; + PushImageDescriptors(texture_cache, update_descriptor_queue, info, rescaling, samplers_it, + views_it); if (!is_built.load(std::memory_order::relaxed)) { // Wait for the pipeline to be built @@ -192,11 +200,18 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, }); } const void* const descriptor_data{update_descriptor_queue.UpdateData()}; - scheduler.Record([this, descriptor_data](vk::CommandBuffer cmdbuf) { + const bool is_rescaling = !info.texture_descriptors.empty() || !info.image_descriptors.empty(); + scheduler.Record([this, descriptor_data, is_rescaling, + rescaling_data = rescaling.Data()](vk::CommandBuffer cmdbuf) { cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline); if (!descriptor_set_layout) { return; } + if (is_rescaling) { + cmdbuf.PushConstants(*pipeline_layout, VK_SHADER_STAGE_COMPUTE_BIT, + RESCALING_LAYOUT_WORDS_OFFSET, sizeof(rescaling_data), + rescaling_data.data()); + } const VkDescriptorSet descriptor_set{descriptor_allocator.Commit()}; const vk::Device& dev{device.GetLogical()}; dev.UpdateDescriptorSet(descriptor_set, *descriptor_update_template, descriptor_data); diff --git a/src/video_core/renderer_vulkan/vk_fsr.cpp b/src/video_core/renderer_vulkan/vk_fsr.cpp new file mode 100644 index 000000000..73629d229 --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_fsr.cpp @@ -0,0 +1,553 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <cmath> +#include "common/bit_cast.h" +#include "common/common_types.h" +#include "common/div_ceil.h" + +#include "video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp16_comp_spv.h" +#include "video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp32_comp_spv.h" +#include "video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp16_comp_spv.h" +#include "video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp32_comp_spv.h" +#include "video_core/renderer_vulkan/vk_fsr.h" +#include "video_core/renderer_vulkan/vk_scheduler.h" +#include "video_core/renderer_vulkan/vk_shader_util.h" +#include "video_core/vulkan_common/vulkan_device.h" + +namespace Vulkan { +namespace { +// Reimplementations of the constant generating functions in ffx_fsr1.h +// GCC generated a lot of warnings when using the official header. +u32 AU1_AH1_AF1(f32 f) { + static constexpr u32 base[512]{ + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, + 0x0080, 0x0100, 0x0200, 0x0400, 0x0800, 0x0c00, 0x1000, 0x1400, 0x1800, 0x1c00, 0x2000, + 0x2400, 0x2800, 0x2c00, 0x3000, 0x3400, 0x3800, 0x3c00, 0x4000, 0x4400, 0x4800, 0x4c00, + 0x5000, 0x5400, 0x5800, 0x5c00, 0x6000, 0x6400, 0x6800, 0x6c00, 0x7000, 0x7400, 0x7800, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8001, 0x8002, 0x8004, 0x8008, + 0x8010, 0x8020, 0x8040, 0x8080, 0x8100, 0x8200, 0x8400, 0x8800, 0x8c00, 0x9000, 0x9400, + 0x9800, 0x9c00, 0xa000, 0xa400, 0xa800, 0xac00, 0xb000, 0xb400, 0xb800, 0xbc00, 0xc000, + 0xc400, 0xc800, 0xcc00, 0xd000, 0xd400, 0xd800, 0xdc00, 0xe000, 0xe400, 0xe800, 0xec00, + 0xf000, 0xf400, 0xf800, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + }; + static constexpr s8 shift[512]{ + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x17, 0x16, + 0x15, 0x14, 0x13, 0x12, 0x11, 0x10, 0x0f, 0x0e, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, + 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, + 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x17, + 0x16, 0x15, 0x14, 0x13, 0x12, 0x11, 0x10, 0x0f, 0x0e, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, + 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, + 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, + }; + const u32 u = Common::BitCast<u32>(f); + const u32 i = u >> 23; + return base[i] + ((u & 0x7fffff) >> shift[i]); +} + +u32 AU1_AH2_AF2(f32 a[2]) { + return AU1_AH1_AF1(a[0]) + (AU1_AH1_AF1(a[1]) << 16); +} + +void FsrEasuCon(u32 con0[4], u32 con1[4], u32 con2[4], u32 con3[4], f32 inputViewportInPixelsX, + f32 inputViewportInPixelsY, f32 inputSizeInPixelsX, f32 inputSizeInPixelsY, + f32 outputSizeInPixelsX, f32 outputSizeInPixelsY) { + con0[0] = Common::BitCast<u32>(inputViewportInPixelsX / outputSizeInPixelsX); + con0[1] = Common::BitCast<u32>(inputViewportInPixelsY / outputSizeInPixelsY); + con0[2] = Common::BitCast<u32>(0.5f * inputViewportInPixelsX / outputSizeInPixelsX - 0.5f); + con0[3] = Common::BitCast<u32>(0.5f * inputViewportInPixelsY / outputSizeInPixelsY - 0.5f); + con1[0] = Common::BitCast<u32>(1.0f / inputSizeInPixelsX); + con1[1] = Common::BitCast<u32>(1.0f / inputSizeInPixelsY); + con1[2] = Common::BitCast<u32>(1.0f / inputSizeInPixelsX); + con1[3] = Common::BitCast<u32>(-1.0f / inputSizeInPixelsY); + con2[0] = Common::BitCast<u32>(-1.0f / inputSizeInPixelsX); + con2[1] = Common::BitCast<u32>(2.0f / inputSizeInPixelsY); + con2[2] = Common::BitCast<u32>(1.0f / inputSizeInPixelsX); + con2[3] = Common::BitCast<u32>(2.0f / inputSizeInPixelsY); + con3[0] = Common::BitCast<u32>(0.0f / inputSizeInPixelsX); + con3[1] = Common::BitCast<u32>(4.0f / inputSizeInPixelsY); + con3[2] = con3[3] = 0; +} + +void FsrEasuConOffset(u32 con0[4], u32 con1[4], u32 con2[4], u32 con3[4], + f32 inputViewportInPixelsX, f32 inputViewportInPixelsY, + f32 inputSizeInPixelsX, f32 inputSizeInPixelsY, f32 outputSizeInPixelsX, + f32 outputSizeInPixelsY, f32 inputOffsetInPixelsX, f32 inputOffsetInPixelsY) { + FsrEasuCon(con0, con1, con2, con3, inputViewportInPixelsX, inputViewportInPixelsY, + inputSizeInPixelsX, inputSizeInPixelsY, outputSizeInPixelsX, outputSizeInPixelsY); + con0[2] = Common::BitCast<u32>(0.5f * inputViewportInPixelsX / outputSizeInPixelsX - 0.5f + + inputOffsetInPixelsX); + con0[3] = Common::BitCast<u32>(0.5f * inputViewportInPixelsY / outputSizeInPixelsY - 0.5f + + inputOffsetInPixelsY); +} + +void FsrRcasCon(u32* con, f32 sharpness) { + sharpness = std::exp2f(-sharpness); + f32 hSharp[2]{sharpness, sharpness}; + con[0] = Common::BitCast<u32>(sharpness); + con[1] = AU1_AH2_AF2(hSharp); + con[2] = 0; + con[3] = 0; +} +} // Anonymous namespace + +FSR::FSR(const Device& device_, MemoryAllocator& memory_allocator_, size_t image_count_, + VkExtent2D output_size_) + : device{device_}, memory_allocator{memory_allocator_}, image_count{image_count_}, + output_size{output_size_} { + + CreateImages(); + CreateSampler(); + CreateShaders(); + CreateDescriptorPool(); + CreateDescriptorSetLayout(); + CreateDescriptorSets(); + CreatePipelineLayout(); + CreatePipeline(); +} + +VkImageView FSR::Draw(VKScheduler& scheduler, size_t image_index, VkImageView image_view, + VkExtent2D input_image_extent, const Common::Rectangle<int>& crop_rect) { + + UpdateDescriptorSet(image_index, image_view); + + scheduler.Record([this, image_index, input_image_extent, crop_rect](vk::CommandBuffer cmdbuf) { + const VkImageMemoryBarrier base_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = 0, + .dstAccessMask = 0, + .oldLayout = VK_IMAGE_LAYOUT_GENERAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = {}, + .subresourceRange = + { + .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .baseMipLevel = 0, + .levelCount = 1, + .baseArrayLayer = 0, + .layerCount = 1, + }, + }; + + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *easu_pipeline); + + std::array<u32, 4 * 4> push_constants; + FsrEasuConOffset( + push_constants.data() + 0, push_constants.data() + 4, push_constants.data() + 8, + push_constants.data() + 12, + + static_cast<f32>(crop_rect.GetWidth()), static_cast<f32>(crop_rect.GetHeight()), + static_cast<f32>(input_image_extent.width), static_cast<f32>(input_image_extent.height), + static_cast<f32>(output_size.width), static_cast<f32>(output_size.height), + static_cast<f32>(crop_rect.left), static_cast<f32>(crop_rect.top)); + cmdbuf.PushConstants(*pipeline_layout, VK_SHADER_STAGE_COMPUTE_BIT, push_constants); + + { + VkImageMemoryBarrier fsr_write_barrier = base_barrier; + fsr_write_barrier.image = *images[image_index], + fsr_write_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED; + + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, fsr_write_barrier); + } + + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline_layout, 0, + descriptor_sets[image_index * 2], {}); + cmdbuf.Dispatch(Common::DivCeil(output_size.width, 16u), + Common::DivCeil(output_size.height, 16u), 1); + + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *rcas_pipeline); + + FsrRcasCon(push_constants.data(), 0.25f); + cmdbuf.PushConstants(*pipeline_layout, VK_SHADER_STAGE_COMPUTE_BIT, push_constants); + + { + std::array<VkImageMemoryBarrier, 2> barriers; + auto& fsr_read_barrier = barriers[0]; + auto& blit_write_barrier = barriers[1]; + + fsr_read_barrier = base_barrier; + fsr_read_barrier.image = *images[image_index]; + fsr_read_barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT; + fsr_read_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT; + + blit_write_barrier = base_barrier; + blit_write_barrier.image = *images[image_count + image_index]; + blit_write_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED; + blit_write_barrier.newLayout = VK_IMAGE_LAYOUT_GENERAL; + + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, + VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, {}, {}, barriers); + } + + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline_layout, 0, + descriptor_sets[image_index * 2 + 1], {}); + cmdbuf.Dispatch(Common::DivCeil(output_size.width, 16u), + Common::DivCeil(output_size.height, 16u), 1); + + { + std::array<VkImageMemoryBarrier, 1> barriers; + auto& blit_read_barrier = barriers[0]; + + blit_read_barrier = base_barrier; + blit_read_barrier.image = *images[image_count + image_index]; + blit_read_barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT; + blit_read_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT; + + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, + VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, {}, {}, barriers); + } + }); + + return *image_views[image_count + image_index]; +} + +void FSR::CreateDescriptorPool() { + const std::array<VkDescriptorPoolSize, 2> pool_sizes{{ + { + .type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .descriptorCount = static_cast<u32>(image_count * 2), + }, + { + .type = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, + .descriptorCount = static_cast<u32>(image_count * 2), + }, + }}; + + const VkDescriptorPoolCreateInfo ci{ + .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO, + .pNext = nullptr, + .flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT, + .maxSets = static_cast<u32>(image_count * 2), + .poolSizeCount = static_cast<u32>(pool_sizes.size()), + .pPoolSizes = pool_sizes.data(), + }; + descriptor_pool = device.GetLogical().CreateDescriptorPool(ci); +} + +void FSR::CreateDescriptorSetLayout() { + const std::array<VkDescriptorSetLayoutBinding, 2> layout_bindings{{ + { + .binding = 0, + .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .descriptorCount = 1, + .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, + .pImmutableSamplers = sampler.address(), + }, + { + .binding = 1, + .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, + .descriptorCount = 1, + .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, + .pImmutableSamplers = sampler.address(), + }, + }}; + + const VkDescriptorSetLayoutCreateInfo ci{ + .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .bindingCount = static_cast<u32>(layout_bindings.size()), + .pBindings = layout_bindings.data(), + }; + + descriptor_set_layout = device.GetLogical().CreateDescriptorSetLayout(ci); +} + +void FSR::CreateDescriptorSets() { + const u32 sets = static_cast<u32>(image_count * 2); + const std::vector layouts(sets, *descriptor_set_layout); + + const VkDescriptorSetAllocateInfo ai{ + .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO, + .pNext = nullptr, + .descriptorPool = *descriptor_pool, + .descriptorSetCount = sets, + .pSetLayouts = layouts.data(), + }; + + descriptor_sets = descriptor_pool.Allocate(ai); +} + +void FSR::CreateImages() { + images.resize(image_count * 2); + image_views.resize(image_count * 2); + buffer_commits.resize(image_count * 2); + + for (size_t i = 0; i < image_count * 2; ++i) { + images[i] = device.GetLogical().CreateImage(VkImageCreateInfo{ + .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .imageType = VK_IMAGE_TYPE_2D, + .format = VK_FORMAT_R16G16B16A16_SFLOAT, + .extent = + { + .width = output_size.width, + .height = output_size.height, + .depth = 1, + }, + .mipLevels = 1, + .arrayLayers = 1, + .samples = VK_SAMPLE_COUNT_1_BIT, + .tiling = VK_IMAGE_TILING_OPTIMAL, + .usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_STORAGE_BIT | + VK_IMAGE_USAGE_SAMPLED_BIT, + .sharingMode = VK_SHARING_MODE_EXCLUSIVE, + .queueFamilyIndexCount = 0, + .pQueueFamilyIndices = nullptr, + .initialLayout = VK_IMAGE_LAYOUT_UNDEFINED, + }); + buffer_commits[i] = memory_allocator.Commit(images[i], MemoryUsage::DeviceLocal); + image_views[i] = device.GetLogical().CreateImageView(VkImageViewCreateInfo{ + .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .image = *images[i], + .viewType = VK_IMAGE_VIEW_TYPE_2D, + .format = VK_FORMAT_R16G16B16A16_SFLOAT, + .components = + { + .r = VK_COMPONENT_SWIZZLE_IDENTITY, + .g = VK_COMPONENT_SWIZZLE_IDENTITY, + .b = VK_COMPONENT_SWIZZLE_IDENTITY, + .a = VK_COMPONENT_SWIZZLE_IDENTITY, + }, + .subresourceRange = + { + .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .baseMipLevel = 0, + .levelCount = 1, + .baseArrayLayer = 0, + .layerCount = 1, + }, + }); + } +} + +void FSR::CreatePipelineLayout() { + VkPushConstantRange push_const{ + .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, + .offset = 0, + .size = sizeof(std::array<u32, 4 * 4>), + }; + VkPipelineLayoutCreateInfo ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .setLayoutCount = 1, + .pSetLayouts = descriptor_set_layout.address(), + .pushConstantRangeCount = 1, + .pPushConstantRanges = &push_const, + }; + + pipeline_layout = device.GetLogical().CreatePipelineLayout(ci); +} + +void FSR::UpdateDescriptorSet(std::size_t image_index, VkImageView image_view) const { + const auto fsr_image_view = *image_views[image_index]; + const auto blit_image_view = *image_views[image_count + image_index]; + + const VkDescriptorImageInfo image_info{ + .sampler = VK_NULL_HANDLE, + .imageView = image_view, + .imageLayout = VK_IMAGE_LAYOUT_GENERAL, + }; + const VkDescriptorImageInfo fsr_image_info{ + .sampler = VK_NULL_HANDLE, + .imageView = fsr_image_view, + .imageLayout = VK_IMAGE_LAYOUT_GENERAL, + }; + const VkDescriptorImageInfo blit_image_info{ + .sampler = VK_NULL_HANDLE, + .imageView = blit_image_view, + .imageLayout = VK_IMAGE_LAYOUT_GENERAL, + }; + + VkWriteDescriptorSet sampler_write{ + .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET, + .pNext = nullptr, + .dstSet = descriptor_sets[image_index * 2], + .dstBinding = 0, + .dstArrayElement = 0, + .descriptorCount = 1, + .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .pImageInfo = &image_info, + .pBufferInfo = nullptr, + .pTexelBufferView = nullptr, + }; + + VkWriteDescriptorSet output_write{ + .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET, + .pNext = nullptr, + .dstSet = descriptor_sets[image_index * 2], + .dstBinding = 1, + .dstArrayElement = 0, + .descriptorCount = 1, + .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, + .pImageInfo = &fsr_image_info, + .pBufferInfo = nullptr, + .pTexelBufferView = nullptr, + }; + + device.GetLogical().UpdateDescriptorSets(std::array{sampler_write, output_write}, {}); + + sampler_write.dstSet = descriptor_sets[image_index * 2 + 1]; + sampler_write.pImageInfo = &fsr_image_info; + output_write.dstSet = descriptor_sets[image_index * 2 + 1]; + output_write.pImageInfo = &blit_image_info; + + device.GetLogical().UpdateDescriptorSets(std::array{sampler_write, output_write}, {}); +} + +void FSR::CreateSampler() { + const VkSamplerCreateInfo ci{ + .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .magFilter = VK_FILTER_LINEAR, + .minFilter = VK_FILTER_LINEAR, + .mipmapMode = VK_SAMPLER_MIPMAP_MODE_LINEAR, + .addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE, + .addressModeV = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE, + .addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE, + .mipLodBias = 0.0f, + .anisotropyEnable = VK_FALSE, + .maxAnisotropy = 0.0f, + .compareEnable = VK_FALSE, + .compareOp = VK_COMPARE_OP_NEVER, + .minLod = 0.0f, + .maxLod = 0.0f, + .borderColor = VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK, + .unnormalizedCoordinates = VK_FALSE, + }; + + sampler = device.GetLogical().CreateSampler(ci); +} + +void FSR::CreateShaders() { + if (device.IsFloat16Supported()) { + easu_shader = BuildShader(device, VULKAN_FIDELITYFX_FSR_EASU_FP16_COMP_SPV); + rcas_shader = BuildShader(device, VULKAN_FIDELITYFX_FSR_RCAS_FP16_COMP_SPV); + } else { + easu_shader = BuildShader(device, VULKAN_FIDELITYFX_FSR_EASU_FP32_COMP_SPV); + rcas_shader = BuildShader(device, VULKAN_FIDELITYFX_FSR_RCAS_FP32_COMP_SPV); + } +} + +void FSR::CreatePipeline() { + VkPipelineShaderStageCreateInfo shader_stage_easu{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_COMPUTE_BIT, + .module = *easu_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }; + + VkPipelineShaderStageCreateInfo shader_stage_rcas{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_COMPUTE_BIT, + .module = *rcas_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }; + + VkComputePipelineCreateInfo pipeline_ci_easu{ + .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = shader_stage_easu, + .layout = *pipeline_layout, + .basePipelineHandle = VK_NULL_HANDLE, + .basePipelineIndex = 0, + }; + + VkComputePipelineCreateInfo pipeline_ci_rcas{ + .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = shader_stage_rcas, + .layout = *pipeline_layout, + .basePipelineHandle = VK_NULL_HANDLE, + .basePipelineIndex = 0, + }; + + easu_pipeline = device.GetLogical().CreateComputePipeline(pipeline_ci_easu); + rcas_pipeline = device.GetLogical().CreateComputePipeline(pipeline_ci_rcas); +} + +} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_fsr.h b/src/video_core/renderer_vulkan/vk_fsr.h new file mode 100644 index 000000000..6bbec3d36 --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_fsr.h @@ -0,0 +1,54 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include "common/math_util.h" +#include "video_core/vulkan_common/vulkan_memory_allocator.h" +#include "video_core/vulkan_common/vulkan_wrapper.h" + +namespace Vulkan { + +class Device; +class VKScheduler; + +class FSR { +public: + explicit FSR(const Device& device, MemoryAllocator& memory_allocator, size_t image_count, + VkExtent2D output_size); + VkImageView Draw(VKScheduler& scheduler, size_t image_index, VkImageView image_view, + VkExtent2D input_image_extent, const Common::Rectangle<int>& crop_rect); + +private: + void CreateDescriptorPool(); + void CreateDescriptorSetLayout(); + void CreateDescriptorSets(); + void CreateImages(); + void CreateSampler(); + void CreateShaders(); + void CreatePipeline(); + void CreatePipelineLayout(); + + void UpdateDescriptorSet(std::size_t image_index, VkImageView image_view) const; + + const Device& device; + MemoryAllocator& memory_allocator; + size_t image_count; + VkExtent2D output_size; + + vk::DescriptorPool descriptor_pool; + vk::DescriptorSetLayout descriptor_set_layout; + vk::DescriptorSets descriptor_sets; + vk::PipelineLayout pipeline_layout; + vk::ShaderModule easu_shader; + vk::ShaderModule rcas_shader; + vk::Pipeline easu_pipeline; + vk::Pipeline rcas_pipeline; + vk::Sampler sampler; + std::vector<vk::Image> images; + std::vector<vk::ImageView> image_views; + std::vector<MemoryCommit> buffer_commits; +}; + +} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index 8634c3316..616a7b457 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -32,6 +32,8 @@ namespace { using boost::container::small_vector; using boost::container::static_vector; using Shader::ImageBufferDescriptor; +using Shader::Backend::SPIRV::RESCALING_LAYOUT_DOWN_FACTOR_OFFSET; +using Shader::Backend::SPIRV::RESCALING_LAYOUT_WORDS_OFFSET; using Tegra::Texture::TexturePair; using VideoCore::Surface::PixelFormat; using VideoCore::Surface::PixelFormatFromDepthFormat; @@ -235,6 +237,7 @@ GraphicsPipeline::GraphicsPipeline( stage_infos[stage] = *info; enabled_uniform_buffer_masks[stage] = info->constant_buffer_mask; std::ranges::copy(info->constant_buffer_used_sizes, uniform_buffer_sizes[stage].begin()); + num_textures += Shader::NumDescriptors(info->texture_descriptors); } auto func{[this, shader_notify, &render_pass_cache, &descriptor_pool, pipeline_statistics] { DescriptorLayoutBuilder builder{MakeBuilder(device, stage_infos)}; @@ -277,11 +280,10 @@ void GraphicsPipeline::AddTransition(GraphicsPipeline* transition) { template <typename Spec> void GraphicsPipeline::ConfigureImpl(bool is_indexed) { - std::array<ImageId, MAX_IMAGE_ELEMENTS> image_view_ids; - std::array<u32, MAX_IMAGE_ELEMENTS> image_view_indices; + std::array<VideoCommon::ImageViewInOut, MAX_IMAGE_ELEMENTS> views; std::array<VkSampler, MAX_IMAGE_ELEMENTS> samplers; size_t sampler_index{}; - size_t image_index{}; + size_t view_index{}; texture_cache.SynchronizeGraphicsDescriptors(); @@ -322,26 +324,30 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { } return TexturePair(gpu_memory.Read<u32>(addr), via_header_index); }}; - const auto add_image{[&](const auto& desc) { + const auto add_image{[&](const auto& desc, bool blacklist) LAMBDA_FORCEINLINE { for (u32 index = 0; index < desc.count; ++index) { const auto handle{read_handle(desc, index)}; - image_view_indices[image_index++] = handle.first; + views[view_index++] = { + .index = handle.first, + .blacklist = blacklist, + .id = {}, + }; } }}; if constexpr (Spec::has_texture_buffers) { for (const auto& desc : info.texture_buffer_descriptors) { - add_image(desc); + add_image(desc, false); } } if constexpr (Spec::has_image_buffers) { for (const auto& desc : info.image_buffer_descriptors) { - add_image(desc); + add_image(desc, false); } } for (const auto& desc : info.texture_descriptors) { for (u32 index = 0; index < desc.count; ++index) { const auto handle{read_handle(desc, index)}; - image_view_indices[image_index++] = handle.first; + views[view_index++] = {handle.first}; Sampler* const sampler{texture_cache.GetGraphicsSampler(handle.second)}; samplers[sampler_index++] = sampler->Handle(); @@ -349,7 +355,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { } if constexpr (Spec::has_images) { for (const auto& desc : info.image_descriptors) { - add_image(desc); + add_image(desc, desc.is_written); } } }}; @@ -368,10 +374,9 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { if constexpr (Spec::enabled_stages[4]) { config_stage(4); } - const std::span indices_span(image_view_indices.data(), image_index); - texture_cache.FillGraphicsImageViews(indices_span, image_view_ids); + texture_cache.FillGraphicsImageViews<Spec::has_images>(std::span(views.data(), view_index)); - ImageId* texture_buffer_index{image_view_ids.data()}; + VideoCommon::ImageViewInOut* texture_buffer_it{views.data()}; const auto bind_stage_info{[&](size_t stage) LAMBDA_FORCEINLINE { size_t index{}; const auto add_buffer{[&](const auto& desc) { @@ -381,12 +386,12 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { if constexpr (is_image) { is_written = desc.is_written; } - ImageView& image_view{texture_cache.GetImageView(*texture_buffer_index)}; + ImageView& image_view{texture_cache.GetImageView(texture_buffer_it->id)}; buffer_cache.BindGraphicsTextureBuffer(stage, index, image_view.GpuAddr(), image_view.BufferSize(), image_view.format, is_written, is_image); ++index; - ++texture_buffer_index; + ++texture_buffer_it; } }}; buffer_cache.UnbindGraphicsTextureBuffers(stage); @@ -402,13 +407,9 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { add_buffer(desc); } } - for (const auto& desc : info.texture_descriptors) { - texture_buffer_index += desc.count; - } + texture_buffer_it += Shader::NumDescriptors(info.texture_descriptors); if constexpr (Spec::has_images) { - for (const auto& desc : info.image_descriptors) { - texture_buffer_index += desc.count; - } + texture_buffer_it += Shader::NumDescriptors(info.image_descriptors); } }}; if constexpr (Spec::enabled_stages[0]) { @@ -432,12 +433,13 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { update_descriptor_queue.Acquire(); + RescalingPushConstant rescaling; const VkSampler* samplers_it{samplers.data()}; - const ImageId* views_it{image_view_ids.data()}; + const VideoCommon::ImageViewInOut* views_it{views.data()}; const auto prepare_stage{[&](size_t stage) LAMBDA_FORCEINLINE { buffer_cache.BindHostStageBuffers(stage); - PushImageDescriptors(stage_infos[stage], samplers_it, views_it, texture_cache, - update_descriptor_queue); + PushImageDescriptors(texture_cache, update_descriptor_queue, stage_infos[stage], rescaling, + samplers_it, views_it); }}; if constexpr (Spec::enabled_stages[0]) { prepare_stage(0); @@ -454,10 +456,10 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { if constexpr (Spec::enabled_stages[4]) { prepare_stage(4); } - ConfigureDraw(); + ConfigureDraw(rescaling); } -void GraphicsPipeline::ConfigureDraw() { +void GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling) { texture_cache.UpdateRenderTargets(false); scheduler.RequestRenderpass(texture_cache.GetFramebuffer()); @@ -468,12 +470,25 @@ void GraphicsPipeline::ConfigureDraw() { build_condvar.wait(lock, [this] { return is_built.load(std::memory_order::relaxed); }); }); } + const bool is_rescaling{texture_cache.IsRescaling()}; + const bool update_rescaling{scheduler.UpdateRescaling(is_rescaling)}; const bool bind_pipeline{scheduler.UpdateGraphicsPipeline(this)}; const void* const descriptor_data{update_descriptor_queue.UpdateData()}; - scheduler.Record([this, descriptor_data, bind_pipeline](vk::CommandBuffer cmdbuf) { + scheduler.Record([this, descriptor_data, bind_pipeline, rescaling_data = rescaling.Data(), + is_rescaling, update_rescaling](vk::CommandBuffer cmdbuf) { if (bind_pipeline) { cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline); } + cmdbuf.PushConstants(*pipeline_layout, VK_SHADER_STAGE_ALL_GRAPHICS, + RESCALING_LAYOUT_WORDS_OFFSET, sizeof(rescaling_data), + rescaling_data.data()); + if (update_rescaling) { + const f32 config_down_factor{Settings::values.resolution_info.down_factor}; + const f32 scale_down_factor{is_rescaling ? config_down_factor : 1.0f}; + cmdbuf.PushConstants(*pipeline_layout, VK_SHADER_STAGE_ALL_GRAPHICS, + RESCALING_LAYOUT_DOWN_FACTOR_OFFSET, sizeof(scale_down_factor), + &scale_down_factor); + } if (!descriptor_set_layout) { return; } @@ -826,18 +841,10 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) { void GraphicsPipeline::Validate() { size_t num_images{}; for (const auto& info : stage_infos) { - for (const auto& desc : info.texture_buffer_descriptors) { - num_images += desc.count; - } - for (const auto& desc : info.image_buffer_descriptors) { - num_images += desc.count; - } - for (const auto& desc : info.texture_descriptors) { - num_images += desc.count; - } - for (const auto& desc : info.image_descriptors) { - num_images += desc.count; - } + num_images += Shader::NumDescriptors(info.texture_buffer_descriptors); + num_images += Shader::NumDescriptors(info.image_buffer_descriptors); + num_images += Shader::NumDescriptors(info.texture_descriptors); + num_images += Shader::NumDescriptors(info.image_descriptors); } ASSERT(num_images <= MAX_IMAGE_ELEMENTS); } diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.h b/src/video_core/renderer_vulkan/vk_graphics_pipeline.h index 1c780e944..a0c1d8f07 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.h +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.h @@ -62,6 +62,7 @@ namespace Vulkan { class Device; class PipelineStatistics; class RenderPassCache; +class RescalingPushConstant; class VKScheduler; class VKUpdateDescriptorQueue; @@ -113,7 +114,7 @@ private: template <typename Spec> void ConfigureImpl(bool is_indexed); - void ConfigureDraw(); + void ConfigureDraw(const RescalingPushConstant& rescaling); void MakePipeline(VkRenderPass render_pass); @@ -138,6 +139,7 @@ private: std::array<Shader::Info, NUM_STAGES> stage_infos; std::array<u32, 5> enabled_uniform_buffer_masks{}; VideoCommon::UniformBufferSizes uniform_buffer_sizes{}; + u32 num_textures{}; vk::DescriptorSetLayout descriptor_set_layout; DescriptorAllocator descriptor_allocator; diff --git a/src/video_core/renderer_vulkan/vk_master_semaphore.h b/src/video_core/renderer_vulkan/vk_master_semaphore.h index 0886b7da8..9be9c9bed 100644 --- a/src/video_core/renderer_vulkan/vk_master_semaphore.h +++ b/src/video_core/renderer_vulkan/vk_master_semaphore.h @@ -70,7 +70,9 @@ public: return; } // If none of the above is hit, fallback to a regular wait - semaphore.Wait(tick); + while (!semaphore.Wait(tick)) { + } + Refresh(); } private: diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 30b47a7a0..fd334a146 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -58,18 +58,28 @@ struct DrawParams { bool is_indexed; }; -VkViewport GetViewportState(const Device& device, const Maxwell& regs, size_t index) { +VkViewport GetViewportState(const Device& device, const Maxwell& regs, size_t index, float scale) { const auto& src = regs.viewport_transform[index]; - const float width = src.scale_x * 2.0f; - float y = src.translate_y - src.scale_y; - float height = src.scale_y * 2.0f; + const auto conv = [scale](float value) { + float new_value = value * scale; + if (scale < 1.0f) { + const bool sign = std::signbit(value); + new_value = std::round(std::abs(new_value)); + new_value = sign ? -new_value : new_value; + } + return new_value; + }; + const float x = conv(src.translate_x - src.scale_x); + const float width = conv(src.scale_x * 2.0f); + float y = conv(src.translate_y - src.scale_y); + float height = conv(src.scale_y * 2.0f); if (regs.screen_y_control.y_negate) { y += height; height = -height; } const float reduce_z = regs.depth_mode == Maxwell::DepthMode::MinusOneToOne ? 1.0f : 0.0f; VkViewport viewport{ - .x = src.translate_x - src.scale_x, + .x = x, .y = y, .width = width != 0.0f ? width : 1.0f, .height = height != 0.0f ? height : 1.0f, @@ -83,14 +93,27 @@ VkViewport GetViewportState(const Device& device, const Maxwell& regs, size_t in return viewport; } -VkRect2D GetScissorState(const Maxwell& regs, size_t index) { +VkRect2D GetScissorState(const Maxwell& regs, size_t index, u32 up_scale = 1, u32 down_shift = 0) { const auto& src = regs.scissor_test[index]; VkRect2D scissor; + const auto scale_up = [&](s32 value) -> s32 { + if (value == 0) { + return 0U; + } + const s32 upset = value * up_scale; + s32 acumm = 0; + if ((up_scale >> down_shift) == 0) { + acumm = upset % 2; + } + const s32 converted_value = (value * up_scale) >> down_shift; + return value < 0 ? std::min<s32>(converted_value - acumm, -1) + : std::max<s32>(converted_value + acumm, 1); + }; if (src.enable) { - scissor.offset.x = static_cast<s32>(src.min_x); - scissor.offset.y = static_cast<s32>(src.min_y); - scissor.extent.width = src.max_x - src.min_x; - scissor.extent.height = src.max_y - src.min_y; + scissor.offset.x = scale_up(static_cast<s32>(src.min_x)); + scissor.offset.y = scale_up(static_cast<s32>(src.min_y)); + scissor.extent.width = scale_up(src.max_x - src.min_x); + scissor.extent.height = scale_up(src.max_y - src.min_y); } else { scissor.offset.x = 0; scissor.offset.y = 0; @@ -199,7 +222,7 @@ void RasterizerVulkan::Clear() { query_cache.UpdateCounters(); - const auto& regs = maxwell3d.regs; + auto& regs = maxwell3d.regs; const bool use_color = regs.clear_buffers.R || regs.clear_buffers.G || regs.clear_buffers.B || regs.clear_buffers.A; const bool use_depth = regs.clear_buffers.Z; @@ -214,8 +237,16 @@ void RasterizerVulkan::Clear() { const VkExtent2D render_area = framebuffer->RenderArea(); scheduler.RequestRenderpass(framebuffer); + u32 up_scale = 1; + u32 down_shift = 0; + if (texture_cache.IsRescaling()) { + up_scale = Settings::values.resolution_info.up_scale; + down_shift = Settings::values.resolution_info.down_shift; + } + UpdateViewportsState(regs); + VkClearRect clear_rect{ - .rect = GetScissorState(regs, 0), + .rect = GetScissorState(regs, 0, up_scale, down_shift), .baseArrayLayer = regs.clear_buffers.layer, .layerCount = 1, }; @@ -230,7 +261,38 @@ void RasterizerVulkan::Clear() { const u32 color_attachment = regs.clear_buffers.RT; if (use_color && framebuffer->HasAspectColorBit(color_attachment)) { VkClearValue clear_value; - std::memcpy(clear_value.color.float32, regs.clear_color, sizeof(regs.clear_color)); + bool is_integer = false; + bool is_signed = false; + size_t int_size = 8; + for (std::size_t i = 0; i < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets; ++i) { + const auto& this_rt = regs.rt[i]; + if (this_rt.Address() == 0) { + continue; + } + if (this_rt.format == Tegra::RenderTargetFormat::NONE) { + continue; + } + const auto format = + VideoCore::Surface::PixelFormatFromRenderTargetFormat(this_rt.format); + is_integer = IsPixelFormatInteger(format); + is_signed = IsPixelFormatSignedInteger(format); + int_size = PixelComponentSizeBitsInteger(format); + break; + } + if (!is_integer) { + std::memcpy(clear_value.color.float32, regs.clear_color, sizeof(regs.clear_color)); + } else if (!is_signed) { + for (size_t i = 0; i < 4; i++) { + clear_value.color.uint32[i] = static_cast<u32>( + static_cast<f32>(static_cast<u64>(int_size) << 1U) * regs.clear_color[i]); + } + } else { + for (size_t i = 0; i < 4; i++) { + clear_value.color.int32[i] = + static_cast<s32>(static_cast<f32>(static_cast<s64>(int_size - 1) << 1) * + (regs.clear_color[i] - 0.5f)); + } + } scheduler.Record([color_attachment, clear_value, clear_rect](vk::CommandBuffer cmdbuf) { const VkClearAttachment attachment{ @@ -595,15 +657,17 @@ void RasterizerVulkan::UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& reg if (!state_tracker.TouchViewports()) { return; } + const bool is_rescaling{texture_cache.IsRescaling()}; + const float scale = is_rescaling ? Settings::values.resolution_info.up_factor : 1.0f; const std::array viewports{ - GetViewportState(device, regs, 0), GetViewportState(device, regs, 1), - GetViewportState(device, regs, 2), GetViewportState(device, regs, 3), - GetViewportState(device, regs, 4), GetViewportState(device, regs, 5), - GetViewportState(device, regs, 6), GetViewportState(device, regs, 7), - GetViewportState(device, regs, 8), GetViewportState(device, regs, 9), - GetViewportState(device, regs, 10), GetViewportState(device, regs, 11), - GetViewportState(device, regs, 12), GetViewportState(device, regs, 13), - GetViewportState(device, regs, 14), GetViewportState(device, regs, 15), + GetViewportState(device, regs, 0, scale), GetViewportState(device, regs, 1, scale), + GetViewportState(device, regs, 2, scale), GetViewportState(device, regs, 3, scale), + GetViewportState(device, regs, 4, scale), GetViewportState(device, regs, 5, scale), + GetViewportState(device, regs, 6, scale), GetViewportState(device, regs, 7, scale), + GetViewportState(device, regs, 8, scale), GetViewportState(device, regs, 9, scale), + GetViewportState(device, regs, 10, scale), GetViewportState(device, regs, 11, scale), + GetViewportState(device, regs, 12, scale), GetViewportState(device, regs, 13, scale), + GetViewportState(device, regs, 14, scale), GetViewportState(device, regs, 15, scale), }; scheduler.Record([viewports](vk::CommandBuffer cmdbuf) { cmdbuf.SetViewport(0, viewports); }); } @@ -612,13 +676,29 @@ void RasterizerVulkan::UpdateScissorsState(Tegra::Engines::Maxwell3D::Regs& regs if (!state_tracker.TouchScissors()) { return; } + u32 up_scale = 1; + u32 down_shift = 0; + if (texture_cache.IsRescaling()) { + up_scale = Settings::values.resolution_info.up_scale; + down_shift = Settings::values.resolution_info.down_shift; + } const std::array scissors{ - GetScissorState(regs, 0), GetScissorState(regs, 1), GetScissorState(regs, 2), - GetScissorState(regs, 3), GetScissorState(regs, 4), GetScissorState(regs, 5), - GetScissorState(regs, 6), GetScissorState(regs, 7), GetScissorState(regs, 8), - GetScissorState(regs, 9), GetScissorState(regs, 10), GetScissorState(regs, 11), - GetScissorState(regs, 12), GetScissorState(regs, 13), GetScissorState(regs, 14), - GetScissorState(regs, 15), + GetScissorState(regs, 0, up_scale, down_shift), + GetScissorState(regs, 1, up_scale, down_shift), + GetScissorState(regs, 2, up_scale, down_shift), + GetScissorState(regs, 3, up_scale, down_shift), + GetScissorState(regs, 4, up_scale, down_shift), + GetScissorState(regs, 5, up_scale, down_shift), + GetScissorState(regs, 6, up_scale, down_shift), + GetScissorState(regs, 7, up_scale, down_shift), + GetScissorState(regs, 8, up_scale, down_shift), + GetScissorState(regs, 9, up_scale, down_shift), + GetScissorState(regs, 10, up_scale, down_shift), + GetScissorState(regs, 11, up_scale, down_shift), + GetScissorState(regs, 12, up_scale, down_shift), + GetScissorState(regs, 13, up_scale, down_shift), + GetScissorState(regs, 14, up_scale, down_shift), + GetScissorState(regs, 15, up_scale, down_shift), }; scheduler.Record([scissors](vk::CommandBuffer cmdbuf) { cmdbuf.SetScissor(0, scissors); }); } diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index 0c11c814f..3bfdf41ba 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -128,6 +128,15 @@ bool VKScheduler::UpdateGraphicsPipeline(GraphicsPipeline* pipeline) { return true; } +bool VKScheduler::UpdateRescaling(bool is_rescaling) { + if (state.rescaling_defined && is_rescaling == state.is_rescaling) { + return false; + } + state.rescaling_defined = true; + state.is_rescaling = is_rescaling; + return true; +} + void VKScheduler::WorkerThread(std::stop_token stop_token) { Common::SetCurrentThreadName("yuzu:VulkanWorker"); do { @@ -227,6 +236,7 @@ void VKScheduler::AllocateNewContext() { void VKScheduler::InvalidateState() { state.graphics_pipeline = nullptr; + state.rescaling_defined = false; state_tracker.InvalidateCommandBufferState(); } diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h index 85fc1712f..1b06c9296 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.h +++ b/src/video_core/renderer_vulkan/vk_scheduler.h @@ -56,6 +56,9 @@ public: /// Update the pipeline to the current execution context. bool UpdateGraphicsPipeline(GraphicsPipeline* pipeline); + /// Update the rescaling state. Returns true if the state has to be updated. + bool UpdateRescaling(bool is_rescaling); + /// Invalidates current command buffer state except for render passes void InvalidateState(); @@ -185,6 +188,8 @@ private: VkFramebuffer framebuffer = nullptr; VkExtent2D render_area = {0, 0}; GraphicsPipeline* graphics_pipeline = nullptr; + bool is_rescaling = false; + bool rescaling_defined = false; }; void WorkerThread(std::stop_token stop_token); diff --git a/src/video_core/renderer_vulkan/vk_state_tracker.h b/src/video_core/renderer_vulkan/vk_state_tracker.h index 2f2d6b31f..40a149832 100644 --- a/src/video_core/renderer_vulkan/vk_state_tracker.h +++ b/src/video_core/renderer_vulkan/vk_state_tracker.h @@ -71,11 +71,15 @@ public: } bool TouchViewports() { - return Exchange(Dirty::Viewports, false); + const bool dirty_viewports = Exchange(Dirty::Viewports, false); + const bool rescale_viewports = Exchange(VideoCommon::Dirty::RescaleViewports, false); + return dirty_viewports || rescale_viewports; } bool TouchScissors() { - return Exchange(Dirty::Scissors, false); + const bool dirty_scissors = Exchange(Dirty::Scissors, false); + const bool rescale_scissors = Exchange(VideoCommon::Dirty::RescaleScissors, false); + return dirty_scissors || rescale_scissors; } bool TouchDepthBias() { diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 06c5fb867..407fd2a15 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -32,10 +32,12 @@ using Tegra::Engines::Fermi2D; using Tegra::Texture::SwizzleSource; using Tegra::Texture::TextureMipmapFilter; using VideoCommon::BufferImageCopy; +using VideoCommon::ImageFlagBits; using VideoCommon::ImageInfo; using VideoCommon::ImageType; using VideoCommon::SubresourceRange; using VideoCore::Surface::IsPixelFormatASTC; +using VideoCore::Surface::IsPixelFormatInteger; namespace { constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { @@ -588,8 +590,158 @@ struct RangedBarrierRange { UNREACHABLE_MSG("Invalid image format={}", format); return VK_FORMAT_R32_UINT; } + +void BlitScale(VKScheduler& scheduler, VkImage src_image, VkImage dst_image, const ImageInfo& info, + VkImageAspectFlags aspect_mask, const Settings::ResolutionScalingInfo& resolution, + bool up_scaling = true) { + const bool is_2d = info.type == ImageType::e2D; + const auto resources = info.resources; + const VkExtent2D extent{ + .width = info.size.width, + .height = info.size.height, + }; + // Depth and integer formats must use NEAREST filter for blits. + const bool is_color{aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT}; + const bool is_bilinear{is_color && !IsPixelFormatInteger(info.format)}; + const VkFilter vk_filter = is_bilinear ? VK_FILTER_LINEAR : VK_FILTER_NEAREST; + + scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.Record([dst_image, src_image, extent, resources, aspect_mask, resolution, is_2d, + vk_filter, up_scaling](vk::CommandBuffer cmdbuf) { + const VkOffset2D src_size{ + .x = static_cast<s32>(up_scaling ? extent.width : resolution.ScaleUp(extent.width)), + .y = static_cast<s32>(is_2d && up_scaling ? extent.height + : resolution.ScaleUp(extent.height)), + }; + const VkOffset2D dst_size{ + .x = static_cast<s32>(up_scaling ? resolution.ScaleUp(extent.width) : extent.width), + .y = static_cast<s32>(is_2d && up_scaling ? resolution.ScaleUp(extent.height) + : extent.height), + }; + boost::container::small_vector<VkImageBlit, 4> regions; + regions.reserve(resources.levels); + for (s32 level = 0; level < resources.levels; level++) { + regions.push_back({ + .srcSubresource{ + .aspectMask = aspect_mask, + .mipLevel = static_cast<u32>(level), + .baseArrayLayer = 0, + .layerCount = static_cast<u32>(resources.layers), + }, + .srcOffsets{ + { + .x = 0, + .y = 0, + .z = 0, + }, + { + .x = std::max(1, src_size.x >> level), + .y = std::max(1, src_size.y >> level), + .z = 1, + }, + }, + .dstSubresource{ + .aspectMask = aspect_mask, + .mipLevel = static_cast<u32>(level), + .baseArrayLayer = 0, + .layerCount = static_cast<u32>(resources.layers), + }, + .dstOffsets{ + { + .x = 0, + .y = 0, + .z = 0, + }, + { + .x = std::max(1, dst_size.x >> level), + .y = std::max(1, dst_size.y >> level), + .z = 1, + }, + }, + }); + } + const VkImageSubresourceRange subresource_range{ + .aspectMask = aspect_mask, + .baseMipLevel = 0, + .levelCount = VK_REMAINING_MIP_LEVELS, + .baseArrayLayer = 0, + .layerCount = VK_REMAINING_ARRAY_LAYERS, + }; + const std::array read_barriers{ + VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, + .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .oldLayout = VK_IMAGE_LAYOUT_GENERAL, + .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = src_image, + .subresourceRange = subresource_range, + }, + VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | + VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED, // Discard contents + .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = dst_image, + .subresourceRange = subresource_range, + }, + }; + const std::array write_barriers{ + VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = 0, + .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT | VK_ACCESS_MEMORY_READ_BIT, + .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = src_image, + .subresourceRange = subresource_range, + }, + VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT | VK_ACCESS_MEMORY_READ_BIT, + .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = dst_image, + .subresourceRange = subresource_range, + }, + }; + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, + 0, nullptr, nullptr, read_barriers); + cmdbuf.BlitImage(src_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_image, + VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, regions, vk_filter); + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + 0, nullptr, nullptr, write_barriers); + }); +} } // Anonymous namespace +TextureCacheRuntime::TextureCacheRuntime(const Device& device_, VKScheduler& scheduler_, + MemoryAllocator& memory_allocator_, + StagingBufferPool& staging_buffer_pool_, + BlitImageHelper& blit_image_helper_, + ASTCDecoderPass& astc_decoder_pass_, + RenderPassCache& render_pass_cache_) + : device{device_}, scheduler{scheduler_}, memory_allocator{memory_allocator_}, + staging_buffer_pool{staging_buffer_pool_}, blit_image_helper{blit_image_helper_}, + astc_decoder_pass{astc_decoder_pass_}, render_pass_cache{render_pass_cache_}, + resolution{Settings::values.resolution_info} {} + void TextureCacheRuntime::Finish() { scheduler.Finish(); } @@ -614,8 +766,8 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst return; } if (aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT && !is_src_msaa && !is_dst_msaa) { - blit_image_helper.BlitColor(dst_framebuffer, src, dst_region, src_region, filter, - operation); + blit_image_helper.BlitColor(dst_framebuffer, src.Handle(Shader::TextureType::Color2D), + dst_region, src_region, filter, operation); return; } if (aspect_mask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) { @@ -719,26 +871,29 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst }); } -void TextureCacheRuntime::ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view) { +void TextureCacheRuntime::ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view, + bool rescaled) { + const u32 up_scale = rescaled ? resolution.up_scale : 1; + const u32 down_shift = rescaled ? resolution.down_shift : 0; switch (dst_view.format) { case PixelFormat::R16_UNORM: if (src_view.format == PixelFormat::D16_UNORM) { - return blit_image_helper.ConvertD16ToR16(dst, src_view); + return blit_image_helper.ConvertD16ToR16(dst, src_view, up_scale, down_shift); } break; case PixelFormat::R32_FLOAT: if (src_view.format == PixelFormat::D32_FLOAT) { - return blit_image_helper.ConvertD32ToR32(dst, src_view); + return blit_image_helper.ConvertD32ToR32(dst, src_view, up_scale, down_shift); } break; case PixelFormat::D16_UNORM: if (src_view.format == PixelFormat::R16_UNORM) { - return blit_image_helper.ConvertR16ToD16(dst, src_view); + return blit_image_helper.ConvertR16ToD16(dst, src_view, up_scale, down_shift); } break; case PixelFormat::D32_FLOAT: if (src_view.format == PixelFormat::R32_FLOAT) { - return blit_image_helper.ConvertR32ToD32(dst, src_view); + return blit_image_helper.ConvertR32ToD32(dst, src_view, up_scale, down_shift); } break; default: @@ -840,36 +995,39 @@ u64 TextureCacheRuntime::GetDeviceLocalMemory() const { return device.GetDeviceLocalMemory(); } -Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_addr_, +void TextureCacheRuntime::TickFrame() {} + +Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu_addr_, VAddr cpu_addr_) - : VideoCommon::ImageBase(info_, gpu_addr_, cpu_addr_), scheduler{&runtime.scheduler}, - image(MakeImage(runtime.device, info)), - commit(runtime.memory_allocator.Commit(image, MemoryUsage::DeviceLocal)), + : VideoCommon::ImageBase(info_, gpu_addr_, cpu_addr_), scheduler{&runtime_.scheduler}, + runtime{&runtime_}, original_image(MakeImage(runtime_.device, info)), + commit(runtime_.memory_allocator.Commit(original_image, MemoryUsage::DeviceLocal)), aspect_mask(ImageAspectMask(info.format)) { - if (IsPixelFormatASTC(info.format) && !runtime.device.IsOptimalAstcSupported()) { + if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported()) { if (Settings::values.accelerate_astc.GetValue()) { flags |= VideoCommon::ImageFlagBits::AcceleratedUpload; } else { flags |= VideoCommon::ImageFlagBits::Converted; } } - if (runtime.device.HasDebuggingToolAttached()) { - image.SetObjectNameEXT(VideoCommon::Name(*this).c_str()); + if (runtime->device.HasDebuggingToolAttached()) { + original_image.SetObjectNameEXT(VideoCommon::Name(*this).c_str()); } static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO, .pNext = nullptr, .usage = VK_IMAGE_USAGE_STORAGE_BIT, }; - if (IsPixelFormatASTC(info.format) && !runtime.device.IsOptimalAstcSupported()) { - const auto& device = runtime.device.GetLogical(); + current_image = *original_image; + if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported()) { + const auto& device = runtime->device.GetLogical(); storage_image_views.reserve(info.resources.levels); for (s32 level = 0; level < info.resources.levels; ++level) { storage_image_views.push_back(device.CreateImageView(VkImageViewCreateInfo{ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, .pNext = &storage_image_view_usage_create_info, .flags = 0, - .image = *image, + .image = *original_image, .viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY, .format = VK_FORMAT_A8B8G8R8_UNORM_PACK32, .components{ @@ -890,26 +1048,39 @@ Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_ } } +Image::Image(const VideoCommon::NullImageParams& params) : VideoCommon::ImageBase{params} {} + Image::~Image() = default; void Image::UploadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) { // TODO: Move this to another API + const bool is_rescaled = True(flags & ImageFlagBits::Rescaled); + if (is_rescaled) { + ScaleDown(true); + } scheduler->RequestOutsideRenderPassOperationContext(); std::vector vk_copies = TransformBufferImageCopies(copies, map.offset, aspect_mask); const VkBuffer src_buffer = map.buffer; - const VkImage vk_image = *image; + const VkImage vk_image = *original_image; const VkImageAspectFlags vk_aspect_mask = aspect_mask; const bool is_initialized = std::exchange(initialized, true); scheduler->Record([src_buffer, vk_image, vk_aspect_mask, is_initialized, vk_copies](vk::CommandBuffer cmdbuf) { CopyBufferToImage(cmdbuf, src_buffer, vk_image, vk_aspect_mask, is_initialized, vk_copies); }); + if (is_rescaled) { + ScaleUp(); + } } void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) { + const bool is_rescaled = True(flags & ImageFlagBits::Rescaled); + if (is_rescaled) { + ScaleDown(); + } std::vector vk_copies = TransformBufferImageCopies(copies, map.offset, aspect_mask); scheduler->RequestOutsideRenderPassOperationContext(); - scheduler->Record([buffer = map.buffer, image = *image, aspect_mask = aspect_mask, + scheduler->Record([buffer = map.buffer, image = *original_image, aspect_mask = aspect_mask, vk_copies](vk::CommandBuffer cmdbuf) { const VkImageMemoryBarrier read_barrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, @@ -959,6 +1130,146 @@ void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferIm cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, memory_write_barrier, nullptr, image_write_barrier); }); + if (is_rescaled) { + ScaleUp(true); + } +} + +bool Image::ScaleUp(bool ignore) { + if (True(flags & ImageFlagBits::Rescaled)) { + return false; + } + ASSERT(info.type != ImageType::Linear); + flags |= ImageFlagBits::Rescaled; + const auto& resolution = runtime->resolution; + if (!resolution.active) { + return false; + } + has_scaled = true; + const auto& device = runtime->device; + if (!scaled_image) { + const bool is_2d = info.type == ImageType::e2D; + const u32 scaled_width = resolution.ScaleUp(info.size.width); + const u32 scaled_height = is_2d ? resolution.ScaleUp(info.size.height) : info.size.height; + auto scaled_info = info; + scaled_info.size.width = scaled_width; + scaled_info.size.height = scaled_height; + scaled_image = MakeImage(device, scaled_info); + auto& allocator = runtime->memory_allocator; + scaled_commit = MemoryCommit(allocator.Commit(scaled_image, MemoryUsage::DeviceLocal)); + ignore = false; + } + current_image = *scaled_image; + if (ignore) { + return true; + } + + if (aspect_mask == 0) { + aspect_mask = ImageAspectMask(info.format); + } + static constexpr auto OPTIMAL_FORMAT = FormatType::Optimal; + const PixelFormat format = StorageFormat(info.format); + const auto vk_format = MaxwellToVK::SurfaceFormat(device, OPTIMAL_FORMAT, false, format).format; + const auto blit_usage = VK_FORMAT_FEATURE_BLIT_SRC_BIT | VK_FORMAT_FEATURE_BLIT_DST_BIT; + if (device.IsFormatSupported(vk_format, blit_usage, OPTIMAL_FORMAT)) { + BlitScale(*scheduler, *original_image, *scaled_image, info, aspect_mask, resolution); + } else { + return BlitScaleHelper(true); + } + return true; +} + +bool Image::ScaleDown(bool ignore) { + if (False(flags & ImageFlagBits::Rescaled)) { + return false; + } + ASSERT(info.type != ImageType::Linear); + flags &= ~ImageFlagBits::Rescaled; + const auto& resolution = runtime->resolution; + if (!resolution.active) { + return false; + } + current_image = *original_image; + if (ignore) { + return true; + } + if (aspect_mask == 0) { + aspect_mask = ImageAspectMask(info.format); + } + static constexpr auto OPTIMAL_FORMAT = FormatType::Optimal; + const PixelFormat format = StorageFormat(info.format); + const auto& device = runtime->device; + const auto vk_format = MaxwellToVK::SurfaceFormat(device, OPTIMAL_FORMAT, false, format).format; + const auto blit_usage = VK_FORMAT_FEATURE_BLIT_SRC_BIT | VK_FORMAT_FEATURE_BLIT_DST_BIT; + if (device.IsFormatSupported(vk_format, blit_usage, OPTIMAL_FORMAT)) { + BlitScale(*scheduler, *scaled_image, *original_image, info, aspect_mask, resolution, false); + } else { + return BlitScaleHelper(false); + } + return true; +} + +bool Image::BlitScaleHelper(bool scale_up) { + using namespace VideoCommon; + static constexpr auto BLIT_OPERATION = Tegra::Engines::Fermi2D::Operation::SrcCopy; + const bool is_color{aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT}; + const bool is_bilinear{is_color && !IsPixelFormatInteger(info.format)}; + const auto operation = is_bilinear ? Tegra::Engines::Fermi2D::Filter::Bilinear + : Tegra::Engines::Fermi2D::Filter::Point; + + const bool is_2d = info.type == ImageType::e2D; + const auto& resolution = runtime->resolution; + const u32 scaled_width = resolution.ScaleUp(info.size.width); + const u32 scaled_height = is_2d ? resolution.ScaleUp(info.size.height) : info.size.height; + std::unique_ptr<ImageView>& blit_view = scale_up ? scale_view : normal_view; + std::unique_ptr<Framebuffer>& blit_framebuffer = + scale_up ? scale_framebuffer : normal_framebuffer; + if (!blit_view) { + const auto view_info = ImageViewInfo(ImageViewType::e2D, info.format); + blit_view = std::make_unique<ImageView>(*runtime, view_info, NULL_IMAGE_ID, *this); + } + + const u32 src_width = scale_up ? info.size.width : scaled_width; + const u32 src_height = scale_up ? info.size.height : scaled_height; + const u32 dst_width = scale_up ? scaled_width : info.size.width; + const u32 dst_height = scale_up ? scaled_height : info.size.height; + const Region2D src_region{ + .start = {0, 0}, + .end = {static_cast<s32>(src_width), static_cast<s32>(src_height)}, + }; + const Region2D dst_region{ + .start = {0, 0}, + .end = {static_cast<s32>(dst_width), static_cast<s32>(dst_height)}, + }; + const VkExtent2D extent{ + .width = std::max(scaled_width, info.size.width), + .height = std::max(scaled_height, info.size.width), + }; + + auto* view_ptr = blit_view.get(); + if (aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT) { + if (!blit_framebuffer) { + blit_framebuffer = std::make_unique<Framebuffer>(*runtime, view_ptr, nullptr, extent); + } + const auto color_view = blit_view->Handle(Shader::TextureType::Color2D); + + runtime->blit_image_helper.BlitColor(blit_framebuffer.get(), color_view, dst_region, + src_region, operation, BLIT_OPERATION); + } else if (!runtime->device.IsBlitDepthStencilSupported() && + aspect_mask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) { + if (!blit_framebuffer) { + blit_framebuffer = std::make_unique<Framebuffer>(*runtime, nullptr, view_ptr, extent); + } + runtime->blit_image_helper.BlitDepthStencil(blit_framebuffer.get(), blit_view->DepthView(), + blit_view->StencilView(), dst_region, + src_region, operation, BLIT_OPERATION); + } else { + // TODO: Use helper blits where applicable + flags &= ~ImageFlagBits::Rescaled; + LOG_ERROR(Render_Vulkan, "Device does not support scaling format {}", info.format); + return false; + } + return true; } ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewInfo& info, @@ -1052,9 +1363,11 @@ ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, : VideoCommon::ImageViewBase{info, view_info}, gpu_addr{gpu_addr_}, buffer_size{VideoCommon::CalculateGuestSizeInBytes(info)} {} -ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::NullImageParams& params) +ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::NullImageViewParams& params) : VideoCommon::ImageViewBase{params} {} +ImageView::~ImageView() = default; + VkImageView ImageView::DepthView() { if (depth_view) { return *depth_view; @@ -1137,7 +1450,8 @@ Sampler::Sampler(TextureCacheRuntime& runtime, const Tegra::Texture::TSCEntry& t LOG_WARNING(Render_Vulkan, "VK_EXT_sampler_filter_minmax is required"); } // Some games have samplers with garbage. Sanitize them here. - const float max_anisotropy = std::clamp(tsc.MaxAnisotropy(), 1.0f, 16.0f); + const f32 max_anisotropy = std::clamp(tsc.MaxAnisotropy(), 1.0f, 16.0f); + sampler = device.GetLogical().CreateSampler(VkSamplerCreateInfo{ .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO, .pNext = pnext, @@ -1162,7 +1476,29 @@ Sampler::Sampler(TextureCacheRuntime& runtime, const Tegra::Texture::TSCEntry& t } Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM_RT> color_buffers, - ImageView* depth_buffer, const VideoCommon::RenderTargets& key) { + ImageView* depth_buffer, const VideoCommon::RenderTargets& key) + : render_area{VkExtent2D{ + .width = key.size.width, + .height = key.size.height, + }} { + CreateFramebuffer(runtime, color_buffers, depth_buffer); + if (runtime.device.HasDebuggingToolAttached()) { + framebuffer.SetObjectNameEXT(VideoCommon::Name(key).c_str()); + } +} + +Framebuffer::Framebuffer(TextureCacheRuntime& runtime, ImageView* color_buffer, + ImageView* depth_buffer, VkExtent2D extent) + : render_area{extent} { + std::array<ImageView*, NUM_RT> color_buffers{color_buffer}; + CreateFramebuffer(runtime, color_buffers, depth_buffer); +} + +Framebuffer::~Framebuffer() = default; + +void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime, + std::span<ImageView*, NUM_RT> color_buffers, + ImageView* depth_buffer) { std::vector<VkImageView> attachments; RenderPassKey renderpass_key{}; s32 num_layers = 1; @@ -1200,10 +1536,6 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM renderpass = runtime.render_pass_cache.Get(renderpass_key); - render_area = VkExtent2D{ - .width = key.size.width, - .height = key.size.height, - }; num_color_buffers = static_cast<u32>(num_colors); framebuffer = runtime.device.GetLogical().CreateFramebuffer({ .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO, @@ -1212,13 +1544,10 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM .renderPass = renderpass, .attachmentCount = static_cast<u32>(attachments.size()), .pAttachments = attachments.data(), - .width = key.size.width, - .height = key.size.height, + .width = render_area.width, + .height = render_area.height, .layers = static_cast<u32>(std::max(num_layers, 1)), }); - if (runtime.device.HasDebuggingToolAttached()) { - framebuffer.SetObjectNameEXT(VideoCommon::Name(key).c_str()); - } } void TextureCacheRuntime::AccelerateImageUpload( diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index b09c468e4..ff28b4e96 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -13,6 +13,10 @@ #include "video_core/vulkan_common/vulkan_memory_allocator.h" #include "video_core/vulkan_common/vulkan_wrapper.h" +namespace Settings { +struct ResolutionScalingInfo; +} + namespace Vulkan { using VideoCommon::ImageId; @@ -31,14 +35,14 @@ class RenderPassCache; class StagingBufferPool; class VKScheduler; -struct TextureCacheRuntime { - const Device& device; - VKScheduler& scheduler; - MemoryAllocator& memory_allocator; - StagingBufferPool& staging_buffer_pool; - BlitImageHelper& blit_image_helper; - ASTCDecoderPass& astc_decoder_pass; - RenderPassCache& render_pass_cache; +class TextureCacheRuntime { +public: + explicit TextureCacheRuntime(const Device& device_, VKScheduler& scheduler_, + MemoryAllocator& memory_allocator_, + StagingBufferPool& staging_buffer_pool_, + BlitImageHelper& blit_image_helper_, + ASTCDecoderPass& astc_decoder_pass_, + RenderPassCache& render_pass_cache_); void Finish(); @@ -46,6 +50,10 @@ struct TextureCacheRuntime { StagingBufferRef DownloadStagingBuffer(size_t size); + void TickFrame(); + + u64 GetDeviceLocalMemory() const; + void BlitImage(Framebuffer* dst_framebuffer, ImageView& dst, ImageView& src, const Region2D& dst_region, const Region2D& src_region, Tegra::Engines::Fermi2D::Filter filter, @@ -53,7 +61,7 @@ struct TextureCacheRuntime { void CopyImage(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies); - void ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view); + void ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view, bool rescaled); bool CanAccelerateImageUpload(Image&) const noexcept { return false; @@ -74,13 +82,21 @@ struct TextureCacheRuntime { return true; } - u64 GetDeviceLocalMemory() const; + const Device& device; + VKScheduler& scheduler; + MemoryAllocator& memory_allocator; + StagingBufferPool& staging_buffer_pool; + BlitImageHelper& blit_image_helper; + ASTCDecoderPass& astc_decoder_pass; + RenderPassCache& render_pass_cache; + const Settings::ResolutionScalingInfo& resolution; }; class Image : public VideoCommon::ImageBase { public: explicit Image(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, GPUVAddr gpu_addr, VAddr cpu_addr); + explicit Image(const VideoCommon::NullImageParams&); ~Image(); @@ -97,7 +113,7 @@ public: std::span<const VideoCommon::BufferImageCopy> copies); [[nodiscard]] VkImage Handle() const noexcept { - return *image; + return current_image; } [[nodiscard]] VkImageAspectFlags AspectMask() const noexcept { @@ -113,14 +129,30 @@ public: return std::exchange(initialized, true); } + bool ScaleUp(bool ignore = false); + + bool ScaleDown(bool ignore = false); + private: - VKScheduler* scheduler; - vk::Image image; + bool BlitScaleHelper(bool scale_up); + + VKScheduler* scheduler{}; + TextureCacheRuntime* runtime{}; + + vk::Image original_image; MemoryCommit commit; - vk::ImageView image_view; std::vector<vk::ImageView> storage_image_views; VkImageAspectFlags aspect_mask = 0; bool initialized = false; + vk::Image scaled_image{}; + MemoryCommit scaled_commit{}; + VkImage current_image{}; + + std::unique_ptr<Framebuffer> scale_framebuffer; + std::unique_ptr<ImageView> scale_view; + + std::unique_ptr<Framebuffer> normal_framebuffer; + std::unique_ptr<ImageView> normal_view; }; class ImageView : public VideoCommon::ImageViewBase { @@ -128,7 +160,15 @@ public: explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageViewInfo&, ImageId, Image&); explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo&, const VideoCommon::ImageViewInfo&, GPUVAddr); - explicit ImageView(TextureCacheRuntime&, const VideoCommon::NullImageParams&); + explicit ImageView(TextureCacheRuntime&, const VideoCommon::NullImageViewParams&); + + ~ImageView(); + + ImageView(const ImageView&) = delete; + ImageView& operator=(const ImageView&) = delete; + + ImageView(ImageView&&) = default; + ImageView& operator=(ImageView&&) = default; [[nodiscard]] VkImageView DepthView(); @@ -197,9 +237,23 @@ private: class Framebuffer { public: - explicit Framebuffer(TextureCacheRuntime&, std::span<ImageView*, NUM_RT> color_buffers, + explicit Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM_RT> color_buffers, ImageView* depth_buffer, const VideoCommon::RenderTargets& key); + explicit Framebuffer(TextureCacheRuntime& runtime, ImageView* color_buffer, + ImageView* depth_buffer, VkExtent2D extent); + + ~Framebuffer(); + + Framebuffer(const Framebuffer&) = delete; + Framebuffer& operator=(const Framebuffer&) = delete; + + Framebuffer(Framebuffer&&) = default; + Framebuffer& operator=(Framebuffer&&) = default; + + void CreateFramebuffer(TextureCacheRuntime& runtime, + std::span<ImageView*, NUM_RT> color_buffers, ImageView* depth_buffer); + [[nodiscard]] VkFramebuffer Handle() const noexcept { return *framebuffer; } diff --git a/src/video_core/surface.cpp b/src/video_core/surface.cpp index eb1746265..58d262446 100644 --- a/src/video_core/surface.cpp +++ b/src/video_core/surface.cpp @@ -279,6 +279,80 @@ bool IsPixelFormatSRGB(PixelFormat format) { } } +bool IsPixelFormatInteger(PixelFormat format) { + switch (format) { + case PixelFormat::A8B8G8R8_SINT: + case PixelFormat::A8B8G8R8_UINT: + case PixelFormat::A2B10G10R10_UINT: + case PixelFormat::R8_SINT: + case PixelFormat::R8_UINT: + case PixelFormat::R16G16B16A16_SINT: + case PixelFormat::R16G16B16A16_UINT: + case PixelFormat::R32G32B32A32_UINT: + case PixelFormat::R32G32B32A32_SINT: + case PixelFormat::R32G32_SINT: + case PixelFormat::R16_UINT: + case PixelFormat::R16_SINT: + case PixelFormat::R16G16_UINT: + case PixelFormat::R16G16_SINT: + case PixelFormat::R8G8_SINT: + case PixelFormat::R8G8_UINT: + case PixelFormat::R32G32_UINT: + case PixelFormat::R32_UINT: + case PixelFormat::R32_SINT: + return true; + default: + return false; + } +} + +bool IsPixelFormatSignedInteger(PixelFormat format) { + switch (format) { + case PixelFormat::A8B8G8R8_SINT: + case PixelFormat::R8_SINT: + case PixelFormat::R16G16B16A16_SINT: + case PixelFormat::R32G32B32A32_SINT: + case PixelFormat::R32G32_SINT: + case PixelFormat::R16_SINT: + case PixelFormat::R16G16_SINT: + case PixelFormat::R8G8_SINT: + case PixelFormat::R32_SINT: + return true; + default: + return false; + } +} + +size_t PixelComponentSizeBitsInteger(PixelFormat format) { + switch (format) { + case PixelFormat::A8B8G8R8_SINT: + case PixelFormat::A8B8G8R8_UINT: + case PixelFormat::R8_SINT: + case PixelFormat::R8_UINT: + case PixelFormat::R8G8_SINT: + case PixelFormat::R8G8_UINT: + return 8; + case PixelFormat::A2B10G10R10_UINT: + return 10; + case PixelFormat::R16G16B16A16_SINT: + case PixelFormat::R16G16B16A16_UINT: + case PixelFormat::R16_UINT: + case PixelFormat::R16_SINT: + case PixelFormat::R16G16_UINT: + case PixelFormat::R16G16_SINT: + return 16; + case PixelFormat::R32G32B32A32_UINT: + case PixelFormat::R32G32B32A32_SINT: + case PixelFormat::R32G32_SINT: + case PixelFormat::R32G32_UINT: + case PixelFormat::R32_UINT: + case PixelFormat::R32_SINT: + return 32; + default: + return 0; + } +} + std::pair<u32, u32> GetASTCBlockSize(PixelFormat format) { return {DefaultBlockWidth(format), DefaultBlockHeight(format)}; } diff --git a/src/video_core/surface.h b/src/video_core/surface.h index 1503db81f..2ce7c7d33 100644 --- a/src/video_core/surface.h +++ b/src/video_core/surface.h @@ -460,6 +460,12 @@ bool IsPixelFormatASTC(PixelFormat format); bool IsPixelFormatSRGB(PixelFormat format); +bool IsPixelFormatInteger(PixelFormat format); + +bool IsPixelFormatSignedInteger(PixelFormat format); + +size_t PixelComponentSizeBitsInteger(PixelFormat format); + std::pair<u32, u32> GetASTCBlockSize(PixelFormat format); u64 EstimatedDecompressedSize(u64 base_size, PixelFormat format); diff --git a/src/video_core/texture_cache/image_base.cpp b/src/video_core/texture_cache/image_base.cpp index 6052d148a..3db2fdf34 100644 --- a/src/video_core/texture_cache/image_base.cpp +++ b/src/video_core/texture_cache/image_base.cpp @@ -60,15 +60,17 @@ namespace { ImageBase::ImageBase(const ImageInfo& info_, GPUVAddr gpu_addr_, VAddr cpu_addr_) : info{info_}, guest_size_bytes{CalculateGuestSizeInBytes(info)}, unswizzled_size_bytes{CalculateUnswizzledSizeBytes(info)}, - converted_size_bytes{CalculateConvertedSizeBytes(info)}, gpu_addr{gpu_addr_}, - cpu_addr{cpu_addr_}, cpu_addr_end{cpu_addr + guest_size_bytes}, - mip_level_offsets{CalculateMipLevelOffsets(info)} { + converted_size_bytes{CalculateConvertedSizeBytes(info)}, scale_rating{}, scale_tick{}, + has_scaled{}, gpu_addr{gpu_addr_}, cpu_addr{cpu_addr_}, + cpu_addr_end{cpu_addr + guest_size_bytes}, mip_level_offsets{CalculateMipLevelOffsets(info)} { if (info.type == ImageType::e3D) { slice_offsets = CalculateSliceOffsets(info); slice_subresources = CalculateSliceSubresources(info); } } +ImageBase::ImageBase(const NullImageParams&) {} + ImageMapView::ImageMapView(GPUVAddr gpu_addr_, VAddr cpu_addr_, size_t size_, ImageId image_id_) : gpu_addr{gpu_addr_}, cpu_addr{cpu_addr_}, size{size_}, image_id{image_id_} {} @@ -254,6 +256,8 @@ void AddImageAlias(ImageBase& lhs, ImageBase& rhs, ImageId lhs_id, ImageId rhs_i } lhs.aliased_images.push_back(std::move(lhs_alias)); rhs.aliased_images.push_back(std::move(rhs_alias)); + lhs.flags &= ~ImageFlagBits::IsRescalable; + rhs.flags &= ~ImageFlagBits::IsRescalable; } } // namespace VideoCommon diff --git a/src/video_core/texture_cache/image_base.h b/src/video_core/texture_cache/image_base.h index 0c17a791b..89c111c00 100644 --- a/src/video_core/texture_cache/image_base.h +++ b/src/video_core/texture_cache/image_base.h @@ -33,6 +33,11 @@ enum class ImageFlagBits : u32 { ///< garbage collection priority Alias = 1 << 11, ///< This image has aliases and has priority on garbage ///< collection + + // Rescaler + Rescaled = 1 << 12, + CheckingRescalable = 1 << 13, + IsRescalable = 1 << 14, }; DECLARE_ENUM_FLAG_OPERATORS(ImageFlagBits) @@ -43,8 +48,11 @@ struct AliasedImage { ImageId id; }; +struct NullImageParams {}; + struct ImageBase { explicit ImageBase(const ImageInfo& info, GPUVAddr gpu_addr, VAddr cpu_addr); + explicit ImageBase(const NullImageParams&); [[nodiscard]] std::optional<SubresourceBase> TryFindBase(GPUVAddr other_addr) const noexcept; @@ -68,11 +76,18 @@ struct ImageBase { void CheckBadOverlapState(); void CheckAliasState(); + bool HasScaled() const { + return has_scaled; + } + ImageInfo info; u32 guest_size_bytes = 0; u32 unswizzled_size_bytes = 0; u32 converted_size_bytes = 0; + u32 scale_rating = 0; + u64 scale_tick = 0; + bool has_scaled = false; ImageFlagBits flags = ImageFlagBits::CpuModified; GPUVAddr gpu_addr = 0; diff --git a/src/video_core/texture_cache/image_info.cpp b/src/video_core/texture_cache/image_info.cpp index 64fd7010a..afb94082b 100644 --- a/src/video_core/texture_cache/image_info.cpp +++ b/src/video_core/texture_cache/image_info.cpp @@ -16,6 +16,7 @@ namespace VideoCommon { using Tegra::Texture::TextureType; using Tegra::Texture::TICEntry; using VideoCore::Surface::PixelFormat; +using VideoCore::Surface::SurfaceType; ImageInfo::ImageInfo(const TICEntry& config) noexcept { format = PixelFormatFromTextureInfo(config.format, config.r_type, config.g_type, config.b_type, @@ -31,6 +32,7 @@ ImageInfo::ImageInfo(const TICEntry& config) noexcept { .depth = config.block_depth, }; } + rescaleable = false; tile_width_spacing = config.tile_width_spacing; if (config.texture_type != TextureType::Texture2D && config.texture_type != TextureType::Texture2DNoMipmap) { @@ -41,6 +43,7 @@ ImageInfo::ImageInfo(const TICEntry& config) noexcept { ASSERT(config.BaseLayer() == 0); type = ImageType::e1D; size.width = config.Width(); + resources.layers = 1; break; case TextureType::Texture1DArray: UNIMPLEMENTED_IF(config.BaseLayer() != 0); @@ -52,12 +55,14 @@ ImageInfo::ImageInfo(const TICEntry& config) noexcept { case TextureType::Texture2DNoMipmap: ASSERT(config.Depth() == 1); type = config.IsPitchLinear() ? ImageType::Linear : ImageType::e2D; + rescaleable = !config.IsPitchLinear(); size.width = config.Width(); size.height = config.Height(); resources.layers = config.BaseLayer() + 1; break; case TextureType::Texture2DArray: type = ImageType::e2D; + rescaleable = true; size.width = config.Width(); size.height = config.Height(); resources.layers = config.BaseLayer() + config.Depth(); @@ -82,10 +87,12 @@ ImageInfo::ImageInfo(const TICEntry& config) noexcept { size.width = config.Width(); size.height = config.Height(); size.depth = config.Depth(); + resources.layers = 1; break; case TextureType::Texture1DBuffer: type = ImageType::Buffer; size.width = config.Width(); + resources.layers = 1; break; default: UNREACHABLE_MSG("Invalid texture_type={}", static_cast<int>(config.texture_type.Value())); @@ -95,12 +102,16 @@ ImageInfo::ImageInfo(const TICEntry& config) noexcept { // FIXME: Call this without passing *this layer_stride = CalculateLayerStride(*this); maybe_unaligned_layer_stride = CalculateLayerSize(*this); + rescaleable &= (block.depth == 0) && resources.levels == 1; + rescaleable &= size.height > 256 || GetFormatType(format) != SurfaceType::ColorTexture; + downscaleable = size.height > 512; } } ImageInfo::ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs, size_t index) noexcept { const auto& rt = regs.rt[index]; format = VideoCore::Surface::PixelFormatFromRenderTargetFormat(rt.format); + rescaleable = false; if (rt.tile_mode.is_pitch_linear) { ASSERT(rt.tile_mode.is_3d == 0); type = ImageType::Linear; @@ -126,6 +137,9 @@ ImageInfo::ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs, size_t index) type = ImageType::e3D; size.depth = rt.depth; } else { + rescaleable = block.depth == 0; + rescaleable &= size.height > 256; + downscaleable = size.height > 512; type = ImageType::e2D; resources.layers = rt.depth; } @@ -135,6 +149,7 @@ ImageInfo::ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs) noexcept { format = VideoCore::Surface::PixelFormatFromDepthFormat(regs.zeta.format); size.width = regs.zeta_width; size.height = regs.zeta_height; + rescaleable = false; resources.levels = 1; layer_stride = regs.zeta.layer_stride * 4; maybe_unaligned_layer_stride = layer_stride; @@ -153,6 +168,8 @@ ImageInfo::ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs) noexcept { type = ImageType::e3D; size.depth = regs.zeta_depth; } else { + rescaleable = block.depth == 0; + downscaleable = size.height > 512; type = ImageType::e2D; resources.layers = regs.zeta_depth; } @@ -161,6 +178,7 @@ ImageInfo::ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs) noexcept { ImageInfo::ImageInfo(const Tegra::Engines::Fermi2D::Surface& config) noexcept { UNIMPLEMENTED_IF_MSG(config.layer != 0, "Surface layer is not zero"); format = VideoCore::Surface::PixelFormatFromRenderTargetFormat(config.format); + rescaleable = false; if (config.linear == Tegra::Engines::Fermi2D::MemoryLayout::Pitch) { type = ImageType::Linear; size = Extent3D{ @@ -171,6 +189,7 @@ ImageInfo::ImageInfo(const Tegra::Engines::Fermi2D::Surface& config) noexcept { pitch = config.pitch; } else { type = config.block_depth > 0 ? ImageType::e3D : ImageType::e2D; + block = Extent3D{ .width = config.block_width, .height = config.block_height, @@ -183,6 +202,9 @@ ImageInfo::ImageInfo(const Tegra::Engines::Fermi2D::Surface& config) noexcept { .height = config.height, .depth = 1, }; + rescaleable = block.depth == 0; + rescaleable &= size.height > 256; + downscaleable = size.height > 512; } } diff --git a/src/video_core/texture_cache/image_info.h b/src/video_core/texture_cache/image_info.h index 5049fc36e..5932dcaba 100644 --- a/src/video_core/texture_cache/image_info.h +++ b/src/video_core/texture_cache/image_info.h @@ -15,7 +15,7 @@ using Tegra::Texture::TICEntry; using VideoCore::Surface::PixelFormat; struct ImageInfo { - explicit ImageInfo() = default; + ImageInfo() = default; explicit ImageInfo(const TICEntry& config) noexcept; explicit ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs, size_t index) noexcept; explicit ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs) noexcept; @@ -33,6 +33,8 @@ struct ImageInfo { u32 maybe_unaligned_layer_stride = 0; u32 num_samples = 1; u32 tile_width_spacing = 0; + bool rescaleable = false; + bool downscaleable = false; }; } // namespace VideoCommon diff --git a/src/video_core/texture_cache/image_view_base.cpp b/src/video_core/texture_cache/image_view_base.cpp index 450becbeb..c7b4fc231 100644 --- a/src/video_core/texture_cache/image_view_base.cpp +++ b/src/video_core/texture_cache/image_view_base.cpp @@ -37,14 +37,15 @@ ImageViewBase::ImageViewBase(const ImageViewInfo& info, const ImageInfo& image_i } ImageViewBase::ImageViewBase(const ImageInfo& info, const ImageViewInfo& view_info) - : format{info.format}, type{ImageViewType::Buffer}, size{ - .width = info.size.width, - .height = 1, - .depth = 1, - } { + : image_id{NULL_IMAGE_ID}, format{info.format}, type{ImageViewType::Buffer}, + size{ + .width = info.size.width, + .height = 1, + .depth = 1, + } { ASSERT_MSG(view_info.type == ImageViewType::Buffer, "Expected texture buffer"); } -ImageViewBase::ImageViewBase(const NullImageParams&) {} +ImageViewBase::ImageViewBase(const NullImageViewParams&) : image_id{NULL_IMAGE_ID} {} } // namespace VideoCommon diff --git a/src/video_core/texture_cache/image_view_base.h b/src/video_core/texture_cache/image_view_base.h index 903f715c5..9c24c5359 100644 --- a/src/video_core/texture_cache/image_view_base.h +++ b/src/video_core/texture_cache/image_view_base.h @@ -15,7 +15,7 @@ using VideoCore::Surface::PixelFormat; struct ImageViewInfo; struct ImageInfo; -struct NullImageParams {}; +struct NullImageViewParams {}; enum class ImageViewFlagBits : u16 { PreemtiveDownload = 1 << 0, @@ -28,7 +28,7 @@ struct ImageViewBase { explicit ImageViewBase(const ImageViewInfo& info, const ImageInfo& image_info, ImageId image_id); explicit ImageViewBase(const ImageInfo& info, const ImageViewInfo& view_info); - explicit ImageViewBase(const NullImageParams&); + explicit ImageViewBase(const NullImageViewParams&); [[nodiscard]] bool IsBuffer() const noexcept { return type == ImageViewType::Buffer; diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index f70c1f764..4d2874bf2 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -7,6 +7,7 @@ #include <unordered_set> #include "common/alignment.h" +#include "common/settings.h" #include "video_core/dirty_flags.h" #include "video_core/engines/kepler_compute.h" #include "video_core/texture_cache/image_view_base.h" @@ -44,21 +45,22 @@ TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface& // Make sure the first index is reserved for the null resources // This way the null resource becomes a compile time constant - void(slot_image_views.insert(runtime, NullImageParams{})); + void(slot_images.insert(NullImageParams{})); + void(slot_image_views.insert(runtime, NullImageViewParams{})); void(slot_samplers.insert(runtime, sampler_descriptor)); if constexpr (HAS_DEVICE_MEMORY_INFO) { const auto device_memory = runtime.GetDeviceLocalMemory(); - const u64 possible_expected_memory = (device_memory * 3) / 10; - const u64 possible_critical_memory = (device_memory * 6) / 10; - expected_memory = std::max(possible_expected_memory, DEFAULT_EXPECTED_MEMORY); - critical_memory = std::max(possible_critical_memory, DEFAULT_CRITICAL_MEMORY); + const u64 possible_expected_memory = (device_memory * 4) / 10; + const u64 possible_critical_memory = (device_memory * 7) / 10; + expected_memory = std::max(possible_expected_memory, DEFAULT_EXPECTED_MEMORY - 256_MiB); + critical_memory = std::max(possible_critical_memory, DEFAULT_CRITICAL_MEMORY - 512_MiB); minimum_memory = 0; } else { - // on OGL we can be more conservatives as the driver takes care. + // On OpenGL we can be more conservatives as the driver takes care. expected_memory = DEFAULT_EXPECTED_MEMORY + 512_MiB; critical_memory = DEFAULT_CRITICAL_MEMORY + 1_GiB; - minimum_memory = expected_memory; + minimum_memory = 0; } } @@ -67,7 +69,7 @@ void TextureCache<P>::RunGarbageCollector() { const bool high_priority_mode = total_used_memory >= expected_memory; const bool aggressive_mode = total_used_memory >= critical_memory; const u64 ticks_to_destroy = aggressive_mode ? 10ULL : high_priority_mode ? 25ULL : 100ULL; - size_t num_iterations = aggressive_mode ? 10000 : (high_priority_mode ? 100 : 5); + size_t num_iterations = aggressive_mode ? 300 : (high_priority_mode ? 50 : 10); const auto clean_up = [this, &num_iterations, high_priority_mode](ImageId image_id) { if (num_iterations == 0) { return true; @@ -89,7 +91,7 @@ void TextureCache<P>::RunGarbageCollector() { UntrackImage(image, image_id); } UnregisterImage(image_id); - DeleteImage(image_id); + DeleteImage(image_id, image.scale_tick > frame_tick + 5); return false; }; lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, clean_up); @@ -103,6 +105,7 @@ void TextureCache<P>::TickFrame() { sentenced_images.Tick(); sentenced_framebuffers.Tick(); sentenced_image_view.Tick(); + runtime.TickFrame(); ++frame_tick; } @@ -122,15 +125,14 @@ void TextureCache<P>::MarkModification(ImageId id) noexcept { } template <class P> -void TextureCache<P>::FillGraphicsImageViews(std::span<const u32> indices, - std::span<ImageViewId> image_view_ids) { - FillImageViews(graphics_image_table, graphics_image_view_ids, indices, image_view_ids); +template <bool has_blacklists> +void TextureCache<P>::FillGraphicsImageViews(std::span<ImageViewInOut> views) { + FillImageViews<has_blacklists>(graphics_image_table, graphics_image_view_ids, views); } template <class P> -void TextureCache<P>::FillComputeImageViews(std::span<const u32> indices, - std::span<ImageViewId> image_view_ids) { - FillImageViews(compute_image_table, compute_image_view_ids, indices, image_view_ids); +void TextureCache<P>::FillComputeImageViews(std::span<ImageViewInOut> views) { + FillImageViews<true>(compute_image_table, compute_image_view_ids, views); } template <class P> @@ -190,6 +192,102 @@ void TextureCache<P>::SynchronizeComputeDescriptors() { } template <class P> +bool TextureCache<P>::RescaleRenderTargets(bool is_clear) { + auto& flags = maxwell3d.dirty.flags; + u32 scale_rating = 0; + bool rescaled = false; + std::array<ImageId, NUM_RT> tmp_color_images{}; + ImageId tmp_depth_image{}; + do { + flags[Dirty::RenderTargets] = false; + + has_deleted_images = false; + // Render target control is used on all render targets, so force look ups when this one is + // up + const bool force = flags[Dirty::RenderTargetControl]; + flags[Dirty::RenderTargetControl] = false; + + scale_rating = 0; + bool any_rescaled = false; + bool can_rescale = true; + const auto check_rescale = [&](ImageViewId view_id, ImageId& id_save) { + if (view_id != NULL_IMAGE_VIEW_ID && view_id != ImageViewId{}) { + const auto& view = slot_image_views[view_id]; + const auto image_id = view.image_id; + id_save = image_id; + auto& image = slot_images[image_id]; + can_rescale &= ImageCanRescale(image); + any_rescaled |= True(image.flags & ImageFlagBits::Rescaled) || + GetFormatType(image.info.format) != SurfaceType::ColorTexture; + scale_rating = std::max<u32>(scale_rating, image.scale_tick <= frame_tick + ? image.scale_rating + 1U + : image.scale_rating); + } else { + id_save = CORRUPT_ID; + } + }; + for (size_t index = 0; index < NUM_RT; ++index) { + ImageViewId& color_buffer_id = render_targets.color_buffer_ids[index]; + if (flags[Dirty::ColorBuffer0 + index] || force) { + flags[Dirty::ColorBuffer0 + index] = false; + BindRenderTarget(&color_buffer_id, FindColorBuffer(index, is_clear)); + } + check_rescale(color_buffer_id, tmp_color_images[index]); + } + if (flags[Dirty::ZetaBuffer] || force) { + flags[Dirty::ZetaBuffer] = false; + BindRenderTarget(&render_targets.depth_buffer_id, FindDepthBuffer(is_clear)); + } + check_rescale(render_targets.depth_buffer_id, tmp_depth_image); + + if (can_rescale) { + rescaled = any_rescaled || scale_rating >= 2; + const auto scale_up = [this](ImageId image_id) { + if (image_id != CORRUPT_ID) { + Image& image = slot_images[image_id]; + ScaleUp(image); + } + }; + if (rescaled) { + for (size_t index = 0; index < NUM_RT; ++index) { + scale_up(tmp_color_images[index]); + } + scale_up(tmp_depth_image); + scale_rating = 2; + } + } else { + rescaled = false; + const auto scale_down = [this](ImageId image_id) { + if (image_id != CORRUPT_ID) { + Image& image = slot_images[image_id]; + ScaleDown(image); + } + }; + for (size_t index = 0; index < NUM_RT; ++index) { + scale_down(tmp_color_images[index]); + } + scale_down(tmp_depth_image); + scale_rating = 1; + } + } while (has_deleted_images); + const auto set_rating = [this, scale_rating](ImageId image_id) { + if (image_id != CORRUPT_ID) { + Image& image = slot_images[image_id]; + image.scale_rating = scale_rating; + if (image.scale_tick <= frame_tick) { + image.scale_tick = frame_tick + 1; + } + } + }; + for (size_t index = 0; index < NUM_RT; ++index) { + set_rating(tmp_color_images[index]); + } + set_rating(tmp_depth_image); + + return rescaled; +} + +template <class P> void TextureCache<P>::UpdateRenderTargets(bool is_clear) { using namespace VideoCommon::Dirty; auto& flags = maxwell3d.dirty.flags; @@ -202,24 +300,18 @@ void TextureCache<P>::UpdateRenderTargets(bool is_clear) { PrepareImageView(depth_buffer_id, true, is_clear && IsFullClear(depth_buffer_id)); return; } - flags[Dirty::RenderTargets] = false; - // Render target control is used on all render targets, so force look ups when this one is up - const bool force = flags[Dirty::RenderTargetControl]; - flags[Dirty::RenderTargetControl] = false; + const bool rescaled = RescaleRenderTargets(is_clear); + if (is_rescaling != rescaled) { + flags[Dirty::RescaleViewports] = true; + flags[Dirty::RescaleScissors] = true; + is_rescaling = rescaled; + } for (size_t index = 0; index < NUM_RT; ++index) { ImageViewId& color_buffer_id = render_targets.color_buffer_ids[index]; - if (flags[Dirty::ColorBuffer0 + index] || force) { - flags[Dirty::ColorBuffer0 + index] = false; - BindRenderTarget(&color_buffer_id, FindColorBuffer(index, is_clear)); - } PrepareImageView(color_buffer_id, true, is_clear && IsFullClear(color_buffer_id)); } - if (flags[Dirty::ZetaBuffer] || force) { - flags[Dirty::ZetaBuffer] = false; - BindRenderTarget(&render_targets.depth_buffer_id, FindDepthBuffer(is_clear)); - } const ImageViewId depth_buffer_id = render_targets.depth_buffer_id; PrepareImageView(depth_buffer_id, true, is_clear && IsFullClear(depth_buffer_id)); @@ -227,9 +319,15 @@ void TextureCache<P>::UpdateRenderTargets(bool is_clear) { for (size_t index = 0; index < NUM_RT; ++index) { render_targets.draw_buffers[index] = static_cast<u8>(maxwell3d.regs.rt_control.Map(index)); } + u32 up_scale = 1; + u32 down_shift = 0; + if (is_rescaling) { + up_scale = Settings::values.resolution_info.up_scale; + down_shift = Settings::values.resolution_info.down_shift; + } render_targets.size = Extent2D{ - maxwell3d.regs.render_area.width, - maxwell3d.regs.render_area.height, + (maxwell3d.regs.render_area.width * up_scale) >> down_shift, + (maxwell3d.regs.render_area.height * up_scale) >> down_shift, }; flags[Dirty::DepthBiasGlobal] = true; @@ -241,17 +339,28 @@ typename P::Framebuffer* TextureCache<P>::GetFramebuffer() { } template <class P> +template <bool has_blacklists> void TextureCache<P>::FillImageViews(DescriptorTable<TICEntry>& table, std::span<ImageViewId> cached_image_view_ids, - std::span<const u32> indices, - std::span<ImageViewId> image_view_ids) { - ASSERT(indices.size() <= image_view_ids.size()); + std::span<ImageViewInOut> views) { + bool has_blacklisted; do { has_deleted_images = false; - std::ranges::transform(indices, image_view_ids.begin(), [&](u32 index) { - return VisitImageView(table, cached_image_view_ids, index); - }); - } while (has_deleted_images); + if constexpr (has_blacklists) { + has_blacklisted = false; + } + for (ImageViewInOut& view : views) { + view.id = VisitImageView(table, cached_image_view_ids, view.index); + if constexpr (has_blacklists) { + if (view.blacklist && view.id != NULL_IMAGE_VIEW_ID) { + const ImageViewBase& image_view{slot_image_views[view.id]}; + auto& image = slot_images[image_view.image_id]; + has_blacklisted |= ScaleDown(image); + image.scale_rating = 0; + } + } + } + } while (has_deleted_images || (has_blacklists && has_blacklisted)); } template <class P> @@ -369,8 +478,43 @@ void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst, PrepareImage(src_id, false, false); PrepareImage(dst_id, true, false); - ImageBase& dst_image = slot_images[dst_id]; - const ImageBase& src_image = slot_images[src_id]; + Image& dst_image = slot_images[dst_id]; + Image& src_image = slot_images[src_id]; + bool is_src_rescaled = True(src_image.flags & ImageFlagBits::Rescaled); + bool is_dst_rescaled = True(dst_image.flags & ImageFlagBits::Rescaled); + + const bool is_resolve = src_image.info.num_samples != 1 && dst_image.info.num_samples == 1; + if (is_src_rescaled != is_dst_rescaled) { + if (ImageCanRescale(src_image)) { + ScaleUp(src_image); + is_src_rescaled = True(src_image.flags & ImageFlagBits::Rescaled); + if (is_resolve) { + dst_image.info.rescaleable = true; + for (const auto& alias : dst_image.aliased_images) { + Image& other_image = slot_images[alias.id]; + other_image.info.rescaleable = true; + } + } + } + if (ImageCanRescale(dst_image)) { + ScaleUp(dst_image); + is_dst_rescaled = True(dst_image.flags & ImageFlagBits::Rescaled); + } + } + if (is_resolve && (is_src_rescaled != is_dst_rescaled)) { + // A resolve requires both images to be the same dimensions. Resize down if needed. + ScaleDown(src_image); + ScaleDown(dst_image); + is_src_rescaled = True(src_image.flags & ImageFlagBits::Rescaled); + is_dst_rescaled = True(dst_image.flags & ImageFlagBits::Rescaled); + } + const auto& resolution = Settings::values.resolution_info; + const auto scale_region = [&](Region2D& region) { + region.start.x = resolution.ScaleUp(region.start.x); + region.start.y = resolution.ScaleUp(region.start.y); + region.end.x = resolution.ScaleUp(region.end.x); + region.end.y = resolution.ScaleUp(region.end.y); + }; // TODO: Deduplicate const std::optional src_base = src_image.TryFindBase(src.Address()); @@ -378,20 +522,26 @@ void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst, const ImageViewInfo src_view_info(ImageViewType::e2D, images.src_format, src_range); const auto [src_framebuffer_id, src_view_id] = RenderTargetFromImage(src_id, src_view_info); const auto [src_samples_x, src_samples_y] = SamplesLog2(src_image.info.num_samples); - const Region2D src_region{ + Region2D src_region{ Offset2D{.x = copy.src_x0 >> src_samples_x, .y = copy.src_y0 >> src_samples_y}, Offset2D{.x = copy.src_x1 >> src_samples_x, .y = copy.src_y1 >> src_samples_y}, }; + if (is_src_rescaled) { + scale_region(src_region); + } const std::optional dst_base = dst_image.TryFindBase(dst.Address()); const SubresourceRange dst_range{.base = dst_base.value(), .extent = {1, 1}}; const ImageViewInfo dst_view_info(ImageViewType::e2D, images.dst_format, dst_range); const auto [dst_framebuffer_id, dst_view_id] = RenderTargetFromImage(dst_id, dst_view_info); const auto [dst_samples_x, dst_samples_y] = SamplesLog2(dst_image.info.num_samples); - const Region2D dst_region{ + Region2D dst_region{ Offset2D{.x = copy.dst_x0 >> dst_samples_x, .y = copy.dst_y0 >> dst_samples_y}, Offset2D{.x = copy.dst_x1 >> dst_samples_x, .y = copy.dst_y1 >> dst_samples_y}, }; + if (is_dst_rescaled) { + scale_region(dst_region); + } // Always call this after src_framebuffer_id was queried, as the address might be invalidated. Framebuffer* const dst_framebuffer = &slot_framebuffers[dst_framebuffer_id]; @@ -487,6 +637,20 @@ void TextureCache<P>::PopAsyncFlushes() { } template <class P> +bool TextureCache<P>::IsRescaling() const noexcept { + return is_rescaling; +} + +template <class P> +bool TextureCache<P>::IsRescaling(const ImageViewBase& image_view) const noexcept { + if (image_view.type == ImageViewType::Buffer) { + return false; + } + const ImageBase& image = slot_images[image_view.image_id]; + return True(image.flags & ImageFlagBits::Rescaled); +} + +template <class P> bool TextureCache<P>::IsRegionGpuModified(VAddr addr, size_t size) { bool is_modified = false; ForEachImageInRegion(addr, size, [&is_modified](ImageId, ImageBase& image) { @@ -624,6 +788,105 @@ ImageId TextureCache<P>::FindImage(const ImageInfo& info, GPUVAddr gpu_addr, } template <class P> +bool TextureCache<P>::ImageCanRescale(ImageBase& image) { + if (!image.info.rescaleable) { + return false; + } + if (Settings::values.resolution_info.downscale && !image.info.downscaleable) { + return false; + } + if (True(image.flags & (ImageFlagBits::Rescaled | ImageFlagBits::CheckingRescalable))) { + return true; + } + if (True(image.flags & ImageFlagBits::IsRescalable)) { + return true; + } + image.flags |= ImageFlagBits::CheckingRescalable; + for (const auto& alias : image.aliased_images) { + Image& other_image = slot_images[alias.id]; + if (!ImageCanRescale(other_image)) { + image.flags &= ~ImageFlagBits::CheckingRescalable; + return false; + } + } + image.flags &= ~ImageFlagBits::CheckingRescalable; + image.flags |= ImageFlagBits::IsRescalable; + return true; +} + +template <class P> +void TextureCache<P>::InvalidateScale(Image& image) { + if (image.scale_tick <= frame_tick) { + image.scale_tick = frame_tick + 1; + } + const std::span<const ImageViewId> image_view_ids = image.image_view_ids; + auto& dirty = maxwell3d.dirty.flags; + dirty[Dirty::RenderTargets] = true; + dirty[Dirty::ZetaBuffer] = true; + for (size_t rt = 0; rt < NUM_RT; ++rt) { + dirty[Dirty::ColorBuffer0 + rt] = true; + } + for (const ImageViewId image_view_id : image_view_ids) { + std::ranges::replace(render_targets.color_buffer_ids, image_view_id, ImageViewId{}); + if (render_targets.depth_buffer_id == image_view_id) { + render_targets.depth_buffer_id = ImageViewId{}; + } + } + RemoveImageViewReferences(image_view_ids); + RemoveFramebuffers(image_view_ids); + for (const ImageViewId image_view_id : image_view_ids) { + sentenced_image_view.Push(std::move(slot_image_views[image_view_id])); + slot_image_views.erase(image_view_id); + } + image.image_view_ids.clear(); + image.image_view_infos.clear(); + if constexpr (ENABLE_VALIDATION) { + std::ranges::fill(graphics_image_view_ids, CORRUPT_ID); + std::ranges::fill(compute_image_view_ids, CORRUPT_ID); + } + graphics_image_table.Invalidate(); + compute_image_table.Invalidate(); + has_deleted_images = true; +} + +template <class P> +u64 TextureCache<P>::GetScaledImageSizeBytes(ImageBase& image) { + const u64 scale_up = static_cast<u64>(Settings::values.resolution_info.up_scale * + Settings::values.resolution_info.up_scale); + const u64 down_shift = static_cast<u64>(Settings::values.resolution_info.down_shift + + Settings::values.resolution_info.down_shift); + const u64 image_size_bytes = + static_cast<u64>(std::max(image.guest_size_bytes, image.unswizzled_size_bytes)); + const u64 tentative_size = (image_size_bytes * scale_up) >> down_shift; + const u64 fitted_size = Common::AlignUp(tentative_size, 1024); + return fitted_size; +} + +template <class P> +bool TextureCache<P>::ScaleUp(Image& image) { + const bool has_copy = image.HasScaled(); + const bool rescaled = image.ScaleUp(); + if (!rescaled) { + return false; + } + if (!has_copy) { + total_used_memory += GetScaledImageSizeBytes(image); + } + InvalidateScale(image); + return true; +} + +template <class P> +bool TextureCache<P>::ScaleDown(Image& image) { + const bool rescaled = image.ScaleDown(); + if (!rescaled) { + return false; + } + InvalidateScale(image); + return true; +} + +template <class P> ImageId TextureCache<P>::InsertImage(const ImageInfo& info, GPUVAddr gpu_addr, RelaxedOptions options) { std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); @@ -660,12 +923,18 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA std::vector<ImageId> right_aliased_ids; std::unordered_set<ImageId> ignore_textures; std::vector<ImageId> bad_overlap_ids; + std::vector<ImageId> all_siblings; + const bool this_is_linear = info.type == ImageType::Linear; const auto region_check = [&](ImageId overlap_id, ImageBase& overlap) { if (True(overlap.flags & ImageFlagBits::Remapped)) { ignore_textures.insert(overlap_id); return; } - if (info.type == ImageType::Linear) { + const bool overlap_is_linear = overlap.info.type == ImageType::Linear; + if (this_is_linear != overlap_is_linear) { + return; + } + if (this_is_linear && overlap_is_linear) { if (info.pitch == overlap.info.pitch && gpu_addr == overlap.gpu_addr) { // Alias linear images with the same pitch left_aliased_ids.push_back(overlap_id); @@ -681,6 +950,7 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA cpu_addr = solution->cpu_addr; new_info.resources = solution->resources; overlap_ids.push_back(overlap_id); + all_siblings.push_back(overlap_id); return; } static constexpr auto options = RelaxedOptions::Size | RelaxedOptions::Format; @@ -688,10 +958,12 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA if (IsSubresource(new_info, overlap, gpu_addr, options, broken_views, native_bgr)) { left_aliased_ids.push_back(overlap_id); overlap.flags |= ImageFlagBits::Alias; + all_siblings.push_back(overlap_id); } else if (IsSubresource(overlap.info, new_image_base, overlap.gpu_addr, options, broken_views, native_bgr)) { right_aliased_ids.push_back(overlap_id); overlap.flags |= ImageFlagBits::Alias; + all_siblings.push_back(overlap_id); } else { bad_overlap_ids.push_back(overlap_id); overlap.flags |= ImageFlagBits::BadOverlap; @@ -709,6 +981,32 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA } }; ForEachSparseImageInRegion(gpu_addr, size_bytes, region_check_gpu); + + bool can_rescale = info.rescaleable; + bool any_rescaled = false; + for (const ImageId sibling_id : all_siblings) { + if (!can_rescale) { + break; + } + Image& sibling = slot_images[sibling_id]; + can_rescale &= ImageCanRescale(sibling); + any_rescaled |= True(sibling.flags & ImageFlagBits::Rescaled); + } + + can_rescale &= any_rescaled; + + if (can_rescale) { + for (const ImageId sibling_id : all_siblings) { + Image& sibling = slot_images[sibling_id]; + ScaleUp(sibling); + } + } else { + for (const ImageId sibling_id : all_siblings) { + Image& sibling = slot_images[sibling_id]; + ScaleDown(sibling); + } + } + const ImageId new_image_id = slot_images.insert(runtime, new_info, gpu_addr, cpu_addr); Image& new_image = slot_images[new_image_id]; @@ -731,14 +1029,23 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA // TODO: Only upload what we need RefreshContents(new_image, new_image_id); + if (can_rescale) { + ScaleUp(new_image); + } else { + ScaleDown(new_image); + } + for (const ImageId overlap_id : overlap_ids) { Image& overlap = slot_images[overlap_id]; if (overlap.info.num_samples != new_image.info.num_samples) { LOG_WARNING(HW_GPU, "Copying between images with different samples is not implemented"); } else { + const auto& resolution = Settings::values.resolution_info; const SubresourceBase base = new_image.TryFindBase(overlap.gpu_addr).value(); - const auto copies = MakeShrinkImageCopies(new_info, overlap.info, base); - runtime.CopyImage(new_image, overlap, copies); + const u32 up_scale = can_rescale ? resolution.up_scale : 1; + const u32 down_shift = can_rescale ? resolution.down_shift : 0; + auto copies = MakeShrinkImageCopies(new_info, overlap.info, base, up_scale, down_shift); + runtime.CopyImage(new_image, overlap, std::move(copies)); } if (True(overlap.flags & ImageFlagBits::Tracked)) { UntrackImage(overlap, overlap_id); @@ -1083,13 +1390,6 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) { "Trying to unregister an already registered image"); image.flags &= ~ImageFlagBits::Registered; image.flags &= ~ImageFlagBits::BadOverlap; - u64 tentative_size = std::max(image.guest_size_bytes, image.unswizzled_size_bytes); - if ((IsPixelFormatASTC(image.info.format) && - True(image.flags & ImageFlagBits::AcceleratedUpload)) || - True(image.flags & ImageFlagBits::Converted)) { - tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format); - } - total_used_memory -= Common::AlignUp(tentative_size, 1024); lru_cache.Free(image.lru_index); const auto& clear_page_table = [this, image_id]( @@ -1213,8 +1513,18 @@ void TextureCache<P>::UntrackImage(ImageBase& image, ImageId image_id) { } template <class P> -void TextureCache<P>::DeleteImage(ImageId image_id) { +void TextureCache<P>::DeleteImage(ImageId image_id, bool immediate_delete) { ImageBase& image = slot_images[image_id]; + if (image.HasScaled()) { + total_used_memory -= GetScaledImageSizeBytes(image); + } + u64 tentative_size = std::max(image.guest_size_bytes, image.unswizzled_size_bytes); + if ((IsPixelFormatASTC(image.info.format) && + True(image.flags & ImageFlagBits::AcceleratedUpload)) || + True(image.flags & ImageFlagBits::Converted)) { + tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format); + } + total_used_memory -= Common::AlignUp(tentative_size, 1024); const GPUVAddr gpu_addr = image.gpu_addr; const auto alloc_it = image_allocs_table.find(gpu_addr); if (alloc_it == image_allocs_table.end()) { @@ -1269,10 +1579,14 @@ void TextureCache<P>::DeleteImage(ImageId image_id) { num_removed_overlaps); } for (const ImageViewId image_view_id : image_view_ids) { - sentenced_image_view.Push(std::move(slot_image_views[image_view_id])); + if (!immediate_delete) { + sentenced_image_view.Push(std::move(slot_image_views[image_view_id])); + } slot_image_views.erase(image_view_id); } - sentenced_images.Push(std::move(slot_images[image_id])); + if (!immediate_delete) { + sentenced_images.Push(std::move(slot_images[image_id])); + } slot_images.erase(image_id); alloc_images.erase(alloc_image_it); @@ -1306,6 +1620,9 @@ void TextureCache<P>::RemoveFramebuffers(std::span<const ImageViewId> removed_vi auto it = framebuffers.begin(); while (it != framebuffers.end()) { if (it->first.Contains(removed_views)) { + auto framebuffer_id = it->second; + ASSERT(framebuffer_id); + sentenced_framebuffers.Push(std::move(slot_framebuffers[framebuffer_id])); it = framebuffers.erase(it); } else { ++it; @@ -1322,26 +1639,60 @@ void TextureCache<P>::MarkModification(ImageBase& image) noexcept { template <class P> void TextureCache<P>::SynchronizeAliases(ImageId image_id) { boost::container::small_vector<const AliasedImage*, 1> aliased_images; - ImageBase& image = slot_images[image_id]; + Image& image = slot_images[image_id]; + bool any_rescaled = True(image.flags & ImageFlagBits::Rescaled); u64 most_recent_tick = image.modification_tick; for (const AliasedImage& aliased : image.aliased_images) { ImageBase& aliased_image = slot_images[aliased.id]; if (image.modification_tick < aliased_image.modification_tick) { most_recent_tick = std::max(most_recent_tick, aliased_image.modification_tick); aliased_images.push_back(&aliased); + any_rescaled |= True(aliased_image.flags & ImageFlagBits::Rescaled); } } if (aliased_images.empty()) { return; } + const bool can_rescale = ImageCanRescale(image); + if (any_rescaled) { + if (can_rescale) { + ScaleUp(image); + } else { + ScaleDown(image); + } + } image.modification_tick = most_recent_tick; std::ranges::sort(aliased_images, [this](const AliasedImage* lhs, const AliasedImage* rhs) { const ImageBase& lhs_image = slot_images[lhs->id]; const ImageBase& rhs_image = slot_images[rhs->id]; return lhs_image.modification_tick < rhs_image.modification_tick; }); + const auto& resolution = Settings::values.resolution_info; for (const AliasedImage* const aliased : aliased_images) { - CopyImage(image_id, aliased->id, aliased->copies); + if (!resolution.active | !any_rescaled) { + CopyImage(image_id, aliased->id, aliased->copies); + continue; + } + Image& aliased_image = slot_images[aliased->id]; + if (!can_rescale) { + ScaleDown(aliased_image); + CopyImage(image_id, aliased->id, aliased->copies); + continue; + } + ScaleUp(aliased_image); + + const bool both_2d{image.info.type == ImageType::e2D && + aliased_image.info.type == ImageType::e2D}; + auto copies = aliased->copies; + for (auto copy : copies) { + copy.extent.width = std::max<u32>( + (copy.extent.width * resolution.up_scale) >> resolution.down_shift, 1); + if (both_2d) { + copy.extent.height = std::max<u32>( + (copy.extent.height * resolution.up_scale) >> resolution.down_shift, 1); + } + } + CopyImage(image_id, aliased->id, copies); } } @@ -1377,9 +1728,25 @@ void TextureCache<P>::PrepareImageView(ImageViewId image_view_id, bool is_modifi } template <class P> -void TextureCache<P>::CopyImage(ImageId dst_id, ImageId src_id, std::span<const ImageCopy> copies) { +void TextureCache<P>::CopyImage(ImageId dst_id, ImageId src_id, std::vector<ImageCopy> copies) { Image& dst = slot_images[dst_id]; Image& src = slot_images[src_id]; + const bool is_rescaled = True(src.flags & ImageFlagBits::Rescaled); + if (is_rescaled) { + ASSERT(True(dst.flags & ImageFlagBits::Rescaled)); + const bool both_2d{src.info.type == ImageType::e2D && dst.info.type == ImageType::e2D}; + const auto& resolution = Settings::values.resolution_info; + for (auto& copy : copies) { + copy.src_offset.x = resolution.ScaleUp(copy.src_offset.x); + copy.dst_offset.x = resolution.ScaleUp(copy.dst_offset.x); + copy.extent.width = resolution.ScaleUp(copy.extent.width); + if (both_2d) { + copy.src_offset.y = resolution.ScaleUp(copy.src_offset.y); + copy.dst_offset.y = resolution.ScaleUp(copy.dst_offset.y); + copy.extent.height = resolution.ScaleUp(copy.extent.height); + } + } + } const auto dst_format_type = GetFormatType(dst.info.format); const auto src_format_type = GetFormatType(src.info.format); if (src_format_type == dst_format_type) { @@ -1424,7 +1791,7 @@ void TextureCache<P>::CopyImage(ImageId dst_id, ImageId src_id, std::span<const }; UNIMPLEMENTED_IF(copy.extent != expected_size); - runtime.ConvertImage(dst_framebuffer, dst_view, src_view); + runtime.ConvertImage(dst_framebuffer, dst_view, src_view, is_rescaled); } } @@ -1433,8 +1800,8 @@ void TextureCache<P>::BindRenderTarget(ImageViewId* old_id, ImageViewId new_id) if (*old_id == new_id) { return; } - if (*old_id) { - const ImageViewBase& old_view = slot_image_views[*old_id]; + if (new_id) { + const ImageViewBase& old_view = slot_image_views[new_id]; if (True(old_view.flags & ImageViewFlagBits::PreemtiveDownload)) { uncommitted_downloads.push_back(old_view.image_id); } @@ -1447,10 +1814,18 @@ std::pair<FramebufferId, ImageViewId> TextureCache<P>::RenderTargetFromImage( ImageId image_id, const ImageViewInfo& view_info) { const ImageViewId view_id = FindOrEmplaceImageView(image_id, view_info); const ImageBase& image = slot_images[image_id]; + const bool is_rescaled = True(image.flags & ImageFlagBits::Rescaled); const bool is_color = GetFormatType(image.info.format) == SurfaceType::ColorTexture; const ImageViewId color_view_id = is_color ? view_id : ImageViewId{}; const ImageViewId depth_view_id = is_color ? ImageViewId{} : view_id; - const Extent3D extent = MipSize(image.info.size, view_info.range.base.level); + Extent3D extent = MipSize(image.info.size, view_info.range.base.level); + if (is_rescaled) { + const auto& resolution = Settings::values.resolution_info; + extent.width = resolution.ScaleUp(extent.width); + if (image.info.type == ImageType::e2D) { + extent.height = resolution.ScaleUp(extent.height); + } + } const u32 num_samples = image.info.num_samples; const auto [samples_x, samples_y] = SamplesLog2(num_samples); const FramebufferId framebuffer_id = GetFramebufferId(RenderTargets{ diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h index 2d1893c1c..643ad811c 100644 --- a/src/video_core/texture_cache/texture_cache_base.h +++ b/src/video_core/texture_cache/texture_cache_base.h @@ -21,6 +21,7 @@ #include "video_core/texture_cache/descriptor_table.h" #include "video_core/texture_cache/image_base.h" #include "video_core/texture_cache/image_info.h" +#include "video_core/texture_cache/image_view_base.h" #include "video_core/texture_cache/image_view_info.h" #include "video_core/texture_cache/render_targets.h" #include "video_core/texture_cache/slot_vector.h" @@ -39,6 +40,12 @@ using VideoCore::Surface::PixelFormatFromDepthFormat; using VideoCore::Surface::PixelFormatFromRenderTargetFormat; using namespace Common::Literals; +struct ImageViewInOut { + u32 index{}; + bool blacklist{}; + ImageViewId id{}; +}; + template <class P> class TextureCache { /// Address shift for caching images into a hash table @@ -53,11 +60,6 @@ class TextureCache { /// True when the API can provide info about the memory of the device. static constexpr bool HAS_DEVICE_MEMORY_INFO = P::HAS_DEVICE_MEMORY_INFO; - /// Image view ID for null descriptors - static constexpr ImageViewId NULL_IMAGE_VIEW_ID{0}; - /// Sampler ID for bugged sampler ids - static constexpr SamplerId NULL_SAMPLER_ID{0}; - static constexpr u64 DEFAULT_EXPECTED_MEMORY = 1_GiB; static constexpr u64 DEFAULT_CRITICAL_MEMORY = 2_GiB; @@ -99,11 +101,11 @@ public: void MarkModification(ImageId id) noexcept; /// Fill image_view_ids with the graphics images in indices - void FillGraphicsImageViews(std::span<const u32> indices, - std::span<ImageViewId> image_view_ids); + template <bool has_blacklists> + void FillGraphicsImageViews(std::span<ImageViewInOut> views); /// Fill image_view_ids with the compute images in indices - void FillComputeImageViews(std::span<const u32> indices, std::span<ImageViewId> image_view_ids); + void FillComputeImageViews(std::span<ImageViewInOut> views); /// Get the sampler from the graphics descriptor table in the specified index Sampler* GetGraphicsSampler(u32 index); @@ -117,6 +119,11 @@ public: /// Refresh the state for compute image view and sampler descriptors void SynchronizeComputeDescriptors(); + /// Updates the Render Targets if they can be rescaled + /// @param is_clear True when the render targets are being used for clears + /// @retval True if the Render Targets have been rescaled. + bool RescaleRenderTargets(bool is_clear); + /// Update bound render targets and upload memory if necessary /// @param is_clear True when the render targets are being used for clears void UpdateRenderTargets(bool is_clear); @@ -160,6 +167,10 @@ public: /// Return true when a CPU region is modified from the GPU [[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size); + [[nodiscard]] bool IsRescaling() const noexcept; + + [[nodiscard]] bool IsRescaling(const ImageViewBase& image_view) const noexcept; + std::mutex mutex; private: @@ -198,9 +209,10 @@ private: void RunGarbageCollector(); /// Fills image_view_ids in the image views in indices + template <bool has_blacklists> void FillImageViews(DescriptorTable<TICEntry>& table, - std::span<ImageViewId> cached_image_view_ids, std::span<const u32> indices, - std::span<ImageViewId> image_view_ids); + std::span<ImageViewId> cached_image_view_ids, + std::span<ImageViewInOut> views); /// Find or create an image view in the guest descriptor table ImageViewId VisitImageView(DescriptorTable<TICEntry>& table, @@ -285,7 +297,7 @@ private: void UntrackImage(ImageBase& image, ImageId image_id); /// Delete image from the cache - void DeleteImage(ImageId image); + void DeleteImage(ImageId image, bool immediate_delete = false); /// Remove image views references from the cache void RemoveImageViewReferences(std::span<const ImageViewId> removed_views); @@ -306,7 +318,7 @@ private: void PrepareImageView(ImageViewId image_view_id, bool is_modification, bool invalidate); /// Execute copies from one image to the other, even if they are incompatible - void CopyImage(ImageId dst_id, ImageId src_id, std::span<const ImageCopy> copies); + void CopyImage(ImageId dst_id, ImageId src_id, std::vector<ImageCopy> copies); /// Bind an image view as render target, downloading resources preemtively if needed void BindRenderTarget(ImageViewId* old_id, ImageViewId new_id); @@ -318,6 +330,12 @@ private: /// Returns true if the current clear parameters clear the whole image of a given image view [[nodiscard]] bool IsFullClear(ImageViewId id); + bool ImageCanRescale(ImageBase& image); + void InvalidateScale(Image& image); + bool ScaleUp(Image& image); + bool ScaleDown(Image& image); + u64 GetScaledImageSizeBytes(ImageBase& image); + Runtime& runtime; VideoCore::RasterizerInterface& rasterizer; Tegra::Engines::Maxwell3D& maxwell3d; @@ -349,6 +367,7 @@ private: VAddr virtual_invalid_space{}; bool has_deleted_images = false; + bool is_rescaling = false; u64 total_used_memory = 0; u64 minimum_memory; u64 expected_memory; diff --git a/src/video_core/texture_cache/types.h b/src/video_core/texture_cache/types.h index 47a11cb2f..5c274abdf 100644 --- a/src/video_core/texture_cache/types.h +++ b/src/video_core/texture_cache/types.h @@ -22,6 +22,13 @@ using ImageAllocId = SlotId; using SamplerId = SlotId; using FramebufferId = SlotId; +/// Fake image ID for null image views +constexpr ImageId NULL_IMAGE_ID{0}; +/// Image view ID for null descriptors +constexpr ImageViewId NULL_IMAGE_VIEW_ID{0}; +/// Sampler ID for bugged sampler ids +constexpr SamplerId NULL_SAMPLER_ID{0}; + enum class ImageType : u32 { e1D, e2D, diff --git a/src/video_core/texture_cache/util.cpp b/src/video_core/texture_cache/util.cpp index 59cf2f561..ddc9fb13a 100644 --- a/src/video_core/texture_cache/util.cpp +++ b/src/video_core/texture_cache/util.cpp @@ -723,7 +723,7 @@ ImageViewType RenderTargetImageViewType(const ImageInfo& info) noexcept { } std::vector<ImageCopy> MakeShrinkImageCopies(const ImageInfo& dst, const ImageInfo& src, - SubresourceBase base) { + SubresourceBase base, u32 up_scale, u32 down_shift) { ASSERT(dst.resources.levels >= src.resources.levels); ASSERT(dst.num_samples == src.num_samples); @@ -732,7 +732,7 @@ std::vector<ImageCopy> MakeShrinkImageCopies(const ImageInfo& dst, const ImageIn ASSERT(src.type == ImageType::e3D); ASSERT(src.resources.levels == 1); } - + const bool both_2d{src.type == ImageType::e2D && dst.type == ImageType::e2D}; std::vector<ImageCopy> copies; copies.reserve(src.resources.levels); for (s32 level = 0; level < src.resources.levels; ++level) { @@ -762,6 +762,10 @@ std::vector<ImageCopy> MakeShrinkImageCopies(const ImageInfo& dst, const ImageIn if (is_dst_3d) { copy.extent.depth = src.size.depth; } + copy.extent.width = std::max<u32>((copy.extent.width * up_scale) >> down_shift, 1); + if (both_2d) { + copy.extent.height = std::max<u32>((copy.extent.height * up_scale) >> down_shift, 1); + } } return copies; } @@ -1153,10 +1157,10 @@ void DeduceBlitImages(ImageInfo& dst_info, ImageInfo& src_info, const ImageBase* if (dst && GetFormatType(dst->info.format) != SurfaceType::ColorTexture) { dst_info.format = dst->info.format; } - if (!dst && src && GetFormatType(src->info.format) != SurfaceType::ColorTexture) { + if (src && GetFormatType(src->info.format) != SurfaceType::ColorTexture) { dst_info.format = src->info.format; } - if (!src && dst && GetFormatType(dst->info.format) != SurfaceType::ColorTexture) { + if (dst && GetFormatType(dst->info.format) != SurfaceType::ColorTexture) { src_info.format = dst->info.format; } } diff --git a/src/video_core/texture_cache/util.h b/src/video_core/texture_cache/util.h index 766502908..7af52de2e 100644 --- a/src/video_core/texture_cache/util.h +++ b/src/video_core/texture_cache/util.h @@ -55,7 +55,8 @@ struct OverlapResult { [[nodiscard]] std::vector<ImageCopy> MakeShrinkImageCopies(const ImageInfo& dst, const ImageInfo& src, - SubresourceBase base); + SubresourceBase base, u32 up_scale = 1, + u32 down_shift = 0); [[nodiscard]] bool IsValidEntry(const Tegra::MemoryManager& gpu_memory, const TICEntry& config); diff --git a/src/video_core/textures/texture.cpp b/src/video_core/textures/texture.cpp index a552543ed..06954963d 100644 --- a/src/video_core/textures/texture.cpp +++ b/src/video_core/textures/texture.cpp @@ -51,22 +51,6 @@ constexpr std::array<float, 256> SRGB_CONVERSION_LUT = { 0.917104f, 0.929242f, 0.941493f, 0.953859f, 0.966338f, 1.000000f, 1.000000f, 1.000000f, }; -unsigned SettingsMinimumAnisotropy() noexcept { - switch (static_cast<Anisotropy>(Settings::values.max_anisotropy.GetValue())) { - default: - case Anisotropy::Default: - return 1U; - case Anisotropy::Filter2x: - return 2U; - case Anisotropy::Filter4x: - return 4U; - case Anisotropy::Filter8x: - return 8U; - case Anisotropy::Filter16x: - return 16U; - } -} - } // Anonymous namespace std::array<float, 4> TSCEntry::BorderColor() const noexcept { @@ -78,7 +62,18 @@ std::array<float, 4> TSCEntry::BorderColor() const noexcept { } float TSCEntry::MaxAnisotropy() const noexcept { - return static_cast<float>(std::max(1U << max_anisotropy, SettingsMinimumAnisotropy())); + if (max_anisotropy == 0 && mipmap_filter != TextureMipmapFilter::Linear) { + return 1.0f; + } + const auto anisotropic_settings = Settings::values.max_anisotropy.GetValue(); + u32 added_anisotropic{}; + if (anisotropic_settings == 0) { + added_anisotropic = Settings::values.resolution_info.up_scale >> + Settings::values.resolution_info.down_shift; + } else { + added_anisotropic = Settings::values.max_anisotropy.GetValue() - 1U; + } + return static_cast<float>(1U << (max_anisotropy + added_anisotropic)); } } // namespace Tegra::Texture diff --git a/src/video_core/video_core.cpp b/src/video_core/video_core.cpp index cae543a51..e852c817e 100644 --- a/src/video_core/video_core.cpp +++ b/src/video_core/video_core.cpp @@ -37,6 +37,8 @@ std::unique_ptr<VideoCore::RendererBase> CreateRenderer( namespace VideoCore { std::unique_ptr<Tegra::GPU> CreateGPU(Core::Frontend::EmuWindow& emu_window, Core::System& system) { + Settings::UpdateRescalingInfo(); + const auto nvdec_value = Settings::values.nvdec_emulation.GetValue(); const bool use_nvdec = nvdec_value != Settings::NvdecEmulation::Off; const bool use_async = Settings::values.use_asynchronous_gpu_emulation.GetValue(); @@ -53,11 +55,10 @@ std::unique_ptr<Tegra::GPU> CreateGPU(Core::Frontend::EmuWindow& emu_window, Cor } } -u16 GetResolutionScaleFactor(const RendererBase& renderer) { - return static_cast<u16>( - Settings::values.resolution_factor.GetValue() != 0 - ? Settings::values.resolution_factor.GetValue() - : renderer.GetRenderWindow().GetFramebufferLayout().GetScalingRatio()); +float GetResolutionScaleFactor(const RendererBase& renderer) { + return Settings::values.resolution_info.active + ? Settings::values.resolution_info.up_factor + : renderer.GetRenderWindow().GetFramebufferLayout().GetScalingRatio(); } } // namespace VideoCore diff --git a/src/video_core/video_core.h b/src/video_core/video_core.h index f5c27125d..f86877e86 100644 --- a/src/video_core/video_core.h +++ b/src/video_core/video_core.h @@ -25,6 +25,6 @@ class RendererBase; /// Creates an emulated GPU instance using the given system context. std::unique_ptr<Tegra::GPU> CreateGPU(Core::Frontend::EmuWindow& emu_window, Core::System& system); -u16 GetResolutionScaleFactor(const RendererBase& renderer); +float GetResolutionScaleFactor(const RendererBase& renderer); } // namespace VideoCore diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 2d5daf6cd..10653ac6b 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -40,6 +40,10 @@ public: VkFormat GetSupportedFormat(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage, FormatType format_type) const; + /// Returns true if a format is supported. + bool IsFormatSupported(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage, + FormatType format_type) const; + /// Reports a device loss. void ReportLoss() const; @@ -370,10 +374,6 @@ private: /// Returns true if the device natively supports blitting depth stencil images. bool TestDepthStencilBlits() const; - /// Returns true if a format is supported. - bool IsFormatSupported(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage, - FormatType format_type) const; - VkInstance instance; ///< Vulkan instance. vk::DeviceDispatch dld; ///< Device function pointers. vk::PhysicalDevice physical; ///< Physical device. diff --git a/src/yuzu/bootmanager.cpp b/src/yuzu/bootmanager.cpp index 46ab0603d..976acd176 100644 --- a/src/yuzu/bootmanager.cpp +++ b/src/yuzu/bootmanager.cpp @@ -628,11 +628,9 @@ void GRenderWindow::ReleaseRenderTarget() { main_context.reset(); } -void GRenderWindow::CaptureScreenshot(u32 res_scale, const QString& screenshot_path) { - VideoCore::RendererBase& renderer = system.Renderer(); - if (res_scale == 0) { - res_scale = VideoCore::GetResolutionScaleFactor(renderer); - } +void GRenderWindow::CaptureScreenshot(const QString& screenshot_path) { + auto& renderer = system.Renderer(); + const f32 res_scale = VideoCore::GetResolutionScaleFactor(renderer); const Layout::FramebufferLayout layout{Layout::FrameLayoutFromResolutionScale(res_scale)}; screenshot_image = QImage(QSize(layout.width, layout.height), QImage::Format_RGB32); diff --git a/src/yuzu/bootmanager.h b/src/yuzu/bootmanager.h index e6a0666e9..40fd4a9d6 100644 --- a/src/yuzu/bootmanager.h +++ b/src/yuzu/bootmanager.h @@ -178,7 +178,7 @@ public: bool IsLoadingComplete() const; - void CaptureScreenshot(u32 res_scale, const QString& screenshot_path); + void CaptureScreenshot(const QString& screenshot_path); std::pair<u32, u32> ScaleTouch(const QPointF& pos) const; diff --git a/src/yuzu/configuration/config.cpp b/src/yuzu/configuration/config.cpp index faea5dda1..8227d06bc 100644 --- a/src/yuzu/configuration/config.cpp +++ b/src/yuzu/configuration/config.cpp @@ -824,6 +824,9 @@ void Config::ReadRendererValues() { ReadGlobalSetting(Settings::values.vulkan_device); ReadGlobalSetting(Settings::values.fullscreen_mode); ReadGlobalSetting(Settings::values.aspect_ratio); + ReadGlobalSetting(Settings::values.resolution_setup); + ReadGlobalSetting(Settings::values.scaling_filter); + ReadGlobalSetting(Settings::values.anti_aliasing); ReadGlobalSetting(Settings::values.max_anisotropy); ReadGlobalSetting(Settings::values.use_speed_limit); ReadGlobalSetting(Settings::values.speed_limit); @@ -1364,6 +1367,18 @@ void Config::SaveRendererValues() { static_cast<u32>(Settings::values.fullscreen_mode.GetDefault()), Settings::values.fullscreen_mode.UsingGlobal()); WriteGlobalSetting(Settings::values.aspect_ratio); + WriteSetting(QString::fromStdString(Settings::values.resolution_setup.GetLabel()), + static_cast<u32>(Settings::values.resolution_setup.GetValue(global)), + static_cast<u32>(Settings::values.resolution_setup.GetDefault()), + Settings::values.resolution_setup.UsingGlobal()); + WriteSetting(QString::fromStdString(Settings::values.scaling_filter.GetLabel()), + static_cast<u32>(Settings::values.scaling_filter.GetValue(global)), + static_cast<u32>(Settings::values.scaling_filter.GetDefault()), + Settings::values.scaling_filter.UsingGlobal()); + WriteSetting(QString::fromStdString(Settings::values.anti_aliasing.GetLabel()), + static_cast<u32>(Settings::values.anti_aliasing.GetValue(global)), + static_cast<u32>(Settings::values.anti_aliasing.GetDefault()), + Settings::values.anti_aliasing.UsingGlobal()); WriteGlobalSetting(Settings::values.max_anisotropy); WriteGlobalSetting(Settings::values.use_speed_limit); WriteGlobalSetting(Settings::values.speed_limit); diff --git a/src/yuzu/configuration/config.h b/src/yuzu/configuration/config.h index a7f4a6720..d673c1cdc 100644 --- a/src/yuzu/configuration/config.h +++ b/src/yuzu/configuration/config.h @@ -189,5 +189,8 @@ Q_DECLARE_METATYPE(Settings::CPUAccuracy); Q_DECLARE_METATYPE(Settings::GPUAccuracy); Q_DECLARE_METATYPE(Settings::FullscreenMode); Q_DECLARE_METATYPE(Settings::NvdecEmulation); +Q_DECLARE_METATYPE(Settings::ResolutionSetup); +Q_DECLARE_METATYPE(Settings::ScalingFilter); +Q_DECLARE_METATYPE(Settings::AntiAliasing); Q_DECLARE_METATYPE(Settings::RendererBackend); Q_DECLARE_METATYPE(Settings::ShaderBackend); diff --git a/src/yuzu/configuration/configure_graphics.cpp b/src/yuzu/configuration/configure_graphics.cpp index 8e20cc6f3..59f975a6e 100644 --- a/src/yuzu/configuration/configure_graphics.cpp +++ b/src/yuzu/configuration/configure_graphics.cpp @@ -89,6 +89,7 @@ void ConfigureGraphics::SetConfiguration() { ui->use_asynchronous_gpu_emulation->setEnabled(runtime_lock); ui->use_disk_shader_cache->setEnabled(runtime_lock); ui->nvdec_emulation_widget->setEnabled(runtime_lock); + ui->resolution_combobox->setEnabled(runtime_lock); ui->accelerate_astc->setEnabled(runtime_lock); ui->use_disk_shader_cache->setChecked(Settings::values.use_disk_shader_cache.GetValue()); ui->use_asynchronous_gpu_emulation->setChecked( @@ -102,6 +103,12 @@ void ConfigureGraphics::SetConfiguration() { ui->nvdec_emulation->setCurrentIndex( static_cast<int>(Settings::values.nvdec_emulation.GetValue())); ui->aspect_ratio_combobox->setCurrentIndex(Settings::values.aspect_ratio.GetValue()); + ui->resolution_combobox->setCurrentIndex( + static_cast<int>(Settings::values.resolution_setup.GetValue())); + ui->scaling_filter_combobox->setCurrentIndex( + static_cast<int>(Settings::values.scaling_filter.GetValue())); + ui->anti_aliasing_combobox->setCurrentIndex( + static_cast<int>(Settings::values.anti_aliasing.GetValue())); } else { ConfigurationShared::SetPerGameSetting(ui->api, &Settings::values.renderer_backend); ConfigurationShared::SetHighlight(ui->api_widget, @@ -122,6 +129,21 @@ void ConfigureGraphics::SetConfiguration() { ConfigurationShared::SetHighlight(ui->ar_label, !Settings::values.aspect_ratio.UsingGlobal()); + ConfigurationShared::SetPerGameSetting(ui->resolution_combobox, + &Settings::values.resolution_setup); + ConfigurationShared::SetHighlight(ui->resolution_label, + !Settings::values.resolution_setup.UsingGlobal()); + + ConfigurationShared::SetPerGameSetting(ui->scaling_filter_combobox, + &Settings::values.scaling_filter); + ConfigurationShared::SetHighlight(ui->scaling_filter_label, + !Settings::values.scaling_filter.UsingGlobal()); + + ConfigurationShared::SetPerGameSetting(ui->anti_aliasing_combobox, + &Settings::values.anti_aliasing); + ConfigurationShared::SetHighlight(ui->anti_aliasing_label, + !Settings::values.anti_aliasing.UsingGlobal()); + ui->bg_combobox->setCurrentIndex(Settings::values.bg_red.UsingGlobal() ? 0 : 1); ui->bg_button->setEnabled(!Settings::values.bg_red.UsingGlobal()); ConfigurationShared::SetHighlight(ui->bg_layout, !Settings::values.bg_red.UsingGlobal()); @@ -133,11 +155,22 @@ void ConfigureGraphics::SetConfiguration() { } void ConfigureGraphics::ApplyConfiguration() { + const auto resolution_setup = static_cast<Settings::ResolutionSetup>( + ui->resolution_combobox->currentIndex() - + ((Settings::IsConfiguringGlobal()) ? 0 : ConfigurationShared::USE_GLOBAL_OFFSET)); + + const auto scaling_filter = static_cast<Settings::ScalingFilter>( + ui->scaling_filter_combobox->currentIndex() - + ((Settings::IsConfiguringGlobal()) ? 0 : ConfigurationShared::USE_GLOBAL_OFFSET)); + + const auto anti_aliasing = static_cast<Settings::AntiAliasing>( + ui->anti_aliasing_combobox->currentIndex() - + ((Settings::IsConfiguringGlobal()) ? 0 : ConfigurationShared::USE_GLOBAL_OFFSET)); + ConfigurationShared::ApplyPerGameSetting(&Settings::values.fullscreen_mode, ui->fullscreen_mode_combobox); ConfigurationShared::ApplyPerGameSetting(&Settings::values.aspect_ratio, ui->aspect_ratio_combobox); - ConfigurationShared::ApplyPerGameSetting(&Settings::values.use_disk_shader_cache, ui->use_disk_shader_cache, use_disk_shader_cache); ConfigurationShared::ApplyPerGameSetting(&Settings::values.use_asynchronous_gpu_emulation, @@ -165,7 +198,34 @@ void ConfigureGraphics::ApplyConfiguration() { Settings::values.bg_green.SetValue(static_cast<u8>(bg_color.green())); Settings::values.bg_blue.SetValue(static_cast<u8>(bg_color.blue())); } + if (Settings::values.resolution_setup.UsingGlobal()) { + Settings::values.resolution_setup.SetValue(resolution_setup); + } + if (Settings::values.scaling_filter.UsingGlobal()) { + Settings::values.scaling_filter.SetValue(scaling_filter); + } + if (Settings::values.anti_aliasing.UsingGlobal()) { + Settings::values.anti_aliasing.SetValue(anti_aliasing); + } } else { + if (ui->resolution_combobox->currentIndex() == ConfigurationShared::USE_GLOBAL_INDEX) { + Settings::values.resolution_setup.SetGlobal(true); + } else { + Settings::values.resolution_setup.SetGlobal(false); + Settings::values.resolution_setup.SetValue(resolution_setup); + } + if (ui->scaling_filter_combobox->currentIndex() == ConfigurationShared::USE_GLOBAL_INDEX) { + Settings::values.scaling_filter.SetGlobal(true); + } else { + Settings::values.scaling_filter.SetGlobal(false); + Settings::values.scaling_filter.SetValue(scaling_filter); + } + if (ui->anti_aliasing_combobox->currentIndex() == ConfigurationShared::USE_GLOBAL_INDEX) { + Settings::values.anti_aliasing.SetGlobal(true); + } else { + Settings::values.anti_aliasing.SetGlobal(false); + Settings::values.anti_aliasing.SetValue(anti_aliasing); + } if (ui->api->currentIndex() == ConfigurationShared::USE_GLOBAL_INDEX) { Settings::values.renderer_backend.SetGlobal(true); Settings::values.shader_backend.SetGlobal(true); @@ -312,6 +372,9 @@ void ConfigureGraphics::SetupPerGameUI() { ui->device->setEnabled(Settings::values.renderer_backend.UsingGlobal()); ui->fullscreen_mode_combobox->setEnabled(Settings::values.fullscreen_mode.UsingGlobal()); ui->aspect_ratio_combobox->setEnabled(Settings::values.aspect_ratio.UsingGlobal()); + ui->resolution_combobox->setEnabled(Settings::values.resolution_setup.UsingGlobal()); + ui->scaling_filter_combobox->setEnabled(Settings::values.scaling_filter.UsingGlobal()); + ui->anti_aliasing_combobox->setEnabled(Settings::values.anti_aliasing.UsingGlobal()); ui->use_asynchronous_gpu_emulation->setEnabled( Settings::values.use_asynchronous_gpu_emulation.UsingGlobal()); ui->nvdec_emulation->setEnabled(Settings::values.nvdec_emulation.UsingGlobal()); @@ -340,6 +403,15 @@ void ConfigureGraphics::SetupPerGameUI() { ConfigurationShared::SetColoredComboBox( ui->fullscreen_mode_combobox, ui->fullscreen_mode_label, static_cast<int>(Settings::values.fullscreen_mode.GetValue(true))); + ConfigurationShared::SetColoredComboBox( + ui->resolution_combobox, ui->resolution_label, + static_cast<int>(Settings::values.resolution_setup.GetValue(true))); + ConfigurationShared::SetColoredComboBox( + ui->scaling_filter_combobox, ui->scaling_filter_label, + static_cast<int>(Settings::values.scaling_filter.GetValue(true))); + ConfigurationShared::SetColoredComboBox( + ui->anti_aliasing_combobox, ui->anti_aliasing_label, + static_cast<int>(Settings::values.anti_aliasing.GetValue(true))); ConfigurationShared::InsertGlobalItem( ui->api, static_cast<int>(Settings::values.renderer_backend.GetValue(true))); ConfigurationShared::InsertGlobalItem( diff --git a/src/yuzu/configuration/configure_graphics.ui b/src/yuzu/configuration/configure_graphics.ui index beae74344..660b68c1c 100644 --- a/src/yuzu/configuration/configure_graphics.ui +++ b/src/yuzu/configuration/configure_graphics.ui @@ -310,6 +310,173 @@ </widget> </item> <item> + <widget class="QWidget" name="resolution_layout" native="true"> + <layout class="QHBoxLayout" name="horizontalLayout_5"> + <property name="leftMargin"> + <number>0</number> + </property> + <property name="topMargin"> + <number>0</number> + </property> + <property name="rightMargin"> + <number>0</number> + </property> + <property name="bottomMargin"> + <number>0</number> + </property> + <item> + <widget class="QLabel" name="resolution_label"> + <property name="text"> + <string>Resolution:</string> + </property> + </widget> + </item> + <item> + <widget class="QComboBox" name="resolution_combobox"> + <item> + <property name="text"> + <string>0.5X (360p/540p) [EXPERIMENTAL]</string> + </property> + </item> + <item> + <property name="text"> + <string>0.75X (540p/810p) [EXPERIMENTAL]</string> + </property> + </item> + <item> + <property name="text"> + <string>1X (720p/1080p)</string> + </property> + </item> + <item> + <property name="text"> + <string>2X (1440p/2160p)</string> + </property> + </item> + <item> + <property name="text"> + <string>3X (2160p/3240p)</string> + </property> + </item> + <item> + <property name="text"> + <string>4X (2880p/4320p)</string> + </property> + </item> + <item> + <property name="text"> + <string>5X (3600p/5400p)</string> + </property> + </item> + <item> + <property name="text"> + <string>6X (4320p/6480p)</string> + </property> + </item> + </widget> + </item> + </layout> + </widget> + </item> + <item> + <widget class="QWidget" name="scaling_filter_layout" native="true"> + <layout class="QHBoxLayout" name="horizontalLayout_6"> + <property name="leftMargin"> + <number>0</number> + </property> + <property name="topMargin"> + <number>0</number> + </property> + <property name="rightMargin"> + <number>0</number> + </property> + <property name="bottomMargin"> + <number>0</number> + </property> + <item> + <widget class="QLabel" name="scaling_filter_label"> + <property name="text"> + <string>Window Adapting Filter:</string> + </property> + </widget> + </item> + <item> + <widget class="QComboBox" name="scaling_filter_combobox"> + <item> + <property name="text"> + <string>Nearest Neighbor</string> + </property> + </item> + <item> + <property name="text"> + <string>Bilinear</string> + </property> + </item> + <item> + <property name="text"> + <string>Bicubic</string> + </property> + </item> + <item> + <property name="text"> + <string>Gaussian</string> + </property> + </item> + <item> + <property name="text"> + <string>ScaleForce</string> + </property> + </item> + <item> + <property name="text"> + <string>AMD's FidelityFX™️ Super Resolution [Vulkan Only]</string> + </property> + </item> + </widget> + </item> + </layout> + </widget> + </item> + <item> + <widget class="QWidget" name="anti_aliasing_layout" native="true"> + <layout class="QHBoxLayout" name="horizontalLayout_7"> + <property name="leftMargin"> + <number>0</number> + </property> + <property name="topMargin"> + <number>0</number> + </property> + <property name="rightMargin"> + <number>0</number> + </property> + <property name="bottomMargin"> + <number>0</number> + </property> + <item> + <widget class="QLabel" name="anti_aliasing_label"> + <property name="text"> + <string>Anti-Aliasing Method:</string> + </property> + </widget> + </item> + <item> + <widget class="QComboBox" name="anti_aliasing_combobox"> + <item> + <property name="text"> + <string>None</string> + </property> + </item> + <item> + <property name="text"> + <string>FXAA</string> + </property> + </item> + </widget> + </item> + </layout> + </widget> + </item> + <item> <widget class="QWidget" name="bg_layout" native="true"> <property name="sizePolicy"> <sizepolicy hsizetype="Preferred" vsizetype="Preferred"> diff --git a/src/yuzu/configuration/configure_graphics_advanced.ui b/src/yuzu/configuration/configure_graphics_advanced.ui index d06b45f17..96de0b3d1 100644 --- a/src/yuzu/configuration/configure_graphics_advanced.ui +++ b/src/yuzu/configuration/configure_graphics_advanced.ui @@ -125,27 +125,32 @@ <widget class="QComboBox" name="anisotropic_filtering_combobox"> <item> <property name="text"> + <string>Automatic</string> + </property> + </item> + <item> + <property name="text"> <string>Default</string> </property> </item> <item> <property name="text"> - <string>2x (WILL BREAK THINGS)</string> + <string>2x</string> </property> </item> <item> <property name="text"> - <string>4x (WILL BREAK THINGS)</string> + <string>4x</string> </property> </item> <item> <property name="text"> - <string>8x (WILL BREAK THINGS)</string> + <string>8x</string> </property> </item> <item> <property name="text"> - <string>16x (WILL BREAK THINGS)</string> + <string>16x</string> </property> </item> </widget> diff --git a/src/yuzu/debugger/profiler.cpp b/src/yuzu/debugger/profiler.cpp index 33110685a..a8b254199 100644 --- a/src/yuzu/debugger/profiler.cpp +++ b/src/yuzu/debugger/profiler.cpp @@ -163,7 +163,7 @@ void MicroProfileWidget::mouseReleaseEvent(QMouseEvent* ev) { } void MicroProfileWidget::wheelEvent(QWheelEvent* ev) { - const auto wheel_position = ev->position().toPoint(); + const auto wheel_position = ev->pos(); MicroProfileMousePosition(wheel_position.x() / x_scale, wheel_position.y() / y_scale, ev->angleDelta().y() / 120); ev->accept(); diff --git a/src/yuzu/game_list.cpp b/src/yuzu/game_list.cpp index 6bd0f9ee9..2af95dbe5 100644 --- a/src/yuzu/game_list.cpp +++ b/src/yuzu/game_list.cpp @@ -159,7 +159,7 @@ GameListSearchField::GameListSearchField(GameList* parent) : QWidget{parent} { * @return true if the haystack contains all words of userinput */ static bool ContainsAllWords(const QString& haystack, const QString& userinput) { - const QStringList userinput_split = userinput.split(QLatin1Char{' '}, Qt::SkipEmptyParts); + const QStringList userinput_split = userinput.split(QLatin1Char{' '}, QString::SkipEmptyParts); return std::all_of(userinput_split.begin(), userinput_split.end(), [&haystack](const QString& s) { return haystack.contains(s); }); diff --git a/src/yuzu/main.cpp b/src/yuzu/main.cpp index 5c767e8f2..d057dc889 100644 --- a/src/yuzu/main.cpp +++ b/src/yuzu/main.cpp @@ -299,6 +299,8 @@ GMainWindow::GMainWindow() SDL_EnableScreenSaver(); #endif + Common::Log::Start(); + QStringList args = QApplication::arguments(); if (args.size() < 2) { @@ -745,6 +747,8 @@ void GMainWindow::InitializeWidgets() { shader_building_label = new QLabel(); shader_building_label->setToolTip(tr("The amount of shaders currently being built")); + res_scale_label = new QLabel(); + res_scale_label->setToolTip(tr("The current selected resolution scaling multiplier.")); emu_speed_label = new QLabel(); emu_speed_label->setToolTip( tr("Current emulation speed. Values higher or lower than 100% " @@ -757,8 +761,8 @@ void GMainWindow::InitializeWidgets() { tr("Time taken to emulate a Switch frame, not counting framelimiting or v-sync. For " "full-speed emulation this should be at most 16.67 ms.")); - for (auto& label : - {shader_building_label, emu_speed_label, game_fps_label, emu_frametime_label}) { + for (auto& label : {shader_building_label, res_scale_label, emu_speed_label, game_fps_label, + emu_frametime_label}) { label->setVisible(false); label->setFrameStyle(QFrame::NoFrame); label->setContentsMargins(4, 0, 4, 0); @@ -770,6 +774,55 @@ void GMainWindow::InitializeWidgets() { tas_label->setFocusPolicy(Qt::NoFocus); statusBar()->insertPermanentWidget(0, tas_label); + // setup AA button + aa_status_button = new QPushButton(); + aa_status_button->setObjectName(QStringLiteral("TogglableStatusBarButton")); + aa_status_button->setFocusPolicy(Qt::NoFocus); + connect(aa_status_button, &QPushButton::clicked, [&] { + auto aa_mode = Settings::values.anti_aliasing.GetValue(); + if (aa_mode == Settings::AntiAliasing::LastAA) { + aa_mode = Settings::AntiAliasing::None; + } else { + aa_mode = static_cast<Settings::AntiAliasing>(static_cast<u32>(aa_mode) + 1); + } + Settings::values.anti_aliasing.SetValue(aa_mode); + aa_status_button->setChecked(true); + UpdateAAText(); + }); + UpdateAAText(); + aa_status_button->setCheckable(true); + aa_status_button->setChecked(true); + statusBar()->insertPermanentWidget(0, aa_status_button); + + // Setup Filter button + filter_status_button = new QPushButton(); + filter_status_button->setObjectName(QStringLiteral("TogglableStatusBarButton")); + filter_status_button->setFocusPolicy(Qt::NoFocus); + connect(filter_status_button, &QPushButton::clicked, [&] { + auto filter = Settings::values.scaling_filter.GetValue(); + if (filter == Settings::ScalingFilter::LastFilter) { + filter = Settings::ScalingFilter::NearestNeighbor; + } else { + filter = static_cast<Settings::ScalingFilter>(static_cast<u32>(filter) + 1); + } + if (Settings::values.renderer_backend.GetValue() == Settings::RendererBackend::OpenGL && + filter == Settings::ScalingFilter::Fsr) { + filter = Settings::ScalingFilter::NearestNeighbor; + } + Settings::values.scaling_filter.SetValue(filter); + filter_status_button->setChecked(true); + UpdateFilterText(); + }); + auto filter = Settings::values.scaling_filter.GetValue(); + if (Settings::values.renderer_backend.GetValue() == Settings::RendererBackend::OpenGL && + filter == Settings::ScalingFilter::Fsr) { + Settings::values.scaling_filter.SetValue(Settings::ScalingFilter::NearestNeighbor); + } + UpdateFilterText(); + filter_status_button->setCheckable(true); + filter_status_button->setChecked(true); + statusBar()->insertPermanentWidget(0, filter_status_button); + // Setup Dock button dock_status_button = new QPushButton(); dock_status_button->setObjectName(QStringLiteral("TogglableStatusBarButton")); @@ -840,6 +893,11 @@ void GMainWindow::InitializeWidgets() { Settings::values.renderer_backend.SetValue(Settings::RendererBackend::Vulkan); } else { Settings::values.renderer_backend.SetValue(Settings::RendererBackend::OpenGL); + const auto filter = Settings::values.scaling_filter.GetValue(); + if (filter == Settings::ScalingFilter::Fsr) { + Settings::values.scaling_filter.SetValue(Settings::ScalingFilter::NearestNeighbor); + UpdateFilterText(); + } } system->ApplySettings(); @@ -1533,6 +1591,7 @@ void GMainWindow::ShutdownGame() { // Disable status bar updates status_bar_update_timer.stop(); shader_building_label->setVisible(false); + res_scale_label->setVisible(false); emu_speed_label->setVisible(false); game_fps_label->setVisible(false); emu_frametime_label->setVisible(false); @@ -2887,8 +2946,7 @@ void GMainWindow::OnCaptureScreenshot() { } } #endif - render_window->CaptureScreenshot(UISettings::values.screenshot_resolution_factor.GetValue(), - filename); + render_window->CaptureScreenshot(filename); } // TODO: Written 2020-10-01: Remove per-game config migration code when it is irrelevant @@ -2979,6 +3037,11 @@ void GMainWindow::UpdateStatusBar() { shader_building_label->setVisible(false); } + const auto res_info = Settings::values.resolution_info; + const auto res_scale = res_info.up_factor; + res_scale_label->setText( + tr("Scale: %1x", "%1 is the resolution scaling factor").arg(res_scale)); + if (Settings::values.use_speed_limit.GetValue()) { emu_speed_label->setText(tr("Speed: %1% / %2%") .arg(results.emulation_speed * 100.0, 0, 'f', 0) @@ -2994,6 +3057,7 @@ void GMainWindow::UpdateStatusBar() { } emu_frametime_label->setText(tr("Frame: %1 ms").arg(results.frametime * 1000.0, 0, 'f', 2)); + res_scale_label->setVisible(true); emu_speed_label->setVisible(!Settings::values.use_multi_core.GetValue()); game_fps_label->setVisible(true); emu_frametime_label->setVisible(true); @@ -3023,11 +3087,55 @@ void GMainWindow::UpdateGPUAccuracyButton() { } } +void GMainWindow::UpdateFilterText() { + const auto filter = Settings::values.scaling_filter.GetValue(); + switch (filter) { + case Settings::ScalingFilter::NearestNeighbor: + filter_status_button->setText(tr("NEAREST")); + break; + case Settings::ScalingFilter::Bilinear: + filter_status_button->setText(tr("BILINEAR")); + break; + case Settings::ScalingFilter::Bicubic: + filter_status_button->setText(tr("BICUBIC")); + break; + case Settings::ScalingFilter::Gaussian: + filter_status_button->setText(tr("GAUSSIAN")); + break; + case Settings::ScalingFilter::ScaleForce: + filter_status_button->setText(tr("SCALEFORCE")); + break; + case Settings::ScalingFilter::Fsr: + filter_status_button->setText(tr("AMD'S FIDELITYFX SR")); + break; + default: + filter_status_button->setText(tr("BILINEAR")); + break; + } +} + +void GMainWindow::UpdateAAText() { + const auto aa_mode = Settings::values.anti_aliasing.GetValue(); + switch (aa_mode) { + case Settings::AntiAliasing::Fxaa: + aa_status_button->setText(tr("FXAA")); + break; + case Settings::AntiAliasing::None: + aa_status_button->setText(tr("NO AA")); + break; + default: + aa_status_button->setText(tr("FXAA")); + break; + } +} + void GMainWindow::UpdateStatusButtons() { dock_status_button->setChecked(Settings::values.use_docked_mode.GetValue()); renderer_status_button->setChecked(Settings::values.renderer_backend.GetValue() == Settings::RendererBackend::Vulkan); UpdateGPUAccuracyButton(); + UpdateFilterText(); + UpdateAAText(); } void GMainWindow::UpdateUISettings() { diff --git a/src/yuzu/main.h b/src/yuzu/main.h index 981102daa..24633ff2d 100644 --- a/src/yuzu/main.h +++ b/src/yuzu/main.h @@ -302,6 +302,8 @@ private: void MigrateConfigFiles(); void UpdateWindowTitle(std::string_view title_name = {}, std::string_view title_version = {}, std::string_view gpu_vendor = {}); + void UpdateFilterText(); + void UpdateAAText(); void UpdateStatusBar(); void UpdateGPUAccuracyButton(); void UpdateStatusButtons(); @@ -328,6 +330,7 @@ private: // Status bar elements QLabel* message_label = nullptr; QLabel* shader_building_label = nullptr; + QLabel* res_scale_label = nullptr; QLabel* emu_speed_label = nullptr; QLabel* game_fps_label = nullptr; QLabel* emu_frametime_label = nullptr; @@ -335,6 +338,8 @@ private: QPushButton* gpu_accuracy_button = nullptr; QPushButton* renderer_status_button = nullptr; QPushButton* dock_status_button = nullptr; + QPushButton* filter_status_button = nullptr; + QPushButton* aa_status_button = nullptr; QTimer status_bar_update_timer; std::unique_ptr<Config> config; diff --git a/src/yuzu/uisettings.h b/src/yuzu/uisettings.h index cac19452f..936914ef3 100644 --- a/src/yuzu/uisettings.h +++ b/src/yuzu/uisettings.h @@ -68,7 +68,6 @@ struct Values { Settings::BasicSetting<bool> enable_discord_presence{true, "enable_discord_presence"}; Settings::BasicSetting<bool> enable_screenshot_save_as{true, "enable_screenshot_save_as"}; - Settings::BasicSetting<u16> screenshot_resolution_factor{0, "screenshot_resolution_factor"}; QString roms_path; QString symbols_path; diff --git a/src/yuzu_cmd/config.cpp b/src/yuzu_cmd/config.cpp index 0b8fde691..33241ea98 100644 --- a/src/yuzu_cmd/config.cpp +++ b/src/yuzu_cmd/config.cpp @@ -451,6 +451,9 @@ void Config::ReadValues() { ReadSetting("Renderer", Settings::values.disable_shader_loop_safety_checks); ReadSetting("Renderer", Settings::values.vulkan_device); + ReadSetting("Renderer", Settings::values.resolution_setup); + ReadSetting("Renderer", Settings::values.scaling_filter); + ReadSetting("Renderer", Settings::values.anti_aliasing); ReadSetting("Renderer", Settings::values.fullscreen_mode); ReadSetting("Renderer", Settings::values.aspect_ratio); ReadSetting("Renderer", Settings::values.max_anisotropy); diff --git a/src/yuzu_cmd/default_ini.h b/src/yuzu_cmd/default_ini.h index 339dca766..ecdc271a8 100644 --- a/src/yuzu_cmd/default_ini.h +++ b/src/yuzu_cmd/default_ini.h @@ -236,6 +236,29 @@ disable_shader_loop_safety_checks = # Which Vulkan physical device to use (defaults to 0) vulkan_device = +# 0: 0.5x (360p/540p) [EXPERIMENTAL] +# 1: 0.75x (540p/810p) [EXPERIMENTAL] +# 2 (default): 1x (720p/1080p) +# 3: 2x (1440p/2160p) +# 4: 3x (2160p/3240p) +# 5: 4x (2880p/4320p) +# 6: 5x (3600p/5400p) +# 7: 6x (4320p/6480p) +resolution_setup = + +# Pixel filter to use when up- or down-sampling rendered frames. +# 0: Nearest Neighbor +# 1 (default): Bilinear +# 2: Bicubic +# 3: Gaussian +# 4: ScaleForce +# 5: AMD FidelityFX™️ Super Resolution [Vulkan Only] +scaling_filter = + +# Anti-Aliasing (AA) +# 0 (default): None, 1: FXAA +anti_aliasing = + # Whether to use fullscreen or borderless window mode # 0 (Windows default): Borderless window, 1 (All other default): Exclusive fullscreen fullscreen_mode = |