2 files changed, 145 insertions, 9 deletions

diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt
index 07b94dcc8..91a30fef7 100644
--- a/src/video_core/CMakeLists.txt
+++ b/src/video_core/CMakeLists.txt
@@ -237,7 +237,6 @@ target_include_directories(video_core PRIVATE ${FFmpeg_INCLUDE_DIR})
 target_link_libraries(video_core PRIVATE ${FFmpeg_LIBRARIES})
 target_link_options(video_core PRIVATE ${FFmpeg_LDFLAGS})
 
-target_link_libraries(video_core PRIVATE ffx-fsr)
 add_dependencies(video_core host_shaders)
 target_include_directories(video_core PRIVATE ${HOST_SHADERS_INCLUDE})
 target_include_directories(video_core PRIVATE sirit ../../externals/Vulkan-Headers/include)
diff --git a/src/video_core/renderer_vulkan/vk_fsr.cpp b/src/video_core/renderer_vulkan/vk_fsr.cpp
index 9288aa7c2..2feaa9d37 100644
--- a/src/video_core/renderer_vulkan/vk_fsr.cpp
+++ b/src/video_core/renderer_vulkan/vk_fsr.cpp
@@ -13,9 +13,146 @@
 #include "video_core/renderer_vulkan/vk_shader_util.h"
 #include "video_core/vulkan_common/vulkan_device.h"
 
-#define A_CPU
-#include <ffx_a.h>
-#include <ffx_fsr1.h>
+// Reimplementations of the constant generating functions in ffx_fsr1.h
+// GCC generated a lot of warnings when using the official header.
+static 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,
+    };
+    auto u = std::bit_cast<u32>(f);
+    u32 i = u >> 23;
+    return base[i] + ((u & 0x7fffff) >> shift[i]);
+}
+
+static u32 AU1_AH2_AF2(f32 a[2]) {
+    return AU1_AH1_AF1(a[0]) + (AU1_AH1_AF1(a[1]) << 16);
+}
+
+static 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] = std::bit_cast<u32>(inputViewportInPixelsX / outputSizeInPixelsX);
+    con0[1] = std::bit_cast<u32>(inputViewportInPixelsY / outputSizeInPixelsY);
+    con0[2] = std::bit_cast<u32>(0.5f * inputViewportInPixelsX / outputSizeInPixelsX - 0.5f);
+    con0[3] = std::bit_cast<u32>(0.5f * inputViewportInPixelsY / outputSizeInPixelsY - 0.5f);
+    con1[0] = std::bit_cast<u32>(1.0f / inputSizeInPixelsX);
+    con1[1] = std::bit_cast<u32>(1.0f / inputSizeInPixelsY);
+    con1[2] = std::bit_cast<u32>(1.0f / inputSizeInPixelsX);
+    con1[3] = std::bit_cast<u32>(-1.0f / inputSizeInPixelsY);
+    con2[0] = std::bit_cast<u32>(-1.0f / inputSizeInPixelsX);
+    con2[1] = std::bit_cast<u32>(2.0f / inputSizeInPixelsY);
+    con2[2] = std::bit_cast<u32>(1.0f / inputSizeInPixelsX);
+    con2[3] = std::bit_cast<u32>(2.0f / inputSizeInPixelsY);
+    con3[0] = std::bit_cast<u32>(0.0f / inputSizeInPixelsX);
+    con3[1] = std::bit_cast<u32>(4.0f / inputSizeInPixelsY);
+    con3[2] = con3[3] = 0;
+}
+
+static 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] = std::bit_cast<u32>(0.5f * inputViewportInPixelsX / outputSizeInPixelsX - 0.5f +
+                                 inputOffsetInPixelsX);
+    con0[3] = std::bit_cast<u32>(0.5f * inputViewportInPixelsY / outputSizeInPixelsY - 0.5f +
+                                 inputOffsetInPixelsY);
+}
+
+static void FsrRcasCon(u32* con, f32 sharpness) {
+    sharpness = std::exp2f(-sharpness);
+    f32 hSharp[2]{sharpness, sharpness};
+    con[0] = std::bit_cast<u32>(sharpness);
+    con[1] = AU1_AH2_AF2(hSharp);
+    con[2] = 0;
+    con[3] = 0;
+}
 
 namespace Vulkan {
 
@@ -62,15 +199,15 @@ VkImageView FSR::Draw(VKScheduler& scheduler, size_t image_index, VkImageView im
 
         cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *easu_pipeline);
 
-        std::array<AU1, 4 * 4> push_constants;
+        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<AF1>(crop_rect.GetWidth()), static_cast<AF1>(crop_rect.GetHeight()),
-            static_cast<AF1>(input_image_extent.width), static_cast<AF1>(input_image_extent.height),
-            static_cast<AF1>(output_size.width), static_cast<AF1>(output_size.height),
-            static_cast<AF1>(crop_rect.left), static_cast<AF1>(crop_rect.top));
+            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);
 
         {