7 files changed, 175 insertions, 136 deletions

diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt
index 6036d6ed3..b5a327936 100644
--- a/src/video_core/CMakeLists.txt
+++ b/src/video_core/CMakeLists.txt
@@ -94,6 +94,8 @@ add_library(video_core STATIC
     surface.h
     textures/astc.cpp
     textures/astc.h
+    textures/convert.cpp
+    textures/convert.h
     textures/decoders.cpp
     textures/decoders.h
     textures/texture.h
diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
index e6d47ce41..642ccb269 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
@@ -20,7 +20,7 @@
 #include "video_core/renderer_opengl/gl_rasterizer_cache.h"
 #include "video_core/renderer_opengl/utils.h"
 #include "video_core/surface.h"
-#include "video_core/textures/astc.h"
+#include "video_core/textures/convert.h"
 #include "video_core/textures/decoders.h"
 
 namespace OpenGL {
@@ -594,103 +594,6 @@ CachedSurface::CachedSurface(const SurfaceParams& params)
     }
 }
 
-static void ConvertS8Z24ToZ24S8(std::vector<u8>& data, u32 width, u32 height, bool reverse) {
-    union S8Z24 {
-        BitField<0, 24, u32> z24;
-        BitField<24, 8, u32> s8;
-    };
-    static_assert(sizeof(S8Z24) == 4, "S8Z24 is incorrect size");
-
-    union Z24S8 {
-        BitField<0, 8, u32> s8;
-        BitField<8, 24, u32> z24;
-    };
-    static_assert(sizeof(Z24S8) == 4, "Z24S8 is incorrect size");
-
-    S8Z24 s8z24_pixel{};
-    Z24S8 z24s8_pixel{};
-    constexpr auto bpp{GetBytesPerPixel(PixelFormat::S8Z24)};
-    for (std::size_t y = 0; y < height; ++y) {
-        for (std::size_t x = 0; x < width; ++x) {
-            const std::size_t offset{bpp * (y * width + x)};
-            if (reverse) {
-                std::memcpy(&z24s8_pixel, &data[offset], sizeof(Z24S8));
-                s8z24_pixel.s8.Assign(z24s8_pixel.s8);
-                s8z24_pixel.z24.Assign(z24s8_pixel.z24);
-                std::memcpy(&data[offset], &s8z24_pixel, sizeof(S8Z24));
-            } else {
-                std::memcpy(&s8z24_pixel, &data[offset], sizeof(S8Z24));
-                z24s8_pixel.s8.Assign(s8z24_pixel.s8);
-                z24s8_pixel.z24.Assign(s8z24_pixel.z24);
-                std::memcpy(&data[offset], &z24s8_pixel, sizeof(Z24S8));
-            }
-        }
-    }
-}
-
-/**
- * Helper function to perform software conversion (as needed) when loading a buffer from Switch
- * memory. This is for Maxwell pixel formats that cannot be represented as-is in OpenGL or with
- * typical desktop GPUs.
- */
-static void ConvertFormatAsNeeded_LoadGLBuffer(std::vector<u8>& data, PixelFormat pixel_format,
-                                               u32 width, u32 height, u32 depth) {
-    switch (pixel_format) {
-    case PixelFormat::ASTC_2D_4X4:
-    case PixelFormat::ASTC_2D_8X8:
-    case PixelFormat::ASTC_2D_8X5:
-    case PixelFormat::ASTC_2D_5X4:
-    case PixelFormat::ASTC_2D_5X5:
-    case PixelFormat::ASTC_2D_4X4_SRGB:
-    case PixelFormat::ASTC_2D_8X8_SRGB:
-    case PixelFormat::ASTC_2D_8X5_SRGB:
-    case PixelFormat::ASTC_2D_5X4_SRGB:
-    case PixelFormat::ASTC_2D_5X5_SRGB:
-    case PixelFormat::ASTC_2D_10X8:
-    case PixelFormat::ASTC_2D_10X8_SRGB: {
-        // Convert ASTC pixel formats to RGBA8, as most desktop GPUs do not support ASTC.
-        u32 block_width{};
-        u32 block_height{};
-        std::tie(block_width, block_height) = GetASTCBlockSize(pixel_format);
-        data =
-            Tegra::Texture::ASTC::Decompress(data, width, height, depth, block_width, block_height);
-        break;
-    }
-    case PixelFormat::S8Z24:
-        // Convert the S8Z24 depth format to Z24S8, as OpenGL does not support S8Z24.
-        ConvertS8Z24ToZ24S8(data, width, height, false);
-        break;
-    }
-}
-
-/**
- * Helper function to perform software conversion (as needed) when flushing a buffer from OpenGL to
- * Switch memory. This is for Maxwell pixel formats that cannot be represented as-is in OpenGL or
- * with typical desktop GPUs.
- */
-static void ConvertFormatAsNeeded_FlushGLBuffer(std::vector<u8>& data, PixelFormat pixel_format,
-                                                u32 width, u32 height) {
-    switch (pixel_format) {
-    case PixelFormat::ASTC_2D_4X4:
-    case PixelFormat::ASTC_2D_8X8:
-    case PixelFormat::ASTC_2D_4X4_SRGB:
-    case PixelFormat::ASTC_2D_8X8_SRGB:
-    case PixelFormat::ASTC_2D_5X5:
-    case PixelFormat::ASTC_2D_5X5_SRGB:
-    case PixelFormat::ASTC_2D_10X8:
-    case PixelFormat::ASTC_2D_10X8_SRGB: {
-        LOG_CRITICAL(HW_GPU, "Conversion of format {} after texture flushing is not implemented",
-                     static_cast<u32>(pixel_format));
-        UNREACHABLE();
-        break;
-    }
-    case PixelFormat::S8Z24:
-        // Convert the Z24S8 depth format to S8Z24, as OpenGL does not support S8Z24.
-        ConvertS8Z24ToZ24S8(data, width, height, true);
-        break;
-    }
-}
-
 MICROPROFILE_DEFINE(OpenGL_SurfaceLoad, "OpenGL", "Surface Load", MP_RGB(128, 192, 64));
 void CachedSurface::LoadGLBuffer() {
     MICROPROFILE_SCOPE(OpenGL_SurfaceLoad);
@@ -719,8 +622,16 @@ void CachedSurface::LoadGLBuffer() {
         }
     }
     for (u32 i = 0; i < params.max_mip_level; i++) {
-        ConvertFormatAsNeeded_LoadGLBuffer(gl_buffer[i], params.pixel_format, params.MipWidth(i),
-                                           params.MipHeight(i), params.MipDepth(i));
+        const u32 width = params.MipWidth(i);
+        const u32 height = params.MipHeight(i);
+        const u32 depth = params.MipDepth(i);
+        if (VideoCore::Surface::IsPixelFormatASTC(params.pixel_format)) {
+            // Reserve size for RGBA8 conversion
+            constexpr std::size_t rgba_bpp = 4;
+            gl_buffer[i].resize(std::max(gl_buffer[i].size(), width * height * depth * rgba_bpp));
+        }
+        Tegra::Texture::ConvertFromGuestToHost(gl_buffer[i].data(), params.pixel_format, width,
+                                               height, depth, true, true);
     }
 }
 
@@ -743,8 +654,8 @@ void CachedSurface::FlushGLBuffer() {
     glGetTextureImage(texture.handle, 0, tuple.format, tuple.type,
                       static_cast<GLsizei>(gl_buffer[0].size()), gl_buffer[0].data());
     glPixelStorei(GL_PACK_ROW_LENGTH, 0);
-    ConvertFormatAsNeeded_FlushGLBuffer(gl_buffer[0], params.pixel_format, params.width,
-                                        params.height);
+    Tegra::Texture::ConvertFromHostToGuest(gl_buffer[0].data(), params.pixel_format, params.width,
+                                           params.height, params.depth, true, true);
     const u8* const texture_src_data = Memory::GetPointer(params.addr);
     ASSERT(texture_src_data);
     if (params.is_tiled) {
diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp
index bc50a4876..b508d64e9 100644
--- a/src/video_core/textures/astc.cpp
+++ b/src/video_core/textures/astc.cpp
@@ -23,28 +23,12 @@
 
 #include "video_core/textures/astc.h"
 
-class BitStream {
+class InputBitStream {
 public:
-    explicit BitStream(unsigned char* ptr, int nBits = 0, int start_offset = 0)
+    explicit InputBitStream(const unsigned char* ptr, int nBits = 0, int start_offset = 0)
         : m_NumBits(nBits), m_CurByte(ptr), m_NextBit(start_offset % 8) {}
 
-    ~BitStream() = default;
-
-    int GetBitsWritten() const {
-        return m_BitsWritten;
-    }
-
-    void WriteBitsR(unsigned int val, unsigned int nBits) {
-        for (unsigned int i = 0; i < nBits; i++) {
-            WriteBit((val >> (nBits - i - 1)) & 1);
-        }
-    }
-
-    void WriteBits(unsigned int val, unsigned int nBits) {
-        for (unsigned int i = 0; i < nBits; i++) {
-            WriteBit((val >> i) & 1);
-        }
-    }
+    ~InputBitStream() = default;
 
     int GetBitsRead() const {
         return m_BitsRead;
@@ -71,6 +55,38 @@ public:
     }
 
 private:
+    const int m_NumBits;
+    const unsigned char* m_CurByte;
+    int m_NextBit = 0;
+    int m_BitsRead = 0;
+
+    bool done = false;
+};
+
+class OutputBitStream {
+public:
+    explicit OutputBitStream(unsigned char* ptr, int nBits = 0, int start_offset = 0)
+        : m_NumBits(nBits), m_CurByte(ptr), m_NextBit(start_offset % 8) {}
+
+    ~OutputBitStream() = default;
+
+    int GetBitsWritten() const {
+        return m_BitsWritten;
+    }
+
+    void WriteBitsR(unsigned int val, unsigned int nBits) {
+        for (unsigned int i = 0; i < nBits; i++) {
+            WriteBit((val >> (nBits - i - 1)) & 1);
+        }
+    }
+
+    void WriteBits(unsigned int val, unsigned int nBits) {
+        for (unsigned int i = 0; i < nBits; i++) {
+            WriteBit((val >> i) & 1);
+        }
+    }
+
+private:
     void WriteBit(int b) {
 
         if (done)
@@ -238,8 +254,8 @@ public:
     // Fills result with the values that are encoded in the given
     // bitstream. We must know beforehand what the maximum possible
     // value is, and how many values we're decoding.
-    static void DecodeIntegerSequence(std::vector<IntegerEncodedValue>& result, BitStream& bits,
-                                      uint32_t maxRange, uint32_t nValues) {
+    static void DecodeIntegerSequence(std::vector<IntegerEncodedValue>& result,
+                                      InputBitStream& bits, uint32_t maxRange, uint32_t nValues) {
         // Determine encoding parameters
         IntegerEncodedValue val = IntegerEncodedValue::CreateEncoding(maxRange);
 
@@ -267,7 +283,7 @@ public:
     }
 
 private:
-    static void DecodeTritBlock(BitStream& bits, std::vector<IntegerEncodedValue>& result,
+    static void DecodeTritBlock(InputBitStream& bits, std::vector<IntegerEncodedValue>& result,
                                 uint32_t nBitsPerValue) {
         // Implement the algorithm in section C.2.12
         uint32_t m[5];
@@ -327,7 +343,7 @@ private:
         }
     }
 
-    static void DecodeQuintBlock(BitStream& bits, std::vector<IntegerEncodedValue>& result,
+    static void DecodeQuintBlock(InputBitStream& bits, std::vector<IntegerEncodedValue>& result,
                                  uint32_t nBitsPerValue) {
         // Implement the algorithm in section C.2.12
         uint32_t m[3];
@@ -406,7 +422,7 @@ struct TexelWeightParams {
     }
 };
 
-static TexelWeightParams DecodeBlockInfo(BitStream& strm) {
+static TexelWeightParams DecodeBlockInfo(InputBitStream& strm) {
     TexelWeightParams params;
 
     // Read the entire block mode all at once
@@ -605,7 +621,7 @@ static TexelWeightParams DecodeBlockInfo(BitStream& strm) {
     return params;
 }
 
-static void FillVoidExtentLDR(BitStream& strm, uint32_t* const outBuf, uint32_t blockWidth,
+static void FillVoidExtentLDR(InputBitStream& strm, uint32_t* const outBuf, uint32_t blockWidth,
                               uint32_t blockHeight) {
     // Don't actually care about the void extent, just read the bits...
     for (int i = 0; i < 4; ++i) {
@@ -821,7 +837,7 @@ static void DecodeColorValues(uint32_t* out, uint8_t* data, const uint32_t* mode
 
     // We now have enough to decode our integer sequence.
     std::vector<IntegerEncodedValue> decodedColorValues;
-    BitStream colorStream(data);
+    InputBitStream colorStream(data);
     IntegerEncodedValue::DecodeIntegerSequence(decodedColorValues, colorStream, range, nValues);
 
     // Once we have the decoded values, we need to dequantize them to the 0-255 range
@@ -1365,9 +1381,9 @@ static void ComputeEndpoints(Pixel& ep1, Pixel& ep2, const uint32_t*& colorValue
 #undef READ_INT_VALUES
 }
 
-static void DecompressBlock(uint8_t inBuf[16], const uint32_t blockWidth,
+static void DecompressBlock(const uint8_t inBuf[16], const uint32_t blockWidth,
                             const uint32_t blockHeight, uint32_t* outBuf) {
-    BitStream strm(inBuf);
+    InputBitStream strm(inBuf);
     TexelWeightParams weightParams = DecodeBlockInfo(strm);
 
     // Was there an error?
@@ -1421,7 +1437,7 @@ static void DecompressBlock(uint8_t inBuf[16], const uint32_t blockWidth,
     // Define color data.
     uint8_t colorEndpointData[16];
     memset(colorEndpointData, 0, sizeof(colorEndpointData));
-    BitStream colorEndpointStream(colorEndpointData, 16 * 8, 0);
+    OutputBitStream colorEndpointStream(colorEndpointData, 16 * 8, 0);
 
     // Read extra config data...
     uint32_t baseCEM = 0;
@@ -1549,7 +1565,7 @@ static void DecompressBlock(uint8_t inBuf[16], const uint32_t blockWidth,
     memset(texelWeightData + clearByteStart, 0, 16 - clearByteStart);
 
     std::vector<IntegerEncodedValue> texelWeightValues;
-    BitStream weightStream(texelWeightData);
+    InputBitStream weightStream(texelWeightData);
 
     IntegerEncodedValue::DecodeIntegerSequence(texelWeightValues, weightStream,
                                                weightParams.m_MaxWeight,
@@ -1597,7 +1613,7 @@ static void DecompressBlock(uint8_t inBuf[16], const uint32_t blockWidth,
 
 namespace Tegra::Texture::ASTC {
 
-std::vector<uint8_t> Decompress(std::vector<uint8_t>& data, uint32_t width, uint32_t height,
+std::vector<uint8_t> Decompress(const uint8_t* data, uint32_t width, uint32_t height,
                                 uint32_t depth, uint32_t block_width, uint32_t block_height) {
     uint32_t blockIdx = 0;
     std::vector<uint8_t> outData(height * width * depth * 4);
@@ -1605,7 +1621,7 @@ std::vector<uint8_t> Decompress(std::vector<uint8_t>& data, uint32_t width, uint
         for (uint32_t j = 0; j < height; j += block_height) {
             for (uint32_t i = 0; i < width; i += block_width) {
 
-                uint8_t* blockPtr = data.data() + blockIdx * 16;
+                const uint8_t* blockPtr = data + blockIdx * 16;
 
                 // Blocks can be at most 12x12
                 uint32_t uncompData[144];
diff --git a/src/video_core/textures/astc.h b/src/video_core/textures/astc.h
index d419dd025..991cdba72 100644
--- a/src/video_core/textures/astc.h
+++ b/src/video_core/textures/astc.h
@@ -9,7 +9,7 @@
 
 namespace Tegra::Texture::ASTC {
 
-std::vector<uint8_t> Decompress(std::vector<uint8_t>& data, uint32_t width, uint32_t height,
+std::vector<uint8_t> Decompress(const uint8_t* data, uint32_t width, uint32_t height,
                                 uint32_t depth, uint32_t block_width, uint32_t block_height);
 
 } // namespace Tegra::Texture::ASTC
diff --git a/src/video_core/textures/convert.cpp b/src/video_core/textures/convert.cpp
new file mode 100644
index 000000000..5e439f036
--- /dev/null
+++ b/src/video_core/textures/convert.cpp
@@ -0,0 +1,92 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <cstring>
+#include <tuple>
+#include <vector>
+
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "common/logging/log.h"
+#include "video_core/textures/astc.h"
+#include "video_core/textures/convert.h"
+
+namespace Tegra::Texture {
+
+using VideoCore::Surface::PixelFormat;
+
+template <bool reverse>
+void SwapS8Z24ToZ24S8(u8* data, u32 width, u32 height) {
+    union S8Z24 {
+        BitField<0, 24, u32> z24;
+        BitField<24, 8, u32> s8;
+    };
+    static_assert(sizeof(S8Z24) == 4, "S8Z24 is incorrect size");
+
+    union Z24S8 {
+        BitField<0, 8, u32> s8;
+        BitField<8, 24, u32> z24;
+    };
+    static_assert(sizeof(Z24S8) == 4, "Z24S8 is incorrect size");
+
+    S8Z24 s8z24_pixel{};
+    Z24S8 z24s8_pixel{};
+    constexpr auto bpp{
+        VideoCore::Surface::GetBytesPerPixel(VideoCore::Surface::PixelFormat::S8Z24)};
+    for (std::size_t y = 0; y < height; ++y) {
+        for (std::size_t x = 0; x < width; ++x) {
+            const std::size_t offset{bpp * (y * width + x)};
+            if constexpr (reverse) {
+                std::memcpy(&z24s8_pixel, &data[offset], sizeof(Z24S8));
+                s8z24_pixel.s8.Assign(z24s8_pixel.s8);
+                s8z24_pixel.z24.Assign(z24s8_pixel.z24);
+                std::memcpy(&data[offset], &s8z24_pixel, sizeof(S8Z24));
+            } else {
+                std::memcpy(&s8z24_pixel, &data[offset], sizeof(S8Z24));
+                z24s8_pixel.s8.Assign(s8z24_pixel.s8);
+                z24s8_pixel.z24.Assign(s8z24_pixel.z24);
+                std::memcpy(&data[offset], &z24s8_pixel, sizeof(Z24S8));
+            }
+        }
+    }
+}
+
+static void ConvertS8Z24ToZ24S8(u8* data, u32 width, u32 height) {
+    SwapS8Z24ToZ24S8<false>(data, width, height);
+}
+
+static void ConvertZ24S8ToS8Z24(u8* data, u32 width, u32 height) {
+    SwapS8Z24ToZ24S8<true>(data, width, height);
+}
+
+void ConvertFromGuestToHost(u8* data, PixelFormat pixel_format, u32 width, u32 height, u32 depth,
+                            bool convert_astc, bool convert_s8z24) {
+    if (convert_astc && IsPixelFormatASTC(pixel_format)) {
+        // Convert ASTC pixel formats to RGBA8, as most desktop GPUs do not support ASTC.
+        u32 block_width{};
+        u32 block_height{};
+        std::tie(block_width, block_height) = GetASTCBlockSize(pixel_format);
+        const std::vector<u8> rgba8_data =
+            Tegra::Texture::ASTC::Decompress(data, width, height, depth, block_width, block_height);
+        std::copy(rgba8_data.begin(), rgba8_data.end(), data);
+
+    } else if (convert_s8z24 && pixel_format == PixelFormat::S8Z24) {
+        Tegra::Texture::ConvertS8Z24ToZ24S8(data, width, height);
+    }
+}
+
+void ConvertFromHostToGuest(u8* data, PixelFormat pixel_format, u32 width, u32 height, u32 depth,
+                            bool convert_astc, bool convert_s8z24) {
+    if (convert_astc && IsPixelFormatASTC(pixel_format)) {
+        LOG_CRITICAL(HW_GPU, "Conversion of format {} after texture flushing is not implemented",
+                     static_cast<u32>(pixel_format));
+        UNREACHABLE();
+
+    } else if (convert_s8z24 && pixel_format == PixelFormat::S8Z24) {
+        Tegra::Texture::ConvertZ24S8ToS8Z24(data, width, height);
+    }
+}
+
+} // namespace Tegra::Texture
\ No newline at end of file
diff --git a/src/video_core/textures/convert.h b/src/video_core/textures/convert.h
new file mode 100644
index 000000000..07cd8b5da
--- /dev/null
+++ b/src/video_core/textures/convert.h
@@ -0,0 +1,18 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "common/common_types.h"
+#include "video_core/surface.h"
+
+namespace Tegra::Texture {
+
+void ConvertFromGuestToHost(u8* data, VideoCore::Surface::PixelFormat pixel_format, u32 width,
+                            u32 height, u32 depth, bool convert_astc, bool convert_s8z24);
+
+void ConvertFromHostToGuest(u8* data, VideoCore::Surface::PixelFormat pixel_format, u32 width,
+                            u32 height, u32 depth, bool convert_astc, bool convert_s8z24);
+
+} // namespace Tegra::Texture
\ No newline at end of file
diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp
index 150612aed..cad7340f5 100644
--- a/src/video_core/textures/decoders.cpp
+++ b/src/video_core/textures/decoders.cpp
@@ -154,7 +154,7 @@ void SwizzledData(u8* const swizzled_data, u8* const unswizzled_data, const bool
             for (u32 xb = 0; xb < blocks_on_x; xb++) {
                 const u32 x_start = xb * block_x_elements;
                 const u32 x_end = std::min(width, x_start + block_x_elements);
-                if (fast) {
+                if constexpr (fast) {
                     FastProcessBlock(swizzled_data, unswizzled_data, unswizzle, x_start, y_start,
                                      z_start, x_end, y_end, z_end, tile_offset, xy_block_size,
                                      layer_z, stride_x, bytes_per_pixel, out_bytes_per_pixel);