8 files changed, 170 insertions, 47 deletions

diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp
index a55097e5f..197c9f02c 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp
@@ -23,6 +23,7 @@ using VideoCore::MortonSwizzleMode;
 
 using VideoCore::Surface::ComponentType;
 using VideoCore::Surface::PixelFormat;
+using VideoCore::Surface::SurfaceCompression;
 using VideoCore::Surface::SurfaceTarget;
 using VideoCore::Surface::SurfaceType;
 
@@ -242,10 +243,10 @@ void CachedSurface::DownloadTexture(std::vector<u8>& staging_buffer) {
     MICROPROFILE_SCOPE(OpenGL_Texture_Download);
 
     // TODO(Rodrigo): Optimize alignment
-    glPixelStorei(GL_PACK_ALIGNMENT, 1);
     SCOPE_EXIT({ glPixelStorei(GL_PACK_ROW_LENGTH, 0); });
 
     for (u32 level = 0; level < params.num_levels; ++level) {
+        glPixelStorei(GL_PACK_ALIGNMENT, std::min(8U, params.GetRowAlignment(level)));
         glPixelStorei(GL_PACK_ROW_LENGTH, static_cast<GLint>(params.GetMipWidth(level)));
         const std::size_t mip_offset = params.GetHostMipmapLevelOffset(level);
         if (is_compressed) {
@@ -270,10 +271,14 @@ void CachedSurface::UploadTexture(std::vector<u8>& staging_buffer) {
 
 void CachedSurface::UploadTextureMipmap(u32 level, std::vector<u8>& staging_buffer) {
     // TODO(Rodrigo): Optimize alignment
-    glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
+    glPixelStorei(GL_UNPACK_ALIGNMENT, std::min(8U, params.GetRowAlignment(level)));
     glPixelStorei(GL_UNPACK_ROW_LENGTH, static_cast<GLint>(params.GetMipWidth(level)));
 
-    const std::size_t mip_offset = params.GetHostMipmapLevelOffset(level);
+    auto compression_type = params.GetCompressionType();
+
+    const std::size_t mip_offset = compression_type == SurfaceCompression::Converted
+                                       ? params.GetConvertedMipmapOffset(level)
+                                       : params.GetHostMipmapLevelOffset(level);
     u8* buffer{staging_buffer.data() + mip_offset};
     if (is_compressed) {
         const auto image_size{static_cast<GLsizei>(params.GetHostMipmapSize(level))};
diff --git a/src/video_core/surface.h b/src/video_core/surface.h
index 8e98033f3..5d49214e5 100644
--- a/src/video_core/surface.h
+++ b/src/video_core/surface.h
@@ -197,7 +197,7 @@ inline constexpr std::array<u32, MaxPixelFormat> compression_factor_shift_table
  */
 inline constexpr u32 GetCompressionFactorShift(PixelFormat format) {
     DEBUG_ASSERT(format != PixelFormat::Invalid);
-    DEBUG_ASSERT(static_cast<std::size_t>(format) < compression_factor_table.size());
+    DEBUG_ASSERT(static_cast<std::size_t>(format) < compression_factor_shift_table.size());
     return compression_factor_shift_table[static_cast<std::size_t>(format)];
 }
 
@@ -438,6 +438,88 @@ static constexpr u32 GetBytesPerPixel(PixelFormat pixel_format) {
     return GetFormatBpp(pixel_format) / CHAR_BIT;
 }
 
+enum class SurfaceCompression : u8 {
+    None = 0,
+    Compressed = 1,
+    Converted = 2,
+    Rearranged = 3,
+};
+
+inline constexpr std::array<SurfaceCompression, MaxPixelFormat> compression_type_table = {{
+    SurfaceCompression::None,           // ABGR8U
+    SurfaceCompression::None,           // ABGR8S
+    SurfaceCompression::None,           // ABGR8UI
+    SurfaceCompression::None,           // B5G6R5U
+    SurfaceCompression::None,           // A2B10G10R10U
+    SurfaceCompression::None,           // A1B5G5R5U
+    SurfaceCompression::None,           // R8U
+    SurfaceCompression::None,           // R8UI
+    SurfaceCompression::None,           // RGBA16F
+    SurfaceCompression::None,           // RGBA16U
+    SurfaceCompression::None,           // RGBA16UI
+    SurfaceCompression::None,           // R11FG11FB10F
+    SurfaceCompression::None,           // RGBA32UI
+    SurfaceCompression::Compressed,     // DXT1
+    SurfaceCompression::Compressed,     // DXT23
+    SurfaceCompression::Compressed,     // DXT45
+    SurfaceCompression::Compressed,     // DXN1
+    SurfaceCompression::Compressed,     // DXN2UNORM
+    SurfaceCompression::Compressed,     // DXN2SNORM
+    SurfaceCompression::Compressed,     // BC7U
+    SurfaceCompression::Compressed,     // BC6H_UF16
+    SurfaceCompression::Compressed,     // BC6H_SF16
+    SurfaceCompression::Converted,      // ASTC_2D_4X4
+    SurfaceCompression::None,           // BGRA8
+    SurfaceCompression::None,           // RGBA32F
+    SurfaceCompression::None,           // RG32F
+    SurfaceCompression::None,           // R32F
+    SurfaceCompression::None,           // R16F
+    SurfaceCompression::None,           // R16U
+    SurfaceCompression::None,           // R16S
+    SurfaceCompression::None,           // R16UI
+    SurfaceCompression::None,           // R16I
+    SurfaceCompression::None,           // RG16
+    SurfaceCompression::None,           // RG16F
+    SurfaceCompression::None,           // RG16UI
+    SurfaceCompression::None,           // RG16I
+    SurfaceCompression::None,           // RG16S
+    SurfaceCompression::None,           // RGB32F
+    SurfaceCompression::None,           // RGBA8_SRGB
+    SurfaceCompression::None,           // RG8U
+    SurfaceCompression::None,           // RG8S
+    SurfaceCompression::None,           // RG32UI
+    SurfaceCompression::None,           // R32UI
+    SurfaceCompression::Converted,      // ASTC_2D_8X8
+    SurfaceCompression::Converted,      // ASTC_2D_8X5
+    SurfaceCompression::Converted,      // ASTC_2D_5X4
+    SurfaceCompression::None,           // BGRA8_SRGB
+    SurfaceCompression::Compressed,     // DXT1_SRGB
+    SurfaceCompression::Compressed,     // DXT23_SRGB
+    SurfaceCompression::Compressed,     // DXT45_SRGB
+    SurfaceCompression::Compressed,     // BC7U_SRGB
+    SurfaceCompression::Converted,      // ASTC_2D_4X4_SRGB
+    SurfaceCompression::Converted,      // ASTC_2D_8X8_SRGB
+    SurfaceCompression::Converted,      // ASTC_2D_8X5_SRGB
+    SurfaceCompression::Converted,      // ASTC_2D_5X4_SRGB
+    SurfaceCompression::Converted,      // ASTC_2D_5X5
+    SurfaceCompression::Converted,      // ASTC_2D_5X5_SRGB
+    SurfaceCompression::Converted,      // ASTC_2D_10X8
+    SurfaceCompression::Converted,      // ASTC_2D_10X8_SRGB
+    SurfaceCompression::None,           // Z32F
+    SurfaceCompression::None,           // Z16
+    SurfaceCompression::None,           // Z24S8
+    SurfaceCompression::Rearranged,     // S8Z24
+    SurfaceCompression::None,           // Z32FS8
+}};
+
+static constexpr SurfaceCompression GetFormatCompressionType(PixelFormat format) {
+    if (format == PixelFormat::Invalid)
+        return SurfaceCompression::None;
+
+    ASSERT(static_cast<std::size_t>(format) < compression_type_table.size());
+    return compression_type_table[static_cast<std::size_t>(format)];
+}
+
 SurfaceTarget SurfaceTargetFromTextureType(Tegra::Texture::TextureType texture_type);
 
 bool SurfaceTargetIsLayered(SurfaceTarget target);
diff --git a/src/video_core/texture_cache/surface_base.cpp b/src/video_core/texture_cache/surface_base.cpp
index 510d1aef5..ceff51043 100644
--- a/src/video_core/texture_cache/surface_base.cpp
+++ b/src/video_core/texture_cache/surface_base.cpp
@@ -17,6 +17,7 @@ MICROPROFILE_DEFINE(GPU_Flush_Texture, "GPU", "Texture Flush", MP_RGB(128, 192,
 
 using Tegra::Texture::ConvertFromGuestToHost;
 using VideoCore::MortonSwizzleMode;
+using VideoCore::Surface::SurfaceCompression;
 
 SurfaceBaseImpl::SurfaceBaseImpl(GPUVAddr gpu_addr, const SurfaceParams& params)
     : params{params}, mipmap_sizes(params.num_levels),
@@ -102,9 +103,20 @@ void SurfaceBaseImpl::LoadBuffer(Tegra::MemoryManager& memory_manager,
         }
     }
 
-    for (u32 level = 0; level < params.num_levels; ++level) {
-        const std::size_t host_offset{params.GetHostMipmapLevelOffset(level)};
-        ConvertFromGuestToHost(staging_buffer.data() + host_offset, params.pixel_format,
+    auto compression_type = params.GetCompressionType();
+    if (compression_type == SurfaceCompression::None ||
+        compression_type == SurfaceCompression::Compressed)
+        return;
+
+    for (u32 level_up = params.num_levels; level_up > 0; --level_up) {
+        const u32 level = level_up - 1;
+        const std::size_t in_host_offset{params.GetHostMipmapLevelOffset(level)};
+        const std::size_t out_host_offset = compression_type == SurfaceCompression::Rearranged
+                                                ? in_host_offset
+                                                : params.GetConvertedMipmapOffset(level);
+        u8* in_buffer = staging_buffer.data() + in_host_offset;
+        u8* out_buffer = staging_buffer.data() + out_host_offset;
+        ConvertFromGuestToHost(in_buffer, out_buffer, params.pixel_format,
                                params.GetMipWidth(level), params.GetMipHeight(level),
                                params.GetMipDepth(level), true, true);
     }
diff --git a/src/video_core/texture_cache/surface_base.h b/src/video_core/texture_cache/surface_base.h
index 78db2d665..cb7f22706 100644
--- a/src/video_core/texture_cache/surface_base.h
+++ b/src/video_core/texture_cache/surface_base.h
@@ -93,6 +93,10 @@ public:
         return mipmap_sizes[level];
     }
 
+    bool IsLinear() const {
+        return !params.is_tiled;
+    }
+
     bool MatchFormat(VideoCore::Surface::PixelFormat pixel_format) const {
         return params.pixel_format == pixel_format;
     }
diff --git a/src/video_core/texture_cache/surface_params.cpp b/src/video_core/texture_cache/surface_params.cpp
index 8472b69dc..d9d157d02 100644
--- a/src/video_core/texture_cache/surface_params.cpp
+++ b/src/video_core/texture_cache/surface_params.cpp
@@ -76,17 +76,14 @@ SurfaceParams SurfaceParams::CreateForTexture(Core::System& system,
     params.type = GetFormatType(params.pixel_format);
     // TODO: on 1DBuffer we should use the tic info.
     params.target = TextureType2SurfaceTarget(entry.GetType(), entry.IsArray());
-    params.width =
-        Common::AlignBits(config.tic.Width(), GetCompressionFactorShift(params.pixel_format));
-    params.height =
-        Common::AlignBits(config.tic.Height(), GetCompressionFactorShift(params.pixel_format));
+    params.width = config.tic.Width();
+    params.height = config.tic.Height();
     params.depth = config.tic.Depth();
     if (params.target == SurfaceTarget::TextureCubemap ||
         params.target == SurfaceTarget::TextureCubeArray) {
         params.depth *= 6;
     }
     params.pitch = params.is_tiled ? 0 : config.tic.Pitch();
-    params.unaligned_height = config.tic.Height();
     params.num_levels = config.tic.max_mip_level + 1;
     params.is_layered = params.IsLayered();
     return params;
@@ -108,7 +105,6 @@ SurfaceParams SurfaceParams::CreateForDepthBuffer(
     params.type = GetFormatType(params.pixel_format);
     params.width = zeta_width;
     params.height = zeta_height;
-    params.unaligned_height = zeta_height;
     params.target = SurfaceTarget::Texture2D;
     params.depth = 1;
     params.pitch = 0;
@@ -141,7 +137,6 @@ SurfaceParams SurfaceParams::CreateForFramebuffer(Core::System& system, std::siz
     }
     params.height = config.height;
     params.depth = 1;
-    params.unaligned_height = config.height;
     params.target = SurfaceTarget::Texture2D;
     params.num_levels = 1;
     params.is_layered = false;
@@ -164,7 +159,6 @@ SurfaceParams SurfaceParams::CreateForFermiCopySurface(
     params.width = config.width;
     params.height = config.height;
     params.pitch = config.pitch;
-    params.unaligned_height = config.height;
     // TODO(Rodrigo): Try to guess the surface target from depth and layer parameters
     params.target = SurfaceTarget::Texture2D;
     params.depth = 1;
@@ -185,18 +179,18 @@ bool SurfaceParams::IsLayered() const {
     }
 }
 
+// Auto block resizing algorithm from:
+// https://cgit.freedesktop.org/mesa/mesa/tree/src/gallium/drivers/nouveau/nv50/nv50_miptree.c
 u32 SurfaceParams::GetMipBlockHeight(u32 level) const {
-    // Auto block resizing algorithm from:
-    // https://cgit.freedesktop.org/mesa/mesa/tree/src/gallium/drivers/nouveau/nv50/nv50_miptree.c
     if (level == 0) {
         return this->block_height;
     }
 
-    const u32 height{GetMipHeight(level)};
+    const u32 height_new{GetMipHeight(level)};
     const u32 default_block_height{GetDefaultBlockHeight()};
-    const u32 blocks_in_y{(height + default_block_height - 1) / default_block_height};
-    const u32 block_height = Common::Log2Ceil32(blocks_in_y);
-    return std::clamp(block_height, 3U, 8U) - 3U;
+    const u32 blocks_in_y{(height_new + default_block_height - 1) / default_block_height};
+    const u32 block_height_new = Common::Log2Ceil32(blocks_in_y);
+    return std::clamp(block_height_new, 3U, 7U) - 3U;
 }
 
 u32 SurfaceParams::GetMipBlockDepth(u32 level) const {
@@ -207,12 +201,12 @@ u32 SurfaceParams::GetMipBlockDepth(u32 level) const {
         return 0;
     }
 
-    const u32 depth{GetMipDepth(level)};
-    const u32 block_depth = Common::Log2Ceil32(depth);
-    if (block_depth > 4) {
+    const u32 depth_new{GetMipDepth(level)};
+    const u32 block_depth_new = Common::Log2Ceil32(depth_new);
+    if (block_depth_new > 4) {
         return 5 - (GetMipBlockHeight(level) >= 2);
     }
-    return block_depth;
+    return block_depth_new;
 }
 
 std::size_t SurfaceParams::GetGuestMipmapLevelOffset(u32 level) const {
@@ -231,6 +225,14 @@ std::size_t SurfaceParams::GetHostMipmapLevelOffset(u32 level) const {
     return offset;
 }
 
+std::size_t SurfaceParams::GetConvertedMipmapOffset(u32 level) const {
+    std::size_t offset = 0;
+    for (u32 i = 0; i < level; i++) {
+        offset += GetConvertedMipmapSize(i);
+    }
+    return offset;
+}
+
 std::size_t SurfaceParams::GetGuestMipmapSize(u32 level) const {
     return GetInnerMipmapMemorySize(level, false, false);
 }
@@ -239,6 +241,14 @@ std::size_t SurfaceParams::GetHostMipmapSize(u32 level) const {
     return GetInnerMipmapMemorySize(level, true, false) * GetNumLayers();
 }
 
+std::size_t SurfaceParams::GetConvertedMipmapSize(u32 level) const {
+    constexpr std::size_t rgb8_bpp = 4ULL;
+    const std::size_t width_t = GetMipWidth(level);
+    const std::size_t height_t = GetMipHeight(level);
+    const std::size_t depth_t = is_layered ? depth : GetMipDepth(level);
+    return width_t * height_t * depth_t * rgb8_bpp;
+}
+
 std::size_t SurfaceParams::GetGuestLayerSize() const {
     return GetLayerSize(false, false);
 }
@@ -287,12 +297,10 @@ std::size_t SurfaceParams::Hash() const {
 
 bool SurfaceParams::operator==(const SurfaceParams& rhs) const {
     return std::tie(is_tiled, block_width, block_height, block_depth, tile_width_spacing, width,
-                    height, depth, pitch, unaligned_height, num_levels, pixel_format,
-                    component_type, type, target) ==
+                    height, depth, pitch, num_levels, pixel_format, component_type, type, target) ==
            std::tie(rhs.is_tiled, rhs.block_width, rhs.block_height, rhs.block_depth,
                     rhs.tile_width_spacing, rhs.width, rhs.height, rhs.depth, rhs.pitch,
-                    rhs.unaligned_height, rhs.num_levels, rhs.pixel_format, rhs.component_type,
-                    rhs.type, rhs.target);
+                    rhs.num_levels, rhs.pixel_format, rhs.component_type, rhs.type, rhs.target);
 }
 
 std::string SurfaceParams::TargetName() const {
diff --git a/src/video_core/texture_cache/surface_params.h b/src/video_core/texture_cache/surface_params.h
index 7c48782c7..b3082173f 100644
--- a/src/video_core/texture_cache/surface_params.h
+++ b/src/video_core/texture_cache/surface_params.h
@@ -7,6 +7,7 @@
 #include <map>
 
 #include "common/alignment.h"
+#include "common/bit_util.h"
 #include "common/common_types.h"
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/engines/maxwell_3d.h"
@@ -16,6 +17,8 @@
 
 namespace VideoCommon {
 
+using VideoCore::Surface::SurfaceCompression;
+
 class SurfaceParams {
 public:
     /// Creates SurfaceCachedParams from a texture configuration.
@@ -50,17 +53,12 @@ public:
 
     std::size_t GetHostSizeInBytes() const {
         std::size_t host_size_in_bytes;
-        if (IsPixelFormatASTC(pixel_format)) {
+        if (GetCompressionType() == SurfaceCompression::Converted) {
             constexpr std::size_t rgb8_bpp = 4ULL;
             // ASTC is uncompressed in software, in emulated as RGBA8
             host_size_in_bytes = 0;
             for (u32 level = 0; level < num_levels; ++level) {
-                const std::size_t width =
-                    Common::AlignUp(GetMipWidth(level), GetDefaultBlockWidth());
-                const std::size_t height =
-                    Common::AlignUp(GetMipHeight(level), GetDefaultBlockHeight());
-                const std::size_t depth = is_layered ? this->depth : GetMipDepth(level);
-                host_size_in_bytes += width * height * depth * rgb8_bpp;
+                host_size_in_bytes += GetConvertedMipmapSize(level);
             }
         } else {
             host_size_in_bytes = GetInnerMemorySize(true, false, false);
@@ -93,6 +91,12 @@ public:
     /// Returns the block depth of a given mipmap level.
     u32 GetMipBlockDepth(u32 level) const;
 
+    u32 GetRowAlignment(u32 level) const {
+        const u32 bpp =
+            GetCompressionType() == SurfaceCompression::Converted ? 4 : GetBytesPerPixel();
+        return 1U << Common::CountTrailingZeroes32(GetMipWidth(level) * bpp);
+    }
+
     // Helper used for out of class size calculations
     static std::size_t AlignLayered(const std::size_t out_size, const u32 block_height,
                                     const u32 block_depth) {
@@ -106,12 +110,16 @@ public:
     /// Returns the offset in bytes in host memory (linear) of a given mipmap level.
     std::size_t GetHostMipmapLevelOffset(u32 level) const;
 
+    std::size_t GetConvertedMipmapOffset(u32 level) const;
+
     /// Returns the size in bytes in guest memory of a given mipmap level.
     std::size_t GetGuestMipmapSize(u32 level) const;
 
     /// Returns the size in bytes in host memory (linear) of a given mipmap level.
     std::size_t GetHostMipmapSize(u32 level) const;
 
+    std::size_t GetConvertedMipmapSize(u32 level) const;
+
     /// Returns the size of a layer in bytes in guest memory.
     std::size_t GetGuestLayerSize() const;
 
@@ -141,6 +149,10 @@ public:
     /// Returns true if the pixel format is a depth and/or stencil format.
     bool IsPixelFormatZeta() const;
 
+    SurfaceCompression GetCompressionType() const {
+        return VideoCore::Surface::GetFormatCompressionType(pixel_format);
+    }
+
     std::string TargetName() const;
 
     bool is_tiled;
@@ -154,7 +166,6 @@ public:
     u32 height;
     u32 depth;
     u32 pitch;
-    u32 unaligned_height;
     u32 num_levels;
     VideoCore::Surface::PixelFormat pixel_format;
     VideoCore::Surface::ComponentType component_type;
diff --git a/src/video_core/textures/convert.cpp b/src/video_core/textures/convert.cpp
index 82050bd51..f3efa7eb0 100644
--- a/src/video_core/textures/convert.cpp
+++ b/src/video_core/textures/convert.cpp
@@ -62,19 +62,19 @@ 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) {
+void ConvertFromGuestToHost(u8* in_data, u8* out_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);
+        const std::vector<u8> rgba8_data = Tegra::Texture::ASTC::Decompress(
+            in_data, width, height, depth, block_width, block_height);
+        std::copy(rgba8_data.begin(), rgba8_data.end(), out_data);
 
     } else if (convert_s8z24 && pixel_format == PixelFormat::S8Z24) {
-        Tegra::Texture::ConvertS8Z24ToZ24S8(data, width, height);
+        Tegra::Texture::ConvertS8Z24ToZ24S8(in_data, width, height);
     }
 }
 
@@ -90,4 +90,4 @@ void ConvertFromHostToGuest(u8* data, PixelFormat pixel_format, u32 width, u32 h
     }
 }
 
-} // namespace Tegra::Texture
\ No newline at end of file
+} // namespace Tegra::Texture
diff --git a/src/video_core/textures/convert.h b/src/video_core/textures/convert.h
index 12542e71c..d5d6c77bb 100644
--- a/src/video_core/textures/convert.h
+++ b/src/video_core/textures/convert.h
@@ -12,10 +12,11 @@ enum class PixelFormat;
 
 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 ConvertFromGuestToHost(u8* in_data, u8* out_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
+} // namespace Tegra::Texture