Merge pull request #7349 from ameerj/ogl-convert-image

gl_texture_cache: Implement pixel format conversions for copies

5 files changed, 49 insertions, 28 deletions

diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp
index 2f7d98d8b..6956535e5 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp
@@ -396,6 +396,10 @@ OGLTexture MakeImage(const VideoCommon::ImageInfo& info, GLenum gl_internal_form
     UNREACHABLE_MSG("Invalid image format={}", format);
     return GL_R32UI;
 }
+
+[[nodiscard]] u32 NextPow2(u32 value) {
+    return 1U << (32U - std::countl_zero(value - 1U));
+}
 } // Anonymous namespace
 
 ImageBufferMap::~ImageBufferMap() {
@@ -522,6 +526,12 @@ void TextureCacheRuntime::CopyImage(Image& dst_image, Image& src_image,
     }
 }
 
+void TextureCacheRuntime::ConvertImage(Image& dst, Image& src,
+                                       std::span<const VideoCommon::ImageCopy> copies) {
+    LOG_DEBUG(Render_OpenGL, "Converting {} to {}", src.info.format, dst.info.format);
+    format_conversion_pass.ConvertImage(dst, src, copies);
+}
+
 bool TextureCacheRuntime::CanImageBeCopied(const Image& dst, const Image& src) {
     if (dst.info.type == ImageType::e3D && dst.info.format == PixelFormat::BC4_UNORM) {
         return false;
@@ -538,7 +548,7 @@ void TextureCacheRuntime::EmulateCopyImage(Image& dst, Image& src,
         ASSERT(src.info.type == ImageType::e3D);
         util_shaders.CopyBC4(dst, src, copies);
     } else if (IsPixelFormatBGR(dst.info.format) || IsPixelFormatBGR(src.info.format)) {
-        bgr_copy_pass.CopyBGR(dst, src, copies);
+        format_conversion_pass.ConvertImage(dst, src, copies);
     } else {
         UNREACHABLE();
     }
@@ -1286,35 +1296,37 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM
 
 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};
+void FormatConversionPass::ConvertImage(Image& dst_image, Image& src_image,
+                                        std::span<const VideoCommon::ImageCopy> copies) {
+    const GLenum dst_target = ImageTarget(dst_image.info);
+    const GLenum src_target = ImageTarget(src_image.info);
     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);
+        const auto src_origin = MakeCopyOrigin(copy.src_offset, copy.src_subresource, src_target);
+        const auto dst_origin = MakeCopyOrigin(copy.dst_offset, copy.dst_subresource, dst_target);
+        const auto region = MakeCopyRegion(copy.extent, copy.dst_subresource, dst_target);
+        const u32 copy_size = region.width * region.height * region.depth * img_bpp;
+        if (pbo_size < copy_size) {
+            intermediate_pbo.Create();
+            pbo_size = NextPow2(copy_size);
+            glNamedBufferData(intermediate_pbo.handle, 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,
-                             num_src_layers, src_image.GlFormat(), src_image.GlType(),
-                             static_cast<GLsizei>(bgr_pbo_size), nullptr);
+        glBindBuffer(GL_PIXEL_PACK_BUFFER, intermediate_pbo.handle);
+        glGetTextureSubImage(src_image.Handle(), src_origin.level, src_origin.x, src_origin.y,
+                             src_origin.z, region.width, region.height, region.depth,
+                             src_image.GlFormat(), src_image.GlType(),
+                             static_cast<GLsizei>(pbo_size), nullptr);
 
         // Copy from PBO to destination in desired GL format
         glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
         glPixelStorei(GL_UNPACK_ROW_LENGTH, copy.extent.width);
-        glBindBuffer(GL_PIXEL_UNPACK_BUFFER, bgr_pbo.handle);
-        glTextureSubImage3D(dst_image.Handle(), 0, 0, 0, 0, copy.extent.width, copy.extent.height,
-                            copy.dst_subresource.num_layers, dst_image.GlFormat(),
-                            dst_image.GlType(), nullptr);
+        glBindBuffer(GL_PIXEL_UNPACK_BUFFER, intermediate_pbo.handle);
+        glTextureSubImage3D(dst_image.Handle(), dst_origin.level, dst_origin.x, dst_origin.y,
+                            dst_origin.z, region.width, region.height, region.depth,
+                            dst_image.GlFormat(), dst_image.GlType(), nullptr);
     }
 }
 
diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h
index 1bb762568..578f8d523 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.h
+++ b/src/video_core/renderer_opengl/gl_texture_cache.h
@@ -52,17 +52,17 @@ struct FormatProperties {
     bool is_compressed;
 };
 
-class BGRCopyPass {
+class FormatConversionPass {
 public:
-    BGRCopyPass() = default;
-    ~BGRCopyPass() = default;
+    FormatConversionPass() = default;
+    ~FormatConversionPass() = default;
 
-    void CopyBGR(Image& dst_image, Image& src_image,
-                 std::span<const VideoCommon::ImageCopy> copies);
+    void ConvertImage(Image& dst_image, Image& src_image,
+                      std::span<const VideoCommon::ImageCopy> copies);
 
 private:
-    OGLBuffer bgr_pbo;
-    size_t bgr_pbo_size{};
+    OGLBuffer intermediate_pbo;
+    size_t pbo_size{};
 };
 
 class TextureCacheRuntime {
@@ -86,6 +86,8 @@ public:
 
     void CopyImage(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies);
 
+    void ConvertImage(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies);
+
     void ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view, bool rescaled) {
         UNIMPLEMENTED();
     }
@@ -144,7 +146,7 @@ private:
     const Device& device;
     StateTracker& state_tracker;
     UtilShaders util_shaders;
-    BGRCopyPass bgr_copy_pass;
+    FormatConversionPass format_conversion_pass;
 
     std::array<std::unordered_map<GLenum, FormatProperties>, 3> format_properties;
     bool has_broken_texture_view_formats = false;
@@ -336,6 +338,7 @@ struct TextureCacheParams {
     static constexpr bool FRAMEBUFFER_BLITS = true;
     static constexpr bool HAS_EMULATED_COPIES = true;
     static constexpr bool HAS_DEVICE_MEMORY_INFO = true;
+    static constexpr bool HAS_PIXEL_FORMAT_CONVERSIONS = true;
 
     using Runtime = OpenGL::TextureCacheRuntime;
     using Image = OpenGL::Image;
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h
index ff28b4e96..f5f8f9a74 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.h
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.h
@@ -316,6 +316,7 @@ struct TextureCacheParams {
     static constexpr bool FRAMEBUFFER_BLITS = false;
     static constexpr bool HAS_EMULATED_COPIES = false;
     static constexpr bool HAS_DEVICE_MEMORY_INFO = true;
+    static constexpr bool HAS_PIXEL_FORMAT_CONVERSIONS = false;
 
     using Runtime = Vulkan::TextureCacheRuntime;
     using Image = Vulkan::Image;
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index 4d2874bf2..241f71a91 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -1759,6 +1759,9 @@ void TextureCache<P>::CopyImage(ImageId dst_id, ImageId src_id, std::vector<Imag
     }
     UNIMPLEMENTED_IF(dst.info.type != ImageType::e2D);
     UNIMPLEMENTED_IF(src.info.type != ImageType::e2D);
+    if constexpr (HAS_PIXEL_FORMAT_CONVERSIONS) {
+        return runtime.ConvertImage(dst, src, copies);
+    }
     for (const ImageCopy& copy : copies) {
         UNIMPLEMENTED_IF(copy.dst_subresource.num_layers != 1);
         UNIMPLEMENTED_IF(copy.src_subresource.num_layers != 1);
diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h
index 643ad811c..a9504c0e8 100644
--- a/src/video_core/texture_cache/texture_cache_base.h
+++ b/src/video_core/texture_cache/texture_cache_base.h
@@ -59,6 +59,8 @@ class TextureCache {
     static constexpr bool HAS_EMULATED_COPIES = P::HAS_EMULATED_COPIES;
     /// 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;
+    /// True when the API provides utilities for pixel format conversions.
+    static constexpr bool HAS_PIXEL_FORMAT_CONVERSIONS = P::HAS_PIXEL_FORMAT_CONVERSIONS;
 
     static constexpr u64 DEFAULT_EXPECTED_MEMORY = 1_GiB;
     static constexpr u64 DEFAULT_CRITICAL_MEMORY = 2_GiB;