Mirrors/yuzu
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge pull request #9913 from ameerj/acc-dma-refactor

Browse source 
AccelerateDMA: Refactor Buffer/Image copy code and implement for OGL
This commit is contained in:
Fernando S 2023-03-11 20:04:19 +01:00 • committed by GitHub
commit 49643d8134
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
10 changed files with 207 additions and 259 deletions
Download patch file Download diff file Expand all files Collapse all files

45
src/video_core/renderer_opengl/gl_rasterizer.cpp
View file

@ -63,7 +63,7 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra
buffer_cache(*this, cpu_memory_, buffer_cache_runtime),
shader_cache(*this, emu_window_, device, texture_cache, buffer_cache, program_manager,
state_tracker, gpu.ShaderNotify()),
query_cache(*this), accelerate_dma(buffer_cache),
query_cache(*this), accelerate_dma(buffer_cache, texture_cache),
fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache),
blit_image(program_manager_) {}
@ -1262,7 +1262,8 @@ void RasterizerOpenGL::ReleaseChannel(s32 channel_id) {
query_cache.EraseChannel(channel_id);
}
AccelerateDMA::AccelerateDMA(BufferCache& buffer_cache_) : buffer_cache{buffer_cache_} {}
AccelerateDMA::AccelerateDMA(BufferCache& buffer_cache_, TextureCache& texture_cache_)
: buffer_cache{buffer_cache_}, texture_cache{texture_cache_} {}
bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) {
std::scoped_lock lock{buffer_cache.mutex};
@ -1274,4 +1275,44 @@ bool AccelerateDMA::BufferClear(GPUVAddr src_address, u64 amount, u32 value) {
return buffer_cache.DMAClear(src_address, amount, value);
}
template <bool IS_IMAGE_UPLOAD>
bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info,
const Tegra::DMA::BufferOperand& buffer_operand,
const Tegra::DMA::ImageOperand& image_operand) {
std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
const auto image_id = texture_cache.DmaImageId(image_operand);
if (image_id == VideoCommon::NULL_IMAGE_ID) {
return false;
}
const u32 buffer_size = static_cast<u32>(buffer_operand.pitch * buffer_operand.height);
static constexpr auto sync_info = VideoCommon::ObtainBufferSynchronize::FullSynchronize;
const auto post_op = IS_IMAGE_UPLOAD ? VideoCommon::ObtainBufferOperation::DoNothing
: VideoCommon::ObtainBufferOperation::MarkAsWritten;
const auto [buffer, offset] =
buffer_cache.ObtainBuffer(buffer_operand.address, buffer_size, sync_info, post_op);
const auto [image, copy] = texture_cache.DmaBufferImageCopy(
copy_info, buffer_operand, image_operand, image_id, IS_IMAGE_UPLOAD);
const std::span copy_span{&copy, 1};
if constexpr (IS_IMAGE_UPLOAD) {
image->UploadMemory(buffer->Handle(), offset, copy_span);
} else {
image->DownloadMemory(buffer->Handle(), offset, copy_span);
}
return true;
}
bool AccelerateDMA::ImageToBuffer(const Tegra::DMA::ImageCopy& copy_info,
const Tegra::DMA::ImageOperand& image_operand,
const Tegra::DMA::BufferOperand& buffer_operand) {
return DmaBufferImageCopy<false>(copy_info, buffer_operand, image_operand);
}
bool AccelerateDMA::BufferToImage(const Tegra::DMA::ImageCopy& copy_info,
const Tegra::DMA::BufferOperand& buffer_operand,
const Tegra::DMA::ImageOperand& image_operand) {
return DmaBufferImageCopy<true>(copy_info, buffer_operand, image_operand);
}
} // namespace OpenGL

16
src/video_core/renderer_opengl/gl_rasterizer.h
View file

@ -50,24 +50,26 @@ static_assert(sizeof(BindlessSSBO) * CHAR_BIT == 128);
class AccelerateDMA : public Tegra::Engines::AccelerateDMAInterface {
public:
explicit AccelerateDMA(BufferCache& buffer_cache);
explicit AccelerateDMA(BufferCache& buffer_cache, TextureCache& texture_cache);
bool BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) override;
bool BufferClear(GPUVAddr src_address, u64 amount, u32 value) override;
bool ImageToBuffer(const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::ImageOperand& src,
const Tegra::DMA::BufferOperand& dst) override {
return false;
}
const Tegra::DMA::BufferOperand& dst) override;
bool BufferToImage(const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& src,
const Tegra::DMA::ImageOperand& dst) override {
return false;
}
const Tegra::DMA::ImageOperand& dst) override;
private:
template <bool IS_IMAGE_UPLOAD>
bool DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info,
const Tegra::DMA::BufferOperand& src,
const Tegra::DMA::ImageOperand& dst);
BufferCache& buffer_cache;
TextureCache& texture_cache;
};
class RasterizerOpenGL : public VideoCore::RasterizerAccelerated,

24
src/video_core/renderer_opengl/gl_texture_cache.cpp
View file

@ -763,14 +763,14 @@ Image::Image(const VideoCommon::NullImageParams& params) : VideoCommon::ImageBas
Image::~Image() = default;
void Image::UploadMemory(const ImageBufferMap& map,
void Image::UploadMemory(GLuint buffer_handle, size_t buffer_offset,
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);
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, buffer_handle);
glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, buffer_offset, unswizzled_size_bytes);
glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
@ -789,21 +789,26 @@ void Image::UploadMemory(const ImageBufferMap& map,
current_image_height = copy.buffer_image_height;
glPixelStorei(GL_UNPACK_IMAGE_HEIGHT, current_image_height);
}
CopyBufferToImage(copy, map.offset);
CopyBufferToImage(copy, buffer_offset);
}
if (is_rescaled) {
ScaleUp();
}
}
void Image::DownloadMemory(ImageBufferMap& map,
void Image::UploadMemory(const ImageBufferMap& map,
std::span<const VideoCommon::BufferImageCopy> copies) {
UploadMemory(map.buffer, map.offset, copies);
}
void Image::DownloadMemory(GLuint buffer_handle, size_t buffer_offset,
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);
glBindBuffer(GL_PIXEL_PACK_BUFFER, buffer_handle);
glPixelStorei(GL_PACK_ALIGNMENT, 1);
u32 current_row_length = std::numeric_limits<u32>::max();
@ -821,13 +826,18 @@ void Image::DownloadMemory(ImageBufferMap& map,
current_image_height = copy.buffer_image_height;
glPixelStorei(GL_PACK_IMAGE_HEIGHT, current_image_height);
}
CopyImageToBuffer(copy, map.offset);
CopyImageToBuffer(copy, buffer_offset);
}
if (is_rescaled) {
ScaleUp(true);
}
}
void Image::DownloadMemory(ImageBufferMap& map,
std::span<const VideoCommon::BufferImageCopy> copies) {
DownloadMemory(map.buffer, map.offset, copies);
}
GLuint Image::StorageHandle() noexcept {
switch (info.format) {
case PixelFormat::A8B8G8R8_SRGB:

6
src/video_core/renderer_opengl/gl_texture_cache.h
View file

@ -206,9 +206,15 @@ public:
Image(Image&&) = default;
Image& operator=(Image&&) = default;
void UploadMemory(GLuint buffer_handle, size_t buffer_offset,
std::span<const VideoCommon::BufferImageCopy> copies);
void UploadMemory(const ImageBufferMap& map,
std::span<const VideoCommon::BufferImageCopy> copies);
void DownloadMemory(GLuint buffer_handle, size_t buffer_offset,
std::span<const VideoCommon::BufferImageCopy> copies);
void DownloadMemory(ImageBufferMap& map, std::span<const VideoCommon::BufferImageCopy> copies);
GLuint StorageHandle() noexcept;