From 6b9cc0ed23b15a1b96b322b03feff2153e44a4a9 Mon Sep 17 00:00:00 2001 From: ameerj <52414509+ameerj@users.noreply.github.com> Date: Mon, 6 Mar 2023 21:16:17 -0500 Subject: Refactor AccelerateDMA code --- src/video_core/texture_cache/texture_cache.h | 86 ++++++++++++++++++----- src/video_core/texture_cache/texture_cache_base.h | 10 ++- 2 files changed, 76 insertions(+), 20 deletions(-) (limited to 'src/video_core/texture_cache') diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 335338434..8e8b9a5e6 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -744,6 +744,25 @@ void TextureCache

::PopAsyncFlushes() { } } +template +ImageId TextureCache

::DmaImageId(const Tegra::DMA::ImageOperand& operand) { + const ImageInfo dst_info(operand); + const ImageId dst_id = FindDMAImage(dst_info, operand.address); + if (!dst_id) { + return NULL_IMAGE_ID; + } + const auto& image = slot_images[dst_id]; + if (False(image.flags & ImageFlagBits::GpuModified)) { + // No need to waste time on an image that's synced with guest + return NULL_IMAGE_ID; + } + const auto base = image.TryFindBase(operand.address); + if (!base) { + return NULL_IMAGE_ID; + } + return dst_id; +} + template bool TextureCache

::IsRescaling() const noexcept { return is_rescaling; @@ -771,6 +790,49 @@ bool TextureCache

::IsRegionGpuModified(VAddr addr, size_t size) { return is_modified; } +template +std::pair::Image*, BufferImageCopy> TextureCache

::DmaBufferImageCopy( + const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& buffer_operand, + const Tegra::DMA::ImageOperand& image_operand, ImageId image_id, bool modifies_image) { + const auto [level, base] = PrepareDmaImage(image_id, image_operand.address, modifies_image); + auto* image = &slot_images[image_id]; + const u32 buffer_size = static_cast(buffer_operand.pitch * buffer_operand.height); + const u32 bpp = VideoCore::Surface::BytesPerBlock(image->info.format); + const auto convert = [old_bpp = image_operand.bytes_per_pixel, bpp](u32 value) { + return (old_bpp * value) / bpp; + }; + const u32 base_x = convert(image_operand.params.origin.x.Value()); + const u32 base_y = image_operand.params.origin.y.Value(); + const u32 length_x = convert(copy_info.length_x); + const u32 length_y = copy_info.length_y; + + const BufferImageCopy copy{ + .buffer_offset = 0, + .buffer_size = buffer_size, + .buffer_row_length = convert(buffer_operand.pitch), + .buffer_image_height = buffer_operand.height, + .image_subresource = + { + .base_level = static_cast(level), + .base_layer = static_cast(base), + .num_layers = 1, + }, + .image_offset = + { + .x = static_cast(base_x), + .y = static_cast(base_y), + .z = 0, + }, + .image_extent = + { + .width = length_x, + .height = length_y, + .depth = 1, + }, + }; + return {image, copy}; +} + template void TextureCache

::RefreshContents(Image& image, ImageId image_id) { if (False(image.flags & ImageFlagBits::CpuModified)) { @@ -1405,26 +1467,14 @@ ImageId TextureCache

::FindDMAImage(const ImageInfo& info, GPUVAddr gpu_addr) } template -std::optional::Image*, std::pair>> -TextureCache

::ObtainImage(const Tegra::DMA::ImageOperand& operand, bool mark_as_modified) { - ImageInfo dst_info(operand); - ImageId dst_id = FindDMAImage(dst_info, operand.address); - if (!dst_id) { - return std::nullopt; - } - auto& image = slot_images[dst_id]; - auto base = image.TryFindBase(operand.address); - if (!base) { - return std::nullopt; - } - if (False(image.flags & ImageFlagBits::GpuModified)) { - // No need to waste time on an image that's synced with guest - return std::nullopt; - } +std::pair TextureCache

::PrepareDmaImage(ImageId dst_id, GPUVAddr base_addr, + bool mark_as_modified) { + const auto& image = slot_images[dst_id]; + const auto base = image.TryFindBase(base_addr); PrepareImage(dst_id, mark_as_modified, false); - auto& new_image = slot_images[dst_id]; + const auto& new_image = slot_images[dst_id]; lru_cache.Touch(new_image.lru_index, frame_tick); - return std::make_pair(&new_image, std::make_pair(base->level, base->layer)); + return std::make_pair(base->level, base->layer); } template diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h index 848a5d9ea..5a5b4179c 100644 --- a/src/video_core/texture_cache/texture_cache_base.h +++ b/src/video_core/texture_cache/texture_cache_base.h @@ -209,8 +209,11 @@ public: /// Pop asynchronous downloads void PopAsyncFlushes(); - [[nodiscard]] std::optional>> ObtainImage( - const Tegra::DMA::ImageOperand& operand, bool mark_as_modified); + [[nodiscard]] ImageId DmaImageId(const Tegra::DMA::ImageOperand& operand); + + [[nodiscard]] std::pair DmaBufferImageCopy( + const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& buffer_operand, + const Tegra::DMA::ImageOperand& image_operand, ImageId image_id, bool modifies_image); /// Return true when a CPU region is modified from the GPU [[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size); @@ -386,6 +389,9 @@ private: /// Returns true if the current clear parameters clear the whole image of a given image view [[nodiscard]] bool IsFullClear(ImageViewId id); + [[nodiscard]] std::pair PrepareDmaImage(ImageId dst_id, GPUVAddr base_addr, + bool mark_as_modified); + bool ImageCanRescale(ImageBase& image); void InvalidateScale(Image& image); bool ScaleUp(Image& image); -- cgit v1.2.3