hdkv/yuzu
1
Fork 0
Code Issues Pull Requests Releases 1 Activity

early-access version 3568

Browse Source
This commit is contained in:
pineappleEA
2023-05-08 02:44:49 +02:00
parent acf7b81b9a
commit 7ae07b908c
13 changed files with 39 additions and 51 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
src/video_core/buffer_cache/buffer_cache.h
View File

@@ -203,11 +203,8 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
const VAddr new_base_address = *cpu_dest_address + diff;
const IntervalType add_interval{new_base_address, new_base_address + size};
tmp_intervals.push_back(add_interval);
if (!Settings::values.use_reactive_flushing.GetValue() ||
memory_tracker.IsRegionPreflushable(new_base_address, new_base_address + size)) {
uncommitted_ranges.add(add_interval);
pending_ranges.add(add_interval);
}
uncommitted_ranges.add(add_interval);
pending_ranges.add(add_interval);
};
ForEachInRangeSet(common_ranges, *cpu_src_address, amount, mirror);
// This subtraction in this order is important for overlapping copies.
@@ -1234,10 +1231,6 @@ void BufferCache<P>::MarkWrittenBuffer(BufferId buffer_id, VAddr cpu_addr, u32 s
const IntervalType base_interval{cpu_addr, cpu_addr + size};
common_ranges.add(base_interval);
if (Settings::values.use_reactive_flushing.GetValue() &&
!memory_tracker.IsRegionPreflushable(cpu_addr, cpu_addr + size)) {
return;
}
uncommitted_ranges.add(base_interval);
pending_ranges.add(base_interval);
}

2
src/video_core/engines/maxwell_dma.cpp
View File

@@ -288,7 +288,7 @@ void MaxwellDMA::CopyPitchToBlockLinear() {
write_buffer.resize_destructive(dst_size);
memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size);
memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size);
memory_manager.ReadBlockUnsafe(regs.offset_out, write_buffer.data(), dst_size);
// If the input is linear and the output is tiled, swizzle the input and copy it over.
SwizzleSubrect(write_buffer, read_buffer, bytes_per_pixel, width, height, depth, x_offset,

2
src/video_core/renderer_opengl/gl_rasterizer.cpp
View File

@@ -1304,7 +1304,7 @@ 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);
const auto image_id = texture_cache.DmaImageId(image_operand, IS_IMAGE_UPLOAD);
if (image_id == VideoCommon::NULL_IMAGE_ID) {
return false;
}

9
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View File

@@ -510,13 +510,6 @@ VideoCore::RasterizerDownloadArea RasterizerVulkan::GetFlushArea(VAddr addr, u64
return *area;
}
}
{
std::scoped_lock lock{buffer_cache.mutex};
auto area = buffer_cache.GetFlushArea(addr, size);
if (area) {
return *area;
}
}
VideoCore::RasterizerDownloadArea new_area{
.start_address = Common::AlignDown(addr, Core::Memory::YUZU_PAGESIZE),
.end_address = Common::AlignUp(addr + size, Core::Memory::YUZU_PAGESIZE),
@@ -800,7 +793,7 @@ 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);
const auto image_id = texture_cache.DmaImageId(image_operand, IS_IMAGE_UPLOAD);
if (image_id == VideoCommon::NULL_IMAGE_ID) {
return false;
}

12
src/video_core/texture_cache/texture_cache.h
View File

@@ -811,7 +811,7 @@ void TextureCache<P>::PopAsyncFlushes() {
}
template <class P>
ImageId TextureCache<P>::DmaImageId(const Tegra::DMA::ImageOperand& operand) {
ImageId TextureCache<P>::DmaImageId(const Tegra::DMA::ImageOperand& operand, bool is_upload) {
const ImageInfo dst_info(operand);
const ImageId dst_id = FindDMAImage(dst_info, operand.address);
if (!dst_id) {
@@ -822,7 +822,7 @@ ImageId TextureCache<P>::DmaImageId(const Tegra::DMA::ImageOperand& operand) {
// No need to waste time on an image that's synced with guest
return NULL_IMAGE_ID;
}
if (!image.info.dma_downloaded) {
if (!is_upload && !image.info.dma_downloaded) {
// Force a full sync.
image.info.dma_downloaded = true;
return NULL_IMAGE_ID;
@@ -1323,7 +1323,6 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
all_siblings.push_back(overlap_id);
} else {
bad_overlap_ids.push_back(overlap_id);
overlap.flags |= ImageFlagBits::BadOverlap;
}
};
ForEachImageInRegion(cpu_addr, size_bytes, region_check);
@@ -1434,7 +1433,12 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
ImageBase& aliased = slot_images[aliased_id];
aliased.overlapping_images.push_back(new_image_id);
new_image.overlapping_images.push_back(aliased_id);
new_image.flags |= ImageFlagBits::BadOverlap;
if (aliased.info.resources.levels == 1 && aliased.overlapping_images.size() > 1) {
aliased.flags |= ImageFlagBits::BadOverlap;
}
if (new_image.info.resources.levels == 1 && new_image.overlapping_images.size() > 1) {
new_image.flags |= ImageFlagBits::BadOverlap;
}
}
RegisterImage(new_image_id);
return new_image_id;

2
src/video_core/texture_cache/texture_cache_base.h
View File

@@ -207,7 +207,7 @@ public:
/// Pop asynchronous downloads
void PopAsyncFlushes();
[[nodiscard]] ImageId DmaImageId(const Tegra::DMA::ImageOperand& operand);
[[nodiscard]] ImageId DmaImageId(const Tegra::DMA::ImageOperand& operand, bool is_upload);
[[nodiscard]] std::pair<Image*, BufferImageCopy> DmaBufferImageCopy(
const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& buffer_operand,