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

Texture Cache: fix memory managment and optimize scaled downloads, uploads.

Browse source
This commit is contained in:
Fernando Sahmkow 2021-10-18 22:56:36 +02:00
parent c2ca55c9d5
commit 3b61de74e6
7 changed files with 57 additions and 28 deletions
Download patch file Download diff file Expand all files Collapse all files

2
src/video_core/texture_cache/image_base.cpp
View file

@ -61,7 +61,7 @@ ImageBase::ImageBase(const ImageInfo& info_, GPUVAddr gpu_addr_, VAddr cpu_addr_
: info{info_}, guest_size_bytes{CalculateGuestSizeInBytes(info)},
unswizzled_size_bytes{CalculateUnswizzledSizeBytes(info)},
converted_size_bytes{CalculateConvertedSizeBytes(info)}, scale_rating{}, scale_tick{},
scale_count{}, gpu_addr{gpu_addr_}, cpu_addr{cpu_addr_},
has_scaled{}, gpu_addr{gpu_addr_}, cpu_addr{cpu_addr_},
cpu_addr_end{cpu_addr + guest_size_bytes}, mip_level_offsets{CalculateMipLevelOffsets(info)} {
if (info.type == ImageType::e3D) {
slice_offsets = CalculateSliceOffsets(info);

6
src/video_core/texture_cache/image_base.h
View file

@ -77,8 +77,8 @@ struct ImageBase {
void CheckBadOverlapState();
void CheckAliasState();
bool HasScaled() {
return scale_count > 0;
bool HasScaled() const {
return has_scaled;
}
ImageInfo info;
@ -88,7 +88,7 @@ struct ImageBase {
u32 converted_size_bytes = 0;
u32 scale_rating = 0;
u64 scale_tick = 0;
u32 scale_count = 0;
bool has_scaled = false;
ImageFlagBits flags = ImageFlagBits::CpuModified;
GPUVAddr gpu_addr = 0;

22
src/video_core/texture_cache/texture_cache.h
View file

@ -60,7 +60,7 @@ TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface&
// On OpenGL we can be more conservatives as the driver takes care.
expected_memory = DEFAULT_EXPECTED_MEMORY + 512_MiB;
critical_memory = DEFAULT_CRITICAL_MEMORY + 1_GiB;
minimum_memory = expected_memory;
minimum_memory = 0;
}
}
@ -1464,16 +1464,6 @@ template <class P>
void TextureCache<P>::TrackImage(ImageBase& image, ImageId image_id) {
ASSERT(False(image.flags & ImageFlagBits::Tracked));
image.flags |= ImageFlagBits::Tracked;
if (image.HasScaled()) {
total_used_memory -= GetScaledImageSizeBytes(image);
}
u64 tentative_size = std::max(image.guest_size_bytes, image.unswizzled_size_bytes);
if ((IsPixelFormatASTC(image.info.format) &&
True(image.flags & ImageFlagBits::AcceleratedUpload)) ||
True(image.flags & ImageFlagBits::Converted)) {
tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format);
}
total_used_memory -= Common::AlignUp(tentative_size, 1024);
if (False(image.flags & ImageFlagBits::Sparse)) {
rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, 1);
return;
@ -1519,6 +1509,16 @@ void TextureCache<P>::UntrackImage(ImageBase& image, ImageId image_id) {
template <class P>
void TextureCache<P>::DeleteImage(ImageId image_id) {
ImageBase& image = slot_images[image_id];
if (image.HasScaled()) {
total_used_memory -= GetScaledImageSizeBytes(image);
}
u64 tentative_size = std::max(image.guest_size_bytes, image.unswizzled_size_bytes);
if ((IsPixelFormatASTC(image.info.format) &&
True(image.flags & ImageFlagBits::AcceleratedUpload)) ||
True(image.flags & ImageFlagBits::Converted)) {
tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format);
}
total_used_memory -= Common::AlignUp(tentative_size, 1024);
const GPUVAddr gpu_addr = image.gpu_addr;
const auto alloc_it = image_allocs_table.find(gpu_addr);
if (alloc_it == image_allocs_table.end()) {