From 425ab9ef4b982213f4ee0d53196f5474e255374f Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Sun, 17 Oct 2021 18:01:18 +0200 Subject: Texture Cache: Fix downscaling and correct memory comsumption. --- src/video_core/texture_cache/image_base.cpp | 4 +-- src/video_core/texture_cache/image_base.h | 5 ++++ src/video_core/texture_cache/texture_cache.h | 31 +++++++++++++++-------- src/video_core/texture_cache/texture_cache_base.h | 2 +- 4 files changed, 28 insertions(+), 14 deletions(-) (limited to 'src/video_core/texture_cache') diff --git a/src/video_core/texture_cache/image_base.cpp b/src/video_core/texture_cache/image_base.cpp index 1909c9ecb..3db2ec825 100644 --- a/src/video_core/texture_cache/image_base.cpp +++ b/src/video_core/texture_cache/image_base.cpp @@ -60,8 +60,8 @@ namespace { 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{}, gpu_addr{gpu_addr_}, cpu_addr{cpu_addr_}, + converted_size_bytes{CalculateConvertedSizeBytes(info)}, scale_rating{}, scale_tick{}, + scale_count{}, 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); diff --git a/src/video_core/texture_cache/image_base.h b/src/video_core/texture_cache/image_base.h index bab290ac7..cd4b5f636 100644 --- a/src/video_core/texture_cache/image_base.h +++ b/src/video_core/texture_cache/image_base.h @@ -77,6 +77,10 @@ struct ImageBase { void CheckBadOverlapState(); void CheckAliasState(); + bool HasScaled() { + return scale_count > 0; + } + ImageInfo info; u32 guest_size_bytes = 0; @@ -84,6 +88,7 @@ struct ImageBase { u32 converted_size_bytes = 0; u32 scale_rating = 0; u64 scale_tick = 0; + u32 scale_count = 0; ImageFlagBits flags = ImageFlagBits::CpuModified; GPUVAddr gpu_addr = 0; diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index a035d2b18..cf0d33a45 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -854,8 +854,8 @@ void TextureCache
::InvalidateScale(Image& image) {
}
template ::GetScaledImageSizeBytes(Image& image) {
- const f32 add_to_size = Settings::values.resolution_info.up_factor - 1.0f;
+u64 TextureCache ::GetScaledImageSizeBytes(ImageBase& image) {
+ const f32 add_to_size = Settings::values.resolution_info.up_factor;
const bool sign = std::signbit(add_to_size);
const u32 image_size_bytes = std::max(image.guest_size_bytes, image.unswizzled_size_bytes);
const u64 tentative_size = image_size_bytes * static_cast ::GetScaledImageSizeBytes(Image& image) {
template ::ScaleUp(Image& image) {
+ const bool has_copy = image.HasScaled();
const bool rescaled = image.ScaleUp();
if (!rescaled) {
return false;
}
- total_used_memory += GetScaledImageSizeBytes(image);
+ if (!has_copy) {
+ total_used_memory += GetScaledImageSizeBytes(image);
+ }
InvalidateScale(image);
return true;
}
@@ -880,7 +883,10 @@ bool TextureCache ::ScaleDown(Image& image) {
if (!rescaled) {
return false;
}
- total_used_memory -= GetScaledImageSizeBytes(image);
+ const bool has_copy = image.HasScaled();
+ if (!has_copy) {
+ total_used_memory -= GetScaledImageSizeBytes(image);
+ }
InvalidateScale(image);
return true;
}
@@ -1391,13 +1397,6 @@ void TextureCache ::UnregisterImage(ImageId image_id) {
"Trying to unregister an already registered image");
image.flags &= ~ImageFlagBits::Registered;
image.flags &= ~ImageFlagBits::BadOverlap;
- 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);
lru_cache.Free(image.lru_index);
const auto& clear_page_table =
[this, image_id](
@@ -1478,6 +1477,16 @@ template ::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;
diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h
index 4dbe050af..e210393ba 100644
--- a/src/video_core/texture_cache/texture_cache_base.h
+++ b/src/video_core/texture_cache/texture_cache_base.h
@@ -331,7 +331,7 @@ private:
void InvalidateScale(Image& image);
bool ScaleUp(Image& image);
bool ScaleDown(Image& image);
- u64 GetScaledImageSizeBytes(Image& image);
+ u64 GetScaledImageSizeBytes(ImageBase& image);
Runtime& runtime;
VideoCore::RasterizerInterface& rasterizer;
--
cgit v1.2.3