aboutsummaryrefslogtreecommitdiff
path: root/src/video_core/texture_cache
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core/texture_cache')
-rw-r--r--src/video_core/texture_cache/texture_cache.h53
-rw-r--r--src/video_core/texture_cache/texture_cache_base.h4
-rw-r--r--src/video_core/texture_cache/util.cpp92
3 files changed, 125 insertions, 24 deletions
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index b24086fce..fe13cac93 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -49,8 +49,8 @@ TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface&
if constexpr (HAS_DEVICE_MEMORY_INFO) {
const s64 device_memory = static_cast<s64>(runtime.GetDeviceLocalMemory());
- const s64 min_spacing_expected = device_memory - 1_GiB - 512_MiB;
- const s64 min_spacing_critical = device_memory - 1_GiB;
+ const s64 min_spacing_expected = device_memory - 1_GiB;
+ const s64 min_spacing_critical = device_memory - 512_MiB;
const s64 mem_threshold = std::min(device_memory, TARGET_THRESHOLD);
const s64 min_vacancy_expected = (6 * mem_threshold) / 10;
const s64 min_vacancy_critical = (3 * mem_threshold) / 10;
@@ -86,10 +86,12 @@ void TextureCache<P>::RunGarbageCollector() {
// used by the async decoder thread.
return false;
}
+ if (!aggressive_mode && True(image.flags & ImageFlagBits::CostlyLoad)) {
+ return false;
+ }
const bool must_download =
image.IsSafeDownload() && False(image.flags & ImageFlagBits::BadOverlap);
- if (!high_priority_mode &&
- (must_download || True(image.flags & ImageFlagBits::CostlyLoad))) {
+ if (!high_priority_mode && must_download) {
return false;
}
if (must_download) {
@@ -137,7 +139,6 @@ void TextureCache<P>::TickFrame() {
TickAsyncDecode();
runtime.TickFrame();
- critical_gc = 0;
++frame_tick;
if constexpr (IMPLEMENTS_ASYNC_DOWNLOADS) {
@@ -184,6 +185,42 @@ void TextureCache<P>::FillComputeImageViews(std::span<ImageViewInOut> views) {
}
template <class P>
+void TextureCache<P>::CheckFeedbackLoop(std::span<const ImageViewInOut> views) {
+ const bool requires_barrier = [&] {
+ for (const auto& view : views) {
+ if (!view.id) {
+ continue;
+ }
+ auto& image_view = slot_image_views[view.id];
+
+ // Check color targets
+ for (const auto& ct_view_id : render_targets.color_buffer_ids) {
+ if (ct_view_id) {
+ auto& ct_view = slot_image_views[ct_view_id];
+ if (image_view.image_id == ct_view.image_id) {
+ return true;
+ }
+ }
+ }
+
+ // Check zeta target
+ if (render_targets.depth_buffer_id) {
+ auto& zt_view = slot_image_views[render_targets.depth_buffer_id];
+ if (image_view.image_id == zt_view.image_id) {
+ return true;
+ }
+ }
+ }
+
+ return false;
+ }();
+
+ if (requires_barrier) {
+ runtime.BarrierFeedbackLoop();
+ }
+}
+
+template <class P>
typename P::Sampler* TextureCache<P>::GetGraphicsSampler(u32 index) {
if (index > channel_state->graphics_sampler_table.Limit()) {
LOG_DEBUG(HW_GPU, "Invalid sampler index={}", index);
@@ -1469,7 +1506,7 @@ std::optional<typename TextureCache<P>::BlitImages> TextureCache<P>::GetBlitImag
if (!copy.must_accelerate) {
do {
if (!src_id && !dst_id) {
- break;
+ return std::nullopt;
}
if (src_id && True(slot_images[src_id].flags & ImageFlagBits::GpuModified)) {
break;
@@ -1847,10 +1884,6 @@ void TextureCache<P>::RegisterImage(ImageId image_id) {
tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format);
}
total_used_memory += Common::AlignUp(tentative_size, 1024);
- if (total_used_memory > critical_memory && critical_gc < GC_EMERGENCY_COUNTS) {
- RunGarbageCollector();
- critical_gc++;
- }
image.lru_index = lru_cache.Insert(image_id, frame_tick);
ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, [this, image_id](u64 page) {
diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h
index 0720494e5..cc27286f7 100644
--- a/src/video_core/texture_cache/texture_cache_base.h
+++ b/src/video_core/texture_cache/texture_cache_base.h
@@ -148,6 +148,9 @@ public:
/// Fill image_view_ids with the compute images in indices
void FillComputeImageViews(std::span<ImageViewInOut> views);
+ /// Handle feedback loops during draws.
+ void CheckFeedbackLoop(std::span<const ImageViewInOut> views);
+
/// Get the sampler from the graphics descriptor table in the specified index
Sampler* GetGraphicsSampler(u32 index);
@@ -424,7 +427,6 @@ private:
u64 minimum_memory;
u64 expected_memory;
u64 critical_memory;
- size_t critical_gc;
struct BufferDownload {
GPUVAddr address;
diff --git a/src/video_core/texture_cache/util.cpp b/src/video_core/texture_cache/util.cpp
index f1071aa23..95a5b47d8 100644
--- a/src/video_core/texture_cache/util.cpp
+++ b/src/video_core/texture_cache/util.cpp
@@ -18,6 +18,8 @@
#include "common/bit_util.h"
#include "common/common_types.h"
#include "common/div_ceil.h"
+#include "common/scratch_buffer.h"
+#include "common/settings.h"
#include "video_core/compatible_formats.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/memory_manager.h"
@@ -28,6 +30,7 @@
#include "video_core/texture_cache/samples_helper.h"
#include "video_core/texture_cache/util.h"
#include "video_core/textures/astc.h"
+#include "video_core/textures/bcn.h"
#include "video_core/textures/decoders.h"
namespace VideoCommon {
@@ -120,7 +123,9 @@ template <u32 GOB_EXTENT>
return {
.width = AdjustMipBlockSize<GOB_SIZE_X>(num_tiles.width, block_size.width, level),
.height = AdjustMipBlockSize<GOB_SIZE_Y>(num_tiles.height, block_size.height, level),
- .depth = AdjustMipBlockSize<GOB_SIZE_Z>(num_tiles.depth, block_size.depth, level),
+ .depth = level == 0
+ ? block_size.depth
+ : AdjustMipBlockSize<GOB_SIZE_Z>(num_tiles.depth, block_size.depth, level),
};
}
@@ -162,6 +167,13 @@ template <u32 GOB_EXTENT>
}
[[nodiscard]] constexpr Extent3D TileShift(const LevelInfo& info, u32 level) {
+ if (level == 0) {
+ return Extent3D{
+ .width = info.block.width,
+ .height = info.block.height,
+ .depth = info.block.depth,
+ };
+ }
const Extent3D blocks = NumLevelBlocks(info, level);
return Extent3D{
.width = AdjustTileSize(info.block.width, GOB_SIZE_X, blocks.width),
@@ -585,6 +597,21 @@ u32 CalculateConvertedSizeBytes(const ImageInfo& info) noexcept {
return info.size.width * BytesPerBlock(info.format);
}
static constexpr Extent2D TILE_SIZE{1, 1};
+ if (IsPixelFormatASTC(info.format) && Settings::values.astc_recompression.GetValue() !=
+ Settings::AstcRecompression::Uncompressed) {
+ const u32 bpp_div =
+ Settings::values.astc_recompression.GetValue() == Settings::AstcRecompression::Bc1 ? 2
+ : 1;
+ // NumBlocksPerLayer doesn't account for this correctly, so we have to do it manually.
+ u32 output_size = 0;
+ for (s32 i = 0; i < info.resources.levels; i++) {
+ const auto mip_size = AdjustMipSize(info.size, i);
+ const u32 plane_dim =
+ Common::AlignUp(mip_size.width, 4U) * Common::AlignUp(mip_size.height, 4U);
+ output_size += (plane_dim * info.size.depth * info.resources.layers) / bpp_div;
+ }
+ return output_size;
+ }
return NumBlocksPerLayer(info, TILE_SIZE) * info.resources.layers * CONVERTED_BYTES_PER_BLOCK;
}
@@ -885,6 +912,7 @@ BufferCopy UploadBufferCopy(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr,
void ConvertImage(std::span<const u8> input, const ImageInfo& info, std::span<u8> output,
std::span<BufferImageCopy> copies) {
u32 output_offset = 0;
+ Common::ScratchBuffer<u8> decode_scratch;
const Extent2D tile_size = DefaultBlockSize(info.format);
for (BufferImageCopy& copy : copies) {
@@ -895,22 +923,58 @@ void ConvertImage(std::span<const u8> input, const ImageInfo& info, std::span<u8
ASSERT(copy.image_extent == mip_size);
ASSERT(copy.buffer_row_length == Common::AlignUp(mip_size.width, tile_size.width));
ASSERT(copy.buffer_image_height == Common::AlignUp(mip_size.height, tile_size.height));
- if (IsPixelFormatASTC(info.format)) {
+
+ const auto input_offset = input.subspan(copy.buffer_offset);
+ copy.buffer_offset = output_offset;
+ copy.buffer_row_length = mip_size.width;
+ copy.buffer_image_height = mip_size.height;
+
+ const auto recompression_setting = Settings::values.astc_recompression.GetValue();
+ const bool astc = IsPixelFormatASTC(info.format);
+
+ if (astc && recompression_setting == Settings::AstcRecompression::Uncompressed) {
Tegra::Texture::ASTC::Decompress(
- input.subspan(copy.buffer_offset), copy.image_extent.width,
- copy.image_extent.height,
+ input_offset, copy.image_extent.width, copy.image_extent.height,
copy.image_subresource.num_layers * copy.image_extent.depth, tile_size.width,
tile_size.height, output.subspan(output_offset));
+
+ output_offset += copy.image_extent.width * copy.image_extent.height *
+ copy.image_subresource.num_layers * CONVERTED_BYTES_PER_BLOCK;
+ } else if (astc) {
+ // BC1 uses 0.5 bytes per texel
+ // BC3 uses 1 byte per texel
+ const auto compress = recompression_setting == Settings::AstcRecompression::Bc1
+ ? Tegra::Texture::BCN::CompressBC1
+ : Tegra::Texture::BCN::CompressBC3;
+ const auto bpp_div = recompression_setting == Settings::AstcRecompression::Bc1 ? 2 : 1;
+
+ const u32 plane_dim = copy.image_extent.width * copy.image_extent.height;
+ const u32 level_size = plane_dim * copy.image_extent.depth *
+ copy.image_subresource.num_layers * CONVERTED_BYTES_PER_BLOCK;
+ decode_scratch.resize_destructive(level_size);
+
+ Tegra::Texture::ASTC::Decompress(
+ input_offset, copy.image_extent.width, copy.image_extent.height,
+ copy.image_subresource.num_layers * copy.image_extent.depth, tile_size.width,
+ tile_size.height, decode_scratch);
+
+ compress(decode_scratch, copy.image_extent.width, copy.image_extent.height,
+ copy.image_subresource.num_layers * copy.image_extent.depth,
+ output.subspan(output_offset));
+
+ const u32 aligned_plane_dim = Common::AlignUp(copy.image_extent.width, 4) *
+ Common::AlignUp(copy.image_extent.height, 4);
+
+ copy.buffer_size =
+ (aligned_plane_dim * copy.image_extent.depth * copy.image_subresource.num_layers) /
+ bpp_div;
+ output_offset += static_cast<u32>(copy.buffer_size);
} else {
- DecompressBC4(input.subspan(copy.buffer_offset), copy.image_extent,
- output.subspan(output_offset));
- }
- copy.buffer_offset = output_offset;
- copy.buffer_row_length = mip_size.width;
- copy.buffer_image_height = mip_size.height;
+ DecompressBC4(input_offset, copy.image_extent, output.subspan(output_offset));
- output_offset += copy.image_extent.width * copy.image_extent.height *
- copy.image_subresource.num_layers * CONVERTED_BYTES_PER_BLOCK;
+ output_offset += copy.image_extent.width * copy.image_extent.height *
+ copy.image_subresource.num_layers * CONVERTED_BYTES_PER_BLOCK;
+ }
}
}
@@ -1233,7 +1297,9 @@ u32 MapSizeBytes(const ImageBase& image) {
static_assert(CalculateLevelSize(LevelInfo{{1920, 1080, 1}, {0, 2, 0}, {1, 1}, 2, 0}, 0) ==
0x7f8000);
-static_assert(CalculateLevelSize(LevelInfo{{32, 32, 1}, {0, 0, 4}, {1, 1}, 4, 0}, 0) == 0x4000);
+static_assert(CalculateLevelSize(LevelInfo{{32, 32, 1}, {0, 0, 4}, {1, 1}, 4, 0}, 0) == 0x40000);
+
+static_assert(CalculateLevelSize(LevelInfo{{128, 8, 1}, {0, 4, 0}, {1, 1}, 4, 0}, 0) == 0x40000);
static_assert(CalculateLevelOffset(PixelFormat::R8_SINT, {1920, 1080, 1}, {0, 2, 0}, 0, 7) ==
0x2afc00);
generated by cgit v1.2.3 (git 2.39.1) at 2026-09-04 03:52:35 +0000