aboutsummaryrefslogtreecommitdiff
path: root/src/video_core
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core')
-rw-r--r--src/video_core/buffer_cache/buffer_cache.h46
-rw-r--r--src/video_core/renderer_opengl/gl_texture_cache.cpp14
-rw-r--r--src/video_core/renderer_vulkan/vk_texture_cache.cpp16
-rw-r--r--src/video_core/texture_cache/util.cpp41
-rw-r--r--src/video_core/vulkan_common/vulkan_memory_allocator.cpp8
5 files changed, 64 insertions, 61 deletions
diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h
index 9e90c587c..9b2698fad 100644
--- a/src/video_core/buffer_cache/buffer_cache.h
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -544,7 +544,7 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
it++;
}
- boost::container::small_vector<std::pair<BufferCopy, BufferId>, 1> downloads;
+ boost::container::small_vector<std::pair<BufferCopy, BufferId>, 16> downloads;
u64 total_size_bytes = 0;
u64 largest_copy = 0;
for (const IntervalSet& intervals : committed_ranges) {
@@ -914,6 +914,11 @@ void BufferCache<P>::BindHostGraphicsStorageBuffers(size_t stage) {
const u32 offset = buffer.Offset(binding.cpu_addr);
const bool is_written = ((channel_state->written_storage_buffers[stage] >> index) & 1) != 0;
+
+ if (is_written) {
+ MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, size);
+ }
+
if constexpr (NEEDS_BIND_STORAGE_INDEX) {
runtime.BindStorageBuffer(stage, binding_index, buffer, offset, size, is_written);
++binding_index;
@@ -931,6 +936,11 @@ void BufferCache<P>::BindHostGraphicsTextureBuffers(size_t stage) {
const u32 size = binding.size;
SynchronizeBuffer(buffer, binding.cpu_addr, size);
+ const bool is_written = ((channel_state->written_texture_buffers[stage] >> index) & 1) != 0;
+ if (is_written) {
+ MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, size);
+ }
+
const u32 offset = buffer.Offset(binding.cpu_addr);
const PixelFormat format = binding.format;
if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) {
@@ -962,6 +972,8 @@ void BufferCache<P>::BindHostTransformFeedbackBuffers() {
const u32 size = binding.size;
SynchronizeBuffer(buffer, binding.cpu_addr, size);
+ MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, size);
+
const u32 offset = buffer.Offset(binding.cpu_addr);
host_bindings.buffers.push_back(&buffer);
host_bindings.offsets.push_back(offset);
@@ -1011,6 +1023,11 @@ void BufferCache<P>::BindHostComputeStorageBuffers() {
const u32 offset = buffer.Offset(binding.cpu_addr);
const bool is_written =
((channel_state->written_compute_storage_buffers >> index) & 1) != 0;
+
+ if (is_written) {
+ MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, size);
+ }
+
if constexpr (NEEDS_BIND_STORAGE_INDEX) {
runtime.BindComputeStorageBuffer(binding_index, buffer, offset, size, is_written);
++binding_index;
@@ -1028,6 +1045,12 @@ void BufferCache<P>::BindHostComputeTextureBuffers() {
const u32 size = binding.size;
SynchronizeBuffer(buffer, binding.cpu_addr, size);
+ const bool is_written =
+ ((channel_state->written_compute_texture_buffers >> index) & 1) != 0;
+ if (is_written) {
+ MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, size);
+ }
+
const u32 offset = buffer.Offset(binding.cpu_addr);
const PixelFormat format = binding.format;
if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) {
@@ -1201,16 +1224,11 @@ void BufferCache<P>::UpdateUniformBuffers(size_t stage) {
template <class P>
void BufferCache<P>::UpdateStorageBuffers(size_t stage) {
- const u32 written_mask = channel_state->written_storage_buffers[stage];
ForEachEnabledBit(channel_state->enabled_storage_buffers[stage], [&](u32 index) {
// Resolve buffer
Binding& binding = channel_state->storage_buffers[stage][index];
const BufferId buffer_id = FindBuffer(binding.cpu_addr, binding.size);
binding.buffer_id = buffer_id;
- // Mark buffer as written if needed
- if (((written_mask >> index) & 1) != 0) {
- MarkWrittenBuffer(buffer_id, binding.cpu_addr, binding.size);
- }
});
}
@@ -1219,10 +1237,6 @@ void BufferCache<P>::UpdateTextureBuffers(size_t stage) {
ForEachEnabledBit(channel_state->enabled_texture_buffers[stage], [&](u32 index) {
Binding& binding = channel_state->texture_buffers[stage][index];
binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size);
- // Mark buffer as written if needed
- if (((channel_state->written_texture_buffers[stage] >> index) & 1) != 0) {
- MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size);
- }
});
}
@@ -1252,7 +1266,6 @@ void BufferCache<P>::UpdateTransformFeedbackBuffer(u32 index) {
.size = size,
.buffer_id = buffer_id,
};
- MarkWrittenBuffer(buffer_id, *cpu_addr, size);
}
template <class P>
@@ -1279,10 +1292,6 @@ void BufferCache<P>::UpdateComputeStorageBuffers() {
// Resolve buffer
Binding& binding = channel_state->compute_storage_buffers[index];
binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size);
- // Mark as written if needed
- if (((channel_state->written_compute_storage_buffers >> index) & 1) != 0) {
- MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size);
- }
});
}
@@ -1291,18 +1300,11 @@ void BufferCache<P>::UpdateComputeTextureBuffers() {
ForEachEnabledBit(channel_state->enabled_compute_texture_buffers, [&](u32 index) {
Binding& binding = channel_state->compute_texture_buffers[index];
binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size);
- // Mark as written if needed
- if (((channel_state->written_compute_texture_buffers >> index) & 1) != 0) {
- MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size);
- }
});
}
template <class P>
void BufferCache<P>::MarkWrittenBuffer(BufferId buffer_id, VAddr cpu_addr, u32 size) {
- if (memory_tracker.IsRegionCpuModified(cpu_addr, size)) {
- SynchronizeBuffer(slot_buffers[buffer_id], cpu_addr, size);
- }
memory_tracker.MarkRegionAsGpuModified(cpu_addr, size);
const IntervalType base_interval{cpu_addr, cpu_addr + size};
diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp
index 9cafd2983..512eef575 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp
@@ -1048,6 +1048,10 @@ void Image::Scale(bool up_scale) {
}
bool Image::ScaleUp(bool ignore) {
+ const auto& resolution = runtime->resolution;
+ if (!resolution.active) {
+ return false;
+ }
if (True(flags & ImageFlagBits::Rescaled)) {
return false;
}
@@ -1060,9 +1064,6 @@ bool Image::ScaleUp(bool ignore) {
return false;
}
flags |= ImageFlagBits::Rescaled;
- if (!runtime->resolution.active) {
- return false;
- }
has_scaled = true;
if (ignore) {
current_texture = upscaled_backup.handle;
@@ -1073,13 +1074,14 @@ bool Image::ScaleUp(bool ignore) {
}
bool Image::ScaleDown(bool ignore) {
- if (False(flags & ImageFlagBits::Rescaled)) {
+ const auto& resolution = runtime->resolution;
+ if (!resolution.active) {
return false;
}
- flags &= ~ImageFlagBits::Rescaled;
- if (!runtime->resolution.active) {
+ if (False(flags & ImageFlagBits::Rescaled)) {
return false;
}
+ flags &= ~ImageFlagBits::Rescaled;
if (ignore) {
current_texture = texture.handle;
return true;
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
index 00ab47268..1b8bb9662 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
@@ -1530,15 +1530,15 @@ bool Image::IsRescaled() const noexcept {
}
bool Image::ScaleUp(bool ignore) {
+ const auto& resolution = runtime->resolution;
+ if (!resolution.active) {
+ return false;
+ }
if (True(flags & ImageFlagBits::Rescaled)) {
return false;
}
ASSERT(info.type != ImageType::Linear);
flags |= ImageFlagBits::Rescaled;
- const auto& resolution = runtime->resolution;
- if (!resolution.active) {
- return false;
- }
has_scaled = true;
if (!scaled_image) {
const bool is_2d = info.type == ImageType::e2D;
@@ -1567,15 +1567,15 @@ bool Image::ScaleUp(bool ignore) {
}
bool Image::ScaleDown(bool ignore) {
+ const auto& resolution = runtime->resolution;
+ if (!resolution.active) {
+ return false;
+ }
if (False(flags & ImageFlagBits::Rescaled)) {
return false;
}
ASSERT(info.type != ImageType::Linear);
flags &= ~ImageFlagBits::Rescaled;
- const auto& resolution = runtime->resolution;
- if (!resolution.active) {
- return false;
- }
current_image = *original_image;
if (ignore) {
return true;
diff --git a/src/video_core/texture_cache/util.cpp b/src/video_core/texture_cache/util.cpp
index 0a86ce139..8151cabf0 100644
--- a/src/video_core/texture_cache/util.cpp
+++ b/src/video_core/texture_cache/util.cpp
@@ -68,6 +68,7 @@ struct LevelInfo {
Extent2D tile_size;
u32 bpp_log2;
u32 tile_width_spacing;
+ u32 num_levels;
};
[[nodiscard]] constexpr u32 AdjustTileSize(u32 shift, u32 unit_factor, u32 dimension) {
@@ -118,11 +119,11 @@ template <u32 GOB_EXTENT>
}
[[nodiscard]] constexpr Extent3D AdjustMipBlockSize(Extent3D num_tiles, Extent3D block_size,
- u32 level) {
+ u32 level, u32 num_levels) {
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 = level == 0
+ .depth = level == 0 && num_levels == 1
? block_size.depth
: AdjustMipBlockSize<GOB_SIZE_Z>(num_tiles.depth, block_size.depth, level),
};
@@ -166,13 +167,6 @@ 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),
@@ -257,7 +251,7 @@ template <u32 GOB_EXTENT>
}
[[nodiscard]] constexpr LevelInfo MakeLevelInfo(PixelFormat format, Extent3D size, Extent3D block,
- u32 tile_width_spacing) {
+ u32 tile_width_spacing, u32 num_levels) {
const u32 bytes_per_block = BytesPerBlock(format);
return {
.size =
@@ -270,16 +264,18 @@ template <u32 GOB_EXTENT>
.tile_size = DefaultBlockSize(format),
.bpp_log2 = BytesPerBlockLog2(bytes_per_block),
.tile_width_spacing = tile_width_spacing,
+ .num_levels = num_levels,
};
}
[[nodiscard]] constexpr LevelInfo MakeLevelInfo(const ImageInfo& info) {
- return MakeLevelInfo(info.format, info.size, info.block, info.tile_width_spacing);
+ return MakeLevelInfo(info.format, info.size, info.block, info.tile_width_spacing,
+ info.resources.levels);
}
[[nodiscard]] constexpr u32 CalculateLevelOffset(PixelFormat format, Extent3D size, Extent3D block,
u32 tile_width_spacing, u32 level) {
- const LevelInfo info = MakeLevelInfo(format, size, block, tile_width_spacing);
+ const LevelInfo info = MakeLevelInfo(format, size, block, tile_width_spacing, level);
u32 offset = 0;
for (u32 current_level = 0; current_level < level; ++current_level) {
offset += CalculateLevelSize(info, current_level);
@@ -466,7 +462,7 @@ template <u32 GOB_EXTENT>
};
const u32 bpp_log2 = BytesPerBlockLog2(info.format);
const u32 alignment = StrideAlignment(num_tiles, info.block, bpp_log2, info.tile_width_spacing);
- const Extent3D mip_block = AdjustMipBlockSize(num_tiles, info.block, 0);
+ const Extent3D mip_block = AdjustMipBlockSize(num_tiles, info.block, 0, info.resources.levels);
return Extent3D{
.width = Common::AlignUpLog2(num_tiles.width, alignment),
.height = Common::AlignUpLog2(num_tiles.height, GOB_SIZE_Y_SHIFT + mip_block.height),
@@ -533,7 +529,8 @@ void SwizzleBlockLinearImage(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr
UNIMPLEMENTED_IF(copy.image_extent != level_size);
const Extent3D num_tiles = AdjustTileSize(level_size, tile_size);
- const Extent3D block = AdjustMipBlockSize(num_tiles, level_info.block, level);
+ const Extent3D block =
+ AdjustMipBlockSize(num_tiles, level_info.block, level, level_info.num_levels);
size_t host_offset = copy.buffer_offset;
@@ -698,7 +695,7 @@ u32 CalculateLevelStrideAlignment(const ImageInfo& info, u32 level) {
const Extent2D tile_size = DefaultBlockSize(info.format);
const Extent3D level_size = AdjustMipSize(info.size, level);
const Extent3D num_tiles = AdjustTileSize(level_size, tile_size);
- const Extent3D block = AdjustMipBlockSize(num_tiles, info.block, level);
+ const Extent3D block = AdjustMipBlockSize(num_tiles, info.block, level, info.resources.levels);
const u32 bpp_log2 = BytesPerBlockLog2(info.format);
return StrideAlignment(num_tiles, block, bpp_log2, info.tile_width_spacing);
}
@@ -887,7 +884,8 @@ boost::container::small_vector<BufferImageCopy, 16> UnswizzleImage(Tegra::Memory
.image_extent = level_size,
};
const Extent3D num_tiles = AdjustTileSize(level_size, tile_size);
- const Extent3D block = AdjustMipBlockSize(num_tiles, level_info.block, level);
+ const Extent3D block =
+ AdjustMipBlockSize(num_tiles, level_info.block, level, level_info.num_levels);
const u32 stride_alignment = StrideAlignment(num_tiles, info.block, gob, bpp_log2);
size_t guest_layer_offset = 0;
@@ -1041,7 +1039,7 @@ Extent3D MipBlockSize(const ImageInfo& info, u32 level) {
const Extent2D tile_size = DefaultBlockSize(info.format);
const Extent3D level_size = AdjustMipSize(info.size, level);
const Extent3D num_tiles = AdjustTileSize(level_size, tile_size);
- return AdjustMipBlockSize(num_tiles, level_info.block, level);
+ return AdjustMipBlockSize(num_tiles, level_info.block, level, level_info.num_levels);
}
boost::container::small_vector<SwizzleParameters, 16> FullUploadSwizzles(const ImageInfo& info) {
@@ -1063,7 +1061,8 @@ boost::container::small_vector<SwizzleParameters, 16> FullUploadSwizzles(const I
for (s32 level = 0; level < num_levels; ++level) {
const Extent3D level_size = AdjustMipSize(size, level);
const Extent3D num_tiles = AdjustTileSize(level_size, tile_size);
- const Extent3D block = AdjustMipBlockSize(num_tiles, level_info.block, level);
+ const Extent3D block =
+ AdjustMipBlockSize(num_tiles, level_info.block, level, level_info.num_levels);
params[level] = SwizzleParameters{
.num_tiles = num_tiles,
.block = block,
@@ -1292,11 +1291,11 @@ u32 MapSizeBytes(const ImageBase& image) {
}
}
-static_assert(CalculateLevelSize(LevelInfo{{1920, 1080, 1}, {0, 2, 0}, {1, 1}, 2, 0}, 0) ==
+static_assert(CalculateLevelSize(LevelInfo{{1920, 1080, 1}, {0, 2, 0}, {1, 1}, 2, 0, 1}, 0) ==
0x7f8000);
-static_assert(CalculateLevelSize(LevelInfo{{32, 32, 1}, {0, 0, 4}, {1, 1}, 4, 0}, 0) == 0x40000);
+static_assert(CalculateLevelSize(LevelInfo{{32, 32, 1}, {0, 0, 4}, {1, 1}, 4, 0, 1}, 0) == 0x4000);
-static_assert(CalculateLevelSize(LevelInfo{{128, 8, 1}, {0, 4, 0}, {1, 1}, 4, 0}, 0) == 0x40000);
+static_assert(CalculateLevelSize(LevelInfo{{128, 8, 1}, {0, 4, 0}, {1, 1}, 4, 0, 1}, 0) == 0x4000);
static_assert(CalculateLevelOffset(PixelFormat::R8_SINT, {1920, 1080, 1}, {0, 2, 0}, 0, 7) ==
0x2afc00);
diff --git a/src/video_core/vulkan_common/vulkan_memory_allocator.cpp b/src/video_core/vulkan_common/vulkan_memory_allocator.cpp
index 82767fdf0..8dd1667f3 100644
--- a/src/video_core/vulkan_common/vulkan_memory_allocator.cpp
+++ b/src/video_core/vulkan_common/vulkan_memory_allocator.cpp
@@ -66,9 +66,10 @@ struct Range {
switch (usage) {
case MemoryUsage::Upload:
case MemoryUsage::Stream:
- return VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT;
+ return VMA_ALLOCATION_CREATE_MAPPED_BIT |
+ VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT;
case MemoryUsage::Download:
- return VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT;
+ return VMA_ALLOCATION_CREATE_MAPPED_BIT | VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT;
case MemoryUsage::DeviceLocal:
return {};
}
@@ -252,8 +253,7 @@ vk::Image MemoryAllocator::CreateImage(const VkImageCreateInfo& ci) const {
vk::Buffer MemoryAllocator::CreateBuffer(const VkBufferCreateInfo& ci, MemoryUsage usage) const {
const VmaAllocationCreateInfo alloc_ci = {
- .flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | VMA_ALLOCATION_CREATE_MAPPED_BIT |
- MemoryUsageVmaFlags(usage),
+ .flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage),
.usage = MemoryUsageVma(usage),
.requiredFlags = 0,
.preferredFlags = MemoryUsagePreferedVmaFlags(usage),
generated by cgit v1.2.3 (git 2.25.1) at 2025-12-09 21:02:45 +0000