Catcrafts/Crafter.Graphics
2
0
Fork 0
Code Issues 1 Pull requests Projects Releases Packages Wiki Activity Actions

perf(font): upload only the dirty sub-rect of the glyph atlas (#51)

Browse source 
FontAtlas::Update re-uploaded the entire 1024x1024 R8 atlas (1 MiB)
whenever a single glyph was rasterized, so warm-up / language-switch /
size-churn cost roughly one full-atlas copy per frame that added a
codepoint.

Accumulate a dirty bounding box (FontAtlas::DirtyRect) in MarkDirty as
glyphs are placed, and copy only that sub-rect. On Vulkan this is a new
ImageVulkan::UpdateRegion: the staging buffer keeps the full atlas row
stride, so bufferRowLength stays the atlas width and bufferOffset points
at the rect origin while imageOffset/imageExtent carve out the rect.
Layout transitions stay whole-image (cheap); only the copy extent
shrinks. WebGPU passes the rect straight to wgpuWriteAtlasRegion, which
already took x/y/w/h.

The freshly-zeroed atlas marks its whole extent dirty in Initialize so
the first Update clears the image once; thereafter every untouched texel
stays the zero it was uploaded as, keeping the image byte-identical to
the staging copy.

Tested: new FontAtlasDirtyRect unit test covers the accumulation /
clamp / lockstep-with-`dirty` math (no GPU needed); HelloUI renders text
correctly on both Vulkan (NVIDIA, validation-clean) and WebGPU.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
catbot 2026-06-16 15:40:37 +00:00
commit 97c79c23ee
5 changed files with 276 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

28
interfaces/Crafter.Graphics-ImageVulkan.cppm
View file

@ -162,6 +162,34 @@ export namespace Crafter {
}
}
// Upload only the sub-rectangle [x, x+w) × [y, y+h) of the staging
// buffer into the image, instead of the whole extent. The staging
// buffer keeps the full image row stride, so bufferRowLength stays
// `width` and bufferOffset just points at the rect's first texel —
// imageOffset/imageExtent then carve out the rect on the GPU side.
// Layout transitions stay whole-image (cheap); only the copy extent
// shrinks. Single mip level only: the partial-upload path is for
// CPU-streamed atlases (FontAtlas) which carry no mips, so there is
// no blit chain to regenerate.
void UpdateRegion(VkCommandBuffer cmd, VkImageLayout layout, std::uint32_t x, std::uint32_t y, std::uint32_t w, std::uint32_t h) {
buffer.FlushDevice(cmd, VK_ACCESS_MEMORY_READ_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT);
TransitionImageLayout(cmd, image, layout, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_PIPELINE_STAGE_RAY_TRACING_SHADER_BIT_KHR, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_ACCESS_SHADER_READ_BIT, VK_ACCESS_TRANSFER_WRITE_BIT, 0, mipLevels);
VkBufferImageCopy region{};
region.bufferOffset = (static_cast<VkDeviceSize>(y) * width + x) * sizeof(PixelType);
region.bufferRowLength = width;
region.bufferImageHeight = 0;
region.imageSubresource.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
region.imageSubresource.mipLevel = 0;
region.imageSubresource.baseArrayLayer = 0;
region.imageSubresource.layerCount = 1;
region.imageOffset = { static_cast<std::int32_t>(x), static_cast<std::int32_t>(y), 0 };
region.imageExtent = { w, h, 1 };
vkCmdCopyBufferToImage(cmd, buffer.buffer, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &region);
TransitionImageLayout(cmd, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, layout, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_RAY_TRACING_SHADER_BIT_KHR, VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_SHADER_READ_BIT, 0, mipLevels);
}
// GPU compressed-asset Update: stage compressed bytes, decompress
// into `buffer` via VK_EXT_memory_decompression, then copy buffer→image
// and transition to `layout`. Falls back to CPU decode + the existing