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

Merge pull request 'perf(sync): scope frame-loop barriers to swapchain image + real per-pass stages (#115)' (#137) from claude/issue-115 into master

Browse source
This commit is contained in:
catbot 2026-06-17 21:45:13 +02:00
commit 82fd6916d9
6 changed files with 316 additions and 8 deletions
Download patch file Download diff file Expand all files Collapse all files

35
implementations/Crafter.Graphics-Window.cpp
View file

@ -815,7 +815,16 @@ void Window::Render() {
: VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;
acquireBarrier.image = images[currentBuffer];
vkCmdPipelineBarrier(drawCmdBuffers[currentBuffer], VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, 0, nullptr, 0, nullptr, 1, &acquireBarrier);
// Real per-pass stage union for this frame's passes — the swapchain image
// is only ever written by compute/RT (and possibly transfer), never by the
// whole pipeline, so the frame-edge barriers wait on exactly those stages
// instead of ALL_COMMANDS. Derived once here and reused for the present
// barrier below (passes don't change within a Render()).
const VkPipelineStageFlags swapWriterStages = SwapchainStageUnion(passes);
// The acquire->GENERAL transition only needs to be visible before the first
// pass writes the image, i.e. at the writer stages — not ALL_COMMANDS.
vkCmdPipelineBarrier(drawCmdBuffers[currentBuffer], VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, swapWriterStages, 0, 0, nullptr, 0, nullptr, 1, &acquireBarrier);
// Synthesise key-repeat events before listeners run, so the focused
// widget's OnTextInput / OnKeyDown sees them in the same frame.
@ -879,18 +888,28 @@ void Window::Render() {
passes[i]->Record(drawCmdBuffers[currentBuffer], currentBuffer, *this);
if (i + 1 < passes.size()) {
VkMemoryBarrier mb {
.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT,
};
vkCmdPipelineBarrier(drawCmdBuffers[currentBuffer], VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, 1, &mb, 0, nullptr, 0, nullptr);
// Make pass i's writes to the swapchain image visible to pass i+1.
// The only resource the inter-pass hazard touches is that one image
// (later passes overdraw earlier ones), so use an image memory
// barrier scoped to its subresource instead of a queue-wide
// VkMemoryBarrier — only this image's caches round-trip, exactly as
// the intra-pass UI barrier already does. The stage masks come from
// each pass's own SwapchainStage(), so a compute->compute edge only
// serialises COMPUTE_SHADER while an RT pass on either side pulls in
// RAY_TRACING_SHADER — never the ALL_COMMANDS full-pipeline stall.
VkImageMemoryBarrier ib = BuildSwapchainInterPassBarrier(images[currentBuffer]);
vkCmdPipelineBarrier(drawCmdBuffers[currentBuffer],
passes[i]->SwapchainStage(), passes[i + 1]->SwapchainStage(),
0, 0, nullptr, 0, nullptr, 1, &ib);
}
}
presentBarrier.image = images[currentBuffer];
vkCmdPipelineBarrier(drawCmdBuffers[currentBuffer], VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, 0, 0, nullptr, 0, nullptr, 1, &presentBarrier);
// The ->PRESENT_SRC transition only needs to wait on the stages that
// actually wrote the image (the per-pass union), not ALL_COMMANDS; present
// itself is gated by the render-finished semaphore, so dst stays BOTTOM.
vkCmdPipelineBarrier(drawCmdBuffers[currentBuffer], swapWriterStages, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, 0, 0, nullptr, 0, nullptr, 1, &presentBarrier);
Device::CheckVkResult(vkEndCommandBuffer(drawCmdBuffers[currentBuffer]));