Various sync fixes

src/render/ForwardRenderer.cpp

@@ -488,8 +488,7 @@ void ForwardRenderer::recordDraw(
 
     const size_t pipelineIndex = options.transparents ? 1 : 0;
 
-    const uint32_t dsIndex =
-        nextFrame * MAX_FRAMES_IN_FLIGHT * 2 + m_nextFrameRecord;
+    const uint32_t dsIndex = nextFrame * 2 * 2 + m_nextFrameRecord;
     const vk::DescriptorSet ds = m_meshSets[dsIndex];
 
     updateDescriptorSet(

src/render/Renderer.cpp

@@ -47,14 +47,10 @@ void blitFinalComposite(
     const StaticArray barriers{{
         *gRenderResources.images->transitionBarrier(
             finalComposite, ImageState::TransferSrc, true),
+        // https://github.com/KhronosGroup/Vulkan-Docs/wiki/Synchronization-Examples#combined-graphicspresent-queue
         vk::ImageMemoryBarrier2{
-            // TODO:
-            // What's the tight stage for this? Synchronization validation
-            // complained about a hazard after color attachment write which
-            // seems like an oddly specific stage for present source access to
-            // happen in.
-            .srcStageMask = vk::PipelineStageFlagBits2::eBottomOfPipe,
-            .srcAccessMask = vk::AccessFlags2{},
+            .srcStageMask = vk::PipelineStageFlagBits2::eColorAttachmentOutput,
+            .srcAccessMask = vk::AccessFlagBits2::eNone,
             .dstStageMask = vk::PipelineStageFlagBits2::eTransfer,
             .dstAccessMask = vk::AccessFlagBits2::eTransferWrite,
             .oldLayout = vk::ImageLayout::eUndefined,
@@ -106,10 +102,11 @@ void blitFinalComposite(
     }
 
     {
+        // https://github.com/KhronosGroup/Vulkan-Docs/wiki/Synchronization-Examples#combined-graphicspresent-queue
         const vk::ImageMemoryBarrier2 barrier{
             .srcStageMask = vk::PipelineStageFlagBits2::eTransfer,
             .srcAccessMask = vk::AccessFlagBits2::eTransferWrite,
-            .dstStageMask = vk::PipelineStageFlagBits2::eNone,
+            .dstStageMask = vk::PipelineStageFlagBits2::eColorAttachmentOutput,
             .dstAccessMask = vk::AccessFlagBits2::eNone,
             .oldLayout = vk::ImageLayout::eTransferDstOptimal,
             .newLayout = vk::ImageLayout::ePresentSrcKHR,

src/utils/Profiler.cpp

@@ -37,7 +37,7 @@ GpuFrameProfiler::Scope::Scope(
         .pLabelName = name,
     });
     cb.writeTimestamp2(
-        vk::PipelineStageFlagBits2::eTopOfPipe, m_pools.timestamps,
+        vk::PipelineStageFlagBits2::eAllCommands, m_pools.timestamps,
         m_queryIndex * 2);
     if (m_hasStatistics)
         cb.beginQuery(
@@ -51,7 +51,7 @@ GpuFrameProfiler::Scope::~Scope()
         if (m_hasStatistics)
             m_cb.endQuery(m_pools.statistics, m_queryIndex);
         m_cb.writeTimestamp2(
-            vk::PipelineStageFlagBits2::eBottomOfPipe, m_pools.timestamps,
+            vk::PipelineStageFlagBits2::eAllCommands, m_pools.timestamps,
             m_queryIndex * 2 + 1);
         m_cb.endDebugUtilsLabelEXT();
     }
@@ -161,6 +161,23 @@ void GpuFrameProfiler::startFrame()
 
 void GpuFrameProfiler::endFrame(vk::CommandBuffer cb)
 {
+    // Full pipeline barrier as a workaround for vk::QueryResultFlagBits::eWait
+    // causing a device lost.
+    // TODO:
+    // Why does wait misbehave?
+    const vk::MemoryBarrier2KHR memoryBarrier = {
+        .srcStageMask = vk::PipelineStageFlagBits2::eAllCommands,
+        .srcAccessMask = vk::AccessFlagBits2::eMemoryRead |
+                         vk::AccessFlagBits2::eMemoryWrite,
+        .dstStageMask = vk::PipelineStageFlagBits2::eAllCommands,
+        .dstAccessMask = vk::AccessFlagBits2::eMemoryRead |
+                         vk::AccessFlagBits2::eMemoryWrite,
+    };
+    cb.pipelineBarrier2(vk::DependencyInfo{
+        .memoryBarrierCount = 1,
+        .pMemoryBarriers = &memoryBarrier,
+    });
+
     cb.copyQueryPoolResults(
         m_pools.timestamps, 0,
         asserted_cast<uint32_t>(m_queryScopeIndices.size() * 2),