Add VulkanAddWorkToEmptyResolvePass Toggle
This toggle is enabled by default on Qualcomm GPUs and adds a small
amount of work (a timestamp write) to render passes which contain no
draw commands. This works around a driver bug where the resolve targets
of the empty pass don't get resolved.
This case is covered by the CTS with
https://gpuweb.github.io/cts/standalone/?runnow=1&q=webgpu:api,operation,render_pass,resolve:render_pass_resolve:*
which was observed to fail on a Pixel 4 but with this fix now passes.
Bug: 411656647
Change-Id: If89e57df5289e18cf7314a9025d2ac985792ba8f
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/238334
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Commit-Queue: Brandon Jones <bajones@chromium.org>
Reviewed-by: Loko Kung <lokokung@google.com>
diff --git a/src/dawn/native/Toggles.cpp b/src/dawn/native/Toggles.cpp
index d85eb16..d1adc69 100644
--- a/src/dawn/native/Toggles.cpp
+++ b/src/dawn/native/Toggles.cpp
@@ -639,6 +639,11 @@
{Toggle::UseVulkanMemoryModel,
{"use_vulkan_memory_model", "Use the Vulkan Memory Model if available.",
"https://crbug.com/392606604", ToggleStage::Adapter}},
+ {Toggle::VulkanAddWorkToEmptyResolvePass,
+ {"vulkan_add_work_to_empty_resolve_pass",
+ "Adds a small amount of work to empty render passes which perform a resolve. This toggle is "
+ "enabled by default on Qualcomm GPUs, where it is needed to force the resolve to complete.",
+ "https://crbug.com/411656647", ToggleStage::Device}},
{Toggle::NoWorkaroundSampleMaskBecomesZeroForAllButLastColorTarget,
{"no_workaround_sample_mask_becomes_zero_for_all_but_last_color_target",
"MacOS 12.0+ Intel has a bug where the sample mask is only applied for the last color "
diff --git a/src/dawn/native/Toggles.h b/src/dawn/native/Toggles.h
index 0d86dce..a761e87 100644
--- a/src/dawn/native/Toggles.h
+++ b/src/dawn/native/Toggles.h
@@ -151,6 +151,7 @@
D3D12RelaxMinSubgroupSizeTo8,
D3D12RelaxBufferTextureCopyPitchAndOffsetAlignment,
UseVulkanMemoryModel,
+ VulkanAddWorkToEmptyResolvePass,
// Unresolved issues.
NoWorkaroundSampleMaskBecomesZeroForAllButLastColorTarget,
diff --git a/src/dawn/native/vulkan/CommandBufferVk.cpp b/src/dawn/native/vulkan/CommandBufferVk.cpp
index e795dec..cc444b1 100644
--- a/src/dawn/native/vulkan/CommandBufferVk.cpp
+++ b/src/dawn/native/vulkan/CommandBufferVk.cpp
@@ -1261,9 +1261,13 @@
DescriptorSetTracker descriptorSets = {};
RenderPipeline* lastPipeline = nullptr;
+ // Tracks the number of commands that do significant GPU work (a draw or query write) this pass.
+ uint32_t workCommandCount = 0;
+
auto EncodeRenderBundleCommand = [&](CommandIterator* iter, Command type) {
switch (type) {
case Command::Draw: {
+ workCommandCount++;
DrawCmd* draw = iter->NextCommand<DrawCmd>();
descriptorSets.Apply(device, recordingContext, VK_PIPELINE_BIND_POINT_GRAPHICS);
@@ -1274,6 +1278,7 @@
}
case Command::DrawIndexed: {
+ workCommandCount++;
DrawIndexedCmd* draw = iter->NextCommand<DrawIndexedCmd>();
descriptorSets.Apply(device, recordingContext, VK_PIPELINE_BIND_POINT_GRAPHICS);
@@ -1284,6 +1289,7 @@
}
case Command::DrawIndirect: {
+ workCommandCount++;
DrawIndirectCmd* draw = iter->NextCommand<DrawIndirectCmd>();
Buffer* buffer = ToBackend(draw->indirectBuffer.Get());
@@ -1295,6 +1301,7 @@
}
case Command::DrawIndexedIndirect: {
+ workCommandCount++;
DrawIndexedIndirectCmd* draw = iter->NextCommand<DrawIndexedIndirectCmd>();
Buffer* buffer = ToBackend(draw->indirectBuffer.Get());
DAWN_ASSERT(buffer != nullptr);
@@ -1308,6 +1315,7 @@
}
case Command::MultiDrawIndirect: {
+ workCommandCount++;
MultiDrawIndirectCmd* cmd = iter->NextCommand<MultiDrawIndirectCmd>();
Buffer* indirectBuffer = ToBackend(cmd->indirectBuffer.Get());
@@ -1333,6 +1341,7 @@
break;
}
case Command::MultiDrawIndexedIndirect: {
+ workCommandCount++;
MultiDrawIndexedIndirectCmd* cmd = iter->NextCommand<MultiDrawIndexedIndirectCmd>();
Buffer* indirectBuffer = ToBackend(cmd->indirectBuffer.Get());
@@ -1476,6 +1485,17 @@
case Command::EndRenderPass: {
mCommands.NextCommand<EndRenderPassCmd>();
+ // If no work-producing commands were executed during the render pass and the
+ // VulkanAddWorkToEmptyResolvePass toggle is enabled, add a small amount of work
+ // in the form of performing an occlusion query before ending the pass. This avoids
+ // a driver bug that fails to resolve render targets in empty passes.
+ if (workCommandCount == 0 &&
+ device->IsToggleEnabled(Toggle::VulkanAddWorkToEmptyResolvePass)) {
+ QuerySetBase* querySet = device->GetEmptyPassQuerySet();
+ device->fn.CmdBeginQuery(commands, ToBackend(querySet)->GetHandle(), 0, 0);
+ device->fn.CmdEndQuery(commands, ToBackend(querySet)->GetHandle(), 0);
+ }
+
device->fn.CmdEndRenderPass(commands);
// Write timestamp at the end of render pass if it's set.
@@ -1561,6 +1581,7 @@
}
case Command::BeginOcclusionQuery: {
+ workCommandCount++;
BeginOcclusionQueryCmd* cmd = mCommands.NextCommand<BeginOcclusionQueryCmd>();
device->fn.CmdBeginQuery(commands, ToBackend(cmd->querySet.Get())->GetHandle(),
@@ -1569,6 +1590,7 @@
}
case Command::EndOcclusionQuery: {
+ workCommandCount++;
EndOcclusionQueryCmd* cmd = mCommands.NextCommand<EndOcclusionQueryCmd>();
device->fn.CmdEndQuery(commands, ToBackend(cmd->querySet.Get())->GetHandle(),
@@ -1577,6 +1599,7 @@
}
case Command::WriteTimestamp: {
+ workCommandCount++;
WriteTimestampCmd* cmd = mCommands.NextCommand<WriteTimestampCmd>();
RecordWriteTimestampCmd(recordingContext, device, cmd->querySet.Get(),
diff --git a/src/dawn/native/vulkan/DeviceVk.cpp b/src/dawn/native/vulkan/DeviceVk.cpp
index 570c07da..67b7792 100644
--- a/src/dawn/native/vulkan/DeviceVk.cpp
+++ b/src/dawn/native/vulkan/DeviceVk.cpp
@@ -1100,4 +1100,17 @@
return true;
}
+// Gets or creates an occlusion Query object for use with Toggle::VulkanAddWorkToEmptyResolvePass.
+QuerySetBase* Device::GetEmptyPassQuerySet() {
+ DAWN_ASSERT(IsToggleEnabled(Toggle::VulkanAddWorkToEmptyResolvePass));
+
+ if (!mEmptyPassQuerySet) {
+ QuerySetDescriptor descriptor;
+ descriptor.type = wgpu::QueryType::Occlusion;
+ descriptor.count = 1;
+ mEmptyPassQuerySet = APICreateQuerySet(&descriptor);
+ }
+ return mEmptyPassQuerySet.Get();
+}
+
} // namespace dawn::native::vulkan
diff --git a/src/dawn/native/vulkan/DeviceVk.h b/src/dawn/native/vulkan/DeviceVk.h
index 99e43dc..764078e 100644
--- a/src/dawn/native/vulkan/DeviceVk.h
+++ b/src/dawn/native/vulkan/DeviceVk.h
@@ -132,6 +132,8 @@
wgpu::BufferUsage originalUsage,
size_t bufferSize) const override;
+ QuerySetBase* GetEmptyPassQuerySet();
+
private:
Device(AdapterBase* adapter,
const UnpackedPtr<DeviceDescriptor>& descriptor,
@@ -210,6 +212,8 @@
Ref<PipelineCache> mMonolithicPipelineCache;
+ Ref<QuerySetBase> mEmptyPassQuerySet;
+
bool mSupportsMappableStorageBuffer = false;
MaybeError ImportExternalImage(const ExternalImageDescriptorVk* descriptor,
diff --git a/src/dawn/native/vulkan/PhysicalDeviceVk.cpp b/src/dawn/native/vulkan/PhysicalDeviceVk.cpp
index c34da20..1817abb 100644
--- a/src/dawn/native/vulkan/PhysicalDeviceVk.cpp
+++ b/src/dawn/native/vulkan/PhysicalDeviceVk.cpp
@@ -792,6 +792,11 @@
// texture. Work around it by resolving into a single level texture and then copying into
// the intended layer.
deviceToggles->Default(Toggle::AlwaysResolveIntoZeroLevelAndLayer, true);
+
+ // chromium:411656647: Qualcomm devices have a bug where an empty render pass that has a
+ // resolve target doesn't perform the resolve. To work around it, add a small amount of work
+ // to the pass to force it to execute.
+ deviceToggles->Default(Toggle::VulkanAddWorkToEmptyResolvePass, true);
}
if (IsAndroidARM()) {
