Vulkan: Optimize Attribute Change Perf (2/5)

This patch series optimizes programs that use the pattern:

for (;;) {
  glVertexAttribPointer(...)
  glDraw(...)
}

Change 2: Micro-optimize XFB resume CPU overhead.

We don't need to set resume on every new command buffer. We only need
to set the dirty bit when we have an unexpected pause.

In total the patch series reduces test iteration time by 25%.

Test: DrawCallPerfBenchmark.Run/vulkan_attrib_change
Bug: angleproject:5045
Bug: b/168493024
Change-Id: I8f6c68ff0513be4f405276e395d80bc1a185a061
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2409174
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>

src/libANGLE/renderer/vulkan/ContextVk.cpp

@@ -694,16 +694,6 @@ ContextVk::ContextVk(const gl::State &state, gl::ErrorSet *errorSet, RendererVk
     mNewGraphicsCommandBufferDirtyBits.set(DIRTY_BIT_VERTEX_BUFFERS);
     mNewGraphicsCommandBufferDirtyBits.set(DIRTY_BIT_INDEX_BUFFER);
     mNewGraphicsCommandBufferDirtyBits.set(DIRTY_BIT_SHADER_RESOURCES);
-    if (getFeatures().supportsTransformFeedbackExtension.enabled ||
-        getFeatures().emulateTransformFeedback.enabled)
-    {
-        mNewGraphicsCommandBufferDirtyBits.set(DIRTY_BIT_TRANSFORM_FEEDBACK_BUFFERS);
-    }
-    if (getFeatures().supportsTransformFeedbackExtension.enabled)
-    {
-        mNewGraphicsCommandBufferDirtyBits.set(DIRTY_BIT_TRANSFORM_FEEDBACK_STATE);
-        mNewGraphicsCommandBufferDirtyBits.set(DIRTY_BIT_TRANSFORM_FEEDBACK_RESUME);
-    }
     mNewGraphicsCommandBufferDirtyBits.set(DIRTY_BIT_DESCRIPTOR_SETS);
     mNewGraphicsCommandBufferDirtyBits.set(DIRTY_BIT_DRIVER_UNIFORMS_BINDING);
 
@@ -713,12 +703,6 @@ ContextVk::ContextVk(const gl::State &state, gl::ErrorSet *errorSet, RendererVk
     mNewComputeCommandBufferDirtyBits.set(DIRTY_BIT_DESCRIPTOR_SETS);
     mNewComputeCommandBufferDirtyBits.set(DIRTY_BIT_DRIVER_UNIFORMS_BINDING);
 
-    mNewGraphicsPipelineDirtyBits.set(DIRTY_BIT_PIPELINE);
-    if (getFeatures().supportsTransformFeedbackExtension.enabled)
-    {
-        mNewGraphicsPipelineDirtyBits.set(DIRTY_BIT_TRANSFORM_FEEDBACK_RESUME);
-    }
-
     mGraphicsDirtyBitHandlers[DIRTY_BIT_DEFAULT_ATTRIBS] =
         &ContextVk::handleDirtyGraphicsDefaultAttribs;
     mGraphicsDirtyBitHandlers[DIRTY_BIT_PIPELINE] = &ContextVk::handleDirtyGraphicsPipeline;
@@ -1354,7 +1338,9 @@ angle::Result ContextVk::handleDirtyGraphicsPipeline(const gl::Context *context,
 
         mGraphicsPipelineTransition.reset();
     }
-    mRenderPassCommands->pauseTransformFeedbackIfStarted();
+
+    resumeTransformFeedbackIfStarted();
+
     commandBuffer->bindGraphicsPipeline(mCurrentGraphicsPipeline->getPipeline());
     // Update the queue serial for the pipeline object.
     ASSERT(mCurrentGraphicsPipeline && mCurrentGraphicsPipeline->valid());
@@ -1646,7 +1632,10 @@ angle::Result ContextVk::handleDirtyGraphicsTransformFeedbackResume(
     const gl::Context *context,
     vk::CommandBuffer *commandBuffer)
 {
-    mRenderPassCommands->resumeTransformFeedbackIfStarted();
+    if (mRenderPassCommands->isTransformFeedbackStarted())
+    {
+        mRenderPassCommands->resumeTransformFeedback();
+    }
     return angle::Result::Continue;
 }
 
@@ -3194,6 +3183,17 @@ angle::Result ContextVk::onMakeCurrent(const gl::Context *context)
     updateSurfaceRotationReadFramebuffer(glState);
     invalidateDriverUniforms();
 
+    const gl::ProgramExecutable *executable = mState.getProgramExecutable();
+    if (executable && executable->hasTransformFeedbackOutput() &&
+        mState.isTransformFeedbackActive())
+    {
+        onTransformFeedbackStateChanged();
+        if (getFeatures().supportsTransformFeedbackExtension.enabled)
+        {
+            mGraphicsDirtyBits.set(DIRTY_BIT_TRANSFORM_FEEDBACK_RESUME);
+        }
+    }
+
     return angle::Result::Continue;
 }
 
@@ -3442,7 +3442,7 @@ void ContextVk::onTransformFeedbackStateChanged()
     if (getFeatures().supportsTransformFeedbackExtension.enabled)
     {
         mGraphicsDirtyBits.set(DIRTY_BIT_TRANSFORM_FEEDBACK_STATE);
-        invalidateCurrentTransformFeedbackBuffers();
+        mGraphicsDirtyBits.set(DIRTY_BIT_TRANSFORM_FEEDBACK_BUFFERS);
     }
     else if (getFeatures().emulateTransformFeedback.enabled)
     {
@@ -3492,7 +3492,10 @@ void ContextVk::onEndTransformFeedback()
 {
     if (getFeatures().supportsTransformFeedbackExtension.enabled)
     {
-        mRenderPassCommands->endTransformFeedback();
+        if (mRenderPassCommands->isTransformFeedbackStarted())
+        {
+            mRenderPassCommands->endTransformFeedback();
+        }
     }
     else if (getFeatures().emulateTransformFeedback.enabled)
     {
@@ -3508,6 +3511,7 @@ angle::Result ContextVk::onPauseTransformFeedback()
     }
     else if (getFeatures().emulateTransformFeedback.enabled)
     {
+        invalidateCurrentTransformFeedbackBuffers();
         return flushCommandsAndEndRenderPass();
     }
     return angle::Result::Continue;
@@ -3676,6 +3680,16 @@ void ContextVk::writeAtomicCounterBufferDriverUniformOffsets(uint32_t *offsetsOu
     }
 }
 
+ANGLE_INLINE void ContextVk::resumeTransformFeedbackIfStarted()
+{
+    if (mRenderPassCommands->isTransformFeedbackStarted())
+    {
+        ASSERT(getFeatures().supportsTransformFeedbackExtension.enabled);
+        mGraphicsDirtyBits.set(DIRTY_BIT_TRANSFORM_FEEDBACK_RESUME);
+        mRenderPassCommands->pauseTransformFeedback();
+    }
+}
+
 angle::Result ContextVk::handleDirtyGraphicsDriverUniforms(const gl::Context *context,
                                                            vk::CommandBuffer *commandBuffer)
 {
@@ -4664,6 +4678,8 @@ angle::Result ContextVk::flushCommandsAndEndRenderPass()
 
     addOverlayUsedBuffersCount(mRenderPassCommands);
 
+    resumeTransformFeedbackIfStarted();
+
     mRenderPassCommands->endRenderPass(this);
 
     if (mRenderer->getFeatures().enableCommandProcessingThread.enabled)

src/libANGLE/renderer/vulkan/ContextVk.h

@@ -795,7 +795,7 @@ class ContextVk : public ContextImpl, public vk::Context
 
     ANGLE_INLINE void invalidateCurrentGraphicsPipeline()
     {
-        mGraphicsDirtyBits |= mNewGraphicsPipelineDirtyBits;
+        mGraphicsDirtyBits.set(DIRTY_BIT_PIPELINE);
     }
 
     ANGLE_INLINE void invalidateCurrentComputePipeline()
@@ -919,6 +919,7 @@ class ContextVk : public ContextImpl, public vk::Context
     void populateTransformFeedbackBufferSet(
         size_t bufferCount,
         const gl::TransformFeedbackBuffersArray<vk::BufferHelper *> &buffers);
+    void resumeTransformFeedbackIfStarted();
 
     // DescriptorSet writes
     template <typename T, const T *VkWriteDescriptorSet::*pInfo>
@@ -964,7 +965,6 @@ class ContextVk : public ContextImpl, public vk::Context
     DirtyBits mIndexedDirtyBitsMask;
     DirtyBits mNewGraphicsCommandBufferDirtyBits;
     DirtyBits mNewComputeCommandBufferDirtyBits;
-    DirtyBits mNewGraphicsPipelineDirtyBits;
 
     // Cached back-end objects.
     VertexArrayVk *mVertexArray;

src/libANGLE/renderer/vulkan/vk_helpers.cpp

@@ -898,8 +898,6 @@ void CommandBufferHelper::restartRenderPassWithReadOnlyDepth(const Framebuffer &
 
 void CommandBufferHelper::endRenderPass(ContextVk *contextVk)
 {
-    pauseTransformFeedbackIfStarted();
-
     if (mDepthStencilAttachmentIndex == kAttachmentIndexInvalid)
     {
         return;
@@ -965,7 +963,7 @@ void CommandBufferHelper::beginTransformFeedback(size_t validBufferCount,
 void CommandBufferHelper::endTransformFeedback()
 {
     ASSERT(mIsRenderPassCommandBuffer);
-    pauseTransformFeedbackIfStarted();
+    pauseTransformFeedback();
     mValidTransformFeedbackBufferCount = 0;
 }
 
@@ -1164,13 +1162,10 @@ void CommandBufferHelper::releaseToContextQueue(ContextVk *contextVk)
     contextVk->recycleCommandBuffer(this);
 }
 
-void CommandBufferHelper::resumeTransformFeedbackIfStarted()
+void CommandBufferHelper::resumeTransformFeedback()
 {
     ASSERT(mIsRenderPassCommandBuffer);
-    if (mValidTransformFeedbackBufferCount == 0)
-    {
-        return;
-    }
+    ASSERT(isTransformFeedbackStarted());
 
     uint32_t numCounterBuffers =
         mRebindTransformFeedbackBuffers ? 0 : mValidTransformFeedbackBufferCount;
@@ -1181,14 +1176,10 @@ void CommandBufferHelper::resumeTransformFeedbackIfStarted()
                                           mTransformFeedbackCounterBuffers.data());
 }
 
-void CommandBufferHelper::pauseTransformFeedbackIfStarted()
+void CommandBufferHelper::pauseTransformFeedback()
 {
     ASSERT(mIsRenderPassCommandBuffer);
-    if (mValidTransformFeedbackBufferCount == 0)
-    {
-        return;
-    }
-
+    ASSERT(isTransformFeedbackStarted());
     mCommandBuffer.endTransformFeedback(mValidTransformFeedbackBufferCount,
                                         mTransformFeedbackCounterBuffers.data());
 }

src/libANGLE/renderer/vulkan/vk_helpers.h

@@ -1047,8 +1047,9 @@ class CommandBufferHelper : angle::NonCopyable
         return mRenderArea;
     }
 
-    void resumeTransformFeedbackIfStarted();
-    void pauseTransformFeedbackIfStarted();
+    void resumeTransformFeedback();
+    void pauseTransformFeedback();
+    bool isTransformFeedbackStarted() const { return mValidTransformFeedbackBufferCount > 0; }
 
     uint32_t getAndResetCounter()
     {

src/tests/gl_tests/TransformFeedbackTest.cpp

@@ -582,7 +582,8 @@ TEST_P(TransformFeedbackTest, MultiContext)
         GLuint buffer;
         size_t primitiveCounts[passCount];
     };
-    ContextInfo contexts[32];
+    static constexpr uint32_t kContextCount = 32;
+    ContextInfo contexts[kContextCount];
 
     const size_t maxDrawSize = 512;