Add basic begin/end support for perf counters

The AMD_performance_monitor extension has explicit begin/end calls to
capture counters.  This was not implemented in ANGLE and the tests were
relying on ANGLE always capturing counters (incurring a small overhead).

This change does not complete the implementation of that extension, but
does add basic support for starting and stopping perf counter
measurements.  While inactive, most counters are not updated.

Bug: angleproject:42267038
Change-Id: I3ff6448b22ca247c217401cb2d76ef4142c9d759
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/5639343
Commit-Queue: Shahbaz Youssefi <[email protected]>
Reviewed-by: Cody Northrop <[email protected]>
Reviewed-by: Amirali Abdolrashidi <[email protected]>

src/libANGLE/Context.cpp

@@ -9674,11 +9674,17 @@ void Context::finishImmutable() const
  ANGLE_CONTEXT_TRY(mImplementation->finish(this));
 }
 
-void Context::beginPerfMonitor(GLuint monitor) {}
+void Context::beginPerfMonitor(GLuint monitor)
+{
+ getMutablePrivateState()->setPerfMonitorActive(true);
+}
 
 void Context::deletePerfMonitors(GLsizei n, GLuint *monitors) {}
 
-void Context::endPerfMonitor(GLuint monitor) {}
+void Context::endPerfMonitor(GLuint monitor)
+{
+ getMutablePrivateState()->setPerfMonitorActive(false);
+}
 
 void Context::genPerfMonitors(GLsizei n, GLuint *monitors)
 {

src/libANGLE/ErrorStrings.h

@@ -498,6 +498,7 @@ MSG kOffsetOverflow = "Offset overflows texture dimensions.";
 MSG kOtherQueryActive = "Other query is active.";
 MSG kOutsideOfBounds = "Parameter outside of bounds.";
 MSG kParamOverflow = "The provided parameters overflow with the provided buffer.";
+MSG kPerfMonitorNotActive = "Perf monitor is not started.";
 MSG kPixelDataNotNull = "Pixel data must be null.";
 MSG kPixelDataNull = "Pixel data cannot be null.";
 MSG kPixelPackBufferBoundForTransformFeedback = "It is undefined behavior to use a pixel pack buffer that is bound for transform feedback.";

src/libANGLE/State.cpp

@@ -384,6 +384,7 @@ PrivateState::PrivateState(const EGLenum clientType,
  mShadingRatePreserveAspectRatio(false),
  mShadingRate(ShadingRate::Undefined),
  mFetchPerSample(false),
+ mIsPerfMonitorActive(false),
  mBindGeneratesResource(bindGeneratesResourceCHROMIUM),
  mClientArraysEnabled(clientArraysEnabled),
  mRobustResourceInit(robustResourceInit),

src/libANGLE/State.h

@@ -597,6 +597,9 @@ class PrivateState : angle::NonCopyable
  const state::DirtyObjects &getDirtyObjects() const { return mDirtyObjects; }
  void clearDirtyObjects() { mDirtyObjects.reset(); }
 
+ void setPerfMonitorActive(bool active) { mIsPerfMonitorActive = active; }
+ bool isPerfMonitorActive() const { return mIsPerfMonitorActive; }
+
  private:
  bool hasConstantColor(GLenum sourceRGB, GLenum destRGB) const;
  bool hasConstantAlpha(GLenum sourceRGB, GLenum destRGB) const;
@@ -725,6 +728,9 @@ class PrivateState : angle::NonCopyable
  // GL_ARM_shader_framebuffer_fetch
  bool mFetchPerSample;
 
+ // Whether perf monitoring is enabled through GL_AMD_performance_monitor.
+ bool mIsPerfMonitorActive;
+
  const bool mBindGeneratesResource;
  const bool mClientArraysEnabled;
  const bool mRobustResourceInit;
@@ -1406,6 +1412,7 @@ class State : angle::NonCopyable
  }
  bool isLogicOpEnabled() const { return mPrivateState.isLogicOpEnabled(); }
  LogicalOperation getLogicOp() const { return mPrivateState.getLogicOp(); }
+ bool isPerfMonitorActive() const { return mPrivateState.isPerfMonitorActive(); }
  const Debug &getDebug() const { return mPrivateState.getDebug(); }
  Debug &getDebug() { return mPrivateState.getDebug(); }
  bool getEnableFeature(GLenum feature) const { return mPrivateState.getEnableFeature(feature); }

src/libANGLE/capture/capture_gles_ext_params.cpp

@@ -4175,7 +4175,7 @@ void CaptureDeletePerfMonitorsAMD_monitors(const State &glState,
  GLuint *monitors,
  angle::ParamCapture *paramCapture)
 {
- UNIMPLEMENTED();
+ CaptureArray(monitors, n, paramCapture);
 }
 
 void CaptureGenPerfMonitorsAMD_monitors(const State &glState,
@@ -4184,7 +4184,7 @@ void CaptureGenPerfMonitorsAMD_monitors(const State &glState,
  GLuint *monitors,
  angle::ParamCapture *paramCapture)
 {
- UNIMPLEMENTED();
+ CaptureArray(monitors, n, paramCapture);
 }
 
 void CaptureGetPerfMonitorCounterDataAMD_data(const State &glState,

src/libANGLE/renderer/vulkan/ContextVk.cpp

@@ -3437,6 +3437,11 @@ angle::Result ContextVk::handleDirtyDescriptorSetsImpl(CommandBufferHelperT *com
 
 void ContextVk::syncObjectPerfCounters(const angle::VulkanPerfCounters &commandQueuePerfCounters)
 {
+ if (!mState.isPerfMonitorActive())
+ {
+ return;
+ }
+
  mPerfCounters.descriptorSetCacheTotalSize = 0;
  mPerfCounters.descriptorSetCacheKeySizeBytes = 0;
  mPerfCounters.uniformsAndXfbDescriptorSetCacheHits = 0;

src/libANGLE/validationESEXT.cpp

@@ -4188,8 +4188,7 @@ bool ValidateBeginPerfMonitorAMD(const Context *context,
  return false;
  }
 
- UNIMPLEMENTED();
- return false;
+ return true;
 }
 
 bool ValidateDeletePerfMonitorsAMD(const Context *context,
@@ -4203,8 +4202,8 @@ bool ValidateDeletePerfMonitorsAMD(const Context *context,
  return false;
  }
 
- UNIMPLEMENTED();
- return false;
+ // Note: ANGLE does not really create monitor objects or track ids.
+ return true;
 }
 
 bool ValidateEndPerfMonitorAMD(const Context *context, angle::EntryPoint entryPoint, GLuint monitor)
@@ -4215,8 +4214,13 @@ bool ValidateEndPerfMonitorAMD(const Context *context, angle::EntryPoint entryPo
  return false;
  }
 
- UNIMPLEMENTED();
- return false;
+ if (!context->getState().isPerfMonitorActive())
+ {
+ ANGLE_VALIDATION_ERROR(GL_INVALID_OPERATION, kPerfMonitorNotActive);
+ return false;
+ }
+
+ return true;
 }
 
 bool ValidateGenPerfMonitorsAMD(const Context *context,
@@ -4230,8 +4234,7 @@ bool ValidateGenPerfMonitorsAMD(const Context *context,
  return false;
  }
 
- UNIMPLEMENTED();
- return false;
+ return true;
 }
 
 bool ValidateGetPerfMonitorCounterDataAMD(const Context *context,

src/tests/gl_tests/ImageTest.cpp

@@ -4654,6 +4654,9 @@ TEST_P(ImageTestES3, AHBImportReleaseStress)
 
  glFinish();
 
+ GLPerfMonitor monitor;
+ glBeginPerfMonitorAMD(monitor);
+
  const uint64_t initialPendingSubmissionGarbageObjects =
  getPerfCounters().pendingSubmissionGarbageObjects;
 
@@ -4680,6 +4683,8 @@ TEST_P(ImageTestES3, AHBImportReleaseStress)
  destroyAndroidHardwareBuffer(ahb);
  }
 
+ glEndPerfMonitorAMD(monitor);
+
  EXPECT_LE(getPerfCounters().pendingSubmissionGarbageObjects,
  initialPendingSubmissionGarbageObjects + 10);
 }

src/tests/gl_tests/TextureTest.cpp

@@ -3579,6 +3579,9 @@ TEST_P(Texture2DMemoryTestES3, TextureDataInLoopUntilFlush)
  // submissions.
  ANGLE_SKIP_TEST_IF(getEGLWindow()->isFeatureEnabled(Feature::SupportsHostImageCopy));
 
+ GLPerfMonitor monitor;
+ glBeginPerfMonitorAMD(monitor);
+
  uint64_t expectedSubmitCalls = getPerfCounters().commandQueueSubmitCallsTotal + 1;
 
  // Set up program
@@ -3639,6 +3642,8 @@ void main()
  break;
  }
  }
+ glEndPerfMonitorAMD(monitor);
+
  EXPECT_EQ(getPerfCounters().commandQueueSubmitCallsTotal, expectedSubmitCalls);
  ASSERT_GL_NO_ERROR();
  EXPECT_PIXEL_NEAR(0, 0, 128, 191, 64, 255, 1);
@@ -3651,6 +3656,10 @@ TEST_P(Texture2DMemoryTestES3, TextureDataInLoopManyTimes)
 {
  // Run this test for Vulkan only.
  ANGLE_SKIP_TEST_IF(!IsVulkan());
+
+ GLPerfMonitor monitor;
+ glBeginPerfMonitorAMD(monitor);
+
  uint64_t expectedSubmitCalls = getPerfCounters().commandQueueSubmitCallsTotal + 1;
  uint64_t expectedDeviceMemoryFallbacks = getPerfCounters().deviceMemoryImageAllocationFallbacks;
 
@@ -3714,6 +3723,9 @@ void main()
  break;
  }
  }
+
+ glEndPerfMonitorAMD(monitor);
+
  EXPECT_EQ(getPerfCounters().commandQueueSubmitCallsTotal, expectedSubmitCalls);
  EXPECT_EQ(getPerfCounters().deviceMemoryImageAllocationFallbacks,
  expectedDeviceMemoryFallbacks);

src/tests/gl_tests/UniformBufferTest.cpp

@@ -3762,6 +3762,9 @@ class UniformBufferMemoryTest : public UniformBufferTest
 // supposedly to issue flush if needed.
 TEST_P(UniformBufferMemoryTest, BufferDataInLoopManyTimes)
 {
+ GLPerfMonitor monitor;
+ glBeginPerfMonitorAMD(monitor);
+
  // Run this test for Vulkan only.
  ANGLE_SKIP_TEST_IF(!IsVulkan());
  uint64_t expectedSubmitCalls = getPerfCounters().commandQueueSubmitCallsTotal + 1;
@@ -3801,6 +3804,8 @@ TEST_P(UniformBufferMemoryTest, BufferDataInLoopManyTimes)
  break;
  }
  }
+ glEndPerfMonitorAMD(monitor);
+
  EXPECT_EQ(getPerfCounters().commandQueueSubmitCallsTotal, expectedSubmitCalls);
  ASSERT_GL_NO_ERROR();
  EXPECT_PIXEL_NEAR(0, 0, 128, 191, 64, 255, 1);

src/tests/gl_tests/VulkanImageTest.cpp

@@ -550,6 +550,9 @@ TEST_P(VulkanMemoryTest, AllocateVMAImageWhenDeviceOOM)
 {
  ANGLE_SKIP_TEST_IF(!getEGLWindow()->isFeatureEnabled(Feature::UseVmaForImageSuballocation));
 
+ GLPerfMonitor monitor;
+ glBeginPerfMonitorAMD(monitor);
+
  VulkanHelper helper;
  helper.initializeFromANGLE();
  uint64_t expectedAllocationFallbacks =
@@ -596,6 +599,8 @@ TEST_P(VulkanMemoryTest, AllocateVMAImageWhenDeviceOOM)
  EXPECT_EQ(getPerfCounters().deviceMemoryImageAllocationFallbacks,
  expectedAllocationFallbacksAfterLastTexture);
 
+ glEndPerfMonitorAMD(monitor);
+
  GLFramebuffer fbo;
  glBindFramebuffer(GL_FRAMEBUFFER, fbo);
  glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, texture, 0);
@@ -609,6 +614,9 @@ TEST_P(VulkanMemoryTest, AllocateVMAImageAfterFreeing2DArrayGarbageWhenDeviceOOM
 {
  ANGLE_SKIP_TEST_IF(!getEGLWindow()->isFeatureEnabled(Feature::UseVmaForImageSuballocation));
 
+ GLPerfMonitor monitor;
+ glBeginPerfMonitorAMD(monitor);
+
  VulkanHelper helper;
  helper.initializeFromANGLE();
  uint64_t expectedAllocationFallbacks =
@@ -689,6 +697,8 @@ TEST_P(VulkanMemoryTest, AllocateVMAImageAfterFreeing2DArrayGarbageWhenDeviceOOM
  GL_UNSIGNED_BYTE, lastTextureColor.data());
  EXPECT_EQ(getPerfCounters().deviceMemoryImageAllocationFallbacks, expectedAllocationFallbacks);
 
+ glEndPerfMonitorAMD(monitor);
+
  GLFramebuffer fbo;
  glBindFramebuffer(GL_FRAMEBUFFER, fbo);
  glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, lastTexture, 0);
@@ -702,6 +712,9 @@ TEST_P(VulkanMemoryTest, AllocateVMAImageAfterFreeingFinished2DGarbageWhenDevice
 {
  ANGLE_SKIP_TEST_IF(!getEGLWindow()->isFeatureEnabled(Feature::UseVmaForImageSuballocation));
 
+ GLPerfMonitor monitor;
+ glBeginPerfMonitorAMD(monitor);
+
  VulkanHelper helper;
  helper.initializeFromANGLE();
  uint64_t expectedAllocationFallbacks =
@@ -797,6 +810,8 @@ TEST_P(VulkanMemoryTest, AllocateVMAImageAfterFreeingFinished2DGarbageWhenDevice
  EXPECT_EQ(getPerfCounters().deviceMemoryImageAllocationFallbacks, expectedAllocationFallbacks);
  EXPECT_EQ(getPerfCounters().commandQueueSubmitCallsTotal, expectedSubmitCalls);
 
+ glEndPerfMonitorAMD(monitor);
+
  GLFramebuffer fbo;
  glBindFramebuffer(GL_FRAMEBUFFER, fbo);
  glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, lastTexture, 0);
@@ -810,6 +825,9 @@ TEST_P(VulkanMemoryTest, AllocateBufferAfterFreeing2DGarbageWhenDeviceOOM)
 {
  ANGLE_SKIP_TEST_IF(!getEGLWindow()->isFeatureEnabled(Feature::UseVmaForImageSuballocation));
 
+ GLPerfMonitor monitor;
+ glBeginPerfMonitorAMD(monitor);
+
  VulkanHelper helper;
  helper.initializeFromANGLE();
  uint64_t expectedAllocationFallbacks =
@@ -867,6 +885,8 @@ TEST_P(VulkanMemoryTest, AllocateBufferAfterFreeing2DGarbageWhenDeviceOOM)
  }
  EXPECT_EQ(getPerfCounters().deviceMemoryImageAllocationFallbacks, expectedAllocationFallbacks);
 
+ glEndPerfMonitorAMD(monitor);
+
  // Wait until GPU finishes execution.
  GLsync sync = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
  glWaitSync(sync, 0, GL_TIMEOUT_IGNORED);

src/tests/gl_tests/VulkanPerformanceCounterTest.cpp

@@ -146,6 +146,18 @@ class VulkanPerformanceCounterTest : public ANGLETest<>
 
  static constexpr GLsizei kOpsTestSize = 16;
 
+ void testSetUp() override
+ {
+ glGenPerfMonitorsAMD(1, &monitor);
+ glBeginPerfMonitorAMD(monitor);
+ }
+
+ void testTearDown() override
+ {
+ glEndPerfMonitorAMD(monitor);
+ glDeletePerfMonitorsAMD(1, &monitor);
+ }
+
  void setupForColorOpsTest(GLFramebuffer *framebuffer, GLTexture *texture)
  {
  // Setup the framebuffer
@@ -416,6 +428,7 @@ class VulkanPerformanceCounterTest : public ANGLETest<>
  !isFeatureEnabled(Feature::DisableDepthStencilResolveThroughAttachment);
  }
 
+ GLuint monitor;
  CounterNameToIndexMap mIndexMap;
 };
 

src/tests/perf_tests/ANGLEPerfTest.cpp

@@ -302,7 +302,8 @@ ANGLEPerfTest::ANGLEPerfTest(const std::string &name,
  mTrialNumStepsPerformed(0),
  mTotalNumStepsPerformed(0),
  mIterationsPerStep(iterationsPerStep),
- mRunning(true)
+ mRunning(true),
+ mPerfMonitor(0)
 {
  if (mStory == "")
  {
@@ -1043,6 +1044,12 @@ void ANGLERenderTest::TearDown()
 {
  ASSERT(mTimestampQueries.empty());
 
+ if (!mPerfCounterInfo.empty())
+ {
+ glDeletePerfMonitorsAMD(1, &mPerfMonitor);
+ mPerfMonitor = 0;
+ }
+
  if (!mSkipTest)
  {
  destroyBenchmark();
@@ -1130,6 +1137,13 @@ void ANGLERenderTest::initPerfCounters()
  fprintf(stderr, "'%s' does not match any available perf counters.\n", counter.c_str());
  }
  }
+
+ if (!mPerfCounterInfo.empty())
+ {
+ glGenPerfMonitorsAMD(1, &mPerfMonitor);
+ // Note: technically, glSelectPerfMonitorCountersAMD should be used to select the counters,
+ // but currently ANGLE always captures all counters.
+ }
 }
 
 void ANGLERenderTest::updatePerfCounters()
@@ -1295,7 +1309,13 @@ void ANGLERenderTest::computeGPUTime()
  }
 }
 
-void ANGLERenderTest::startTest() {}
+void ANGLERenderTest::startTest()
+{
+ if (!mPerfCounterInfo.empty())
+ {
+ glBeginPerfMonitorAMD(mPerfMonitor);
+ }
+}
 
 void ANGLERenderTest::finishTest()
 {
@@ -1304,6 +1324,11 @@ void ANGLERenderTest::finishTest()
  {
  FinishAndCheckForContextLoss();
  }
+
+ if (!mPerfCounterInfo.empty())
+ {
+ glEndPerfMonitorAMD(mPerfMonitor);
+ }
 }
 
 bool ANGLERenderTest::popEvent(Event *event)