content: perf tests for GpuMemoryBuffers mapping and data coherency

content/child/child_thread_impl_perftest.cc

Index: content/child/child_thread_impl_perftest.cc
diff --git a/content/child/child_thread_impl_perftest.cc b/content/child/child_thread_impl_perftest.cc
new file mode 100644
index 0000000000000000000000000000000000000000..abe88032fa5b8c51b9e83cf72de94ffdad5a0aef
--- /dev/null
+++ b/content/child/child_thread_impl_perftest.cc
@@ -0,0 +1,222 @@
+// Copyright 2015 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+// The idea is to benchmark how the hardware, on different usages of
+// GpuMemoryBuffer, performs when the native buffer object (bo) is mapped into
+// the CPU. In particular this test aims to capture the effects of data
+// coherency and answer the following:
+//
+// - measure memory mapping performance of GpuMemoryBuffer using shared memory
+//  (fallback case) and also native implementation of it.
+// - what if the Renderer process (client) just writes into the buffer object?
+// - what's the effect of reading from a write-combining (WC) memory? can we
+//   avoid read backs?
+// - should it be UC and/or WC mapped, to get a faster access?
+// - what the effect of clients doing sequential writes or non-sequential?  If
+//   the latter, a WC mapping may end up being very slow.
+
+#include "base/bind.h"
+#include "base/command_line.h"
+#include "base/memory/scoped_vector.h"
+#include "base/time/time.h"
+#include "content/child/child_gpu_memory_buffer_manager.h"
+#include "content/child/child_thread_impl.h"
+#include "content/common/gpu/client/gpu_memory_buffer_impl.h"
+#include "content/public/common/content_switches.h"
+#include "content/public/test/content_browser_test.h"
+#include "content/public/test/content_browser_test_utils.h"
+#include "content/shell/browser/shell.h"
+#include "testing/perf/perf_test.h"
+#include "url/gurl.h"
+
+namespace content {
+namespace {
+
+ChildGpuMemoryBufferManager* child_gpu_memory_buffer_manager_ = NULL;

[dshwang, 2015/06/25 10:52:11]

nullptr.
why don't BufferPerfTest have this as member?

+
+static const int kNumRuns = 30;

[dshwang, 2015/06/25 10:52:11]

'static' is not needed because of anonymous namespace

+
+enum NativeBufferFlag { kDisableNativeBuffers, kEnableNativeBuffers };
+
+std::string NativeBufferFlagName(NativeBufferFlag flag) {
+  switch (flag) {
+    case kDisableNativeBuffers:
+      return "";
+    case kEnableNativeBuffers:
+      return "_native";
+  }
+
+  NOTREACHED();
+  return "";
+}
+
+static NativeBufferFlag native_buffer_flag_;

[dshwang, 2015/06/25 10:52:11]

why don't ChildThreadImplGpuMemoryBufferPerfTest have this as member?

+
+enum MemoryOperation { kMemoryOperationWrite, kMemoryOperationNoop };

[dshwang, 2015/06/25 10:52:11]

unused

+
+class BufferPerfTest {
+ public:
+  BufferPerfTest ()
+      : gpu_memory_buffer_(nullptr),
+        num_planes_(0),
+        format_(gfx::GpuMemoryBuffer::BGRA_8888),
+        buffer_size_(4, 4) {}
+
+  void Allocate(void) {
+    gpu_memory_buffer_ =
+      child_gpu_memory_buffer_manager_->AllocateGpuMemoryBuffer(
+          buffer_size_, format_, gfx::GpuMemoryBuffer::MAP);
+    ASSERT_TRUE(gpu_memory_buffer_);
+
+    EXPECT_EQ(format_, gpu_memory_buffer_->GetFormat());
+
+    num_planes_ =
+      GpuMemoryBufferImpl::NumberOfPlanesForGpuMemoryBufferFormat(format_);
+  }
+
+  void Map(scoped_ptr<void* []> const &planes, std::string operation_name) {
+    std::string flag_name = NativeBufferFlagName(native_buffer_flag_);
+
+    base::TimeTicks start = base::TimeTicks::Now();
+    bool rv = gpu_memory_buffer_->Map(planes.get());
+    base::TimeTicks end = base::TimeTicks::Now();
+    ASSERT_TRUE(rv);
+    EXPECT_TRUE(gpu_memory_buffer_->IsMapped());
+
+    // TODO(vignatti): get the mean time and print to stdout only once. At the
+    // moment it's being useful to check individual runs though cause for
+    // example VGEM has way worse performance on its first runs (got check why).
+    perf_test::PrintResult(

[dshwang, 2015/06/25 10:52:11]

This test measures only Map time even in "time_to_execute_map_native: Write"
case. In addition, it's too verbose. Could you print avg or sum time after
kNumRuns times execution?

+        "time_to_execute_map",
+        flag_name,
+        operation_name,
+        static_cast<size_t>((end - start).InMicroseconds()),
+        "us", true);
+  }
+
+  void Unmap(void) {
+    gpu_memory_buffer_->Unmap();
+    EXPECT_FALSE(gpu_memory_buffer_->IsMapped());
+  }
+
+  void Write(scoped_ptr<void* []> const &planes) {
+    // Get stride.
+    scoped_ptr<int[]> strides(new int[num_planes_]);
+    gpu_memory_buffer_->GetStride(strides.get());
+
+    for (size_t plane = 0; plane < num_planes_; ++plane) {
+      size_t row_size_in_bytes = 0;
+      EXPECT_TRUE(GpuMemoryBufferImpl::RowSizeInBytes(buffer_size_.width(),
+            format_, plane, &row_size_in_bytes));
+
+      scoped_ptr<char[]> data(new char[row_size_in_bytes]);
+      memset(data.get(), 0x2a + plane, row_size_in_bytes);
+
+      size_t height = buffer_size_.height() /
+        GpuMemoryBufferImpl::SubsamplingFactor(format_, plane);
+      for (size_t y = 0; y < height; ++y) {
+        // Copy |data| to row |y| of |plane| and verify result.
+        memcpy(static_cast<char*>(planes[plane]) + y * strides[plane],
+               data.get(),
+               row_size_in_bytes);
+#if defined(NDEBUG)
+        EXPECT_EQ(memcmp(static_cast<char*>(planes[plane]) + y * strides[plane],
+              data.get(), row_size_in_bytes),
+            0);
+#endif
+      }
+    }
+  }
+
+  size_t GetNumPlanes() { return num_planes_; }
+
+ private:
+  scoped_ptr<gfx::GpuMemoryBuffer> gpu_memory_buffer_;
+  size_t num_planes_;
+  gfx::GpuMemoryBuffer::Format format_;
+  gfx::Size buffer_size_;
+};
+
+class ChildThreadImplBrowserTest : public ContentBrowserTest {
+ public:
+  ChildThreadImplBrowserTest() {}
+
+  // Overridden from BrowserTestBase:
+  void SetUpCommandLine(base::CommandLine* command_line) override {
+    command_line->AppendSwitch(switches::kSingleProcess);
+  }
+  void SetUpOnMainThread() override {
+    NavigateToURL(shell(), GURL(url::kAboutBlankURL));
+    PostTaskToInProcessRendererAndWait(
+        base::Bind(&ChildThreadImplBrowserTest::SetUpOnChildThread, this));
+  }
+
+ private:
+  void SetUpOnChildThread() {
+    child_gpu_memory_buffer_manager_ =
+        ChildThreadImpl::current()->gpu_memory_buffer_manager();
+  }
+
+};
+
+class ChildThreadImplGpuMemoryBufferPerfTest
+    : public ChildThreadImplBrowserTest,
+      public testing::WithParamInterface<
+          ::testing::tuple<NativeBufferFlag>> {
+ public:
+  ChildThreadImplGpuMemoryBufferPerfTest() {}
+
+  // Overridden from BrowserTestBase:
+  void SetUpCommandLine(base::CommandLine* command_line) override {
+    ChildThreadImplBrowserTest::SetUpCommandLine(command_line);
+    native_buffer_flag_ = ::testing::get<0>(GetParam());
+    switch (native_buffer_flag_) {
+      case kEnableNativeBuffers:
+        command_line->AppendSwitch(switches::kEnableNativeGpuMemoryBuffers);
+        break;
+      case kDisableNativeBuffers:
+        break;
+    }
+  }
+
+ protected:
+  scoped_ptr<BufferPerfTest> buffer_;
+
+ private:
+  DISALLOW_COPY_AND_ASSIGN(ChildThreadImplGpuMemoryBufferPerfTest);
+};
+
+IN_PROC_BROWSER_TEST_P(ChildThreadImplGpuMemoryBufferPerfTest,
+                       Write) {
+  buffer_.reset(new BufferPerfTest());
+  buffer_->Allocate();
+
+  scoped_ptr<void* []> planes(new void* [buffer_->GetNumPlanes()]);
+
+  for (int i = 0; i < kNumRuns; ++i) {
+    buffer_->Map(planes, "Write");
+    buffer_->Write(planes);

[dshwang, 2015/06/25 10:52:11]

I think Map and Unmap should be out of 'for block', because Map seems to be as
expensive as Write.

In addition, this test print out only Map time.

+    buffer_->Unmap();
+  }
+}
+
+IN_PROC_BROWSER_TEST_P(ChildThreadImplGpuMemoryBufferPerfTest,
+                       Map) {
+  buffer_.reset(new BufferPerfTest());
+  buffer_->Allocate();
+
+  scoped_ptr<void* []> planes(new void* [buffer_->GetNumPlanes()]);
+
+  for (int i = 0; i < kNumRuns; ++i) {
+    buffer_->Map(planes, "Map");
+    buffer_->Unmap();
+  }
+}
+
+INSTANTIATE_TEST_CASE_P(
+    ChildThreadImplGpuMemoryBufferPerfTests,
+    ChildThreadImplGpuMemoryBufferPerfTest,
+    ::testing::Values(kDisableNativeBuffers, kEnableNativeBuffers));
+}  // namespace
+}  // namespace content