content: perf tests for GpuMemoryBuffers mapping and data coherency

content/common/gpu/client/gpu_memory_buffer_impl_perftest.cc

+// 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 a native buffer object is mapped into the
+// CPU. In particular this test aims to capture the effects of data coherency
+// and could be extended further to answer the following:
+//
+// - measure memory mapping performance of GpuMemoryBuffer using shared memory
+//  (fallback case) and also native implementation.
+// - 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 "content/common/gpu/client/gpu_memory_buffer_impl.h"
+
+#include "base/bind.h"
+#include "content/common/gpu/gpu_memory_buffer_factory.h"
+#include "testing/gtest/include/gtest/gtest.h"
+#include "testing/perf/perf_test.h"
+
+namespace content {
+namespace {
+
+const int kNumRuns = 1000;
+const int kClientId = 1;
+
+// Do we want to move this into a public API?
+std::string GpuMemoryBufferTypeName(gfx::GpuMemoryBufferType type) {
+  switch (type) {
+    case gfx::SHARED_MEMORY_BUFFER:
+      return "shared_memory";
+    case gfx::IO_SURFACE_BUFFER:
+      return "io_surface";
+    case gfx::SURFACE_TEXTURE_BUFFER:
+      return "surface_texture";
+    case gfx::OZONE_NATIVE_BUFFER:
+      return "ozone_native";
+    default:
+      NOTREACHED();
+      return "";
+  }
+}
+
+class GpuMemoryBufferPerfTest
+    : public testing::TestWithParam<gfx::GpuMemoryBufferType> {
+ public:
+  GpuMemoryBufferPerfTest()
+    : buffer_count_(0),
+      buffer_size_(128, 128),
+      buffer_(nullptr),
+      factory_(nullptr) {}
+
+  // Overridden from testing::Test:
+  void SetUp() override {
+    factory_ = GpuMemoryBufferFactory::Create(GetParam());
+  }
+  void TearDown() override { factory_.reset(); }
+
+  gfx::GpuMemoryBufferHandle CreateGpuMemoryBuffer(
+      gfx::GpuMemoryBufferId id,
+      const gfx::Size& size,
+      gfx::GpuMemoryBuffer::Format format,
+      gfx::GpuMemoryBuffer::Usage usage) {
+    ++buffer_count_;
+    return factory_->CreateGpuMemoryBuffer(id, size, format, usage, kClientId,
+                                           gfx::kNullPluginWindow);
+  }
+
+  void DestroyGpuMemoryBuffer(gfx::GpuMemoryBufferId id, uint32 sync_point) {
+    factory_->DestroyGpuMemoryBuffer(id, kClientId);
+    DCHECK_GT(buffer_count_, 0);
+    --buffer_count_;
+  }
+
+  void CreateTestGpuMemoryBuffer(void) {
+    const int kBufferId = 1;
+    const gfx::GpuMemoryBuffer::Format format =
+      gfx::GpuMemoryBuffer::BGRA_8888;
+    const gfx::GpuMemoryBuffer::Usage usage = gfx::GpuMemoryBuffer::MAP;
+    total_time_ = 0;
+
+    handle_ =
+      CreateGpuMemoryBuffer(kBufferId, buffer_size_, format, usage);
+
+    buffer_ =
+        GpuMemoryBufferImpl::CreateFromHandle(
+          handle_,
+          buffer_size_, format, usage,
+          base::Bind(&GpuMemoryBufferPerfTest::DestroyGpuMemoryBuffer,
+            base::Unretained(this), kBufferId));
+    ASSERT_TRUE(buffer_);
+    EXPECT_FALSE(buffer_->IsMapped());
+  }
+
+  void DestroyTestGpuMemoryBuffer(void) {
+    buffer_.reset();
+  }
+
+  void BufferMap(scoped_ptr<void* []> const &buffers) {
+    base::TimeTicks start = base::TimeTicks::Now();
+    bool rv = buffer_->Map(buffers.get());
+    total_time_ +=
+      static_cast<size_t>((base::TimeTicks::Now() - start).InMicroseconds());
+    ASSERT_TRUE(rv);
+    EXPECT_TRUE(buffer_->IsMapped());
+  }
+
+  void BufferUnmap(void) {
+    buffer_->Unmap();
+    EXPECT_FALSE(buffer_->IsMapped());
+  }
+
+  void BufferRead(volatile uint32_t *ptr) {
+    int j;
+    int size = buffer_size_.width() * buffer_size_.height();
+    int x = 0;
+
+    for (j = 0; j < static_cast<int>(size/sizeof(*ptr)); j++)
+      x += ptr[j];
+
+    /* force overtly clever gcc to actually compute x */
+    ptr[0] = x;
+  }
+
+  void BufferWrite(volatile uint32_t *ptr) {
+    int j;
+    int size = buffer_size_.width() * buffer_size_.height();
+
+    for (j = 0; j < static_cast<int>(size/sizeof(*ptr)); j++)
+      ptr[j] = j;
+  }
+
+  void PrintTestGpuMemoryBuffer(std::string name) {
+    // Sometimes the first runs have worse performance so it might be useful to
+    // print out the standard deviation as well to capture this discrepancy.
+    perf_test::PrintResult(
+        "gpu_memory_buffer_time_",
+        GpuMemoryBufferTypeName(handle_.type),
+        name,
+        total_time_ / static_cast<double>(kNumRuns),
+        "us/task",
+        true);
+  }
+  double total_time_;
+
+ private:
+  int buffer_count_;
+  gfx::Size buffer_size_;
+  gfx::GpuMemoryBufferHandle handle_;
+  scoped_ptr<GpuMemoryBufferImpl> buffer_;
+  scoped_ptr<GpuMemoryBufferFactory> factory_;
+};
+
+TEST_P(GpuMemoryBufferPerfTest, MapUnmap) {
+  scoped_ptr<void*[]> mapped_buffers(new void*[0]);
+
+  CreateTestGpuMemoryBuffer();
+
+  base::TimeTicks start = base::TimeTicks::Now();
+  for (int i = 0; i < kNumRuns; ++i) {
+    BufferMap(mapped_buffers);
+    BufferUnmap();
+  }
+  total_time_ +=
+    static_cast<size_t>((base::TimeTicks::Now() - start).InMicroseconds());
+
+  DestroyTestGpuMemoryBuffer();
+  PrintTestGpuMemoryBuffer("map & unmap ");
+}
+
+TEST_P(GpuMemoryBufferPerfTest, Read) {
+  scoped_ptr<void*[]> mapped_buffers(new void*[0]);
+
+  CreateTestGpuMemoryBuffer();
+
+  base::TimeTicks start = base::TimeTicks::Now();
+  for (int i = 0; i < kNumRuns; ++i) {
+    BufferMap(mapped_buffers);
+    BufferRead(static_cast<uint32_t*>(mapped_buffers[0]));
+    BufferUnmap();
+  }
+  total_time_ +=
+    static_cast<size_t>((base::TimeTicks::Now() - start).InMicroseconds());
+
+  DestroyTestGpuMemoryBuffer();
+  PrintTestGpuMemoryBuffer("read ");
+}
+
+TEST_P(GpuMemoryBufferPerfTest, Write) {
+  scoped_ptr<void*[]> mapped_buffers(new void*[0]);
+
+  CreateTestGpuMemoryBuffer();
+
+  base::TimeTicks start = base::TimeTicks::Now();
+  for (int i = 0; i < kNumRuns; ++i) {
+    BufferMap(mapped_buffers);
+    BufferWrite(static_cast<uint32_t*>(mapped_buffers[0]));
+    BufferUnmap();
+  }
+  total_time_ +=
+    static_cast<size_t>((base::TimeTicks::Now() - start).InMicroseconds());
+
+  DestroyTestGpuMemoryBuffer();
+  PrintTestGpuMemoryBuffer("write ");
+}
+
+// TODO: description
+TEST_P(GpuMemoryBufferPerfTest, ReadWrite) {
+  scoped_ptr<void*[]> mapped_buffers(new void*[0]);
+
+  CreateTestGpuMemoryBuffer();
+
+  base::TimeTicks start = base::TimeTicks::Now();
+  for (int i = 0; i < kNumRuns; ++i) {
+    BufferMap(mapped_buffers);
+    BufferRead(static_cast<uint32_t*>(mapped_buffers[0]));
+    BufferWrite(static_cast<uint32_t*>(mapped_buffers[0]));
+    BufferUnmap();
+  }
+  total_time_ +=
+    static_cast<size_t>((base::TimeTicks::Now() - start).InMicroseconds());
+
+  DestroyTestGpuMemoryBuffer();
+  PrintTestGpuMemoryBuffer("read & write ");
+}
+
+std::vector<gfx::GpuMemoryBufferType> GetSupportedGpuMemoryBufferTypes() {
+  std::vector<gfx::GpuMemoryBufferType> supported_types;
+  GpuMemoryBufferFactory::GetSupportedTypes(&supported_types);
+  return supported_types;
+}
+
+INSTANTIATE_TEST_CASE_P(
+    GpuMemoryBufferPerfTests,
+    GpuMemoryBufferPerfTest,
+    ::testing::ValuesIn(GetSupportedGpuMemoryBufferTypes()));
+}  // namespace
+}  // namespace content