OmxVideoDecodeAcceleratorTest improvements:

content/common/gpu/omx_video_decode_accelerator_unittest.cc

Index: content/common/gpu/omx_video_decode_accelerator_unittest.cc
diff --git a/content/common/gpu/omx_video_decode_accelerator_unittest.cc b/content/common/gpu/omx_video_decode_accelerator_unittest.cc
index 70f878a03c70fce3762051fcacaa300a06e0e34c..691d6da0320e6158059d37559498a1667d219965 100644
--- a/content/common/gpu/omx_video_decode_accelerator_unittest.cc
+++ b/content/common/gpu/omx_video_decode_accelerator_unittest.cc
@@ -13,9 +13,10 @@
 // - Finally actual TEST cases are at the bottom of this file, using the above
 //   infrastructure.
 
-#include <sys/types.h>
-#include <sys/stat.h>
 #include <fcntl.h>
+#include <math.h>
+#include <sys/stat.h>
+#include <sys/types.h>
 
 // Include gtest.h out of order because <X11/X.h> #define's Bool & None, which
 // gtest uses as struct names (inside a namespace).  This means that
@@ -48,9 +49,9 @@ enum {
   kFrameHeight = 240,
 };
 
-// Helper for managing X11, EGL, and GLES2 resources.  Because GL state is
-// thread-specific, all the methods of this class (except for ctor/dtor) CHECK
-// for being run on a single thread.
+// Helper for managing X11, EGL, and GLES2 resources.  Xlib is not thread-safe,
+// and GL state is thread-specific, so all the methods of this class (except for
+// ctor/dtor) ensure they're being run on a single thread.
 //
 // TODO(fischman): consider moving this into media/ if we can de-dup some of the
 // code that ends up getting copy/pasted all over the place (esp. the GL setup
@@ -60,46 +61,73 @@ class RenderingHelper {
   explicit RenderingHelper();
   ~RenderingHelper();
 
-  // Initialize all structures to prepare to render to a window of the specified
-  // dimensions.  CHECK-fails if any initialization step fails.  After this
-  // returns, texture creation and rendering (swaps) can be requested.
-  // This method can be called multiple times, in which case all
-  // previously-acquired resources and initializations are discarded.
-  void Initialize(int width, int height, base::WaitableEvent* done);
+  // Initialize all structures to prepare to render to one or more windows of
+  // the specified dimensions.  CHECK-fails if any initialization step fails.
+  // After this returns, texture creation and rendering can be requested.  This
+  // method can be called multiple times, in which case all previously-acquired
+  // resources and initializations are discarded.  If |suppress_swap_to_display|
+  // then all the usual work is done, except for the final swap of the EGL
+  // surface to the display.  This cuts test times over 50% so is worth doing
+  // when testing non-rendering-related aspects.
+  void Initialize(bool suppress_swap_to_display, int num_windows,
+                  int width, int height, base::WaitableEvent* done);
 
   // Undo the effects of Initialize() and signal |*done|.
   void UnInitialize(base::WaitableEvent* done);
 
-  // Return a newly-created GLES2 texture id.
-  GLuint CreateTexture();
+  // Return a newly-created GLES2 texture id rendering to a specific window, and
+  // signal |*done|.
+  void CreateTexture(int window_id, GLuint* texture_id,
+                     base::WaitableEvent* done);
 
-  // Render |texture_id| to the screen.
+  // Render |texture_id| to the screen (unless |suppress_swap_to_display_|).
   void RenderTexture(GLuint texture_id);
 
+  // Delete |texture_id|.
+  void DeleteTexture(GLuint texture_id);
+
   EGLDisplay egl_display() { return egl_display_; }
   EGLContext egl_context() { return egl_context_; }
 
  private:
+  // Zero-out internal state.  Helper for ctor & UnInitialize().
+  void Clear();
+
+  bool suppress_swap_to_display_;
   int width_;
   int height_;
   Display* x_display_;
-  Window x_window_;
+  std::vector<Window> x_windows_;
   EGLDisplay egl_display_;
   EGLContext egl_context_;
-  EGLSurface egl_surface_;
-  // Since GL carries per-thread state, we ensure all operations are carried out
-  // on the same thread by remembering where we were Initialized.
+  std::vector<EGLSurface> egl_surfaces_;
+  std::map<GLuint, int> texture_id_to_surface_index_;
+  // We ensure all operations are carried out on the same thread by remembering
+  // where we were Initialized.
   MessageLoop* message_loop_;
 };
 
 RenderingHelper::RenderingHelper() {
-  memset(this, 0, sizeof(this));
+  Clear();
 }
 
 RenderingHelper::~RenderingHelper() {
   CHECK_EQ(width_, 0) << "Must call UnInitialize before dtor.";
 }
 
+void RenderingHelper::Clear() {
+  suppress_swap_to_display_ = false;
+  width_ = 0;
+  height_ = 0;
+  x_display_ = NULL;
+  x_windows_.clear();
+  egl_display_ = EGL_NO_DISPLAY;
+  egl_context_ = EGL_NO_CONTEXT;
+  egl_surfaces_.clear();
+  texture_id_to_surface_index_.clear();
+  message_loop_ = NULL;
+}
+
 // Helper for Shader creation.
 static void CreateShader(
     GLuint program, GLenum type, const char* source, int size) {
@@ -119,7 +147,10 @@ static void CreateShader(
 }
 
 void RenderingHelper::Initialize(
-    int width, int height, base::WaitableEvent* done) {
+    bool suppress_swap_to_display,
+    int num_windows,
+    int width, int height,
+    base::WaitableEvent* done) {
   // Use width_ != 0 as a proxy for the class having already been
   // Initialize()'d, and UnInitialize() before continuing.
   if (width_) {
@@ -128,36 +159,27 @@ void RenderingHelper::Initialize(
     done.Wait();
   }
 
+  suppress_swap_to_display_ = suppress_swap_to_display;
   CHECK_GT(width, 0);
   CHECK_GT(height, 0);
   width_ = width;
   height_ = height;
   message_loop_ = MessageLoop::current();
+  CHECK_GT(num_windows, 0);
 
-  // X11 initialization.
+  // Per-display X11 & EGL initialization.
   CHECK(x_display_ = XOpenDisplay(NULL));
   int depth = DefaultDepth(x_display_, DefaultScreen(x_display_));
   XSetWindowAttributes window_attributes;
   window_attributes.background_pixel =
       BlackPixel(x_display_, DefaultScreen(x_display_));
   window_attributes.override_redirect = true;
-  x_window_ = XCreateWindow(
-      x_display_, DefaultRootWindow(x_display_),
-      100, 100, /* x/y of top-left corner */
-      width_, height_,
-      0 /* border width */,
-      depth, CopyFromParent /* class */, CopyFromParent /* visual */,
-      (CWBackPixel | CWOverrideRedirect), &window_attributes);
-  XStoreName(x_display_, x_window_, "OmxVideoDecodeAcceleratorTest");
-  XSelectInput(x_display_, x_window_, ExposureMask);
-  XMapWindow(x_display_, x_window_);
 
-  // EGL initialization.
   egl_display_ = eglGetDisplay(x_display_);
   EGLint major;
   EGLint minor;
   CHECK(eglInitialize(egl_display_, &major, &minor)) << eglGetError();
-  EGLint rgba8888[] = {
+  static EGLint rgba8888[] = {
     EGL_RED_SIZE, 8,
     EGL_GREEN_SIZE, 8,
     EGL_BLUE_SIZE, 8,
@@ -165,20 +187,39 @@ void RenderingHelper::Initialize(
     EGL_SURFACE_TYPE, EGL_WINDOW_BIT,
     EGL_NONE,
   };
-
   EGLConfig egl_config;
   int num_configs;
   CHECK(eglChooseConfig(egl_display_, rgba8888, &egl_config, 1, &num_configs))
       << eglGetError();
   CHECK_GE(num_configs, 1);
-  egl_surface_ =
-      eglCreateWindowSurface(egl_display_, egl_config, x_window_, NULL);
-  CHECK_NE(egl_surface_, EGL_NO_SURFACE);
-  EGLint context_attribs[] = {EGL_CONTEXT_CLIENT_VERSION, 2, EGL_NONE};
+  static EGLint context_attribs[] = {EGL_CONTEXT_CLIENT_VERSION, 2, EGL_NONE};
   egl_context_ = eglCreateContext(
       egl_display_, egl_config, EGL_NO_CONTEXT, context_attribs);
-  CHECK(eglMakeCurrent(egl_display_, egl_surface_, egl_surface_, egl_context_))
-      << eglGetError();
+  CHECK_NE(egl_context_, EGL_NO_CONTEXT) << eglGetError();
+
+  // Per-window/surface X11 & EGL initialization.
+  for (int i = 0; i < num_windows; ++i) {
+    // Arrange X windows side by side whimsically.
+    int top_left_x = (width + 50) * i;
+    int top_left_y = 100 + (i % 2) * 250;
+    Window x_window = XCreateWindow(
+      x_display_, DefaultRootWindow(x_display_),
+      top_left_x, top_left_y, width_, height_,
+      0 /* border width */,
+      depth, CopyFromParent /* class */, CopyFromParent /* visual */,
+      (CWBackPixel | CWOverrideRedirect), &window_attributes);
+    x_windows_.push_back(x_window);
+    XStoreName(x_display_, x_window, "OmxVideoDecodeAcceleratorTest");
+    XSelectInput(x_display_, x_window, ExposureMask);
+    XMapWindow(x_display_, x_window);
+
+    EGLSurface egl_surface =
+        eglCreateWindowSurface(egl_display_, egl_config, x_window, NULL);
+    egl_surfaces_.push_back(egl_surface);
+    CHECK_NE(egl_surface, EGL_NO_SURFACE);
+  }
+  CHECK(eglMakeCurrent(egl_display_, egl_surfaces_[0],
+                       egl_surfaces_[0], egl_context_)) << eglGetError();
 
   // GLES2 initialization.  Note: This is pretty much copy/pasted from
   // media/tools/player_x11/gles_video_renderer.cc, with some simplification
@@ -204,9 +245,10 @@ void RenderingHelper::Initialize(
       }
                                                      );
   GLuint program = glCreateProgram();
-  CreateShader(program, GL_VERTEX_SHADER, kVertexShader, sizeof(kVertexShader));
+  CreateShader(program, GL_VERTEX_SHADER,
+               kVertexShader, arraysize(kVertexShader));
   CreateShader(program, GL_FRAGMENT_SHADER,
-               kFragmentShaderEgl, sizeof(kFragmentShaderEgl));
+               kFragmentShaderEgl, arraysize(kFragmentShaderEgl));
   glLinkProgram(program);
   int result = GL_FALSE;
   glGetProgramiv(program, GL_LINK_STATUS, &result);
@@ -226,35 +268,50 @@ void RenderingHelper::Initialize(
   glEnableVertexAttribArray(tc_location);
   glVertexAttribPointer(tc_location, 2, GL_FLOAT, GL_FALSE, 0,
                         kTextureCoordsEgl);
+
   done->Signal();
 }
 
 void RenderingHelper::UnInitialize(base::WaitableEvent* done) {
   CHECK_EQ(MessageLoop::current(), message_loop_);
   // Destroy resources acquired in Initialize, in reverse-acquisition order.
-  CHECK(eglMakeCurrent(
-      egl_display_, EGL_NO_SURFACE, EGL_NO_SURFACE, EGL_NO_CONTEXT));
+  CHECK(eglMakeCurrent(egl_display_, EGL_NO_SURFACE, EGL_NO_SURFACE,
+                       EGL_NO_CONTEXT)) << eglGetError();
   CHECK(eglDestroyContext(egl_display_, egl_context_));
-  CHECK(eglDestroySurface(egl_display_, egl_surface_));
+  for (size_t i = 0; i < egl_surfaces_.size(); ++i)
+    CHECK(eglDestroySurface(egl_display_, egl_surfaces_[i]));
   CHECK(eglTerminate(egl_display_));
-  CHECK(XUnmapWindow(x_display_, x_window_));
-  CHECK(XDestroyWindow(x_display_, x_window_));
+  for (size_t i = 0; i < x_windows_.size(); ++i) {
+    CHECK(XUnmapWindow(x_display_, x_windows_[i]));
+    CHECK(XDestroyWindow(x_display_, x_windows_[i]));
+  }
   // Mimic newly-created object.
-  memset(this, 0, sizeof(this));
+  Clear();
   done->Signal();
 }
 
-GLuint RenderingHelper::CreateTexture() {
-  CHECK_EQ(MessageLoop::current(), message_loop_);
-  GLuint texture_id;
-  glGenTextures(1, &texture_id);
-  glBindTexture(GL_TEXTURE_2D, texture_id);
+void RenderingHelper::CreateTexture(int window_id, GLuint* texture_id,
+                                    base::WaitableEvent* done) {
+  if (MessageLoop::current() != message_loop_) {
+    message_loop_->PostTask(
+        FROM_HERE,
+        base::Bind(&RenderingHelper::CreateTexture, base::Unretained(this),
+                   window_id, texture_id, done));
+    return;
+  }
+  CHECK(eglMakeCurrent(egl_display_, egl_surfaces_[window_id],
+                       egl_surfaces_[window_id], egl_context_))
+      << eglGetError();
+  glGenTextures(1, texture_id);
+  glBindTexture(GL_TEXTURE_2D, *texture_id);
   glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, width_, height_, 0, GL_RGBA,
                GL_UNSIGNED_BYTE, NULL);
   glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
   glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
   CHECK_EQ(static_cast<int>(glGetError()), GL_NO_ERROR);
-  return texture_id;
+  CHECK(texture_id_to_surface_index_.insert(
+      std::make_pair(*texture_id, window_id)).second);
+  done->Signal();
 }
 
 void RenderingHelper::RenderTexture(GLuint texture_id) {
@@ -263,8 +320,20 @@ void RenderingHelper::RenderTexture(GLuint texture_id) {
   glBindTexture(GL_TEXTURE_2D, texture_id);
   glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
   DCHECK_EQ(static_cast<int>(glGetError()), GL_NO_ERROR);
-  eglSwapBuffers(egl_display_, egl_surface_);
   DCHECK_EQ(static_cast<int>(eglGetError()), EGL_SUCCESS);
+  if (!suppress_swap_to_display_) {
+    int window_id = texture_id_to_surface_index_[texture_id];
+    CHECK(eglMakeCurrent(egl_display_, egl_surfaces_[window_id],
+                         egl_surfaces_[window_id], egl_context_))
+        << eglGetError();
+    eglSwapBuffers(egl_display_, egl_surfaces_[window_id]);
+  }
+  DCHECK_EQ(static_cast<int>(eglGetError()), EGL_SUCCESS);
+}
+
+void RenderingHelper::DeleteTexture(GLuint texture_id) {
+  glDeleteTextures(1, &texture_id);
+  DCHECK_EQ(static_cast<int>(glGetError()), GL_NO_ERROR);
 }
 
 // State of the EglRenderingVDAClient below.
@@ -322,8 +391,12 @@ class EglRenderingVDAClient : public VideoDecodeAccelerator::Client {
  public:
   // Doesn't take ownership of |note| or |encoded_data|, which must outlive
   // |*this|.
-  EglRenderingVDAClient(ClientStateNotification* note,
-                        std::string* encoded_data);
+  EglRenderingVDAClient(RenderingHelper* rendering_helper,
+                        int rendering_window_id,
+                        ClientStateNotification* note,
+                        std::string* encoded_data,
+                        int num_NALUs_per_decode,
+                        bool suppress_swap_to_display);
   virtual ~EglRenderingVDAClient();
 
   // VideoDecodeAccelerator::Client implementation.
@@ -350,20 +423,29 @@ class EglRenderingVDAClient : public VideoDecodeAccelerator::Client {
   VideoDecodeAccelerator::Error error() { return error_; }
   int num_done_bitstream_buffers() { return num_done_bitstream_buffers_; }
   int num_decoded_frames() { return num_decoded_frames_; }
-  MessageLoop* message_loop() { return thread_.message_loop(); }
-  EGLDisplay egl_display() { return rendering_helper_.egl_display(); }
-  EGLContext egl_context() { return rendering_helper_.egl_context(); }
+  EGLDisplay egl_display() { return rendering_helper_->egl_display(); }
+  EGLContext egl_context() { return rendering_helper_->egl_context(); }
+  double frames_per_second();
 
  private:
+  typedef std::map<int, media::GLESBuffer*> PictureBufferById;
+
   void SetState(ClientState new_state) {
     note_->Notify(new_state);
     state_ = new_state;
   }
 
-  // Request decode of the next NALU in the encoded data.
-  void DecodeNextNALU();
+  // Compute & return in |*end_pos| the end position for the next batch of NALUs
+  // to ship to the decoder (based on |start_pos| & |num_NALUs_per_decode_|).
+  void GetRangeForNextNALUs(size_t start_pos, size_t* end_pos);
+
+  // Request decode of the next batch of NALUs in the encoded data.
+  void DecodeNextNALUs();
 
+  RenderingHelper* rendering_helper_;
+  int rendering_window_id_;
   const std::string* encoded_data_;
+  const int num_NALUs_per_decode_;
   size_t encoded_data_next_pos_to_decode_;
   int next_bitstream_buffer_id_;
   ClientStateNotification* note_;
@@ -372,41 +454,28 @@ class EglRenderingVDAClient : public VideoDecodeAccelerator::Client {
   VideoDecodeAccelerator::Error error_;
   int num_decoded_frames_;
   int num_done_bitstream_buffers_;
-  std::map<int, media::GLESBuffer*> picture_buffers_by_id_;
-  // Required for Thread to work.  Not used otherwise.
-  base::ShadowingAtExitManager at_exit_manager_;
-  base::Thread thread_;
-  RenderingHelper rendering_helper_;
+  PictureBufferById picture_buffers_by_id_;
+  base::TimeTicks initialize_done_ticks_;
+  base::TimeTicks last_frame_delivered_ticks_;
 };
 
-EglRenderingVDAClient::EglRenderingVDAClient(ClientStateNotification* note,
-                                             std::string* encoded_data)
-    : encoded_data_(encoded_data), encoded_data_next_pos_to_decode_(0),
-      next_bitstream_buffer_id_(0), note_(note), decoder_(NULL),
-      state_(CS_CREATED),
+EglRenderingVDAClient::EglRenderingVDAClient(RenderingHelper* rendering_helper,
+                                             int rendering_window_id,
+                                             ClientStateNotification* note,
+                                             std::string* encoded_data,
+                                             int num_NALUs_per_decode,
+                                             bool suppress_swap_to_display)
+    : rendering_helper_(rendering_helper),
+      rendering_window_id_(rendering_window_id),
+      encoded_data_(encoded_data), num_NALUs_per_decode_(num_NALUs_per_decode),
+      encoded_data_next_pos_to_decode_(0), next_bitstream_buffer_id_(0),
+      note_(note), decoder_(NULL), state_(CS_CREATED),
       error_(VideoDecodeAccelerator::VIDEODECODERERROR_NONE),
-      num_decoded_frames_(0), num_done_bitstream_buffers_(0),
-      thread_("EglRenderingVDAClientThread") {
-  CHECK(thread_.Start());
-  base::WaitableEvent done(false, false);
-  message_loop()->PostTask(
-      FROM_HERE,
-      base::Bind(&RenderingHelper::Initialize,
-                 base::Unretained(&rendering_helper_),
-                 static_cast<int>(kFrameWidth), static_cast<int>(kFrameHeight),
-                 &done));
-  done.Wait();
+      num_decoded_frames_(0), num_done_bitstream_buffers_(0) {
+  CHECK_GT(num_NALUs_per_decode, 0);
 }
 
 EglRenderingVDAClient::~EglRenderingVDAClient() {
-  base::WaitableEvent done(false, false);
-  message_loop()->PostTask(
-      FROM_HERE,
-      base::Bind(&RenderingHelper::UnInitialize,
-                 base::Unretained(&rendering_helper_),
-                 &done));
-  done.Wait();
-  thread_.Stop();
   STLDeleteValues(&picture_buffers_by_id_);
   SetState(CS_DESTROYED);
 }
@@ -415,7 +484,6 @@ void EglRenderingVDAClient::ProvidePictureBuffers(
     uint32 requested_num_of_buffers,
     const gfx::Size& dimensions,
     VideoDecodeAccelerator::MemoryType type) {
-  CHECK_EQ(message_loop(), MessageLoop::current());
   CHECK_EQ(type, VideoDecodeAccelerator::PICTUREBUFFER_MEMORYTYPE_GL_TEXTURE);
   std::vector<media::GLESBuffer> buffers;
   CHECK_EQ(dimensions.width(), kFrameWidth);
@@ -423,7 +491,10 @@ void EglRenderingVDAClient::ProvidePictureBuffers(
 
   for (uint32 i = 0; i < requested_num_of_buffers; ++i) {
     uint32 id = picture_buffers_by_id_.size();
-    GLuint texture_id = rendering_helper_.CreateTexture();
+    GLuint texture_id;
+    base::WaitableEvent done(false, false);
+    rendering_helper_->CreateTexture(rendering_window_id_, &texture_id, &done);
+    done.Wait();
     // TODO(fischman): context_id is always 0.  Can it be removed from the API?
     // (since it's always inferrable from context).
     media::GLESBuffer* buffer =
@@ -437,60 +508,58 @@ void EglRenderingVDAClient::ProvidePictureBuffers(
 }
 
 void EglRenderingVDAClient::DismissPictureBuffer(int32 picture_buffer_id) {
-  CHECK_EQ(message_loop(), MessageLoop::current());
-  delete picture_buffers_by_id_[picture_buffer_id];
-  CHECK_EQ(1U, picture_buffers_by_id_.erase(picture_buffer_id));
+  PictureBufferById::iterator it =
+      picture_buffers_by_id_.find(picture_buffer_id);
+  DCHECK(it != picture_buffers_by_id_.end());
+  rendering_helper_->DeleteTexture(it->second->texture_id());
+  delete it->second;
+  picture_buffers_by_id_.erase(it);
 }
 
 void EglRenderingVDAClient::PictureReady(const media::Picture& picture) {
-  CHECK_EQ(message_loop(), MessageLoop::current());
+  last_frame_delivered_ticks_ = base::TimeTicks::Now();
 
-  // Because we feed the decoder one NALU at a time, we can be sure each frame
-  // comes from a bitstream buffer numbered at least as high as our current
-  // decoded frame's index, and less than the id of the next bitstream buffer
-  // we'll send for decoding.  Assert that.
-  CHECK_GE(picture.bitstream_buffer_id(), num_decoded_frames_);
+  // Because we feed the decoder a limited number of NALUs at a time, we can be
+  // sure each the bitstream buffer from which a frame comes has a limited
+  // range.  Assert that.
+  CHECK_GE((picture.bitstream_buffer_id() + 1) * num_NALUs_per_decode_,
+           num_decoded_frames_);
   CHECK_LE(picture.bitstream_buffer_id(), next_bitstream_buffer_id_);
   ++num_decoded_frames_;
 
   media::GLESBuffer* gles_buffer =
       picture_buffers_by_id_[picture.picture_buffer_id()];
   CHECK(gles_buffer);
-  rendering_helper_.RenderTexture(gles_buffer->texture_id());
+  rendering_helper_->RenderTexture(gles_buffer->texture_id());
 
   decoder_->ReusePictureBuffer(picture.picture_buffer_id());
 }
 
 void EglRenderingVDAClient::NotifyInitializeDone() {
-  CHECK_EQ(message_loop(), MessageLoop::current());
   SetState(CS_INITIALIZED);
-  DecodeNextNALU();
+  initialize_done_ticks_ = base::TimeTicks::Now();
+  DecodeNextNALUs();
 }
 
 void EglRenderingVDAClient::NotifyEndOfStream() {
-  CHECK_EQ(message_loop(), MessageLoop::current());
   SetState(CS_DONE);
 }
 
 void EglRenderingVDAClient::NotifyEndOfBitstreamBuffer(
     int32 bitstream_buffer_id) {
-  CHECK_EQ(message_loop(), MessageLoop::current());
   ++num_done_bitstream_buffers_;
-  DecodeNextNALU();
+  DecodeNextNALUs();
 }
 
 void EglRenderingVDAClient::NotifyFlushDone() {
-  CHECK_EQ(message_loop(), MessageLoop::current());
   SetState(CS_FLUSHED);
 }
 
 void EglRenderingVDAClient::NotifyAbortDone() {
-  CHECK_EQ(message_loop(), MessageLoop::current());
   SetState(CS_ABORTED);
 }
 
 void EglRenderingVDAClient::NotifyError(VideoDecodeAccelerator::Error error) {
-  CHECK_EQ(message_loop(), MessageLoop::current());
   SetState(CS_ERROR);
   error_ = error;
 }
@@ -506,20 +575,31 @@ static bool LookingAtNAL(const std::string& encoded, size_t pos) {
       encoded[pos + 2] == 0 && encoded[pos + 3] == 1;
 }
 
-void EglRenderingVDAClient::DecodeNextNALU() {
+void EglRenderingVDAClient::GetRangeForNextNALUs(
+    size_t start_pos, size_t* end_pos) {
+  *end_pos = start_pos;
+  CHECK(LookingAtNAL(*encoded_data_, start_pos));
+  for (int i = 0; i < num_NALUs_per_decode_; ++i) {
+    *end_pos += 4;
+    while (*end_pos + 3 < encoded_data_->size() &&
+           !LookingAtNAL(*encoded_data_, *end_pos)) {
+      ++*end_pos;
+    }
+    if (*end_pos + 3 >= encoded_data_->size()) {
+      *end_pos = encoded_data_->size();
+      return;
+    }
+  }
+}
+
+void EglRenderingVDAClient::DecodeNextNALUs() {
   if (encoded_data_next_pos_to_decode_ == encoded_data_->size()) {
     decoder_->Flush();
     return;
   }
   size_t start_pos = encoded_data_next_pos_to_decode_;
-  CHECK(LookingAtNAL(*encoded_data_, start_pos));
-  size_t end_pos = encoded_data_next_pos_to_decode_ + 4;
-  while (end_pos + 3 < encoded_data_->size() &&
-         !LookingAtNAL(*encoded_data_, end_pos)) {
-    ++end_pos;
-  }
-  if (end_pos + 3 >= encoded_data_->size())
-    end_pos = encoded_data_->size();
+  size_t end_pos;
+  GetRangeForNextNALUs(start_pos, &end_pos);
 
   // Populate the shared memory buffer w/ the NALU, duplicate its handle, and
   // hand it off to the decoder.
@@ -535,52 +615,139 @@ void EglRenderingVDAClient::DecodeNextNALU() {
   encoded_data_next_pos_to_decode_ = end_pos;
 }
 
+double EglRenderingVDAClient::frames_per_second() {
+  base::TimeDelta delta = last_frame_delivered_ticks_ - initialize_done_ticks_;
+  CHECK_GT(delta.InSecondsF(), 0);
+  return num_decoded_frames_ / delta.InSecondsF();
+}
+
+// Test parameters:
+// - Number of NALUs per Decode() call.
+// - Number of concurrent decoders.
+class OmxVideoDecodeAcceleratorTest
+    : public ::testing::TestWithParam<std::pair<int, int> > {
+};
+
 // Test the most straightforward case possible: data is decoded from a single
 // chunk and rendered to the screen.
-TEST(OmxVideoDecodeAcceleratorTest, TestSimpleDecode) {
-  logging::SetMinLogLevel(-1);
-  ClientStateNotification note;
-  ClientState state;
-  // Read in the video data and hand it off to the client for later decoding.
+TEST_P(OmxVideoDecodeAcceleratorTest, TestSimpleDecode) {
+  // Can be useful for debugging VLOGs from OVDA.
+  // logging::SetMinLogLevel(-1);
+
+  // Required for Thread to work.  Not used otherwise.
+  base::ShadowingAtExitManager at_exit_manager;
+
+  const int num_NALUs_per_decode = GetParam().first;
+  const size_t num_concurrent_decoders = GetParam().second;
+
+  // Suppress EGL surface swapping in all but a few tests, to cut down overall
+  // test runtime.
+  const bool suppress_swap_to_display = num_NALUs_per_decode > 1;
+
+  std::vector<ClientStateNotification*> notes(num_concurrent_decoders, NULL);
+  std::vector<EglRenderingVDAClient*> clients(num_concurrent_decoders, NULL);
+  std::vector<OmxVideoDecodeAccelerator*> decoders(
+      num_concurrent_decoders, NULL);
+  // Read in the video data.
   std::string data_str;
   CHECK(file_util::ReadFileToString(FilePath(std::string("test-25fps.h264")),
                                     &data_str));
-  EglRenderingVDAClient client(&note, &data_str);
-  OmxVideoDecodeAccelerator decoder(&client, client.message_loop());
-  client.SetDecoder(&decoder);
-  decoder.SetEglState(client.egl_display(), client.egl_context());
-  ASSERT_EQ((state = note.Wait()), CS_DECODER_SET);
-
-
-  // Configure the decoder.
-  int32 config_array[] = {
-    media::VIDEOATTRIBUTEKEY_BITSTREAMFORMAT_FOURCC,
-    media::VIDEOCODECFOURCC_H264,
-    media::VIDEOATTRIBUTEKEY_BITSTREAMFORMAT_WIDTH, kFrameWidth,
-    media::VIDEOATTRIBUTEKEY_BITSTREAMFORMAT_HEIGHT, kFrameHeight,
-    media::VIDEOATTRIBUTEKEY_VIDEOCOLORFORMAT, media::VIDEOCOLORFORMAT_RGBA,
-    media::VIDEOATTRIBUTEKEY_TERMINATOR
-  };
-  std::vector<uint32> config(
-      config_array, config_array + arraysize(config_array));
-  CHECK(decoder.Initialize(config));
-  ASSERT_EQ((state = note.Wait()), CS_INITIALIZED);
-  // InitializeDone kicks off decoding inside the client, so we just need to
-  // wait for Flush.
-  ASSERT_EQ((state = note.Wait()), CS_FLUSHED);
-
-  EXPECT_EQ(client.num_decoded_frames(), 25 /* fps */ * 10 /* seconds */);
-  // Hard-coded the number of NALUs in the stream.  Icky.
-  EXPECT_EQ(client.num_done_bitstream_buffers(), 258);
-}
+
+  // Initialize the rendering helper.
+  base::Thread rendering_thread("EglRenderingVDAClientThread");
+  rendering_thread.Start();
+  RenderingHelper rendering_helper;
+
+  base::WaitableEvent done(false, false);
+  rendering_thread.message_loop()->PostTask(
+      FROM_HERE,
+      base::Bind(&RenderingHelper::Initialize,
+                 base::Unretained(&rendering_helper),
+                 suppress_swap_to_display, num_concurrent_decoders,
+                 static_cast<int>(kFrameWidth), static_cast<int>(kFrameHeight),
+                 &done));
+  done.Wait();
+
+  // First kick off all the decoders.
+  for (size_t index = 0; index < num_concurrent_decoders; ++index) {
+    ClientStateNotification* note = new ClientStateNotification();
+    notes[index] = note;
+    EglRenderingVDAClient* client = new EglRenderingVDAClient(
+        &rendering_helper, index,
+        note, &data_str, num_NALUs_per_decode,
+        suppress_swap_to_display);
+    clients[index] = client;
+    OmxVideoDecodeAccelerator* decoder =
+        new OmxVideoDecodeAccelerator(client, rendering_thread.message_loop());
+    decoders[index] = decoder;
+    client->SetDecoder(decoder);
+    decoder->SetEglState(client->egl_display(), client->egl_context());
+    ASSERT_EQ(note->Wait(), CS_DECODER_SET);
+
+    // Configure the decoder.
+    int32 config_array[] = {
+      media::VIDEOATTRIBUTEKEY_BITSTREAMFORMAT_FOURCC,
+      media::VIDEOCODECFOURCC_H264,
+      media::VIDEOATTRIBUTEKEY_BITSTREAMFORMAT_WIDTH, kFrameWidth,
+      media::VIDEOATTRIBUTEKEY_BITSTREAMFORMAT_HEIGHT, kFrameHeight,
+      media::VIDEOATTRIBUTEKEY_VIDEOCOLORFORMAT, media::VIDEOCOLORFORMAT_RGBA,
+      media::VIDEOATTRIBUTEKEY_TERMINATOR
+    };
+    std::vector<uint32> config(
+        config_array, config_array + arraysize(config_array));
+    CHECK(decoder->Initialize(config));
+  }
+  // Then wait for all the decodes to finish.
+  for (size_t i = 0; i < num_concurrent_decoders; ++i) {
+    ClientStateNotification* note = notes[i];
+    ASSERT_EQ(note->Wait(), CS_INITIALIZED);
+    // InitializeDone kicks off decoding inside the client, so we just need to
+    // wait for Flush.
+    ASSERT_EQ(note->Wait(), CS_FLUSHED);
+  }
+  // Finally assert that decoding went as expected.
+  for (size_t i = 0; i < num_concurrent_decoders; ++i) {
+    EglRenderingVDAClient* client = clients[i];
+    EXPECT_EQ(client->num_decoded_frames(), 25 /* fps */ * 10 /* seconds */);
+    // Hard-coded the number of NALUs in the stream.  Icky.
+    EXPECT_EQ(client->num_done_bitstream_buffers(),
+              ceil(258.0 / num_NALUs_per_decode));
+    // These numbers are pulled out of a hat, but seem to be safe with current
+    // hardware.
+    double min_expected_fps = suppress_swap_to_display ? 175 : 50;
+    min_expected_fps /= num_concurrent_decoders;
+    LOG(INFO) << "Decoder " << i << " fps: " << client->frames_per_second();
+    EXPECT_GT(client->frames_per_second(), min_expected_fps);
+  }
+  STLDeleteElements(&decoders);
+  STLDeleteElements(&clients);
+  STLDeleteElements(&notes);
+
+  rendering_thread.message_loop()->PostTask(
+      FROM_HERE,
+      base::Bind(&RenderingHelper::UnInitialize,
+                 base::Unretained(&rendering_helper),
+                 &done));
+  done.Wait();
+  rendering_thread.Stop();
+};
+
+// TODO(fischman): using 16 and higher below breaks decode - visual artifacts
+// are introduced as well as spurious frames are delivered (more pictures are
+// returned than NALUs are fed to the decoder).  Increase the "15" below when
+// http://code.google.com/p/chrome-os-partner/issues/detail?id=4378 is fixed.
+INSTANTIATE_TEST_CASE_P(
+    DecodeVariations, OmxVideoDecodeAcceleratorTest,
+    ::testing::Values(
+        std::make_pair(1, 1), std::make_pair(1, 3), std::make_pair(2, 1),
+        std::make_pair(3, 1), std::make_pair(5, 1), std::make_pair(8, 1),
+        std::make_pair(15, 1)));
 
 // TODO(fischman, vrk): add more tests!  In particular:
-// - Test that chunking Decode() calls differently works.
+// - Test life-cycle: Seek/Stop/Pause/Play/RePlay for a single decoder.
 // - Test for memory leaks (valgrind)
-// - Test decode speed.  Ideally we can beat 60fps esp on simple test.mp4.
 // - Test alternate configurations
 // - Test failure conditions.
-// - Test multiple decodes; sequentially & concurrently.
 // - Test frame size changes mid-stream
 
 }  // namespace