cc: Make async readback path use async glRreadPixels.

cc/output/gl_renderer.cc

 // Copyright 2010 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.
 
 #include "cc/output/gl_renderer.h"
 
 #include <algorithm>
 #include <set>
 #include <string>
 #include <vector>
@@ skipping 12 matching lines @@
 #include "cc/output/gl_frame_data.h"
 #include "cc/output/output_surface.h"
 #include "cc/output/render_surface_filters.h"
 #include "cc/quads/picture_draw_quad.h"
 #include "cc/quads/render_pass.h"
 #include "cc/quads/stream_video_draw_quad.h"
 #include "cc/quads/texture_draw_quad.h"
 #include "cc/resources/layer_quad.h"
 #include "cc/resources/priority_calculator.h"
 #include "cc/resources/scoped_resource.h"
+#include "cc/resources/sync_point_helper.h"
 #include "cc/trees/damage_tracker.h"
 #include "cc/trees/proxy.h"
 #include "cc/trees/single_thread_proxy.h"
 #include "gpu/GLES2/gl2extchromium.h"
 #include "third_party/WebKit/Source/Platform/chromium/public/WebGraphicsContext3D.h"
 #include "third_party/khronos/GLES2/gl2.h"
 #include "third_party/khronos/GLES2/gl2ext.h"
 #include "third_party/skia/include/core/SkBitmap.h"
 #include "third_party/skia/include/core/SkColor.h"
 #include "third_party/skia/include/core/SkColorFilter.h"
@@ skipping 37 matching lines @@
   return false;
 #endif
 }
 
 // Smallest unit that impact anti-aliasing output. We use this to
 // determine when anti-aliasing is unnecessary.
 const float kAntiAliasingEpsilon = 1.0f / 1024.0f;
 
 }  // anonymous namespace
 
+struct GLRenderer::PendingAsyncReadPixels {
+  PendingAsyncReadPixels() : buffer(0) {}
+
+  CopyRenderPassCallback copy_callback;
+  base::CancelableClosure finished_read_pixels_callback;
+  unsigned buffer;
+
+ private:
+  DISALLOW_COPY_AND_ASSIGN(PendingAsyncReadPixels);
+};
+
 scoped_ptr<GLRenderer> GLRenderer::Create(RendererClient* client,
                                           OutputSurface* output_surface,
                                           ResourceProvider* resource_provider,
                                           int highp_threshold_min) {
   scoped_ptr<GLRenderer> renderer(new GLRenderer(
       client, output_surface, resource_provider, highp_threshold_min));
   if (!renderer->Initialize())
     return scoped_ptr<GLRenderer>();
 
   return renderer.Pass();
@@ skipping 85 matching lines @@
 
   if (!InitializeSharedObjects())
     return false;
 
   // Make sure the viewport and context gets initialized, even if it is to zero.
   ViewportChanged();
   return true;
 }
 
 GLRenderer::~GLRenderer() {
+  while (!pending_async_read_pixels_.empty()) {
+    pending_async_read_pixels_.back()->finished_read_pixels_callback.Cancel();
+    pending_async_read_pixels_.back()->copy_callback.Run(
+        scoped_ptr<SkBitmap>());
+    pending_async_read_pixels_.pop_back();
+  }
+
   context_->setMemoryAllocationChangedCallbackCHROMIUM(NULL);
   CleanupSharedObjects();
 }
 
 const RendererCapabilities& GLRenderer::Capabilities() const {
   return capabilities_;
 }
 
 WebGraphicsContext3D* GLRenderer::Context() { return context_; }
 
@@ skipping 1565 matching lines @@
 
 void GLRenderer::EnsureScissorTestDisabled() {
   if (!is_scissor_enabled_)
     return;
 
   FlushTextureQuadCache();
   GLC(context_, context_->disable(GL_SCISSOR_TEST));
   is_scissor_enabled_ = false;
 }
 
-void GLRenderer::CopyCurrentRenderPassToBitmap(DrawingFrame* frame,
-                                               SkBitmap* bitmap) {
-  gfx::Size render_pass_size = frame->current_render_pass->output_rect.size();
-  bitmap->setConfig(SkBitmap::kARGB_8888_Config,
-                    render_pass_size.width(),
-                    render_pass_size.height());
-  if (bitmap->allocPixels()) {
-    bitmap->lockPixels();
-    GetFramebufferPixels(bitmap->getPixels(), gfx::Rect(render_pass_size));
-    bitmap->unlockPixels();
-  }
+void GLRenderer::CopyCurrentRenderPassToBitmap(
+    DrawingFrame* frame,
+    const CopyRenderPassCallback& callback) {
+  GetFramebufferPixelsAsync(frame->current_render_pass->output_rect, callback);
 }
 
 void GLRenderer::ToGLMatrix(float* gl_matrix, const gfx::Transform& transform) {
   transform.matrix().asColMajorf(gl_matrix);
 }
 
 void GLRenderer::SetShaderQuadF(const gfx::QuadF& quad, int quad_location) {
   if (quad_location == -1)
     return;
 
@@ skipping 181 matching lines @@
 
 void GLRenderer::EnsureBackbuffer() {
   if (!is_backbuffer_discarded_)
     return;
 
   output_surface_->EnsureBackbuffer();
   is_backbuffer_discarded_ = false;
 }
 
 void GLRenderer::GetFramebufferPixels(void* pixels, gfx::Rect rect) {
+  if (!pixels || rect.IsEmpty())
+    return;
+
+  scoped_ptr<PendingAsyncReadPixels> pending_read(new PendingAsyncReadPixels);
+  pending_async_read_pixels_.insert(pending_async_read_pixels_.begin(),
+                                    pending_read.Pass());
+
+  // This is a syncronous call since the callback is null.
+  DoGetFramebufferPixels(static_cast<uint8*>(pixels),
+                         rect,
+                         AsyncGetFramebufferPixelsCleanupCallback());
+}
+
+void GLRenderer::GetFramebufferPixelsAsync(gfx::Rect rect,
+                                           CopyRenderPassCallback callback) {
+  if (callback.is_null())
+    return;
+  if (rect.IsEmpty()) {
+    callback.Run(scoped_ptr<SkBitmap>());
+    return;
+  }
+
+  scoped_ptr<SkBitmap> bitmap(new SkBitmap);
+  bitmap->setConfig(SkBitmap::kARGB_8888_Config, rect.width(), rect.height());
+  bitmap->allocPixels();
+
+  scoped_ptr<SkAutoLockPixels> lock(new SkAutoLockPixels(*bitmap));
+
+  // Save a pointer to the pixels, the bitmap is owned by the cleanup_callback.
+  uint8* pixels = static_cast<uint8*>(bitmap->getPixels());
+
+  AsyncGetFramebufferPixelsCleanupCallback cleanup_callback = base::Bind(
+      &GLRenderer::PassOnSkBitmap,
+      base::Unretained(this),
+      base::Passed(&bitmap),
+      base::Passed(&lock),
+      callback);
+
+  scoped_ptr<PendingAsyncReadPixels> pending_read(new PendingAsyncReadPixels);
+  pending_read->copy_callback = callback;
+  pending_async_read_pixels_.insert(pending_async_read_pixels_.begin(),
+                                    pending_read.Pass());
+
+  // This is an asyncronous call since the callback is not null.
+  DoGetFramebufferPixels(pixels, rect, cleanup_callback);
+}
+
+void GLRenderer::DoGetFramebufferPixels(
+    uint8* dest_pixels,
+    gfx::Rect rect,
+    const AsyncGetFramebufferPixelsCleanupCallback& cleanup_callback) {
   DCHECK(rect.right() <= ViewportWidth());
   DCHECK(rect.bottom() <= ViewportHeight());
 
-  if (!pixels)
-    return;
+  bool is_async = !cleanup_callback.is_null();
 
   MakeContextCurrent();
 
   bool do_workaround = NeedsIOSurfaceReadbackWorkaround();
 
-  GLuint temporary_texture = 0;
-  GLuint temporary_fbo = 0;
+  unsigned temporary_texture = 0;
+  unsigned temporary_fbo = 0;
 
   if (do_workaround) {
     // On Mac OS X, calling glReadPixels() against an FBO whose color attachment
     // is an IOSurface-backed texture causes corruption of future glReadPixels()
     // calls, even those on different OpenGL contexts. It is believed that this
     // is the root cause of top crasher
     // http://crbug.com/99393. <rdar://problem/10949687>
 
     temporary_texture = context_->createTexture();
     GLC(context_, context_->bindTexture(GL_TEXTURE_2D, temporary_texture));
@@ skipping 27 matching lines @@
         context_->framebufferTexture2D(GL_FRAMEBUFFER,
                                        GL_COLOR_ATTACHMENT0,
                                        GL_TEXTURE_2D,
                                        temporary_texture,
                                        0));
 
     DCHECK(context_->checkFramebufferStatus(GL_FRAMEBUFFER) ==
            GL_FRAMEBUFFER_COMPLETE);
   }
 
-  scoped_ptr<uint8_t[]> src_pixels(
-      new uint8_t[rect.width() * rect.height() * 4]);
+  unsigned buffer = context_->createBuffer();
+  GLC(context_, context_->bindBuffer(GL_PIXEL_PACK_TRANSFER_BUFFER_CHROMIUM,
+                                     buffer));
+  GLC(context_, context_->bufferData(GL_PIXEL_PACK_TRANSFER_BUFFER_CHROMIUM,
+                                     4 * rect.size().GetArea(),
+                                     NULL,
+                                     GL_STREAM_READ));
+
   GLC(context_,
       context_->readPixels(rect.x(),
                            ViewportSize().height() - rect.bottom(),
                            rect.width(),
                            rect.height(),
                            GL_RGBA,
                            GL_UNSIGNED_BYTE,
-                           src_pixels.get()));
+                           NULL));
 
-  uint8_t* dest_pixels = static_cast<uint8_t*>(pixels);
-  size_t row_bytes = rect.width() * 4;
-  int num_rows = rect.height();
-  size_t total_bytes = num_rows * row_bytes;
-  for (size_t dest_y = 0; dest_y < total_bytes; dest_y += row_bytes) {
-    // Flip Y axis.
-    size_t src_y = total_bytes - dest_y - row_bytes;
-    // Swizzle BGRA -> RGBA.
-    for (size_t x = 0; x < row_bytes; x += 4) {
-      dest_pixels[dest_y + (x + 0)] = src_pixels.get()[src_y + (x + 2)];
-      dest_pixels[dest_y + (x + 1)] = src_pixels.get()[src_y + (x + 1)];
-      dest_pixels[dest_y + (x + 2)] = src_pixels.get()[src_y + (x + 0)];
-      dest_pixels[dest_y + (x + 3)] = src_pixels.get()[src_y + (x + 3)];
-    }
-  }
+  GLC(context_, context_->bindBuffer(GL_PIXEL_PACK_TRANSFER_BUFFER_CHROMIUM,
+                                     0));
 
   if (do_workaround) {
     // Clean up.
     GLC(context_, context_->bindFramebuffer(GL_FRAMEBUFFER, 0));
     GLC(context_, context_->bindTexture(GL_TEXTURE_2D, 0));
     GLC(context_, context_->deleteFramebuffer(temporary_fbo));
     GLC(context_, context_->deleteTexture(temporary_texture));
   }
 
+  base::Closure finished_callback =
+      base::Bind(&GLRenderer::FinishedReadback,
+                 base::Unretained(this),
+                 cleanup_callback,
+                 buffer,
+                 dest_pixels,
+                 rect.size());
+  // Save the finished_callback so it can be cancelled.
+  pending_async_read_pixels_.front()->finished_read_pixels_callback.Reset(
+      finished_callback);
+
+  // Save the buffer to verify the callbacks happen in the expected order.
+  pending_async_read_pixels_.front()->buffer = buffer;
+
+  if (is_async) {
+    unsigned sync_point = context_->insertSyncPoint();
+    SyncPointHelper::SignalSyncPoint(
+        context_,
+        sync_point,
+        finished_callback);
+  } else {
+    resource_provider_->Finish();
+    finished_callback.Run();
+  }
+
   EnforceMemoryPolicy();
 }
 
+void GLRenderer::FinishedReadback(
+    const AsyncGetFramebufferPixelsCleanupCallback& cleanup_callback,
+    unsigned source_buffer,
+    uint8* dest_pixels,
+    gfx::Size size) {
+  DCHECK(!pending_async_read_pixels_.empty());
+  DCHECK_EQ(source_buffer, pending_async_read_pixels_.back()->buffer);
+
+  uint8* src_pixels = NULL;
+
+  if (source_buffer != 0) {
+    GLC(context_, context_->bindBuffer(GL_PIXEL_PACK_TRANSFER_BUFFER_CHROMIUM,
+                                       source_buffer));
+    src_pixels = static_cast<uint8*>(
+        context_->mapBufferCHROMIUM(GL_PIXEL_PACK_TRANSFER_BUFFER_CHROMIUM,
+                                    GL_READ_ONLY));
+
+    if (src_pixels) {
+      memcpy(dest_pixels, src_pixels, 4 * size.GetArea());
+      GLC(context_, context_->unmapBufferCHROMIUM(
+          GL_PIXEL_PACK_TRANSFER_BUFFER_CHROMIUM));
+    }
+    GLC(context_, context_->bindBuffer(GL_PIXEL_PACK_TRANSFER_BUFFER_CHROMIUM,
+                                       0));
+    GLC(context_, context_->deleteBuffer(source_buffer));
+  }
+
+  if (src_pixels) {

[piman, 2013/04/29 17:29:13]

Mmh, you unmapped the buffer here. src_pixel isn't guaranteed to point to valid
data anymore...

[danakj, 2013/04/29 17:30:54]

Oh.. right. This is silly. I will do this instead of the memcpy.

+    size_t row_bytes = size.width() * 4;
+    int num_rows = size.height();
+    size_t total_bytes = num_rows * row_bytes;
+    for (size_t dest_y = 0; dest_y < total_bytes; dest_y += row_bytes) {
+      // Flip Y axis.
+      size_t src_y = total_bytes - dest_y - row_bytes;
+      // Swizzle BGRA -> RGBA.
+      for (size_t x = 0; x < row_bytes; x += 4) {
+        dest_pixels[dest_y + (x + 0)] = src_pixels[src_y + (x + 2)];

[piman, 2013/04/29 17:29:13]

Should all this be replacing the memcpy?
Why do the memcpy at all?

[danakj, 2013/04/30 01:30:52]

Done.

+        dest_pixels[dest_y + (x + 1)] = src_pixels[src_y + (x + 1)];
+        dest_pixels[dest_y + (x + 2)] = src_pixels[src_y + (x + 0)];
+        dest_pixels[dest_y + (x + 3)] = src_pixels[src_y + (x + 3)];
+      }
+    }
+  }
+
+  // TODO(danakj): This can go away when synchronous readback is no more and its
+  // contents can just move here.
+  if (!cleanup_callback.is_null())
+    cleanup_callback.Run(src_pixels != NULL);
+
+  pending_async_read_pixels_.pop_back();
+}
+
+void GLRenderer::PassOnSkBitmap(
+    scoped_ptr<SkBitmap> bitmap,
+    scoped_ptr<SkAutoLockPixels> lock,
+    const CopyRenderPassCallback& callback,
+    bool success) {
+  DCHECK(callback.Equals(pending_async_read_pixels_.back()->copy_callback));
+
+  lock.reset();
+  if (success)
+    callback.Run(bitmap.Pass());
+  else
+    callback.Run(scoped_ptr<SkBitmap>());
+}
+
 bool GLRenderer::GetFramebufferTexture(ScopedResource* texture,
                                        gfx::Rect device_rect) {
   DCHECK(!texture->id() || (texture->size() == device_rect.size() &&
                             texture->format() == GL_RGB));
 
   if (!texture->id() && !texture->Allocate(device_rect.size(),
                                            GL_RGB,
                                            ResourceProvider::TextureUsageAny))
     return false;
 
@@ skipping 536 matching lines @@
     resource_provider_->DeleteResource(on_demand_tile_raster_resource_id_);
 
   ReleaseRenderPassTextures();
 }
 
 bool GLRenderer::IsContextLost() {
   return (context_->getGraphicsResetStatusARB() != GL_NO_ERROR);
 }
 
 }  // namespace cc