gpu: Reuse transfer buffers more aggresively

gpu/command_buffer/service/gles2_cmd_decoder.cc

 // Copyright (c) 2012 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 "gpu/command_buffer/service/gles2_cmd_decoder.h"
 
 #include <stdio.h>
 
 #include <algorithm>
 #include <list>
 #include <map>
 #include <stack>
 #include <string>
 #include <vector>
 
 #include "base/at_exit.h"
 #include "base/bind.h"
+#include "base/callback_helpers.h"
 #include "base/command_line.h"
 #include "base/debug/trace_event.h"
 #include "base/debug/trace_event_synthetic_delay.h"
 #include "base/memory/scoped_ptr.h"
 #include "base/strings/string_number_conversions.h"
 #include "base/strings/string_split.h"
 #include "build/build_config.h"
 #define GLES2_GPU_SERVICE 1
 #include "gpu/command_buffer/common/debug_marker_manager.h"
 #include "gpu/command_buffer/common/gles2_cmd_format.h"
@@ skipping 444 matching lines @@
 
 struct FenceCallback {
   explicit FenceCallback()
       : fence(gfx::GLFence::Create()) {
     DCHECK(fence);
   }
   std::vector<base::Closure> callbacks;
   scoped_ptr<gfx::GLFence> fence;
 };
 
+class AsyncUploadTokenCompletionObserver
+    : public AsyncPixelTransferCompletionObserver {
+ public:
+  explicit AsyncUploadTokenCompletionObserver(uint32 async_upload_token)
+      : async_upload_token_(async_upload_token) {
+  }
+
+  virtual void DidComplete(const AsyncMemoryParams& mem_params) OVERRIDE {
+    void* data = static_cast<int8*>(mem_params.shared_memory->memory()) +
+        mem_params.shm_data_offset;
+    AsyncUploadSync* sync = static_cast<AsyncUploadSync*>(data);
+    sync->SetAsyncUploadToken(async_upload_token_);
+  }
+
+ private:
+  uint32 async_upload_token_;
+
+  DISALLOW_COPY_AND_ASSIGN(AsyncUploadTokenCompletionObserver);
+};
+
 // }  // anonymous namespace.
 
 bool GLES2Decoder::GetServiceTextureId(uint32 client_texture_id,
                                        uint32* service_texture_id) {
   return false;
 }
 
 GLES2Decoder::GLES2Decoder()
     : initialized_(false),
       debug_(false),
@@ skipping 174 matching lines @@
   void DeleteBuffersHelper(GLsizei n, const GLuint* client_ids);
   bool GenFramebuffersHelper(GLsizei n, const GLuint* client_ids);
   void DeleteFramebuffersHelper(GLsizei n, const GLuint* client_ids);
   bool GenRenderbuffersHelper(GLsizei n, const GLuint* client_ids);
   void DeleteRenderbuffersHelper(GLsizei n, const GLuint* client_ids);
   bool GenQueriesEXTHelper(GLsizei n, const GLuint* client_ids);
   void DeleteQueriesEXTHelper(GLsizei n, const GLuint* client_ids);
   bool GenVertexArraysOESHelper(GLsizei n, const GLuint* client_ids);
   void DeleteVertexArraysOESHelper(GLsizei n, const GLuint* client_ids);
 
+  // Helper for async upload token completion notification callback.
+  base::Closure AsyncUploadTokenCompletionClosure(uint32 async_upload_token,
+                                                  uint32 sync_data_shm_id,
+                                                  uint32 sync_data_shm_offset);
+
+
+
   // Workarounds
   void OnFboChanged() const;
   void OnUseFramebuffer() const;
 
   // TODO(gman): Cache these pointers?
   BufferManager* buffer_manager() {
     return group_->buffer_manager();
   }
 
   RenderbufferManager* renderbuffer_manager() {
@@ skipping 9597 matching lines @@
   if (!texture_ref ||
       async_pixel_transfer_manager_->AsyncTransferIsInProgress(texture_ref)) {
     LOCAL_SET_GL_ERROR(
         GL_INVALID_OPERATION,
         function_name, "transfer already in progress");
     return false;
   }
   return true;
 }
 
+base::Closure GLES2DecoderImpl::AsyncUploadTokenCompletionClosure(
+    uint32 async_upload_token,
+    uint32 sync_data_shm_id,
+    uint32 sync_data_shm_offset) {
+  AsyncMemoryParams mem_params;
+  gpu::Buffer buffer = GetSharedMemoryBuffer(sync_data_shm_id);
+  if (!buffer.shared_memory)
+    return base::Closure();
+  mem_params.shared_memory = buffer.shared_memory;
+  mem_params.shm_size = buffer.size;
+  mem_params.shm_data_offset = sync_data_shm_offset;
+  mem_params.shm_data_size = sizeof(AsyncUploadSync);
+
+  if (mem_params.shm_data_offset + mem_params.shm_data_size >

[piman, 2014/03/20 21:18:26]

Beware of overflows. If shm_data_offset is UINT32_MAX, then this will pass, but
we'll dereference out-of-bounds later.

We can use CheckedNumerics, like we did in
https://codereview.chromium.org/196633009

+      mem_params.shm_size)
+    return base::Closure();
+
+  scoped_refptr<AsyncUploadTokenCompletionObserver> observer(
+      new AsyncUploadTokenCompletionObserver(async_upload_token));
+
+  return base::Bind(
+      &AsyncPixelTransferManager::AsyncNotifyCompletion,
+      base::Unretained(GetAsyncPixelTransferManager()),
+      mem_params,
+      observer);
+}
+
 error::Error GLES2DecoderImpl::HandleAsyncTexImage2DCHROMIUM(
     uint32 immediate_data_size, const cmds::AsyncTexImage2DCHROMIUM& c) {
   TRACE_EVENT0("gpu", "GLES2DecoderImpl::HandleAsyncTexImage2DCHROMIUM");
   GLenum target = static_cast<GLenum>(c.target);
   GLint level = static_cast<GLint>(c.level);
   // TODO(kloveless): Change HandleAsyncTexImage2DCHROMIUM command to use
   // unsigned integer for internalformat.
   GLenum internal_format = static_cast<GLenum>(c.internalformat);
   GLsizei width = static_cast<GLsizei>(c.width);
   GLsizei height = static_cast<GLsizei>(c.height);
   GLint border = static_cast<GLint>(c.border);
   GLenum format = static_cast<GLenum>(c.format);
   GLenum type = static_cast<GLenum>(c.type);
   uint32 pixels_shm_id = static_cast<uint32>(c.pixels_shm_id);
   uint32 pixels_shm_offset = static_cast<uint32>(c.pixels_shm_offset);
   uint32 pixels_size;
+  uint32 async_upload_token = static_cast<uint32>(c.async_upload_token);
+  uint32 sync_data_shm_id = static_cast<uint32>(c.sync_data_shm_id);
+  uint32 sync_data_shm_offset = static_cast<uint32>(c.sync_data_shm_offset);
+
+  base::ScopedClosureRunner scoped_completion_callback;
+  if (async_upload_token) {
+    base::Closure completion_closure =
+        AsyncUploadTokenCompletionClosure(async_upload_token,
+                                          sync_data_shm_id,
+                                          sync_data_shm_offset);
+    if (completion_closure.is_null())
+      return error::kInvalidArguments;
+
+    scoped_completion_callback.Reset(completion_closure);
+  }
 
   // TODO(epenner): Move this and copies of this memory validation
   // into ValidateTexImage2D step.
   if (!GLES2Util::ComputeImageDataSizes(
       width, height, format, type, state_.unpack_alignment, &pixels_size, NULL,
       NULL)) {
     return error::kOutOfBounds;
   }
   const void* pixels = NULL;
   if (pixels_shm_id != 0 || pixels_shm_offset != 0) {
@@ skipping 74 matching lines @@
     uint32 immediate_data_size, const cmds::AsyncTexSubImage2DCHROMIUM& c) {
   TRACE_EVENT0("gpu", "GLES2DecoderImpl::HandleAsyncTexSubImage2DCHROMIUM");
   GLenum target = static_cast<GLenum>(c.target);
   GLint level = static_cast<GLint>(c.level);
   GLint xoffset = static_cast<GLint>(c.xoffset);
   GLint yoffset = static_cast<GLint>(c.yoffset);
   GLsizei width = static_cast<GLsizei>(c.width);
   GLsizei height = static_cast<GLsizei>(c.height);
   GLenum format = static_cast<GLenum>(c.format);
   GLenum type = static_cast<GLenum>(c.type);
+  uint32 async_upload_token = static_cast<uint32>(c.async_upload_token);
+  uint32 sync_data_shm_id = static_cast<uint32>(c.sync_data_shm_id);
+  uint32 sync_data_shm_offset = static_cast<uint32>(c.sync_data_shm_offset);
+
+  base::ScopedClosureRunner scoped_completion_callback;
+  if (async_upload_token) {
+    base::Closure completion_closure =
+        AsyncUploadTokenCompletionClosure(async_upload_token,
+                                          sync_data_shm_id,
+                                          sync_data_shm_offset);
+    if (completion_closure.is_null())
+      return error::kInvalidArguments;
+
+    scoped_completion_callback.Reset(completion_closure);
+  }
 
   // TODO(epenner): Move this and copies of this memory validation
   // into ValidateTexSubImage2D step.
   uint32 data_size;
   if (!GLES2Util::ComputeImageDataSizes(
       width, height, format, type, state_.unpack_alignment, &data_size,
       NULL, NULL)) {
     return error::kOutOfBounds;
   }
   const void* pixels = GetSharedMemoryAs<const void*>(
@@ skipping 90 matching lines @@
       LOCAL_SET_GL_ERROR(
           GL_INVALID_OPERATION,
           "glWaitAsyncTexImage2DCHROMIUM", "No async transfer started");
     return error::kNoError;
   }
   delegate->WaitForTransferCompletion();
   ProcessFinishedAsyncTransfers();
   return error::kNoError;
 }
 
+error::Error GLES2DecoderImpl::HandleWaitAllAsyncTexImage2DCHROMIUM(
+    uint32 immediate_data_size, const cmds::WaitAllAsyncTexImage2DCHROMIUM& c) {
+  TRACE_EVENT0("gpu", "GLES2DecoderImpl::HandleWaitAsyncTexImage2DCHROMIUM");
+
+  GetAsyncPixelTransferManager()->WaitAllAsyncTexImage2D();
+  ProcessFinishedAsyncTransfers();
+  return error::kNoError;
+}
+
 void GLES2DecoderImpl::OnTextureRefDetachedFromFramebuffer(
     TextureRef* texture_ref) {
   Texture* texture = texture_ref->texture();
   DoDidUseTexImageIfNeeded(texture, texture->target());
 }
 
 // Include the auto-generated part of this file. We split this because it means
 // we can easily edit the non-auto generated parts right here in this file
 // instead of having to edit some template or the code generator.
 #include "gpu/command_buffer/service/gles2_cmd_decoder_autogen.h"
 
 }  // namespace gles2
 }  // namespace gpu