cc: Set the mapped memory reclaim limit for the renderer compositor on Android

content/common/gpu/client/webgraphicscontext3d_command_buffer_impl.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 "content/common/gpu/client/webgraphicscontext3d_command_buffer_impl.h"
 
 #include "third_party/khronos/GLES2/gl2.h"
 #ifndef GL_GLEXT_PROTOTYPES
 #define GL_GLEXT_PROTOTYPES 1
 #endif
@@ skipping 463 matching lines @@
       bind_generates_resources_,
       NULL));
   gl_ = real_gl_.get();
 
   if (attributes_.shareResources) {
     // Don't add ourselves to the list before others can get to our ShareGroup.
     g_all_shared_contexts.Pointer()->insert(this);
     lock.reset();
   }
 
+  size_t gl_mapped_memory_limit = gpu::gles2::GLES2Implementation::kNoLimit;
+#if defined(OS_ANDROID)
+  gl_mapped_memory_limit = mapped_memory_limit_;
+#endif

[piman, 2013/08/30 03:31:07]

noooo!

This is taking the problem upside down.

You want to decide the amount of uploads the compositor is going to do per
frame, and from that, size the command buffer so that it's at least big enough
so that you don't have to stall.

So, instead of setting the limit on the context, then extracting that value in
the compositor and then ignoring the limit on !ANDROID, why don't we:
- decide the amount of uploads per frame. pass that explicitly to the compositor
- figure out the limit in the context. On Android we may want it tight because
we're memory constrained, so set the limit to the kMaxTransferBufferUsageBytes.
On other platforms, where we're not constrained, don't set the limit (or set it
to a larger value) so that we don't risk stalling.

In either case, this decision should be made in the embedder (RenderWidget, so
be it), not as an ifdef that affects all contexts for no good reason.

+
   if (!real_gl_->Initialize(
       start_transfer_buffer_size_,
       min_transfer_buffer_size_,
       max_transfer_buffer_size_,
-      mapped_memory_limit_)) {
+      gl_mapped_memory_limit)) {
     return false;
   }
 
   if (CommandLine::ForCurrentProcess()->HasSwitch(
       switches::kEnableGpuClientTracing)) {
     trace_gl_.reset(new gpu::gles2::GLES2TraceImplementation(gl_));
     gl_ = trace_gl_.get();
   }
 
   return true;
@@ skipping 1092 matching lines @@
 
 void WebGraphicsContext3DCommandBufferImpl::OnErrorMessage(
     const std::string& message, int id) {
   if (error_message_callback_) {
     WebKit::WebString str = WebKit::WebString::fromUTF8(message.c_str());
     error_message_callback_->onErrorMessage(str, id);
   }
 }
 
 }  // namespace content