Add a transform feedback workaround for Intel GPUs on MacOSX

gpu/command_buffer/service/gles2_cmd_decoder.cc

Index: gpu/command_buffer/service/gles2_cmd_decoder.cc
diff --git a/gpu/command_buffer/service/gles2_cmd_decoder.cc b/gpu/command_buffer/service/gles2_cmd_decoder.cc
index 866c3a35d5c34ad64765c9f21ff05b3e8e948332..28de9a6a367af3b07cc48749bdfc51d0057b8703 100644
--- a/gpu/command_buffer/service/gles2_cmd_decoder.cc
+++ b/gpu/command_buffer/service/gles2_cmd_decoder.cc
@@ -5713,6 +5713,11 @@ void GLES2DecoderImpl::DoResumeTransformFeedback() {
                        "transform feedback is not active or not paused");
     return;
   }
+  if (feature_info_->workarounds().use_transform_feedback_temp_unbind) {
+    glBindTransformFeedback(GL_TRANSFORM_FEEDBACK, 0);
+    state_.bound_transform_feedback->DoBindTransformFeedback(

[Zhenyao Mo, 2016/08/18 17:39:31]

I think you should just do glBindTransformFeedback(GL_TRANSFORM_FEEDBACK,
state_.bound_transform_feedback->service_id()) here to avoid all the unnecessary
steps.

[Zhenyao Mo, 2016/08/18 18:49:05]

1) I am still worried that this is caused by Chrome command buffer, so can you
do a glGetIntegerv(GL_TRANSFORM_FEEDBACK_BINDING) and see if the ID is lost? If
not, 2)

2) Do we have to Bind to 0 first then the currently bound transformfeedback?  If
we just bind it, will that work around the issue?

[Ken Russell (switch to Gerrit), 2016/08/19 00:13:35]

I think this is very likely a driver bug. These tests pass on my MacBook Pro
Retina with NVIDIA GPU.

[jchen10, 2016/08/19 01:01:23]

Anyway the worry makes sense to me. I have tried them. It turned out that 
    ID queried from GL_TRANSFORM_FEEDBACK_BINDING was same as
state_.bound_transform_feedback->service_id(), and
    the walk around didn't work without binding to 0 first.
So I think it's more likely a driver bug.

+        GL_TRANSFORM_FEEDBACK);
+  }
   state_.bound_transform_feedback->DoResumeTransformFeedback();
 }