gpu: Reuse transfer buffers more aggresively

gpu/command_buffer/build_gles2_cmd_buffer.py

Index: gpu/command_buffer/build_gles2_cmd_buffer.py
diff --git a/gpu/command_buffer/build_gles2_cmd_buffer.py b/gpu/command_buffer/build_gles2_cmd_buffer.py
index 0f2e49a31f115d1d8835c45ec5070e23c7b83d18..fc14f176affeb1415758cf41fcd9ba23646b4a1f 100755
--- a/gpu/command_buffer/build_gles2_cmd_buffer.py
+++ b/gpu/command_buffer/build_gles2_cmd_buffer.py
@@ -828,6 +828,7 @@ _ENUM_LISTS = {
       'GL_COMMANDS_ISSUED_CHROMIUM',
       'GL_LATENCY_QUERY_CHROMIUM',
       'GL_ASYNC_PIXEL_UNPACK_COMPLETED_CHROMIUM',
+      'GL_ASYNC_PIXEL_UNPACK_COMPLETED_PRIVATE_CHROMIUM',

[reveman, 2014/01/02 01:31:29]

Are clients prevented from using this somehow?

Did you consider adding a new pair of Begin/EndQuery commands instead? ie.
BeginInternalQueryEXT and EndInternalQueryEXT. Or a "bool private" parameter to
the existing Begin/EndQuery commands.

[jadahl, 2014/01/02 10:59:54]

The idea is that this target should be private, and not allowed to be used by
some client. Not sure how to enforce that rule except with error reporting. Any
ideas here?

What we'd gain from having a "bool private"? Would it be applicable to all types
of queries? Also the assumption that one type of target only has one running
query would longer hold, and any such assumptions throughout the code would no
longer be true.

[reveman, 2014/01/02 11:56:44]

The idea is that this would make it easy to keep it internal as we'd simply not
expose it in the client API. It could be limited to ASYNC_PIXEL_UNPACK_COMPLETED
for now but it would be trivial to add support for other targets if necessary.

The key type for the QueryMap you introduced in this patch would instead be a
std::pair<bool, GLuint>. Where "bool" is whether the query is private. Are we
making more assumptions than this related to running queries?

       'GL_ASYNC_PIXEL_PACK_COMPLETED_CHROMIUM',
     ],
   },