gpu_perftests: Use zero size gl surface for offscreen rendering

gpu/perftests/texture_upload_perftest.cc

Index: gpu/perftests/texture_upload_perftest.cc
diff --git a/gpu/perftests/texture_upload_perftest.cc b/gpu/perftests/texture_upload_perftest.cc
index 3a1af627acd344cb5e79dae661b594da677437e9..27444a4c432ca278ab20396033fb0a3e77b0fa55 100644
--- a/gpu/perftests/texture_upload_perftest.cc
+++ b/gpu/perftests/texture_upload_perftest.cc
@@ -24,6 +24,10 @@
 #include "ui/gl/gpu_timing.h"
 #include "ui/gl/scoped_make_current.h"
 
+#if defined(USE_OZONE)
+#include "base/message_loop/message_loop.h"
+#endif
+
 namespace gpu {
 namespace {
 
@@ -171,10 +175,15 @@ class TextureUploadPerfTest : public testing::Test {
 
   // Overridden from testing::Test
   void SetUp() override {
+#if defined(USE_OZONE)
+    // On Ozone, the backend initializes the event system using a UI
+    // thread.
+    base::MessageLoopForUI main_loop;

[Daniele Castagna, 2015/04/14 15:38:38]

Where in the backend is this message loop used? It'd be nice to avoid this.

[vignatti (out of this project), 2015/04/14 16:39:16]

A few Ozone implementations like the GBM requires early initialization of the
message loop because GL stuff, display management and etc (most of hardware
accesses) happen in the GPU process. So we need the message loop in place
already for the GPU process initial handshake behave correctly.

That said, a couple of weeks ago I've been trying to put an end to all of this,
but it went too far from what I've originally have intended and gave up. More
info here:

https://codereview.chromium.org/1043233003/

+#endif
     static bool gl_initialized = gfx::GLSurface::InitializeOneOff();
     DCHECK(gl_initialized);
     // Initialize an offscreen surface and a gl context.
-    surface_ = gfx::GLSurface::CreateOffscreenGLSurface(gfx::Size(4, 4));
+    surface_ = gfx::GLSurface::CreateOffscreenGLSurface(gfx::Size());
     gl_context_ = gfx::GLContext::CreateGLContext(NULL,  // share_group
                                                   surface_.get(),
                                                   gfx::PreferIntegratedGpu);