Add support for converting cc::FilterOperations into an SkImageFilter

cc/output/software_renderer.cc

Index: cc/output/software_renderer.cc
diff --git a/cc/output/software_renderer.cc b/cc/output/software_renderer.cc
index 77c44eadc3d2e62c70ef9080d8bac02d47c4109a..1554aec67c415ab6d97d4584986c01233c382de1 100644
--- a/cc/output/software_renderer.cc
+++ b/cc/output/software_renderer.cc
@@ -11,6 +11,7 @@
 #include "cc/output/compositor_frame_metadata.h"
 #include "cc/output/copy_output_request.h"
 #include "cc/output/output_surface.h"
+#include "cc/output/render_surface_filters.h"
 #include "cc/output/software_output_device.h"
 #include "cc/quads/checkerboard_draw_quad.h"
 #include "cc/quads/debug_border_draw_quad.h"
@@ -433,9 +434,11 @@ void SoftwareRenderer::DrawRenderPassQuad(const DrawingFrame* frame,
   shader->setLocalMatrix(content_mat);
   current_paint_.setShader(shader.get());
 
-  SkImageFilter* filter = quad->filter.get();
-  if (filter)
-    current_paint_.setImageFilter(filter);
+  if (quad->filters.HasReferenceFilter()) {

[ajuma, 2013/09/09 17:35:36]

I'm checking for a reference filter here just to match the existing behavior.
But note that there doesn't appear to be any existing code handling the other
case.

[Stephen White, 2013/09/09 18:51:01]

IIRC, the software renderer only ever used setFilter(), for both the CSS
shorthands and for reference filters. Maybe skaslev@ can refresh my memory. If
so, we may want to skip the check and simply call BuildImageFilter()
unconditionally.

[danakj, 2013/09/09 22:04:50]

I think you should drop this if statement. It's not like we would be applying
the filters somewhere else.. Also, it sounds like we don't have any pixel tests
with software comp and filters. Would you care to make some/all of the
LayerTreeHostPixelTestFilterTest variants work with software comp too?

[ajuma, 2013/09/10 21:17:58]

Done.
All but one of those tests use background filters, which the software renderer
ignores entirely right now (there's a TODO in
SoftwareRenderer::DrawRenderPassQuad for fixing that). The other test (that uses
a foreground filter) fails badly with the software compositor (all pixels
incorrect) both with and without this patch. There's an open bug for getting
filters working with the software compositor (http://crbug.com/160302). Perhaps
getting these pixel tests working with the software compositor should be done
once that work is complete (or done as part of that work).

[danakj, 2013/09/10 21:22:38]

Do you know how the foreground filter test fails? It seems like this is the work
to get it going.

+    skia::RefPtr<SkImageFilter> filter = RenderSurfaceFilters::BuildImageFilter(
+        quad->filters, content_texture->size());
+    current_paint_.setImageFilter(filter.get());
+  }
 
   if (quad->mask_resource_id) {
     ResourceProvider::ScopedReadLockSoftware mask_lock(resource_provider_,