Prefer GrAAConvexPathRenderer over ccpr

Even though ccpr can be faster than the convex renderer for small
paths, there is value in a simpler, more consistent approach. The
convex renderer is single-pass, doesn't deal with caching, and is
unaffected by the less favorable flush pattern in Chrome. We can
address the regressions from this CL by optimizing single-pass convex
path rendering, rather than trying to use ccpr more.

Bug: chromium:860021
Change-Id: I44ceefa43cdad7fb25c620faed8b525901fb1f0c
Reviewed-on: https://skia-review.googlesource.com/c/171525
Reviewed-by: Chris Dalton <csmartdalton@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Chris Dalton <csmartdalton@google.com>

src/gpu/GrPathRendererChain.cpp

@@ -34,6 +34,9 @@ GrPathRendererChain::GrPathRendererChain(GrContext* context, const Options& opti
             fChain.push_back(std::move(pr));
         }
     }
+    if (options.fGpuPathRenderers & GpuPathRenderers::kAAConvex) {
+        fChain.push_back(sk_make_sp<GrAAConvexPathRenderer>());
+    }
     if (options.fGpuPathRenderers & GpuPathRenderers::kCoverageCounting) {
         using AllowCaching = GrCoverageCountingPathRenderer::AllowCaching;
         if (auto ccpr = GrCoverageCountingPathRenderer::CreateIfSupported(
@@ -46,9 +49,6 @@ GrPathRendererChain::GrPathRendererChain(GrContext* context, const Options& opti
     if (options.fGpuPathRenderers & GpuPathRenderers::kAAHairline) {
         fChain.push_back(sk_make_sp<GrAAHairLinePathRenderer>());
     }
-    if (options.fGpuPathRenderers & GpuPathRenderers::kAAConvex) {
-        fChain.push_back(sk_make_sp<GrAAConvexPathRenderer>());
-    }
     if (options.fGpuPathRenderers & GpuPathRenderers::kAALinearizing) {
         fChain.push_back(sk_make_sp<GrAALinearizingConvexPathRenderer>());
     }