Defer DDL destruction until after SwapBuffers

This reordering allows us to issue SwapBuffers a little bit sooner
by moving DDL destructor after.

Bug: 974359
Change-Id: Iba9a81815cf1166d3d31bd9312adc8e0834bf512
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1797175Reviewed-by: Vasiliy Telezhnikov <vasilyt@chromium.org>
Commit-Queue: Jonathan Backer <backer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#695762}

components/viz/service/display_embedder/skia_output_surface_impl_on_gpu.cc

@@ -725,13 +725,13 @@ void SkiaOutputSurfaceImplOnGpu::FinishPaintCurrentFrame(
 
     if (!output_sk_surface()->draw(ddl.get()))
       DLOG(ERROR) << "output_sk_surface()->draw() failed.";
-    ddl = nullptr;
+    destroy_after_swap_.emplace_back(std::move(ddl));
 
     if (overdraw_ddl) {
       sk_sp<SkSurface> overdraw_surface = SkSurface::MakeRenderTarget(
           gr_context(), overdraw_ddl->characterization(), SkBudgeted::kNo);
       overdraw_surface->draw(overdraw_ddl.get());
-      overdraw_ddl = nullptr;
+      destroy_after_swap_.emplace_back(std::move(overdraw_ddl));
 
       SkPaint paint;
       sk_sp<SkImage> overdraw_image = overdraw_surface->makeImageSnapshot();
@@ -833,6 +833,8 @@ void SkiaOutputSurfaceImplOnGpu::SwapBuffers(
     output_device_->SwapBuffers(buffer_presented_callback_,
                                 std::move(frame.latency_info));
   }
+
+  destroy_after_swap_.clear();
 }
 
 void SkiaOutputSurfaceImplOnGpu::FinishPaintRenderPass(
@@ -877,6 +879,8 @@ void SkiaOutputSurfaceImplOnGpu::FinishPaintRenderPass(
       DCHECK(result);
     }
     offscreen.surface()->draw(ddl.get());
+    destroy_after_swap_.emplace_back(std::move(ddl));
+
     GrFlushInfo flush_info = {
         .fFlags = kNone_GrFlushFlags,
         .fNumSemaphores = scoped_promise_image_access.end_semaphores().size(),
@@ -921,6 +925,11 @@ void SkiaOutputSurfaceImplOnGpu::CopyOutput(
   // TODO(crbug.com/898595): Do this on the GPU instead of CPU with Vulkan.
   DCHECK_CALLED_ON_VALID_THREAD(thread_checker_);
 
+  // Clear |destroy_after_swap_| if we CopyOutput without SwapBuffers.
+  base::ScopedClosureRunner cleanup(
+      base::BindOnce([](std::vector<std::unique_ptr<SkDeferredDisplayList>>) {},
+                     std::move(destroy_after_swap_)));
+
   if (deferred_framebuffer_draw_closure) {
     std::move(deferred_framebuffer_draw_closure).Run();
     DCHECK(context_state_->IsCurrent(nullptr /* surface */));

components/viz/service/display_embedder/skia_output_surface_impl_on_gpu.h

@@ -263,6 +263,9 @@ class SkiaOutputSurfaceImplOnGpu : public gpu::ImageTransportSurfaceDelegate {
   gl::GLApi* api_ = nullptr;
   bool supports_alpha_ = false;
 
+  // Micro-optimization to get to issuing GPU SwapBuffers as soon as possible.
+  std::vector<std::unique_ptr<SkDeferredDisplayList>> destroy_after_swap_;
+
   THREAD_CHECKER(thread_checker_);
 
   base::WeakPtr<SkiaOutputSurfaceImplOnGpu> weak_ptr_;