Mac Overlays: Add GPU back-pressure

content/common/gpu/image_transport_surface_overlay_mac.mm

Index: content/common/gpu/image_transport_surface_overlay_mac.mm
diff --git a/content/common/gpu/image_transport_surface_overlay_mac.mm b/content/common/gpu/image_transport_surface_overlay_mac.mm
index 1a528333529330c9b8872a8f6c71be59970c170b..487e851b3f5de01852a23a5116794f99d844b84b 100644
--- a/content/common/gpu/image_transport_surface_overlay_mac.mm
+++ b/content/common/gpu/image_transport_surface_overlay_mac.mm
@@ -4,27 +4,66 @@
 
 #include "content/common/gpu/image_transport_surface_overlay_mac.h"
 
-#include <OpenGL/gl.h>
+// This type consistently causes problem on Mac, and needs to be dealt with
+// in a systemic way.
+// http://crbug.com/517208
+#ifndef GL_OES_EGL_image
+typedef void* GLeglImageOES;
+#endif
 
 #include "content/common/gpu/gpu_messages.h"
 #include "ui/accelerated_widget_mac/surface_handle_types.h"
 #include "ui/base/cocoa/animation_utils.h"
 #include "ui/base/cocoa/remote_layer_api.h"
 #include "ui/gfx/geometry/dip_util.h"
+#include "ui/gl/gl_fence_apple.h"
 #include "ui/gl/gl_image_io_surface.h"
+#include "ui/gl/scoped_api.h"
+#include "ui/gl/scoped_cgl.h"
+
+namespace {
+
+void CheckGLErrors(const char* msg) {
+  GLenum gl_error;
+  while ((gl_error = glGetError()) != GL_NO_ERROR) {
+    LOG(ERROR) << "OpenGL error hit " << msg << ": " << gl_error;
+  }
+}
+
+}  // namespace
 
 namespace content {
 
+class ImageTransportSurfaceOverlayMac::PendingSwap :
+    public base::RefCounted<PendingSwap> {
+ public:
+  PendingSwap() {}
+
+  // The IOSurface with new content for this swap.
+  base::ScopedCFTypeRef<IOSurfaceRef> io_surface;
+
+  // A fence object, and the CGL context it was issued in.
+  base::ScopedTypeRef<CGLContextObj> cgl_context;
+  scoped_ptr<gfx::GLFenceAPPLE> fence;
+
+  // The size of the full frame, in dip.
+  gfx::Size dip_size;
+
+ protected:
+  friend class base::RefCounted<PendingSwap>;
+
+  virtual ~PendingSwap() { DCHECK(!fence); }
+};
+
 ImageTransportSurfaceOverlayMac::ImageTransportSurfaceOverlayMac(
     GpuChannelManager* manager,
     GpuCommandBufferStub* stub,
     gfx::PluginWindowHandle handle)
-    : scale_factor_(1), pending_overlay_image_(nullptr) {
+    : scale_factor_(1), pending_overlay_image_(nullptr), weak_factory_(this) {
   helper_.reset(new ImageTransportHelper(this, manager, stub, handle));
 }
 
 ImageTransportSurfaceOverlayMac::~ImageTransportSurfaceOverlayMac() {
-  gfx::GLImageIOSurface::SetLayerForWidget(widget_, nil);
 }
 
 bool ImageTransportSurfaceOverlayMac::Initialize() {
@@ -37,46 +76,98 @@ bool ImageTransportSurfaceOverlayMac::Initialize() {
   ca_context_.reset(
       [[CAContext contextWithCGSConnection:connection_id options:@{}] retain]);
   layer_.reset([[CALayer alloc] init]);
+  [layer_ setGeometryFlipped:YES];
   [ca_context_ setLayer:layer_];
 
-  // Register the CALayer so that it can be picked up in GLImageIOSurface.
-  static intptr_t previous_widget = 0;
-  previous_widget += 1;
-  widget_ = reinterpret_cast<gfx::AcceleratedWidget>(previous_widget);
-  gfx::GLImageIOSurface::SetLayerForWidget(widget_, layer_);
-
   return true;
 }
 
-void ImageTransportSurfaceOverlayMac::Destroy() {}
+void ImageTransportSurfaceOverlayMac::Destroy() {
+  FinishAllPendingSwaps();
+}
 
 bool ImageTransportSurfaceOverlayMac::IsOffscreen() {
   return false;
 }
 
-gfx::SwapResult ImageTransportSurfaceOverlayMac::SwapBuffers() {
+gfx::SwapResult ImageTransportSurfaceOverlayMac::SwapBuffersInternal(
+    const gfx::Rect& pixel_damage_rect) {
   TRACE_EVENT0("gpu", "ImageTransportSurfaceOverlayMac::SwapBuffers");
 
-  // A flush is required to ensure that all content appears in the layer.
-  {
-    TRACE_EVENT0("gpu", "ImageTransportSurfaceOverlayMac::glFlush");
-    glFlush();
-  }
+  // The remainder of the function will populate the PendingSwap structure and
+  // then enqueue it.
+  scoped_refptr<PendingSwap> this_swap(new PendingSwap);
 
   // There should exist only one overlay image, and it should cover the whole
   // surface.
   DCHECK(pending_overlay_image_);
   if (pending_overlay_image_) {
-    TRACE_EVENT0("gpu", "ImageTransportSurfaceOverlayMac::setContents");
-    ScopedCAActionDisabler disabler;
-    gfx::Rect dip_bounds = gfx::ConvertRectToDIP(
-        scale_factor_, gfx::Rect(pixel_size_));
-    gfx::RectF crop_rect(0, 0, 1, 1);
-    pending_overlay_image_->ScheduleOverlayPlane(
-        widget_, 0, gfx::OVERLAY_TRANSFORM_NONE, dip_bounds, crop_rect);
+    gfx::GLImageIOSurface* pending_overlay_image_io_surface =
+        static_cast<gfx::GLImageIOSurface*>(pending_overlay_image_);
+    this_swap->io_surface = pending_overlay_image_io_surface->io_surface();
     pending_overlay_image_ = nullptr;
   }
 
+  // A flush is required to ensure that all content appears in the layer.
+  {
+    gfx::ScopedSetGLToRealGLApi scoped_set_gl_api;
+    TRACE_EVENT1("gpu", "ImageTransportSurfaceOverlayMac::glFlush", "surface",
+                 this_swap->io_surface.get());
+    CheckGLErrors("before flushing frame");
+    this_swap->cgl_context.reset(CGLGetCurrentContext(),
+                                base::scoped_policy::RETAIN);
+    this_swap->fence.reset(new gfx::GLFenceAPPLE);
+    CheckGLErrors("while flushing frame");
+  }
+
+  this_swap->dip_size = gfx::ConvertSizeToDIP(scale_factor_, pixel_size_);
+  pending_swaps_.push_back(this_swap);
+
+  PostCheckAndDisplayPendingSwaps();
+  return gfx::SwapResult::SWAP_ACK;
+}
+
+void ImageTransportSurfaceOverlayMac::CheckAndDisplayPendingSwaps(
+    bool force_immediate_display) {
+  TRACE_EVENT0("gpu",
+               "ImageTransportSurfaceOverlayMac::CheckAndDisplayPendingSwaps");
+  if (pending_swaps_.empty())
+    return;
+
+  // Check if the pending work has finished (and early-out if it is not, and
+  // this is not forced).
+  scoped_refptr<PendingSwap> this_swap = pending_swaps_.front();
+  {
+    gfx::ScopedSetGLToRealGLApi scoped_set_gl_api;
+    gfx::ScopedCGLSetCurrentContext scoped_set_current(this_swap->cgl_context);
+
+    CheckGLErrors("before testing fence");
+    bool result = this_swap->fence->HasCompleted();
+    CheckGLErrors("after testing fence");
+    if (!result && !force_immediate_display) {
+      PostCheckAndDisplayPendingSwaps();
+      return;
+    }
+    this_swap->fence.reset();
+    CheckGLErrors("while deleting fence");
+  }
+
+  // Update the CALayer hierarchy.
+  {
+    TRACE_EVENT1("gpu", "ImageTransportSurfaceOverlayMac::setContents",
+                 "surface", this_swap->io_surface.get());
+    ScopedCAActionDisabler disabler;
+    id new_contents = static_cast<id>(this_swap->io_surface.get());
+    [layer_ setContents:new_contents];
+    CGRect new_frame = gfx::Rect(this_swap->dip_size).ToCGRect();
+    if (!CGRectEqualToRect([layer_ frame], new_frame))
+      [layer_ setFrame:new_frame];
+  }
+
+  // Remove this swap from the queue.
+  pending_swaps_.pop_front();
+
+  // Send acknowledgement to the browser.
   GpuHostMsg_AcceleratedSurfaceBuffersSwapped_Params params;
   params.surface_handle =
       ui::SurfaceHandleFromCAContextID([ca_context_ contextId]);
@@ -84,14 +175,51 @@ gfx::SwapResult ImageTransportSurfaceOverlayMac::SwapBuffers() {
   params.scale_factor = scale_factor_;
   params.latency_info.swap(latency_info_);
   helper_->SendAcceleratedSurfaceBuffersSwapped(params);
-  return gfx::SwapResult::SWAP_ACK;
+}
+
+void ImageTransportSurfaceOverlayMac::PostCheckAndDisplayPendingSwaps() {
+  bool force_immediate_display = true;
+  base::TimeDelta delay;
+  if (display_link_mac_) {
+    base::TimeTicks timebase;
+    base::TimeDelta interval;
+    if (display_link_mac_->GetVSyncParameters(&timebase, &interval)) {
+      base::TimeTicks now = base::TimeTicks::Now();
+      // Make the timer fire halfway through the vsync interval (since that
+      // makes actual vsync alignment the most likely).
+      timebase -= interval / 2;
+      // Make sure that timebase is before now (to simplify the below
+      // computations).
+      if (timebase >= now)
+        timebase -= (1 + (timebase - now) / interval) * interval;
+      // Compute the target time.
+      base::TimeTicks target =
+        timebase + (1 + (now - timebase) / interval) * interval;

[tapted, 2015/08/06 00:56:49]

nit: indent 2 extra spaces

[ccameron, 2015/08/06 01:42:47]

Done.

+      delay = target - now;
+      force_immediate_display = false;
+    }
+  }
+  base::MessageLoop::current()->PostDelayedTask(
+      FROM_HERE,
+      base::Bind(&ImageTransportSurfaceOverlayMac::CheckAndDisplayPendingSwaps,
+                     weak_factory_.GetWeakPtr(), force_immediate_display),
+                 delay);

[Ken Russell (switch to Gerrit), 2015/08/05 21:55:23]

This will busy-loop while the browser is waiting for the pending swaps to
complete, correct? I wonder whether we can get into pathological situations
where every new swapped frame just misses the fence check above, causing this
delayed task to be issued for every frame.

[ccameron, 2015/08/06 01:42:47]

That could very well happen -- the effect of it would be that there would be 1
extra frame of latency before content appears on-screen, which I think is okay.

+}
+
+void ImageTransportSurfaceOverlayMac::FinishAllPendingSwaps() {
+  while (!pending_swaps_.empty())
+    CheckAndDisplayPendingSwaps(true);

[Ken Russell (switch to Gerrit), 2015/08/05 21:55:23]

Would it better match the intended behavior to actually wait for the
most-recently-issued fence to be completed, rather than just ignoring whether
they've passed or not? I'm not arguing for doing this, only asking. Your current
approach is probably less prone to strange deadlocks.

[ccameron, 2015/08/06 01:42:47]

Doing a finish on the GLFenceAPPLE would have this effect.

One thing that I didn't make clear in the CL description (fixing that) is that I
do the GLFence so that the timing of things showing up on-screen is more
predictable (and also for back-pressure).

If you do
  glDoSomethingReallyComplicatedAndSlow()
  glFlush
  [myLayer setContentsTo:complicated_thing_just_drawn]
then you will always see the correct content -- it's just that it may be a while
before it actually appears on-screen (which makes aligning with vsync hard).

+}
+
+gfx::SwapResult ImageTransportSurfaceOverlayMac::SwapBuffers() {
+  return SwapBuffersInternal(gfx::Rect(pixel_size_));
 }
 
 gfx::SwapResult ImageTransportSurfaceOverlayMac::PostSubBuffer(int x,
                                                                int y,
                                                                int width,
                                                                int height) {
-  return SwapBuffers();
+  return SwapBuffersInternal(gfx::Rect(x, y, width, height));
 }
 
 bool ImageTransportSurfaceOverlayMac::SupportsPostSubBuffer() {
@@ -129,10 +257,16 @@ bool ImageTransportSurfaceOverlayMac::IsSurfaceless() const {
 }
 
 void ImageTransportSurfaceOverlayMac::OnBufferPresented(
-    const AcceleratedSurfaceMsg_BufferPresented_Params& params) {}
+    const AcceleratedSurfaceMsg_BufferPresented_Params& params) {
+  if (display_link_mac_ && display_link_mac_->display_id() == params.display_id)
+    return;
+  display_link_mac_ = ui::DisplayLinkMac::GetForDisplay(params.display_id);
+}
 
 void ImageTransportSurfaceOverlayMac::OnResize(gfx::Size pixel_size,
                                                float scale_factor) {
+  // Flush through any pending frames.
+  FinishAllPendingSwaps();
   pixel_size_ = pixel_size;
   scale_factor_ = scale_factor;
 }