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..c9fbe56e67e952a2bbe859d677470fcb4f024d04 100644
--- a/content/common/gpu/image_transport_surface_overlay_mac.mm
+++ b/content/common/gpu/image_transport_surface_overlay_mac.mm
@@ -5,6 +5,7 @@
 #include "content/common/gpu/image_transport_surface_overlay_mac.h"
 
 #include <OpenGL/gl.h>
+#include <OpenGL/glext.h>
 
 #include "content/common/gpu/gpu_messages.h"
 #include "ui/accelerated_widget_mac/surface_handle_types.h"
@@ -12,6 +13,16 @@
 #include "ui/base/cocoa/remote_layer_api.h"
 #include "ui/gfx/geometry/dip_util.h"
 #include "ui/gl/gl_image_io_surface.h"
+#include "ui/gl/scoped_api.h"
+#include "ui/gl/scoped_cgl.h"
+
+#define LOG_GL_ERRORS(msg) \

[tapted, 2015/08/05 07:55:57]

Not sure a macro is right for this. There's a `ClearGLErrors` in
android_webview/browser/scoped_app_gl_state_restore.cc which is similar.
gl_fence_arb.cc also has something (less similar).

[ccameron, 2015/08/05 21:00:43]

Changed this to a function.

+  do { \
+    for (GLenum gl_error = glGetError(); gl_error != GL_NO_ERROR; \
+         gl_error = glGetError()) { \
+      LOG(ERROR) << "OpenGL error hit " << msg << ": " << gl_error; \
+    } \
+  } while (0)
 
 namespace content {
 
@@ -19,12 +30,11 @@ 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 +47,99 @@ 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() {
+  CheckAndDisplayPendingSwaps(true);
+}
 
 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.
+  PendingSwap this_swap;
 
   // 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());
+    LOG_GL_ERRORS("before flushing frame");
+    this_swap.cgl_context.reset(CGLGetCurrentContext(),
+                                base::scoped_policy::RETAIN);
+    glGenFencesAPPLE(1, &this_swap.fence);
+    glSetFenceAPPLE(this_swap.fence);
+    glFlush();
+    LOG_GL_ERRORS("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).
+  const PendingSwap& this_swap = pending_swaps_.front();
+  {
+    gfx::ScopedSetGLToRealGLApi scoped_set_gl_api;
+    gfx::ScopedCGLSetCurrentContext scoped_set_current(this_swap.cgl_context);
+
+    LOG_GL_ERRORS("before testing fence");
+    bool result = !glIsFenceAPPLE(this_swap.fence) ||

[tapted, 2015/08/05 07:55:57]

Should `glIsFenceAPPLE` be a DCHECK?

[ccameron, 2015/08/05 21:00:43]

Done.

+                  glTestFenceAPPLE(this_swap.fence);
+    LOG_GL_ERRORS("after testing fence");
+    if (!result && !force_immediate_display) {
+      PostCheckAndDisplayPendingSwaps();
+      return;
+    }
+    glDeleteFencesAPPLE(1, &this_swap.fence);
+    LOG_GL_ERRORS("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];
+    [layer_ setFrame:gfx::Rect(this_swap.dip_size).ToCGRect()];

[tapted, 2015/08/05 07:55:57]

I wonder if there's a downside to calling this when the size hasn't changed..
(I'm often surprised with the lack of early exit cases in foundation libraries
when I've traced such things)

[ccameron, 2015/08/05 21:00:43]

Sounds like a good idea. Updated.

+  }
+
+  // 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 +147,50 @@ gfx::SwapResult ImageTransportSurfaceOverlayMac::SwapBuffers() {
   params.scale_factor = scale_factor_;
   params.latency_info.swap(latency_info_);
   helper_->SendAcceleratedSurfaceBuffersSwapped(params);
-  return gfx::SwapResult::SWAP_ACK;
+
+  // If this was a forced display, ensure that all pending swaps are flushed.
+  if (force_immediate_display && !pending_swaps_.empty())
+    CheckAndDisplayPendingSwaps(true);

[tapted, 2015/08/05 07:55:57]

tail recursion scares me a bit - maybe a wrapper function that loops while
!empty?

[ccameron, 2015/08/05 21:00:43]

Sounds reasonable. Done.

+}
+
+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;
+      delay = target - now;
+      force_immediate_display = false;
+    }
+  }
+  base::MessageLoop::current()->PostDelayedTask(
+      FROM_HERE,
+      base::Bind(&ImageTransportSurfaceOverlayMac::CheckAndDisplayPendingSwaps,
+                     weak_factory_.GetWeakPtr(), force_immediate_display),
+                 delay);
+}
+
+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 +228,16 @@ 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.
+  CheckAndDisplayPendingSwaps(true);
   pixel_size_ = pixel_size;
   scale_factor_ = scale_factor;
 }
@@ -145,4 +250,12 @@ void ImageTransportSurfaceOverlayMac::SetLatencyInfo(
 
 void ImageTransportSurfaceOverlayMac::WakeUpGpu() {}
 
+
+ImageTransportSurfaceOverlayMac::PendingSwap::PendingSwap()
+    : fence(0) {
+}
+
+ImageTransportSurfaceOverlayMac::PendingSwap::~PendingSwap() {
+}
+
 }  // namespace content