gpu: Reuse existing code for shared memory allocation.

base/memory/shared_memory_mac.cc

Index: base/memory/shared_memory_mac.cc
diff --git a/base/memory/shared_memory_mac.cc b/base/memory/shared_memory_mac.cc
index 0b3a62f0e7a44643c786fe239ec5862377e657e6..5cc3c68461cac8bcef475364653877d6429e6fd0 100644
--- a/base/memory/shared_memory_mac.cc
+++ b/base/memory/shared_memory_mac.cc
@@ -177,6 +177,11 @@ SharedMemoryHandle SharedMemory::handle() const {
   return shm_;
 }
 
+SharedMemoryHandle SharedMemory::TakeHandle() {
+  NOTREACHED();

[sadrul, 2016/11/11 02:00:30]

Can I do return std::move(shm_);?

[Nico, 2016/11/14 18:53:51]

erikchen wrote this. Erik, can you answer?

[erikchen, 2016/11/14 19:37:53]

SharedMemoryHandle doesn't have ownership semantics, so std::move doesn't make
much sense.

Duplicate()/Close() increments and decrements a mach port reference count so you
could theoretically skip it, but Duplicate() also modifies
ownership_passes_to_ipc_. To be more similar to the other platforms, I recommend
just keeping the Duplicate()/Close().

+  return SharedMemoryHandle();
+}
+
 void SharedMemory::Close() {
   shm_.Close();
   shm_ = SharedMemoryHandle();