Add BarrierArriveAndDetachExceptLast().

Provide a way for one process to continue the remaining phases of a
(previously) parallel computation alone.  Later patches will use this to
extend Parallel Hash Join.

Author: Melanie Plageman <melanieplageman@gmail.com>
Reviewed-by: Thomas Munro <thomas.munro@gmail.com>
Discussion: https://postgr.es/m/CA%2BhUKG%2BA6ftXPz4oe92%2Bx8Er%2BxpGZqto70-Q_ERwRaSyA%3DafNg%40mail.gmail.com

src/backend/storage/ipc/barrier.c

@@ -205,6 +205,28 @@ BarrierArriveAndDetach(Barrier *barrier)
 	return BarrierDetachImpl(barrier, true);
 }
 
+/*
+ * Arrive at a barrier, and detach all but the last to arrive.  Returns true if
+ * the caller was the last to arrive, and is therefore still attached.
+ */
+bool
+BarrierArriveAndDetachExceptLast(Barrier *barrier)
+{
+	SpinLockAcquire(&barrier->mutex);
+	if (barrier->participants > 1)
+	{
+		--barrier->participants;
+		SpinLockRelease(&barrier->mutex);
+
+		return false;
+	}
+	Assert(barrier->participants == 1);
+	++barrier->phase;
+	SpinLockRelease(&barrier->mutex);
+
+	return true;
+}
+
 /*
  * Attach to a barrier.  All waiting participants will now wait for this
  * participant to call BarrierArriveAndWait(), BarrierDetach() or

src/include/storage/barrier.h

@@ -37,6 +37,7 @@ typedef struct Barrier
 extern void BarrierInit(Barrier *barrier, int num_workers);
 extern bool BarrierArriveAndWait(Barrier *barrier, uint32 wait_event_info);
 extern bool BarrierArriveAndDetach(Barrier *barrier);
+extern bool BarrierArriveAndDetachExceptLast(Barrier *barrier);
 extern int	BarrierAttach(Barrier *barrier);
 extern bool BarrierDetach(Barrier *barrier);
 extern int	BarrierPhase(Barrier *barrier);