From a0fc306023d77e5605203c14ca92f368bdbce3ae Mon Sep 17 00:00:00 2001
From: Austin Clements <austin@google.com>
Date: Wed, 13 May 2015 17:08:16 -0400
Subject: [PATCH] runtime: eliminate runqvictims and a copy from runqsteal

Currently, runqsteal steals Gs from another P into an intermediate
buffer and then copies those Gs into the current P's run queue. This
intermediate buffer itself was moved from the stack to the P in commit
c4fe503 to eliminate the cost of zeroing it on every steal.

This commit follows up c4fe503 by stealing directly into the current
P's run queue, which eliminates the copy and the need for the
intermediate buffer. The update to the tail pointer is only committed
once the entire steal operation has succeeded, so the semantics of
stealing do not change.

Change-Id: Icdd7a0eb82668980bf42c0154b51eef6419fdd51
Reviewed-on: https://go-review.googlesource.com/9998
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
---
 src/runtime/proc1.go    | 21 ++++++++++-----------
 src/runtime/runtime2.go |  7 +++----
 2 files changed, 13 insertions(+), 15 deletions(-)

diff --git a/src/runtime/proc1.go b/src/runtime/proc1.go
index 8aeacee747..4ce756b692 100644
--- a/src/runtime/proc1.go
+++ b/src/runtime/proc1.go
@@ -3460,10 +3460,11 @@ func runqget(_p_ *p) (gp *g, inheritTime bool) {
 	}
 }
 
-// Grabs a batch of goroutines from local runnable queue.
-// batch array must be of size len(p->runq)/2. Returns number of grabbed goroutines.
+// Grabs a batch of goroutines from _p_'s runnable queue into batch.
+// Batch is a ring buffer starting at batchHead.
+// Returns number of grabbed goroutines.
 // Can be executed by any P.
-func runqgrab(_p_ *p, batch []*g, stealRunNextG bool) uint32 {
+func runqgrab(_p_ *p, batch *[256]*g, batchHead uint32, stealRunNextG bool) uint32 {
 	for {
 		h := atomicload(&_p_.runqhead) // load-acquire, synchronize with other consumers
 		t := atomicload(&_p_.runqtail) // load-acquire, synchronize with the producer
@@ -3484,7 +3485,7 @@ func runqgrab(_p_ *p, batch []*g, stealRunNextG bool) uint32 {
 					if !_p_.runnext.cas(next, 0) {
 						continue
 					}
-					batch[0] = next.ptr()
+					batch[batchHead%uint32(len(batch))] = next.ptr()
 					return 1
 				}
 			}
@@ -3494,7 +3495,8 @@ func runqgrab(_p_ *p, batch []*g, stealRunNextG bool) uint32 {
 			continue
 		}
 		for i := uint32(0); i < n; i++ {
-			batch[i] = _p_.runq[(h+i)%uint32(len(_p_.runq))]
+			g := _p_.runq[(h+i)%uint32(len(_p_.runq))]
+			batch[(batchHead+i)%uint32(len(batch))] = g
 		}
 		if cas(&_p_.runqhead, h, h+n) { // cas-release, commits consume
 			return n
@@ -3506,23 +3508,20 @@ func runqgrab(_p_ *p, batch []*g, stealRunNextG bool) uint32 {
 // and put onto local runnable queue of p.
 // Returns one of the stolen elements (or nil if failed).
 func runqsteal(_p_, p2 *p, stealRunNextG bool) *g {
-	n := runqgrab(p2, _p_.runqvictims[:], stealRunNextG)
+	t := _p_.runqtail
+	n := runqgrab(p2, &_p_.runq, t, stealRunNextG)
 	if n == 0 {
 		return nil
 	}
 	n--
-	gp := _p_.runqvictims[n]
+	gp := _p_.runq[(t+n)%uint32(len(_p_.runq))]
 	if n == 0 {
 		return gp
 	}
 	h := atomicload(&_p_.runqhead) // load-acquire, synchronize with consumers
-	t := _p_.runqtail
 	if t-h+n >= uint32(len(_p_.runq)) {
 		throw("runqsteal: runq overflow")
 	}
-	for i := uint32(0); i < n; i++ {
-		_p_.runq[(t+i)%uint32(len(_p_.runq))] = _p_.runqvictims[i]
-	}
 	atomicstore(&_p_.runqtail, t+n) // store-release, makes the item available for consumption
 	return gp
 }
diff --git a/src/runtime/runtime2.go b/src/runtime/runtime2.go
index ae93bb8dcb..8dfece5845 100644
--- a/src/runtime/runtime2.go
+++ b/src/runtime/runtime2.go
@@ -353,10 +353,9 @@ type p struct {
 	goidcacheend uint64
 
 	// Queue of runnable goroutines. Accessed without lock.
-	runqhead    uint32
-	runqtail    uint32
-	runq        [256]*g
-	runqvictims [128]*g // Used to stage victims from another p's runq
+	runqhead uint32
+	runqtail uint32
+	runq     [256]*g
 	// runnext, if non-nil, is a runnable G that was ready'd by
 	// the current G and should be run next instead of what's in
 	// runq if there's time remaining in the running G's time