runtime: replace typedmemmmove with bulkBarrierPreWrite and memmove in growslice

A bulkBarrierPreWrite together with a memmove as used in typedslicecopy
is faster than a typedmemmove for each element of the old slice that
needs to be copied to the new slice.

typedslicecopy is not used here as runtime functions should not call
other instrumented runtime functions and some conditions like dst == src
or the destination slice not being large enought that are checked for
in typedslicecopy can not happen in growslice.

Append                         13.5ns ± 6%  13.3ns ± 3%     ~     (p=0.304 n=10+10)
AppendGrowByte                 1.18ms ± 2%  1.19ms ± 1%     ~     (p=0.113 n=10+9)
AppendGrowString                123ms ± 1%    73ms ± 1%  -40.39%  (p=0.000 n=9+8)
AppendSlice/1Bytes             3.81ns ± 1%  3.78ns ± 1%     ~     (p=0.116 n=10+10)
AppendSlice/4Bytes             3.71ns ± 1%  3.70ns ± 0%     ~     (p=0.095 n=10+9)
AppendSlice/7Bytes             3.73ns ± 0%  3.75ns ± 1%     ~     (p=0.442 n=10+10)
AppendSlice/8Bytes             4.00ns ± 1%  4.01ns ± 1%     ~     (p=0.330 n=10+10)
AppendSlice/15Bytes            4.29ns ± 1%  4.28ns ± 1%     ~     (p=0.536 n=10+10)
AppendSlice/16Bytes            4.28ns ± 1%  4.31ns ± 1%   +0.75%  (p=0.019 n=10+10)
AppendSlice/32Bytes            4.57ns ± 2%  4.58ns ± 2%     ~     (p=0.236 n=10+10)
AppendSliceLarge/1024Bytes      305ns ± 2%   306ns ± 1%     ~     (p=0.236 n=10+10)
AppendSliceLarge/4096Bytes     1.06µs ± 1%  1.06µs ± 0%     ~     (p=1.000 n=9+10)
AppendSliceLarge/16384Bytes    3.12µs ± 2%  3.11µs ± 1%     ~     (p=0.493 n=10+10)
AppendSliceLarge/65536Bytes    5.61µs ± 5%  5.36µs ± 2%   -4.58%  (p=0.003 n=10+8)
AppendSliceLarge/262144Bytes   21.0µs ± 1%  19.5µs ± 1%   -7.09%  (p=0.000 n=8+10)
AppendSliceLarge/1048576Bytes  78.4µs ± 1%  78.7µs ± 2%     ~     (p=0.315 n=8+10)
AppendStr/1Bytes               3.96ns ± 6%  3.99ns ± 9%     ~     (p=0.591 n=10+10)
AppendStr/4Bytes               3.98ns ± 1%  3.99ns ± 1%     ~     (p=0.515 n=9+9)
AppendStr/8Bytes               4.27ns ± 1%  4.27ns ± 1%     ~     (p=0.633 n=10+10)
AppendStr/16Bytes              4.56ns ± 2%  4.55ns ± 1%     ~     (p=0.869 n=10+10)
AppendStr/32Bytes              4.85ns ± 1%  4.89ns ± 1%   +0.71%  (p=0.003 n=10+8)
AppendSpecialCase              18.7ns ± 1%  18.7ns ± 1%     ~     (p=0.144 n=10+10)
AppendInPlace/NoGrow/Byte       438ns ± 1%   439ns ± 1%     ~     (p=0.135 n=10+8)
AppendInPlace/NoGrow/1Ptr      1.05µs ± 2%  1.05µs ± 1%     ~     (p=0.469 n=10+10)
AppendInPlace/NoGrow/2Ptr      1.77µs ± 1%  1.78µs ± 2%     ~     (p=0.469 n=10+10)
AppendInPlace/NoGrow/3Ptr      1.94µs ± 1%  1.93µs ± 2%     ~     (p=0.517 n=10+10)
AppendInPlace/NoGrow/4Ptr      3.18µs ± 1%  3.17µs ± 0%     ~     (p=0.483 n=10+9)
AppendInPlace/Grow/Byte         382ns ± 2%   383ns ± 2%     ~     (p=0.705 n=9+10)
AppendInPlace/Grow/1Ptr         383ns ± 1%   384ns ± 1%     ~     (p=0.844 n=10+10)
AppendInPlace/Grow/2Ptr         459ns ± 2%   467ns ± 2%   +1.74%  (p=0.001 n=10+10)
AppendInPlace/Grow/3Ptr         593ns ± 1%   597ns ± 2%     ~     (p=0.195 n=10+10)
AppendInPlace/Grow/4Ptr         583ns ± 2%   589ns ± 2%     ~     (p=0.084 n=10+10)

Change-Id: I629872f065a22b29267c1adbfc578aaedd36d365
Reviewed-on: https://go-review.googlesource.com/115755
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

1 files changed, 3 insertions, 7 deletions

diff --git a/src/runtime/slice.go b/src/runtime/slice.go
index fd5d08b52c..737aab5704 100644
--- a/src/runtime/slice.go
+++ b/src/runtime/slice.go
@@ -195,21 +195,17 @@ func growslice(et *_type, old slice, cap int) slice {
 	var p unsafe.Pointer
 	if et.kind&kindNoPointers != 0 {
 		p = mallocgc(capmem, nil, false)
-		memmove(p, old.array, lenmem)
 		// The append() that calls growslice is going to overwrite from old.len to cap (which will be the new length).
 		// Only clear the part that will not be overwritten.
 		memclrNoHeapPointers(add(p, newlenmem), capmem-newlenmem)
 	} else {
 		// Note: can't use rawmem (which avoids zeroing of memory), because then GC can scan uninitialized memory.
 		p = mallocgc(capmem, et, true)
-		if !writeBarrier.enabled {
-			memmove(p, old.array, lenmem)
-		} else {
-			for i := uintptr(0); i < lenmem; i += et.size {
-				typedmemmove(et, add(p, i), add(old.array, i))
-			}
+		if writeBarrier.enabled {
+			bulkBarrierPreWrite(uintptr(p), uintptr(old.array), lenmem)
 		}
 	}
+	memmove(p, old.array, lenmem)
 
 	return slice{p, old.len, newcap}
 }