code review 12440044: runtime: tune append crossover on amd64 and 386

runtime: tune append crossover on amd64 and 386

Fixes issue 4963.

Sets the append crossover to 0 on intel platforms.

Results for linux/amd64 Core i5 SNB

benchmark                     old ns/op    new ns/op    delta
BenchmarkAppend                     102          104   +1.96%
BenchmarkAppend1Byte                 10           11   +0.92%
BenchmarkAppend4Bytes                15           11  -28.10%
BenchmarkAppend7Bytes                17           12  -32.58%
BenchmarkAppend8Bytes                18           12  -36.17%
BenchmarkAppend15Bytes               24           11  -55.02%
BenchmarkAppend16Bytes               25           11  -56.03%
BenchmarkAppend32Bytes               11           12   +4.31%
BenchmarkAppendStr1Byte               8            9  +13.99%
BenchmarkAppendStr4Bytes             11            9  -17.52%
BenchmarkAppendStr8Bytes             14            9  -35.70%
BenchmarkAppendStr16Bytes            21            9  -55.19%
BenchmarkAppendStr32Bytes            10           10   -5.66%
BenchmarkAppendSpecialCase           49           52   +7.96%

Results for linux/386 Atom(TM) CPU 330 @ 1.60GHz

benchmark                     old ns/op    new ns/op    delta
BenchmarkAppend                     219          218   -0.46%
BenchmarkAppend1Byte                 75           72   -3.44%
BenchmarkAppend4Bytes                92           73  -19.87%
BenchmarkAppend7Bytes               108           74  -31.20%
BenchmarkAppend8Bytes               116           74  -35.95%
BenchmarkAppend15Bytes              162           77  -52.22%
BenchmarkAppend16Bytes              169           77  -54.20%
BenchmarkAppend32Bytes               88           86   -2.38%
BenchmarkAppendStr1Byte              57           59   +3.32%
BenchmarkAppendStr4Bytes             72           59  -17.40%
BenchmarkAppendStr8Bytes             92           60  -34.70%
BenchmarkAppendStr16Bytes           141           63  -54.89%
BenchmarkAppendStr32Bytes            75           73   -2.64%
BenchmarkAppendSpecialCase          270          270   +0.00%

[dave_cheney.net, 2013-08-05 10:24:26]

Hello golang-dev@googlegroups.com,

I'd like you to review this change to
https://code.google.com/p/go

[dave_cheney.net, 2013-08-05 10:25:32]

On 2013/08/05 10:24:26, dfc wrote:
> Hello mailto:golang-dev@googlegroups.com,
> 
> I'd like you to review this change to
> https://code.google.com/p/go

Reducing the arm crossover from 8 to 4 did not significantly improve the
benchmarks.

[r, 2013-08-05 10:34:02]

I'm suspicious. What happened to memmove?

-rob

[r, 2013-08-05 10:34:19]

https://codereview.appspot.com/12440044/diff/10001/src/pkg/runtime/slice.c
File src/pkg/runtime/slice.c (right):

https://codereview.appspot.com/12440044/diff/10001/src/pkg/runtime/slice.c#ne...
src/pkg/runtime/slice.c:98: if(appendCrossover && w <= appendCrossover) {
appendCrossover is not a boolean.

[dave_cheney.net, 2013-08-05 10:39:09]

> I'm suspicious. What happened to memmove?

khr made it better after tip reopened in May,
https://code.google.com/p/go/source/detail?r=4cb93e2900d0

[dave_cheney.net, 2013-08-05 10:49:01]

Hello golang-dev@googlegroups.com, r@golang.org (cc:
golang-dev@googlegroups.com),

Please take another look.

[r, 2013-08-05 12:24:59]

LGTM

[dave_cheney.net, 2013-08-05 21:51:54]

*** Submitted as https://code.google.com/p/go/source/detail?r=e285a7b170aa ***

runtime: tune append crossover on amd64 and 386

Fixes issue 4963.

Sets the append crossover to 0 on intel platforms.

Results for linux/amd64 Core i5 SNB

benchmark                     old ns/op    new ns/op    delta
BenchmarkAppend                     102          104   +1.96%
BenchmarkAppend1Byte                 10           11   +0.92%
BenchmarkAppend4Bytes                15           11  -28.10%
BenchmarkAppend7Bytes                17           12  -32.58%
BenchmarkAppend8Bytes                18           12  -36.17%
BenchmarkAppend15Bytes               24           11  -55.02%
BenchmarkAppend16Bytes               25           11  -56.03%
BenchmarkAppend32Bytes               11           12   +4.31%
BenchmarkAppendStr1Byte               8            9  +13.99%
BenchmarkAppendStr4Bytes             11            9  -17.52%
BenchmarkAppendStr8Bytes             14            9  -35.70%
BenchmarkAppendStr16Bytes            21            9  -55.19%
BenchmarkAppendStr32Bytes            10           10   -5.66%
BenchmarkAppendSpecialCase           49           52   +7.96%

Results for linux/386 Atom(TM) CPU 330 @ 1.60GHz

benchmark                     old ns/op    new ns/op    delta
BenchmarkAppend                     219          218   -0.46%
BenchmarkAppend1Byte                 75           72   -3.44%
BenchmarkAppend4Bytes                92           73  -19.87%
BenchmarkAppend7Bytes               108           74  -31.20%
BenchmarkAppend8Bytes               116           74  -35.95%
BenchmarkAppend15Bytes              162           77  -52.22%
BenchmarkAppend16Bytes              169           77  -54.20%
BenchmarkAppend32Bytes               88           86   -2.38%
BenchmarkAppendStr1Byte              57           59   +3.32%
BenchmarkAppendStr4Bytes             72           59  -17.40%
BenchmarkAppendStr8Bytes             92           60  -34.70%
BenchmarkAppendStr16Bytes           141           63  -54.89%
BenchmarkAppendStr32Bytes            75           73   -2.64%
BenchmarkAppendSpecialCase          270          270   +0.00%

R=golang-dev, r
CC=golang-dev
https://codereview.appspot.com/12440044