runtime: add memory prefetching to GC (f09e63a2) · Commits · go / golang

Commit f09e63a2 authored Apr 07, 2012 by

Dmitriy Vyukov

runtime: add memory prefetching to GC

benchmark old ns/op new ns/op delta

garbage.BenchmarkParser 4448988000 4370531000 -1.76%
garbage.BenchmarkParser-2 4086045000 4023083000 -1.54%
garbage.BenchmarkParser-4 3677365000 3667020000 -0.28%
garbage.BenchmarkParser-8 3517253000 3543946000 +0.76%
garbage.BenchmarkParser-16 3506562000 3512518000 +0.17%

garbage.BenchmarkTree 494435529 505784058 +2.30%
garbage.BenchmarkTree-2 499652705 502774823 +0.62%
garbage.BenchmarkTree-4 468482117 465713352 -0.59%
garbage.BenchmarkTree-8 488533235 482287000 -1.28%
garbage.BenchmarkTree-16 507835176 500654882 -1.41%

garbage.BenchmarkTree2 31453900 28804600 -8.42%
garbage.BenchmarkTree2-2 21440600 19065800 -11.08%
garbage.BenchmarkTree2-4 10982000 10009100 -8.86%
garbage.BenchmarkTree2-8 7544700 6479800 -14.11%
garbage.BenchmarkTree2-16 7049500 6163200 -12.57%

garbage.BenchmarkParserPause 135815000 125360666 -7.70%
garbage.BenchmarkParserPause-2 92691523 84365476 -8.98%
garbage.BenchmarkParserPause-4 53392190 46995809 -11.98%
garbage.BenchmarkParserPause-8 36059523 30998900 -14.03%
garbage.BenchmarkParserPause-16 30174300 27613350 -8.49%

garbage.BenchmarkTreePause 20969784 22568102 +7.62%
garbage.BenchmarkTreePause-2 20215875 20975130 +3.76%
garbage.BenchmarkTreePause-4 17240709 17180666 -0.35%
garbage.BenchmarkTreePause-8 18196386 18205870 +0.05%
garbage.BenchmarkTreePause-16 20621158 20486867 -0.65%

garbage.BenchmarkTree2Pause 173992142 159995285 -8.04%
garbage.BenchmarkTree2Pause-2 131281904 118013714 -10.11%
garbage.BenchmarkTree2Pause-4 93484952 85092666 -8.98%
garbage.BenchmarkTree2Pause-8 88950523 77340809 -13.05%
garbage.BenchmarkTree2Pause-16 86071238 76557952 -11.05%

garbage.BenchmarkParserLastPause 327247000 288205000 -11.93%
garbage.BenchmarkParserLastPause-2 217039000 187336000 -13.69%
garbage.BenchmarkParserLastPause-4 119722000 105069000 -12.24%
garbage.BenchmarkParserLastPause-8 70806000 64755000 -8.55%
garbage.BenchmarkParserLastPause-16 62813000 53486000 -14.85%

garbage.BenchmarkTreeLastPause 28420000 29735000 +4.63%
garbage.BenchmarkTreeLastPause-2 23514000 25427000 +8.14%
garbage.BenchmarkTreeLastPause-4 21773000 19548000 -10.22%
garbage.BenchmarkTreeLastPause-8 24072000 24046000 -0.11%
garbage.BenchmarkTreeLastPause-16 25149000 25291000 +0.56%

garbage.BenchmarkTree2LastPause 314491000 287988000 -8.43%
garbage.BenchmarkTree2LastPause-2 214363000 190616000 -11.08%
garbage.BenchmarkTree2LastPause-4 109778000 100052000 -8.86%
garbage.BenchmarkTree2LastPause-8 75390000 64753000 -14.11%
garbage.BenchmarkTree2LastPause-16 70333000 61484000 -12.58%

FTR, below are result with the empty prefetch function,
that is, single RET but no real prefetching.
It suggests that inlinable PREFETCH is worth pursuing.

benchmark old ns/op new ns/op delta

garbage.BenchmarkParser 4448988000 4560488000 +2.51%
garbage.BenchmarkParser-2 4086045000 4129728000 +1.07%
garbage.BenchmarkParser-4 3677365000 3728672000 +1.40%
garbage.BenchmarkParser-8 3517253000 3583968000 +1.90%
garbage.BenchmarkParser-16 3506562000 3591414000 +2.42%

garbage.BenchmarkTree 494435529 499580882 +1.04%
garbage.BenchmarkTree-4 468482117 467387294 -0.23%
garbage.BenchmarkTree-8 488533235 478311117 -2.09%
garbage.BenchmarkTree-2 499652705 499324235 -0.07%
garbage.BenchmarkTree-16 507835176 502005705 -1.15%

garbage.BenchmarkTree2 31453900 33296800 +5.86%
garbage.BenchmarkTree2-2 21440600 22466400 +4.78%
garbage.BenchmarkTree2-4 10982000 11402700 +3.83%
garbage.BenchmarkTree2-8 7544700 7476500 -0.90%
garbage.BenchmarkTree2-16 7049500 7338200 +4.10%

garbage.BenchmarkParserPause 135815000 139529142 +2.73%
garbage.BenchmarkParserPause-2 92691523 95229190 +2.74%
garbage.BenchmarkParserPause-4 53392190 53083476 -0.58%
garbage.BenchmarkParserPause-8 36059523 34594800 -4.06%
garbage.BenchmarkParserPause-16 30174300 30063300 -0.37%

garbage.BenchmarkTreePause 20969784 21866920 +4.28%
garbage.BenchmarkTreePause-2 20215875 20731125 +2.55%
garbage.BenchmarkTreePause-4 17240709 17275837 +0.20%
garbage.BenchmarkTreePause-8 18196386 17898777 -1.64%
garbage.BenchmarkTreePause-16 20621158 20662772 +0.20%

garbage.BenchmarkTree2Pause 173992142 184336857 +5.95%
garbage.BenchmarkTree2Pause-2 131281904 138005714 +5.12%
garbage.BenchmarkTree2Pause-4 93484952 98449238 +5.31%
garbage.BenchmarkTree2Pause-8 88950523 89286095 +0.38%
garbage.BenchmarkTree2Pause-16 86071238 89568666 +4.06%

garbage.BenchmarkParserLastPause 327247000 342189000 +4.57%
garbage.BenchmarkParserLastPause-2 217039000 217224000 +0.09%
garbage.BenchmarkParserLastPause-4 119722000 121327000 +1.34%
garbage.BenchmarkParserLastPause-8 70806000 71941000 +1.60%
garbage.BenchmarkParserLastPause-16 62813000 60166000 -4.21%

garbage.BenchmarkTreeLastPause 28420000 27840000 -2.04%
garbage.BenchmarkTreeLastPause-2 23514000 27390000 +16.48%
garbage.BenchmarkTreeLastPause-4 21773000 21414000 -1.65%
garbage.BenchmarkTreeLastPause-8 24072000 21705000 -9.83%
garbage.BenchmarkTreeLastPause-16 25149000 23932000 -4.84%

garbage.BenchmarkTree2LastPause 314491000 332894000 +5.85%
garbage.BenchmarkTree2LastPause-2 214363000 224611000 +4.78%
garbage.BenchmarkTree2LastPause-4 109778000 113976000 +3.82%
garbage.BenchmarkTree2LastPause-8 75390000 67223000 -10.83%
garbage.BenchmarkTree2LastPause-16 70333000 73216000 +4.10%

R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/5991057

parent c9529e02

Hide whitespace changes

Inline Side-by-side

View file @ f09e63a2

@@ -268,6 +268,8 @@ scanblock(byte *b, int64 n)
 			if((bits & bitNoPointers) != 0)
 				continue;
 			PREFETCH(obj);
 			// If another proc wants a pointer, give it some.
 			if(nobj > 4 && work.nwait > 0 && work.full == nil) {
 				wbuf->nobj = nobj;
-...

Please register or to comment