qbit/go
1
0
Fork 0
mirror of https://github.com/golang/go synced 2024-11-19 21:04:43 -07:00
Code Releases Activity

runtime: optimize markspan

Browse Source 
Increases throughput by 2x on a memory hungry program on 8-node NUMA machine.

LGTM=rsc
R=rsc
CC=golang-codereviews
https://golang.org/cl/100230043
This commit is contained in:
Dmitriy Vyukov 2014-05-07 19:32:34 +04:00
parent 2e1ddeb136
commit acb03b8028
1 changed files with 12 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
src/pkg/runtime/mgc0.c
View File

@ -2785,7 +2785,7 @@ runtime·checkfreed(void *v, uintptr n)
void
runtime·markspan(void *v, uintptr size, uintptr n, bool leftover)
{
uintptr *b, off, shift, i;
uintptr *b, *b0, off, shift, i, x;
byte *p;
if((byte*)v+size*n > (byte*)runtime·mheap.arena_used || (byte*)v < runtime·mheap.arena_start)
@ -2804,6 +2804,9 @@ runtime·markspan(void *v, uintptr size, uintptr n, bool leftover)
p = v;
if(leftover) // mark a boundary just past end of last block too
n++;
b0 = nil;
x = 0;
for(; n-- > 0; p += size) {
// Okay to use non-atomic ops here, because we control
// the entire span, and each bitmap word has bits for only
@ -2812,8 +2815,15 @@ runtime·markspan(void *v, uintptr size, uintptr n, bool leftover)
off = (uintptr*)p - (uintptr*)runtime·mheap.arena_start; // word offset
b = (uintptr*)runtime·mheap.arena_start - off/wordsPerBitmapWord - 1;
shift = off % wordsPerBitmapWord;
*b = (*b & ~(bitMask<<shift)) | (bitAllocated<<shift);
if(b0 != b) {
if(b0 != nil)
*b0 = x;
b0 = b;
x = 0;
}
x |= bitAllocated<<shift;
}
*b0 = x;
}
// unmark the span of memory at v of length n bytes.