portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-10 06:00:45 +01:00
Code Releases Activity

COntrol the prefetch strategy

Browse Source
This commit is contained in:
paboyle 2016-06-25 12:55:25 -07:00
parent db057cc276
commit b2933a0557
1 changed files with 28 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
lib/simd/Intel512common.h
View File

@ -28,6 +28,14 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#ifndef GRID_ASM_INTEL_COMMON_512_H
#define GRID_ASM_INTEL_COMMON_512_H
////////////////////////////////////////////////////////////////////////////////////////////////////
// Peformance options
////////////////////////////////////////////////////////////////////////////////////////////////////
#define AVX512_PF_L1
#undef AVX512_PF_L2_LINEAR
#undef AVX512_PF_L2_TABLE
#undef AVX512_PF_L2_WRITE
////////////////////////////////////////////////////////////////////////////////////////////////////
// Opcodes common
////////////////////////////////////////////////////////////////////////////////////////////////////
@ -88,10 +96,30 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define VMOVf(A,DEST) "vmovaps " #A ", " #DEST ";\n"
#define VMOVd(A,DEST) "vmovapd " #A ", " #DEST ";\n"
#ifdef AVX512_PF_L1
#define VPREFETCHG(O,A) "prefetcht0 "#O"*64("#A");\n"
#else
#define VPREFETCHG(O,A)
#endif
#ifdef AVX512_PF_L2_LINEAR
#define VPREFETCH2(O,A) "prefetcht1 "#O"*64("#A");\n"
#else
#define VPREFETCH2(O,A)
#endif
#ifdef AVX512_PF_L2_TABLE
#define VPREFETCHP(O,A) "prefetcht1 "#O"*64("#A");\n"
#else
#define VPREFETCHP(O,A)
#endif
#ifdef AVX512_PF_L2_WRITE
#define VPREFETCHW(O,A) "prefetchwt1 "#O"*64("#A");\n"
#else
#define VPREFETCHW(O,A)
#endif
#define VPREFETCHNTA(O,A)
#define VPREFETCH(O,A)