portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-11-04 05:54:32 +00:00
Code Releases Activity

Better Avx512 assembly sequence for SU3 using fmaddsub to get the imag imag sign

Browse Source
This commit is contained in:
paboyle
2016-03-30 08:39:39 +01:00
parent 1ecbf9794d
commit 83b15bfcdd
1 changed files with 10 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
lib/simd/Intel512wilson.h
View File

@@ -290,17 +290,17 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
ZEND2(UChi_02,Z4,Chi_02) \ ZEND2(UChi_02,Z4,Chi_02) \
ZEND2(UChi_12,Z5,Chi_12) ); ZEND2(UChi_12,Z5,Chi_12) );
#define MULT_2SPINa(ptr) MULT_2SPIN_PF(ptr,ptr,VPREFETCHG); #define MULT_2SPIN_PFXM(ptr,pf) MULT_2SPIN(ptr)
#define MULT_2SPIN(ptr) MULT_ADDSUB_2SPIN(ptr); #define MULT_2SPIN_PFYM(ptr,pf) MULT_2SPIN(ptr)
#define MULT_2SPIN_PFZM(ptr,pf) MULT_2SPIN(ptr)
#define MULT_2SPIN_PFTM(ptr,pf) MULT_2SPIN(ptr)
#define MULT_2SPIN_PFTP(ptr,pf) MULT_2SPIN(ptr)
#define MULT_2SPIN_PFZP(ptr,pf) MULT_2SPIN(ptr)
#define MULT_2SPIN_PFYP(ptr,pf) MULT_2SPIN(ptr)
#define MULT_2SPIN_PFXP(ptr,pf) MULT_2SPIN(ptr)
#define MULT_2SPIN_PFXM(ptr,pf) MULT_2SPIN_PF(ptr,pf,VPREFETCHNTA) #define MULT_2SPINa(ptr) MULT_2SPIN_PF(ptr,ptr,VPREFETCHG);
#define MULT_2SPIN_PFYM(ptr,pf) MULT_2SPIN_PF(ptr,pf,VPREFETCHNTA) #define MULT_2SPIN(ptr) MULT_ADDSUB_2SPIN(ptr);
#define MULT_2SPIN_PFZM(ptr,pf) MULT_2SPIN_PF(ptr,pf,VPREFETCHNTA)
#define MULT_2SPIN_PFTM(ptr,pf) MULT_2SPIN_PF(ptr,pf,VPREFETCHNTA)
#define MULT_2SPIN_PFTP(ptr,pf) MULT_2SPIN_PF(ptr,pf,VPREFETCHNTA)
#define MULT_2SPIN_PFZP(ptr,pf) MULT_2SPIN_PF(ptr,pf,VPREFETCHNTA)
#define MULT_2SPIN_PFYP(ptr,pf) MULT_2SPIN_PF(ptr,pf,VPREFETCH)
#define MULT_2SPIN_PFXP(ptr,pf) MULT_2SPIN_PF(ptr,pf,VPREFETCHNTA)
#define MULT_2SPIN_PF(ptr,pf,VPF) \ #define MULT_2SPIN_PF(ptr,pf,VPF) \