portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-25 05:05:56 +01:00
Code Releases Activity

Typo fixes and rotate for CLANG

Browse Source
This commit is contained in:
azusayamaguchi 2016-10-21 22:34:29 +01:00
parent 09fd5c43a7
commit 75ebd3a0d1
3 changed files with 13 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
lib/simd/Grid_avx512.h
View File

@ -41,6 +41,16 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
namespace Grid{
namespace Optimization {
union u512f {
__m512 v;
float f[16];
};
union u512d {
__m512d v;
double f[8];
};
struct Vsplat{
//Complex float
@ -361,7 +371,7 @@ namespace Optimization {
// Some Template specialization
// Hack for CLANG until mm512_reduce_add_ps etc... are implemented in GCC and Clang releases
#undef GNU_CLANG_COMPILER
#define GNU_CLANG_COMPILER
#ifdef GNU_CLANG_COMPILER
//Complex float Reduce
template<>

2
lib/simd/Intel512avx.h
View File

@ -53,7 +53,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define ZMULMEM2SPd(O,P,tmp,B,C,Briir,Biirr,Criir,Ciirr)\
VSHUFMEMd(O,P,tmp) \
VMULMEMd(O,P,B,Biirr) \
VMULMEMd(O,P,B,Biirr) \
VMULMEMd(O,P,C,Ciirr) \
VMULd(tmp,B,Briir) \
VMULd(tmp,C,Criir)

2
lib/simd/Intel512common.h
View File

@ -37,7 +37,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
// Opcodes common
////////////////////////////////////////////////////////////////////////////////////////////////////
#define MASK_REGS \
__asm__ ("mov $0xAAAA, %%eax \n"\
__asm__ ("mov $0xAAAA, %%eax \n"\
"kmovw %%eax, %%k6 \n"\
"mov $0x5555, %%eax \n"\
"kmovw %%eax, %%k7 \n" : : : "%eax");