portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-10 07:55:35 +00:00
Code Releases Activity

Hide internal data

Browse Source
This commit is contained in:
paboyle 2018-01-26 23:03:54 +00:00
parent bf659dfd92
commit 8b371ffa94
3 changed files with 21 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
lib/simd/BGQQPX.h
View File

@ -129,7 +129,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
*********************************************************/
#define LOCK_GAUGE(dir) \
{ \
uint8_t *byte_addr = (uint8_t *)&U._odata[sU](dir); \
uint8_t *byte_addr = (uint8_t *)&U[sU](dir); \
for(int i=0;i< 18*2*BYTES_PER_WORD*8;i+=32){ \
CACHE_LOCK(&byte_addr[i]); \
} \
@ -137,7 +137,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define UNLOCK_GAUGE(dir) \
{ \
uint8_t *byte_addr = (uint8_t *)&U._odata[sU](dir); \
uint8_t *byte_addr = (uint8_t *)&U[sU](dir); \
for(int i=0;i< 18*2*BYTES_PER_WORD*8;i+=32){ \
CACHE_UNLOCK(&byte_addr[i]); \
} \
@ -150,14 +150,14 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define PERMUTE_DIR1
#define PERMUTE_DIR0
#define MULT_2SPIN_DIR_PFXP(A,p) MULT_2SPIN(&U._odata[sU](A),p)
#define MULT_2SPIN_DIR_PFYP(A,p) MULT_2SPIN(&U._odata[sU](A),p)
#define MULT_2SPIN_DIR_PFZP(A,p) MULT_2SPIN(&U._odata[sU](A),p)
#define MULT_2SPIN_DIR_PFTP(A,p) MULT_2SPIN(&U._odata[sU](A),p)
#define MULT_2SPIN_DIR_PFXM(A,p) MULT_2SPIN(&U._odata[sU](A),p)
#define MULT_2SPIN_DIR_PFYM(A,p) MULT_2SPIN(&U._odata[sU](A),p)
#define MULT_2SPIN_DIR_PFZM(A,p) MULT_2SPIN(&U._odata[sU](A),p)
#define MULT_2SPIN_DIR_PFTM(A,p) MULT_2SPIN(&U._odata[sU](A),p)
#define MULT_2SPIN_DIR_PFXP(A,p) MULT_2SPIN(&U[sU](A),p)
#define MULT_2SPIN_DIR_PFYP(A,p) MULT_2SPIN(&U[sU](A),p)
#define MULT_2SPIN_DIR_PFZP(A,p) MULT_2SPIN(&U[sU](A),p)
#define MULT_2SPIN_DIR_PFTP(A,p) MULT_2SPIN(&U[sU](A),p)
#define MULT_2SPIN_DIR_PFXM(A,p) MULT_2SPIN(&U[sU](A),p)
#define MULT_2SPIN_DIR_PFYM(A,p) MULT_2SPIN(&U[sU](A),p)
#define MULT_2SPIN_DIR_PFZM(A,p) MULT_2SPIN(&U[sU](A),p)
#define MULT_2SPIN_DIR_PFTM(A,p) MULT_2SPIN(&U[sU](A),p)
#define MULT_SPIN(ptr,p) { \
uint64_t ub = ((uint64_t)base); \
@ -540,7 +540,7 @@ void testme(int osites,int ssU)
int local,perm, ptype;
uint64_t base;
uint64_t basep;
const uint64_t plocal =(uint64_t) & in._odata[0];
const uint64_t plocal =(uint64_t) & in[0];
// vComplexF isigns[2] = { signs[0], signs[1] };
//COMPLEX_TYPE is vComplexF of vComplexD depending
@ -675,7 +675,7 @@ void testme(int osites,int ssU)
// Xm
////////////////////////////////
#ifndef STREAM_STORE
basep= (uint64_t) &out._odata[ss];
basep= (uint64_t) &out[ss];
#endif
// basep= GetPFInfo(nent,plocal); nent++;
if ( local ) {
@ -770,7 +770,7 @@ void testme(int osites,int ssU)
} else {
LOAD_CHI(base);
}
base= (uint64_t) &out._odata[ss];
base= (uint64_t) &out[ss];
#ifndef STREAM_STORE
PREFETCH_CHIMU(base);
#endif

10
lib/simd/IBM_qpx.h
View File

@ -155,8 +155,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
*********************************************************/
#define LOCK_GAUGE(dir) \
{ \
uint64_t byte_addr = (uint64_t)&U._odata[sU]; \
int count = (sizeof(U._odata[0])+63)/64; \
uint64_t byte_addr = (uint64_t)&U[sU]; \
int count = (sizeof(U[0])+63)/64; \
asm (" mtctr %0 \n" \
" mr " HASH(REP) ", %1\n" \
" li " HASH(IMM) ", 64\n" \
@ -169,8 +169,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define UNLOCK_GAUGE(dir) \
{ \
uint64_t byte_addr = (uint64_t)&U._odata[sU]; \
int count = (sizeof(U._odata[0])+63)/64; \
uint64_t byte_addr = (uint64_t)&U[sU]; \
int count = (sizeof(U[0])+63)/64; \
asm (" mtctr %0 \n" \
" mr " HASH(REP) ", %1\n" \
" li " HASH(IMM) ", 64\n" \
@ -256,7 +256,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
}
#define MULT_2SPIN_DIR_PF(A,p) MULT_2SPIN_PF(&U._odata[sU](A),p)
#define MULT_2SPIN_DIR_PF(A,p) MULT_2SPIN_PF(&U[sU](A),p)
#define MULT_2SPIN_PF(ptr,pf) MULT_2SPIN(ptr,pf)
#define SAVE_RESULT(base,basep) { \

6
lib/simd/Intel512wilson.h
View File

@ -101,7 +101,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define LOCK_GAUGE(dir)
#define UNLOCK_GAUGE(dir)
// const SiteSpinor * ptr = & in._odata[offset];
// const SiteSpinor * ptr = & in[offset];
#define LOAD_CHIMU(PTR) LOAD64(%r8,PTR) __asm__ ( LOAD_CHIMUi );
#define LOAD_CHI(PTR) LOAD64(%r8,PTR) __asm__ ( LOAD_CHIi );
#define SAVE_UCHI(PTR) SAVE_UCHIi(PTR)
@ -179,7 +179,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
VSTORE(4,%r8,Chi_11) \
VSTORE(5,%r8,Chi_12) );
#define MULT_2SPIN_DIR_PF(A,p) MULT_2SPIN_PF(&U._odata[sU](A),p)
#define MULT_2SPIN_DIR_PF(A,p) MULT_2SPIN_PF(&U[sU](A),p)
#define MULT_2SPIN_PF(ptr,pf) MULT_2SPIN(ptr,pf)
//////////////////////////////////////////////////////////////////
@ -561,7 +561,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#endif
#define PF_GAUGE(A) \
LOAD64(%r8,&U._odata[sU](A)) \
LOAD64(%r8,&U[sU](A)) \
__asm__ ( \
VPREFETCH_G1(0,%r8) VPREFETCH_G1(1,%r8) \
VPREFETCH_G1(2,%r8) VPREFETCH_G1(3,%r8) \