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

Debugged assemply split phase with interior suppression

Browse Source
This commit is contained in:
Peter Boyle 2017-04-23 19:30:19 -04:00
parent e3d0e31525
commit 3accb1ef89
1 changed files with 11 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
lib/qcd/action/fermion/WilsonKernelsAsmBody.h
View File

@ -75,22 +75,20 @@
LOAD64(%r10,isigns); \ LOAD64(%r10,isigns); \
PROJ(base); \ PROJ(base); \
MAYBEPERM(PERMUTE_DIR,perm); \ MAYBEPERM(PERMUTE_DIR,perm); \
} else if ( st.same_node[Dir] ) { \ }else if ( st.same_node[Dir] ) {LOAD_CHI(base);} \
LOAD_CHI(base); \
} \
base = st.GetInfo(ptype,local,perm,NxtDir,ent,plocal); ent++; \
PREFETCH_CHIMU(base); \
if ( local || st.same_node[Dir] ) { \ if ( local || st.same_node[Dir] ) { \
MULT_2SPIN_DIR_PF(Dir,basep); \ MULT_2SPIN_DIR_PF(Dir,basep); \
LOAD64(%r10,isigns); \ LOAD64(%r10,isigns); \
RECON; \ RECON; \
} } \
base = st.GetInfo(ptype,local,perm,NxtDir,ent,plocal); ent++; \
PREFETCH_CHIMU(base); \
#define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \ #define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \
base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++; \ base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++; \
PF_GAUGE(Xp); \ PF_GAUGE(Xp); \
PREFETCH1_CHIMU(base); \ PREFETCH1_CHIMU(base);{ ZERO_PSI; } \
ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) else { ZERO_PSI; } ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)
#define RESULT(base,basep) SAVE_RESULT(base,basep); #define RESULT(base,basep) SAVE_RESULT(base,basep);
@ -100,8 +98,10 @@
//////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////
#ifdef EXTERIOR #ifdef EXTERIOR
#define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \ #define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \
base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++; \ base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++; \
/*if((!local) ) {*/ \
if((!local)&&(!st.same_node[Dir]) ) { \ if((!local)&&(!st.same_node[Dir]) ) { \
LOAD_CHI(base); \ LOAD_CHI(base); \
MULT_2SPIN_DIR_PF(Dir,base); \ MULT_2SPIN_DIR_PF(Dir,base); \
@ -111,8 +111,8 @@
} }
#define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \ #define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \
nmu=0; \ nmu=0;{ ZERO_PSI; } \
ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) else { ZERO_PSI; } ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)
#define RESULT(base,basep) if (nmu){ ADD_RESULT(base,base);} #define RESULT(base,basep) if (nmu){ ADD_RESULT(base,base);}
@ -151,8 +151,7 @@
ASM_LEG(Tm,Xp,PERMUTE_DIR0,DIR7_PROJMEM,DIR7_RECON); ASM_LEG(Tm,Xp,PERMUTE_DIR0,DIR7_PROJMEM,DIR7_RECON);
#ifndef EXTERIOR #ifndef EXTERIOR
//Early out if no work if (nmu==0) break; //Early out if no work
//if (nmu==0) break;
#endif #endif
base = (uint64_t) &out._odata[ss]; base = (uint64_t) &out._odata[ss];
basep= st.GetPFInfo(nent,plocal); nent++; basep= st.GetPFInfo(nent,plocal); nent++;