1
0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-18 07:47:06 +01:00

Global changes to parallel_for structure.

Move the comms flags to more sensible names
This commit is contained in:
paboyle
2017-02-21 05:24:27 -05:00
parent 3906cd2149
commit 3ae92fa2e6
43 changed files with 271 additions and 513 deletions

View File

@ -54,8 +54,8 @@ void CayleyFermion5D<Impl>::M5D(const FermionField &psi,
// Flops = 6.0*(Nc*Ns) *Ls*vol
M5Dcalls++;
M5Dtime-=usecond();
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
for(int s=0;s<Ls;s++){
auto tmp = psi._odata[0];
if ( s==0 ) {
@ -98,8 +98,8 @@ void CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi,
// Flops = 6.0*(Nc*Ns) *Ls*vol
M5Dcalls++;
M5Dtime-=usecond();
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
auto tmp = psi._odata[0];
for(int s=0;s<Ls;s++){
if ( s==0 ) {
@ -137,8 +137,7 @@ void CayleyFermion5D<Impl>::MooeeInv (const FermionField &psi, FermionField &
MooeeInvCalls++;
MooeeInvTime-=usecond();
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
auto tmp = psi._odata[0];
// flops = 12*2*Ls + 12*2*Ls + 3*12*Ls + 12*2*Ls = 12*Ls * (9) = 108*Ls flops
@ -184,8 +183,7 @@ void CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi, FermionField &
MooeeInvCalls++;
MooeeInvTime-=usecond();
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
auto tmp = psi._odata[0];