portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-19 08:17:05 +01:00
Code Releases Activity

Merge branch 'develop' into feature/distil

Browse Source 
* develop:
  Hadron WeakEye and A2ALoop bug fixes, and WWVVContraction bug fix
  DiskVector: fix of memory bug triggering segfault when the cache is accessed following a certain pattern
  MFermion::GaugeProp fix for 4d fields
This commit is contained in:
Michael Marshall
2019-05-14 13:10:40 +01:00
parent c16916cc45 d8512b03f8
commit ec7d96ce3b
5 changed files with 36 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
Grid/qcd/utils/A2Autils.h
View File

@ -1385,17 +1385,18 @@ void A2Autils<FImpl>::ContractWWVV(std::vector<PropagatorField> &WWVV,
for(int t=0;t<N_t;t++){
for(int s=0;s<N_s;s++){
auto tmp1 = vs[s]._odata[ss];
vobj tmp2 = zero;
vobj tmp2 = zero;
vobj tmp3 = zero;
for(int d=d_o;d<MIN(d_o+d_unroll,N_d);d++){
Scalar_v coeff = WW_sd(t,s,d);
mac(&tmp2 ,& coeff, & vd[d]._odata[ss]);
}
tmp3 = conjugate(vd[d]._odata[ss]);
mac(&tmp2, &coeff, &tmp3);
}
//////////////////////////
// Fast outer product of tmp1 with a sum of terms suppressed by d_unroll
//////////////////////////
tmp2 = conjugate(tmp2);
for(int s1=0;s1<Ns;s1++){
for(int s2=0;s2<Ns;s2++){
WWVV[t]._odata[ss]()(s1,s2)(0,0) += tmp1()(s1)(0)*tmp2()(s2)(0);