1
0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-17 15:27:06 +01:00

Improvements to the assembler interface that let us move chunks of the

site and s loop into the kernels. This will save on function call overhead and
guarantee L2 prefetching strategy is right since OMP can't distribute the
sub-chunks of work.
This commit is contained in:
paboyle
2016-06-09 01:12:36 -07:00
parent d9408893b3
commit 55f65b81b5
10 changed files with 77 additions and 87 deletions

View File

@ -2,19 +2,24 @@
int locala,perma, ptypea;
int localb,permb, ptypeb;
uint64_t basea, baseb;
uint64_t basex;
const uint64_t plocal =(uint64_t) & in._odata[0];
// vComplexF isigns[2] = { signs[0], signs[1] };
vComplexF *isigns = &signs[0];
MASK_REGS;
for(int site=0;site<Ns;site++) {
for(int s=0;s<Ls;s++) {
////////////////////////////////
// Xp
////////////////////////////////
int ent=ss*8;// 2*Ndim
basea = st.GetInfo(ptypea,locala,perma,Xp,ent,plocal); ent++;
baseb = st.GetInfo(ptypeb,localb,permb,Yp,ent,plocal); ent++;
basex = basea;
if ( locala ) {
LOAD64(%r10,isigns);
@ -149,5 +154,8 @@
TP_RECON_ACCUM;
SAVE_RESULT(&out._odata[ss]);
ss++;
}
sU++;
}
}