portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-09 21:50:45 +01:00
Code Releases Activity

Use explicit linalg calls to get coalesce optimisations on GPU

Browse Source
This commit is contained in:
Peter Boyle 2020-01-27 12:44:51 -05:00
parent 8cec294ec9
commit 49e123dbda
1 changed files with 9 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
Grid/algorithms/approx/Chebyshev.h
View File

@ -252,20 +252,20 @@ public:
RealD xscale = 2.0/(hi-lo); RealD xscale = 2.0/(hi-lo);
RealD mscale = -(hi+lo)/(hi-lo); RealD mscale = -(hi+lo)/(hi-lo);
Linop.HermOp(T0,y); Linop.HermOp(T0,y);
T1=y*xscale+in*mscale; axpby(T1,xscale,mscale,y,in);
// sum = .5 c[0] T0 + c[1] T1 // sum = .5 c[0] T0 + c[1] T1
out = (0.5*Coeffs[0])*T0 + Coeffs[1]*T1; // out = ()*T0 + Coeffs[1]*T1;
axpby(out,0.5*Coeffs[0],Coeffs[1],T0,T1);
for(int n=2;n<order;n++){ for(int n=2;n<order;n++){
Linop.HermOp(*Tn,y); Linop.HermOp(*Tn,y);
// y=xscale*y+mscale*(*Tn);
y=xscale*y+mscale*(*Tn); // *Tnp=2.0*y-(*Tnm);
// out=out+Coeffs[n]* (*Tnp);
*Tnp=2.0*y-(*Tnm); axpby(y,xscale,mscale,y,(*Tn));
axpby(*Tnp,2.0,-1.0,y,(*Tnm));
out=out+Coeffs[n]* (*Tnp); axpy(out,Coeffs[n],*Tnp,out);
// Cycle pointers to avoid copies // Cycle pointers to avoid copies
Field *swizzle = Tnm; Field *swizzle = Tnm;
Tnm =Tn; Tnm =Tn;