Investigate results from finufft and cufinufft
Overview A comparison of model slices calculation (forward).
Run Command:
./scripts/run_summit_mult.sh -m -n 1 -a 1 -g 1 -r 1 -d 1 -c -f -t 1 -l tests/forward_and_adjoint.py
SINGLE PRECISION INPUT (H,K,L)
finufft 1.0 and cufinufft
Compare Forward cpu and gpu:
dtype cpu:float64 gpu:float64
shape cpu:(10485760,) gpu:(10485760,)
equal? False
cc: 0.9999999999999977
r: 4.212172146783729e-09
finufft 2.0 and cufinufft
Compare Forward cpu and gpu:
dtype cpu:float64 gpu:float64
shape cpu:(10485760,) gpu:(10485760,)
equal? False
cc: 0.9999999999985627
r: 0.999999709550877
DOUBLE PRECISION INPUT
finufft 1.0 and cufinufft
Compare Forward cpu and gpu:
dtype cpu:float64 gpu:float64
shape cpu:(10485760,) gpu:(10485760,)
equal? False
cc: 0.9999999999999982
r: 4.212172103799584e-09
finufft 2.0 and cufinufft
No support yet in finufft 2.0.
Note: Current we use finufft 1.0 with fortran order. Spinfel arrays need rearrangement to c-order to switch to finufft 2.0