lattice/latticegpu.jl
1
0
Fork 0
mirror of https://igit.ific.uv.es/alramos/latticegpu.jl.git synced 2025-05-15 03:33:42 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

Working version. Comparison with one field seem ok.

Browse source
This commit is contained in:
Alberto Ramos 2021-10-15 15:52:21 +02:00
parent 30baeade9f
commit f0ca0bbde1
6 changed files with 85 additions and 72 deletions
Download patch file Download diff file Expand all files Collapse all files

15
src/Scalar/ScalarFields.jl
View file

@ -11,9 +11,11 @@
function randomize!(f, sp::ScalarParm{NS}, lp::SpaceParm, ymws::YMworkspace) where {NS}
m = CUDA.randn(ymws.PRC, lp.bsz, 4, NS, lp.rsz)
CUDA.@sync begin
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_assign_SU2fund!(f,m,sp,lp)
@timeit "Randomize scalar field" begin
m = CUDA.randn(ymws.PRC, lp.bsz, 4, NS, lp.rsz)
CUDA.@sync begin
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_assign_SU2fund!(f,m,sp,lp)
end
end
return nothing
@ -21,10 +23,13 @@ end
function krnl_assign_SU2fund!(f::AbstractArray{T}, m, sp::ScalarParm{NS}, lp::SpaceParm) where {T, NS}
# Think about precision here
SR2 = 1.4142135623730951
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
for i in 1:NS
f[b,i,r] = SU2fund(complex(m[b,1,i,r], m[b,2,i,r]),
complex(m[b,3,i,r], m[b,4,i,r]))
f[b,i,r] = SU2fund(complex(m[b,1,i,r]*SR2, m[b,2,i,r]*SR2),
complex(m[b,3,i,r]*SR2, m[b,4,i,r]*SR2))
end
return nothing