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

Work in scalar observables

Browse source
This commit is contained in:
Alberto Ramos 2021-10-20 13:36:39 +02:00
parent 38c861e042
commit 0654f15564
4 changed files with 117 additions and 87 deletions
Download patch file Download diff file Expand all files Collapse all files

67
src/Scalar/ScalarObs.jl
View file

@ -3,45 +3,33 @@
### Alberto Ramos wrote this file. As long as you retain this
### notice you can do whatever you want with this stuff. If we meet some
### day, and you think this stuff is worth it, you can buy me a beer in
### return. <alberto.ramos@cern.ch>
### return. <>
###
### file: YMact.jl
### created: Mon Jul 12 18:31:19 2021
###
"""
each instance defines fields to store observables in each lattice point
"""
struct Obs{T}
rho2 #ρ^2
lphi #L_\phi
lalpha #L_\alpha
function Obs(::Type{T}, lp::SpaceParm, sp::ScalarParm, gp::GaugeParm) where {T <: AbstractFloat}
rho2n = nscalar_field(Complex{T}, length(sp.kap), lp)
lphin = nscalar_field(Complex{T}, length(sp.kap), lp)
lalphan = nscalar_field(Complex{T}, length(sp.kap), lp)
return new{T}(rho2n, lphin, lalphan)
end
end
"""
computes global observables by calling krnl_obs! and summing
for all lattice points
"""
function updt_obs!(obs::Obs{T}, U, Phi, sp::ScalarParm, lp::SpaceParm) where {T <: AbstractFloat}
function scalar_obs(U, Phi, sp::ScalarParm, lp::SpaceParm, ymws::YMworkspace)
CUDA.@sync begin
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_obs!(obs, U, Phi, sp, lp)
@timeit "Scalar observables" begin
CUDA.@sync begin
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_obs!(ymws.rm, ymws.cm, U, Phi, sp, lp)
end
V = prod(lp.iL)
#summation of global observables
rho2 = CUDA.mapreduce(norm2, +, Phi)/V
lphi = CUDA.reduce(+, ymws.rm)/(lp.ndim*V)
lalpha = CUDA.mapreduce(real, +, ymws.cm)/(lp.ndim*V)
end
#summation of global observables
rho2 = CUDA.reduce(+, obs.rho2)
lphi = CUDA.reduce(+, obs.lphi)
lalpha = CUDA.reduce(+, obs.lalpha)
return (rho2,lphi,lalpha)
return rho2, lphi, lalpha
end
"""
@ -49,14 +37,13 @@ end
for each lattice point
"""
function krnl_obs!(obs::Obs{T}, U::AbstractArray{TG}, Phi::AbstractArray{TS}, sp::ScalarParm{NP,T}, lp::SpaceParm{N,M,D}) where {TG,TS,NP,T,N,M,D}
function krnl_obs!(rm, cm, U::AbstractArray{TG}, Phi::AbstractArray{TS}, sp::ScalarParm{NP,T}, lp::SpaceParm{N,M,D}) where {TG,TS,NP,T,N,M,D}
#thread/block coordinate
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
Ush = @cuStaticSharedMem(TG, (D,N)) #Links - array (lattice points)x(spacetime dimension)
Psh = @cuStaticSharedMem(TS, (D,NP)) #Phield - array (lattice points)x(scalalr dimension)
Ush = @cuStaticSharedMem(TG, (D,N))
Psh = @cuStaticSharedMem(TS, (D,NP))
for id in 1:N
Ush[b,id] = U[b,id,r]
@ -66,16 +53,14 @@ function krnl_obs!(obs::Obs{T}, U::AbstractArray{TG}, Phi::AbstractArray{TS}, sp
end
sync_threads()
IX = point_coord((b,r), lp)
rm[IX] = zero(eltype(rm))
cm[IX] = zero(eltype(cm))
#compute obs
for i in 1:NP
obs.rho2[b,i,r] = norm2( Psh[b,i] )
obs.lphi[b,i,r] = zero( obs.lphi[b,i,r] )
obs.lalpha[b,i,r] = zero( obs.lalpha[b,i,r] )
norm = norm( Psh[b,i] )
for mu in 1:N
#up fields
psq = norm( Psh[b,i] )
for id in 1:N
bu, ru = up((b, r), id, lp) #thread/block coordinate of up point
if (ru == r)
phiup = Psh[bu,i]
@ -83,10 +68,10 @@ function krnl_obs!(obs::Obs{T}, U::AbstractArray{TG}, Phi::AbstractArray{TS}, sp
phiup = Phi[bu,i,ru]
end
normup = norm(phiup)
psqup = norm(phiup)
obs.lphi[b,i,r] += tr( Psh[b,i], Ush[b,i]*phiup )
obs.lalpha[b,i,r] += tr( Psh[b,i]/norm, Ush[b,i]*phiup/normup )
rm[IX] += dot( Psh[b,i], Ush[b,id]*phiup )
cm[IX] += complex(rm[IX])/(psq*psqup)
end
end