mirror of
https://igit.ific.uv.es/alramos/latticegpu.jl.git
synced 2025-05-15 03:33:42 +02:00
Last version
This commit is contained in:
Alberto Ramos
parent
c378648508
commit
76d0b66b4b
9 changed files with 515 additions and 322 deletions
122
src/YM/YMhmc.jl
122
src/YM/YMhmc.jl
|
|
@ -9,10 +9,10 @@
|
|||
### created: Thu Jul 15 11:27:28 2021
|
||||
###
|
||||
|
||||
function gauge_action(U, lp::SpaceParm, gp::GaugeParm, kp::KernelParm, ymws::YMworkspace)
|
||||
function gauge_action(U, lp::SpaceParm, gp::GaugeParm, ymws::YMworkspace)
|
||||
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=kp.threads blocks=kp.blocks krnl_plaq!(ymws.cm, U, lp)
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_plaq!(ymws.cm, U, lp)
|
||||
end
|
||||
S = gp.beta*( prod(lp.iL)*lp.npls -
|
||||
CUDA.mapreduce(real, +, ymws.cm)/gp.ng )
|
||||
|
|
@ -20,32 +20,34 @@ function gauge_action(U, lp::SpaceParm, gp::GaugeParm, kp::KernelParm, ymws::YMw
|
|||
return S
|
||||
end
|
||||
|
||||
function plaquette(U, lp::SpaceParm, gp::GaugeParm, kp::KernelParm, ymws::YMworkspace)
|
||||
function plaquette(U, lp::SpaceParm, gp::GaugeParm, ymws::YMworkspace)
|
||||
|
||||
println("Entro!")
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=kp.threads blocks=kp.blocks krnl_plaq!(ymws.cm, U, lp)
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_plaq!(ymws.cm, U, lp)
|
||||
end
|
||||
println("Salgo!")
|
||||
|
||||
return CUDA.mapreduce(real, +, real.(ymws.cm))/(prod(lp.iL)*lp.npls)
|
||||
return CUDA.mapreduce(real, +, ymws.cm)/(prod(lp.iL)*lp.npls)
|
||||
end
|
||||
|
||||
function hamiltonian(mom, U, lp, gp, kp, ymws)
|
||||
K = norm2(mom)/2.0
|
||||
V = gauge_action(U, lp, gp, kp, ymws)
|
||||
function hamiltonian(mom, U, lp, gp, ymws)
|
||||
K = norm_field(mom)/2.0
|
||||
V = gauge_action(U, lp, gp, ymws)
|
||||
println("K: ", K, " V: ", V)
|
||||
return K+V
|
||||
end
|
||||
|
||||
function HMC!(U, eps, ns, lp::SpaceParm, gp::GaugeParm, kp::KernelParm, ymws::YMworkspace; noacc=false)
|
||||
function HMC!(U, eps, ns, lp::SpaceParm, gp::GaugeParm, ymws::YMworkspace; noacc=false)
|
||||
|
||||
ymws.U1 .= U
|
||||
|
||||
randomn!(ymws.mom)
|
||||
hini = hamiltonian(ymws.mom, U, lp, gp, kp, ymws)
|
||||
randomn!(ymws.mom, lp)
|
||||
hini = hamiltonian(ymws.mom, U, lp, gp, ymws)
|
||||
|
||||
OMF4!(ymws.mom, U, eps, ns, lp, gp, kp, ymws)
|
||||
OMF4!(ymws.mom, U, eps, ns, lp, gp, ymws)
|
||||
|
||||
dh = hamiltonian(ymws.mom, U, lp, gp, kp, ymws) - hini
|
||||
dh = hamiltonian(ymws.mom, U, lp, gp, ymws) - hini
|
||||
pacc = exp(-dh)
|
||||
|
||||
acc = true
|
||||
|
|
@ -66,18 +68,17 @@ end
|
|||
|
||||
function krnl_updt!(mom, frc1, frc2, eps1, U, eps2, lp::SpaceParm)
|
||||
|
||||
X = map2latt((CUDA.threadIdx().x,CUDA.threadIdx().y,CUDA.threadIdx().z),
|
||||
(CUDA.blockIdx().x,CUDA.blockIdx().y,CUDA.blockIdx().z))
|
||||
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
|
||||
|
||||
@inbounds for id in 1:lp.ndim
|
||||
mom[X,id] = mom[X,id] + eps1 * (frc1[X,id]+frc2[X,id])
|
||||
U[X,id] = expm(U[X,id],mom[X,id], eps2)
|
||||
mom[b,id,r] = mom[b,id,r] + eps1 * (frc1[b,id,r]+frc2[b,id,r])
|
||||
U[b,id,r] = expm(U[b,id,r], mom[b,id,r], eps2)
|
||||
end
|
||||
|
||||
return nothing
|
||||
end
|
||||
|
||||
function OMF4!(mom, U, eps, ns, lp::SpaceParm, gp::GaugeParm, kp::KernelParm, ymws::YMworkspace)
|
||||
function OMF4!(mom, U, eps, ns, lp::SpaceParm, gp::GaugeParm, ymws::YMworkspace)
|
||||
|
||||
r1::Float64 = 0.08398315262876693
|
||||
r2::Float64 = 0.2539785108410595
|
||||
|
|
@ -86,43 +87,78 @@ function OMF4!(mom, U, eps, ns, lp::SpaceParm, gp::GaugeParm, kp::KernelParm, ym
|
|||
r5::Float64 = 0.5-r1-r3
|
||||
r6::Float64 = 1.0-2.0*(r2+r4)
|
||||
|
||||
# ee = eps*gp.beta/gp.ng
|
||||
# @device_code_warntype force0_wilson!(ymws.frc1,ymws.frc2, U, lp, gp)
|
||||
# force0_wilson_pln!(ymws.frc1, ymws.frc2, U, lp)
|
||||
# zero!(ymws.frc2, lp)
|
||||
# for i in 1:ns
|
||||
# # STEP 1
|
||||
# CUDA.@sync begin
|
||||
# CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_updt!(ymws.mom, ymws.frc1,ymws.frc2, r1*ee, U, eps*r2, lp)
|
||||
# end
|
||||
#
|
||||
# # STEP 2
|
||||
# force0_wilson_pln!(ymws.frc1, ymws.frc2, U, lp)
|
||||
# CUDA.@sync begin
|
||||
# CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_updt!(ymws.mom, ymws.frc1,ymws.frc2, r3*ee, U, eps*r4, lp)
|
||||
# end
|
||||
#
|
||||
# # STEP 3
|
||||
# force0_wilson_pln!(ymws.frc1, ymws.frc2, U, lp)
|
||||
# CUDA.@sync begin
|
||||
# CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_updt!(ymws.mom, ymws.frc1,ymws.frc2, r5*ee, U, eps*r6, lp)
|
||||
# end
|
||||
#
|
||||
# # STEP 4
|
||||
# force0_wilson_pln!(ymws.frc1, ymws.frc2, U, lp)
|
||||
# CUDA.@sync begin
|
||||
# CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_updt!(ymws.mom, ymws.frc1,ymws.frc2, r5*ee, U, eps*r4, lp)
|
||||
# end
|
||||
#
|
||||
# # STEP 5
|
||||
# force0_wilson_pln!(ymws.frc1, ymws.frc2, U, lp)
|
||||
# CUDA.@sync begin
|
||||
# CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_updt!(ymws.mom, ymws.frc1,ymws.frc2, r3*ee, U, eps*r2, lp)
|
||||
# end
|
||||
#
|
||||
# # STEP 6
|
||||
# force0_wilson_pln!(ymws.frc1, ymws.frc2, U, lp)
|
||||
# CUDA.@sync begin
|
||||
# CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_updt!(ymws.mom, ymws.frc1,ymws.frc2, r1*ee, U, 0.0, lp)
|
||||
# end
|
||||
# end
|
||||
|
||||
ee = eps*gp.beta/gp.ng
|
||||
force0_wilson!(ymws.frc1,ymws.frc2, U, lp, gp, kp)
|
||||
force0_wilson!(ymws.frc1, U, lp)
|
||||
zero!(ymws.frc2, lp)
|
||||
for i in 1:ns
|
||||
# STEP 1
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=kp.threads blocks=kp.blocks krnl_updt!(ymws.mom, ymws.frc1,ymws.frc2, r1*ee, U, eps*r2, lp)
|
||||
end
|
||||
|
||||
mom .= mom .+ (r1*ee) .* (ymws.frc1 .+ ymws.frc2)
|
||||
U .= expm.(U, mom, eps*r2)
|
||||
|
||||
# STEP 2
|
||||
force0_wilson!(ymws.frc1,ymws.frc2, U, lp, gp, kp)
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=kp.threads blocks=kp.blocks krnl_updt!(ymws.mom, ymws.frc1,ymws.frc2, r3*ee, U, eps*r4, lp)
|
||||
end
|
||||
force0_wilson!(ymws.frc1, U, lp)
|
||||
mom .= mom .+ (r3*ee) .* (ymws.frc1 .+ ymws.frc2)
|
||||
U .= expm.(U, mom, eps*r4)
|
||||
|
||||
# STEP 3
|
||||
force0_wilson!(ymws.frc1,ymws.frc2, U, lp, gp, kp)
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=kp.threads blocks=kp.blocks krnl_updt!(ymws.mom, ymws.frc1,ymws.frc2, r5*ee, U, eps*r6, lp)
|
||||
end
|
||||
force0_wilson!(ymws.frc1, U, lp)
|
||||
mom .= mom .+ (r5*ee) .* (ymws.frc1 .+ ymws.frc2)
|
||||
U .= expm.(U, mom, eps*r6)
|
||||
|
||||
# STEP 4
|
||||
force0_wilson!(ymws.frc1,ymws.frc2, U, lp, gp, kp)
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=kp.threads blocks=kp.blocks krnl_updt!(ymws.mom, ymws.frc1,ymws.frc2, r5*ee, U, eps*r4, lp)
|
||||
end
|
||||
force0_wilson!(ymws.frc1, U, lp)
|
||||
mom .= mom .+ (r5*ee) .* (ymws.frc1 .+ ymws.frc2)
|
||||
U .= expm.(U, mom, eps*r4)
|
||||
|
||||
# STEP 5
|
||||
force0_wilson!(ymws.frc1,ymws.frc2, U, lp, gp, kp)
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=kp.threads blocks=kp.blocks krnl_updt!(ymws.mom, ymws.frc1,ymws.frc2, r3*ee, U, eps*r2, lp)
|
||||
end
|
||||
force0_wilson!(ymws.frc1, U, lp)
|
||||
mom .= mom .+ (r3*ee) .* (ymws.frc1 .+ ymws.frc2)
|
||||
U .= expm.(U, mom, eps*r2)
|
||||
|
||||
# STEP 6
|
||||
force0_wilson!(ymws.frc1,ymws.frc2, U, lp, gp, kp)
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=kp.threads blocks=kp.blocks krnl_updt!(ymws.mom, ymws.frc1,ymws.frc2, r1*ee, U, 0.0, lp)
|
||||
end
|
||||
force0_wilson!(ymws.frc1, U, lp)
|
||||
mom .= mom .+ (r1*ee) .* (ymws.frc1 .+ ymws.frc2)
|
||||
end
|
||||
|
||||
return nothing
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue