mirror of
https://igit.ific.uv.es/alramos/latticegpu.jl.git
synced 2025-05-14 19:23:42 +02:00
Added support for improved actions
This commit is contained in:
Alberto Ramos
parent
cdad0f2be7
commit
3d464a0325
4 changed files with 128 additions and 112 deletions
17
src/YM/YM.jl
17
src/YM/YM.jl
|
|
@ -20,6 +20,7 @@ import Base.show
|
||||||
|
|
||||||
struct GaugeParm{T}
|
struct GaugeParm{T}
|
||||||
beta::T
|
beta::T
|
||||||
|
c0::T
|
||||||
cG::NTuple{2,T}
|
cG::NTuple{2,T}
|
||||||
ng::Int64
|
ng::Int64
|
||||||
end
|
end
|
||||||
|
|
@ -27,6 +28,7 @@ export GaugeParm
|
||||||
function Base.show(io::IO, gp::GaugeParm)
|
function Base.show(io::IO, gp::GaugeParm)
|
||||||
|
|
||||||
println(io, "beta: ", gp.beta)
|
println(io, "beta: ", gp.beta)
|
||||||
|
println(io, "c0: ", gp.c0)
|
||||||
println(io, "Ngauge: ", gp.ng)
|
println(io, "Ngauge: ", gp.ng)
|
||||||
|
|
||||||
return nothing
|
return nothing
|
||||||
|
|
@ -38,23 +40,17 @@ struct YMworkspace{T}
|
||||||
PRC
|
PRC
|
||||||
frc1
|
frc1
|
||||||
frc2
|
frc2
|
||||||
fpln
|
|
||||||
mom
|
mom
|
||||||
U1
|
U1
|
||||||
cm # complex of volume
|
cm # complex of volume
|
||||||
rm # float of volume
|
rm # float of volume
|
||||||
function YMworkspace(::Type{G}, ::Type{T}, lp::SpaceParm; save_mem = false) where {G <: Group, T <: AbstractFloat}
|
function YMworkspace(::Type{G}, ::Type{T}, lp::SpaceParm) where {G <: Group, T <: AbstractFloat}
|
||||||
|
|
||||||
if (G == SU2)
|
if (G == SU2)
|
||||||
GRP = SU2
|
GRP = SU2
|
||||||
ALG = SU2alg
|
ALG = SU2alg
|
||||||
f1 = field(SU2alg{T}, lp)
|
f1 = field(SU2alg{T}, lp)
|
||||||
f2 = field(SU2alg{T}, lp)
|
f2 = field(SU2alg{T}, lp)
|
||||||
if save_mem
|
|
||||||
fp = nothing
|
|
||||||
else
|
|
||||||
fp = field_pln(SU2alg{T}, lp)
|
|
||||||
end
|
|
||||||
mm = field(SU2alg{T}, lp)
|
mm = field(SU2alg{T}, lp)
|
||||||
u1 = field(SU2{T}, lp)
|
u1 = field(SU2{T}, lp)
|
||||||
end
|
end
|
||||||
|
|
@ -64,18 +60,13 @@ struct YMworkspace{T}
|
||||||
ALG = SU3alg
|
ALG = SU3alg
|
||||||
f1 = field(SU3alg{T}, lp)
|
f1 = field(SU3alg{T}, lp)
|
||||||
f2 = field(SU3alg{T}, lp)
|
f2 = field(SU3alg{T}, lp)
|
||||||
if save_mem
|
|
||||||
fp = nothing
|
|
||||||
else
|
|
||||||
fp = field_pln(SU2alg{T}, lp)
|
|
||||||
end
|
|
||||||
mm = field(SU3alg{T}, lp)
|
mm = field(SU3alg{T}, lp)
|
||||||
u1 = field(SU3{T}, lp)
|
u1 = field(SU3{T}, lp)
|
||||||
end
|
end
|
||||||
cs = CuArray{Complex{T}, 2}(undef, lp.bsz,lp.rsz)
|
cs = CuArray{Complex{T}, 2}(undef, lp.bsz,lp.rsz)
|
||||||
rs = CuArray{T, 2}(undef, lp.bsz,lp.rsz)
|
rs = CuArray{T, 2}(undef, lp.bsz,lp.rsz)
|
||||||
|
|
||||||
return new{T}(GRP,ALG,T,f1, f2, fp, mm, u1, cs, rs)
|
return new{T}(GRP,ALG,T,f1, f2, mm, u1, cs, rs)
|
||||||
end
|
end
|
||||||
end
|
end
|
||||||
export YMworkspace
|
export YMworkspace
|
||||||
|
|
|
||||||
189
src/YM/YMact.jl
189
src/YM/YMact.jl
|
|
@ -9,6 +9,88 @@
|
||||||
### created: Mon Jul 12 18:31:19 2021
|
### created: Mon Jul 12 18:31:19 2021
|
||||||
###
|
###
|
||||||
|
|
||||||
|
function krnl_impr!(plx, U::AbstractArray{T}, c0, c1, lp::SpaceParm{N,M,D}) where {T,N,M,D}
|
||||||
|
|
||||||
|
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
|
||||||
|
|
||||||
|
Ush = @cuStaticSharedMem(T, (D,2))
|
||||||
|
|
||||||
|
plx[b,r] = zero(plx[b,r])
|
||||||
|
for id1 in 1:N-1
|
||||||
|
bu1, ru1 = up((b, r), id1, lp)
|
||||||
|
Ush[b,1] = U[b,id1,r]
|
||||||
|
|
||||||
|
for id2 = id1+1:N
|
||||||
|
bu2, ru2 = up((b, r), id2, lp)
|
||||||
|
Ush[b,2] = U[b,id2,r]
|
||||||
|
sync_threads()
|
||||||
|
|
||||||
|
# H2 staple
|
||||||
|
(b1, r1) = up((b,r), id1, lp)
|
||||||
|
if r1 == r
|
||||||
|
ga = Ush[b1,1]
|
||||||
|
else
|
||||||
|
ga = U[b1,id1,r1]
|
||||||
|
end
|
||||||
|
|
||||||
|
(b2, r2) = up((b1,r1), id1, lp)
|
||||||
|
if r2 == r
|
||||||
|
gb = Ush[b2,2]
|
||||||
|
else
|
||||||
|
gb = U[b2,id2,r2]
|
||||||
|
end
|
||||||
|
|
||||||
|
(b2, r2) = up((b1,r1), id2, lp)
|
||||||
|
if r2 == r
|
||||||
|
gc = Ush[b2,1]
|
||||||
|
else
|
||||||
|
gc = U[b2,id1,r2]
|
||||||
|
end
|
||||||
|
h2 = (ga*gb)/gc
|
||||||
|
|
||||||
|
# H3 staple
|
||||||
|
(b1, r1) = up((b,r), id2, lp)
|
||||||
|
if r1 == r
|
||||||
|
ga = Ush[b1,2]
|
||||||
|
else
|
||||||
|
ga = U[b1,id2,r1]
|
||||||
|
end
|
||||||
|
|
||||||
|
(b2, r2) = up((b1,r1), id2, lp)
|
||||||
|
if r2 == r
|
||||||
|
gb = Ush[b2,1]
|
||||||
|
else
|
||||||
|
gb = U[b2,id1,r2]
|
||||||
|
end
|
||||||
|
|
||||||
|
(b2, r2) = up((b1,r1), id1, lp)
|
||||||
|
if r2 == r
|
||||||
|
gc = Ush[b2,2]
|
||||||
|
else
|
||||||
|
gc = U[b2,id2,r2]
|
||||||
|
end
|
||||||
|
h3 = (ga*gb)/gc
|
||||||
|
|
||||||
|
if ru2 == r
|
||||||
|
gb = Ush[bu2,1]
|
||||||
|
else
|
||||||
|
gb = U[bu2,id1,ru2]
|
||||||
|
end
|
||||||
|
if ru1 == r
|
||||||
|
ga = Ush[bu1,2]
|
||||||
|
else
|
||||||
|
ga = U[bu1,id2,ru1]
|
||||||
|
end
|
||||||
|
|
||||||
|
g2 = Ush[b,2]\Ush[b,1]
|
||||||
|
|
||||||
|
plx[b,r] += c0*tr(g2*ga/gb) + c1*( tr(g2*h2/gb) + tr(g2*ga/h3))
|
||||||
|
end
|
||||||
|
end
|
||||||
|
|
||||||
|
return nothing
|
||||||
|
end
|
||||||
|
|
||||||
function krnl_plaq!(plx, U::AbstractArray{T}, lp::SpaceParm{N,M,D}) where {T,N,M,D}
|
function krnl_plaq!(plx, U::AbstractArray{T}, lp::SpaceParm{N,M,D}) where {T,N,M,D}
|
||||||
|
|
||||||
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
|
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
|
||||||
|
|
@ -121,14 +203,14 @@ function krnl_force_impr_pln!(frc1, frc2, U::AbstractArray{T}, c0, c1, ipl, lp::
|
||||||
if r1 == r
|
if r1 == r
|
||||||
ga = Ush[b1,1]
|
ga = Ush[b1,1]
|
||||||
else
|
else
|
||||||
ga = U[b1,id2,r1]
|
ga = U[b1,id1,r1]
|
||||||
end
|
end
|
||||||
|
|
||||||
(b2, r2) = up((b1,r1), id1, lp)
|
(b2, r2) = up((b1,r1), id1, lp)
|
||||||
if r2 == r
|
if r2 == r
|
||||||
gc = Ush[b2,2]
|
gb = Ush[b2,2]
|
||||||
else
|
else
|
||||||
gc = U[b2,id2,r2]
|
gb = U[b2,id2,r2]
|
||||||
end
|
end
|
||||||
|
|
||||||
(b2, r2) = up((b1,r1), id2, lp)
|
(b2, r2) = up((b1,r1), id2, lp)
|
||||||
|
|
@ -149,9 +231,9 @@ function krnl_force_impr_pln!(frc1, frc2, U::AbstractArray{T}, c0, c1, ipl, lp::
|
||||||
|
|
||||||
(b2, r2) = up((b1,r1), id2, lp)
|
(b2, r2) = up((b1,r1), id2, lp)
|
||||||
if r2 == r
|
if r2 == r
|
||||||
gc = Ush[b2,1]
|
gb = Ush[b2,1]
|
||||||
else
|
else
|
||||||
gc = U[b2,id1,r2]
|
gb = U[b2,id1,r2]
|
||||||
end
|
end
|
||||||
|
|
||||||
(b2, r2) = up((b1,r1), id1, lp)
|
(b2, r2) = up((b1,r1), id1, lp)
|
||||||
|
|
@ -160,7 +242,7 @@ function krnl_force_impr_pln!(frc1, frc2, U::AbstractArray{T}, c0, c1, ipl, lp::
|
||||||
else
|
else
|
||||||
gc = U[b2,id2,r2]
|
gc = U[b2,id2,r2]
|
||||||
end
|
end
|
||||||
h3 = (ga\gb)/gc
|
h3 = (ga*gb)/gc
|
||||||
|
|
||||||
# H4 staple
|
# H4 staple
|
||||||
(b1, r1) = dw((b,r), id1, lp)
|
(b1, r1) = dw((b,r), id1, lp)
|
||||||
|
|
@ -207,67 +289,7 @@ function krnl_force_impr_pln!(frc1, frc2, U::AbstractArray{T}, c0, c1, ipl, lp::
|
||||||
|
|
||||||
frc2[bu2,id1,ru2] += c0*projalg(g2*g1) + c1*( projalg((Ush[b,2]\h1)*g1) +
|
frc2[bu2,id1,ru2] += c0*projalg(g2*g1) + c1*( projalg((Ush[b,2]\h1)*g1) +
|
||||||
projalg(g2*h2/gb) +
|
projalg(g2*h2/gb) +
|
||||||
projalg(h4*Ush[b,1]*g1) )
|
projalg(h4\Ush[b,1]*g1) )
|
||||||
end
|
|
||||||
|
|
||||||
return nothing
|
|
||||||
end
|
|
||||||
|
|
||||||
function krnl_force_wilson_nw!(fpl, U::AbstractArray{T}, lp::SpaceParm{N,M,D}) where {T,N,M,D}
|
|
||||||
|
|
||||||
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
|
|
||||||
|
|
||||||
Ush = @cuStaticSharedMem(T, (D,N))
|
|
||||||
|
|
||||||
for id in 1:N
|
|
||||||
@inbounds Ush[b,id] = U[b,id,r]
|
|
||||||
end
|
|
||||||
sync_threads()
|
|
||||||
|
|
||||||
@inbounds for ipl in 1:lp.npls
|
|
||||||
id1, id2 = lp.plidx[ipl]
|
|
||||||
|
|
||||||
bu1, ru1 = up((b, r), id1, lp)
|
|
||||||
bu2, ru2 = up((b, r), id2, lp)
|
|
||||||
|
|
||||||
if ru2 == r
|
|
||||||
gt2 = Ush[bu2,id1]
|
|
||||||
else
|
|
||||||
gt2 = U[bu2,id1,ru2]
|
|
||||||
end
|
|
||||||
if ru1 == r
|
|
||||||
gt1 = Ush[bu1,id2]
|
|
||||||
else
|
|
||||||
gt1 = U[bu1,id2,ru1]
|
|
||||||
end
|
|
||||||
|
|
||||||
g1 = gt1/gt2
|
|
||||||
g2 = Ush[b,id2]\Ush[b,id1]
|
|
||||||
|
|
||||||
fpl[b ,ipl, r ,1] = projalg(Ush[b,id1]*g1/Ush[b,id2])
|
|
||||||
fpl[bu1,ipl, ru1,2] = projalg(g1*g2)
|
|
||||||
fpl[bu2,ipl, ru2,3] = projalg(g2*g1)
|
|
||||||
|
|
||||||
end
|
|
||||||
|
|
||||||
return nothing
|
|
||||||
end
|
|
||||||
|
|
||||||
function krnl_add_force_plns!(frc::AbstractArray{T}, fpl, lp::SpaceParm{N,M,D}) where {T,N,M,D}
|
|
||||||
|
|
||||||
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
|
|
||||||
|
|
||||||
Fsh = @cuStaticSharedMem(T, (D,M))
|
|
||||||
@inbounds for j in 1:M
|
|
||||||
Fsh[b,j] = fpl[b,j,r,1]
|
|
||||||
end
|
|
||||||
sync_threads()
|
|
||||||
|
|
||||||
@inbounds for ipl in 1:lp.npls
|
|
||||||
id1, id2 = lp.plidx[ipl]
|
|
||||||
|
|
||||||
frc[b,id1,r] += -Fsh[b,ipl] + fpl[b,ipl,r,3]
|
|
||||||
frc[b,id2,r] += Fsh[b,ipl] - fpl[b,ipl,r,2]
|
|
||||||
end
|
end
|
||||||
|
|
||||||
return nothing
|
return nothing
|
||||||
|
|
@ -279,25 +301,34 @@ end
|
||||||
Computes the force deriving from the Wilson plaquette action, without
|
Computes the force deriving from the Wilson plaquette action, without
|
||||||
the prefactor 1/g0^2, and assign it to the workspace force `ymws.frc1`
|
the prefactor 1/g0^2, and assign it to the workspace force `ymws.frc1`
|
||||||
"""
|
"""
|
||||||
function force_wilson(ymws::YMworkspace, U, lp::SpaceParm)
|
function force_gauge(ymws::YMworkspace, U, c0, lp::SpaceParm)
|
||||||
|
|
||||||
if ymws.fpln == nothing
|
if abs(c0-1) < 1.0E-10
|
||||||
force_wilson_pln!(ymws.frc1, ymws.frc2, U, lp::SpaceParm)
|
force_wilson_pln!(ymws.frc1, ymws.frc2, U, lp::SpaceParm)
|
||||||
else
|
else
|
||||||
force_wilson_nw!(ymws.frc1, ymws.fpln, U, lp::SpaceParm)
|
force_wilson_pln!(ymws.frc1, ymws.frc2, U, lp::SpaceParm, c0)
|
||||||
end
|
end
|
||||||
return nothing
|
return nothing
|
||||||
end
|
end
|
||||||
|
|
||||||
|
force_wilson(ymws::YMworkspace, U, lp::SpaceParm) = force_gauge(ymws, U, 1, lp)
|
||||||
|
|
||||||
function force_wilson_pln!(frc1, ftmp, U, lp::SpaceParm)
|
|
||||||
|
function force_wilson_pln!(frc1, ftmp, U, lp::SpaceParm, c0=1)
|
||||||
|
|
||||||
fill!(frc1, zero(eltype(frc1)))
|
fill!(frc1, zero(eltype(frc1)))
|
||||||
fill!(ftmp, zero(eltype(ftmp)))
|
fill!(ftmp, zero(eltype(ftmp)))
|
||||||
|
if c0 == 1
|
||||||
for i in 1:lp.npls
|
for i in 1:lp.npls
|
||||||
CUDA.@sync begin
|
CUDA.@sync begin
|
||||||
# CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_force_wilson_pln!(frc1,ftmp,U,i,lp)
|
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_force_wilson_pln!(frc1,ftmp,U,i,lp)
|
||||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_force_impr_pln!(frc1,ftmp,U,1.0,0.0,i,lp)
|
end
|
||||||
|
end
|
||||||
|
else
|
||||||
|
for i in 1:lp.npls
|
||||||
|
CUDA.@sync begin
|
||||||
|
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_force_impr_pln!(frc1,ftmp,U,c0,(1-c0)/8,i,lp)
|
||||||
|
end
|
||||||
end
|
end
|
||||||
end
|
end
|
||||||
frc1 .= frc1 .+ ftmp
|
frc1 .= frc1 .+ ftmp
|
||||||
|
|
@ -305,15 +336,3 @@ function force_wilson_pln!(frc1, ftmp, U, lp::SpaceParm)
|
||||||
return nothing
|
return nothing
|
||||||
end
|
end
|
||||||
|
|
||||||
function force_wilson_nw!(frc1, fpln, U, lp::SpaceParm)
|
|
||||||
|
|
||||||
CUDA.@sync begin
|
|
||||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_force_wilson_nw!(fpln,U,lp)
|
|
||||||
end
|
|
||||||
fill!(frc1, zero(eltype(frc1)))
|
|
||||||
CUDA.@sync begin
|
|
||||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_add_force_plns!(frc1, fpln,lp)
|
|
||||||
end
|
|
||||||
|
|
||||||
return nothing
|
|
||||||
end
|
|
||||||
|
|
|
||||||
|
|
@ -17,10 +17,16 @@ Returns the value of the Wilson plaquette action for the configuration U.
|
||||||
"""
|
"""
|
||||||
function gauge_action(U, lp::SpaceParm, gp::GaugeParm{T}, ymws::YMworkspace{T}) where T <: AbstractFloat
|
function gauge_action(U, lp::SpaceParm, gp::GaugeParm{T}, ymws::YMworkspace{T}) where T <: AbstractFloat
|
||||||
|
|
||||||
|
if abs(gp.c0-1) < 1.0E-10
|
||||||
CUDA.@sync begin
|
CUDA.@sync begin
|
||||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_plaq!(ymws.cm, U, lp)
|
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_plaq!(ymws.cm, U, lp)
|
||||||
end
|
end
|
||||||
S = gp.beta*( prod(lp.iL)*lp.npls -
|
else
|
||||||
|
CUDA.@sync begin
|
||||||
|
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_impr!(ymws.cm, U, gp.c0, (1-gp.c0)/8, lp)
|
||||||
|
end
|
||||||
|
end
|
||||||
|
S = gp.beta*( prod(lp.iL)*lp.npls*(gp.c0 + (1-gp.c0)/8) -
|
||||||
CUDA.mapreduce(real, +, ymws.cm)/gp.ng )
|
CUDA.mapreduce(real, +, ymws.cm)/gp.ng )
|
||||||
|
|
||||||
return S
|
return S
|
||||||
|
|
@ -80,28 +86,28 @@ function OMF4!(mom, U, eps, ns, lp::SpaceParm, gp::GaugeParm{T}, ymws::YMworkspa
|
||||||
r6::T = 1.0-2.0*(r2+r4)
|
r6::T = 1.0-2.0*(r2+r4)
|
||||||
|
|
||||||
ee = eps*gp.beta/gp.ng
|
ee = eps*gp.beta/gp.ng
|
||||||
force_wilson(ymws, U, lp)
|
force_gauge(ymws, U, gp.c0, lp)
|
||||||
for i in 1:ns
|
for i in 1:ns
|
||||||
mom .= mom .+ (r1*ee) .* ymws.frc1
|
mom .= mom .+ (r1*ee) .* ymws.frc1
|
||||||
U .= expm.(U, mom, eps*r2)
|
U .= expm.(U, mom, eps*r2)
|
||||||
|
|
||||||
force_wilson(ymws, U, lp)
|
force_gauge(ymws, U, gp.c0, lp)
|
||||||
mom .= mom .+ (r3*ee) .* ymws.frc1
|
mom .= mom .+ (r3*ee) .* ymws.frc1
|
||||||
U .= expm.(U, mom, eps*r4)
|
U .= expm.(U, mom, eps*r4)
|
||||||
|
|
||||||
force_wilson(ymws, U, lp)
|
force_gauge(ymws, U, gp.c0, lp)
|
||||||
mom .= mom .+ (r5*ee) .* ymws.frc1
|
mom .= mom .+ (r5*ee) .* ymws.frc1
|
||||||
U .= expm.(U, mom, eps*r6)
|
U .= expm.(U, mom, eps*r6)
|
||||||
|
|
||||||
force_wilson(ymws, U, lp)
|
force_gauge(ymws, U, gp.c0, lp)
|
||||||
mom .= mom .+ (r5*ee) .* ymws.frc1
|
mom .= mom .+ (r5*ee) .* ymws.frc1
|
||||||
U .= expm.(U, mom, eps*r4)
|
U .= expm.(U, mom, eps*r4)
|
||||||
|
|
||||||
force_wilson(ymws, U, lp)
|
force_gauge(ymws, U, gp.c0, lp)
|
||||||
mom .= mom .+ (r3*ee) .* ymws.frc1
|
mom .= mom .+ (r3*ee) .* ymws.frc1
|
||||||
U .= expm.(U, mom, eps*r2)
|
U .= expm.(U, mom, eps*r2)
|
||||||
|
|
||||||
force_wilson(ymws, U, lp)
|
force_gauge(ymws, U, gp.c0, lp)
|
||||||
mom .= mom .+ (r1*ee) .* ymws.frc1
|
mom .= mom .+ (r1*ee) .* ymws.frc1
|
||||||
end
|
end
|
||||||
|
|
||||||
|
|
|
||||||
|
|
@ -13,7 +13,7 @@ GRP = SU3
|
||||||
ALG = SU3alg
|
ALG = SU3alg
|
||||||
PREC = Float64
|
PREC = Float64
|
||||||
lp = SpaceParm{4}((32,32,32,32), (4,4,4,4))
|
lp = SpaceParm{4}((32,32,32,32), (4,4,4,4))
|
||||||
gp = GaugeParm{PREC}(6.0, (0.0,0.0), 3)
|
gp = GaugeParm{PREC}(6.0, 5.0/3.0, (0.0,0.0), 3)
|
||||||
|
|
||||||
println("Space Parameters: ", lp)
|
println("Space Parameters: ", lp)
|
||||||
println("Gauge Parameters: ", gp)
|
println("Gauge Parameters: ", gp)
|
||||||
|
|
@ -33,7 +33,7 @@ println("Take to take the configuration to memory: ")
|
||||||
|
|
||||||
|
|
||||||
println("Allocating YM workspace")
|
println("Allocating YM workspace")
|
||||||
ymws = YMworkspace(GRP, PREC, lp, save_mem = true)
|
ymws = YMworkspace(GRP, PREC, lp)
|
||||||
|
|
||||||
#CUDA.@sync begin
|
#CUDA.@sync begin
|
||||||
# @device_code_warntype CUDA.@cuda threads=lp.bsz blocks=lp.rsz LatticeGPU.krnl_plaq!(ymws.cm, U, lp)
|
# @device_code_warntype CUDA.@cuda threads=lp.bsz blocks=lp.rsz LatticeGPU.krnl_plaq!(ymws.cm, U, lp)
|
||||||
|
|
|
||||||
Loading…
Add table
Add a link
Reference in a new issue