mirror of
https://igit.ific.uv.es/alramos/latticegpu.jl.git
synced 2025-05-15 03:33:42 +02:00
Full support for twisted bc
This commit is contained in:
Alberto Ramos
parent
26b5dfab01
commit
410a6d7b25
9 changed files with 214 additions and 100 deletions
24
src/YM/YM.jl
24
src/YM/YM.jl
|
|
@ -104,15 +104,29 @@ function Base.show(io::IO, ymws::YMworkspace)
|
|||
println(io, "Workspace for Group: ", ymws.GRP)
|
||||
println(io, " Algebra: ", ymws.ALG)
|
||||
println(io, "Precision: ", ymws.PRC)
|
||||
if ymws.fpln == nothing
|
||||
println(io, " - Running in memory efficient mode")
|
||||
else
|
||||
println(io, " - Running in computing efficient mode")
|
||||
end
|
||||
|
||||
return nothing
|
||||
end
|
||||
|
||||
|
||||
function ztwist(gp::GaugeParm{T,G}, lp::SpaceParm{N,M,B,D}) where {T,G,N,M,B,D}
|
||||
|
||||
function plnf(ipl)
|
||||
id1, id2 = lp.plidx[ipl]
|
||||
return convert(Complex{T},exp(2im * pi * lp.ntw[ipl]/(lp.iL[id1]*lp.iL[id2]*gp.ng)))
|
||||
end
|
||||
|
||||
return ntuple(i->plnf(i), M)
|
||||
end
|
||||
|
||||
function ztwist(gp::GaugeParm{T,G}, lp::SpaceParm{N,M,B,D}, ipl::Int) where {T,G,N,M,B,D}
|
||||
|
||||
id1, id2 = lp.plidx[ipl]
|
||||
return convert(Complex{T},exp(2im * pi * lp.ntw[ipl]/(lp.iL[id1]*lp.iL[id2]*gp.ng)))
|
||||
end
|
||||
export ztwist
|
||||
|
||||
|
||||
include("YMfields.jl")
|
||||
export randomize!, zero!, norm2
|
||||
|
||||
|
|
|
|||
|
|
@ -9,13 +9,14 @@
|
|||
### created: Mon Jul 12 18:31:19 2021
|
||||
###
|
||||
|
||||
function krnl_impr!(plx, U::AbstractArray{T}, c0, c1, Ubnd::T, cG, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
|
||||
function krnl_impr!(plx, U::AbstractArray{T}, c0, c1, Ubnd::T, cG, ztw, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
|
||||
|
||||
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
|
||||
it = point_time((b, r), lp)
|
||||
|
||||
Ush = @cuStaticSharedMem(T, (D,2))
|
||||
|
||||
|
||||
ipl = 0
|
||||
S = zero(eltype(plx))
|
||||
for id1 in N:-1:1
|
||||
bu1, ru1 = up((b, r), id1, lp)
|
||||
|
|
@ -26,7 +27,8 @@ function krnl_impr!(plx, U::AbstractArray{T}, c0, c1, Ubnd::T, cG, lp::SpaceParm
|
|||
bu2, ru2 = up((b, r), id2, lp)
|
||||
Ush[b,2] = U[b,id2,r]
|
||||
sync_threads()
|
||||
|
||||
ipl = ipl + 1
|
||||
|
||||
# H2 staple
|
||||
(b1, r1) = up((b,r), id1, lp)
|
||||
if r1 == r
|
||||
|
|
@ -100,7 +102,8 @@ function krnl_impr!(plx, U::AbstractArray{T}, c0, c1, Ubnd::T, cG, lp::SpaceParm
|
|||
elseif (it == 1) && SFBC
|
||||
S += cG*(c0*tr(g2*ga/gb) + (3*c1/2)*tr(g2*ga/h3)) + c1*tr(g2*h2/gb)
|
||||
else
|
||||
S += c0*tr(g2*ga/gb) + c1*( tr(g2*h2/gb) + tr(g2*ga/h3))
|
||||
S += ztw[ipl]*c0*tr(g2*ga/gb) +
|
||||
(ztw[ipl]^2*c1)*( tr(g2*h2/gb) + tr(g2*ga/h3))
|
||||
end
|
||||
|
||||
end
|
||||
|
|
@ -112,14 +115,14 @@ function krnl_impr!(plx, U::AbstractArray{T}, c0, c1, Ubnd::T, cG, lp::SpaceParm
|
|||
return nothing
|
||||
end
|
||||
|
||||
function krnl_plaq!(plx, U::AbstractArray{T}, Ubnd::T, cG, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
|
||||
function krnl_plaq!(plx, U::AbstractArray{T}, Ubnd::T, cG, ztw, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
|
||||
|
||||
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
|
||||
it = point_time((b, r), lp)
|
||||
|
||||
Ush = @cuStaticSharedMem(T, (D,2))
|
||||
|
||||
S = zero(eltype(plx))
|
||||
ipl = 0
|
||||
for id1 in N:-1:1
|
||||
bu1, ru1 = up((b, r), id1, lp)
|
||||
Ush[b,1] = U[b,id1,r]
|
||||
|
|
@ -131,7 +134,8 @@ function krnl_plaq!(plx, U::AbstractArray{T}, Ubnd::T, cG, lp::SpaceParm{N,M,B,D
|
|||
bu2, ru2 = up((b, r), id2, lp)
|
||||
Ush[b,2] = U[b,id2,r]
|
||||
sync_threads()
|
||||
|
||||
ipl = ipl + 1
|
||||
|
||||
if ru1 == r
|
||||
gt1 = Ush[bu1,2]
|
||||
else
|
||||
|
|
@ -150,7 +154,7 @@ function krnl_plaq!(plx, U::AbstractArray{T}, Ubnd::T, cG, lp::SpaceParm{N,M,B,D
|
|||
if SFBND
|
||||
S += cG*tr(Ush[b,1]*gt1 / (Ush[b,2]*gt2))
|
||||
else
|
||||
S += tr(Ush[b,1]*gt1 / (Ush[b,2]*gt2))
|
||||
S += ztw[ipl]*tr(Ush[b,1]*gt1 / (Ush[b,2]*gt2))
|
||||
end
|
||||
end
|
||||
end
|
||||
|
|
@ -161,10 +165,10 @@ function krnl_plaq!(plx, U::AbstractArray{T}, Ubnd::T, cG, lp::SpaceParm{N,M,B,D
|
|||
return nothing
|
||||
end
|
||||
|
||||
function krnl_force_wilson_pln!(frc1, frc2, U::AbstractArray{T}, Ubnd::T, cG, ipl, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
|
||||
function krnl_force_wilson_pln!(frc1, frc2, U::AbstractArray{T}, Ubnd::T, cG, ztw, ipl, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
|
||||
|
||||
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
|
||||
it = point_time((b, r), lp)
|
||||
it = point_time((b,r), lp)
|
||||
|
||||
Ush = @cuStaticSharedMem(T, (D,2))
|
||||
|
||||
|
|
@ -198,32 +202,31 @@ function krnl_force_wilson_pln!(frc1, frc2, U::AbstractArray{T}, Ubnd::T, cG, ip
|
|||
g2 = Ush[b,2]\Ush[b,1]
|
||||
|
||||
if SFBC && (it == 1)
|
||||
X = cG*projalg(Ush[b,1]*g1/Ush[b,2])
|
||||
X = cG*projalg(ztw,Ush[b,1]*g1/Ush[b,2])
|
||||
|
||||
frc1[b ,id1, r ] -= X
|
||||
frc2[bu1,id2,ru1] -= cG*projalg(g1*g2)
|
||||
frc2[bu2,id1,ru2] += cG*projalg(g2*g1)
|
||||
frc2[bu1,id2,ru1] -= cG*projalg(ztw,g1*g2)
|
||||
frc2[bu2,id1,ru2] += cG*projalg(ztw,g2*g1)
|
||||
elseif SFBC && (it == lp.iL[end])
|
||||
X = cG*projalg(Ush[b,1]*g1/Ush[b,2])
|
||||
X = cG*projalg(ztw,Ush[b,1]*g1/Ush[b,2])
|
||||
|
||||
frc1[b ,id1, r ] -= X
|
||||
frc1[b ,id2, r ] += X
|
||||
frc2[bu2,id1,ru2] += cG*projalg(g2*g1)
|
||||
frc2[bu2,id1,ru2] += cG*projalg(ztw,g2*g1)
|
||||
else
|
||||
X = projalg(Ush[b,1]*g1/Ush[b,2])
|
||||
X = projalg(ztw,Ush[b,1]*g1/Ush[b,2])
|
||||
|
||||
frc1[b ,id1, r ] -= X
|
||||
frc1[b ,id2, r ] += X
|
||||
frc2[bu1,id2,ru1] -= projalg(g1*g2)
|
||||
frc2[bu2,id1,ru2] += projalg(g2*g1)
|
||||
frc2[bu1,id2,ru1] -= projalg(ztw,g1*g2)
|
||||
frc2[bu2,id1,ru2] += projalg(ztw,g2*g1)
|
||||
end
|
||||
|
||||
end
|
||||
|
||||
return nothing
|
||||
end
|
||||
|
||||
function krnl_force_impr_pln!(frc1, frc2, U::AbstractArray{T}, c0, c1, Ubnd::T, cG, ipl, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
|
||||
function krnl_force_impr_pln!(frc1, frc2, U::AbstractArray{T}, c0, c1, Ubnd::T, cG, ztw, ipl, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
|
||||
|
||||
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
|
||||
it = point_time((b, r), lp)
|
||||
|
|
@ -371,19 +374,18 @@ function krnl_force_impr_pln!(frc1, frc2, U::AbstractArray{T}, c0, c1, Ubnd::T,
|
|||
frc2[bu2,id1,ru2] += (cG*c0)*projalg(g2*g1) + (3*c1*cG/2) * projalg((Ush[b,2]\h1)*g1) +
|
||||
c1 * projalg(h4\Ush[b,1]*g1)
|
||||
else
|
||||
X = c0*projalg(Ush[b,1]*g1/Ush[b,2]) + c1 * (projalg(Ush[b,1]*h2/(Ush[b,2]*gb)) +
|
||||
projalg(Ush[b,1]*ga/(Ush[b,2]*h3)))
|
||||
zsq = ztw[ipl]^2
|
||||
X = projalg(c0*ztw[ipl],Ush[b,1]*g1/Ush[b,2]) + projalg(zsq*c1,Ush[b,1]*h2/(Ush[b,2]*gb)) +
|
||||
projalg(zsq*c1,Ush[b,1]*ga/(Ush[b,2]*h3))
|
||||
|
||||
frc1[b,id1,r] -= X + c1*projalg(Ush[b,1]*g1/h4)
|
||||
frc1[b,id2,r] += X + c1*projalg(h1*g1/Ush[b,2])
|
||||
frc1[b,id1,r] -= X + projalg(zsq*c1,Ush[b,1]*g1/h4)
|
||||
frc1[b,id2,r] += X + projalg(zsq*c1,h1*g1/Ush[b,2])
|
||||
|
||||
frc2[bu1,id2,ru1] -= c0*projalg(g1*g2) + c1*( projalg((ga/h3)*g2) +
|
||||
projalg((g1/h4)*Ush[b,1]) +
|
||||
projalg((g1/Ush[b,2])*h1) )
|
||||
frc2[bu1,id2,ru1] -= projalg(c0*ztw[ipl],g1*g2) + projalg(zsq*c1,(ga/h3)*g2) +
|
||||
projalg(zsq*c1,(g1/h4)*Ush[b,1]) + projalg(zsq*c1,(g1/Ush[b,2])*h1)
|
||||
|
||||
frc2[bu2,id1,ru2] += c0*projalg(g2*g1) + c1*( projalg((Ush[b,2]\h1)*g1) +
|
||||
projalg(g2*h2/gb) +
|
||||
projalg(h4\Ush[b,1]*g1) )
|
||||
frc2[bu2,id1,ru2] += projalg(c0*ztw[ipl],g2*g1) + projalg(zsq*c1,(Ush[b,2]\h1)*g1) +
|
||||
projalg(zsq*c1,g2*h2/gb) + projalg(zsq*c1,h4\Ush[b,1]*g1)
|
||||
end
|
||||
|
||||
end
|
||||
|
|
@ -399,13 +401,14 @@ the prefactor 1/g0^2, and assign it to the workspace force `ymws.frc1`
|
|||
"""
|
||||
function force_gauge(ymws::YMworkspace, U, c0, cG, gp::GaugeParm, lp::SpaceParm)
|
||||
|
||||
ztw = ztwist(gp, lp)
|
||||
if abs(c0-1) < 1.0E-10
|
||||
@timeit "Wilson gauge force" begin
|
||||
force_pln!(ymws.frc1, ymws.frc2, U, gp.Ubnd, cG, lp::SpaceParm)
|
||||
force_pln!(ymws.frc1, ymws.frc2, U, gp.Ubnd, cG, ztw, lp::SpaceParm)
|
||||
end
|
||||
else
|
||||
@timeit "Improved gauge force" begin
|
||||
force_pln!(ymws.frc1, ymws.frc2, U, gp.Ubnd, cG, lp::SpaceParm, c0)
|
||||
force_pln!(ymws.frc1, ymws.frc2, U, gp.Ubnd, cG, ztw, lp::SpaceParm, c0)
|
||||
end
|
||||
end
|
||||
return nothing
|
||||
|
|
@ -416,20 +419,20 @@ force_wilson(ymws::YMworkspace, U, gp::GaugeParm, lp::SpaceParm) = force_gauge(y
|
|||
force_wilson(ymws::YMworkspace, U, cG, gp::GaugeParm, lp::SpaceParm) = force_gauge(ymws, U, 1, gp.cG[1], gp, lp)
|
||||
|
||||
|
||||
function force_pln!(frc1, ftmp, U, Ubnd, cG, lp::SpaceParm, c0=1)
|
||||
function force_pln!(frc1, ftmp, U, Ubnd, cG, ztw, lp::SpaceParm, c0=1)
|
||||
|
||||
fill!(frc1, zero(eltype(frc1)))
|
||||
fill!(ftmp, zero(eltype(ftmp)))
|
||||
if c0 == 1
|
||||
for i in 1:lp.npls
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_force_wilson_pln!(frc1,ftmp,U, Ubnd, cG,i,lp)
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_force_wilson_pln!(frc1,ftmp,U, Ubnd, cG, ztw[i], 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,Ubnd, cG, i,lp)
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_force_impr_pln!(frc1,ftmp,U,c0,(1-c0)/8,Ubnd, cG, ztw[i], i,lp)
|
||||
end
|
||||
end
|
||||
end
|
||||
|
|
|
|||
109
src/YM/YMflow.jl
109
src/YM/YMflow.jl
|
|
@ -146,6 +146,9 @@ function Eoft_plaq(Eslc, U, gp::GaugeParm{T}, lp::SpaceParm{N,M,B,D}, ymws::YMwo
|
|||
|
||||
@timeit "E(t) plaquette measurement" begin
|
||||
|
||||
ztw = ztwist(gp, lp)
|
||||
SFBC = ((B == BC_SF_AFWB) || (B == BC_SF_ORBI) )
|
||||
|
||||
tp = ntuple(i->i, N-1)
|
||||
V3 = prod(lp.iL[1:end-1])
|
||||
|
||||
|
|
@ -154,7 +157,7 @@ function Eoft_plaq(Eslc, U, gp::GaugeParm{T}, lp::SpaceParm{N,M,B,D}, ymws::YMwo
|
|||
for ipl in 1:M
|
||||
fill!(Etmp, zero(T))
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_plaq_pln!(ymws.cm, U, ipl, lp)
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_plaq_pln!(ymws.cm, U, gp.Ubnd, ztw[ipl], ipl, lp)
|
||||
end
|
||||
|
||||
Etmp .= (gp.ng .- reshape(Array(CUDA.mapreduce(real, +, ymws.cm;dims=tp)),lp.iL[end])/V3 )
|
||||
|
|
@ -162,7 +165,9 @@ function Eoft_plaq(Eslc, U, gp::GaugeParm{T}, lp::SpaceParm{N,M,B,D}, ymws::YMwo
|
|||
for it in 2:lp.iL[end]
|
||||
Eslc[it,ipl] = Etmp[it] + Etmp[it-1]
|
||||
end
|
||||
Eslc[1,ipl] = Etmp[1] + Etmp[end]
|
||||
if !SFBC
|
||||
Eslc[1,ipl] = Etmp[1] + Etmp[end]
|
||||
end
|
||||
else
|
||||
for it in 1:lp.iL[end]
|
||||
Eslc[it,ipl] = 2*Etmp[it]
|
||||
|
|
@ -179,17 +184,24 @@ end
|
|||
Eoft_plaq(U, gp::GaugeParm{T}, lp::SpaceParm{N,M,B,D}, ymws::YMworkspace) where {T,N,M,B,D} = Eoft_plaq(zeros(T,lp.iL[end],M), U, gp, lp, ymws)
|
||||
|
||||
|
||||
function krnl_plaq_pln!(plx, U::AbstractArray{T}, ipl, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
|
||||
function krnl_plaq_pln!(plx, U::AbstractArray{T}, Ubnd::T, ztw, ipl, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
|
||||
|
||||
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
|
||||
|
||||
id1, id2 = lp.plidx[ipl]
|
||||
SFBC = ((B == BC_SF_AFWB) || (B == BC_SF_ORBI)) && (id1 == lp.iL[end])
|
||||
|
||||
bu1, ru1 = up((b, r), id1, lp)
|
||||
bu2, ru2 = up((b, r), id2, lp)
|
||||
|
||||
|
||||
if SFBC && (ru1 != r)
|
||||
gt = Ubnd
|
||||
else
|
||||
gt = U[bu1,id2,ru1]
|
||||
end
|
||||
|
||||
I = point_coord((b,r), lp)
|
||||
plx[I] = tr(U[b,1,r]*U[bu1,id2,ru1] / (U[b,2,r]*U[bu2,id1,ru2]))
|
||||
plx[I] = ztw*tr(U[b,1,r]*gt / (U[b,2,r]*U[bu2,id1,ru2]))
|
||||
|
||||
return nothing
|
||||
end
|
||||
|
|
@ -200,39 +212,40 @@ end
|
|||
Measure the topological charge `Q` of the configuration `U`. If the argument `Qslc` is present
|
||||
the contribution for each Euclidean time slice are returned.
|
||||
"""
|
||||
function Qtop(Qslc, U, lp::SpaceParm{4,M,B,D}, ymws::YMworkspace) where {M,B,D}
|
||||
function Qtop(Qslc, U, gp::GaugeParm, lp::SpaceParm{4,M,B,D}, ymws::YMworkspace) where {M,B,D}
|
||||
|
||||
@timeit "Qtop measurement" begin
|
||||
|
||||
ztw = ztwist(gp, lp)
|
||||
tp = (1,2,3)
|
||||
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_field_tensor!(ymws.frc1, ymws.frc2, U, 1,5, lp)
|
||||
end
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_add_qd!(ymws.rm, +, ymws.frc1, ymws.frc2, U, lp)
|
||||
end
|
||||
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_field_tensor!(ymws.frc1, ymws.frc2, U, 2,4, lp)
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_field_tensor!(ymws.frc1, ymws.frc2, U, gp.Ubnd, 1,6, ztw[1], ztw[5], lp)
|
||||
end
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_add_qd!(ymws.rm, -, ymws.frc1, ymws.frc2, U, lp)
|
||||
end
|
||||
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_field_tensor!(ymws.frc1, ymws.frc2, U, 3,6, lp)
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_field_tensor!(ymws.frc1, ymws.frc2, U, gp.Ubnd, 2,5, ztw[2], ztw[4], lp)
|
||||
end
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_add_qd!(ymws.rm, +, ymws.frc1, ymws.frc2, U, lp)
|
||||
end
|
||||
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_field_tensor!(ymws.frc1, ymws.frc2, U, gp.Ubnd, 3,4, ztw[3], ztw[6], lp)
|
||||
end
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_add_qd!(ymws.rm, -, ymws.frc1, ymws.frc2, U, lp)
|
||||
end
|
||||
|
||||
Qslc .= reshape(Array(CUDA.reduce(+, ymws.rm; dims=tp)),lp.iL[end])./(32*pi^2)
|
||||
end
|
||||
|
||||
return sum(Qslc)
|
||||
end
|
||||
Qtop(U, lp::SpaceParm{4,M,D}, ymws::YMworkspace{T}) where {T,M,D} = Qtop(zeros(T,lp.iL[end],M), U, lp, ymws)
|
||||
Qtop(U, gp::GaugeParm, lp::SpaceParm{4,M,D}, ymws::YMworkspace{T}) where {T,M,D} = Qtop(zeros(T,lp.iL[end],M), U, gp, lp, ymws)
|
||||
|
||||
|
||||
"""
|
||||
|
|
@ -241,7 +254,7 @@ Qtop(U, lp::SpaceParm{4,M,D}, ymws::YMworkspace{T}) where {T,M,D} = Qtop(zeros(T
|
|||
Measure the action density `E(t)` using the clover discretization. If the argument `Eslc`
|
||||
the contribution for each Euclidean time slice and plane are returned.
|
||||
"""
|
||||
function Eoft_clover(Eslc, U, lp::SpaceParm{4,M,B,D}, ymws::YMworkspace{T}) where {T,M,B,D}
|
||||
function Eoft_clover(Eslc, U, gp::GaugeParm, lp::SpaceParm{4,M,B,D}, ymws::YMworkspace{T}) where {T,M,B,D}
|
||||
|
||||
function acum(ipl1, ipl2, Etmp)
|
||||
|
||||
|
|
@ -270,22 +283,23 @@ function Eoft_clover(Eslc, U, lp::SpaceParm{4,M,B,D}, ymws::YMworkspace{T}) wher
|
|||
|
||||
@timeit "E(t) clover measurement" begin
|
||||
|
||||
ztw = ztwist(gp, lp)
|
||||
fill!(Eslc,zero(T))
|
||||
Etmp = zeros(T,lp.iL[end])
|
||||
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_field_tensor!(ymws.frc1, ymws.frc2, U, 1,2, lp)
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_field_tensor!(ymws.frc1, ymws.frc2, U, gp.Ubnd, 1,2, ztw[1], ztw[2], lp)
|
||||
end
|
||||
acum(1,2,Etmp)
|
||||
|
||||
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_field_tensor!(ymws.frc1, ymws.frc2, U, 3,4, lp)
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_field_tensor!(ymws.frc1, ymws.frc2, U, gp.Ubnd, 3,4, ztw[3], ztw[4], lp)
|
||||
end
|
||||
acum(3,4,Etmp)
|
||||
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_field_tensor!(ymws.frc1, ymws.frc2, U, 5,6, lp)
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_field_tensor!(ymws.frc1, ymws.frc2, U, gp.Ubnd, 5,6, ztw[5], ztw[6], lp)
|
||||
end
|
||||
acum(5,6,Etmp)
|
||||
|
||||
|
|
@ -293,7 +307,7 @@ function Eoft_clover(Eslc, U, lp::SpaceParm{4,M,B,D}, ymws::YMworkspace{T}) wher
|
|||
|
||||
return sum(Eslc)/lp.iL[end]
|
||||
end
|
||||
Eoft_clover(U, lp::SpaceParm{N,M,B,D}, ymws::YMworkspace{T}) where {T,N,M,B,D} = Eoft_clover(zeros(T,lp.iL[end],M), U, lp, ymws)
|
||||
Eoft_clover(U, gp::GaugeParm, lp::SpaceParm{N,M,B,D}, ymws::YMworkspace{T}) where {T,N,M,B,D} = Eoft_clover(zeros(T,lp.iL[end],M), U, gp, lp, ymws)
|
||||
|
||||
function krnl_add_et!(rm, op, frc1, U, lp::SpaceParm{4,M,B,D}) where {M,B,D}
|
||||
|
||||
|
|
@ -312,14 +326,16 @@ function krnl_add_qd!(rm, op, frc1, frc2, U, lp::SpaceParm{4,M,B,D}) where {M,B,
|
|||
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
|
||||
|
||||
I = point_coord((b,r), lp)
|
||||
rm[I] = op(dot( (frc1[b,1,r]+frc1[b,2,r]+frc1[b,3,r]+frc1[b,4,r]),
|
||||
(frc2[b,1,r]+frc2[b,2,r]+frc2[b,3,r]+frc2[b,4,r]) ) )
|
||||
rm[I] += op(dot( (frc1[b,1,r]+frc1[b,2,r]+frc1[b,3,r]+frc1[b,4,r]),
|
||||
(frc2[b,1,r]+frc2[b,2,r]+frc2[b,3,r]+frc2[b,4,r]) ) )
|
||||
return nothing
|
||||
end
|
||||
|
||||
function krnl_field_tensor!(frc1, frc2, U::AbstractArray{T}, ipl1, ipl2, lp::SpaceParm{4,M,B,D}) where {T,M,B,D}
|
||||
function krnl_field_tensor!(frc1, frc2, U::AbstractArray{T}, Ubnd::T, ipl1, ipl2, ztw1, ztw2, lp::SpaceParm{4,M,B,D}) where {T,M,B,D}
|
||||
|
||||
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
|
||||
it = point_time((b,r), lp)
|
||||
SFBC = ((B == BC_SF_AFWB) || (B == BC_SF_ORBI) )
|
||||
|
||||
Ush = @cuStaticSharedMem(T, (D,2))
|
||||
|
||||
|
|
@ -329,34 +345,48 @@ function krnl_field_tensor!(frc1, frc2, U::AbstractArray{T}, ipl1, ipl2, lp::Spa
|
|||
Ush[b,2] = U[b,id2,r]
|
||||
sync_threads()
|
||||
|
||||
SFBC = ((B == BC_SF_AFWB) || (B == BC_SF_ORBI) ) && (id1 == N)
|
||||
|
||||
bu1, ru1 = up((b, r), id1, lp)
|
||||
bu2, ru2 = up((b, r), id2, lp)
|
||||
bd, rd = up((bu1, ru1), id2, lp)
|
||||
if ru1 == r
|
||||
gt1 = Ush[bu1,2]
|
||||
else
|
||||
|
||||
gt1 = U[bu1,id2,ru1]
|
||||
end
|
||||
if ru2 == r
|
||||
gt2 = Ush[bu2,1]
|
||||
else
|
||||
gt2 = U[bu2,id1,ru2]
|
||||
if SFBC && (it == lp.iL[end])
|
||||
gt2 = Ubnd
|
||||
else
|
||||
gt2 = U[bu2,id1,ru2]
|
||||
end
|
||||
end
|
||||
|
||||
l1 = gt1/gt2
|
||||
l2 = Ush[b,2]\Ush[b,1]
|
||||
|
||||
frc1[b,1,r] = projalg(Ush[b,1]*l1/Ush[b,2])
|
||||
frc1[bu1,2,ru1] = projalg(l1*l2)
|
||||
frc1[bd,3,rd] = projalg(gt2\(l2*gt1))
|
||||
frc1[bu2,4,ru2] = projalg(l2*l1)
|
||||
|
||||
if SFBC && (it == lp.iL[end])
|
||||
frc1[b,1,r] = projalg(Ush[b,1]*l1/Ush[b,2])
|
||||
frc1[bu2,4,ru2] = projalg(l2*l1)
|
||||
else
|
||||
frc1[b,1,r] = projalg(ztw1, Ush[b,1]*l1/Ush[b,2])
|
||||
frc1[bu1,2,ru1] = projalg(ztw1, l1*l2)
|
||||
frc1[bd,3,rd] = projalg(ztw1, gt2\(l2*gt1))
|
||||
frc1[bu2,4,ru2] = projalg(ztw1, l2*l1)
|
||||
end
|
||||
|
||||
# Second plane
|
||||
id1, id2 = lp.plidx[ipl2]
|
||||
Ush[b,1] = U[b,id1,r]
|
||||
Ush[b,2] = U[b,id2,r]
|
||||
sync_threads()
|
||||
|
||||
SFBC = ((B == BC_SF_AFWB) || (B == BC_SF_ORBI) ) && (id1 == N)
|
||||
|
||||
bu1, ru1 = up((b, r), id1, lp)
|
||||
bu2, ru2 = up((b, r), id2, lp)
|
||||
bd, rd = up((bu1, ru1), id2, lp)
|
||||
|
|
@ -368,16 +398,25 @@ function krnl_field_tensor!(frc1, frc2, U::AbstractArray{T}, ipl1, ipl2, lp::Spa
|
|||
if ru2 == r
|
||||
gt2 = Ush[bu2,1]
|
||||
else
|
||||
gt2 = U[bu2,id1,ru2]
|
||||
if SFBC && (it == lp.iL[end])
|
||||
gt2 = Ubnd
|
||||
else
|
||||
gt2 = U[bu2,id1,ru2]
|
||||
end
|
||||
end
|
||||
|
||||
l1 = gt1/gt2
|
||||
l2 = Ush[b,2]\Ush[b,1]
|
||||
|
||||
frc2[b,1,r] = projalg(Ush[b,1]*l1/Ush[b,2])
|
||||
frc2[bu1,2,ru1] = projalg(l1*l2)
|
||||
frc2[bd,3,rd] = projalg(gt2\(l2*gt1))
|
||||
frc2[bu2,4,ru2] = projalg(l2*l1)
|
||||
if SFBC && (it == lp.iL[end])
|
||||
frc2[b,1,r] = projalg(Ush[b,1]*l1/Ush[b,2])
|
||||
frc2[bu2,4,ru2] = projalg(l2*l1)
|
||||
else
|
||||
frc2[b,1,r] = projalg(ztw1, Ush[b,1]*l1/Ush[b,2])
|
||||
frc2[bu1,2,ru1] = projalg(ztw1, l1*l2)
|
||||
frc2[bd,3,rd] = projalg(ztw1, gt2\(l2*gt1))
|
||||
frc2[bu2,4,ru2] = projalg(ztw1, l2*l1)
|
||||
end
|
||||
|
||||
return nothing
|
||||
end
|
||||
|
|
|
|||
|
|
@ -17,16 +17,17 @@ Returns the value of the gauge plaquette action for the configuration U. The par
|
|||
"""
|
||||
function gauge_action(U, lp::SpaceParm, gp::GaugeParm{T}, ymws::YMworkspace{T}) where T <: AbstractFloat
|
||||
|
||||
ztw = ztwist(gp, lp)
|
||||
if abs(gp.c0-1) < 1.0E-10
|
||||
@timeit "Wilson gauge action" begin
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_plaq!(ymws.cm, U, gp.Ubnd, gp.cG[1], lp)
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_plaq!(ymws.cm, U, gp.Ubnd, gp.cG[1], ztw, lp)
|
||||
end
|
||||
end
|
||||
else
|
||||
@timeit "Improved gauge action" begin
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_impr!(ymws.cm, U, gp.c0, (1-gp.c0)/8, gp.Ubnd, gp.cG[1], lp)
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_impr!(ymws.cm, U, gp.c0, (1-gp.c0)/8, gp.Ubnd, gp.cG[1], ztw, lp)
|
||||
end
|
||||
end
|
||||
end
|
||||
|
|
@ -38,9 +39,10 @@ end
|
|||
|
||||
function plaquette(U, lp::SpaceParm, gp::GaugeParm, ymws::YMworkspace)
|
||||
|
||||
ztw = ztwist(gp, lp)
|
||||
@timeit "Plaquette measurement" begin
|
||||
CUDA.@sync begin
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_plaq!(ymws.cm, U, gp.Ubnd, one(gp.cG[1]), lp)
|
||||
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_plaq!(ymws.cm, U, gp.Ubnd, one(gp.cG[1]), ztw, lp)
|
||||
end
|
||||
end
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue