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
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
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue