From 4c560ab0feb5bbf2d1c8a14ecd7725f2837db5f6 Mon Sep 17 00:00:00 2001 From: "Fernando P.Panadero" Date: Thu, 4 Jul 2024 15:22:20 +0200 Subject: [PATCH] SFORBI different from SFAFWB. --- src/YM/YMact.jl | 229 +++++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 227 insertions(+), 2 deletions(-) diff --git a/src/YM/YMact.jl b/src/YM/YMact.jl index f9849bd..4c03097 100644 --- a/src/YM/YMact.jl +++ b/src/YM/YMact.jl @@ -324,13 +324,99 @@ end ## ## SF ## -function krnl_impr!(plx, U::AbstractArray{T}, c0, c1, Ubnd::NTuple{NB,T}, cG, ztw, lp::Union{SpaceParm{N,M,BC_SF_ORBI,D},SpaceParm{N,M,BC_SF_AFWB,D}}) where {T,NB,N,M,D} +function krnl_impr!(plx, U::AbstractArray{T}, c0, c1, Ubnd::NTuple{NB,T}, cG, ztw, lp::SpaceParm{N,M,BC_SF_ORBI,D}) where {T,NB,N,M,D} b = Int64(CUDA.threadIdx().x) r = Int64(CUDA.blockIdx().x) I = point_coord((b,r), lp) it = I[N] + ipl = 0 + S = zero(eltype(plx)) + @inbounds begin + for id1 in N:-1:1 + bu1, ru1 = up((b, r), id1, lp) + SFBC = (id1==N) + + for id2 = 1:id1-1 + bu2, ru2 = up((b, r), id2, lp) + ipl = ipl + 1 + + TWP = (I[id1]==1) && (I[id2]==1) + TWH2 = TWP || ( (I[id1]==lp.iL[id1]) && (I[id2]==1) ) + TWH3 = TWP || ( (I[id1]==1) && (I[id2]==lp.iL[id2]) ) + + # H2 staple + (b1, r1) = up((b,r), id1, lp) + (b2, r2) = up((b1,r1), id1, lp) + if SFBC && (it == lp.iL[end]-1) + gb = Ubnd[id2] + else + gb = U[b2,id2,r2] + end + + (b2, r2) = up((b1,r1), id2, lp) + h2 = (U[b1,id1,r1]*gb)/U[b2,id1,r2] + + # H3 staple + (b1, r1) = up((b,r), id2, lp) + (b2, r2) = up((b1,r1), id2, lp) + + (b3, r3) = up((b1,r1), id1, lp) + if SFBC && (it == lp.iL[end]) + gc = Ubnd[id2] + else + gc = U[b3,id2,r3] + end + h3 = (U[b1,id2,r1]*U[b2,id1,r2])/gc + # END staples + + if SFBC && (it == lp.iL[end]) + ga = Ubnd[id2] + else + ga = U[bu1,id2,ru1] + end + + g2 = U[b,id2,r]\U[b,id1,r] + + if (it == lp.iL[end]) && SFBC + S += cG*(c0*tr(g2*ga/U[bu2,id1,ru2])) + (c1)*tr(g2*ga/h3) + (c1/2)*tr((g2*ga/U[bu2,id1,ru2])*g2*ga/U[bu2,id1,ru2]) + elseif (it == 1) && SFBC + S += cG*(c0*tr(g2*ga/U[bu2,id1,ru2])) + (c1)*tr(g2*ga/h3) + c1*tr(g2*h2/U[bu2,id1,ru2]) + (c1/2)*tr((g2*ga/U[bu2,id1,ru2])*g2*ga/U[bu2,id1,ru2]) + else + if TWP + S += (ztw[ipl]*c0)*tr(g2*ga/U[bu2,id1,ru2]) + else + S += c0*tr(g2*ga/U[bu2,id1,ru2]) + end + if TWH2 + S += (ztw[ipl]*c1)*tr(g2*h2/U[bu2,id1,ru2]) + else + S += c1*tr(g2*h2/U[bu2,id1,ru2]) + end + if TWH3 + S += (ztw[ipl]*c1)*tr(g2*ga/h3) + else + S += c1*tr(g2*ga/h3) + end + end + + end + end + + plx[I] = S + end + + return nothing +end + +function krnl_impr!(plx, U::AbstractArray{T}, c0, c1, Ubnd::NTuple{NB,T}, cG, ztw, lp::SpaceParm{N,M,BC_SF_AFWB,D}) where {T,NB,N,M,D} + + b = Int64(CUDA.threadIdx().x) + r = Int64(CUDA.blockIdx().x) + I = point_coord((b,r), lp) + it = I[N] + ipl = 0 S = zero(eltype(plx)) @inbounds begin @@ -509,7 +595,146 @@ function krnl_force_wilson_pln!(frc1, frc2, U::AbstractArray{T}, Ubnd, cG, ztw, return nothing end -function krnl_force_impr_pln!(frc1, frc2, U::AbstractArray{T}, c0, c1, Ubnd, cG, ztw, ipl, lp::Union{SpaceParm{N,M,BC_SF_ORBI,D},SpaceParm{N,M,BC_SF_AFWB,D}}) where {T,N,M,D} +function krnl_force_impr_pln!(frc1, frc2, U::AbstractArray{T}, c0, c1, Ubnd, cG, ztw, ipl, lp::SpaceParm{N,M,BC_SF_ORBI,D}) where {T,N,M,D} + + b = Int64(CUDA.threadIdx().x) + r = Int64(CUDA.blockIdx().x) + I = point_coord((b,r), lp) + it = I[N] + + @inbounds begin + id1, id2 = lp.plidx[ipl] + bu1, ru1 = up((b, r), id1, lp) + bu2, ru2 = up((b, r), id2, lp) + + SFBC = (id1 == N) + TWP = (I[id1]==1) && (I[id2]==1) + TWH1 = TWP || ( (I[id1]==1) && (I[id2]==2) ) + TWH2 = TWP || ( (I[id1]==lp.iL[id1]) && (I[id2]==1) ) + TWH3 = TWP || ( (I[id1]==1) && (I[id2]==lp.iL[id2]) ) + TWH4 = TWP || ( (I[id1]==2) && (I[id2]==1) ) + + # H1 staple + (b1, r1) = dw((b,r), id2, lp) + (b2, r2) = up((b1,r1), id1, lp) + if SFBC && (it == lp.iL[end]) + gc = Ubnd[id2] + else + gc = U[b2,id2,r2] + end + h1 = (U[b1,id2,r1]\U[b1,id1,r1])*gc + + # H2 staple + (b1, r1) = up((b,r), id1, lp) + (b2, r2) = up((b1,r1), id1, lp) + if SFBC && (it == lp.iL[end]-1) + gb = Ubnd[id2] + else + gb = U[b2,id2,r2] + end + + (b2, r2) = up((b1,r1), id2, lp) + h2 = (U[b1,id1,r1]*gb)/U[b2,id1,r2] + + # H3 staple + (b1, r1) = up((b,r), id2, lp) + (b2, r2) = up((b1,r1), id2, lp) + (b3, r3) = up((b1,r1), id1, lp) + if SFBC && (it == lp.iL[end]) + gc = Ubnd[id2] + else + gc = U[b3,id2,r3] + end + h3 = (U[b1,id2,r1]*U[b2,id1,r2])/gc + + # H4 staple + (b1, r1) = dw((b,r), id1, lp) + (b2, r2) = up((b1,r1), id2, lp) + h4 = (U[b1,id1,r1]\U[b1,id2,r1])*U[b2,id1,r2] + # END staples + + if SFBC && (it == lp.iL[end]) + ga = Ubnd[id2] + else + ga = U[bu1,id2,ru1] + end + + g1 = ga/U[bu2,id1,ru2] + g2 = U[b,id2,r]\U[b,id1,r] + + if SFBC && (it == 1) + X = (cG*c0)*projalg(U[b,id1,r]*g1/U[b,id2,r]) + c1*projalg(U[b,id1,r]*h2/(U[b,id2,r]*U[bu2,id1,ru2])) + + (c1)*projalg(U[b,id1,r]*ga/(U[b,id2,r]*h3)) + c1*(projalg(U[b,id1,r]*g1*g2*g1/U[b,id2,r])) + + frc1[b,id1,r] -= X + + frc2[bu1,id2,ru1] -= (cG*c0)*projalg(g1*g2) + c1*projalg((ga/h3)*g2) + + c1*projalg((g1/U[b,id2,r])*h1) + c1*projalg(g1*g2*g1*g2) + + frc2[bu2,id1,ru2] += (cG*c0)*projalg(g2*g1) + c1*projalg((U[b,id2,r]\h1)*g1) + + c1*projalg(g2*h2/U[bu2,id1,ru2]) + c1*projalg(g2*g1*g2*g1) + elseif SFBC && (it == lp.iL[end]) + X = cG*c0*projalg(U[b,id1,r]*g1/U[b,id2,r]) + + c1 * (projalg(U[b,id1,r]*ga/(U[b,id2,r]*h3))) + c1*(projalg(U[b,id1,r]*g1*g2*g1/U[b,id2,r])) + + frc1[b,id1,r] -= X + c1*projalg(U[b,id1,r]*g1/h4) + frc1[b,id2,r] += X + c1*projalg(h1*g1/U[b,id2,r]) + + frc2[bu2,id1,ru2] += (cG*c0)*projalg(g2*g1) + c1*projalg((U[b,id2,r]\h1)*g1) + + c1 * projalg(h4\U[b,id1,r]*g1) + c1* projalg(g2*g1*g2*g1) + else + if TWP + X = projalg(c0*ztw,U[b,id1,r]*g1/U[b,id2,r]) + frc2[bu1,id2,ru1] -= projalg(c0*ztw,g1*g2) + frc2[bu2,id1,ru2] += projalg(c0*ztw,g2*g1) + else + X = c0*projalg(U[b,id1,r]*g1/U[b,id2,r]) + frc2[bu1,id2,ru1] -= c0*projalg(g1*g2) + frc2[bu2,id1,ru2] += c0*projalg(g2*g1) + end + if TWH1 + frc1[b,id2,r] += projalg(ztw*c1,h1*g1/U[b,id2,r]) + frc2[bu1,id2,ru1] -= projalg(ztw*c1,(g1/U[b,id2,r])*h1) + frc2[bu2,id1,ru2] += projalg(ztw*c1,(U[b,id2,r]\h1)*g1) + else + frc1[b,id2,r] += c1*projalg(h1*g1/U[b,id2,r]) + frc2[bu1,id2,ru1] -= c1*projalg((g1/U[b,id2,r])*h1) + frc2[bu2,id1,ru2] += c1*projalg((U[b,id2,r]\h1)*g1) + end + if TWH2 + X += projalg(ztw*c1,U[b,id1,r]*h2/(U[b,id2,r]*U[bu2,id1,ru2])) + frc2[bu2,id1,ru2] += projalg(ztw*c1,g2*h2/U[bu2,id1,ru2]) + else + X += c1*projalg(U[b,id1,r]*h2/(U[b,id2,r]*U[bu2,id1,ru2])) + frc2[bu2,id1,ru2] += c1*projalg(g2*h2/U[bu2,id1,ru2]) + end + if TWH3 + X += projalg(ztw*c1,U[b,id1,r]*ga/(U[b,id2,r]*h3)) + frc2[bu1,id2,ru1] -= projalg(ztw*c1,(ga/h3)*g2) + else + X += c1*projalg(U[b,id1,r]*ga/(U[b,id2,r]*h3)) + frc2[bu1,id2,ru1] -= c1*projalg((ga/h3)*g2) + end + if TWH4 + frc1[b,id1,r] -= projalg(ztw*c1,U[b,id1,r]*g1/h4) + frc2[bu1,id2,ru1] -= projalg(ztw*c1,(g1/h4)*U[b,id1,r]) + frc2[bu2,id1,ru2] += projalg(ztw*c1,h4\U[b,id1,r]*g1) + else + frc1[b,id1,r] -= c1*projalg(U[b,id1,r]*g1/h4) + frc2[bu1,id2,ru1] -= c1*projalg((g1/h4)*U[b,id1,r]) + frc2[bu2,id1,ru2] += c1*projalg(h4\U[b,id1,r]*g1) + end + frc1[b,id1,r] -= X + frc1[b,id2,r] += X + + end + + end + + return nothing +end + +function krnl_force_impr_pln!(frc1, frc2, U::AbstractArray{T}, c0, c1, Ubnd, cG, ztw, ipl, lp::SpaceParm{N,M,BC_SF_AFWB,D}) where {T,N,M,D} b = Int64(CUDA.threadIdx().x) r = Int64(CUDA.blockIdx().x)