lattice/latticegpu.jl
1
0
Fork 0
mirror of https://igit.ific.uv.es/alramos/latticegpu.jl.git synced 2025-05-14 19:23:42 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

SF fix for the propagators

Browse source
This commit is contained in:
Fernando Pérez Panadero 2023-12-13 11:42:26 +01:00
parent 6a90d74024
commit a7bc21769b
5 changed files with 101 additions and 91 deletions
Download patch file Download diff file Expand all files Collapse all files

5
docs/src/dirac.md
View file

@ -63,6 +63,11 @@ in the first time-slice is zero. To enforce this, we have the function
SF_bndfix!
```
Note that this is not enforced in the Dirac operators, so if the field `so` does not satisfy SF
boundary conditions, it will not (in general) satisfy them after applying [`Dw!`](@ref)
or [`g5Dw!`](@ref). This function is called for the function [`DwdagDw!`](@ref), so in this case
`so` will always be a proper SF field after calling this function.
The function [`Csw!`](@ref) is used to store the clover in `dws.csw`. It is computed
according to the expression

15
src/Dirac/Dirac.jl
View file

@ -28,7 +28,7 @@ The parameters are:
- `m0::T` : Mass of the fermion
- `csw::T` : Improvement coefficient for the Csw term
- `th{Ntuple{4,Complex{T}}}` : Phase for the fermions included in the boundary conditions, reabsorbed in the Dirac operator.
- `tm` : Twisted mass paramete
- `tm` : Twisted mass parameter
- `ct` : Boundary improvement term, only used for Schrödinger Funtional boundary conditions.
"""
struct DiracParam{T,R}
@ -534,12 +534,13 @@ function DwdagDw!(so, U, si, dpar::DiracParam, dws::DiracWorkspace, lp::Union{Sp
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_g5Dwimpr!(dws.st, U, si, dws.csw, dpar.m0, dpar.tm, dpar.th, dpar.csw, dpar.ct, lp)
end
end
SF_bndfix!(dws.st,lp)
@timeit "g5Dw" begin
CUDA.@sync begin
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_g5Dwimpr!(so, U, dws.st, dws.csw, dpar.m0, dpar.tm, dpar.th, dpar.csw, dpar.ct, lp)
end
end
SF_bndfix!(so,lp)
end
else
@timeit "DwdagDw" begin
@ -549,12 +550,13 @@ function DwdagDw!(so, U, si, dpar::DiracParam, dws::DiracWorkspace, lp::Union{Sp
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_g5Dw!(dws.st, U, si, dpar.m0, dpar.tm, dpar.th, dpar.ct, lp)
end
end
SF_bndfix!(dws.st,lp)
@timeit "g5Dw" begin
CUDA.@sync begin
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_g5Dw!(so, U, dws.st, dpar.m0, dpar.tm, dpar.th, dpar.ct, lp)
end
end
SF_bndfix!(so,lp)
end
end
@ -604,10 +606,11 @@ end
Sets all the values of `sp` in the first time slice to zero.
"""
function SF_bndfix!(sp, lp::Union{SpaceParm{4,6,BC_SF_ORBI,D},SpaceParm{4,6,BC_SF_AFWB,D}}) where {D}
CUDA.@sync begin
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_sfbndfix!(sp, lp)
@timeit "SF boundary fix" begin
CUDA.@sync begin
CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_sfbndfix!(sp, lp)
end
end
return nothing
end

8
src/Solvers/Propagators.jl
View file

@ -56,7 +56,7 @@ end
function bndpropagator!(pro,U, dpar::DiracParam{T}, dws::DiracWorkspace, lp::SpaceParm{4,6,1,D}, maxiter::Int64, tol::Float64, c::Int64, s::Int64)
Saves the propagator in from the t=0 boundary to the bulk for the SF boundary conditions for a source with color 'c' and spin 's'. The factor c_t is included while the factor 1/sqrt(V) is not.
Saves the propagator from the t=0 boundary to the bulk for the SF boundary conditions for a source with color 'c' and spin 's' in 'pro'. The factor c_t is included while the factor 1/sqrt(V) is not.
For the propagator from T to the bulk, use the function Tbndpropagator(U, dpar::DiracParam{T}, dws::DiracWorkspace, lp::SpaceParm{4,6,1,D}, maxiter::Int64, tol::Float64, c::Int64, s::Int64)
"""
@ -81,6 +81,7 @@ function bndpropagator!(pro, U, dpar::DiracParam{T}, dws::DiracWorkspace, lp::Sp
return nothing
end
SF_bndfix!(pro,lp)
fill!(dws.sp,zero(eltype(scalar_field(Spinor{4,SU3fund{Float64}},lp))))
CUDA.@sync begin
@ -94,7 +95,7 @@ function bndpropagator!(pro, U, dpar::DiracParam{T}, dws::DiracWorkspace, lp::Sp
g5Dw!(pro,U,dpar.ct*dws.sp,dpar,dws,lp)
CG!(pro,U,DwdagDw!,dpar,lp,dws,maxiter,tol)
return pro
return nothing
end
"""
@ -125,6 +126,7 @@ function Tbndpropagator!(pro, U, dpar::DiracParam{T}, dws::DiracWorkspace, lp::S
return nothing
end
SF_bndfix!(pro,lp)
fill!(dws.sp,zero(eltype(scalar_field(Spinor{4,SU3fund{Float64}},lp))))
CUDA.@sync begin
@ -138,7 +140,7 @@ function Tbndpropagator!(pro, U, dpar::DiracParam{T}, dws::DiracWorkspace, lp::S
g5Dw!(pro,U,dpar.ct*dws.sp,dpar,dws,lp)
CG!(pro,U,DwdagDw!,dpar,lp,dws,maxiter,tol)
return pro
return nothing
end

140
test/dirac/test_fp_fa.jl
View file

@ -8,59 +8,7 @@ using TimerOutputs
function fP_test(;theta = (0.5,0.7,1.0,0.0), m = 1.3, size = (8,8,8,16),prec = 1.0e-16)
@timeit "fP inversion (x12)" begin
lp = SpaceParm{4}(size,(4,4,4,4),1,(0,0,0,0,0,0));
exptheta = exp.(im.*theta./lp.iL);
dpar = DiracParam{Float64}(SU3fund,m,0.0,exptheta,0.0,1.0);
dws = DiracWorkspace(SU3fund,Float64,lp);
U = fill!(vector_field(SU3{Float64},lp),one(SU3{Float64}));
psi = scalar_field(Spinor{4,SU3fund{Float64}},lp);
res = zeros(lp.iL[4])
for s in 1:4 for c in 1:3
bndpropagator!(psi,U,dpar,dws,lp,1000,prec,c,s);
for t in 1:lp.iL[4]
#for i in 1:lp.iL[1] for j in 1:lp.iL[2] for k in 1:lp.iL[3]
i=abs(rand(Int))%lp.iL[1] +1;j=abs(rand(Int))%lp.iL[2] +1;k=abs(rand(Int))%lp.iL[3] +1;
CUDA.@allowscalar (res[t] += norm2(psi[point_index(CartesianIndex{lp.ndim}((i,j,k,t)),lp)...])/2)
#end end end
#res[t] = res[t]/(lp.iL[1]*lp.iL[2]*lp.iL[3])
end
end end
end
@timeit "fP analitical solution" begin
#THEORETICAL SOLUTION: hep-lat/9606016 eq (2.33)
res_th = zeros(lp.iL[4])
pp3 = ntuple(i -> theta[i]/lp.iL[i],3)
omega = 2 * asinh(0.5* sqrt(( sum((sin.(pp3)).^2) + (m + 2*(sum((sin.(pp3./2)).^2) ))^2) / (1+m+2*(sum((sin.(pp3./2)).^2) )) ) )
pp = (-im*omega,pp3...)
Mpp = m + 2* sum((sin.(pp./2)).^2)
Rpp = Mpp*(1-exp(-2*omega*lp.iL[4])) + sinh(omega) * (1+exp(-2*omega*lp.iL[4]))
for i in 2:lp.iL[4]
res_th[i] = (2*3*sinh(omega)/(Rpp^2)) * ( (Mpp + sinh(omega))*exp(-2*omega*(i-1)) - (Mpp - sinh(omega))*exp(-2*omega*(2*lp.iL[4]- (i - 1))) )
end
end
return sum(abs.(res-res_th))
end
function fA_test(;theta = (0.5,0.7,1.0,0.0), m = 1.3, size = (8,8,8,16),prec = 1.0e-16)
@timeit "fA inversion (x12)" begin
@timeit "fP inversion (x12)" begin
lp = SpaceParm{4}(size,(4,4,4,4),1,(0,0,0,0,0,0));
exptheta = exp.(im.*theta./lp.iL);
@ -71,7 +19,7 @@ function fA_test(;theta = (0.5,0.7,1.0,0.0), m = 1.3, size = (8,8,8,16),prec = 1
U = fill!(vector_field(SU3{Float64},lp),one(SU3{Float64}));
psi = scalar_field(Spinor{4,SU3fund{Float64}},lp);
res = im*zeros(lp.iL[4])
res = zeros(lp.iL[4])
for s in 1:4 for c in 1:3
bndpropagator!(psi,U,dpar,dws,lp,1000,prec,c,s);
@ -79,7 +27,7 @@ function fA_test(;theta = (0.5,0.7,1.0,0.0), m = 1.3, size = (8,8,8,16),prec = 1
for t in 1:lp.iL[4]
#for i in 1:lp.iL[1] for j in 1:lp.iL[2] for k in 1:lp.iL[3]
i=abs(rand(Int))%lp.iL[1] +1;j=abs(rand(Int))%lp.iL[2] +1;k=abs(rand(Int))%lp.iL[3] +1;
CUDA.@allowscalar (res[t] += -dot(psi[point_index(CartesianIndex{lp.ndim}((i,j,k,t)),lp)...],dmul(Gamma{4},psi[point_index(CartesianIndex{lp.ndim}((i,j,k,t)),lp)...]))/2)
CUDA.@allowscalar (res[t] += norm2(psi[point_index(CartesianIndex{lp.ndim}((i,j,k,t)),lp)...])/2)
#end end end
#res[t] = res[t]/(lp.iL[1]*lp.iL[2]*lp.iL[3])
@ -87,25 +35,77 @@ function fA_test(;theta = (0.5,0.7,1.0,0.0), m = 1.3, size = (8,8,8,16),prec = 1
end end
end
#THEORETICAL SOLUTION: hep-lat/9606016 eq (2.32)
@timeit "fA analitical solution" begin
res_th = zeros(lp.iL[4])
pp3 = ntuple(i -> theta[i]/lp.iL[i],3)
omega = 2 * asinh(0.5* sqrt(( sum((sin.(pp3)).^2) + (m + 2*(sum((sin.(pp3./2)).^2) ))^2) / (1+m+2*(sum((sin.(pp3./2)).^2) )) ) )
pp = (-im*omega,pp3...)
Mpp = m + 2* sum((sin.(pp./2)).^2)
Rpp = Mpp*(1-exp(-2*omega*lp.iL[4])) + sinh(omega) * (1+exp(-2*omega*lp.iL[4]))
for i in 2:lp.iL[4]
res_th[i] = (6/(Rpp^2)) * ( 2*(Mpp - sinh(omega))*(Mpp + sinh(omega))*exp(-2*omega*lp.iL[4])
- Mpp*((Mpp + sinh(omega))*exp(-2*omega*(i-1)) + (Mpp - sinh(omega))*exp(-2*omega*(2*lp.iL[4]- (i - 1)))))
end
@timeit "fP analitical solution" begin
#THEORETICAL SOLUTION: hep-lat/9606016 eq (2.33)
res_th = zeros(lp.iL[4])
pp3 = ntuple(i -> theta[i]/lp.iL[i],3)
omega = 2 * asinh(0.5* sqrt(( sum((sin.(pp3)).^2) + (m + 2*(sum((sin.(pp3./2)).^2) ))^2) / (1+m+2*(sum((sin.(pp3./2)).^2) )) ) )
pp = (-im*omega,pp3...)
Mpp = m + 2* sum((sin.(pp./2)).^2)
Rpp = Mpp*(1-exp(-2*omega*lp.iL[4])) + sinh(omega) * (1+exp(-2*omega*lp.iL[4]))
for i in 2:lp.iL[4]
res_th[i] = (2*3*sinh(omega)/(Rpp^2)) * ( (Mpp + sinh(omega))*exp(-2*omega*(i-1)) - (Mpp - sinh(omega))*exp(-2*omega*(2*lp.iL[4]- (i - 1))) )
end
end
return sum(abs.(res-res_th))
end
function fA_test(;theta = (0.5,0.7,1.0,0.0), m = 1.3, size = (8,8,8,16),prec = 1.0e-16)
@timeit "fA inversion (x12)" begin
lp = SpaceParm{4}(size,(4,4,4,4),1,(0,0,0,0,0,0));
exptheta = exp.(im.*theta./lp.iL);
dpar = DiracParam{Float64}(SU3fund,m,0.0,exptheta,0.0,1.0);
dws = DiracWorkspace(SU3fund,Float64,lp);
U = fill!(vector_field(SU3{Float64},lp),one(SU3{Float64}));
psi = scalar_field(Spinor{4,SU3fund{Float64}},lp);
res = im*zeros(lp.iL[4])
for s in 1:4 for c in 1:3
bndpropagator!(psi,U,dpar,dws,lp,1000,prec,c,s);
for t in 1:lp.iL[4]
#for i in 1:lp.iL[1] for j in 1:lp.iL[2] for k in 1:lp.iL[3]
i=abs(rand(Int))%lp.iL[1] +1;j=abs(rand(Int))%lp.iL[2] +1;k=abs(rand(Int))%lp.iL[3] +1;
CUDA.@allowscalar (res[t] += -dot(psi[point_index(CartesianIndex{lp.ndim}((i,j,k,t)),lp)...],dmul(Gamma{4},psi[point_index(CartesianIndex{lp.ndim}((i,j,k,t)),lp)...]))/2)
#end end end
#res[t] = res[t]/(lp.iL[1]*lp.iL[2]*lp.iL[3])
end
end end
end
#THEORETICAL SOLUTION: hep-lat/9606016 eq (2.32)
@timeit "fA analitical solution" begin
res_th = zeros(lp.iL[4])
pp3 = ntuple(i -> theta[i]/lp.iL[i],3)
omega = 2 * asinh(0.5* sqrt(( sum((sin.(pp3)).^2) + (m + 2*(sum((sin.(pp3./2)).^2) ))^2) / (1+m+2*(sum((sin.(pp3./2)).^2) )) ) )
pp = (-im*omega,pp3...)
Mpp = m + 2* sum((sin.(pp./2)).^2)
Rpp = Mpp*(1-exp(-2*omega*lp.iL[4])) + sinh(omega) * (1+exp(-2*omega*lp.iL[4]))
for i in 2:lp.iL[4]
res_th[i] = (6/(Rpp^2)) * ( 2*(Mpp - sinh(omega))*(Mpp + sinh(omega))*exp(-2*omega*lp.iL[4])
- Mpp*((Mpp + sinh(omega))*exp(-2*omega*(i-1)) + (Mpp - sinh(omega))*exp(-2*omega*(2*lp.iL[4]- (i - 1)))))
end
end
return sum(abs.(res-res_th))
end

10
test/dirac/test_solver_plw.jl
View file

@ -96,15 +96,15 @@ end
begin
dif = 0.0
diff = 0.0
for i in 1:3 for j in 1:4
dif += Dwpw_test(c=i,s=j)
global diff += Dwpw_test(c=i,s=j)
end end
if dif < 1.0e-15
print("Dwpl test passed with average error ", dif/12,"!\n")
if diff < 1.0e-15
print("Dwpl test passed with average error ", diff/12,"!\n")
else
error("Dwpl test failed with difference: ",dif,"\n")
error("Dwpl test failed with difference: ",diff,"\n")
end