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

Wrap threadIdx, etc in an inline function

Browse source
This commit is contained in:
Alberto Ramos 2022-01-05 20:04:55 +01:00
parent 65a619e554
commit 584b92edb0
7 changed files with 21 additions and 18 deletions
Download patch file Download diff file Expand all files Collapse all files

4
src/Dirac/Dirac.jl
View file

@ -64,7 +64,7 @@ end
function krnl_Dw!(so, U, si, m0, th, lp::SpaceParm{4,6,B,D}) where {B,D}
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
b, r = assign_thx()
# For SF:
# - cttilde affects mass term at x0 = a, T-a
@ -85,7 +85,7 @@ end
function krnl_g5Dw!(so, U, si, m0, th, lp::SpaceParm{4,6,B,D}) where {B,D}
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
b, r = assign_thx()
@inbounds begin
so[b,r] = (4+m0)*si[b,r]

2
src/LatticeGPU.jl
View file

@ -23,7 +23,7 @@ include("Space/Space.jl")
using .Space
export SpaceParm
export up, dw, updw, point_coord, point_index, point_time
export up, dw, updw, point_coord, point_index, point_time, assign_thx
export BC_PERIODIC, BC_OPEN, BC_SF_AFWB, BC_SF_ORBI
include("Fields/Fields.jl")

5
src/Space/Space.jl
View file

@ -333,7 +333,10 @@ Returns the sum of the cartesian coordinates of the point p=(b,r).
return s
end
export up, dw, updw, point_index, point_coord, point_time
@inline assign_thx() = convert(Int64, CUDA.threadIdx().x), convert(Int64, CUDA.blockIdx().x)
export up, dw, updw, point_index, point_coord, point_time, assign_thx
export BC_PERIODIC, BC_OPEN, BC_SF_AFWB, BC_SF_ORBI
end

8
src/YM/YMact.jl
View file

@ -11,7 +11,7 @@
function krnl_impr!(plx, U::AbstractArray{T}, c0, c1, Ubnd::NTuple{NB,T}, cG, ztw, lp::SpaceParm{N,M,B,D}) where {T,NB,N,M,B,D}
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
b, r = assign_thx()
it = point_time((b, r), lp)
Ush = @cuStaticSharedMem(T, (D,2))
@ -121,7 +121,7 @@ end
function krnl_plaq!(plx, U::AbstractArray{T}, Ubnd, cG, ztw, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
b, r = assign_thx()
it = point_time((b, r), lp)
Ush = @cuStaticSharedMem(T, (D,2))
@ -172,7 +172,7 @@ end
function krnl_force_wilson_pln!(frc1, frc2, U::AbstractArray{T}, Ubnd, cG, ztw, ipl, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
b, r = assign_thx()
it = point_time((b,r), lp)
Ush = @cuStaticSharedMem(T, (D,2))
@ -233,7 +233,7 @@ end
function krnl_force_impr_pln!(frc1, frc2, U::AbstractArray{T}, c0, c1, Ubnd, cG, ztw, ipl, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
b, r = assign_thx()
it = point_time((b, r), lp)
Ush = @cuStaticSharedMem(T, (D,2))

6
src/YM/YMfields.jl
View file

@ -36,7 +36,7 @@ end
function krnl_assign_SU3!(frc::AbstractArray{T}, m, lp::SpaceParm{N,M,BC_PERIODIC,D}) where {T,N,M,D}
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
b, r = assign_thx()
for id in 1:lp.ndim
frc[b,id,r] = SU3alg(m[b,id,1,r], m[b,id,2,r], m[b,id,3,r],
m[b,id,4,r], m[b,id,5,r], m[b,id,6,r],
@ -47,7 +47,7 @@ end
function krnl_assign_SU3!(frc::AbstractArray{T}, m, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
b, r = assign_thx()
it = point_time((b,r), lp)
if ((B==BC_SF_AFWB)||(B==BC_SF_ORBI))
@ -72,7 +72,7 @@ end
function krnl_assign_SU2!(frc, m, lp::SpaceParm{N,M,BC_PERIODIC,D}) where {N,M,D}
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
b, r = assign_thx()
for id in 1:lp.ndim
frc[b,id,r] = SU2alg(m[b,id,1,r], m[b,id,2,r], m[b,id,3,r])
end

10
src/YM/YMflow.jl
View file

@ -89,7 +89,7 @@ end
function krnl_add_zth!(frc, frc2::AbstractArray{TA}, U::AbstractArray{TG}, lp::SpaceParm{N,M,B,D}) where {TA,TG,N,M,B,D}
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
b, r = assign_thx()
it = point_time((b, r), lp)
SFBC = ((B == BC_SF_AFWB) || (B == BC_SF_ORBI) )
@ -261,7 +261,7 @@ Eoft_plaq(U, gp::GaugeParm{T,G,NN}, lp::SpaceParm{N,M,B,D}, ymws::YMworkspace) w
function krnl_plaq_pln!(plx, U::AbstractArray{T}, Ubnd, ztw, ipl, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
b, r = assign_thx()
id1, id2 = lp.plidx[ipl]
SFBC = ((B == BC_SF_AFWB) || (B == BC_SF_ORBI)) && (id1 == lp.iL[end])
@ -387,7 +387,7 @@ Eoft_clover(U, gp::GaugeParm, lp::SpaceParm{N,M,B,D}, ymws::YMworkspace{T}) wher
function krnl_add_et!(rm, op, frc1, U, lp::SpaceParm{4,M,B,D}) where {M,B,D}
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
b, r = assign_thx()
X1 = (frc1[b,1,r]+frc1[b,2,r]+frc1[b,3,r]+frc1[b,4,r])
@ -399,7 +399,7 @@ end
function krnl_add_qd!(rm, op, frc1, frc2, U, lp::SpaceParm{4,M,B,D}) where {M,B,D}
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
b, r = assign_thx()
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]),
@ -409,7 +409,7 @@ end
function krnl_field_tensor!(frc1::AbstractArray{TA}, frc2, U::AbstractArray{T}, Ubnd, ipl1, ipl2, ztw1, ztw2, lp::SpaceParm{4,M,B,D}) where {TA,T,M,B,D}
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
b, r = assign_thx()
it = point_time((b,r), lp)
SFBC = ((B == BC_SF_AFWB) || (B == BC_SF_ORBI) )

4
src/YM/YMsf.jl
View file

@ -44,7 +44,7 @@ end
function krnl_sfcoupling!(rm, U::AbstractArray{T}, Ubnd, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
b, r = assign_thx()
I = point_coord((b,r), lp)
it = I[N]
@ -99,7 +99,7 @@ end
function krnl_setbnd_it0!(U, phi1, phi2, lp::SpaceParm{N,M,B,D}) where {N,M,B,D}
b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
b, r = assign_thx()
it = point_time((b,r), lp)
SFBC = (B == BC_SF_AFWB) || (B == BC_SF_ORBI)