Skip to content

Commit

Permalink
Fixes + more dont limit
Browse files Browse the repository at this point in the history
  • Loading branch information
Charlie Kawczynski committed Feb 17, 2025
1 parent 5038012 commit eea802c
Show file tree
Hide file tree
Showing 2 changed files with 7 additions and 11 deletions.
12 changes: 1 addition & 11 deletions ext/cuda/data_layouts_threadblock.jl
Original file line number Diff line number Diff line change
Expand Up @@ -309,22 +309,12 @@ end
@inline function fd_stencil_universal_index(space::Spaces.AbstractSpace, us)
(tv,) = CUDA.threadIdx()
(h, bv, ij) = CUDA.blockIdx()
vid = tv + (bv - 1) * CUDA.blockDim().x
v = tv + (bv - 1) * CUDA.blockDim().x
(Nq, _, _, _, _) = DataLayouts.universal_size(us)
if Nq * Nq < ij
return CartesianIndex((-1, -1, 1, -1, -1))
end
@inbounds (i, j) = CartesianIndices((Nq, Nq))[ij].I
v =
if space isa Spaces.FaceExtrudedFiniteDifferenceSpace ||
space isa Spaces.FaceFiniteDifferenceSpace
v = vid - half
elseif space isa Spaces.CenterExtrudedFiniteDifferenceSpace ||
space isa Spaces.CenterFiniteDifferenceSpace
v = vid
else
error("Invalid space")
end
return CartesianIndex((i, j, 1, v, h))
end
@inline fd_stencil_is_valid_index(I::CI5, us::UniversalSize) =
Expand Down
6 changes: 6 additions & 0 deletions ext/cuda/operators_fd_shmem_common.jl
Original file line number Diff line number Diff line change
Expand Up @@ -286,4 +286,10 @@ if hasfield(Method, :recursion_relation)
for m in methods(_fd_resolve_shmem!)
m.recursion_relation = dont_limit
end
for m in methods(_fd_allocate_shmem)
m.recursion_relation = dont_limit
end
for m in methods(fd_allocate_shmem)
m.recursion_relation = dont_limit
end
end

0 comments on commit eea802c

Please sign in to comment.