Skip to content

Backport fix from PR #47 #48

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
lohedges merged 1 commit into from
Sep 16, 2026
Merged

Backport fix from PR #47 #48

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion src/loch/_kernels.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,7 @@
#define CONSTANT __constant
#define LOCAL __local
#define GET_GLOBAL_ID(dim) get_global_id(dim)
#define BLOCK_ID_Y get_group_id(1) // OpenCL: work-group ID in dimension 1
#define BLOCK_ID_Y get_global_id(1) // OpenCL: y block size is 1, and the runtime picks the work-group shape
// Map CUDA-style function names to OpenCL names
#define sqrtf sqrt
#define powf pow
Expand Down
6 changes: 5 additions & 1 deletion src/loch/_platforms/_opencl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -207,7 +207,11 @@ def wrapper(*args, **kwargs):
grid = kwargs.get("grid", (1, 1, 1))

global_size = tuple(b * g for b, g in zip(block, grid))
local_size = block

# OpenCL caps the work-group size per device (e.g. 256 on AMD),
# below common CUDA block sizes. The kernels use no local memory
# or barriers, so let the runtime pick the partitioning.
local_size = None

processed_args = []
for arg in args:
Expand Down