You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Sometimes kernel will launch gpu kernels even though they have no work to do.
The gpu implementation of kernel tracks the number of blocks/threads requested and a minimum number of blocks/threads required in each dimension to choose appropriate numbers of blocks and threads. It uses 0 as the default number of blocks/threads and then only chooses not to launch a kernel if all dimensions are still 0 after calculating dimensions.
To Reproduce
Use kernel with a 2d loop over ranges of size 1 and 0 respectively.
Expected behavior
Kernel should not launch a kernel if any dimension in a gpu policy has no work.
Compilers & Libraries (please complete the following information):
CUDA/HIP version: any
Additional context
Discovered this when I saw segfaults while trying out new "unchecked" policies with 0 length list segments in the test-kernel-nested-loop-segments-HIP test.
The text was updated successfully, but these errors were encountered:
Describe the bug
Sometimes kernel will launch gpu kernels even though they have no work to do.
The gpu implementation of kernel tracks the number of blocks/threads requested and a minimum number of blocks/threads required in each dimension to choose appropriate numbers of blocks and threads. It uses 0 as the default number of blocks/threads and then only chooses not to launch a kernel if all dimensions are still 0 after calculating dimensions.
To Reproduce
Use kernel with a 2d loop over ranges of size 1 and 0 respectively.
Expected behavior
Kernel should not launch a kernel if any dimension in a gpu policy has no work.
Compilers & Libraries (please complete the following information):
Additional context
Discovered this when I saw segfaults while trying out new "unchecked" policies with 0 length list segments in the test-kernel-nested-loop-segments-HIP test.
The text was updated successfully, but these errors were encountered: