Skip to content

Comments

Change LBnd object to use int64_t#180

Open
Goobley wants to merge 4 commits intomrnorman:mainfrom
Goobley:LBndFix
Open

Change LBnd object to use int64_t#180
Goobley wants to merge 4 commits intomrnorman:mainfrom
Goobley:LBndFix

Conversation

@Goobley
Copy link

@Goobley Goobley commented Apr 2, 2025

Otherwise, this is an int implictly constructed from a size_t in e.g. memset_loc, and then converted back to a size_t in the Kokkos RangePolicy. This results in RangePolicy throwing (e.g. Kokkos::RangePolicy bound type error: an unsafe implicit conversion is performed on a bound (18446744071830503424), which may not preserve its original value.).

This necessitates some changes in loop handling, but they couldn't handle more than int32 number of iterations (but e.g. in cuda we could do 2^31-1 * BlockSize (e.g. 128)). I think for multi-dimensional loops int32 is fine per dimension, but for things like memset_loc, that implicitly flatten, this is a problem. So I've tweaked the code to work around this.

Goobley added 4 commits April 2, 2025 12:04
…constructed from a size_t in e.e.g. memset_loc, and then converted back to a size_t in the Kokkos RangePolicy, and bad things happen
…big flat loop, and add same adjustments to FArray
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant