KyleVaughn / UM2

An unstructured mesh library for automated method of characteristic mesh generation
https://univeristy-of-michigan-unstructured-mesh-code.readthedocs.io/en/main/index.html#
MIT License
7 stars 2 forks source link

Update cuda to 12.2.2+ to avoid Float16 error #130

Closed KyleVaughn closed 4 months ago

KyleVaughn commented 1 year ago

Usingstd::execution::par_unseq seems to import immintrin.h or other SIMD intrinsic headers. These headers have conflicting definitions with CUDA for 16 bit floating point types. However, it seems this issue was fixed in 12.2.2.

We need to wait for Spack to include this version before we update the spack yaml files.

https://forums.developer.nvidia.com/t/including-cub-header-breakes-compilation-with-gcc-12-and-sse2-or-better/255018

KyleVaughn commented 8 months ago

Update: It seems spack now provides CUDA version 12.3 on the development branch, but v0.20 and v0.21 do not. It is probably safer to wait for an official v0.22 or update to v0.21 before addressing this.

KyleVaughn commented 4 months ago

Change implemented on "format" branch. Plan to merge in the next few days.