search

ROCm / clr

MIT License
89 stars 47 forks source link

missing atomicMin/Max for signed long long in HIP (available in CUDA) #2

Closed Epliz closed 1 year ago

Epliz commented 1 year ago

Hi,

As pointed out at https://github.com/ROCm-Developer-Tools/hipamd/issues/65 , there is not version of the atomicMin/Max methods for signed long long arguments. These variants are available in CUDA as seen at https://docs.nvidia.com/cuda/cuda-c-programming-guide/#atomicmin (as well as in OpenCL on AMD GPUs). Therefore it seems like they should be added.

Best regards, Epliz

jatinx commented 1 year ago

Thanks for reporting it. Will look at it.

cjatin commented 1 year ago

Fixed via: https://github.com/ROCm-Developer-Tools/clr/commit/be02e1e524b0592cd34d25994e8d3b3d6094bed0