As I know, the HIP compiler generates native GCN ISA which can be directly executed on the AMD GPUs. So, there should not be any performance penalty for using HIP.
Are any of the other languages more or less performant ?
By default, the hipcc compiler invokes the hcc compiler (Home · RadeonOpenCompute/hcc Wiki · GitHub ). HC C++ API, which is the default C++ compute API for the hcc compiler, may be an alternative (it lacks the CUDA portability though).
By the way, ROCm Github site is the best place to report/post any issue/query related to ROCm. For any query related to HIP, please post here: Issues · ROCm-Developer-Tools/HIP · GitHub
As I know, the HIP compiler generates native GCN ISA which can be directly executed on the AMD GPUs. So, there should not be any performance penalty for using HIP.
Here is a great discussion which explains it in more detail: OpenCL support · Issue #90 · ROCm-Developer-Tools/HIP · GitHub
By default, the hipcc compiler invokes the hcc compiler (Home · RadeonOpenCompute/hcc Wiki · GitHub ). HC C++ API, which is the default C++ compute API for the hcc compiler, may be an alternative (it lacks the CUDA portability though).
By the way, ROCm Github site is the best place to report/post any issue/query related to ROCm. For any query related to HIP, please post here: Issues · ROCm-Developer-Tools/HIP · GitHub
Thanks.