search

openxla / xla

A machine learning compiler for GPUs, CPUs, and ML accelerators
Apache License 2.0
2.61k stars 406 forks source link

OpenCL Support. #12413

Open abhaskumarsinha opened 4 months ago

abhaskumarsinha commented 4 months ago

Currently, XLA seems supporting - CPU, GPU and TPU hardware.

Does OpenXLA team plans to use Open Source OpenCL that is somewhat very similar to CUDA? Or does XLA support RISC-V CPUs currently?

OpenCL is an open and very portable standard framework of code and no framework supporting it properly - PyTorch, TensorFlow or JAX makes me think about it.

cheshire commented 4 months ago

OpenCL is not supported and isn't planned. I'm not an expert, but in my understanding OpenCL offers very limited programmability : no TensorCores, very limited datatypes, etc.

abhaskumarsinha commented 4 months ago

Hello @cheshire

OpenCL is not supported and isn't planned. I'm not an expert, but in my understanding OpenCL offers very limited programmability : no TensorCores, very limited datatypes, etc.

Thank you for your reply. I'm not an expert with OpenCL either.

But I'm more concerned about open architectures and software units that are open source. From the best of my search for an alternative to CUDA, I've found only OpenCL as a viable alternative that can be ported to open architectures like RISC-V.

Do you have a good open-source alternative that can be ported to RISC-V like open architectures and can be optimized too?