First a bit of context. I'm trying to build onnxruntime for raspberry pi using cross-compilation (instructions here). The onnxruntime package depends on pytorch-cpuinfo and fetches and builds it as part of the build process.
Would love to hear if I'm doing something wrong. Or if I can submit a PR for this to allow to override the target architecture from the environment variables.
Hi!
First a bit of context. I'm trying to build onnxruntime for raspberry pi using cross-compilation (instructions here). The onnxruntime package depends on pytorch-cpuinfo and fetches and builds it as part of the build process.
I'm using this command.
This triggers the following error:
My take on this is that pytorch-cpuinfo errorneously tries to compile for x86 (the host for the cross-compile).
Looking at the CMakeList.txt in this project, I think the culprit is that it always assumes that the host architecture is also the target architecture: https://github.com/pytorch/cpuinfo/blob/3dc310302210c1891ffcfb12ae67b11a3ad3a150/CMakeLists.txt#L59
Would love to hear if I'm doing something wrong. Or if I can submit a PR for this to allow to override the target architecture from the environment variables.