Build AMD GPU operator images for ppc64le

[Amulyam24]

Motivation

This PR aims to expand the arch support for the operator images

Technical Details

Added multi-arch support with Docker buildx

Test Plan

Tested the image along with the device plugin and KKM on a ppc64le test cluster

Test Result

Works as expected. Verified by running a sample vLLM pod on 1 AMD GPU

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[yansun1996]

Hi @Amulyam24 thanks for raising the PR, could you please share more details about why you want to expand the image build to ppc64le ?

for example:

1. how's the cluster look like ? based on which k8s version or k8s variant ?
2. what is the host OS , kernel, amdgpu, ROCm version ?
3. how large is the cluster you are going to use ?

Thanks

[Amulyam24]

Hi @Amulyam24 thanks for raising the PR, could you please share more details about why you want to expand the image build to ppc64le ?

Hi @yansun1996, This PR has been raised in a effort to support heterogenous computing with AMD GPUs for IBM Power(ppc64le).
More context: ROCm/TheRock#5518

1. how's the cluster look like ? based on which k8s version or k8s variant ?

Deployed the GPU operator on a single node K8s cluster on a machine with 2 AMD MI210 GPUS
K8s version: 1.34.8

2. what is the host OS , kernel, amdgpu, ROCm version ?

Host OS: Linux ppc64le
Kernel: Patched with amd64 support
ROCm Version: 7.12.0

3. how large is the cluster you are going to use ?

For trials, we have used a single node k8s cluster but there are use cases where it would be on larger clusters.

Let me know if you have further questions, Thanks!