backend_gpu.py crashes on AMD ROCm systems due to hardcoded nvidia-smi call

[adelejjeh]

Description

BackendGPU.get_backend_info() in src/xpu_perf/micro_perf/backends/GPU/backend_gpu.py calls subprocess.run(['nvidia-smi', ...]) unconditionally. On AMD ROCm systems where nvidia-smi is not present, this raises an unhandled FileNotFoundError and crashes the benchmark before any workloads run.

Steps to Reproduce

1. Run xpu-perf on a system with AMD GPUs using ROCm 7.2.2 (PyTorch with torch.version.hip set)
2. Launch any GPU benchmark: python projects/micro_perf/launch.py --backend GPU --device 0
3. Crash occurs in get_backend_info():

   FileNotFoundError: [Errno 2] No such file or directory: 'nvidia-smi'
   

Suggested Fix

Detect the platform and use rocm-smi on AMD ROCm systems to retrieve driver version information, rather than unconditionally calling nvidia-smi.