Add get-rocm-devices for AMD GPUs

[Henryfzh]

The command to run the script is:

cm run script --tags=get,rocm-devices

The script first check if hip-python is installed, if not install it from https://test.pypi.org/simple.
After finishing run the script will write to file tmp-run.out and print the properties of the GPUs to the terminal like this:

GPU 0:
GPU Device ID: 0000:e3:00.0
GPU Name: 0
GPU compute capability: 9.0
ROCM driver version: 60241134
ROCM runtime version: 60241134
Global memory (GiB): 63.984375
Max clock rate: 1700.0 MHz
Total amount of shared memory per block (Bytes): 65536
Total number of registers available per block: 65536
Warp size: 64
Maximum number of threads per multiprocessor: 2048
Maximum number of threads per block: 1024
Max dimension size of a thread block X: 1024
Max dimension size of a thread block Y: 1024
Max dimension size of a thread block Z: 1024
Max dimension size of a grid size X: 2147483647
Max dimension size of a grid size Y: 65536
Max dimension size of a grid size Z: 65536
GPU 1:
GPU Device ID: 0000:c3:00.0
GPU Name: 1
GPU compute capability: 9.0
ROCM driver version: 60241134
ROCM runtime version: 60241134
Global memory (GiB): 63.984375
Max clock rate: 1700.0 MHz
Total amount of shared memory per block (Bytes): 65536
Total number of registers available per block: 65536
Warp size: 64
Maximum number of threads per multiprocessor: 2048
Maximum number of threads per block: 1024
Max dimension size of a thread block X: 1024
Max dimension size of a thread block Y: 1024
Max dimension size of a thread block Z: 1024
Max dimension size of a grid size X: 2147483647
Max dimension size of a grid size Y: 65536
Max dimension size of a grid size Z: 65536

[github-actions]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅

[arjunsuresh]

Instead of CUDA it must be ROCM right? This is the prefix for the ENV keys which are to be exported by this CM script. Same for the state.

[arjunsuresh]

Thank you for the PR