Replace rocm-smi with amd-smi across ROCm build, CI, and docs

[adam360x]

Summary

• fbgemm_gpu/src/topology_utils.cpp: Replace legacy rocm_smi C API with amd_smi. Updates include, macro, init/shutdown calls, and device enumeration (now uses the socket→processor handle pattern required by AMD SMI). The HIP device index → handle mapping (hip_device_to_handle) replaces the old integer-index map.
• ci/utils/gpu_detect.bash: Replace all rocm-smi CLI calls with amd-smi equivalents — vendor detection, GPU model detection (amd-smi static --asic / TARGET_GRAPHICS_VERSION), device count (amd-smi list), and utilization check (amd-smi metric --usage).
• .github/scripts/fbgemm_gpu_test.bash: Replace rocm-smi --showproductname | grep GUID | wc -l with amd-smi list | grep -c "^GPU" for GPU count.
• .github/scripts/utils_system.bash: Replace rocm-smi and rocminfo diagnostic calls with amd-smi.
• .github/scripts/utils_rocm.bash: Replace post-install rocm-smi smoke check with amd-smi.
• cmake/modules/GpuCppLibrary.cmake, fbgemm_gpu/cmake/Hip.cmake: Remove never-defined ROCM_SMI_INCLUDE and ROCRAND_INCLUDE variables (both were always empty and had no effect).
• fbgemm_gpu/docs/: Update InstallationInstructions.rst and TestInstructions.rst to reference amd-smi commands and output format.

rocm-smi has been in maintenance mode since ROCm 7.0 with only critical bug fixes. amd-smi is the supported replacement.

Test plan

• Verify topology_utils.cpp compiles under USE_ROCM=1 with amd-smi-lib installed
• Run source ci/utils/gpu_detect.bash && detect_gpu_vendor && detect_amd_gpu_model on an AMD GPU host
• Confirm amd-smi list | grep -c "^GPU" returns correct GPU count on target hardware

[meta-cla]

Hi @adam360x!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at cla@meta.com. Thanks!

[meta-codesync]

@q10 has imported this pull request. If you are a Meta employee, you can view this in D100035089.

[q10]

@adam360x Looks like one of the runs is failing with:

https://github.com/pytorch/FBGEMM/actions/runs/24141873048/job/70483178549?pr=5597

OSError: /github/home/miniconda/envs/build_binary/lib/python3.14/site-packages/fbgemm_gpu/fbgemm_gpu_py.so: undefined symbol: amdsmi_is_P2P_accessible

I think it may have to do with the host machine itself though.

[adam360x]

@adam360x Looks like one of the runs is failing with:

https://github.com/pytorch/FBGEMM/actions/runs/24141873048/job/70483178549?pr=5597

OSError: /github/home/miniconda/envs/build_binary/lib/python3.14/site-packages/fbgemm_gpu/fbgemm_gpu_py.so: undefined symbol: amdsmi_is_P2P_accessible

I think it may have to do with the host machine itself though.

One of the issues was with linking - fixed.
The other, "upload-manywheel-py3_10-cuda13_2" failed downloading an artifact...

[adam360x]

@q10 looks like ci EADDRINUSE failure. Amd-smi related changes look to be passing as of now