pytorch - 💡(How to fix) Fix RocM: Assumed realtime clock rate [1 comments, 2 participants]

[302/895] Building HIPCC object caffe2/CMakeFiles/torch_hip.dir/__/aten/src/ATen/native/hip/torch_hip_generated_group_norm_kernel.hip.o In file included from /home/ste/code/pytorch/aten/src/ATen/native/hip/group_norm_kernel.hip:8: In file included from /opt/rocm/include/thrust/tuple.h:33: In file included from /opt/rocm/include/thrust/detail/config.h:22: In file included from /opt/rocm/include/thrust/detail/config/config.h:23: In file included from /opt/rocm/include/thrust/detail/config/libcxx.h:44: In file included from /opt/rocm/include/cuda/std/version:31: In file included from /opt/rocm/include/cuda/std/detail/__config:67: /opt/rocm/include/cuda/std/detail/libcxx/include/__config:371:12: warning: Assuming 100 MHz realtime clock rate (TSC) for gfx1100/gfx1101 (according to the RDNA3 ISA). Timing-related APIs (e.g., chrono) or sleep instructions may behave incorrectly! [-W#warnings] 371 | # warning Assuming 100 MHz realtime clock rate (TSC) for gfx1100/gfx1101 (according to the RDNA3 ISA). Timing-related APIs (e.g., chrono) or sleep instructions may behave incorrectly! | ^ 1 warning generated when compiling for gfx1100. ninja: build stopped: subcommand failed. error: subprocess-exited-with-error

× Building editable for torch (pyproject.toml) did not run successfully. │ exit code: 1 ╰─> No available output.

note: This error originates from a subprocess, and is likely not a problem with pip. full command: /home/ste/code/pytorch/.venv/bin/python /home/ste/code/pytorch/.venv/lib/python3.14/site-packages/pip/_vendor/pyproject_hooks/_in_process/_in_process.py build_editable /tmp/tmp5ec9a96l cwd: /home/ste/code/pytorch Building editable for torch (pyproject.toml) ... error ERROR: Failed building editable for torch Failed to build torch error: failed-wheel-build-for-install

× Failed to build installable wheels for some pyproject.toml based projects ╰─> torch

TL;DR

The compilation of PyTorch from source for an AMD gfx1103 GPU is failing due to an assumed 100 MHz realtime clock rate warning that stops the build process.

Guidance

• The warning is related to the gfx1100/gfx1101 GPU architecture, and the build process is treating the gfx1103 as a gfx1100.
• To temporarily bypass the warning and continue the compilation, you can try setting the WARNINGS_AS_ERRORS environment variable to OFF or modifying the CMake configuration to treat warnings as non-fatal errors.
• Verify that the PYTORCH_ROCM_ARCH environment variable is correctly set to gfx1103 to ensure the build process is using the correct architecture.
• Consider modifying the HSA_OVERRIDE_GFX_VERSION environment variable to a version that supports the gfx1103 GPU, if available.

Example

No code snippet is provided as the issue is related to environment variables and build configuration.

Notes

The provided solution is a temporary workaround to bypass the warning and continue the compilation. The actual fix may require updating the ROCm or PyTorch versions to support the gfx1103 GPU architecture.

Recommendation

Apply workaround: temporarily bypass the warning by setting WARNINGS_AS_ERRORS to OFF or modifying the CMake configuration, as the issue is likely due to the assumed clock rate and not a critical error.