rocm-arch / tensorflow-rocm

tensorflow-rocm AUR package
17 stars 12 forks source link

Tensorflow 2.6.0 build failure due to segfault #33

Closed afzalarsalan closed 1 year ago

afzalarsalan commented 3 years ago

The build for Tensorflow 2.6.0 fails with the following

compile tensorflow/core/kernels/mlir_generated/atan2_gpu_f16_f16_kernel_generator_kernel.o failed: (Segmentation fault).

This happens towards the end of the compilation phase

justinkb commented 2 years ago

I bumped it to 2.6.2 and it segfaulted on tensorflow/core/kernels/mlir_generated/is_finite_gpu_f16_i1_kernel_generator_kernel.o for me, likely the very same issue

edit: see https://github.com/tensorflow/tensorflow/issues/50345

riaqn commented 2 years ago

@justinkb does it work for you? I use the workaround mentioned in the link and it doesn't solve the problem.

acxz commented 2 years ago

Can you guys update your error now that the package has been updated to 2.8.0?

mindless728 commented 2 years ago

Currently there are two issues (as of today)

  1. The bazel install from arch is version 5.1.0 and the tensorflow build requires < 5.0.0
  2. Can't download the link for tensorflow, maybe it's wrong?

WARNING: Download from https://storage.googleapis.com/mirror.tensorflow.org/github.com/tensorflow/runtime/archive/c3e082762b7664bbc7ffd2c39e86464928e27c0c.tar.gz failed: class com.google.devtools.build.lib.bazel.repository.downloader.UnrecoverableHttpException GET returned 404 Not Found

acxz commented 1 year ago

Closing this issue as a stale build issue. If you have further issues please open up another issue. Sorry @afzalarsalan @justinkb @riaqn