Error on installation

[Zernez]

🐛 Describe the bug

Hello, after "python setup.py install" the script shows this error and the installation fails:

fatal error: error in backend: Cannot select: intrinsic %llvm.amdgcn.ds.bpermute
clang-14: error: clang frontend command failed with exit code 70 (use -v to see invocation)
AMD clang version 14.0.0 (https://github.com/RadeonOpenCompute/llvm-project roc-5.2.3 22324 d6c88e5a78066d5d7a1e8db6c5e3e9884c6ad10e)
Target: x86_64-unknown-linux-gnu
Thread model: posix
InstalledDir: /opt/rocm/hip/../llvm/bin
clang-14: note: diagnostic msg: Error generating preprocessed source(s).
CMake Error at torch_hip_generated_cub-RadixSortPairs.hip.o.cmake:200 (message):
Error generating file
/home/ferna/pytorch/build/caffe2/CMakeFiles/torch_hip.dir/__/aten/src/ATen/hip/./torch_hip_generated_cub-RadixSortPairs.hip.o

Something is missing? Thanks

Versions

roc-5.2.3
python 3.9

cc @malfet @seemethere @jeffdaily @sunway513 @jithunnair-amd @ROCmSupport @KyleCZH

[malfet]

Looks like ROCm compiler crash to me...

[amdfaa]

This is Faa Diallo from the ROCm team. Can you provide the output from the following script for info about the environment: https://github.com/pytorch/pytorch/blob/master/torch/utils/collect_env.py

Also have you tried using the ROCm release containers?
For 5.2.3, this page might be helpful:
https://docs.amd.com/bundle/ROCm-Release-Notes-v5.2.3/page/ROCm_v5.2.3_Release_Notes.html

[Zernez]

Thanks for reply to me. I run the script:

PyTorch version: N/A
Is debug build: N/A
CUDA used to build PyTorch: N/A
ROCM used to build PyTorch: N/A
OS: Ubuntu 20.04.4 LTS (x86_64)
GCC version: (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0
Clang version: 10.0.0-4ubuntu1
CMake version: version 3.19.6
Libc version: glibc-2.31
Python version: 3.9.12 (main, Apr 5 2022, 06:56:58) [GCC 7.5.0] (64-bit runtime)
Python platform: Linux-5.15.0-46-generic-x86_64-with-glibc2.31
Is CUDA available: N/A
CUDA runtime version: Could not collect
GPU models and configuration: Could not collect
Nvidia driver version: Could not collect
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: N/A
Versions of relevant libraries:
[pip3] mypy-extensions==0.4.3
[pip3] numpy==1.21.5
[pip3] numpydoc==1.2
[conda] blas 1.0 mkl
[conda] mkl 2021.4.0 h06a4308_640
[conda] mkl-include 2022.0.1 h06a4308_117
[conda] mkl-service 2.4.0 py39h7f8727e_0
[conda] mkl_fft 1.3.1 py39hd3c417c_0
[conda] mkl_random 1.2.2 py39h51133e4_0
[conda] numpy 1.21.5 py39he7a7128_1
[conda] numpy-base 1.21.5 py39hf524024_1
[conda] numpydoc 1.2 pyhd3eb1b0_0

And with "apt show rocm-libs -a":

Package: rocm-libs
Version: 5.2.3.50203-109
Priority: optional
Section: devel
Maintainer: ROCm Libs Support rocm-libs.support@amd.com
Installed-Size: 13.3 kB
Depends: hipblas (= 0.51.0.50203-109), hipfft (= 1.0.8.50203-109), hipsolver (= 1.4.0.50203-109), hipsparse (= 2.1.0.50203-109), miopen-hip (= 2.17.0.50203-109), rccl (= 2.12.12.50203-109), rocalution (= 2.0.2.50203-109), rocblas (= 2.44.0.50203-109), rocfft (= 1.0.17.50203-109), rocrand (= 2.10.9.50203-109), rocsolver (= 3.18.0.50203-109), rocsparse (= 2.2.0.50203-109), rocm-core (= 5.2.3.50203-109), hipblas-dev (= 0.51.0.50203-109), hipcub-dev (= 2.10.12.50203-109), hipfft-dev (= 1.0.8.50203-109), hipsolver-dev (= 1.4.0.50203-109), hipsparse-dev (= 2.1.0.50203-109), miopen-hip-dev (= 2.17.0.50203-109), rccl-dev (= 2.12.12.50203-109), rocalution-dev (= 2.0.2.50203-109), rocblas-dev (= 2.44.0.50203-109), rocfft-dev (= 1.0.17.50203-109), rocprim-dev (= 2.10.9.50203-109), rocrand-dev (= 2.10.9.50203-109), rocsolver-dev (= 3.18.0.50203-109), rocsparse-dev (= 2.2.0.50203-109), rocthrust-dev (= 2.10.9.50203-109), rocwmma-dev (= 0.7.0.50203-109)
Homepage: https://github.com/RadeonOpenCompute/ROCm
Download-Size: 986 B
APT-Manual-Installed: yes
APT-Sources: https://repo.radeon.com/rocm/apt/5.2.3 ubuntu/main amd64 Packages
Description: Radeon Open Compute (ROCm) Runtime software stack

[amdfaa]

Can you try the following before the running the setup script:

python tools/amd_build/build_amd.py
export CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname $(which conda))/../"}
I'd advise you to use the 5.2.3 release container mentioned in my above comment.
Let me know if that works!

[Zernez]

Is not working again... Thanks anyways

I've tried to use container but another error occur:
"docker: Error response from daemon: could not select device driver "" with capabilities: [[gpu]]."

My gpu is R9 290x that have from the official ROCm list is: "GPUs that are enabled, but which AMD does not officially support". I don't know what it is mean.

[hongxiayang]

Please check the latest documentations about supported GPUs: https://rocm.docs.amd.com/en/latest/release/gpu_os_support.html