intel / intel/torch-xpu-ops

[B60][stock PyTorch][XPU] Tensor closeness mismatches in Inductor dynamic/select tests

Open
#4,795 0 comments 0 reactions 0 assignees View on GitHub
hw: BMG test: ut
Dominant language
Python
Stars
113
Forks
128
Avg merge
5d 9h
Merged PRs (30d)
112

Description

## Summary

B60 has stock PyTorch XPU numerical mismatches reported as tensor closeness failures. The same tests passed on PVC.

## Failure Category

Tensor closeness mismatch.

Representative messages:

```text
AssertionError: Tensor-likes are not close
Tensor-likes are not close
```

## Scope

- Test suite: stock PyTorch
- Platform comparison: B60 failed; PVC passed
- Affected rows observed: 2

## Failures

```text
test/inductor/test_torchinductor_dynamic_shapes.py::TestInductorDynamicXPU::test_coalescing_analysis_sympy_is_constant_xpu
AssertionError: Tensor-likes are not close

test/inductor/test_select_algorithm.py::TestExternKernelCaller::test_extern_kernel_benchmark_valid_timing
Tensor-likes are not close
```

## Expected Behavior

The compiled XPU output should match eager/reference output within test tolerance, consistent with PVC behavior.

## Notes

The two failures are in different Inductor areas. They are grouped here because the visible failure mode is the same; they can be split if triage identifies separate root causes.

## Full Case List

Total cases: 2

1. `stock pytorch` | `test/inductor/test_torchinductor_dynamic_shapes.py::TestInductorDynamicXPU::test_coalescing_analysis_sympy_is_constant_xpu` | `failed`
- `AssertionError: Tensor-likes are not close`
2. `stock pytorch` | `test/inductor/test_select_algorithm.py::TestExternKernelCaller::test_extern_kernel_benchmark_valid_timing` | `failed`
- `Tensor-likes are not close`

## Reproducer / Environment Setup

```bash
docker run -it -e TZ=Asia/Shanghai --device=/dev/mem --device=/dev/dri --group-add video --privileged -v $(realpath ${HOME}):/home/jenkins --shm-size=8g intelgpu/ubuntu-26.04-rolling:26.18 bash

apt update
apt upgrade -y

curl -LsSf https://astral.sh/uv/install.sh | sh
source $HOME/.local/bin/env
uv venv myvenv --python 3.10 --clear
source myvenv/bin/activate
uv pip install pip wheel "setuptools<81"
uv pip install torch==2.13.0 torchaudio torchvision --index-url https://download.pytorch.org/whl/xpu

pytorch_commit="$(python -c 'import torch; print(torch.version.git_version)')"
git clone https://github.com/pytorch/pytorch
cd pytorch
git checkout ${pytorch_commit}

uv pip install -r .ci/docker/requirements-ci.txt
uv pip install -U typing_extensions
uv pip install pytest pytest-timeout pytest-xdist pytest-rerunfailures

python test/inductor/test_torchinductor_dynamic_shapes.py -k 'test_coalescing_analysis_sympy_is_constant_xpu'
python test/inductor/test_select_algorithm.py -k 'test_extern_kernel_benchmark_valid_timing'
```

Contributor guide

Open the contributing guide

Research direction

Reproduce the two cases with the provided environment and commands: test/inductor/test_torchinductor_dynamic_shapes.py::TestInductorDynamicXPU::test_coalescing_analysis_sympy_is_constant_xpu and test/inductor/test_select_algorithm.py::TestExternKernelCaller::test_extern_kernel_benchmark_valid_timing. Compare compiled XPU output with eager/reference output and investigate the tensor closeness failures separately; done means both tests pass within tolerance on B60.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
compilers, machine-learning, testing
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.