intel / intel/torch-xpu-ops

[upstream_ut] test_zero_dtypes_without_fill_kernel_xpu: fill_xpu not implemented for Bits/Float4

Open
#5,378 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
113
Forks
128
Avg merge
5d 13h
Merged PRs (30d)
107

Description

## Description
test_zero_dtypes_without_fill_kernel_xpu_* fail for bits1x8, bits2x4, bits4x2,
bits8, bits16 and float4_e2m1fn_x2. zero_() on a view of these dtypes dispatches
to fill_xpu, which is not implemented for these types.

## Error log
```
File "test/test_torch.py", line 171, in test_zero_dtypes_without_fill_kernel
raw.view(dtype).zero_()
NotImplementedError: "fill_xpu" not implemented for 'Bits16'/'Bits8'/'Bits1x8'/'Bits2x4'/'Bits4x2'/'Float4_e2m1fn_x2'
```

## Reproduce
```
python test/test_torch.py TestTorchDeviceTypeXPU.test_zero_dtypes_without_fill_kernel_xpu_bits16
```

Contributor guide

Open the contributing guide

Research direction

Start by running test/test_torch.py TestTorchDeviceTypeXPU.test_zero_dtypes_without_fill_kernel_xpu_bits16 and review test_zero_dtypes_without_fill_kernel in that file. Trace the zero_() dispatch to fill_xpu for the listed Bits and Float4 dtypes. Done means the XPU test passes for bits1x8, bits2x4, bits4x2, bits8, bits16, and float4_e2m1fn_x2 without the NotImplementedError.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
backend, testing-qa
Issue type
Bug
Difficulty
3/5
Estimated time
1-2 days
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
68/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.