llvm / llvm/offload-test-suite

Texture2D.SampleBias.test.yaml reads the wrong mip level on DirectX Intel

Open
#1,405 1 comment 0 reactions 0 assignees View on GitHub
Dominant language
C++
Stars
18
Forks
39
Avg merge
2d 18h
Merged PRs (30d)
40

Description

The [Texture2D.SampleBias.test] is getting a few mismatched values on intel:

```
Mismatch at element 12: expected=0.5 (0x1.0000000000000p-1), actual=1 (0x1.0000000000000p+0)
Mismatch at element 13: expected=0.5 (0x1.0000000000000p-1), actual=1 (0x1.0000000000000p+0)
Mismatch at element 14: expected=0.5 (0x1.0000000000000p-1), actual=1 (0x1.0000000000000p+0)
```

These are the values in r3, which is calculated with a `-0.1` bias and a `0.1` clamp:
```hlsl
// 4. Bias = -0.1, Clamp = 0.1 -> LOD = 0.1 -> Min(Linear) -> Grey
float4 r3 = Tex.SampleBias(Samp, sample_uv, -0.1, int2(0, 0), 0.1);
```

The expected values here do note an issue with lavapipe, but there doesn't seem to be a correlated issue so it isn't clear to me why:
```
0.5, 0.5, 0.5, 1.0, # Clamp 0.1 -> Min(Linear) -> Grey (Spec: LOD 0.1 is Minification. Lavapipe might fail this and give White)
```

It isn't clear to me if this is a test issue or a driver issue, so it will need further investigation.

[Texture2D.SampleBias.test]: https://github.com/llvm/offload-test-suite/blob/74cbd37/test/Feature/Textures/Texture2D.SampleBias.test.yaml

Contributor guide

No contributing guide indexed for this repository

Research direction

Start with test/Feature/Textures/Texture2D.SampleBias.test.yaml, especially the r3 values and the DirectX Intel mismatches. Compare the expected LOD behavior with the reported lavapipe note and investigate whether the discrepancy is in the test or the driver. Done when the cause is identified and the test or issue has the appropriate correction.

Written by the indexing model from the issue text.

Assessment

Tech stack
yaml
Domain
testing-qa
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
48/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.