[DirectML] MVN tests have correctness issues

[hariharans29]

Describe the issue

MVN op tests fail in CI sporadically

Sample "passing" run: https://github.com/microsoft/onnxruntime/actions/runs/23826498144/job/69450582096#step:13:21501

Sample "failing" run: https://github.com/microsoft/onnxruntime/actions/runs/23831205376/job/69484574894#step:13:22000

My suspicion is that they are limited to 4-D inputs that include 0 in the axes parameter - but not 100% sure about this.

To reproduce

Notice CI failure for DML CI pipeline

Urgency

No response

Platform

Windows

OS Version

Windows 11

ONNX Runtime Installation

Built from Source

ONNX Runtime Version or Commit ID

main

ONNX Runtime API

Python

Architecture

X64

Execution Provider

DirectML

Execution Provider Library Version

No response

[fdwr]

Hmm, no similar past issues come to mind:

[ FAILED ] MeanVarianceNormalizationTest.AllAxes (1206 ms)

2: Google Test trace:
2: D:\a\_work\onnxruntime\onnxruntime\onnxruntime\test\unittest_util\checkers.cc(649): provider type: DmlExecutionProvider
2: D:\a\_work\onnxruntime\onnxruntime\onnxruntime\test\unittest_util\base_tester.cc(912): registered execution providers: DmlExecutionProvider
2: D:\a\_work\onnxruntime\onnxruntime\onnxruntime\test\common\tensor_op_test_utils.cc(12): ORT test random seed with fixed pattern tensor generator
2: D:\a\_work\onnxruntime\onnxruntime\onnxruntime\test\providers\cpu\tensor\mean_variance_normalization_test.cc(76): shape: {2,2,2,3}
2: 
2: D:\a\_work\onnxruntime\onnxruntime\onnxruntime\test\unittest_util\checkers.cc(471): error: The difference between cur_expected[i] and cur_actual[i] is 0.27156467735767365, which exceeds tolerance, where
2: cur_expected[i] evaluates to -0.35893502831459045,
2: cur_actual[i] evaluates to -0.087370350956916809, and
2: tolerance evaluates to 4.5893502829130739e-05.