[Performance] Upsample operator regression 1.20.0 → 1.21.0

[junghyunpark2001]

Describe the issue

Summary

Upsample (opset 9) shows regression only for uniform 2x scale + multi-channel images.

Test Results

Test Case	Input Shape	Scales	Kernel Change
upsample_nearest_2d_float32	(2,64,28,28)	[1,1,2,2]	+25.06%
upsample_nearest_1d_feature	(8,128)	[1,2]	-22.71%
upsample_nearest_2d_non_uniform	(1,3,224,224)	[1,1,1.5,2]	-49.17%

Regression Pattern

Regressed (float32):

• Uniform scale (2x2) + multi-channel (64ch) + batch > 1

To reproduce

python script_profiling.py upsample_9_9_upsample_nearest_2d_float32 1.20.0 1.21.0

Archive.zip

Urgency

No response

Platform

Linux

OS Version

Ubuntu 24.04.3 LTS

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.21

ONNX Runtime API

Python

Architecture

X64

Execution Provider

Default CPU

Execution Provider Library Version

No response

Model File

No response

Is this a quantized model?

Yes