[Performance] Performance regression in GRU operator between v1.20.0 and v1.21.0

[junghyunpark2001]

Describe the issue

Description

We observed performance regressions in the GRU (Gated Recurrent Unit) operator across multiple configurations between ONNXRuntime v1.20.0 and v1.21.0. The regression is most severe for single time step configurations with +20.77% kernel slowdown.

Affected Operator

GRU

• Opset Version: 21
• Configurations: Various (forward, bidirectional, different batch/sequence sizes)
• Regression: +7% to +20.77% kernel slowdown

To reproduce

python script_profiling.py gru_deepcpugruop_21_gru_forward_single_time_step 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