[Regression] Floating-point overflow with v1.21

[triplef]

Describe the issue

After upgrading from v1.20.2 to v1.21 we’re seeing the following exception thrown in tests on Windows x64. Windows arm64 is unaffected.

Unhandled exception at 0x00007FFC9F7C2A31 (onnxruntime.dll) in MyApp.exe.9596.dmp: 0xC0000091: Floating-point overflow (parameters: 0x0000000000000000, 0x0000000000001B3A).

onnxruntime.dll!00007ffc9f7c2a31()	Unknown
onnxruntime.dll!00007ffc9f7c0e3d()	Unknown
onnxruntime.dll!00007ffc9f7c57c0()	Unknown
onnxruntime.dll!00007ffc9fe53010()	Unknown
onnxruntime.dll!00007ffc9fe5050b()	Unknown
onnxruntime.dll!00007ffc9fe506e9()	Unknown
onnxruntime.dll!00007ffc9fe4f17a()	Unknown
onnxruntime.dll!00007ffc9fe4ef20()	Unknown
onnxruntime.dll!00007ffc9fe51e59()	Unknown
onnxruntime.dll!00007ffc9f7c6e56()	Unknown
onnxruntime.dll!00007ffc9fcc7251()	Unknown
onnxruntime.dll!00007ffc9fcf7524()	Unknown
onnxruntime.dll!00007ffc9fcf68db()	Unknown
onnxruntime.dll!00007ffc9fcc86d2()	Unknown
onnxruntime.dll!00007ffc9fcc7de9()	Unknown
onnxruntime.dll!00007ffc9fc64908()	Unknown
onnxruntime.dll!00007ffc9fc646d7()	Unknown
onnxruntime.dll!00007ffc9fc644d4()	Unknown
onnxruntime.dll!00007ffc9f68aeb1()	Unknown
onnxruntime.dll!00007ffc9f68c1bf()	Unknown
onnxruntime.dll!00007ffc9f621ead()	Unknown
[Inline Frame] MyApp.exe!Ort::detail::SessionImpl<OrtSession>::Run(const Ort::RunOptions & run_options, const char * const * input_names, const Ort::Value * input_values, unsigned __int64 input_count, const char * const * output_names, Ort::Value * output_values, unsigned __int64 output_count) Line 1002	C++
MyApp.exe!Ort::detail::SessionImpl<OrtSession>::Run(const Ort::RunOptions & run_options, const char * const * input_names, const Ort::Value * input_values, unsigned __int64 input_count, const char * const * output_names, unsigned __int64 output_count) Line 992	C++
...

To reproduce

Unknown

Urgency

No response

Platform

Windows

OS Version

24H2

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

v1.21

ONNX Runtime API

C++

Architecture

X64

Execution Provider

Default CPU

Execution Provider Library Version

No response

[xadupre]

I assume you compared the same model with the same input on the same machine. It is possible to know more about the model you are using?

[triplef]

Correct, same model and everything else.

@omareldeeb can you provide some details about the model?

[github-actions]

This issue has been automatically marked as stale due to inactivity and will be closed in 30 days if no further activity occurs. If further support is needed, please provide an update and/or more details.

[omareldeeb]

Hi, yes, apologies for the delayed response. It’s a custom, unquantized Float32 ONNX model with a single image-like input and four outputs of the same rank and dimensions as the input.

[triplef]

@xadupre please let us know if we can provide any further details here.

[snnn]

This issue has been automatically closed because it has been marked as 'stale' for more than 30 days without activity. If you believe this is still an issue, please feel free to reopen it.