ONNX Runtime CPU BitShift incorrectly handles uint64 shift amount == 64

[ALinrunrun]

Describe the issue

ONNX Runtime CPUExecutionProvider produces an incorrect result for ONNX BitShift when the input type is uint64 and the shift amount is 64.

In a minimal model with a single BitShift node using direction="RIGHT", shifting [1000, 255, 1, 42] by [64, 64, 64, 64] returns the original input values instead of [0, 0, 0, 0].

This behavior is consistent with native 64-bit shift masking on x86, where a shift amount of 64 may effectively behave like 0. The issue reproduces with CPUExecutionProvider and ORT_DISABLE_ALL. With ORT_ENABLE_ALL, this constant-input case may be hidden by constant folding.

To reproduce

import numpy as np
import onnx
from onnx import TensorProto, helper
import onnxruntime as ort

vals = np.array([[1000, 255, 1, 42]], dtype=np.uint64)
shifts = np.array([[64, 64, 64, 64]], dtype=np.uint64)
expected = np.zeros_like(vals)

X = helper.make_tensor_value_info("x", TensorProto.UINT64, list(vals.shape))
N = helper.make_tensor_value_info("n", TensorProto.UINT64, list(shifts.shape))
Y = helper.make_tensor_value_info("y", TensorProto.UINT64, list(vals.shape))

node = helper.make_node("BitShift", ["x", "n"], ["y"], direction="RIGHT")
graph = helper.make_graph([node], "bitshift64", [X, N], [Y])
model = helper.make_model(graph, opset_imports=[helper.make_opsetid("", 11)])
model.ir_version = 8
onnx.checker.check_model(model)

feed = {"x": vals, "n": shifts}

so = ort.SessionOptions()
so.graph_optimization_level = ort.GraphOptimizationLevel.ORT_DISABLE_ALL

sess = ort.InferenceSession(
    model.SerializeToString(),
    sess_options=so,
    providers=["CPUExecutionProvider"],
)

out = sess.run(None, feed)[0]

print(f"Input values: {vals[0].tolist()}")
print(f"Shift amount: {shifts[0].tolist()}")
print(f"Expected:     {expected[0].tolist()}")
print(f"ORT output:   {out[0].tolist()}")

PASS = np.array_equal(out, expected)
print(f"PASS={PASS}")

Urgency

Expected output

Input values: [1000, 255, 1, 42]
Shift amount: [64, 64, 64, 64]
Expected:     [0, 0, 0, 0]
ORT output:   [0, 0, 0, 0]
PASS=True

Actual output

Input values: [1000, 255, 1, 42]
Shift amount: [64, 64, 64, 64]
Expected:     [0, 0, 0, 0]
ORT output:   [1000, 255, 1, 42]
PASS=False

Platform

Linux

OS Version

Linux-6.17.0-20-generic-x86_64-with-glibc2.39

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.25.1

ONNX Runtime API

Python

Architecture

X86

Execution Provider

Default CPU

Execution Provider Library Version

No response