[bug] MaxUnpool2d with invalid indices (-1) crashes process instead of raising Python error

[tinywisdom]

Summary

When nn.MaxUnpool2d is called with invalid indices (e.g., all values set to -1), OneFlow does not raise a Python error. Instead, the process aborts due to a C++ CHECKfailure.

Code to reproduce bug

import oneflow as flow
import oneflow.nn as nn
import numpy as np

device = "cpu"
flow.manual_seed(0)
np.random.seed(0)

class M(nn.Module):
    def __init__(self):
        super().__init__()
        self.unpool = nn.MaxUnpool2d(kernel_size=2)

    def forward(self, x, indices):
        return self.unpool(x, indices)

def main():
    m = M().to(device)
    x = flow.tensor(np.random.rand(1, 1, 2, 2), dtype=flow.float32, device=device)

    # Invalid indices: all set to -1
    bad_idx = flow.full_like(x.to(flow.int64), -1)

    print("about to call unpool with invalid indices = -1")
    y = m(x, bad_idx)

    # Force sync to trigger backend error
    print("forcing sync via .numpy() ...")
    _ = y.numpy()

if __name__ == "__main__":
    main()

System Information

• OS: Ubuntu 22.04.4 LTS (x86_64)
• OneFlow version : 1.0.0.dev20250921+cpu
• Python version: 3.10.16