ciflow/trunk/194112: Fix integer overflow in max_pool1d shape checks
- PyTorch: 282 events in the last 90 days
- PyTorch: 279th Release in the last 90 days
- Previous: earlier the same day · ciflow/inductor/195373: Support complex dropout by allocating a real-valued mask
What happened
torch.max_pool1d segfaults for several out-of-domain integer arguments. Two overflows in the shape math remain after the at::divup fix this is stacked on: div_rtn() computed the floor quotient in int regardless of T , although every call site passes div_rtn . The truncated remainder gets the wrong sign, so the floor correction is skipped and the quotient wraps: for #73190 the true output size -2147483694 became +214…
Summary assembled by rule from the sources below