PyTorch CUDA adaptive max-pool backward can corrupt device memory
Unchecked out-of-range indices in the backward kernel become atomic-write destinations, unlike the CPU path which raises an error.
By tensorUnchecked out-of-range indices in the backward kernel become atomic-write destinations, unlike the CPU path which raises an error.
By tensorCoordinate-descent autotune can pick an R0_BLOCK smaller than workspace allocation assumes, overrunning the buffer under torch.compile.
By tensor