frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-20 21:14:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

used guard_or_false instead of guard_size_oblivious in is_int_or_symint (#154167)

Browse Source 
This is a short circuit, that we should not fail on. Before this PR we would not fail on u0, u0+u1,
only if they are size like.  but we will fail on u0-u1.. etc for no need.
guard_or_false seems appropriate for that reason.

This was added in https://github.com/pytorch/pytorch/pull/122145 there was no unit tests for me to verify
why it was added, i could not repo using the associated issue , the example does not work.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/154167
Approved by: https://github.com/bobrenjc93
ghstack dependencies: #154154, #154164
This commit is contained in:
Laith Sakka
2025-05-22 16:32:46 -07:00
committed by PyTorch MergeBot
parent 1da2cc52bc
commit ab5137b048
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
torch/csrc/utils/python_arg_parser.cpp
View File

@ -889,7 +889,7 @@ static bool is_int_or_symint(PyObject* obj) {
// for regular tensors it's redundant with the test below.
if (THPVariable_Check(obj)) {
auto& var = THPVariable_Unpack(obj);
if (TORCH_GUARD_SIZE_OBLIVIOUS(var.sym_numel().sym_eq(1)) &&
if (TORCH_GUARD_OR_FALSE(var.sym_numel().sym_eq(1)) &&
at::isIntegralType(var.dtype().toScalarType(), /*include_bool*/ true)) {
return true;
}