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

[do not revert] Compute contiguity symbolically to avoid dde, and introduce c++ sym_is_contiguous (#155590)

Browse Source 
When we compute contiguity for a tensor with dynamic shapes we first:
1) Try to compute it without guarding.
2) If all shapes hinted, compute it with potentially adding guards.
3) if any input is not hinted, compute it symbolically.

sym_is_contiguous return a SymBool that is then either evaluated or guard_or_false can be called
on it to avoid data dependent errors.

ex:
 bool is_contiguous = input.sym_is_contiguous().guard_or_false(__FILE__, __LINE__);
is_contiguous_or_false is a helper function that does that.

In this PR I only handle default contiguity, will follow up with changes for other formats like  channel_last .
We use this patter in this PR for several locations to avoid DDEs.
Differential Revision: [D77183032](https://our.internmc.facebook.com/intern/diff/D77183032)

Pull Request resolved: https://github.com/pytorch/pytorch/pull/155590
Approved by: https://github.com/ezyang
This commit is contained in:
Laith Sakka
2025-07-01 09:49:10 -07:00
committed by PyTorch MergeBot
parent 22edb457c9
commit d0a9629435
34 changed files with 390 additions and 114 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
test/test_proxy_tensor.py
View File

@ -1370,8 +1370,8 @@ def forward(self, crop_camera_1, mask_1):
view_1 = torch.ops.aten.view.default(expand_1, [sym_size_int, sym_size_int_1, sym_size_int_2]); expand_1 = sym_size_int_1 = sym_size_int_2 = None
bmm = torch.ops.aten.bmm.default(view, view_1); view = view_1 = None
view_2 = torch.ops.aten.view.default(bmm, [sym_size_int, 3, 3]); bmm = None
mul_6 = sym_size_int * 3
view_3 = torch.ops.aten.view.default(view_2, [mul_6, 3]); view_2 = mul_6 = None
mul_9 = sym_size_int * 3
view_3 = torch.ops.aten.view.default(view_2, [mul_9, 3]); view_2 = mul_9 = None
mm = torch.ops.aten.mm.default(view_3, eye); view_3 = eye = None
_unsafe_view = torch.ops.aten._unsafe_view.default(mm, [sym_size_int, 3, 3]); mm = sym_size_int = None
index_put_ = torch.ops.aten.index_put_.default(crop_camera_1, [mask_1], _unsafe_view); crop_camera_1 = mask_1 = _unsafe_view = index_put_ = None