[vmap] symintify alias and squeeze

[kshitij12345]

Following tests now pass (both ops call into alias on certain paths)

PYTORCH_TEST_WITH_DYNAMO=1 pytest test/functorch/test_vmap.py -k test_squeeze -v
PYTORCH_TEST_WITH_DYNAMO=1 pytest test/functorch/test_vmap.py -k test_conj -v

NOTE: Ideally, this symint version should work with non symint version as well but that would mean changes at multiple places. Wanted to get a review for this fix before-hand.

Other sites which use the IntArrayRef overload.

pytorch/aten/src/ATen/native/TensorShape.cpp

Lines 1707 to 1713 in 5f56c4f

	Tensor _reshape_alias(const Tensor& self, IntArrayRef sizes, IntArrayRef strides) {
	// This is only used by `reshape` in cases where it would otherwise have dispatched
	// to `view`. This removes the overhead of calling `view` which duplicates some of
	// the work that's already been done (`infer_size_dv` and `computeStride`).
	return alias_with_sizes_and_strides(self, sizes, strides);
	}

view_impl is called from view and _unsafe_view.

pytorch/aten/src/ATen/native/TensorShape.cpp

Lines 3295 to 3306 in 5f56c4f

	inline Tensor view_impl(const Tensor& self, IntArrayRef size) {
	at::DimVector inferred_size = at::infer_size_dv(size, self.numel());
	auto stride = at::detail::computeStride(self.sizes(),
	self.strides(),
	inferred_size);
	TORCH_CHECK(stride.has_value(), "view size is "
	"not compatible with input tensor's size and stride (at least one dimension"
	" spans across two contiguous subspaces). Use .reshape(...) instead.");
	return alias_with_sizes_and_strides(self, inferred_size, *stride);
	}

cc @ezyang

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/107577

• 📄 Preview Python docs built from this PR
• 📄 Preview C++ docs built from this PR
• ❓ Need help or want to give feedback on the CI? Visit the bot commands wiki or our office hours

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit 2abdba4 with merge base d0f8ee4 ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[zou3519]

Is there a reason why we can't change the ArrayRef<int64_t> and SmallVector<int64_t> template above to use this line?

[kshitij12345]

I just forgot about that! Thanks!

[kshitij12345]

Actually, that won't work because there are two overloads for set_sizes_and_strides

1. void set_sizes_and_strides(SymIntArrayRef sizes, SymIntArrayRef strides, optional<c10::SymInt> storage_offset)
2. void set_sizes_and_strides(IntArrayRef new_size, IntArrayRef new_stride, optional<int64_t> storage_offset)

So, all inputs have to be symbolic or concrete, we can't pass self.sym_storage_offset() to second overload.

[kshitij12345]

Ping @zou3519

[kshitij12345]

Ping @zou3519

[kshitij12345]

@pytorchbot merge

[pytorchmergebot]

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here