Fix PrecomputedValues::bindTensorMetaData for DID loop split

[wujingyue]

in the same way as ExpressionEvaluator::bindTensorDomain and several other places. Caveat: having to fix multiple places in the same way probably indicates a pre-existing duplication of logic.

Fixes #3817

[wujingyue]

!test

[github-actions]

Description

• Fix PrecomputedValues::bindTensorMetaData for DID loop split

• Update tensor size handling in bindTensorMetaData

• Modify test case test_sdpa_loop_split for better parameterization

• Add unshardedSizes utility function usage

---

Changes walkthrough 📝

	Relevant files
Bug fix	evaluator_common.cpp Update tensor size handling in bindTensorMetaData                csrc/evaluator_common.cpp Include multidevice/utils.h Update bindTensorMetaData to use unshardedSizes for tensor sizes Simplify device dimension handling in bindTensorMetaData +6/-9
Enhancement	test_multidevice.py Update test_sdpa_loop_split for better parameterization    tests/python/test_multidevice.py Update test_sdpa_loop_split to use parameterized tensor shapes Reorganize tensor shape definition in test_sdpa_loop_split +6/-5

---

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

🧪 PR contains tests

⚡ Recommended focus areas for review Possible Issue The new code introduces a dependency on multidevice/utils.h which was not previously included. Ensure that this inclusion is necessary and does not introduce any unintended side effects. #include <multidevice/utils.h> Logic Change The logic for binding tensor metadata has been changed to use unshardedSizes and logical_sizes. Verify that this change does not alter the behavior for non-multidevice scenarios. std::vector<int64_t> logical_sizes = unshardedSizes(tv, tensor.sizes()); for (const auto dim : c10::irange(static_cast<int64_t>(logical_domain.size()))) { IterDomain* id = logical_domain[dim]; const auto dim_size = logical_sizes.at(dim); if (id->isBroadcast()) { Test Changes The test test_sdpa_loop_split has been modified to use a different approach for defining tensor shapes. Ensure that these changes do not inadvertently skip valid test cases. d = multidevice_test.size mesh = nvfuser.DeviceMesh(range(d)) class Model(FusionDefinition): def __init__(self, qkv_format: QkvFormat): super().__init__() self._qkv_format = qkv_format def definition(self) -> None: