Add slice operation to direct bindings

[rdspring1]

PR List

• Add threshold and clamp operations to direct_bindings #4876
• Add addcmul operation to direct_bindings #4877
• Add slice operation to direct bindings #4878 << This PR.
• Add Iota operation to direct bindings #4879
• Add full operation to direct bindings #4880
• Add topk operation to direct bindings #4881
• Support alias outputs in direct bindings. #4882
• Add welford operation to direct bindings #4883
• Add scatter operation to direct bindings #4884

[github-actions]

Description

• Added slice operation to direct bindings

• Implemented slice_fn in ops.cpp

• Mapped SliceOp to Python frontend in python_translate.cpp

• Added tests for slice operation in test_python_frontend.py

• Updated opinfos.py to support direct bindings for slice

---

Changes walkthrough 📝

	Relevant files
Enhancement	ops.cpp Implement and bind slice operation                                              python/python_direct/ops.cpp Implemented slice_fn template function Added bindings for slice operation in bindMetadataOps +107/-0  python_translate.cpp Map SliceOp to Python frontend                                                      python/python_direct/python_translate.cpp Mapped SliceOp to Python frontend +35/-0    opinfos.py Update slice_opinfo for direct bindings                                    tests/python/opinfo/opinfos.py Updated slice_opinfo to support direct bindings +1/-0
Tests	test_python_frontend.py Add slice operation tests                                                                tests/python/direct/test_python_frontend.py Added test cases for slice operation +33/-0

---

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

🧪 PR contains tests

⚡ Recommended focus areas for review Stride Limitation The current implementation only supports strides of size 1. This is a significant limitation and should be documented or considered for future enhancement. start_idx->evaluate().as<int64_t>()); NVF_CHECK( !start_idx->isConstInt() || !end_idx->isConstInt() || end_idx->evaluate().as<int64_t>() >= start_idx->evaluate().as<int64_t>(), "Slice operation end_indices must be greater than or equal to " "start_indices. Start Indices: ", start_idx->evaluate().as<int64_t>(), " End Indices: ", end_idx->evaluate().as<int64_t>()); NVF_CHECK( stride_idx->isConstInt() && stride_idx->evaluate().as<int64_t>() == 1, "nvFuser Limitation: All slice operation strides must be of const " "size 1."); vec_slice.push_back({start_idx, end_idx, stride_idx}); Test Coverage The test cases cover static and dynamic slicing but could benefit from additional test cases, especially for edge cases and error handling. x = torch.randn((2, 5, 10), dtype=torch.float32, device="cuda:0") offset = (0, 1, 2) def fusion_func(fd: FusionDefinition) -> None: T0 = fd.define_tensor( shape=[-1, -1, -1], contiguity=[True, True, True], dtype=DataType.Float, is_cpu=False, stride_order=[2, 1, 0], ) T1 = fd.ops.slice( T0, start_indices=offset, end_indices=(2, 5, 10), strides=(1, 1, 1) ) fd.add_output(T1) V_start = list(offset) V_end = T0.shape() T2 = fd.ops.slice(T0, V_start, V_end) fd.add_output(T2) dynamic_start = fd.define_vector(3) dynamic_end = fd.define_vector(3) T3 = fd.ops.slice(T0, dynamic_start, dynamic_end) fd.add_output(T3) inputs = [x, *offset, *x.shape] nvf_out, _ = nvfuser_direct_test.exec_nvfuser(fusion_func, inputs) for out in nvf_out: nvfuser_direct_test.assertTrue(out.allclose(x[:, 1:, 2:])) Manual Normalization The manual_normalization parameter is set to true in PythonTranslator::handle(const SliceOp* sop). This should be reviewed to ensure it aligns with the intended behavior and is correctly documented. isinf, R"( Element-wise infinity check. Parameters ---------- x : Val or TensorView

[rdspring1]

!test

[rdspring1]

!build

[jjsjann123]

LGTM

[jjsjann123]

Neat~

[jjsjann123]

nitpick: felt a bit strange that we have offset but not (2, 5, 10) as a named variable.

[rdspring1]

offset is used for V_start.

[rdspring1]

!build