Add threshold and clamp operations to direct_bindings

[rdspring1]

PR List

• Add threshold and clamp operations to direct_bindings #4876 << This PR.
• Add addcmul operation to direct_bindings #4877
• Add slice operation to direct bindings #4878
• 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]

Review updated until commit 730be69

Description

• Added threshold and clamp operations to direct bindings

• Introduced new macro for threshold-like operations

• Defined clamp and threshold operations with documentation

---

Changes walkthrough 📝

	Relevant files
Enhancement	ops.cpp Add threshold and clamp operations                                              python/python_direct/ops.cpp Introduced NVFUSER_DIRECT_BINDING_THRESHOLD_LIKE_OP macro Added clamp and threshold operations using the new macro Included documentation for clamp and threshold operations +47/-0

---

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

🧪 No relevant tests

⚡ Recommended focus areas for review Possible Issue The macro NVFUSER_DIRECT_BINDING_THRESHOLD_LIKE_OP defines two bindings for the same operation name (NAME), which might lead to conflicts or unexpected behavior. Ensure that the operation names are unique or that the bindings are correctly handled. #define NVFUSER_DIRECT_BINDING_THRESHOLD_LIKE_OP(NAME, OP_NAME, DOCSTRING) \ ops.def( \ NAME, \ [](Val* arg1, Val* arg2, Val* arg3) -> Val* { \ return static_cast<Val* (*)(Val*, Val*, Val*)>(OP_NAME)( \ arg1, arg2, arg3); \ }, \ py::return_value_policy::reference); \ ops.def( \ NAME, \ [](TensorView* arg1, Val* arg2, Val* arg3) -> TensorView* { \ return static_cast<TensorView* (*)(TensorView*, Val*, Val*)>(OP_NAME)( \ arg1, arg2, arg3); \ }, \ DOCSTRING, \ py::return_value_policy::reference); Documentation The documentation for the clamp and threshold operations should specify the types of min, max, threshold, and value parameters more clearly. For example, whether they can be scalars or only tensors. NVFUSER_DIRECT_BINDING_THRESHOLD_LIKE_OP("clamp", clamp, R"( Clamps all elements in input into the range [ min, max ] Parameters ---------- input : Val or TensorView min : Val or TensorView max : Val or TensorView Returns ------- Val or TensorView Clamped values. )") NVFUSER_DIRECT_BINDING_THRESHOLD_LIKE_OP("threshold", threshold, R"( Thresholds each element of the input Tensor. Parameters ---------- input : Val or TensorView threshold : Val or TensorView value : Val or TensorView Returns ------- Val or TensorView Thresholded values. )")

[rdspring1]

!test

[jjsjann123]

nitpick: tests~~

[rdspring1]

Reference #432 to add opinfo tests for threshold and clamp