Implement operator sinc | feat(torchlib) (#1228)

Fixes #1221

onnxscript/function_libs/torch_lib/ops/special.py

@@ -11,14 +11,16 @@
 """
 from __future__ import annotations
 
+import math
 from typing import Optional, Sequence
 
 from onnxscript.function_libs.torch_lib.ops import common as common_ops
 from onnxscript.function_libs.torch_lib.registration import torch_op
-from onnxscript.function_libs.torch_lib.tensor_typing import TFloatOrBFloat16
+from onnxscript.function_libs.torch_lib.tensor_typing import TFloat, TFloatOrBFloat16
 from onnxscript.onnx_opset import opset18 as op
 from onnxscript.onnx_types import TensorType
 
+_MATH_PI = math.pi
 IsScalar = common_ops.IsScalar
 
 
@@ -338,10 +340,15 @@ def aten_special_shifted_chebyshev_polynomial_w(x: TensorType, n: TensorType) ->
     raise NotImplementedError()
 
 
-def aten_special_sinc(self: TensorType) -> TensorType:
+@torch_op(("aten::special_sinc", "aten::sinc"))
+def aten_special_sinc(self: TFloat) -> TFloat:
     """special_sinc(Tensor self) -> Tensor"""
 
-    raise NotImplementedError()
+    # This computes the normalized sinc, where the input is multiplied by pi.
+    # https://pytorch.org/docs/stable/special.html#torch.special.sinc
+    pi_self = self * _MATH_PI
+
+    return op.Where(self == 0.0, op.CastLike(1, self), op.Sin(pi_self) / pi_self)
 
 
 def aten_special_spherical_bessel_j0(x: TensorType) -> TensorType:

onnxscript/tests/function_libs/torch_lib/ops_test_data.py

@@ -1428,6 +1428,9 @@ def _where_input_wrangler(
     TorchLibOpInfo("sigmoid", core_ops.aten_sigmoid),
     TorchLibOpInfo("sign", core_ops.aten_sign),
     TorchLibOpInfo("sin", core_ops.aten_sin),
+    TorchLibOpInfo(
+        "sinc", special_ops.aten_special_sinc, tolerance={torch.float16: (1e-2, 6e-4)}
+    ),
     TorchLibOpInfo("sinh", core_ops.aten_sinh),
     TorchLibOpInfo(
         "softmax",