Move arithmetic ops to separate file/registry

This should be a non-functional change.

Signed-off-by: Greg Bonik <gbonik@nvidia.com>

Author: gbonik

experimental/cuda-lang/src/cuda/lang/_ir/_host_program.py

@@ -7,10 +7,11 @@
 from typing import Literal
 
 from cuda.lang._ir import ir
+from cuda.tile._ir.arithmetic_ops import binary_arithmetic_tensorlike_raw
+from cuda.tile._ir.core_ops import TypedConst
 from cuda.tile._ir.ops import (
-    loosely_typed_const, binary_arithmetic_tensorlike_raw,
-    strictly_typed_const, unary, _UNARY_BOOL_INT, TypedConst, AssumeBounded, AssumeDivBy, Assign,
-    binary_bitwise_tensorlike_raw
+    loosely_typed_const, strictly_typed_const, unary, _UNARY_BOOL_INT, AssumeBounded, AssumeDivBy,
+    Assign, binary_bitwise_tensorlike_raw
 )
 from cuda.tile._ir.typing_support import I32_TY, I64_TY
 

experimental/cuda-lang/src/cuda/lang/_ir/ops.py

@@ -25,9 +25,20 @@
     require_scalar_pointer_type,
 )
 from cuda.tile._ir.type import TensorLikeTy
-from cuda.tile._ir.ops import (
+from cuda.tile._ir.core_ops import (
+    TypedConst,
+)
+from cuda.tile._ir.arithmetic_ops import (
     binary_arithmetic_tensorlike,
     binary_arithmetic_tensorlike_raw,
+    RawBinaryArithmeticOperation,
+    RawComparisonOperation,
+    RawBinaryBitwiseOperation,
+    RawBitwiseShiftOperation,
+    TileAsType,
+    RawWhereOperation,
+)
+from cuda.tile._ir.ops import (
     loosely_typed_const,
     tile_impl_registry,
     bind_method,
@@ -41,14 +52,7 @@
     AssumeDivBy,
     MakeTensorView,
     MakeDummy,
-    RawBinaryArithmeticOperation,
-    RawComparisonOperation,
-    RawBinaryBitwiseOperation,
-    RawBitwiseShiftOperation,
-    TileAsType,
-    TypedConst,
     IfElse,
-    RawWhereOperation,
     EndBranch,
     Loop,
     Continue,