Adds support for NumPy version 2 (#1613)

dace/codegen/cppunparse.py

@@ -751,7 +751,7 @@ def _Repr(self, t):
 
     def _Num(self, t):
         t_n = t.value if sys.version_info >= (3, 8) else t.n
-        repr_n = repr(t_n)
+        repr_n = str(t_n)
         # For complex values, use ``dtype_to_typeclass``
         if isinstance(t_n, complex):
             dtype = dtypes.dtype_to_typeclass(complex)

dace/frontend/python/astutils.py

@@ -229,6 +229,14 @@ def subscript_to_ast_slice_recursive(node):
 
 
 class ExtUnparser(astunparse.Unparser):
+
+    def _Constant(self, t):
+        # NOTE: This is needed since NumPy 2.0 to avoid unparsing NumPy scalars as calls, e.g. `numpy.int32(1)`
+        if isinstance(t.value, numbers.Number):
+            self.write(str(t.value))
+        else:
+            super()._Constant(t)
+
     def _Subscript(self, t):
         self.dispatch(t.value)
         self.write('[')

dace/frontend/python/newast.py

@@ -38,6 +38,8 @@
 import numpy
 import sympy
 
+numpy_version = numpy.lib.NumpyVersion(numpy.__version__)
+
 # register replacements in oprepo
 import dace.frontend.python.replacements
 from dace.frontend.python.replacements import _sym_type, broadcast_to, broadcast_together
@@ -4918,9 +4920,9 @@ def visit_Num(self, node: NumConstant):
         return node.n
 
     def visit_Constant(self, node: ast.Constant):
-        if isinstance(node.value, bool):
+        if isinstance(node.value, bool) and numpy_version < '2.0.0':
             return dace.bool_(node.value)
-        if isinstance(node.value, (int, float, complex)):
+        if isinstance(node.value, (int, float, complex)) and numpy_version < '2.0.0':
             return dtypes.dtype_to_typeclass(type(node.value))(node.value)
         if isinstance(node.value, (str, bytes)):
             return StringLiteral(node.value)

dace/frontend/python/replacements.py

@@ -26,6 +26,8 @@
 import numpy as np
 import sympy as sp
 
+numpy_version = np.lib.NumpyVersion(np.__version__)
+
 Size = Union[int, dace.symbolic.symbol]
 Shape = Sequence[Size]
 if TYPE_CHECKING:
@@ -1680,22 +1682,28 @@ def _result_type(arguments: Sequence[Union[str, Number, symbolic.symbol, sp.Basi
 
     datatypes = []
     dtypes_for_result = []
+    dtypes_for_result_np2 = []
     for arg in arguments:
         if isinstance(arg, (data.Array, data.Stream)):
             datatypes.append(arg.dtype)
             dtypes_for_result.append(arg.dtype.type)
+            dtypes_for_result_np2.append(arg.dtype.type)
         elif isinstance(arg, data.Scalar):
             datatypes.append(arg.dtype)
             dtypes_for_result.append(_representative_num(arg.dtype))
+            dtypes_for_result_np2.append(arg.dtype.type)
         elif isinstance(arg, (Number, np.bool_)):
             datatypes.append(dtypes.dtype_to_typeclass(type(arg)))
             dtypes_for_result.append(arg)
+            dtypes_for_result_np2.append(arg)
         elif symbolic.issymbolic(arg):
             datatypes.append(_sym_type(arg))
             dtypes_for_result.append(_representative_num(_sym_type(arg)))
+            dtypes_for_result_np2.append(_sym_type(arg).type)
         elif isinstance(arg, dtypes.typeclass):
             datatypes.append(arg)
             dtypes_for_result.append(_representative_num(arg))
+            dtypes_for_result_np2.append(arg.type)
         else:
             raise TypeError("Type {t} of argument {a} is not supported".format(t=type(arg), a=arg))
 
@@ -1728,8 +1736,11 @@ def _result_type(arguments: Sequence[Union[str, Number, symbolic.symbol, sp.Basi
         elif (operator in ('Fabs', 'Cbrt', 'Angles', 'SignBit', 'Spacing', 'Modf', 'Floor', 'Ceil', 'Trunc')
               and coarse_types[0] == 3):
             raise TypeError("ufunc '{}' not supported for complex input".format(operator))
+        elif operator in ('Ceil', 'Floor', 'Trunc') and coarse_types[0] < 2 and numpy_version < '2.1.0':
+            result_type = dace.float64
+            casting[0] = _cast_str(result_type)
         elif (operator in ('Fabs', 'Rint', 'Exp', 'Log', 'Sqrt', 'Cbrt', 'Trigonometric', 'Angles', 'FpBoolean',
-                           'Spacing', 'Modf', 'Floor', 'Ceil', 'Trunc') and coarse_types[0] < 2):
+                           'Spacing', 'Modf') and coarse_types[0] < 2):
             result_type = dace.float64
             casting[0] = _cast_str(result_type)
         elif operator in ('Frexp'):
@@ -1809,7 +1820,10 @@ def _result_type(arguments: Sequence[Union[str, Number, symbolic.symbol, sp.Basi
                 result_type = dace.float64
             # All other arithmetic operators and cases of the above operators
             else:
-                result_type = _np_result_type(dtypes_for_result)
+                if numpy_version >= '2.0.0':
+                    result_type = _np_result_type(dtypes_for_result_np2)
+                else:
+                    result_type = _np_result_type(dtypes_for_result)
 
             if dtype1 != result_type:
                 left_cast = _cast_str(result_type)
@@ -2701,7 +2715,7 @@ def _matmult(visitor: ProgramVisitor, sdfg: SDFG, state: SDFGState, op1: str, op
               operator=None,
               inputs=["__in1"],
               outputs=["__out"],
-              code="__out = sign(__in1)",
+              code="__out = sign_numpy_2(__in1)" if numpy_version >= '2.0.0' else "__out = sign(__in1)",
               reduce=None,
               initial=np.sign.identity),
     heaviside=dict(name="_numpy_heaviside_",

dace/runtime/include/dace/math.h

@@ -327,6 +327,16 @@ template<typename T>
 static DACE_CONSTEXPR DACE_HDFI std::complex<T> sign(const std::complex<T>& x) {
     return (x.real() != 0) ? std::complex<T>(sign(x.real()), 0) : std::complex<T>(sign(x.imag()), 0);
 }
+// Numpy v2.0 or higher for complex inputs: sign(x) = x / abs(x)
+template<typename T>
+static DACE_CONSTEXPR DACE_HDFI T sign_numpy_2(const T& x) {
+    return T( (T(0) < x) - (x < T(0)) );
+    // return (x < 0) ? -1 : ( (x > 0) ? 1 : 0); 
+}
+template<typename T>
+static DACE_CONSTEXPR DACE_HDFI std::complex<T> sign_numpy_2(const std::complex<T>& x) {
+    return (x.real() != 0 && x.imag() != 0) ? x / std::abs(x) : std::complex<T>(0, 0);
+}
 
 // Computes the Heaviside step function
 template<typename T>

setup.py

@@ -73,7 +73,7 @@
       },
       include_package_data=True,
       install_requires=[
-          'numpy < 2.0', 'networkx >= 2.5', 'astunparse', 'sympy >= 1.9', 'pyyaml', 'ply',
+          'numpy', 'networkx >= 2.5', 'astunparse', 'sympy >= 1.9', 'pyyaml', 'ply',
           'fparser >= 0.1.3', 'aenum >= 3.1', 'dataclasses; python_version < "3.7"', 'dill',
           'pyreadline;platform_system=="Windows"', 'typing-compat; python_version < "3.8"', 'packaging'
       ] + cmake_requires,

tests/numpy/ufunc_test.py

@@ -978,7 +978,7 @@ def test_ufunc_isfinite_c():
     @compare_numpy_output(check_dtype=True)
     def ufunc_isfinite_c(A: dace.complex64[10]):
         A[0] = np.inf
-        A[1] = np.NaN
+        A[1] = np.nan
         return np.isfinite(A)
 
     args = dace.Config.get('compiler', 'cpu', 'args')
@@ -997,7 +997,7 @@ def test_ufunc_isfinite_f():
     @compare_numpy_output(check_dtype=True)
     def ufunc_isfinite_f(A: dace.float32[10]):
         A[0] = np.inf
-        A[1] = np.NaN
+        A[1] = np.nan
         return np.isfinite(A)
 
     args = dace.Config.get('compiler', 'cpu', 'args')
@@ -1017,7 +1017,7 @@ def ufunc_isfinite_f(A: dace.float32[10]):
 @compare_numpy_output(validation_func=lambda a: np.isfinite(a))
 def test_ufunc_isfinite_u(A: dace.uint32[10]):
     A[0] = np.inf
-    A[1] = np.NaN
+    A[1] = np.nan
     return np.isfinite(A)
 
 
@@ -1026,7 +1026,7 @@ def test_ufunc_isinf_c():
     @compare_numpy_output(check_dtype=True)
     def ufunc_isinf_c(A: dace.complex64[10]):
         A[0] = np.inf
-        A[1] = np.NaN
+        A[1] = np.nan
         return np.isinf(A)
 
     args = dace.Config.get('compiler', 'cpu', 'args')
@@ -1045,7 +1045,7 @@ def test_ufunc_isinf_f():
     @compare_numpy_output(check_dtype=True)
     def ufunc_isinf_f(A: dace.float32[10]):
         A[0] = np.inf
-        A[1] = np.NaN
+        A[1] = np.nan
         return np.isinf(A)
 
     args = dace.Config.get('compiler', 'cpu', 'args')
@@ -1065,7 +1065,7 @@ def ufunc_isinf_f(A: dace.float32[10]):
 @compare_numpy_output(validation_func=lambda a: np.isinf(a))
 def test_ufunc_isinf_u(A: dace.uint32[10]):
     A[0] = np.inf
-    A[1] = np.NaN
+    A[1] = np.nan
     return np.isinf(A)
 
 
@@ -1074,7 +1074,7 @@ def test_ufunc_isnan_c():
     @compare_numpy_output(check_dtype=True)
     def ufunc_isnan_c(A: dace.complex64[10]):
         A[0] = np.inf
-        A[1] = np.NaN
+        A[1] = np.nan
         return np.isnan(A)
 
     args = dace.Config.get('compiler', 'cpu', 'args')
@@ -1093,7 +1093,7 @@ def test_ufunc_isnan_f():
     @compare_numpy_output(check_dtype=True)
     def ufunc_isnan_f(A: dace.float32[10]):
         A[0] = np.inf
-        A[1] = np.NaN
+        A[1] = np.nan
         return np.isnan(A)
 
     args = dace.Config.get('compiler', 'cpu', 'args')
@@ -1113,7 +1113,7 @@ def ufunc_isnan_f(A: dace.float32[10]):
 @compare_numpy_output(validation_func=lambda a: np.isnan(a))
 def test_ufunc_isnan_u(A: dace.uint32[10]):
     A[0] = np.inf
-    A[1] = np.NaN
+    A[1] = np.nan
     return np.isnan(A)
 
 

tests/numpy/unop_test.py

@@ -29,7 +29,10 @@ def test_not():
     B = np.zeros((5, 5), dtype=np.int64).astype(np.bool_)
     regression = np.logical_not(A)
     nottest(A, B)
-    assert np.alltrue(B == regression)
+    if np.lib.NumpyVersion(np.__version__) >= '2.0.0':
+        assert np.all(B == regression)
+    else:
+        assert np.alltrue(B == regression)
 
 
 if __name__ == '__main__':