Build with -ftrapping-math.

This fixes the spurious cast warnings seen in jax-ml#106, which occured when, for example, casting a NaN float to a bfloat16. Casting NaNs between floating point types should not warn in NumPy.

Reverts jax-ml#107, since we now fix rather than suppressing the errors.

CHANGELOG.md

@@ -23,6 +23,9 @@ To release a new version (e.g. from `1.0.0` -> `2.0.0`):
 
 ## [Unreleased]
 
+* Fixed spurious invalid value warnings when casting between floating point
+  types on Mac ARM.
+
 ## [0.3.1] - 2023-09-22
 
 * Added support for int4 casting to wider integers such as int8

ml_dtypes/tests/custom_float_test.py

@@ -214,9 +214,6 @@ def testRoundTripToInt(self, float_type):
       self.assertEqual(v, int(float_type(v)))
       self.assertEqual(-v, int(float_type(-v)))
 
-  @ignore_warning(
-      category=RuntimeWarning, message="invalid value encountered in cast"
-  )
   @ignore_warning(category=RuntimeWarning, message="overflow encountered")
   def testRoundTripToNumpy(self, float_type):
     for dtype in [
@@ -241,7 +238,7 @@ def testRoundTripToNumpy(self, float_type):
               ),
           )
 
-  def testBetweenCustomTypes(self, float_type):
+  def testCastBetweenCustomTypes(self, float_type):
     for dtype in FLOAT_DTYPES:
       x = np.array(FLOAT_VALUES[float_type], dtype=dtype)
       y = x.astype(float_type)
@@ -460,9 +457,6 @@ def testSort(self, float_type):
     sorted_float_type = np.sort(values_to_sort.astype(float_type))  # pylint: disable=too-many-function-args
     np.testing.assert_equal(sorted_f32, np.float32(sorted_float_type))
 
-  @ignore_warning(
-      category=RuntimeWarning, message="invalid value encountered in cast"
-  )
   def testArgmax(self, float_type):
     values_to_sort = np.float32(
         float_type(np.float32(FLOAT_VALUES[float_type]))
@@ -485,9 +479,6 @@ def testArgmaxOnNegativeInfinity(self, float_type):
     inf = np.array([float("-inf")], dtype=np.float32)
     np.testing.assert_equal(np.argmax(inf.astype(float_type)), np.argmax(inf))
 
-  @ignore_warning(
-      category=RuntimeWarning, message="invalid value encountered in cast"
-  )
   def testArgmin(self, float_type):
     values_to_sort = np.float32(
         float_type(np.float32(FLOAT_VALUES[float_type]))

setup.py

@@ -30,6 +30,11 @@
   COMPILE_ARGS = [
       "-std=c++17",
       "-DEIGEN_MPL2_ONLY",
+      # -ftrapping-math is necessary because NumPy looks at floating point
+      # exception state to determine whether to emit, e.g., invalid value
+      # warnings. Without this setting, on Mac ARM we see spurious "invalid
+      # value" warnings when running the tests.
+      "-ftrapping-math",
   ]
 
 exclude = ["third_party*"]