update a utility function in FFT module

dpnp/fft/dpnp_iface_fft.py

@@ -38,11 +38,7 @@
 
 import dpnp
 
-from .dpnp_utils_fft import (
-    dpnp_fft,
-    dpnp_fftn,
-    dpnp_fillfreq,
-)
+from .dpnp_utils_fft import dpnp_fft, dpnp_fftn, dpnp_fillfreq, swap_direction
 
 __all__ = [
     "fft",
@@ -66,24 +62,6 @@
 ]
 
 
-_SWAP_DIRECTION_MAP = {
-    "backward": "forward",
-    None: "forward",
-    "ortho": "ortho",
-    "forward": "backward",
-}
-
-
-def _swap_direction(norm):
-    try:
-        return _SWAP_DIRECTION_MAP[norm]
-    except KeyError:
-        raise ValueError(
-            f'Invalid norm value {norm}; should be None, "backward", '
-            '"ortho" or "forward".'
-        ) from None
-
-
 def fft(a, n=None, axis=-1, norm=None, out=None):
     """
     Compute the one-dimensional discrete Fourier Transform.
@@ -644,7 +622,7 @@ def hfft(a, n=None, axis=-1, norm=None, out=None):
 
     """
 
-    new_norm = _swap_direction(norm)
+    new_norm = swap_direction(norm)
     return irfft(dpnp.conjugate(a), n=n, axis=axis, norm=new_norm, out=out)
 
 
@@ -1073,7 +1051,7 @@ def ihfft(a, n=None, axis=-1, norm=None, out=None):
 
     """
 
-    new_norm = _swap_direction(norm)
+    new_norm = swap_direction(norm)
     res = rfft(a, n=n, axis=axis, norm=new_norm, out=out)
     return dpnp.conjugate(res, out=out)
 

dpnp/fft/dpnp_utils_fft.py

@@ -57,11 +57,7 @@
     _standardize_strides_to_nonzero,
 )
 
-__all__ = [
-    "dpnp_fft",
-    "dpnp_fftn",
-    "dpnp_fillfreq",
-]
+__all__ = ["dpnp_fft", "dpnp_fftn", "dpnp_fillfreq", "swap_direction"]
 
 
 def _check_norm(norm):
@@ -584,7 +580,6 @@ def dpnp_fft(a, forward, real, n=None, axis=-1, norm=None, out=None):
     if n < 1:
         raise ValueError(f"Invalid number of FFT data points ({n}) specified")
 
-    _check_norm(norm)
     a = _truncate_or_pad(a, (n,), (axis,))
     _validate_out_keyword(a, out, (n,), (axis,), c2c, c2r, r2c)
     # if input array is copied, in-place FFT can be used
@@ -714,3 +709,17 @@ def dpnp_fillfreq(a, m, n, val):
     ht_lin_ev, lin_ev = ti._linspace_step(m - n, 1, a[m:].get_array(), exec_q)
     _manager.add_event_pair(ht_lin_ev, lin_ev)
     return a * val
+
+
+def swap_direction(norm):
+    """Swap the direction of the FFT."""
+
+    _check_norm(norm)
+    _swap_direction_map = {
+        "backward": "forward",
+        None: "forward",
+        "ortho": "ortho",
+        "forward": "backward",
+    }
+
+    return _swap_direction_map[norm]

dpnp/tests/test_fft.py

@@ -14,6 +14,7 @@
     get_all_dtypes,
     get_complex_dtypes,
     get_float_dtypes,
+    has_support_aspect16,
 )
 
 
@@ -926,6 +927,16 @@ def test_rfft_1D_on_2D_array_out(self, dtype, n, axis, norm, order):
         expected = numpy.fft.rfft(a_np, n=n, axis=axis, norm=norm)
         assert_dtype_allclose(result, expected, check_only_type_kind=True)
 
+    @pytest.mark.skipif(not has_support_aspect16(), reason="no fp16 support")
+    def test_float16(self):
+        a = numpy.ones(10, dtype=numpy.float16)
+        ia = dpnp.array(a)
+
+        result = numpy.fft.rfft(a)
+        expected = dpnp.fft.rfft(ia)
+        # check_only_type_kind=True since Intel NumPy returns complex128
+        assert_dtype_allclose(expected, result, check_only_type_kind=True)
+
     @pytest.mark.parametrize("xp", [numpy, dpnp])
     def test_rfft_error(self, xp):
         a = xp.ones((4, 3), dtype=xp.complex64)

dpnp/tests/test_umath.py

@@ -18,7 +18,6 @@
     get_all_dtypes,
     get_float_complex_dtypes,
     get_float_dtypes,
-    has_support_aspect16,
     has_support_aspect64,
 )