import queue import threading import multiprocessing import numpy as np import pytest from numpy.random import random from numpy.testing import assert_array_almost_equal, assert_allclose from pytest import raises as assert_raises import scipy.fft as fft from scipy.conftest import array_api_compatible from scipy._lib._array_api import ( array_namespace, size, xp_assert_close, xp_assert_equal ) pytestmark = [array_api_compatible, pytest.mark.usefixtures("skip_if_array_api")] skip_if_array_api = pytest.mark.skip_if_array_api # Expected input dtypes. Note that `scipy.fft` is more flexible for numpy, # but for C2C transforms like `fft.fft`, the array API standard only mandates # that complex dtypes should work, float32/float64 aren't guaranteed to. def get_expected_input_dtype(func, xp): if func in [fft.fft, fft.fftn, fft.fft2, fft.ifft, fft.ifftn, fft.ifft2, fft.hfft, fft.hfftn, fft.hfft2, fft.irfft, fft.irfftn, fft.irfft2]: dtype = xp.complex128 elif func in [fft.rfft, fft.rfftn, fft.rfft2, fft.ihfft, fft.ihfftn, fft.ihfft2]: dtype = xp.float64 else: raise ValueError(f'Unknown FFT function: {func}') return dtype def fft1(x): L = len(x) phase = -2j*np.pi*(np.arange(L)/float(L)) phase = np.arange(L).reshape(-1, 1) * phase return np.sum(x*np.exp(phase), axis=1) class TestFFTShift: def test_fft_n(self, xp): x = xp.asarray([1, 2, 3], dtype=xp.complex128) if xp.__name__ == 'torch': assert_raises(RuntimeError, fft.fft, x, 0) else: assert_raises(ValueError, fft.fft, x, 0) class TestFFT1D: def test_identity(self, xp): maxlen = 512 x = xp.asarray(random(maxlen) + 1j*random(maxlen)) xr = xp.asarray(random(maxlen)) for i in range(1, maxlen): xp_assert_close(fft.ifft(fft.fft(x[0:i])), x[0:i], rtol=1e-9, atol=0) xp_assert_close(fft.irfft(fft.rfft(xr[0:i]), i), xr[0:i], rtol=1e-9, atol=0) def test_fft(self, xp): x = random(30) + 1j*random(30) expect = xp.asarray(fft1(x)) x = xp.asarray(x) xp_assert_close(fft.fft(x), expect) xp_assert_close(fft.fft(x, norm="backward"), expect) xp_assert_close(fft.fft(x, norm="ortho"), expect / xp.sqrt(xp.asarray(30, dtype=xp.float64)),) xp_assert_close(fft.fft(x, norm="forward"), expect / 30) def test_ifft(self, xp): x = xp.asarray(random(30) + 1j*random(30)) xp_assert_close(fft.ifft(fft.fft(x)), x) for norm in ["backward", "ortho", "forward"]: xp_assert_close(fft.ifft(fft.fft(x, norm=norm), norm=norm), x) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) def test_fft2(self, xp): x = xp.asarray(random((30, 20)) + 1j*random((30, 20))) expect = fft.fft(fft.fft(x, axis=1), axis=0) xp_assert_close(fft.fft2(x), expect) xp_assert_close(fft.fft2(x, norm="backward"), expect) xp_assert_close(fft.fft2(x, norm="ortho"), expect / xp.sqrt(xp.asarray(30 * 20, dtype=xp.float64))) xp_assert_close(fft.fft2(x, norm="forward"), expect / (30 * 20)) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) def test_ifft2(self, xp): x = xp.asarray(random((30, 20)) + 1j*random((30, 20))) expect = fft.ifft(fft.ifft(x, axis=1), axis=0) xp_assert_close(fft.ifft2(x), expect) xp_assert_close(fft.ifft2(x, norm="backward"), expect) xp_assert_close(fft.ifft2(x, norm="ortho"), expect * xp.sqrt(xp.asarray(30 * 20, dtype=xp.float64))) xp_assert_close(fft.ifft2(x, norm="forward"), expect * (30 * 20)) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) def test_fftn(self, xp): x = xp.asarray(random((30, 20, 10)) + 1j*random((30, 20, 10))) expect = fft.fft(fft.fft(fft.fft(x, axis=2), axis=1), axis=0) xp_assert_close(fft.fftn(x), expect) xp_assert_close(fft.fftn(x, norm="backward"), expect) xp_assert_close(fft.fftn(x, norm="ortho"), expect / xp.sqrt(xp.asarray(30 * 20 * 10, dtype=xp.float64))) xp_assert_close(fft.fftn(x, norm="forward"), expect / (30 * 20 * 10)) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) def test_ifftn(self, xp): x = xp.asarray(random((30, 20, 10)) + 1j*random((30, 20, 10))) expect = fft.ifft(fft.ifft(fft.ifft(x, axis=2), axis=1), axis=0) xp_assert_close(fft.ifftn(x), expect) xp_assert_close(fft.ifftn(x, norm="backward"), expect) xp_assert_close( fft.ifftn(x, norm="ortho"), fft.ifftn(x) * xp.sqrt(xp.asarray(30 * 20 * 10, dtype=xp.float64)) ) xp_assert_close(fft.ifftn(x, norm="forward"), expect * (30 * 20 * 10)) def test_rfft(self, xp): x = xp.asarray(random(29), dtype=xp.float64) for n in [size(x), 2*size(x)]: for norm in [None, "backward", "ortho", "forward"]: xp_assert_close(fft.rfft(x, n=n, norm=norm), fft.fft(xp.asarray(x, dtype=xp.complex128), n=n, norm=norm)[:(n//2 + 1)]) xp_assert_close( fft.rfft(x, n=n, norm="ortho"), fft.rfft(x, n=n) / xp.sqrt(xp.asarray(n, dtype=xp.float64)) ) def test_irfft(self, xp): x = xp.asarray(random(30)) xp_assert_close(fft.irfft(fft.rfft(x)), x) for norm in ["backward", "ortho", "forward"]: xp_assert_close(fft.irfft(fft.rfft(x, norm=norm), norm=norm), x) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) def test_rfft2(self, xp): x = xp.asarray(random((30, 20)), dtype=xp.float64) expect = fft.fft2(xp.asarray(x, dtype=xp.complex128))[:, :11] xp_assert_close(fft.rfft2(x), expect) xp_assert_close(fft.rfft2(x, norm="backward"), expect) xp_assert_close(fft.rfft2(x, norm="ortho"), expect / xp.sqrt(xp.asarray(30 * 20, dtype=xp.float64))) xp_assert_close(fft.rfft2(x, norm="forward"), expect / (30 * 20)) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) def test_irfft2(self, xp): x = xp.asarray(random((30, 20))) xp_assert_close(fft.irfft2(fft.rfft2(x)), x) for norm in ["backward", "ortho", "forward"]: xp_assert_close(fft.irfft2(fft.rfft2(x, norm=norm), norm=norm), x) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) def test_rfftn(self, xp): x = xp.asarray(random((30, 20, 10)), dtype=xp.float64) expect = fft.fftn(xp.asarray(x, dtype=xp.complex128))[:, :, :6] xp_assert_close(fft.rfftn(x), expect) xp_assert_close(fft.rfftn(x, norm="backward"), expect) xp_assert_close(fft.rfftn(x, norm="ortho"), expect / xp.sqrt(xp.asarray(30 * 20 * 10, dtype=xp.float64))) xp_assert_close(fft.rfftn(x, norm="forward"), expect / (30 * 20 * 10)) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) def test_irfftn(self, xp): x = xp.asarray(random((30, 20, 10))) xp_assert_close(fft.irfftn(fft.rfftn(x)), x) for norm in ["backward", "ortho", "forward"]: xp_assert_close(fft.irfftn(fft.rfftn(x, norm=norm), norm=norm), x) def test_hfft(self, xp): x = random(14) + 1j*random(14) x_herm = np.concatenate((random(1), x, random(1))) x = np.concatenate((x_herm, x[::-1].conj())) x = xp.asarray(x) x_herm = xp.asarray(x_herm) expect = xp.real(fft.fft(x)) xp_assert_close(fft.hfft(x_herm), expect) xp_assert_close(fft.hfft(x_herm, norm="backward"), expect) xp_assert_close(fft.hfft(x_herm, norm="ortho"), expect / xp.sqrt(xp.asarray(30, dtype=xp.float64))) xp_assert_close(fft.hfft(x_herm, norm="forward"), expect / 30) def test_ihfft(self, xp): x = random(14) + 1j*random(14) x_herm = np.concatenate((random(1), x, random(1))) x = np.concatenate((x_herm, x[::-1].conj())) x = xp.asarray(x) x_herm = xp.asarray(x_herm) xp_assert_close(fft.ihfft(fft.hfft(x_herm)), x_herm) for norm in ["backward", "ortho", "forward"]: xp_assert_close(fft.ihfft(fft.hfft(x_herm, norm=norm), norm=norm), x_herm) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) def test_hfft2(self, xp): x = xp.asarray(random((30, 20))) xp_assert_close(fft.hfft2(fft.ihfft2(x)), x) for norm in ["backward", "ortho", "forward"]: xp_assert_close(fft.hfft2(fft.ihfft2(x, norm=norm), norm=norm), x) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) def test_ihfft2(self, xp): x = xp.asarray(random((30, 20)), dtype=xp.float64) expect = fft.ifft2(xp.asarray(x, dtype=xp.complex128))[:, :11] xp_assert_close(fft.ihfft2(x), expect) xp_assert_close(fft.ihfft2(x, norm="backward"), expect) xp_assert_close( fft.ihfft2(x, norm="ortho"), expect * xp.sqrt(xp.asarray(30 * 20, dtype=xp.float64)) ) xp_assert_close(fft.ihfft2(x, norm="forward"), expect * (30 * 20)) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) def test_hfftn(self, xp): x = xp.asarray(random((30, 20, 10))) xp_assert_close(fft.hfftn(fft.ihfftn(x)), x) for norm in ["backward", "ortho", "forward"]: xp_assert_close(fft.hfftn(fft.ihfftn(x, norm=norm), norm=norm), x) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) def test_ihfftn(self, xp): x = xp.asarray(random((30, 20, 10)), dtype=xp.float64) expect = fft.ifftn(xp.asarray(x, dtype=xp.complex128))[:, :, :6] xp_assert_close(expect, fft.ihfftn(x)) xp_assert_close(expect, fft.ihfftn(x, norm="backward")) xp_assert_close( fft.ihfftn(x, norm="ortho"), expect * xp.sqrt(xp.asarray(30 * 20 * 10, dtype=xp.float64)) ) xp_assert_close(fft.ihfftn(x, norm="forward"), expect * (30 * 20 * 10)) def _check_axes(self, op, xp): dtype = get_expected_input_dtype(op, xp) x = xp.asarray(random((30, 20, 10)), dtype=dtype) axes = [(0, 1, 2), (0, 2, 1), (1, 0, 2), (1, 2, 0), (2, 0, 1), (2, 1, 0)] xp_test = array_namespace(x) for a in axes: op_tr = op(xp_test.permute_dims(x, axes=a)) tr_op = xp_test.permute_dims(op(x, axes=a), axes=a) xp_assert_close(op_tr, tr_op) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) @pytest.mark.parametrize("op", [fft.fftn, fft.ifftn, fft.rfftn, fft.irfftn]) def test_axes_standard(self, op, xp): self._check_axes(op, xp) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) @pytest.mark.parametrize("op", [fft.hfftn, fft.ihfftn]) def test_axes_non_standard(self, op, xp): self._check_axes(op, xp) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) @pytest.mark.parametrize("op", [fft.fftn, fft.ifftn, fft.rfftn, fft.irfftn]) def test_axes_subset_with_shape_standard(self, op, xp): dtype = get_expected_input_dtype(op, xp) x = xp.asarray(random((16, 8, 4)), dtype=dtype) axes = [(0, 1, 2), (0, 2, 1), (1, 2, 0)] xp_test = array_namespace(x) for a in axes: # different shape on the first two axes shape = tuple([2*x.shape[ax] if ax in a[:2] else x.shape[ax] for ax in range(x.ndim)]) # transform only the first two axes op_tr = op(xp_test.permute_dims(x, axes=a), s=shape[:2], axes=(0, 1)) tr_op = xp_test.permute_dims(op(x, s=shape[:2], axes=a[:2]), axes=a) xp_assert_close(op_tr, tr_op) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) @pytest.mark.parametrize("op", [fft.fft2, fft.ifft2, fft.rfft2, fft.irfft2, fft.hfft2, fft.ihfft2, fft.hfftn, fft.ihfftn]) def test_axes_subset_with_shape_non_standard(self, op, xp): dtype = get_expected_input_dtype(op, xp) x = xp.asarray(random((16, 8, 4)), dtype=dtype) axes = [(0, 1, 2), (0, 2, 1), (1, 2, 0)] xp_test = array_namespace(x) for a in axes: # different shape on the first two axes shape = tuple([2*x.shape[ax] if ax in a[:2] else x.shape[ax] for ax in range(x.ndim)]) # transform only the first two axes op_tr = op(xp_test.permute_dims(x, axes=a), s=shape[:2], axes=(0, 1)) tr_op = xp_test.permute_dims(op(x, s=shape[:2], axes=a[:2]), axes=a) xp_assert_close(op_tr, tr_op) def test_all_1d_norm_preserving(self, xp): # verify that round-trip transforms are norm-preserving x = xp.asarray(random(30), dtype=xp.float64) xp_test = array_namespace(x) x_norm = xp_test.linalg.vector_norm(x) n = size(x) * 2 func_pairs = [(fft.rfft, fft.irfft), # hfft: order so the first function takes x.size samples # (necessary for comparison to x_norm above) (fft.ihfft, fft.hfft), # functions that expect complex dtypes at the end (fft.fft, fft.ifft), ] for forw, back in func_pairs: if forw == fft.fft: x = xp.asarray(x, dtype=xp.complex128) x_norm = xp_test.linalg.vector_norm(x) for n in [size(x), 2*size(x)]: for norm in ['backward', 'ortho', 'forward']: tmp = forw(x, n=n, norm=norm) tmp = back(tmp, n=n, norm=norm) xp_assert_close(xp_test.linalg.vector_norm(tmp), x_norm) @skip_if_array_api(np_only=True) @pytest.mark.parametrize("dtype", [np.float16, np.longdouble]) def test_dtypes_nonstandard(self, dtype): x = random(30).astype(dtype) out_dtypes = {np.float16: np.complex64, np.longdouble: np.clongdouble} x_complex = x.astype(out_dtypes[dtype]) res_fft = fft.ifft(fft.fft(x)) res_rfft = fft.irfft(fft.rfft(x)) res_hfft = fft.hfft(fft.ihfft(x), x.shape[0]) # Check both numerical results and exact dtype matches assert_array_almost_equal(res_fft, x_complex) assert_array_almost_equal(res_rfft, x) assert_array_almost_equal(res_hfft, x) assert res_fft.dtype == x_complex.dtype assert res_rfft.dtype == np.result_type(np.float32, x.dtype) assert res_hfft.dtype == np.result_type(np.float32, x.dtype) @pytest.mark.parametrize("dtype", ["float32", "float64"]) def test_dtypes_real(self, dtype, xp): x = xp.asarray(random(30), dtype=getattr(xp, dtype)) res_rfft = fft.irfft(fft.rfft(x)) res_hfft = fft.hfft(fft.ihfft(x), x.shape[0]) # Check both numerical results and exact dtype matches rtol = {"float32": 1.2e-4, "float64": 1e-8}[dtype] xp_assert_close(res_rfft, x, rtol=rtol, atol=0) xp_assert_close(res_hfft, x, rtol=rtol, atol=0) @pytest.mark.parametrize("dtype", ["complex64", "complex128"]) def test_dtypes_complex(self, dtype, xp): x = xp.asarray(random(30), dtype=getattr(xp, dtype)) res_fft = fft.ifft(fft.fft(x)) # Check both numerical results and exact dtype matches rtol = {"complex64": 1.2e-4, "complex128": 1e-8}[dtype] xp_assert_close(res_fft, x, rtol=rtol, atol=0) @skip_if_array_api(np_only=True) @pytest.mark.parametrize( "dtype", [np.float32, np.float64, np.longdouble, np.complex64, np.complex128, np.clongdouble]) @pytest.mark.parametrize("order", ["F", 'non-contiguous']) @pytest.mark.parametrize( "fft", [fft.fft, fft.fft2, fft.fftn, fft.ifft, fft.ifft2, fft.ifftn]) def test_fft_with_order(dtype, order, fft): # Check that FFT/IFFT produces identical results for C, Fortran and # non contiguous arrays rng = np.random.RandomState(42) X = rng.rand(8, 7, 13).astype(dtype, copy=False) if order == 'F': Y = np.asfortranarray(X) else: # Make a non contiguous array Y = X[::-1] X = np.ascontiguousarray(X[::-1]) if fft.__name__.endswith('fft'): for axis in range(3): X_res = fft(X, axis=axis) Y_res = fft(Y, axis=axis) assert_array_almost_equal(X_res, Y_res) elif fft.__name__.endswith(('fft2', 'fftn')): axes = [(0, 1), (1, 2), (0, 2)] if fft.__name__.endswith('fftn'): axes.extend([(0,), (1,), (2,), None]) for ax in axes: X_res = fft(X, axes=ax) Y_res = fft(Y, axes=ax) assert_array_almost_equal(X_res, Y_res) else: raise ValueError class TestFFTThreadSafe: threads = 16 input_shape = (800, 200) def _test_mtsame(self, func, *args, xp=None): def worker(args, q): q.put(func(*args)) q = queue.Queue() expected = func(*args) # Spin off a bunch of threads to call the same function simultaneously t = [threading.Thread(target=worker, args=(args, q)) for i in range(self.threads)] [x.start() for x in t] [x.join() for x in t] # Make sure all threads returned the correct value for i in range(self.threads): xp_assert_equal( q.get(timeout=5), expected, err_msg='Function returned wrong value in multithreaded context' ) def test_fft(self, xp): a = xp.ones(self.input_shape, dtype=xp.complex128) self._test_mtsame(fft.fft, a, xp=xp) def test_ifft(self, xp): a = xp.full(self.input_shape, 1+0j) self._test_mtsame(fft.ifft, a, xp=xp) def test_rfft(self, xp): a = xp.ones(self.input_shape) self._test_mtsame(fft.rfft, a, xp=xp) def test_irfft(self, xp): a = xp.full(self.input_shape, 1+0j) self._test_mtsame(fft.irfft, a, xp=xp) def test_hfft(self, xp): a = xp.ones(self.input_shape, dtype=xp.complex64) self._test_mtsame(fft.hfft, a, xp=xp) def test_ihfft(self, xp): a = xp.ones(self.input_shape) self._test_mtsame(fft.ihfft, a, xp=xp) @skip_if_array_api(np_only=True) @pytest.mark.parametrize("func", [fft.fft, fft.ifft, fft.rfft, fft.irfft]) def test_multiprocess(func): # Test that fft still works after fork (gh-10422) with multiprocessing.Pool(2) as p: res = p.map(func, [np.ones(100) for _ in range(4)]) expect = func(np.ones(100)) for x in res: assert_allclose(x, expect) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) class TestIRFFTN: def test_not_last_axis_success(self, xp): ar, ai = np.random.random((2, 16, 8, 32)) a = ar + 1j*ai a = xp.asarray(a) axes = (-2,) # Should not raise error fft.irfftn(a, axes=axes) @skip_if_array_api('torch', reasons=['torch.fft not yet implemented by array-api-compat']) @pytest.mark.parametrize("func", [fft.fft, fft.ifft, fft.rfft, fft.irfft, fft.fftn, fft.ifftn, fft.rfftn, fft.irfftn, fft.hfft, fft.ihfft]) def test_non_standard_params(func, xp): if func in [fft.rfft, fft.rfftn, fft.ihfft]: dtype = xp.float64 else: dtype = xp.complex128 if xp.__name__ != 'numpy': x = xp.asarray([1, 2, 3], dtype=dtype) # func(x) should not raise an exception func(x) assert_raises(ValueError, func, x, workers=2) # `plan` param is not tested since SciPy does not use it currently # but should be tested if it comes into use