test_arithmetic.py

import itertools
import warnings

import numpy
import pytest

import cupy
from cupy import testing


float_types = [numpy.float16, numpy.float32, numpy.float64]
complex_types = [numpy.complex64, numpy.complex128]
signed_int_types = [numpy.int8, numpy.int16, numpy.int32, numpy.int64]
unsigned_int_types = [numpy.uint8, numpy.uint16, numpy.uint32, numpy.uint64]
int_types = signed_int_types + unsigned_int_types
all_types = [numpy.bool_] + float_types + int_types + complex_types
negative_types = (
    [numpy.bool_] + float_types + signed_int_types + complex_types)
negative_types_wo_fp16 = (
    [numpy.bool_] + [numpy.float32, numpy.float64]
    + [numpy.int16, numpy.int32, numpy.int64] + complex_types)
negative_no_complex_types = [numpy.bool_] + float_types + signed_int_types
no_complex_types = [numpy.bool_] + float_types + int_types


@testing.parameterize(*(
    testing.product({
        'nargs': [1],
        'name': ['reciprocal', 'conj', 'conjugate', 'angle'],
    }) + testing.product({
        'nargs': [2],
        'name': [
            'add', 'multiply', 'divide', 'power', 'subtract', 'true_divide',
            'floor_divide', 'float_power', 'fmod', 'remainder'],
    })
))
class TestArithmeticRaisesWithNumpyInput:

    def test_raises_with_numpy_input(self):
        nargs = self.nargs
        name = self.name

        # Check TypeError is raised if numpy.ndarray is given as input
        func = getattr(cupy, name)
        for input_xp_list in itertools.product(*[[numpy, cupy]] * nargs):
            if all(xp is cupy for xp in input_xp_list):
                # We don't test all-cupy-array inputs here
                continue
            arys = [xp.array([2, -3]) for xp in input_xp_list]
            with pytest.raises(TypeError):
                func(*arys)


@testing.parameterize(*(
    testing.product({
        'arg1': ([testing.shaped_arange((2, 3), numpy, dtype=d)
                  for d in all_types
                  ] + [0, 0.0j, 0j, 2, 2.0, 2j, True, False]),
        'name': ['conj', 'conjugate', 'real', 'imag'],
    }) + testing.product({
        'arg1': ([testing.shaped_arange((2, 3), numpy, dtype=d)
                  for d in all_types
                  ] + [0, 0.0j, 0j, 2, 2.0, 2j, True, False]),
        'deg': [True, False],
        'name': ['angle'],
    }) + testing.product({
        'arg1': ([numpy.array([-3, -2, -1, 1, 2, 3], dtype=d)
                  for d in negative_types_wo_fp16
                  ] + [0, 0.0j, 0j, 2, 2.0, 2j, -2, -2.0, -2j, True, False]),
        'deg': [True, False],
        'name': ['angle'],
    }) + testing.product({
        'arg1': ([testing.shaped_arange((2, 3), numpy, dtype=d) + 1
                  for d in all_types
                  ] + [2, 2.0, 2j, True]),
        'name': ['reciprocal'],
    })
))
class TestArithmeticUnary:

    @testing.numpy_cupy_allclose(atol=1e-5)
    def test_unary(self, xp):
        arg1 = self.arg1
        if isinstance(arg1, numpy.ndarray):
            arg1 = xp.asarray(arg1)

        if self.name in {'angle'}:
            y = getattr(xp, self.name)(arg1, self.deg)
        else:
            y = getattr(xp, self.name)(arg1)

        if self.name in ('real', 'imag'):
            # Some NumPy functions return Python scalars for Python scalar
            # inputs.
            # We need to convert them to arrays to compare with CuPy outputs.
            if xp is numpy and isinstance(arg1, (bool, int, float, complex)):
                y = xp.asarray(y)

            # TODO(niboshi): Fix this
            # numpy.real and numpy.imag return Python int if the input is
            # Python bool. CuPy should return an array of dtype=int32 or
            # dtype=int64 (depending on the platform) in such cases, instead
            # of an array of dtype=bool.
            if xp is cupy and isinstance(arg1, bool):
                y = y.astype(int)

        return y


@testing.parameterize(*testing.product({
    'shape': [(3, 2), (), (3, 0, 2)],
}))
class TestComplex:

    @testing.for_all_dtypes(no_complex=True)
    @testing.numpy_cupy_array_equal()
    def test_real_ndarray_nocomplex(self, xp, dtype):
        x = testing.shaped_arange(self.shape, xp, dtype=dtype)
        real = x.real
        assert real is x  # real returns self
        return real

    @testing.for_all_dtypes(no_complex=True)
    @testing.numpy_cupy_array_equal()
    def test_real_nocomplex(self, xp, dtype):
        x = testing.shaped_arange(self.shape, xp, dtype=dtype)
        real = xp.real(x)
        assert real is x  # real returns self
        return real

    @testing.for_all_dtypes(no_complex=True)
    @testing.numpy_cupy_array_equal()
    def test_imag_ndarray_nocomplex(self, xp, dtype):
        x = testing.shaped_arange(self.shape, xp, dtype=dtype)
        imag = x.imag
        return imag

    @testing.for_all_dtypes(no_complex=True)
    @testing.numpy_cupy_array_equal()
    def test_imag_nocomplex(self, xp, dtype):
        x = testing.shaped_arange(self.shape, xp, dtype=dtype)
        imag = xp.imag(x)
        return imag

    @testing.for_complex_dtypes()
    @testing.numpy_cupy_array_equal()
    def test_real_ndarray_complex(self, xp, dtype):
        x = testing.shaped_arange(self.shape, xp, dtype=dtype)
        x_ = x.copy()
        real = x_.real
        # real returns a view
        assert real.base is x_
        x_ += 1 + 1j
        testing.assert_array_equal(real, x.real + 1)
        return real

    @testing.for_complex_dtypes()
    @testing.numpy_cupy_array_equal()
    def test_real_complex(self, xp, dtype):
        x = testing.shaped_arange(self.shape, xp, dtype=dtype)
        x_ = x.copy()
        real = xp.real(x_)
        # real returns a view
        assert real.base is x_
        x_ += 1 + 1j
        testing.assert_array_equal(real, x.real + 1)
        return real

    @testing.for_complex_dtypes()
    @testing.numpy_cupy_array_equal()
    def test_imag_ndarray_complex(self, xp, dtype):
        x = testing.shaped_arange(self.shape, xp, dtype=dtype)
        x_ = x.copy()
        imag = x_.imag
        # imag returns a view
        assert imag.base is x_
        x_ += 1 + 1j
        testing.assert_array_equal(imag, x.imag + 1)
        return imag

    @testing.for_complex_dtypes()
    @testing.numpy_cupy_array_equal()
    def test_imag_complex(self, xp, dtype):
        x = testing.shaped_arange(self.shape, xp, dtype=dtype)
        x_ = x.copy()
        imag = xp.imag(x_)
        # imag returns a view
        assert imag.base is x_
        x_ += 1 + 1j
        testing.assert_array_equal(imag, x.imag + 1)
        return imag


class ArithmeticBinaryBase:

    @testing.numpy_cupy_allclose(atol=1e-4)
    def check_binary(self, xp):
        arg1 = self.arg1
        arg2 = self.arg2
        np1 = numpy.asarray(arg1)
        np2 = numpy.asarray(arg2)
        dtype1 = np1.dtype
        dtype2 = np2.dtype

        if self.name == 'power' or self.name == 'float_power':
            # TODO(niboshi): Fix this: power(0, 1j)
            #     numpy => 1+0j
            #     cupy => 0j
            if dtype2 in complex_types and (np1 == 0).any():
                return xp.array(True)

        # TODO(niboshi): Fix this: xp.add(0j, xp.array([2.], 'f')).dtype
        #     numpy => complex64
        #     cupy => complex128
        if isinstance(arg1, complex):
            if dtype2 in (numpy.float16, numpy.float32):
                return xp.array(True)

        if isinstance(arg1, numpy.ndarray):
            arg1 = xp.asarray(arg1)
        if isinstance(arg2, numpy.ndarray):
            arg2 = xp.asarray(arg2)

        # Subtraction between booleans is not allowed.
        if (self.name == 'subtract'
                and dtype1 == numpy.bool_
                and dtype2 == numpy.bool_):
            return xp.array(True)

        func = getattr(xp, self.name)
        with numpy.errstate(divide='ignore'):
            with warnings.catch_warnings():
                warnings.filterwarnings('ignore')
                if self.use_dtype:
                    y = func(arg1, arg2, dtype=self.dtype)
                else:
                    y = func(arg1, arg2)

        # TODO(niboshi): Fix this. If rhs is a Python complex,
        #    numpy returns complex64
        #    cupy returns complex128
        if (xp is cupy and isinstance(arg2, complex)
                and self.name != 'float_power'):
            if dtype1 in (numpy.float16, numpy.float32):
                y = y.astype(numpy.complex64)

        # NumPy returns different values (nan/inf) on division by zero
        # depending on the architecture.
        # As it is not possible for CuPy to replicate this behavior, we ignore
        # the difference here.
        if self.name in ('floor_divide', 'remainder'):
            if y.dtype in (float_types + complex_types) and (np2 == 0).any():
                y = xp.asarray(y)
                y[y == numpy.inf] = numpy.nan
                y[y == -numpy.inf] = numpy.nan

        return y


@testing.parameterize(*(
    testing.product({
        # TODO(unno): boolean subtract causes DeprecationWarning in numpy>=1.13
        'arg1': [testing.shaped_arange((2, 3), numpy, dtype=d)
                 for d in all_types
                 ] + [0, 0.0, 0j, 2, 2.0, 2j, True, False],
        'arg2': [testing.shaped_reverse_arange((2, 3), numpy, dtype=d)
                 for d in all_types
                 ] + [0, 0.0, 0j, 2, 2.0, 2j, True, False],
        'name': ['add', 'multiply', 'power', 'subtract', 'float_power'],
    }) + testing.product({
        'arg1': [numpy.array([-3, -2, -1, 1, 2, 3], dtype=d)
                 for d in negative_types
                 ] + [0, 0.0, 0j, 2, 2.0, 2j, -2, -2.0, -2j, True, False],
        'arg2': [numpy.array([-3, -2, -1, 1, 2, 3], dtype=d)
                 for d in negative_types
                 ] + [0, 0.0, 0j, 2, 2.0, 2j, -2, -2.0, -2j, True, False],
        'name': ['divide', 'true_divide', 'subtract'],
    })
))
class TestArithmeticBinary(ArithmeticBinaryBase):

    def test_binary(self):
        self.use_dtype = False
        self.check_binary()


@testing.parameterize(*(
    testing.product({
        'arg1': [numpy.array([3, 2, 1, 1, 2, 3], dtype=d)
                 for d in unsigned_int_types
                 ] + [0, 0.0, 2, 2.0, -2, -2.0, True, False],
        'arg2': [numpy.array([3, 2, 1, 1, 2, 3], dtype=d)
                 for d in unsigned_int_types
                 ] + [0, 0.0, 2, 2.0, -2, -2.0, True, False],
        'name': ['true_divide'],
        'dtype': [numpy.float64],
        'use_dtype': [True, False],
    }) + testing.product({
        'arg1': [numpy.array([-3, -2, -1, 1, 2, 3], dtype=d)
                 for d in signed_int_types
                 ] + [0, 0.0, 2, 2.0, -2, -2.0, True, False],
        'arg2': [numpy.array([-3, -2, -1, 1, 2, 3], dtype=d)
                 for d in signed_int_types
                 ] + [0, 0.0, 2, 2.0, -2, -2.0, True, False],
        'name': ['true_divide'],
        'dtype': [numpy.float64],
        'use_dtype': [True, False],
    }) + testing.product({
        'arg1': [numpy.array([-3, -2, -1, 1, 2, 3], dtype=d)
                 for d in float_types] + [0.0, 2.0, -2.0],
        'arg2': [numpy.array([-3, -2, -1, 1, 2, 3], dtype=d)
                 for d in float_types] + [0.0, 2.0, -2.0],
        'name': ['power', 'true_divide', 'subtract', 'float_power'],
        'dtype': [numpy.float64],
        'use_dtype': [True, False],
    }) + testing.product({
        'arg1': [testing.shaped_arange((2, 3), numpy, dtype=d)
                 for d in no_complex_types
                 ] + [0, 0.0, 2, 2.0, -2, -2.0, True, False],
        'arg2': [testing.shaped_reverse_arange((2, 3), numpy, dtype=d)
                 for d in no_complex_types
                 ] + [0, 0.0, 2, 2.0, -2, -2.0, True, False],
        'name': ['floor_divide', 'fmod', 'remainder'],
        'dtype': [numpy.float64],
        'use_dtype': [True, False],
    }) + testing.product({
        'arg1': [numpy.array([-3, -2, -1, 1, 2, 3], dtype=d)
                 for d in negative_no_complex_types
                 ] + [0, 0.0, 2, 2.0, -2, -2.0, True, False],
        'arg2': [numpy.array([-3, -2, -1, 1, 2, 3], dtype=d)
                 for d in negative_no_complex_types
                 ] + [0, 0.0, 2, 2.0, -2, -2.0, True, False],
        'name': ['floor_divide', 'fmod', 'remainder'],
        'dtype': [numpy.float64],
        'use_dtype': [True, False],
    })
))
class TestArithmeticBinary2(ArithmeticBinaryBase):

    def test_binary(self):
        self.check_binary()


class UfuncTestBase:

    @testing.numpy_cupy_allclose(accept_error=TypeError)
    def check_casting_out(self, in0_type, in1_type, out_type, casting, xp):
        a = testing.shaped_arange((2, 3), xp, in0_type)
        b = testing.shaped_arange((2, 3), xp, in1_type)
        c = xp.zeros((2, 3), out_type)
        if casting != 'unsafe':
            # may raise TypeError
            return xp.add(a, b, out=c, casting=casting)

        with warnings.catch_warnings(record=True) as ws:
            warnings.simplefilter('always')
            ret = xp.add(a, b, out=c, casting=casting)
        ws = [w.category for w in ws]
        assert all([w == numpy.ComplexWarning for w in ws]), str(ws)
        return ret, xp.array(len(ws))

    @testing.numpy_cupy_allclose(accept_error=TypeError)
    def check_casting_dtype(self, in0_type, in1_type, dtype, casting, xp):
        a = testing.shaped_arange((2, 3), xp, in0_type)
        b = testing.shaped_arange((2, 3), xp, in1_type)
        if casting != 'unsafe':
            # may raise TypeError
            return xp.add(a, b, dtype=dtype, casting=casting)

        with warnings.catch_warnings(record=True) as ws:
            warnings.simplefilter('always')
            ret = xp.add(a, b, dtype=dtype, casting='unsafe')
        ws = [w.category for w in ws]
        assert all([w == numpy.ComplexWarning for w in ws]), str(ws)
        return ret, xp.array(len(ws))

    # delete this, once check_casting_dtype passes
    @testing.numpy_cupy_allclose()
    def check_casting_dtype_unsafe_ignore_warnings(
            self, in0_type, in1_type, dtype, xp):
        a = testing.shaped_arange((2, 3), xp, in0_type)
        b = testing.shaped_arange((2, 3), xp, in1_type)
        with warnings.catch_warnings():
            warnings.simplefilter('ignore')
            return xp.add(a, b, dtype=dtype, casting='unsafe')


class TestUfunc(UfuncTestBase):

    @pytest.mark.parametrize('casting', [
        'no',
        'equiv',
        'safe',
        'same_kind',
        'unsafe',
    ])
    @testing.for_all_dtypes_combination(
        names=['in_type', 'out_type'])
    def test_casting_out_only(self, in_type, out_type, casting):
        self.check_casting_out(in_type, in_type, out_type, casting)

    @pytest.mark.parametrize('casting', [
        pytest.param('no', marks=pytest.mark.skip('flaky xfail')),
        pytest.param('equiv', marks=pytest.mark.skip('flaky xfail')),
        'safe',
        'same_kind',
        'unsafe',
    ])
    @testing.for_all_dtypes_combination(
        names=['in0_type', 'in1_type', 'out_type'], full=False)
    def test_casting_in_out(self, in0_type, in1_type, out_type, casting):
        self.check_casting_out(in0_type, in1_type, out_type, casting)

    @pytest.mark.xfail()
    @pytest.mark.parametrize('casting', [
        'no',
        'equiv',
    ])
    @pytest.mark.parametrize(('in0_type', 'in1_type', 'out_type'), [
        (numpy.int16, numpy.int32, numpy.int32),
    ])
    def test_casting_in_xfail1(self, in0_type, in1_type, out_type, casting):
        self.check_casting_out(in0_type, in1_type, out_type, casting)

    @pytest.mark.skip('flaky xfail')
    @pytest.mark.parametrize('casting', [
        'no',
        'equiv',
        'safe',
        'same_kind',
        'unsafe',
    ])
    @testing.for_all_dtypes_combination(
        names=['in0_type', 'in1_type', 'dtype'], full=False)
    def test_casting_dtype(self, in0_type, in1_type, dtype, casting):
        self.check_casting_dtype(in0_type, in1_type, dtype, casting)

    @pytest.mark.xfail()
    @pytest.mark.parametrize('casting', [
        'no',
        'equiv',
    ])
    @pytest.mark.parametrize(('in0_type', 'in1_type', 'dtype'), [
        (numpy.int16, numpy.int32, numpy.int32),
    ])
    def test_casting_dtype_xfail1(self, in0_type, in1_type, dtype, casting):
        self.check_casting_dtype(in0_type, in1_type, dtype, casting)

    @pytest.mark.xfail()
    @pytest.mark.parametrize('casting', [
        'no',
        'equiv',
        'safe',
        'same_kind',
    ])
    @pytest.mark.parametrize(('in0_type', 'in1_type', 'dtype'), [
        (numpy.int32, numpy.int32, numpy.bool_),
        (numpy.float64, numpy.float64, numpy.int32),
    ])
    def test_casting_dtype_xfail2(self, in0_type, in1_type, dtype, casting):
        self.check_casting_dtype(in0_type, in1_type, dtype, casting)

    @testing.for_all_dtypes_combination(
        names=['in0_type', 'in1_type', 'dtype'], full=False)
    def test_casting_dtype_unsafe_ignore_warnings(
            self, in0_type, in1_type, dtype):
        self.check_casting_dtype_unsafe_ignore_warnings(
            in0_type, in1_type, dtype
        )


@testing.slow
class TestUfuncSlow(UfuncTestBase):
    @pytest.mark.parametrize('casting', [
        pytest.param('no', marks=pytest.mark.xfail()),
        pytest.param('equiv', marks=pytest.mark.xfail()),
        'safe',
        'same_kind',
        'unsafe',
    ])
    @testing.for_all_dtypes_combination(
        names=['in0_type', 'in1_type', 'out_type'], full=True)
    def test_casting_out(self, in0_type, in1_type, out_type, casting):
        self.check_casting_out(in0_type, in1_type, out_type, casting)

    @pytest.mark.xfail()
    @pytest.mark.parametrize('casting', [
        'no',
        'equiv',
        'safe',
        'same_kind',
        'unsafe',
    ])
    @testing.for_all_dtypes_combination(
        names=['in0_type', 'in1_type', 'dtype'], full=True)
    def test_casting_dtype(self, in0_type, in1_type, dtype, casting):
        self.check_casting_dtype(in0_type, in1_type, dtype, casting)

    @testing.for_all_dtypes_combination(
        names=['in0_type', 'in1_type', 'dtype'], full=True)
    def test_casting_dtype_unsafe_ignore_warnings(
            self, in0_type, in1_type, dtype):
        self.check_casting_dtype_unsafe_ignore_warnings(
            in0_type, in1_type, dtype
        )


class TestArithmeticModf:

    @testing.for_float_dtypes()
    @testing.numpy_cupy_allclose()
    def test_modf(self, xp, dtype):
        a = xp.array([-2.5, -1.5, -0.5, 0, 0.5, 1.5, 2.5], dtype=dtype)
        b, c = xp.modf(a)
        d = xp.empty((2, 7), dtype=dtype)
        d[0] = b
        d[1] = c
        return d


@testing.parameterize(*testing.product({
    'xp': [numpy, cupy],
    'shape': [(3, 2), (), (3, 0, 2)]
}))
class TestBoolSubtract:

    def test_bool_subtract(self):
        xp = self.xp
        shape = self.shape
        x = testing.shaped_random(shape, xp, dtype=numpy.bool_)
        y = testing.shaped_random(shape, xp, dtype=numpy.bool_)
        with pytest.raises(TypeError):
            xp.subtract(x, y)