def combine_complex():
"""
Returns a transformation which joins two real inputs into complex output
(1 output, 2 inputs): ``output = input1 + 1j * input2``.
"""
return Transformation(
inputs=2, outputs=1,
derive_o_from_is=lambda i1, i2: dtypes.complex_for(i1),
code="${o1.store}(COMPLEX_CTR(${o1.ctype})(${i1.load}, ${i2.load}));")
def combine_complex():
"""
Returns a transformation which joins two real inputs into complex output
(1 output, 2 inputs): ``output = input1 + 1j * input2``.
"""
return Transformation(
inputs=2,
outputs=1,
derive_o_from_is=lambda i1, i2: dtypes.complex_for(i1),
code="${o1.store}(COMPLEX_CTR(${o1.ctype})(${i1.load}, ${i2.load}));")
def split_complex():
"""
Returns a transformation which splits complex input into two real outputs
(2 outputs, 1 input): ``output1 = Re(input1), output2 = Im(input1)``.
"""
return Transformation(
inputs=1, outputs=2,
derive_i_from_os=lambda o1, o2: dtypes.complex_for(o1),
code="""
${o1.store}(${i1.load}.x);
${o2.store}(${i1.load}.y);
""")
def split_complex():
"""
Returns a transformation which splits complex input into two real outputs
(2 outputs, 1 input): ``output1 = Re(input1), output2 = Im(input1)``.
"""
return Transformation(
inputs=1,
outputs=2,
derive_i_from_os=lambda o1, o2: dtypes.complex_for(o1),
code="""
${o1.store}(${i1.load}.x);
${o2.store}(${i1.load}.y);
""")
def test_errors(ctx_and_double, shapes, arg_dtypes):
ctx, double = ctx_and_double
s1, s2 = shapes
c1, c2 = arg_dtypes
dtype = numpy.float64 if double else numpy.float32
dtype1 = dtypes.complex_for(dtype) if c1 else dtype
dtype2 = dtypes.complex_for(dtype) if c2 else dtype
a = get_test_array(s1, dtype1)
b = get_test_array(s2, dtype2)
res_ref = ref_dot(a, b)
a_dev = ctx.to_device(a)
b_dev = ctx.to_device(b)
res_dev = ctx.empty_like(res_ref)
dot = MatrixMul(ctx).prepare_for(res_dev, a_dev, b_dev)
dot(res_dev, a_dev, b_dev)
assert diff_is_negligible(ctx.from_device(res_dev), res_ref)
def test_errors(ctx_and_double, shapes, arg_dtypes):
ctx, double = ctx_and_double
s1, s2 = shapes
c1, c2 = arg_dtypes
dtype = numpy.float64 if double else numpy.float32
dtype1 = dtypes.complex_for(dtype) if c1 else dtype
dtype2 = dtypes.complex_for(dtype) if c2 else dtype
a = get_test_array(s1, dtype1)
b = get_test_array(s2, dtype2)
res_ref = ref_dot(a, b)
a_dev = ctx.to_device(a)
b_dev = ctx.to_device(b)
res_dev = ctx.empty_like(res_ref)
dot = MatrixMul(ctx).prepare_for(res_dev, a_dev, b_dev)
dot(res_dev, a_dev, b_dev)
assert diff_is_negligible(ctx.from_device(res_dev), res_ref)
