def test_dot(ctx_factory):
from pytest import importorskip
importorskip("mako")
context = ctx_factory()
queue = cl.CommandQueue(context)
dtypes = [np.float32, np.complex64]
if has_double_support(context.devices[0]):
dtypes.extend([np.float64, np.complex128])
for a_dtype in dtypes:
for b_dtype in dtypes:
print(a_dtype, b_dtype)
a_gpu = general_clrand(queue, (200000,), a_dtype)
a = a_gpu.get()
b_gpu = general_clrand(queue, (200000,), b_dtype)
b = b_gpu.get()
dot_ab = np.dot(a, b)
dot_ab_gpu = cl_array.dot(a_gpu, b_gpu).get()
assert abs(dot_ab_gpu - dot_ab) / abs(dot_ab) < 1e-4
vdot_ab = np.vdot(a, b)
vdot_ab_gpu = cl_array.vdot(a_gpu, b_gpu).get()
assert abs(vdot_ab_gpu - vdot_ab) / abs(vdot_ab) < 1e-4
def test_dot(ctx_factory):
from pytest import importorskip
importorskip("mako")
context = ctx_factory()
queue = cl.CommandQueue(context)
dtypes = [np.float32, np.complex64]
if has_double_support(context.devices[0]):
dtypes.extend([np.float64, np.complex128])
for a_dtype in dtypes:
for b_dtype in dtypes:
print(a_dtype, b_dtype)
a_gpu = general_clrand(queue, (200000, ), a_dtype)
a = a_gpu.get()
b_gpu = general_clrand(queue, (200000, ), b_dtype)
b = b_gpu.get()
dot_ab = np.dot(a, b)
dot_ab_gpu = cl_array.dot(a_gpu, b_gpu).get()
assert abs(dot_ab_gpu - dot_ab) / abs(dot_ab) < 1e-4
vdot_ab = np.vdot(a, b)
vdot_ab_gpu = cl_array.vdot(a_gpu, b_gpu).get()
assert abs(vdot_ab_gpu - vdot_ab) / abs(vdot_ab) < 1e-4
def test_sum(ctx_factory):
from pytest import importorskip
importorskip("mako")
context = ctx_factory()
queue = cl.CommandQueue(context)
n = 200000
for dtype in [np.float32, np.complex64]:
a_gpu = general_clrand(queue, (n,), dtype)
a = a_gpu.get()
for slc in [
slice(None),
slice(1000, 3000),
slice(1000, -3000),
slice(1000, None),
slice(1000, None, 3),
slice(1000, 1000),
]:
sum_a = np.sum(a[slc])
if sum_a:
ref_divisor = abs(sum_a)
else:
ref_divisor = 1
if slc.step is None:
sum_a_gpu = cl_array.sum(a_gpu[slc]).get()
assert abs(sum_a_gpu - sum_a) / ref_divisor < 1e-4
sum_a_gpu_2 = cl_array.sum(a_gpu, slice=slc).get()
assert abs(sum_a_gpu_2 - sum_a) / ref_divisor < 1e-4
def test_sum(ctx_factory):
from pytest import importorskip
importorskip("mako")
context = ctx_factory()
queue = cl.CommandQueue(context)
n = 200000
for dtype in [np.float32, np.complex64]:
a_gpu = general_clrand(queue, (n,), dtype)
a = a_gpu.get()
for slc in [
slice(None),
slice(1000, 3000),
slice(1000, -3000),
slice(1000, None),
slice(1000, None, 3),
]:
sum_a = np.sum(a[slc])
if slc.step is None:
sum_a_gpu = cl_array.sum(a_gpu[slc]).get()
assert abs(sum_a_gpu - sum_a) / abs(sum_a) < 1e-4
sum_a_gpu_2 = cl_array.sum(a_gpu, slice=slc).get()
assert abs(sum_a_gpu_2 - sum_a) / abs(sum_a) < 1e-4
def test_dot(ctx_factory):
from pytest import importorskip
importorskip("mako")
context = ctx_factory()
queue = cl.CommandQueue(context)
dev = context.devices[0]
dtypes = [np.float32, np.complex64]
if has_double_support(dev):
if has_struct_arg_count_bug(dev) == "apple":
dtypes.extend([np.float64])
else:
dtypes.extend([np.float64, np.complex128])
for a_dtype in dtypes:
for b_dtype in dtypes:
print(a_dtype, b_dtype)
a_gpu = general_clrand(queue, (200000,), a_dtype)
a = a_gpu.get()
b_gpu = general_clrand(queue, (200000,), b_dtype)
b = b_gpu.get()
dot_ab = np.dot(a, b)
dot_ab_gpu = cl_array.dot(a_gpu, b_gpu).get()
assert abs(dot_ab_gpu - dot_ab) / abs(dot_ab) < 1e-4
try:
vdot_ab = np.vdot(a, b)
except NotImplementedError:
import sys
is_pypy = "__pypy__" in sys.builtin_module_names
if is_pypy:
print("PYPY: VDOT UNIMPLEMENTED")
continue
else:
raise
vdot_ab_gpu = cl_array.vdot(a_gpu, b_gpu).get()
rel_err = abs(vdot_ab_gpu - vdot_ab) / abs(vdot_ab)
assert rel_err < 1e-4, rel_err
def test_dot(ctx_factory):
from pytest import importorskip
importorskip("mako")
context = ctx_factory()
queue = cl.CommandQueue(context)
dev = context.devices[0]
dtypes = [np.float32, np.complex64]
if has_double_support(dev):
if has_struct_arg_count_bug(dev) == "apple":
dtypes.extend([np.float64])
else:
dtypes.extend([np.float64, np.complex128])
for a_dtype in dtypes:
for b_dtype in dtypes:
print(a_dtype, b_dtype)
a_gpu = general_clrand(queue, (200000,), a_dtype)
a = a_gpu.get()
b_gpu = general_clrand(queue, (200000,), b_dtype)
b = b_gpu.get()
dot_ab = np.dot(a, b)
dot_ab_gpu = cl_array.dot(a_gpu, b_gpu).get()
assert abs(dot_ab_gpu - dot_ab) / abs(dot_ab) < 1e-4
try:
vdot_ab = np.vdot(a, b)
except NotImplementedError:
import sys
is_pypy = '__pypy__' in sys.builtin_module_names
if is_pypy:
print("PYPY: VDOT UNIMPLEMENTED")
continue
else:
raise
vdot_ab_gpu = cl_array.vdot(a_gpu, b_gpu).get()
rel_err = abs(vdot_ab_gpu - vdot_ab) / abs(vdot_ab)
assert rel_err < 1e-4, rel_err
def test_sum(ctx_factory):
from pytest import importorskip
importorskip("mako")
context = ctx_factory()
queue = cl.CommandQueue(context)
n = 200000
for dtype in [np.float32, np.complex64]:
a_gpu = general_clrand(queue, (n,), dtype)
a = a_gpu.get()
sum_a = np.sum(a)
sum_a_gpu = cl_array.sum(a_gpu).get()
assert abs(sum_a_gpu - sum_a) / abs(sum_a) < 1e-4
def test_sum(ctx_factory):
from pytest import importorskip
importorskip("mako")
context = ctx_factory()
queue = cl.CommandQueue(context)
n = 200000
for dtype in [np.float32, np.complex64]:
a_gpu = general_clrand(queue, (n, ), dtype)
a = a_gpu.get()
sum_a = np.sum(a)
sum_a_gpu = cl_array.sum(a_gpu).get()
assert abs(sum_a_gpu - sum_a) / abs(sum_a) < 1e-4