def main():
parser = argparse.ArgumentParser(description="Black-Scholes")
parser.add_argument("--options",
dest="options",
type=int,
default=10000000)
args = parser.parse_args()
options = args.options
if dpctl.has_gpu_queues():
print("\nScheduling on OpenCL GPU\n")
with dpctl.device_context("opencl:gpu") as gpu_queue:
run(10)
else:
print("\nSkip scheduling on OpenCL GPU\n")
# if dpctl.has_gpu_queues(dpctl.backend_type.level_zero):
# print("\nScheduling on Level Zero GPU\n")
# with dpctl.device_context("level0:gpu") as gpu_queue:
# run(10)
# else:
# print("\nSkip scheduling on Level Zero GPU\n")
if dpctl.has_cpu_queues():
print("\nScheduling on OpenCL CPU\n")
with dpctl.device_context("opencl:cpu") as cpu_queue:
run(10)
else:
print("\nSkip scheduling on OpenCL CPU\n")
def test_trigonometric_fn(filter_str, trig_op, input_arrays):
# FIXME: Why does archcosh fail on Gen12 discrete graphics card?
if trig_op == "arccosh" and is_gen12(filter_str):
pytest.skip()
a, b = input_arrays
trig_fn = getattr(np, trig_op)
actual = np.empty(shape=a.shape, dtype=a.dtype)
expected = np.empty(shape=a.shape, dtype=a.dtype)
if trig_op == "arctan2":
@njit
def f(a, b):
return trig_fn(a, b)
device = dpctl.SyclDevice(filter_str)
with dpctl.device_context(device), assert_auto_offloading():
actual = f(a, b)
expected = trig_fn(a, b)
else:
@njit
def f(a):
return trig_fn(a)
device = dpctl.SyclDevice(filter_str)
with dpctl.device_context(device), assert_auto_offloading():
actual = f(a)
expected = trig_fn(a)
np.testing.assert_allclose(actual, expected, rtol=1e-5, atol=0)
def test_get_current_device_type_inside_nested_device_ctxt():
assert dpctl.get_current_device_type() is not None
with dpctl.device_context("opencl:cpu:0"):
assert dpctl.get_current_device_type() == dpctl.device_type.cpu
with dpctl.device_context("opencl:gpu:0"):
assert dpctl.get_current_device_type() == dpctl.device_type.gpu
assert dpctl.get_current_device_type() == dpctl.device_type.cpu
assert dpctl.get_current_device_type() is not None
def test_get_current_device_type_inside_nested_device_ctxt(self):
self.assertNotEqual(dpctl.get_current_device_type(), None)
with dpctl.device_context("opencl:cpu:0"):
self.assertEqual(dpctl.get_current_device_type(), dpctl.device_type.cpu)
with dpctl.device_context("opencl:gpu:0"):
self.assertEqual(dpctl.get_current_device_type(), dpctl.device_type.gpu)
self.assertEqual(dpctl.get_current_device_type(), dpctl.device_type.cpu)
self.assertNotEqual(dpctl.get_current_device_type(), None)
def test_caching_kernel_using_same_queue(filter_str):
"""Test kernel caching when the same queue is used to submit a kernel
multiple times.
Args:
filter_str: SYCL filter selector string
"""
global_size = 10
N = global_size
def data_parallel_sum(a, b, c):
i = dppy.get_global_id(0)
c[i] = a[i] + b[i]
a = np.array(np.random.random(N), dtype=np.float32)
b = np.array(np.random.random(N), dtype=np.float32)
c = np.ones_like(a)
with dpctl.device_context(filter_str) as gpu_queue:
func = dppy.kernel(data_parallel_sum)
cached_kernel = func[global_size, dppy.DEFAULT_LOCAL_SIZE].specialize(
func._get_argtypes(a, b, c), gpu_queue)
for i in range(10):
_kernel = func[global_size, dppy.DEFAULT_LOCAL_SIZE].specialize(
func._get_argtypes(a, b, c), gpu_queue)
assert _kernel == cached_kernel
def test_all(dtype, shape, filter_str):
size = 1
for i in range(len(shape)):
size *= shape[i]
for i in range(2**size):
t = i
a = np.empty(size, dtype=dtype)
for j in range(size):
a[j] = 0 if t % 2 == 0 else j + 1
t = t >> 1
a = a.reshape(shape)
def fn(a):
return np.all(a)
f = njit(fn)
with dpctl.device_context(filter_str), dpnp_debug():
actual = f(a)
expected = fn(a)
np.testing.assert_allclose(actual, expected, rtol=1e-3, atol=0)
def test_dppy_fallback_false(self):
@numba.jit
def fill_value(i):
return i
def inner_call_fallback():
x = 10
a = np.empty(shape=x, dtype=np.float32)
for i in numba.prange(x):
a[i] = fill_value(i)
return a
try:
config.DEBUG = 1
config.FALLBACK_ON_CPU = 0
with warnings.catch_warnings(record=True) as w:
device = dpctl.SyclDevice("opencl:gpu")
with dpctl.device_context(device):
dppy = numba.njit(parallel=True)(inner_call_fallback)
dppy_fallback_false = dppy()
finally:
ref_result = inner_call_fallback()
config.FALLBACK_ON_CPU = 1
config.DEBUG = 0
not np.testing.assert_array_equal(dppy_fallback_false, ref_result)
assert "Failed to offload parfor" not in str(w[-1].message)
def run_kmeans(
arrayP,
arrayPclusters,
arrayC,
arrayCsum,
arrayCnumpoint,
NUMBER_OF_POINTS,
NUMBER_OF_CENTROIDS,
):
with dpctl.device_context(base_kmeans_gpu_graph.get_device_selector()):
for i in range(REPEAT):
# for i1 in range(NUMBER_OF_CENTROIDS):
# arrayC[i1, 0] = arrayP[i1, 0]
# arrayC[i1, 1] = arrayP[i1, 1]
arrayC, arrayCsum, arrayCnumpoint = kmeans(
arrayP,
arrayPclusters,
arrayC,
arrayCsum,
arrayCnumpoint,
NUMBER_OF_POINTS,
NUMBER_OF_CENTROIDS,
)
def test_multiple_prange(self):
@njit
def f(a, b):
# dimensions must be provided as scalar
m, n = a.shape
for i in prange(m):
val = 10
for j in prange(n):
b[i, j] = a[i, j] * val
for i in prange(m):
for j in prange(n):
a[i, j] = a[i, j] * 10
m = 8
n = 8
a = np.ones((m, n))
b = np.ones((m, n))
device = dpctl.SyclDevice("opencl:gpu")
with assert_auto_offloading(
parfor_offloaded=2), dpctl.device_context(device):
f(a, b)
assert np.all(b == 10)
assert np.all(a == 10)
def main():
# Use the environment variable SYCL_DEVICE_FILTER to change the default device.
# See https://github.com/intel/llvm/blob/sycl/sycl/doc/EnvironmentVariables.md#sycl_device_filter.
device = dpctl.select_default_device()
print("Using device ...")
device.print_device_info()
with dpctl.device_context(device):
c = f1(a, b)
print("c:", c, hex(c.ctypes.data))
for i in range(N):
for j in range(N):
for k in range(N):
for l in range(N): # noqa
for m in range(N):
if c[i, j, k, l, m] != 2.0:
print(
"First index not equal to 2.0 was",
i,
j,
k,
l,
m,
)
break
print("Done...")
def local_memory():
"""
This example demonstrates the usage of numba-dppy's `local.array`
intrinsic function. The function is used to create a static array
allocated on the devices local address space.
"""
blocksize = 10
@dppy.kernel
def reverse_array(A):
lm = dppy.local.array(shape=10, dtype=float32)
i = dppy.get_global_id(0)
# preload
lm[i] = A[i]
# barrier local or global will both work as we only have one work group
dppy.barrier(dppy.CLK_LOCAL_MEM_FENCE) # local mem fence
# write
A[i] += lm[blocksize - 1 - i]
arr = np.arange(blocksize).astype(np.float32)
print(arr)
# Use the environment variable SYCL_DEVICE_FILTER to change the default device.
# See https://github.com/intel/llvm/blob/sycl/sycl/doc/EnvironmentVariables.md#sycl_device_filter.
device = dpctl.select_default_device()
print("Using device ...")
device.print_device_info()
with dpctl.device_context(device):
reverse_array[blocksize, dppy.DEFAULT_LOCAL_SIZE](arr)
# the output should be `orig[::-1] + orig, i.e. [9, 9, 9, ...]``
print(arr)
def no_arg_barrier_support():
"""
This example demonstrates the usage of numba_dppy's ``barrier``
intrinsic function. The ``barrier`` function is usable only inside
a ``kernel`` and is equivalent to OpenCL's ``barrier`` function.
"""
@dppy.kernel
def twice(A):
i = dppy.get_global_id(0)
d = A[i]
# no argument defaults to global mem fence
dppy.barrier()
A[i] = d * 2
N = 10
arr = np.arange(N).astype(np.float32)
print(arr)
# Use the environment variable SYCL_DEVICE_FILTER to change the default device.
# See https://github.com/intel/llvm/blob/sycl/sycl/doc/EnvironmentVariables.md#sycl_device_filter.
device = dpctl.select_default_device()
print("Using device ...")
device.print_device_info()
with dpctl.device_context(device):
twice[N, dppy.DEFAULT_LOCAL_SIZE](arr)
# the output should be `arr * 2, i.e. [0, 2, 4, 6, ...]`
print(arr)
def main():
size = 9
scale = 3.0
# Use the environment variable SYCL_DEVICE_FILTER to change the default device.
# See https://github.com/intel/llvm/blob/sycl/sycl/doc/EnvironmentVariables.md#sycl_device_filter.
device = dpctl.select_default_device()
print("Using device ...")
device.print_device_info()
with dpctl.device_context(device):
result = rand()
# Random values in a given shape (3, 2)
print(result)
result = random_sample(size)
# Array of shape (9,) with random floats in the
# half-open interval [0.0, 1.0)
print(result)
result = random_exponential(scale, size)
# Array of shape (9,) with samples from an exponential distribution
print(result)
result = random_normal(0.0, 0.1, size)
# Array of shape (9,) with samples from a normal distribution
print(result)
print("Done...")
def gen_data_usm(nopt, dims):
X, Y, D = gen_data(nopt, dims)
with dpctl.device_context(get_device_selector()) as gpu_queue:
X_usm = dpt.usm_ndarray(
X.shape,
dtype=X.dtype,
buffer="device",
buffer_ctor_kwargs={"queue": gpu_queue},
)
Y_usm = dpt.usm_ndarray(
Y.shape,
dtype=Y.dtype,
buffer="device",
buffer_ctor_kwargs={"queue": gpu_queue},
)
D_usm = dpt.usm_ndarray(
D.shape,
dtype=D.dtype,
buffer="device",
buffer_ctor_kwargs={"queue": gpu_queue},
)
X_usm.usm_data.copy_from_host(X.reshape((-1)).view("u1"))
Y_usm.usm_data.copy_from_host(Y.reshape((-1)).view("u1"))
return (X_usm, Y_usm, D_usm)
def test_caching_kernel_using_same_context(filter_str):
"""Test kernel caching for the scenario where different SYCL queues that
share a SYCL context are used to submit a kernel.
Args:
filter_str: SYCL filter selector string
"""
global_size = 10
N = global_size
def data_parallel_sum(a, b, c):
i = dppy.get_global_id(0)
c[i] = a[i] + b[i]
a = np.array(np.random.random(N), dtype=np.float32)
b = np.array(np.random.random(N), dtype=np.float32)
c = np.ones_like(a)
# Set the global queue to the default device so that the cached_kernel gets
# created for that device
dpctl.set_global_queue(filter_str)
func = dppy.kernel(data_parallel_sum)
default_queue = dpctl.get_current_queue()
cached_kernel = func[global_size, dppy.DEFAULT_LOCAL_SIZE].specialize(
func._get_argtypes(a, b, c), default_queue)
for i in range(0, 10):
# Each iteration create a fresh queue that will share the same context
with dpctl.device_context(filter_str) as gpu_queue:
_kernel = func[global_size, dppy.DEFAULT_LOCAL_SIZE].specialize(
func._get_argtypes(a, b, c), gpu_queue)
assert _kernel == cached_kernel
def test_device_array_args_gpu(self):
c = np.ones_like(a)
with dpctl.device_context("opencl:gpu"):
data_parallel_sum[global_size, dppy.DEFAULT_LOCAL_SIZE](a, b, c)
assert np.all(c == d)
def main():
blockdim = 512, 1
griddim = int(math.ceil(float(OPT_N) / blockdim[0])), 1
# Use the environment variable SYCL_DEVICE_FILTER to change the default device.
# See https://github.com/intel/llvm/blob/sycl/sycl/doc/EnvironmentVariables.md#sycl_device_filter.
device = dpctl.select_default_device()
print("Using device ...")
device.print_device_info()
with dpctl.device_context(device):
for i in range(iterations):
black_scholes_dppy[blockdim, griddim](
callResult,
putResult,
stockPrice,
optionStrike,
optionYears,
RISKFREE,
VOLATILITY,
)
print("callResult : \n", callResult)
print("putResult : \n", putResult)
print("Done...")
def test_dpctl_api(filter_str):
device = dpctl.SyclDevice(filter_str)
with dpctl.device_context(device):
dpctl.lsplatform()
dpctl.get_current_queue()
dpctl.get_num_activated_queues()
dpctl.is_in_device_context()
def test_create_program_from_source(self):
oclSrc = " \
kernel void add(global int* a, global int* b, global int* c) { \
size_t index = get_global_id(0); \
c[index] = a[index] + b[index]; \
} \
kernel void axpy(global int* a, global int* b, global int* c, int d) { \
size_t index = get_global_id(0); \
c[index] = a[index] + d*b[index]; \
}"
with dpctl.device_context("opencl:gpu:0"):
q = dpctl.get_current_queue()
prog = dpctl_prog.create_program_from_source(q, oclSrc)
self.assertIsNotNone(prog)
self.assertTrue(prog.has_sycl_kernel("add"))
self.assertTrue(prog.has_sycl_kernel("axpy"))
addKernel = prog.get_sycl_kernel("add")
axpyKernel = prog.get_sycl_kernel("axpy")
self.assertEqual(addKernel.get_function_name(), "add")
self.assertEqual(axpyKernel.get_function_name(), "axpy")
self.assertEqual(addKernel.get_num_args(), 3)
self.assertEqual(axpyKernel.get_num_args(), 4)
def test_proper_lowering(filter_str):
# We perform eager compilation at the site of
# @dppy.kernel. This takes the default dpctl
# queue which is level_zero backed. Level_zero
# is not yet supported on Windows platform and
# hence we skip these tests if the platform is
# Windows regardless of which backend filter_str
# specifies.
if skip_if_win():
pytest.skip()
# This will trigger eager compilation
@dppy.kernel("void(float32[::1])")
def twice(A):
i = dppy.get_global_id(0)
d = A[i]
dppy.barrier(dppy.CLK_LOCAL_MEM_FENCE) # local mem fence
A[i] = d * 2
N = 256
arr = np.random.random(N).astype(np.float32)
orig = arr.copy()
with dpctl.device_context(filter_str):
twice[N, N // 2](arr)
# The computation is correct?
np.testing.assert_allclose(orig * 2, arr)
def test_dppy_func_ndarray(self):
@dppy.func
def g(a):
return a + 1
@dppy.kernel
def f(a, b):
i = dppy.get_global_id(0)
b[i] = g(a[i])
@dppy.kernel
def h(a, b):
i = dppy.get_global_id(0)
b[i] = g(a[i]) + 1
a = np.ones(self.N)
b = np.ones(self.N)
device = dpctl.SyclDevice("opencl:gpu")
with dpctl.device_context(device):
f[self.N, dppy.DEFAULT_LOCAL_SIZE](a, b)
assert np.all(b == 2)
h[self.N, dppy.DEFAULT_LOCAL_SIZE](a, b)
assert np.all(b == 3)
def test_local_memory(filter_str):
blocksize = 10
if skip_if_win():
pytest.skip()
@dppy.kernel("void(float32[::1])")
def reverse_array(A):
lm = dppy.local.array(shape=10, dtype=np.float32)
i = dppy.get_global_id(0)
# preload
lm[i] = A[i]
# barrier local or global will both work as we only have one work group
dppy.barrier(dppy.CLK_LOCAL_MEM_FENCE) # local mem fence
# write
A[i] += lm[blocksize - 1 - i]
arr = np.arange(blocksize).astype(np.float32)
orig = arr.copy()
with dpctl.device_context(filter_str):
reverse_array[blocksize, blocksize](arr)
expected = orig[::-1] + orig
np.testing.assert_allclose(expected, arr)
def test_unary_ops(filter_str, unary_op, input_arrays, get_shape, capfd):
a = input_arrays[0]
op, name = unary_op
if name != "argsort" and name != "copy":
a = np.reshape(a, get_shape)
if name == "cumprod" and (filter_str == "opencl:cpu:0"
or a.dtype == np.int32 or is_gen12(filter_str)):
pytest.skip()
if name == "cumsum" and (filter_str == "opencl:cpu:0"
or a.dtype == np.int32 or is_gen12(filter_str)):
pytest.skip()
if name == "mean" and is_gen12(filter_str):
pytest.skip()
if name == "argmax" and is_gen12(filter_str):
pytest.skip()
actual = np.empty(shape=a.shape, dtype=a.dtype)
expected = np.empty(shape=a.shape, dtype=a.dtype)
f = njit(op)
device = dpctl.SyclDevice(filter_str)
with dpctl.device_context(device), dpnp_debug():
actual = f(a)
captured = capfd.readouterr()
assert "dpnp implementation" in captured.out
expected = op(a)
np.testing.assert_allclose(actual, expected, rtol=1e-3, atol=0)
def sum_reduce(A):
global_size = len(A)
work_group_size = 64
nb_work_groups = global_size // work_group_size
if (global_size % work_group_size) != 0:
nb_work_groups += 1
partial_sums = np.zeros(nb_work_groups).astype(A.dtype)
# Use the environment variable SYCL_DEVICE_FILTER to change the default device.
# See https://github.com/intel/llvm/blob/sycl/sycl/doc/EnvironmentVariables.md#sycl_device_filter.
device = dpctl.select_default_device()
print("Using device ...")
device.print_device_info()
with dpctl.device_context(device):
inp_buf = dpctl_mem.MemoryUSMShared(A.size * A.dtype.itemsize)
inp_ndarray = np.ndarray(A.shape, buffer=inp_buf, dtype=A.dtype)
np.copyto(inp_ndarray, A)
partial_sums_buf = dpctl_mem.MemoryUSMShared(
partial_sums.size * partial_sums.dtype.itemsize)
partial_sums_ndarray = np.ndarray(
partial_sums.shape,
buffer=partial_sums_buf,
dtype=partial_sums.dtype,
)
np.copyto(partial_sums_ndarray, partial_sums)
result = sum_recursive_reduction(global_size, work_group_size,
inp_ndarray, partial_sums_ndarray)
return result
def test_kernel_arg_accessor(filter_str, input_arrays, kernel):
a, b, actual = input_arrays
expected = a + b
device = dpctl.SyclDevice(filter_str)
with dpctl.device_context(device):
call_kernel(global_size, local_size, a, b, actual, kernel)
np.testing.assert_allclose(actual, expected, rtol=1e-5, atol=0)
def test_sum_reduction(self):
# This test will only work for even case
N = 1024
assert N % 2 == 0
A = np.array(np.random.random(N), dtype=np.float32)
A_copy = A.copy()
# at max we will require half the size of A to store sum
R = np.array(np.random.random(math.ceil(N / 2)), dtype=np.float32)
device = dpctl.SyclDevice("opencl:gpu")
with dpctl.device_context(device):
total = N
while total > 1:
# call kernel
global_size = total // 2
reduction_kernel[global_size, dppy.DEFAULT_LOCAL_SIZE](
A, R, global_size
)
total = total // 2
result = A_copy.sum()
max_abs_err = result - R[0]
assert max_abs_err < 1e-4
def main():
global_size = 10
N = global_size
print("N", N)
a = np.array(np.random.random(N), dtype=np.float32)
b = np.array(np.random.random(N), dtype=np.float32)
c = np.ones_like(a)
# Use the environment variable SYCL_DEVICE_FILTER to change the default device.
# See https://github.com/intel/llvm/blob/sycl/sycl/doc/EnvironmentVariables.md#sycl_device_filter.
device = dpctl.select_default_device()
print("Using device ...")
device.print_device_info()
with dpctl.device_context(device):
da = dpt.usm_ndarray(a.shape, dtype=a.dtype, buffer="shared")
da.usm_data.copy_from_host(a.reshape((-1)).view("|u1"))
db = dpt.usm_ndarray(b.shape, dtype=b.dtype, buffer="shared")
db.usm_data.copy_from_host(b.reshape((-1)).view("|u1"))
dc = dpt.usm_ndarray(c.shape, dtype=c.dtype, buffer="shared")
driver(da, db, dc, global_size)
print("Done...")
def main():
global_size = 64
local_size = 32
N = global_size * local_size
print("N", N)
a = np.ones(N, dtype=np.float32)
b = np.ones(N, dtype=np.float32)
print("a:", a, hex(a.ctypes.data))
print("b:", b, hex(b.ctypes.data))
# Use the environment variable SYCL_DEVICE_FILTER to change
# the default device. See
# https://github.com/intel/llvm/blob/sycl/sycl/doc/EnvironmentVariables.md#sycl_device_filter.
device = dpctl.select_default_device()
print("Using device ...")
device.print_device_info()
with dpctl.device_context(device):
c = f1(a, b)
print("RESULT c:", c, hex(c.ctypes.data))
for i in range(N):
if c[i] != 2.0:
print("First index not equal to 2.0 was", i)
break
print("Done...")
def test_create_program_from_source(self):
oclSrc = " \
kernel void axpy(global int* a, global int* b, global int* c, int d) { \
size_t index = get_global_id(0); \
c[index] = d*a[index] + b[index]; \
}"
with dpctl.device_context("opencl:gpu:0"):
q = dpctl.get_current_queue()
prog = dpctl_prog.create_program_from_source(q, oclSrc)
axpyKernel = prog.get_sycl_kernel("axpy")
abuf = dpctl_mem.MemoryUSMShared(1024 * np.dtype("i").itemsize)
bbuf = dpctl_mem.MemoryUSMShared(1024 * np.dtype("i").itemsize)
cbuf = dpctl_mem.MemoryUSMShared(1024 * np.dtype("i").itemsize)
a = np.ndarray((1024), buffer=abuf, dtype="i")
b = np.ndarray((1024), buffer=bbuf, dtype="i")
c = np.ndarray((1024), buffer=cbuf, dtype="i")
a[:] = np.arange(1024)
b[:] = np.arange(1024, 0, -1)
c[:] = 0
d = 2
args = []
args.append(a.base)
args.append(b.base)
args.append(c.base)
args.append(ctypes.c_int(d))
r = [1024]
q.submit(axpyKernel, args, r)
self.assertTrue(np.allclose(c, a * d + b))
def test_kernel_atomic_simple(filter_str, input_arrays, kernel_result_pair):
a, dtype = input_arrays
kernel, expected = kernel_result_pair
device = dpctl.SyclDevice(filter_str)
with dpctl.device_context(device):
kernel[global_size, dppy.DEFAULT_LOCAL_SIZE](a)
assert a[0] == expected
