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_get_current_backend():
dpctl.get_current_backend()
dpctl.get_current_device_type()
q = dpctl.SyclQueue()
dpctl.set_global_queue(q)
if has_gpu():
dpctl.set_global_queue("gpu")
elif has_cpu():
dpctl.set_global_queue("cpu")
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import dpctl
import syclbuffer as sb
import numpy as np
X = np.full((10**4, 4098), 1e-4, dtype="d")
# warm-up
print("=" * 10 + " Executing warm-up " + "=" * 10)
print("NumPy result: ", X.sum(axis=0))
dpctl.set_global_queue("opencl:cpu")
print("SYCL({}) result: {}".format(
dpctl.get_current_queue().get_sycl_device().get_device_name(),
sb.columnwise_total(X),
))
dpctl.set_default_queue("opencl:gpu")
print("SYCL({}) result: {}".format(
dpctl.get_current_queue().get_sycl_device().get_device_name(),
sb.columnwise_total(X),
))
import timeit
print("Times for 'opencl:cpu'")
print(
# GPU computation
Xgpu = bs.black_scholes_price(opts)
# compute prices in CPython
X_ref = np.array([ref_python_black_scholes(*opt) for opt in opts], dtype="d")
print(np.allclose(Xgpu, X_ref, atol=1e-5))
n_opts = 3 * 10**6
# compute on CPU sycl device
import timeit
for _ in range(3):
dpctl.set_global_queue("opencl:cpu:0")
print("Using : {}".format(
dpctl.get_current_queue().get_sycl_device().get_device_name()))
t0 = timeit.default_timer()
opts1 = gen_option_params(n_opts, 20.0, 30.0, 22.0, 29.0, 18.0, 24.0, 0.01,
0.05, 0.01, 0.05, "d")
X1 = bs.black_scholes_price(opts1)
t1 = timeit.default_timer()
print("Elapsed: {}".format(t1 - t0))
# compute on GPU sycl device
dpctl.set_global_queue("level_zero:gpu:0")
print("Using : {}".format(
dpctl.get_current_queue().get_sycl_device().get_device_name()))