#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# 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 numpy as np
import pycuda.driver as drv
from nervanagpu import NervanaGPU
from pycuda.autoinit import context
from scikits.cuda import cublas
print context.get_device().name()
handle = cublas.cublasCreate()
start, end = (drv.Event(), drv.Event())
def cublas_dot(A, B, C, alpha=1.0, beta=0.0, repeat=1):
lda = max(A.strides) // 4
ldb = max(B.strides) // 4
ldc = max(C.strides) // 4
opA = 't' if A.is_trans else 'n'
opB = 't' if B.is_trans else 'n'
op = opB + opA
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# 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 numpy as np
import pycuda.driver as drv
from nervanagpu import NervanaGPU
from pycuda.autoinit import context
from scikits.cuda import cublas
print(context.get_device().name())
start, end = (drv.Event(), drv.Event())
handle = cublas.cublasCreate()
def cublas_dot(A, B, C, alpha=1.0, beta=0.0, repeat=1):
lda = max(A.strides) // 4
ldb = max(B.strides) // 4
ldc = max(C.strides) // 4
opA = 't' if A.is_trans else 'n'
opB = 't' if B.is_trans else 'n'
op = opB + opA
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# 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 numpy as np
import pycuda.driver as drv
from nervanagpu import NervanaGPU
from pycuda.autoinit import context
from operator import mul
print context.get_device().name()
np.set_printoptions(threshold=8193, linewidth=600, formatter={'int':lambda x: "%10d" % x,'float':lambda x: "% .3f" % x})
dtype = np.float16
cpu = 1
repeat = 1
ng = NervanaGPU(stochastic_round=False, bench=True)
pool = ng.pool_layer(
"max",
64, # N
64,1,64,64, # C,D,H,W
4,1,2,2, # J,T,R,S
0,0,0,0, # padding
# You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # 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 numpy as np import pycuda.driver as drv from pycuda.autoinit import context from nervanagpu import NervanaGPU from nervanagpu.layers import DataLayer, ConvLayer, PoolLayer, FullLayer print(context.get_device().name()) # Compare results here: # https://github.com/soumith/convnet-benchmarks # number of full iterations loops = 10 # show bechmark details for each layer layer_bench = 0 # show layer stats after each operation print_stats = 0 ng = NervanaGPU(bench=layer_bench) # don't learn, just benchmark momentum = 0.0