def test_timer():
if gpu.lib.pt_clusterNet == gpu.lib.pt_clusterNetCPU: return
t = gpu.Timer()
A = gpu.rand(100,100)
B = gpu.rand(100,100)
C = gpu.rand(100,100)
time = 0
t.tick()
for i in range(10):
gpu.dot(A,B,C)
time = t.tock()
assert time > 0
time = 0
t.tick("Timer test")
gpu.dot(A,B,C)
time = t.tock("Timer test")
assert time > 0
accumulative_time = 0
for i in range(100):
t.tick('cumulative')
gpu.dot(A,B,C)
t.tick('cumulative')
accumulative_time = t.tock('cumulative')
assert accumulative_time > 5*time
def test_timer():
if gpu.lib.pt_clusterNet == gpu.lib.pt_clusterNetCPU: return
t = gpu.Timer()
A = gpu.rand(100, 100)
B = gpu.rand(100, 100)
C = gpu.rand(100, 100)
time = 0
t.tick()
for i in range(10):
gpu.dot(A, B, C)
time = t.tock()
assert time > 0
time = 0
t.tick("Timer test")
gpu.dot(A, B, C)
time = t.tock("Timer test")
assert time > 0
accumulative_time = 0
for i in range(100):
t.tick('cumulative')
gpu.dot(A, B, C)
t.tick('cumulative')
accumulative_time = t.tock('cumulative')
assert accumulative_time > 5 * time
def test_rand():
A = gpu.rand(100, 100)
B = A.tocpu()
assert np.mean(B) > 0.45 and np.mean(B) < 0.55
assert np.var(B) > 0.08 and np.var(B) < 0.085
gpu.setRandomState(1234)
A1 = gpu.rand(1000, 1000)
gpu.setRandomState(1234)
A2 = gpu.rand(1000, 1000)
B1 = A1.tocpu()
B2 = A2.tocpu()
t.assert_array_equal(B1, B2, "Seed not working!")
def test_rand():
A = gpu.rand(100,100)
B = A.tocpu()
assert np.mean(B) > 0.45 and np.mean(B) < 0.55
assert np.var(B) > 0.08 and np.var(B) < 0.085
gpu.setRandomState(1234)
A1 = gpu.rand(1000,1000)
gpu.setRandomState(1234)
A2 = gpu.rand(1000,1000)
B1 = A1.tocpu()
B2 = A2.tocpu()
t.assert_array_equal(B1, B2, "Seed not working!")
@author: tim
'''
import cluster_net as gpu
import cudanet as gpu2
t = gpu.Timer()
dim1 = 16
dim_inner = 1024
dim_outer = 1024
batch_first_mode = True
if batch_first_mode:
input = gpu.rand(dim1,dim_inner)
W = gpu.rand(dim_inner,dim_outer)
output = gpu.rand(dim1,dim_outer)
input2 = gpu2.random.rand(dim1,dim_inner)
W2 = gpu2.random.rand(dim_inner,dim_outer)
output2 = gpu2.random.rand(dim1,dim_outer)
else:
input = gpu.rand(dim_inner,dim1)
W = gpu.rand(dim_outer,dim_inner)
output = gpu.rand(dim_outer,dim1)
input2 = gpu2.random.rand(dim_inner,dim1)
W2 = gpu2.random.rand(dim_outer,dim_inner)
output2 = gpu2.random.rand(dim_outer,dim1)
'''
Created on Dec 31, 2015
@author: tim
'''
import cluster_net as gpu
import time
t = gpu.Timer()
dim1 = 128
dim_inner = 32
dim_outer = 256
for i in range(1000):
dim_inner += 32
A = gpu.rand(dim1, dim_inner)
B = gpu.rand(dim_inner, dim_outer)
C = gpu.rand(dim1, dim_outer)
if dim_inner > 0: iters = 1000
if dim_inner > 100: iters = 100
if dim_inner > 1000: iters = 10
if dim_inner > 3000: iters = 4
#warmup
for j in range(2):
gpu.dot(A, B, C)
t.tick(str(dim_inner))
for j in range(iters):
gpu.dot(A, B, C)
sec = t.tock(str(dim_inner)) / 1000.
tilesA = (dim1 / 16) * ((dim_inner / 64) +
outputreated on Jan 3, 2016
@author: tim
'''
import cluster_net as gpu
import cudanet as gpu2
t = gpu.Timer()
dim1 = 16
dim_inner = 1024
dim_outer = 1024
batch_first_mode = True
if batch_first_mode:
input = gpu.rand(dim1, dim_inner)
W = gpu.rand(dim_inner, dim_outer)
output = gpu.rand(dim1, dim_outer)
input2 = gpu2.random.rand(dim1, dim_inner)
W2 = gpu2.random.rand(dim_inner, dim_outer)
output2 = gpu2.random.rand(dim1, dim_outer)
else:
input = gpu.rand(dim_inner, dim1)
W = gpu.rand(dim_outer, dim_inner)
output = gpu.rand(dim_outer, dim1)
input2 = gpu2.random.rand(dim_inner, dim1)
W2 = gpu2.random.rand(dim_outer, dim_inner)
output2 = gpu2.random.rand(dim_outer, dim1)
mean_time = 0
'''
Created on Dec 31, 2015
@author: tim
'''
import cluster_net as gpu
import time
t = gpu.Timer()
dim1 = 128
dim_inner = 32
dim_outer = 256
for i in range(1000):
dim_inner += 32
A = gpu.rand(dim1,dim_inner)
B = gpu.rand(dim_inner,dim_outer)
C = gpu.rand(dim1,dim_outer)
if dim_inner > 0: iters = 1000
if dim_inner > 100: iters = 100
if dim_inner > 1000: iters = 10
if dim_inner > 3000: iters = 4
#warmup
for j in range(2):
gpu.dot(A,B,C)
t.tick(str(dim_inner))
for j in range(iters):
gpu.dot(A,B,C)
sec = t.tock(str(dim_inner))/1000.
tilesA = (dim1/16)*((dim_inner/64) + (1 if dim_inner % 64 > 0 else 0))