# module of support routines for testing
import numpy as np
import gpu
# initialize default CUDA device
gpu.device()
# we will be testing linear algebra on CUDA we we need an approximation
def arrays_equal(a, b, epsilon=1e-05):
return np.allclose(a,b, epsilon)
#return (abs(a-b) < epsilon).all()
def close(a, b, rtol=1e-05, atol=1e-08):
return abs(a - b) <= (atol + rtol * abs(b))
def scalars_equal(a, b, epsilon=0.00004):
#print a, b, abs(a-b)
return close(a, b, epsilon)
#return abs(a-b) < epsilon
def with_device():
with gpu.device(0) as cuda0:
A = cn.ones((1024,8), dtype=cn.float32)
print A
