Performance issue on Random Number Generation

[neko-suki]

I found a problem on random number generation. It affects performance.

https://github.com/fixstars/clpy/blob/clpy/clpy/random/generator.py#L175

or self.seed_array.size != numpy.prod(size) means when we change the size of randomly generated array, initialization process will run everytime.

The following is code for measure performance.

import clpy
import time

def generate_random(size):
    tmp = clpy.random.rand(size)
    

beg = time.time()
generate_random(1) # this case is extremely slow
end = time.time()
print("time = {:.5f} msec".format(end - beg))


beg = time.time()
generate_random(100000)
end = time.time()
print("time = {:.5f} msec".format(end - beg))

beg = time.time()
generate_random(100000)
end = time.time()
print("time = {:.5f} msec".format(end - beg))

beg = time.time()
generate_random( 99999)
end = time.time()
print("time = {:.5f} msec".format(end - beg))

beg = time.time()
generate_random( 99999)
end = time.time()
print("time = {:.5f} msec".format(end - beg))

The following is the result of performance measurement on Secondary machine (titanv). When the size of array to generate random number is changed, it takes long time.

time = 0.12419 msec
time = 0.00214 msec
time = 0.00089 msec
time = 0.00152 msec
time = 0.00023 msec

[LWisteria]

@neko-suki Did #166 fix this? If so, please close.

[neko-suki]

@LWisteria OK. I will close this issue.

Improved performance is written here