mirror of
https://github.com/PacktPublishing/Hands-On-GPU-Programming-with-CUDA-C-and-Python-3.x-Second-Edition.git
synced 2025-07-21 21:01:06 +02:00
32 lines
931 B
Python
32 lines
931 B
Python
import numpy as np
|
|
import pycuda.autoinit
|
|
from pycuda import gpuarray
|
|
from time import time
|
|
from pycuda.elementwise import ElementwiseKernel
|
|
|
|
host_data = np.float32( np.random.random(50000000) )
|
|
|
|
gpu_2x_ker = ElementwiseKernel(
|
|
"float *in, float *out",
|
|
"out[i] = 2*in[i];",
|
|
"gpu_2x_ker")
|
|
|
|
def speedcomparison():
|
|
t1 = time()
|
|
host_data_2x = host_data * np.float32(2)
|
|
t2 = time()
|
|
print('total time to compute on CPU: %f' % (t2 - t1))
|
|
device_data = gpuarray.to_gpu(host_data)
|
|
# allocate memory for output
|
|
device_data_2x = gpuarray.empty_like(device_data)
|
|
t1 = time()
|
|
gpu_2x_ker(device_data, device_data_2x)
|
|
t2 = time()
|
|
from_device = device_data_2x.get()
|
|
print('total time to compute on GPU: %f' % (t2 - t1))
|
|
print('Is the host computation the same as the GPU computation? : {}'.format(np.allclose(from_device, host_data_2x) ))
|
|
|
|
|
|
if __name__ == '__main__':
|
|
speedcomparison()
|