Create simple_element_kernel_example0.py

Chapter03/simple_element_kernel_example0.py

@@ -0,0 +1,31 @@
+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()