changes

primesGPU.py

@@ -111,7 +111,7 @@ if __name__ == "__main__":
     parser.add_option("-e", "--end", dest="end",
                       help="numbers to check without even numbers", default="5000000000", type="int")
     parser.add_option("--numbers-per-step", dest="numbers_per_step",
-                      help="amount of uneven numbers checked in each step (even number are skipped)", default="4194304",
+                      help="amount of uneven numbers checked in each step (even number are skipped)", default="8000000",
                       type="int")
     parser.add_option("--output", dest="output",
                       help="name of the file, where the primes should be stored", default="primes.txt", type="string")
@@ -124,7 +124,7 @@ if __name__ == "__main__":
 
     block_size = 1024
     start = 1
-    grid_size = int(math.ceil(options.numbers_per_step / block_size))  # half of numbers per
+    grid_size = int(math.ceil(options.numbers_per_step / block_size))
     resulting_numbers_per_step = block_size * grid_size
     last_number_checked = start - 1
 

simpleCustomKernel.py

@@ -10,15 +10,14 @@ __global__ void scalar_multiply_kernel(float *outvec, float scalar, float *vec)
      int i = threadIdx.x;
      outvec[i] = scalar*vec[i];
 }
-""")
+""")  # compile kernel function
 
-scalar_multiply_gpu = ker.get_function("scalar_multiply_kernel")
+scalar_multiply_gpu = ker.get_function("scalar_multiply_kernel")  # get kernel function reference
 
-testvec = np.random.randn(512).astype(np.float32)
+host_vector = np.random.randn(512).astype(np.float32)  # create array of 512 random numbers
-testvec_gpu = gpuarray.to_gpu(testvec)
+device_vector = gpuarray.to_gpu(host_vector)  # copy into GPUs global memory
-outvec_gpu = gpuarray.empty_like(testvec_gpu)
+out_device_vector = gpuarray.empty_like(device_vector)  # allocate a chunk of empty memory to GPUs global memory
 
-scalar_multiply_gpu(outvec_gpu, np.float32(2), testvec_gpu, block=(512, 1, 1), grid=(1, 1, 1))
+scalar_multiply_gpu(out_device_vector, np.float32(2), device_vector, block=(512, 1, 1), grid=(1, 1, 1))  # launch the kernel
-print("Does our kernel work correctly? : {}".format(np.allclose(outvec_gpu.get(), 2 * testvec)))
+print("Does our kernel work correctly? : {}".format(np.allclose(out_device_vector.get(), 2 * host_vector)))
-print(outvec_gpu.get())
+print(out_device_vector.get())
-print(2 * testvec)