reduction_sum.py 文件源码

def hist_cuda_test():

    histogram_array = src1#np.zeros(vectorSize*BIN_COUNT, dtype=np.int32).reshape(vectorSize,BIN_COUNT)
    histogram = src1[SEARCH_INDEX]#np.zeros(BIN_COUNT, dtype=np.float32)
    results = np.zeros(9, dtype=np.float64)

    # use stream to trigger async memory transfer
    cstream = cuda.stream()
    ts = timer()
    # Increase Counter to measure the Efficiency
    count = 1
    for i in range(count):
        with cstream.auto_synchronize():

            # For Histogram Compairision.
            d_histogram_array = cuda.to_device(histogram_array, stream=cstream)
            d_histogram = cuda.to_device(histogram, stream=cstream)
            d_results = cuda.to_device(results, stream=cstream)

            hist_comp[1, vectorSize, cstream](d_histogram_array,d_histogram,d_results)

            d_histogram_array.copy_to_host(histogram_array, stream=cstream)
            d_histogram.copy_to_host(histogram, stream=cstream)
            d_results.copy_to_host(results, stream=cstream)

    te = timer()
    print('GPU Process ',count," Iterations : in ", te - ts)
    print('histogram is')
    print(results)