Added CUDA-streams to Darknet-Yolo forward inference

AlexeyAB

2017-12-31 9d23aad8696268e8ce3a94fee9490fd1db000dc8

 src/im2col_kernels.cu

@@ -54,7 +54,7 @@
    int width_col = (width + 2 * pad - ksize) / stride + 1;
    int num_kernels = channels * height_col * width_col;
    im2col_gpu_kernel<<<(num_kernels+BLOCK-1)/BLOCK,
        BLOCK>>>(
        BLOCK, 0, get_cuda_stream()>>>(
                num_kernels, im, height, width, ksize, pad,
                stride, height_col,
                width_col, data_col);