CUDA 向右求和

Question

我正在尝试使用 CUDA 实现总和减少，但是我希望减少在右侧而不是左侧.. 我写了下面的代码，但我不确定为什么它不起作用

__global__ void reduce_kernel(
    float *input,
    float *partialSums,
    unsigned int N) 
{
    unsigned int segment = blockIdx.x * blockDim.x * 2;
    unsigned int i = segment + threadIdx.x;
    __shared__ float input_s[BLOCK_DIM];

    input_s[threadIdx.x] = input[i] + input[i + BLOCK_DIM];
    int count = 2;
    __syncthreads();

    for (unsigned int stride = BLOCK_DIM / 2; 
         stride < BLOCK_DIM;
         stride = stride + (BLOCK_DIM / count)) 
    {
        if (threadIdx.x >= stride) {
            count = count * 2;
            input_s[threadIdx.x] += input_s[threadIdx.x - stride];
            printf("%d  ", stride);
            __syncthreads();
            if (stride == BLOCK_DIM - 1) {
                break;
            }
        }
        __syncthreads();
    }

    if (threadIdx.x == BLOCK_DIM - 1) {
        partialSums[blockIdx.x] = input_s[threadIdx.x];
    }
}

任何想法我做错了什么？

Answer 1

只要输入的元素数量是 2 的幂，这应该完全符合您的要求。部分总和应该在右边结束。这种算法的步幅必须从1 增长到BLOCK_DIM / 2（产生更多的warp 分歧）或从BLOCK_DIM / 2 缩小到1。无论哪种方式，它都应该乘以/除以2。

__global__ void reduce_kernel(
    float *input,
    float *partialSums,
    unsigned int N) 
{
    unsigned int segment = blockIdx.x * blockDim.x * 2;
    unsigned int i = segment + threadIdx.x;
    __shared__ float input_s[BLOCK_DIM];

    input_s[threadIdx.x] = input[i] + input[i + BLOCK_DIM];
    __syncthreads();

    for (unsigned int stride = BLOCK_DIM / 2; 
         stride > 0;
         stride /= 2) 
    {
        if (threadIdx.x >= BLOCK_DIM - stride) {
            input_s[threadIdx.x] += input_s[threadIdx.x - stride];
        }
        __syncthreads();
    }

    if (threadIdx.x == BLOCK_DIM - 1) {
        partialSums[blockIdx.x] = input_s[threadIdx.x];
    }
}

条件中的__syncthreads(); 是另一个错误，因为块的所有线程都必须参与同步。否则会导致未定义的行为。