1D Convolution - 阿东阿言

归档25

LeetGPU18 计组6 CPP1

239 字

1 分钟

1D Convolution

2026-07-14

解析#

该题要实现一个执行一维卷积运算的程序。

给定一个输入数组和一个卷积核（滤波器），计算卷积后的输出。卷积运算应满足“有效”边界条件，即卷积核仅应用于与输入完全重叠的区域。

卷积操作其实很简单，就是将卷积核和数组内元素先求积再求和而已。

答案#

1
#include <cuda_runtime.h>
2

3
__global__ void convolution_1d_kernel(const float* input, const float* kernel, float* output,
4
    int input_size, int kernel_size) {
5
        int idx = blockIdx.x * blockDim.x + threadIdx.x;
6
        int output_size = input_size - kernel_size + 1;
7
        if (idx < output_size) {
8
            float sum = 0.0f;
9
            for (int i = 0; i < kernel_size; i++) {
10
                sum += input[idx + i] * kernel[i];
11
            }
12
            output[idx] = sum;
13
        }
14
    }
15

16
// input, kernel, output are device pointers (i.e. pointers to memory on the GPU)
17
extern "C" void solve(const float* input, const float* kernel, float* output, int input_size,
18
                      int kernel_size) {
19
    int output_size = input_size - kernel_size + 1;
20
    int threadsPerBlock = 256;
21
    int blocksPerGrid = (output_size + threadsPerBlock - 1) / threadsPerBlock;
22

23
    convolution_1d_kernel<<<blocksPerGrid, threadsPerBlock>>>(input, kernel, output, input_size,
24
                                                              kernel_size);
25
    cudaDeviceSynchronize();
26
}