Hi, I've been watching these videos in addition to reading the Programming Massively Parallel Processors, My take on the exercise: (for the sake of brevity, I will not include assigning memory or memcpy for now) ```c // Kernel Function for Array Summing __global__ void sumArrays_Kernel(float *A, float *B, float *C, float *D, int Width, int Height, int Depth) { int x = blockIdx.x * blockDim.x + threadIdx.x; int y = blockIdx.y * blockDim.y + threadIdx.y; int z = blockIdx.z * blockDim.z + threadIdx.z; if (x < Width && y < Height && z < Depth) { int index = x + y * Width + z * Width * Height; // Defined as index as used twice in next line D[index] = A[index] + B[y * Width + x] + C[x]; } } void sumArrays_Host(float *A, float *B, float *C, float *D, int X, int Y, int Z) { float *A_d, *B_d, *C_d, *D_d; // Malloc and Memcpy vars (i.e A -> A_d) dim3 block(2, 2, 2); // I'm not massively sure on good sizing here dim3 grid((X + block.x - 1) / block.x, (Y + block.y - 1) / block.y, (Z + block.z - 1) / block.z); sumArraysKernel(d_A, d_B, d_C, d_D, X, Y, Z); // memcpy result back, and then free memory } ``` General idea is that we're using a different index for each input vector, based on the logic you were mentioning earlier, the block and grid logic is just making sure we're in bounds
