Calculate weight matrix in CUDA

gistfile1.c

__global__ void calculateOutputForLayer(float *inputs, float *weight_matrix, int n_inputs, int n_outputs, float *output_matrix) {
    output_matrix[n_outputs * threadIdx.y + threadIdx.x] = inputs[threadIdx.y] * weight_matrix[n_outputs * threadIdx.y + threadIdx.x];
}

int main() {
    // Build inputs and weights
    int n_inputs = 4;
    int n_outputs = 2;
    float inputs[4] = {1,2,3,4};
    float weights[8] = {2,4,2,2,2,7,2,2};

    // Build output matrix
    float *d_output;
    cudaMalloc((void**)&d_output, sizeof(float) * n_inputs * n_outputs);

    // Copy layers
    float *d_inputs;
    cudaMalloc((void**)&d_inputs, sizeof(float) * 4);
    cudaMemcpy(d_inputs, &inputs, sizeof(float) * 4, cudaMemcpyHostToDevice);

    float *d_weights;
    cudaMalloc((void**)&d_weights, sizeof(float) * 8);
    cudaMemcpy(d_weights, &weights, sizeof(float) * 8, cudaMemcpyHostToDevice);

    // Run compute kernel
    dim3 block(n_outputs, n_inputs);
    calculateOutputForLayer<<<1, block>>>(d_inputs, d_weights, n_inputs, n_outputs, d_output);

    // Copy back output
    float *output = (float *)malloc(sizeof(float) * n_inputs * n_outputs);
    cudaMemcpy(output, d_output, sizeof(float) * n_inputs * n_outputs, cudaMemcpyDeviceToHost);
    for (int i=0; i < n_inputs * n_outputs; i++) {
        printf("%f\n", output[i]);
    }

    cudaDeviceReset();

    return 0;
}