#include <stdio.h>
#include <stdlib.h>
#include <ctime>

#include <arm_neon.h>


#define SIZE 1024


void matmul_optimized(float** A, float** B, float** C, int n) {
    for (int i = 0; i < n; i++) {
        for (int j = 0; j < n; j++) {
            float32x4_t vecC = vdupq_n_f32(0);  
            for (int k = 0; k < n; k += 4) {
                
                float32x4_t vecA = vld1q_f32(&A[i][k]);
              
                float32x4_t vecB = vld1q_f32(&B[k][j]);

                
                vecC = vmlaq_f32(vecC, vecA, vecB);
            }
           
            C[i][j] = vgetq_lane_f32(vecC, 0) + vgetq_lane_f32(vecC, 1) +
                      vgetq_lane_f32(vecC, 2) + vgetq_lane_f32(vecC, 3);
        }
    }
}

int main() {
    
    float** A = (float**)malloc(SIZE * sizeof(float*));
    for (int i = 0; i < SIZE; i++) {
        A[i] = (float*)malloc(SIZE * sizeof(float));
    }

    float** B = (float**)malloc(SIZE * sizeof(float*));
    for (int i = 0; i < SIZE; i++) {
        B[i] = (float*)malloc(SIZE * sizeof(float));
    }

    float** C = (float**)malloc(SIZE * sizeof(float*));
    for (int i = 0; i < SIZE; i++) {
        C[i] = (float*)malloc(SIZE * sizeof(float));
    }

    
    for (int i = 0; i < SIZE; i++) {
        for (int j = 0; j < SIZE; j++) {
            A[i][j] = (float)(rand() % 100);
            B[i][j] = (float)(rand() % 100);
        }
    }

    clock_t start_time, end_time;
    
    start_time = clock();

    
    matmul_optimized(A, B, C, SIZE);

    
    end_time = clock();

    
    double elapsed_time = ((double)(end_time - start_time)) / CLOCKS_PER_SEC;
    printf("NEON优化矩阵乘法的运行时间：%f 秒\n", elapsed_time);

    
    for (int i = 0; i < SIZE; i++) {
        free(A[i]);
        free(B[i]);
        free(C[i]);
    }
    free(A);
    free(B);
    free(C);

    return 0;
}