#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include<math.h>
#include <smmintrin.h>

float dot_product(const float *A, const float *B, int startA, int endA, int startB, int endB){
    float f[4];
    f[0] = 0.0;
    f[1] = 0.0;
    f[2] = 0.0;
    f[3] = 0.0;
    int i = 5;
    __m128 result = _mm_loadu_ps (f);
    float res = 0;
    while(startA + 3 < endA) {
        __m128 value1 = _mm_loadu_ps (A+startA);
        __m128 value2 = _mm_loadu_ps (B+startB);
        __m128 val = _mm_dp_ps(value1, value2, 0xF1);
        result = _mm_add_ss(result, val);
        startA +=4;
        startB +=4;
    }
    printf("%d\n", i);
    _mm_storeu_ps (&res, result);
    printf("%d\n", i);
    while (startA < endA) {
        // printf("%d %d %d\n", i, t, endA);
        // printf("%f %f\n", A[i], B[t]);
        res += A[startA]*B[startB];
        ++startA;
        ++startB;
    }
    return res;
}

void multiply(float*A, float*B, float*res, int N, int M) {
    for (int i = 0; i < N; i++) {
        for (int j = 0; j < N; j++) {
            double value = dot_product(A, B, i*M, (i+1)*M, j*M, (j+1)*M);
            res[i*N+j] = value;
        }
    }
}
void Print(float* res, int N, int M) {
    for (int i = 0; i < N; i++) {
        for (int j = 0; j < M; j++) {
            printf("%.4f ", res[i*M + j]);
        }
        printf("\n");
    }
}

int main() {
    int N, M;
   // scanf("%d %d", &N, &M); //N -количество строк, M - количество столбцов
    for(N = 7; N < 10; ++N) {
          for(M = 1; M < 10; M++) {
    printf("%d %d\n", N, M);
    //float *A = (float*) calloc(N*M, 4); //    }

    //float *B = (float*)calloc(N*M, 4);
    //float *res = (float*) calloc(N*M, 4);
    float A[M*N], B[N*M], res[N*M];
    for (int i = 0; i < N; i++) {
        for (int j = 0; j < M; j++) {
           A[i*M+j] = i;//scanf("%f", &A[i*M + j]);
        }
    }

    for (int i = 0; i < M; i++) {
        for (int j = 0; j < N; j++) {
           B[j*M + i] = j;// scanf("%f", &B[j*M + i]);
        }
    }

    multiply(A, B,res,  N, M);
    //Print(res, N, N);
    //free(A);
    //free(B);
    //free(res);
          }
          }
    return 0;
}