#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <arm_neon.h>

typedef struct {
    int* values;      
    int* rowIndex;    
    int* colIndex;    
    int nonZeroCount; 
} SparseMatrix;

void sparseMatmul(SparseMatrix* A, SparseMatrix* B, SparseMatrix* C) {
    int currentIndex = 0;
    int i, j;
    for (i = 0; i < A->nonZeroCount; i++)
    {
        int rowA = A->rowIndex[i];
        int colA = A->colIndex[i];
        float valueA = A->values[i];
        for (j = 0; j < A->nonZeroCount; j++)
        {
            int rowB = B->rowIndex[j];
            int colB = B->colIndex[j];
            float valueB = B->values[j];
            if (colA == rowB)
            {
                float product = valueA * valueB;
                int found = 0;
                int k;
                for (k = 0; k < currentIndex; k++)
                {
                    if (C->rowIndex[k] == rowA && C->colIndex[k] == colB){
                        C->values[k] += product;
                        found = 1;
                        break;
                    }
                }
                if (!found)
                {
                    C->values[currentIndex] = product;
                    C->rowIndex[currentIndex] = rowA;
                    C->colIndex[currentIndex] = colB;
                    currentIndex++;
                }
            }
        }
    }
    C->nonZeroCount = currentIndex;
}

void generate(SparseMatrix* matrix, int rows, int cols, int nonZeroCount){
    matrix->values = (int*)malloc(sizeof(int) * nonZeroCount);
    matrix->rowIndex = (int*)malloc(sizeof(int) * nonZeroCount);
    matrix->colIndex = (int*)malloc(sizeof(int) * nonZeroCount);   
    matrix->nonZeroCount = nonZeroCount;
    int i;
    for (i = 0; i < nonZeroCount; i++)
    {
        matrix->rowIndex[i] = rand() % rows;
        matrix->colIndex[i] = rand() % cols;
        matrix->values[i] = rand() %100;
    }
}

void matmulNEON(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 = vmovq_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);
        }
    }
}

void free_matrix(SparseMatrix* matrix) {
    free(matrix->values);
    free(matrix->rowIndex);
    free(matrix->colIndex);
}

int main() {
    srand(time(NULL));

    // int i;
    int rowsA = 1000;
    int rowsB = 2000;
    int colsB = 1000;
    int nonZeroCountA = 10000;
    int nonZeroCountB = 10000;

    SparseMatrix A, B;
    generate(&A, rowsA, rowsB, nonZeroCountA);
    generate(&B, rowsB, colsB, nonZeroCountB);

    // for (i = 0; i < A.nonZeroCount; ++i) {
    //     printf("A[%d][%d] = %d\n", A.rowIndex[i], A.colIndex[i], A.values[i]);
    // }

    // for (i = 0; i < B.nonZeroCount; ++i) {
    //     printf("B[%d][%d] = %d\n", B.rowIndex[i], B.colIndex[i], B.values[i]);
    // }
    
    int i;
    float** matrixA = (float**)malloc(rowsA * rowsB * sizeof(float*));
    float** matrixB = (float**)malloc(rowsB * colsB * sizeof(float*));
    float** matrixC = (float**)malloc(rowsA * colsB * sizeof(float*));
    for (i = 0; i < nonZeroCountA; i++)
    {
        int row = A.rowIndex[i];
        int col = A.colIndex[i];
        matrixA[row][col] = A.values[i];
    }
    for (i = 0; i < nonZeroCountB; i++)
    {
        int row = B.rowIndex[i];
        int col = B.colIndex[i];
        matrixA[row][col] = B.values[i];
    }
    

    
    SparseMatrix C;
    C.values = (int*)malloc(A.nonZeroCount * B.nonZeroCount * sizeof(int));
    C.rowIndex = (int*)malloc(A.nonZeroCount * B.nonZeroCount * sizeof(int));
    C.colIndex = (int*)malloc(A.nonZeroCount * B.nonZeroCount * sizeof(int));
    C.nonZeroCount = 0;

    clock_t start_time1 = clock();
    sparseMatmul(&A, &B, &C);
    clock_t end_time1 = clock();
    double time1 = (double)(end_time1 - start_time1) / CLOCKS_PER_SEC;
    printf("sparseMatrix %f\n", time1);

    clock_t start_time2 = clock();
    matmulNEON(matrixA, matrixB, matrixC, rowsA * colsB)
    clock_t end_time2 = clock();
    double time2 = (double)(end_time2 - start_time2) / CLOCKS_PER_SEC;
    printf("sparseMatrix %f\n", time2);

    
    // for (i = 0; i < C.nonZeroCount; ++i) {
    //     printf("C[%d][%d] = %d\n", C.rowIndex[i], C.colIndex[i], C.values[i]);
    // }

    free_matrix(&A);
    free_matrix(&B);
    free_matrix(&C);
    return 0;
}