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

#define SIZE 1024

void inArry(float a[]) {
	srand(time(NULL));
	for (int i = 0; i < SIZE; i++)
	{
		a[i] = rand() % 100;
	}
}

void vector_add_optimized(float* A, float* B, float* C, int size) {
	for (int i = 0; i < SIZE; i += 4) {
		float32x4_t vecA = vld1q_f32(&A[i]);
		float32x4_t vecB = vld1q_f32(&B[i]);
		float32x4_t vecC = vaddq_f32(vecA, vecB);
		vst1q_f32(&C[i], vecC);
	}
}

int main() {
	float A[SIZE], B[SIZE], C[SIZE];
	inArry(A);
	inArry(B);
	clock_t start = clock();
	vector_add_optimized(A, B, C, SIZE);
	clock_t end = clock();
	printf("%f", (double)(end - start) / CLOCKS_PER_SEC);
}