parent
b6c8c05f1b
commit
d584c56c32
@ -0,0 +1,31 @@
|
||||
#include <arm_neon.h>
|
||||
#include <stdio.h>
|
||||
#define SIZE 1024
|
||||
#include<ctime>
|
||||
#include<iostream>
|
||||
#include <stdlib.h>
|
||||
#include <time.h>
|
||||
|
||||
|
||||
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];
|
||||
float B[SIZE];
|
||||
float C[SIZE];
|
||||
for(int i=0;i<SIZE;++i){
|
||||
A[i]=(float)rand() / (float)RAND_MAX * 100;
|
||||
B[i]=(float)rand() / (float)RAND_MAX * 100;
|
||||
}
|
||||
clock_t start=clock();
|
||||
vector_add_optimized(A,B,C,SIZE);
|
||||
clock_t end=clock();
|
||||
printf("Initial vector addition time: %lf seconds\n", (double)(end - start) / CLOCKS_PER_SEC);
|
||||
return 0;
|
||||
}
|
Loading…
Reference in new issue