neon.c

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

#define SIZE (4*1000)

void double_elements(unsigned int *ptr, unsigned int size);

int main(void)
{	
	struct timeval tv1,tv2;
	unsigned int i;
	uint32_t input_array[SIZE];
	uint32_t *ptr=input_array;

	for(i=0;i<SIZE;i++)
	{
		input_array[i]=i+1;
#ifdef DEBUG
		printf("Input:%d\n",input_array[i]);
#endif
	}
	gettimeofday(&tv1,NULL);
	double_elements(input_array,SIZE);
	gettimeofday(&tv2,NULL);
	printf("\n");
#ifdef DEBUG
	for(i=0;i<SIZE;i++)
		printf("Output:%d\n",input_array[i]);
#endif
	printf("Time for compute : %d microseconds\n", tv2.tv_usec - tv1.tv_usec +1000000 * (tv2.tv_sec - tv1.tv_sec));

	return 0;
	
}

 void  double_elements(unsigned int *ptr, unsigned int size)
{	
	unsigned int i;
	uint32x4_t Q0,Q1;
	for( i=0;i<(size/4);i++)
	{
		Q0=vld1q_u32(ptr);
		Q1=vshlq_n_u32(Q0,(const int)1);//Q1=vaddq_u32(Q0,Q0);
		vst1q_u32(ptr,Q1);
		ptr+=4;
	}
	return;
}

arm-v7a8-linux-gnueabi-gcc -mfpu=neon neon.c -o neon