CODEWITHSUNDEEP
cfloating-pointarmneonintrinsics
Aurum
Aurum

Reputation: 77

Not able to load Floating point Values using NEON Intrinsics


I am not able to load floating point values in to NEON 128 bit registers no matter what! I tried every possible way to load floating point numbers but the registers remain zero (found via debugging). 

/* neon_example.c - Neon intrinsics example program */
    #include <stdint.h>
    #include <stdio.h>
    #include <arm_neon.h>

    /* fill array with increasing integers beginning with 0 */
    void fill_array(float32_t *array, int size)
    {    int i;
        for (i = 0; i < size; i++)
        {
             array[i] = 2.0f;
             printf("%f",array[i]);
        }
    }

    /* return the sum of all elements in an array. This works by calculating 4 totals (one for each lane) and adding those at the end to get the final total */
    float sum_array(float32_t *array, int size)
    {    
         float32_t a,b,c,d,add;
         float32x4_t acc= vdupq_n_f32(0.0);
         for (; size != 0; size -= 4)
         {
           float32x4_t vec =vdupq_n_f32(0.0f);;
           vec=vld1q_f32(array);
    //The above operation does not load values??????????????????????
           array += 4;
           acc = vaddq_f32(acc,vec);
          }

         vst1q_lane_f32(&a,acc,0);
         vst1q_lane_f32(&a,acc,1);
         vst1q_lane_f32(&a,acc,2);
         vst1q_lane_f32(&a,acc,3);
         add=a+b+c+d;

          //return (int)vget_lane_s64(acc2, 0);
         return add;

    }
    /* main function */
    int main()
    {
          float32_t my_array[100];
          fill_array(my_array, 100);
          printf("Sum was %f\n", sum_array(my_array, 100));
          return 0;
    }

Upvotes: 2

Views: 1814

Answers (1)

Aurelius
Aurelius

Reputation: 11349

I just ran your code and it loads into the registers properly. I built the code using LLVM 4.2 in Xcode 4.6.

Implementing Guy Sirton's changes fixes the bug and results in a more readable function:

float32_t sum_array(float32_t *array, int size)
{
    float32_t arr[4],add;
    float32x4_t acc= vdupq_n_f32(0.0);
    for (; size != 0; size -= 4)
    {
        float32x4_t vec=vld1q_f32(array);
        array += 4;
        acc = vaddq_f32(acc,vec);
    }

    vst1q_f32(arr, acc);

    add = arr[0] + arr[1] + arr[2] + arr[3];

    return add; 
}

Upvotes: 2

Related Questions