0

I have been found a lib but there was not

void atan2fv_neon_hfp(float *y, float *x,float *res,int len)

to calculate len floats once.

How can I write a neon version for atan2fv_neon_hfp ?

Paul R
  • 208,748
  • 37
  • 389
  • 560
WateLemon
  • 11
  • Note [the accuracy of **NEON**](http://stackoverflow.com/questions/7346521/subnormal-ieee-754-floating-point-numbers-support-on-ios-arm-devices-iphone-4) is not the same as the standard libraries. – artless noise Sep 18 '13 at 14:31

1 Answers1

0

Glancing over the implementation of atan2f_neon_hfp it doesn't look like there's anything that wouldn't be possible to trivially vectorize (by loading pairs into d16 and d17, instead of vdup'ing into them), but you'll need to know ARM/NEON assembly. Alternatively, you could just call the scalar version in a loop.

Sneftel
  • 40,271
  • 12
  • 71
  • 104
  • as math-neon,atan2f_neon_hfp is not really fast for multi floats , i want to found a way to calculate 4x floats everytime use neon.I am a newbie in neon.So i want to found a example to learn more skill in neon.scalar version in a loop can not solve my problem.it still slow for multi floats.. – WateLemon Sep 19 '13 at 02:54