ARM NEON C++ Cheat Sheet

November 19, 2015

Newer ARM processors have their own flavor of SIMD instructions called NEON. In my little Android application Arashi, NEON is used a lot to speed up the simulation of particles.

Here is a table explaining some of the NEON functions that are used:

C++ NEON functions

Neon	Explanation	Pseudocode
`vdupq_n_f32(a)`	New NEON value	`a`
`vsubq_f32(a, b)`	Subtract	`a - b`
`vaddq_f32(a, b)`	Add	`a + b`
`vmulq_f32(a, b)`	Multiply	`a * b`
`vmlaq_f32(a, b, c)`	Multiply and add	`a + (b * c)`
`vmlsq_f32(a, b, c)`	Multiply and subtract	`a - (b * c)`
`vrsqrteq_f32(a)`	Reciprocal square root	`1 / sqrt(a)`
`vcgtq_f32(a, b)`	Compare greater than	`a > b ? 1 : 0`
`vcltq_f32(a, b)`	Compare less than	`a < b ? 1 : 0`
`vbslq_f32(mask, a, b)`	Select by mask	`mask != 0 ? a : b`
`vminq_f32(a, b)`	Get minimum	`a < b ? a : b`
`vmaxq_f32(a, b)`	Get maximum	`a > b ? a : b`

Tags

= [android, c++, intrinsics, NEON, SIMD]

Comments

Gal Hai - October 8, 2018

Slight Typo:

vcgtq_f32(a, b) Compare less than a < b ? 1 : 0

Guess you meant vcltq_f32.

Kevin - December 15, 2018 > Replying to Gal Hai

You’re right, thanks for letting me know! I’ve updated the post