Macintosh-Tools
/
Pomme
mirror of https://github.com/jorio/Pomme.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

Rewrite __frsqrte with Arm/NEON counterpart 

__frsqrte is the intrinsic for floating reciprocal square root estimate.
In Arm64, we can implement it with NEON intrinsics. Since Armv8.2,
instruction "FRSQRTE" [1] is provided to calculate an approximate square
root for each vector element in the source SIMD and FP register.

With -O3, generated assembly on Apple Silicon M1:

[original]
	fsqrt	s0, s0
	fmov	s1, #1.00000000
	fdiv	s0, s1, s0

[neon]
	dup.4s	v0, v0[0]
	frsqrte.4s	v0, v0

[1] https://developer.arm.com/documentation/100076/0100/a64-instruction-set-reference/a64-simd-scalar-instructions/frsqrte--scalar-
This commit is contained in:
Jim Huang 2022-01-29 22:21:01 +08:00 committed by Iliyas Jorio
parent 1b0ea49ded
commit 86d35caf94
1 changed files with 8 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
src/Pomme.h
View File

@ -16,7 +16,15 @@
#define __fres(x) (1.0f/x)
#define __fabs(x) fabs(x)
#if defined(__aarch64__)
#include <arm_neon.h>
static inline float __frsqrte(float f)
{
return vrsqrteq_f32(vdupq_n_f32(f))[0];
}
#else
#define __frsqrte(x) (1.0f/sqrtf(x))
#endif
//-----------------------------------------------------------------------------
// Source code compat