procyberian/gcc
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

aarch64: Reimplement vrsqrte* intrinsics with builtins

Browse source 
Another very simple move from inline asm to builtins.
Only two intrinsics this time.

gcc/ChangeLog:

	* config/aarch64/aarch64-simd-builtins.def (ursqrte): Define builtin.
	* config/aarch64/aarch64-simd.md (aarch64_ursqrte<mode>): New pattern.
	* config/aarch64/arm_neon.h (vrsqrte_u32): Reimplement using builtin.
	(vrsqrteq_u32): Likewise.
This commit is contained in:
Kyrylo Tkachov 2021-02-02 14:27:34 +00:00
parent 8fdfd0cfdb
commit d14cf89b94
3 changed files with 13 additions and 12 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
gcc/config/aarch64/aarch64-simd-builtins.def
View file

@ -705,6 +705,9 @@
/* Implemented by aarch64_rsqrts<mode>. */
BUILTIN_VHSDF_HSDF (BINOP, rsqrts, 0, FP)
/* Implemented by aarch64_ursqrte<mode>. */
BUILTIN_VDQ_SI (UNOPU, ursqrte, 0, NONE)
/* Implemented by fabd<mode>3. */
BUILTIN_VHSDF_HSDF (BINOP, fabd, 3, FP)

8
gcc/config/aarch64/aarch64-simd.md
View file

@ -755,6 +755,14 @@
DONE;
})
(define_insn "aarch64_ursqrte<mode>"
[(set (match_operand:VDQ_SI 0 "register_operand" "=w")
(unspec:VDQ_SI [(match_operand:VDQ_SI 1 "register_operand" "w")]
UNSPEC_RSQRTE))]
"TARGET_SIMD"
"ursqrte\\t%<v>0<Vmtype>, %<v>1<Vmtype>"
[(set_attr "type" "neon_fp_rsqrte_<stype><q>")])
(define_insn "*aarch64_mul3_elt_to_64v2df"
[(set (match_operand:DF 0 "register_operand" "=w")
(mult:DF

14
gcc/config/aarch64/arm_neon.h
View file

@ -9387,24 +9387,14 @@ __extension__ extern __inline uint32x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrsqrte_u32 (uint32x2_t __a)
{
uint32x2_t __result;
__asm__ ("ursqrte %0.2s,%1.2s"
: "=w"(__result)
: "w"(__a)
: /* No clobbers */);
return __result;
return __builtin_aarch64_ursqrtev2si_uu (__a);
}
__extension__ extern __inline uint32x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vrsqrteq_u32 (uint32x4_t __a)
{
uint32x4_t __result;
__asm__ ("ursqrte %0.4s,%1.4s"
: "=w"(__result)
: "w"(__a)
: /* No clobbers */);
return __result;
return __builtin_aarch64_ursqrtev4si_uu (__a);
}
__extension__ extern __inline int8x16_t