https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109046
Bug ID: 109046
Summary: 13 regresion caused by complex lower.
Product: gcc
Version: 13.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: crazylht at gmail dot com
Target Milestone: ---
commit f9d4c3b45c5ed5f45c8089c990dbd4e181929c3d
Author: liuhongt <[email protected]>
Date: Tue Jul 19 17:24:52 2022 +0800
Lower complex type move to enable vectorization for complex type
load&store.
2022-07-20 Richard Biener <[email protected]>
Hongtao Liu <[email protected]>
gcc/ChangeLog:
PR tree-optimization/106010
* tree-complex.cc (init_dont_simulate_again): Lower complex
type move.
(expand_complex_move): Also expand COMPLEX_CST for rhs.
My r13-1762-gf9d4c3b45c5ed5 lower complex type for load&store but somehow
caused a regression in phoronix benchmark
#include <complex.h>
extern void push1(void *p, float _Complex x);
void foo(void * _q, float _Complex * _x)
{
push1(_q, *_x);
}
foo(void*, float _Complex*):
vmovss 4(%rsi), %xmm0
vmovss (%rsi), %xmm1
vmovss %xmm0, -12(%rsp)
vmovss %xmm1, -16(%rsp)
vmovq -16(%rsp), %xmm0
jmp push1(void*, float _Complex)
gcc12 generates
foo(void*, float _Complex*):
vmovq (%rsi), %xmm0
jmp push1(void*, float _Complex)
Since complex lower is only for vectorization, maybe it should be gimplified
back to a complex load after vectorization if there're things like
_6 = REALPART_EXPR <*_x_3(D)>;
_7 = IMAGPART_EXPR <*_x_3(D)>;
_1 = COMPLEX_EXPR <_6, _7>;
