https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108073
Bug ID: 108073
Summary: [rs6000] sub-optimal float member accessing on struct
parameter
Product: gcc
Version: 13.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: guojiufu at gcc dot gnu.org
Target Milestone: ---
For the below code:
typedef struct DF {double a[4]; long l; } DF;
double __attribute__ ((noipa)) foo_df (DF arg){return arg.a[3];}
At -O2, with gcc trunk(13.0), we get below sequence:
std 6,-24(1)
ori 2,2,0
lfd 1,-24(1)
blr
Actually, just one "mtvsrd 1, 6" is enough.
In this case, the argument is passed through integer registers.
For below code, it is similar:
typedef struct SF {float a[4];short l; } SF;
float foo (SF arg){return arg.a[3];}
std 4,-24(1)
ori 2,2,0
lfs 1,-20(1)
vs. below seq seems faster.
rldicr 4,4,0,31
mtvsrd 1,4
xscvspdpn 1,1
