https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65421
Bug ID: 65421
Summary: inefficient code returning float aggregates on
powepc64le
Product: gcc
Version: 5.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: msebor at gcc dot gnu.org
When returning homogeneous floating-point aggregates, gcc first loads the
aggregate into GPRs, then saves it to the save area, and then copies the
aggregate from the GPRs to the floating registers used to return it to the
caller. Clang emits much more efficient code (see below).
$ cat x.c && ./gcc/xgcc -B./gcc -O3 -o/dev/tty -S x.c
typedef struct { double a[4]; } A;
A foo (const A *a) { return *a; }
.file "x.c"
.machine power8
.abiversion 2
.section ".toc","aw"
.section ".text"
.align 2
.p2align 4,,15
.globl foo
.type foo, @function
foo:
ld 7,0(3)
ld 8,8(3)
ld 10,16(3)
ld 9,24(3)
std 7,-64(1)
std 8,-56(1)
std 10,-48(1)
std 9,-40(1)
lfd 2,-56(1)
lfd 1,-64(1)
ori 2,2,0
lfd 3,-48(1)
lfd 4,-40(1)
blr
.long 0
.byte 0,0,0,0,0,0,0,0
.size foo,.-foo
.ident "GCC: (GNU) 5.0.0 20150310 (experimental)"
.section .note.GNU-stack,"",@progbits
Clang emits:
.text
.abiversion 2
.file "-"
.globl foo
.align 2
.type foo,@function
foo:
lfd 1, 0(3)
lfd 2, 8(3)
lfd 3, 16(3)
lfd 4, 24(3)
blr
.long 0
.quad 0
.Ltmp0:
.size foo, .Ltmp0-foo
.ident "clang version 3.6.0 (trunk 215935)"
