[Bug target/16494] New: [3.3.2/3.4/3.5] Pessimization with FP conditional branches
ebotcazou at gcc dot gnu dot org
gcc-bugzilla@gcc.gnu.org
Mon Jul 12 12:59:00 GMT 2004
On non-V9 targets, when delay slots are not filled, FP conditional branches are
unnecessarily reversed:
[eric@localhost obj.sparc]$ cat t.c
int foo(float f1, float f2)
{
if (f1 > f2)
return 0;
else
return 1;
}
[eric@localhost obj.sparc]$ gcc/cc1 -Bgcc t.c -mcpu=v7 -O2 -fno-delayed-branch
[eric@localhost obj.sparc]$ cat t.s
.file "t.c"
.section ".text"
.align 4
.global foo
.type foo, #function
.proc 04
foo:
!#PROLOGUE# 0
add %sp, -120, %sp
!#PROLOGUE# 1
st %o0, [%sp+100]
ld [%sp+100], %f9
st %o1, [%sp+100]
ld [%sp+100], %f8
fcmpes %f9, %f8
mov 0, %o0
nop
fbule .+16
nop
b .LL1
nop
mov 1, %o0
.LL1:
retl
sub %sp, -120, %sp
.size foo, .-foo
.ident "GCC: (GNU) 3.3.5 20040703 (prerelease)"
The expected assembly output is:
.file "t.c"
.section ".text"
.align 4
.global foo
.type foo, #function
.proc 04
foo:
!#PROLOGUE# 0
add %sp, -120, %sp
!#PROLOGUE# 1
st %o0, [%sp+100]
ld [%sp+100], %f9
st %o1, [%sp+100]
ld [%sp+100], %f8
fcmpes %f9, %f8
mov 0, %o0
nop
fbg .LL1
nop
mov 1, %o0
.LL1:
retl
sub %sp, -120, %sp
.size foo, .-foo
.ident "GCC: (GNU) 3.3.5 20040703 (prerelease)"
This was inadvertently introduced by:
2003-10-01 Eric Botcazou <ebotcazou@libertysurf.fr>
PR optimization/11753
* config/sparc/sparc.md (length attribute) [fcc branch]: Add 1 to
the length in the non-V9 case.
because there is an undocumented side-effect associated with having a length
greater or equal to 3 for FP conditional branches.
--
Summary: [3.3.2/3.4/3.5] Pessimization with FP conditional
branches
Product: gcc
Version: 3.3.2
Status: UNCONFIRMED
Severity: normal
Priority: P2
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: ebotcazou at gcc dot gnu dot org
CC: gcc-bugs at gcc dot gnu dot org
GCC build triplet: sparc-*-*
GCC host triplet: sparc-*-*
GCC target triplet: sparc-*-*
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16494
More information about the Gcc-bugs
mailing list