https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61300
Bug ID: 61300
Summary: powerpc64le miscompile with K&R-style function
definition at -O0
Product: gcc
Version: 4.9.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: other
Assignee: unassigned at gcc dot gnu.org
Reporter: brooks at gcc dot gnu.org
We've run into a GCC miscompile problem with K&R-style function definitions
that's causing some various old GNU software to fail in some cases.
A simple test case:
$ cat bb.c
struct builtin { char *name; };
int compare_kr (sbp1, sbp2)
struct builtin *sbp1, *sbp2;
{
return sbp1->name[0] - sbp2->name[0];
}
int compare_ansi (struct builtin *sbp1, struct builtin *sbp2)
{
return sbp1->name[0] - sbp2->name[0];
}
On either our Google branch of GCC 4.9 or the Ubuntu system GCC 4.8 on our test
machine, we get something like this (annotated to show the difference):
$ powerpc64le-unknown-linux-gnu-gcc -S -o- bb.c
.file "bb.c"
.abiversion 2
.section ".toc","aw"
.section ".text"
.align 2
.globl compare_kr
.type compare_kr, @function
compare_kr:
std 31,-8(1)
stdu 1,-48(1)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ it allocates 48 bytes for the frame
mr 31,1
std 3,80(31)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ and then writes way past this
std 4,88(31)
ld 9,80(31)
...
.globl compare_ansi
.type compare_ansi, @function
compare_ansi:
std 31,-8(1)
stdu 1,-64(1)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ allocates 64 bytes
mr 31,1
std 3,40(31)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ and saves within the confines of the frame
std 4,32(31)
ld 9,40(31)
...
In this testcase, this seems to go away if we turn on -O2. I'm not sure if
that's always the case or not.
I've also confirmed that this does _not_ happen with the x86_64 compiler built
from the same source branch.
