https://bugs.kde.org/show_bug.cgi?id=492210
Bug ID: 492210
Summary: False positive on x86 with ZF taken directly from
addition
Classification: Developer tools
Product: valgrind
Version: 3.24 GIT
Platform: Other
OS: Linux
Status: REPORTED
Severity: normal
Priority: NOR
Component: memcheck
Assignee: jsew...@acm.org
Reporter: amona...@gmail.com
Target Milestone: ---
This is minimized from Memcheck false positive on GCC bootstrap,
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116458
Take the following testcase:
static int f(short *s)
{
return s[0] + (s[1] << 16);
}
int main()
{
void *p = __builtin_malloc(64);
*(char *)p = 1;
asm("" ::: "memory");
int r = f(p);
asm("" : "+d"(r));
if (!r)
__builtin_abort();
}
Compile with 'gcc -O2 -g1' and run under valgrind. Then remove or comment out
asm("" : "+d"(r));
and compile/run again. valgrind will complain:
==16441== Conditional jump or move depends on uninitialised value(s)
==16441== at 0x10908D: main (in /tmp/vg/a.out)
even though the difference in generated code is
--- 1.s 2024-08-24 09:46:24.719252103 +0300
+++ 2.s 2024-08-24 09:46:35.202670418 +0300
@@ -19,7 +19,6 @@
movswl (%rax), %eax
sall $16, %edx
addl %eax, %edx
- testl %edx, %edx
je .L3
xorl %eax, %eax
addq $8, %rsp
i.e. ZF is now taken directly from addition rather than recomputed with the
TEST instruction. In other words, propagation of known bits in ADD is powerful
enough and proves that low 8 bits of %edx are known, but something goes awry in
deducing ZF.
--
You are receiving this mail because:
You are watching all bug changes.
