https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127129
Bug ID: 127129
Summary: Incorrect vectorization with -march=x86-64-v2
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: middle-end
Assignee: unassigned at gcc dot gnu.org
Reporter: kristerw at gcc dot gnu.org
Blocks: 118443
Target Milestone: ---
The function glob3 below is miscompiled on x86_64 with -O3 -march=x86-64-v2.
The function as called in the test is supposed to write 1 and 0 to the first
two elements of `a`, but the generated code instead writes 1 to the fist 16
elements, followed by a 0.
__attribute__ ((noipa))
void
glob3 (char *sc, short *dc)
{
while (dc && (*dc++ = *sc))
;
}
int main(void)
{
short a[64];
__builtin_memset(a, 1, sizeof(a));
glob3((char*)a + 1, a);
if (a[0] != 1 || a[1] != 0 || a[2] != 257)
__builtin_abort();
return 0;
}
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118443
[Bug 118443] [Meta bug] Bugs triggered by and blocking more smtgcc testing
