https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114075
Bug ID: 114075
Summary: [14 Regression] s390x miscompilation since r14-322
Product: gcc
Version: 14.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: jakub at gcc dot gnu.org
Target Milestone: ---
Since r14-322-g821ef93976e750c118d42a2ad33b96dbd1b9f3a5 the following testcase
is miscompiled at -O2 -fopenmp-simd -march=z13 on s390x-linux:
typedef float V __attribute__((__vector_size__ (16)));
__attribute__((__always_inline__)) inline static float
foo (V a)
{
float r = 0;
#pragma omp simd reduction(+:r)
for (unsigned long i = 0; i < (sizeof (a) / sizeof (float)); i++)
r += a[i];
return r;
}
int
main ()
{
static const struct { V a; float r; } t[] = {
{ (V) { -654.47f, -946.98f, -606.16f, 426.64f }, -1780.97f },
{ (V) { 663.94f, 660.23f, 682.64f, -644.47f }, 1362.34f },
{ (V) { -909.10f, 23.60f, -382.13f, -563.57f }, -1831.20f }
};
for (unsigned long i = 0; i < (sizeof (t) / sizeof (t[0])); i++)
{
float r = foo (t[i].a);
if (r < t[i].r - 1.0f || r > t[i].r + 1.0f)
__builtin_abort ();
}
}
Before that change, all the 3 results were correct, now all of them are +-0.0f.