https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91096
Bug ID: 91096
Summary: Openmp vector max reduction incorrect
Product: gcc
Version: 10.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: wilco at gcc dot gnu.org
Target Milestone: ---
The new gcc.dg/vect/vect-simd-14.c test produces incorrect reduction due to the
vectorizer adding fmax (result, FLT_MAX):
float b[1024], a[1024];
__attribute__((noipa)) float
bar (void)
{
float s = -__builtin_inff ();
#pragma omp simd reduction (inscan, max:s)
for (int i = 0; i < 1024; i++)
{
b[i] = s;
#pragma omp scan exclusive(s)
s = s > a[i] ? s : a[i];
}
return s;
}
When built with -ffast-math -ftree-vectorize -O2 -fdump-tree-vect-details
-fopenmp-simd it shows the following constants being used in the fmax
operations:
vect_cst__30 = { -3.4028234663852885981170418348451692544e+38,
-3.4028234663852885981170418348451692544e+38,
-3.4028234663852885981170418348451692544e+38,
-3.4028234663852885981170418348451692544e+38 };
This is 0xff7fffff / -0x1.fffffep+127 / FLT_MAX. As a result b[0] has the wrong
value - it should be -inff.
