[Bug tree-optimization/107608] New: [13 Regression] Failure on fold-overflow-1.c

[jakub at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107608

            Bug ID: 107608
           Summary: [13 Regression] Failure on fold-overflow-1.c
           Product: gcc
           Version: 13.0
            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: ---

With my https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107569#c20 patch
I get
+FAIL: gcc.dg/fold-overflow-1.c scan-assembler-times 2139095040 2
regression because
float foo(void)
{
  return  __FLT_MAX__ + __FLT_MAX__;
}
no longer raises overflow exception.  This case is canonicalized to
  return __FLT_MAX__ * 2.0f;
and ranger correctly determines the result is [+Inf, +Inf], but then
comes dom2 and happily replaces
  _1 = 3.4028234663852885981170418348451692544e+38 * 2.0e+0;
  return _1;
with
  _1 = 3.4028234663852885981170418348451692544e+38 * 2.0e+0;
  return  Inf;
(I think this is still correct) and then dce3 removes the
  _1 = 3.4028234663852885981170418348451692544e+38 * 2.0e+0;
statement.  That is not correct, because -ftrapping-math is on and this
multiplication is supposed to trap.
Now, my patch isn't needed, if I do:
float bar(void)
{
  return  __FLT_MAX__ + (__FLT_MAX__ / 4.0f);
}
instead, then the same regressed already on vanilla trunk:
  _1 = 3.4028234663852885981170418348451692544e+38 +
8.507058665963221495292604587112923136e+37;
  return _1;
changes in dom2 to
  _1 = 3.4028234663852885981170418348451692544e+38 +
8.507058665963221495292604587112923136e+37;
  return  Inf;
and in dce3 to just
  return  Inf;