[Bug target/67929] New: [4.9/5/6 Regression][arm] Wrong code for FP mult-by-power-of-2 + int conversion

["ktkachov at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67929

            Bug ID: 67929
           Summary: [4.9/5/6 Regression][arm] Wrong code for FP
                    mult-by-power-of-2 + int conversion
           Product: gcc
           Version: 6.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ktkachov at gcc dot gnu.org
  Target Milestone: ---
            Target: arm*

Testcase:

int
foo (float a)
{
  return a * 4.9f;
}


int
main (void)
{
  if (foo (10.0f) != 49)
    __builtin_abort ();

  return 0;
}

Compiled with -Ofast -mfpu=vfpv3 -mfloat-abi=hard -mcpu=cortex-a15 -fno-inline
aborts.

The problem is foo (10.0f) returns 40.
This is because the combine_vcvtf2i triggers where its predicate should have
rejected 4.9f

The vfp3_const_double_for_bits function in arm.c is too liberal and accepts any
FP constant that, when truncated, evaluates to a power of 2 FP constant.