[Bug target/107723] New: RISC-V lround/ceil/floor with -fno-fp-int-builtin-inexact

[Bug target/107723] New: RISC-V lround/ceil/floor with -fno-fp-int-builtin-inexact

[kevinl at rivosinc dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107723

            Bug ID: 107723
           Summary: RISC-V lround/ceil/floor with
                    -fno-fp-int-builtin-inexact
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: kevinl at rivosinc dot com
  Target Milestone: ---

This was observed when solving
https://gcc.gnu.org/pipermail/gcc-patches/2022-November/605481.html. For
example, 

int ceil(double i)
{
  return __builtin_lceil(i);
}

-march=rv64gc -mabi=lp64d -fno-fp-int-builtin-inexact generates

ceil1:
        addi    sp,sp,-16
        sd      ra,8(sp)
        call    ceil
        ld      ra,8(sp)
        fcvt.l.d a0,fa0,rtz
        addi    sp,sp,16
        sext.w  a0,a0
        jr      ra

fcvt can possibly set the inexact exception flag, so this seems to be a
violation.

[Bug middle-end/107723] lround/ceil/floor with -fno-fp-int-builtin-inexact

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107723

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|RISC-V lround/ceil/floor    |lround/ceil/floor with
                   |with                        |-fno-fp-int-builtin-inexact
                   |-fno-fp-int-builtin-inexact |
          Component|target                      |middle-end

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
This is not a target issue though.
As x86_64 produces:
ceilh:
        sub     rsp, 8
        call    ceil
        add     rsp, 8
        cvttsd2si       rax, xmm0
        ret

cvttsd2si will raise inexact also.

[Bug middle-end/107723] lround/ceil/floor with -fno-fp-int-builtin-inexact

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107723

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Hmm, only lround is documented and the other two are not, I filed PR 107724 for
the missing documentation.

[Bug middle-end/107723] lround/ceil/floor with -fno-fp-int-builtin-inexact

[kevinl at rivosinc dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107723

--- Comment #3 from Kevin Lee <kevinl at rivosinc dot com> ---
aarch64 also produces

ceil1:
        fcvtps  x0, d0
        ret

Since it has been changed to middle-end, I'll delete riscv as the target

[Bug middle-end/107723] lround/ceil/floor with -fno-fp-int-builtin-inexact

[xry111 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107723

Xi Ruoyao <xry111 at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2023-11-17
                 CC|                            |xry111 at gcc dot gnu.org
             Status|UNCONFIRMED                 |NEW

--- Comment #4 from Xi Ruoyao <xry111 at gcc dot gnu.org> ---
long x(double y)
{
        return __builtin_trunc(y);
}

In 005t.original (!) it's already "optimized" to

{
  return (long int) y;
}

Despite -fno-fp-int-builtin-inexact is set.

[Bug middle-end/107723] lround/ceil/floor with -fno-fp-int-builtin-inexact

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107723

--- Comment #5 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Xi Ruoyao <xry111@gcc.gnu.org>:

https://gcc.gnu.org/g:99182ea09f10beca8445396cbab491899536f5c3

commit r14-6455-g99182ea09f10beca8445396cbab491899536f5c3
Author: Xi Ruoyao <xry111@xry111.site>
Date:   Fri Nov 24 11:08:19 2023 +0800

    Only allow (int)trunc(x) to (int)x simplification with
-ffp-int-builtin-inexact [PR107723]

    With -fno-fp-int-builtin-inexact, trunc is not allowed to raise
    FE_INEXACT and it should produce an integral result (if the input is not
    NaN or Inf).  Thus FE_INEXACT should not be raised.

    But (int)x may raise FE_INEXACT when x is a non-integer, non-NaN, and
    non-Inf value.  C23 recommends to do so in a footnote.

    Thus we should not simplify (int)trunc(x) to (int)x if
    -fno-fp-int-builtin-inexact is in-effect.

    gcc/ChangeLog:

            PR middle-end/107723
            * convert.cc (convert_to_integer_1) [case BUILT_IN_TRUNC]: Break
            early if !flag_fp_int_builtin_inexact and flag_trapping_math.

    gcc/testsuite/ChangeLog:

            PR middle-end/107723
            * gcc.dg/torture/builtin-fp-int-inexact-trunc.c: New test.