[Bug target/108293] New: Incorrect assembly emitted for float for BPF target

[Bug target/108293] New: Incorrect assembly emitted for float for BPF target

[dkm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108293

            Bug ID: 108293
           Summary: Incorrect assembly emitted for float for BPF target
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: dkm at gcc dot gnu.org
  Target Milestone: ---

https://godbolt.org/z/z9WEs3Ksn

The generated code has identical function for b() and d(), which is obviously
wrong.

float f;
float a() { f = 1.0; return 1.0; }
float b() { f = 2.0; return 2.0; }
float c() { f = 2.0; return 3.0; }
float d() { f = 3.0; return 3.0; }

_Z1av:
        lddw    %r0,65
        lddw    %r1,f
        stxw    [%r1+0],%r0
        exit
_Z1bv:
        lddw    %r0,129
        lddw    %r1,f
        stxw    [%r1+0],%r0
        exit
_Z1cv:
        lddw    %r0,f
        lddw    %r1,129
        stxw    [%r0+0],%r1
        lddw    %r0,129
        exit
_Z1dv:
        lddw    %r0,129
        lddw    %r1,f
        stxw    [%r1+0],%r0
        exit

[Bug target/108293] Incorrect assembly emitted for float for BPF target

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108293

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The bug is in bpf_print_operand:
    case CONST_DOUBLE:
      if (CONST_DOUBLE_HIGH (op))
        fprintf (file, HOST_WIDE_INT_PRINT_DOUBLE_HEX,
                 CONST_DOUBLE_HIGH (op), CONST_DOUBLE_LOW (op));
      else if (CONST_DOUBLE_LOW (op) < 0)
        fprintf (file, HOST_WIDE_INT_PRINT_HEX, CONST_DOUBLE_LOW (op));
      else
        fprintf (file, HOST_WIDE_INT_PRINT_DEC, CONST_DOUBLE_LOW (op));
      break;
Obviously, the above handling is fine only for integral CONST_DOUBLEs, so for
floating point modes one needs to do something that other targets do, like:
  else if (CONST_DOUBLE_P (x) && GET_MODE (x) == SFmode)
    {
      long l;

      REAL_VALUE_TO_TARGET_SINGLE (*CONST_DOUBLE_REAL_VALUE (x), l);
      fprintf (file, "0x%08x", (unsigned int) l);
    }
for SFmode, or

  else if (CONST_DOUBLE_P (x) && GET_MODE (x) == DFmode)
    {
      long l[2];

      REAL_VALUE_TO_TARGET_DOUBLE (*CONST_DOUBLE_REAL_VALUE (x), l);
      fprintf (file, "0x%lx%08lx", l[1] & 0xffffffff, l[0] & 0xffffffff);
    }
for DFmode.

[Bug target/108293] Incorrect assembly emitted for float for BPF target

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108293

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Another thing is that at least for all SFmode constant one could use mov
instead of lddw.

[Bug target/108293] Incorrect assembly emitted for float for BPF target

[jemarch at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108293

--- Comment #3 from Jose E. Marchesi <jemarch at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #2)
> Another thing is that at least for all SFmode constant one could use mov
> instead of lddw.

For this I guess we could expand the "I" constraint to cover const_double that
fits in the 32-bit range.  But then we would need something like IN_RANGE that
works with const_double.

[Bug target/108293] Incorrect assembly emitted for float for BPF target

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108293

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |marxin at gcc dot gnu.org
   Last reconfirmed|                            |2023-01-06
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1

[Bug target/108293] Incorrect assembly emitted for float for BPF target

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108293

--- Comment #4 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by David Faust <dfaust@gcc.gnu.org>:

https://gcc.gnu.org/g:c7279270a2deda81eaeba37a87d721bee0ed6004

commit r13-5110-gc7279270a2deda81eaeba37a87d721bee0ed6004
Author: David Faust <david.faust@oracle.com>
Date:   Tue Jan 10 10:53:12 2023 -0800

    bpf: correct bpf_print_operand for floats [PR108293]

    The existing logic in bpf_print_operand was only correct for integral
    CONST_DOUBLEs, and emitted garbage for floating point modes. Fix it so
    floating point mode operands are correctly handled.

            PR target/108293

    gcc/

            * config/bpf/bpf.cc (bpf_print_operand): Correct handling for
            floating point modes.

    gcc/testsuite/

            * gcc.target/bpf/double-1.c: New test.
            * gcc.target/bpf/double-2.c: New test.
            * gcc.target/bpf/float-1.c: New test.

[Bug target/108293] Incorrect assembly emitted for float for BPF target

[jemarch at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108293

Jose E. Marchesi <jemarch at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED

--- Comment #5 from Jose E. Marchesi <jemarch at gcc dot gnu.org> ---
Fixed.