[Bug rtl-optimization/100378] New: [Regression 9/10/11/12] arm64: lsl + asr used instead of sxth

[Bug rtl-optimization/100378] New: [Regression 9/10/11/12] arm64: lsl + asr used instead of sxth

[luc.vanoostenryck at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100378

            Bug ID: 100378
           Summary: [Regression 9/10/11/12] arm64: lsl + asr used instead
                    of sxth
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: rtl-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: luc.vanoostenryck at gmail dot com
  Target Milestone: ---

Created attachment 50727
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50727&action=edit
testcase

On arm64, when compiling with optimization, for example with -O2,
the following code:

    struct sh {
        short a;
        short b;
        short y[2];
    };
    int fooh(struct sh s) { return s.a; }

produces the following assembly code since GCC9.x:
    fooh:
        lsl     x0, x0, 16
        asr     w0, w0, 16
        ret

but with GCC8.x and before it produces the shorter:
    fooh(sh):
        sxth    w0, w0
        ret


See https://gcc.godbolt.org/z/YrW7E3cro

[Bug target/100378] [9/10/11/12 Regression] arm64: lsl + asr used instead of sxth

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100378

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[Regression 9/10/11/12]     |[9/10/11/12 Regression]
                   |arm64: lsl + asr used       |arm64: lsl + asr used
                   |instead of sxth             |instead of sxth
   Target Milestone|---                         |12.0
           Keywords|                            |missed-optimization

[Bug target/100378] [9/10/11/12 Regression] arm64: lsl + asr used instead of sxth

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100378

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2021-05-02
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Trying 2 -> 6:
    2: r94:DI=r96:DI
      REG_DEAD r96:DI
    6: r95:SI=sign_extend(r94:DI#0)
      REG_DEAD r94:DI
Failed to match this instruction:
(set (reg:SI 95 [ s ])
    (ashiftrt:SI (subreg:SI (ashift:DI (reg:DI 96)
                (const_int 16 [0x10])) 0)
        (const_int 16 [0x10])))

So what is happening is now combine is inserting an extra mov from x0 to a new
register 96 and when it does the combine, it does it badly not into what it was
before:
(insn 6 3 11 2 (set (reg:SI 95 [ s ])
        (sign_extend:SI (subreg:HI (reg/v:DI 94 [ s ]) 0))) "t8877.c":6:37 111
{*extendhisi2_aarch64}
     (expr_list:REG_DEAD (reg/v:DI 94 [ s ])
        (nil)))

Confirmed.  I thought there was a dup of this bug somewhere but I cannot find
it.

[Bug target/100378] [9/10/11/12 Regression] arm64: lsl + asr used instead of sxth

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100378

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|12.0                        |9.4

[Bug target/100378] [9/10/11/12 Regression] arm64: lsl + asr used instead of sxth

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100378

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|9.4                         |9.5

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 9.4 is being released, retargeting bugs to GCC 9.5.

[Bug target/100378] [9/10/11/12 Regression] arm64: lsl + asr used instead of sxth

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100378

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2

[Bug target/100378] [10/11/12/13 Regression] arm64: lsl + asr used instead of sxth

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100378

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|9.5                         |10.4

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 9 branch is being closed

[Bug target/100378] [10/11/12/13 Regression] arm64: lsl + asr used instead of sxth

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100378

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|10.4                        |10.5

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 10.4 is being released, retargeting bugs to GCC 10.5.

[Bug target/100378] [11/12/13/14 Regression] arm64: lsl + asr used instead of sxth

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100378

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|10.5                        |11.5

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 10 branch is being closed.