[Bug target/51697] New: SH Target: Inefficient DImode comparisons for -Os

[Bug target/51697] New: SH Target: Inefficient DImode comparisons for -Os

[oleg.endo@t-online.de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51697

             Bug #: 51697
           Summary: SH Target: Inefficient DImode comparisons for -Os
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: enhancement
          Priority: P3
         Component: target
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: oleg.endo@t-online.de
                CC: kkojima@gcc.gnu.org
            Target: sh*-*-*


For -Os and everything but -m1 DImode comparisons are not optimized properly
which results in redundant SImode comparisons, producing code worse than for
-O1.  
A reduced example:

int test_0 (long long* x)
{
  return *x & 0xFFFFFFFF ? -20 : -40;
}

-Os -m2/-m3/-m4:
        mov    #0,r2           ! 55    movsi_ie/3    [length = 2]
        tst    r2,r2           ! 57    cmpeqsi_t/1    [length = 2]
        bf/s    .L12            ! 58    branch_false    [length = 2]
        mov.l    @(4,r4),r3    ! 12    movsi_ie/7    [length = 2]
        tst    r3,r3           ! 59    cmpeqsi_t/1    [length = 2]
.L12:
        bt/s    .L11            ! 14    branch_true    [length = 2]
        mov    #-40,r0         ! 5    movsi_ie/3    [length = 2]
        mov    #-20,r0         ! 4    movsi_ie/3    [length = 2]
.L11:
        rts    
        nop                     ! 65    *return_i    [length = 4]


-Os -m1:
-O2 -m4:
        mov.l   @(4,r4),r1       ! 10    movsi_i/5    [length = 2]
        mov     #-40,r0         ! 5    movsi_i/3    [length = 2]
        tst     r1,r1           ! 15    cmpeqsi_t/1    [length = 2]
        bt      .L7             ! 16    branch_true    [length = 2]
        mov     #-20,r0         ! 4    movsi_i/3    [length = 2]
.L7:
        rts    
        nop                     ! 61    *return_i    [length = 4]


-O1 -m4:
        mov.l   @(4,r4),r1      ! 10    movsi_ie/7    [length = 2]
        tst     r1,r1           ! 17    cmpeqsi_t/1    [length = 2]
        bt/s    .L6             ! 18    branch_true    [length = 2]
        mov     #-40,r0         ! 5    movsi_ie/3    [length = 2]
        mov     #-20,r0         ! 4    movsi_ie/3    [length = 2]
.L6:
        rts    
        nop                     ! 62    *return_i    [length = 4]


Another example would be:

int test_2 (unsigned long long x)
{
  return x >= 0x100000000LL ? -20 : -40;
}

-Os -m2/-m3/-m4:
        mov     #0,r2           ! 48    movsi_ie/3    [length = 2]
        mov     #-1,r3          ! 49    movsi_ie/3    [length = 2]
        cmp/eq  r2,r4           ! 9    cmpgtudi_t    [length = 8]
        bf/s    .Ldi67
        cmp/hi  r2,r4
        cmp/hi  r3,r5
.Ldi67:
        bf/s    .L16            ! 10    branch_false    [length = 2]
        mov     #-40,r0         ! 5    movsi_ie/3    [length = 2]
        mov     #-20,r0         ! 4    movsi_ie/3    [length = 2]
.L16:
        rts    
        nop                     ! 52    *return_i    [length = 4]


-Os -m1:
        tst     r4,r4           ! 9    cmpeqsi_t/1    [length = 2]
        mov     #-20,r0         ! 4    movsi_i/3    [length = 2]
        bf      .L12            ! 10    branch_false    [length = 2]
        mov     #-40,r0         ! 5    movsi_i/3    [length = 2]
.L12:
        rts    
        nop                     ! 56    *return_i    [length = 4]



The problem does not appear for -m1, only for -Os and -m2*, -m3*, -m4*.

[Bug target/51697] SH Target: Inefficient DImode comparisons for -Os

[olegendo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51697

Oleg Endo <olegendo at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |olegendo at gcc dot gnu.org
         Depends on|                            |52898

--- Comment #1 from Oleg Endo <olegendo at gcc dot gnu.org> 2012-04-12 06:56:41 UTC ---
This is very closely related to PR 52898.

[Bug target/51697] SH Target: Inefficient DImode comparisons for -Os

[olegendo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51697

--- Comment #2 from Oleg Endo <olegendo at gcc dot gnu.org> ---
Author: olegendo
Date: Sun Dec  8 22:15:59 2013
New Revision: 205794

URL: http://gcc.gnu.org/viewcvs?rev=205794&root=gcc&view=rev
Log:
    PR target/52898
    PR target/51697
    * common/config/sh/sh-common.c (sh_option_optimization_table): Remove
    OPT_mcbranchdi entry.
    * config/sh/sh.opt (mcbranchdi, mcmpeqdi): Mark as undocumented and
    emit a warning.
    * config/sh/sh.c (sh_option_override): Initialize TARGET_CBRANCHDI4
    and TARGET_CMPEQDI_T variables.
    * doc/invoke.texi (SH options): Undocument -mcbranchdi and -mcmpeqdi.

    PR target/52898
    PR target/51697
    * gcc.target/sh/pr51697.c: New.


Added:
    trunk/gcc/testsuite/gcc.target/sh/pr51697.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/common/config/sh/sh-common.c
    trunk/gcc/config/sh/sh.c
    trunk/gcc/config/sh/sh.opt
    trunk/gcc/doc/invoke.texi
    trunk/gcc/testsuite/ChangeLog

[Bug target/51697] SH Target: Inefficient DImode comparisons for -Os

[olegendo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51697

Oleg Endo <olegendo at gcc dot gnu.org> changed:

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

--- Comment #3 from Oleg Endo <olegendo at gcc dot gnu.org> ---
I'd like to close this PR.  Further improvements to DImode comparisons should
go in PR 52898.