[Bug target/46092] New: Improve constant handling of thumb2 instructions

[Bug target/46092] New: Improve constant handling of thumb2 instructions

[carrot at google dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46092

           Summary: Improve constant handling of thumb2 instructions
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: carrot@google.com
                CC: carrot@google.com
              Host: i686-linux
            Target: arm-eabi
             Build: i686-linux


Compile the following code with options -march=armv7-a -mthumb -Os

unsigned long tv(unsigned long x)
{
  if (x >= 65521)
    x = x - 65521;
  return x;
}

gcc generates:

tv:
        movw    r3, #65520
        cmp     r0, r3
        ittt    hi
        subhi   r0, r0, #65024   // A
        subhi   r0, r0, #496     // B
        subhi   r0, r0, #1       // C
        bx      lr

Notice that (x = x - 65521) is translated into 3 instructions. Actually 65521
can be loaded into register with one instruction. So the shorter result is:

tv:
        movw    r3, #65520
        cmp     r0, r3
        ittt    hi
        movwhi   r1, #65521
        subhi   r0, r1
        bx      lr

This should be improved by function arm_gen_constant. The function is already
commented with
/* ??? This needs more work for thumb2.  */

[Bug target/46092] Improve constant handling of thumb2 instructions

[ibolton at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46092

Ian Bolton <ibolton at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2010.10.27 16:03:59
                 CC|                            |ibolton at gcc dot gnu.org
     Ever Confirmed|0                           |1
      Known to fail|                            |4.6.0

--- Comment #1 from Ian Bolton <ibolton at gcc dot gnu.org> 2010-10-27 16:03:59 UTC ---
Confirmed on trunk, revision 165513.

In your example, another way would be to subtract r3 (#65520) and then #1.

Note: neither -O3 or -O2 can avoid the three subhi instruction, currently.

[Bug target/46092] Improve constant handling of thumb2 instructions

[ams at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46092

Andrew Stubbs <ams at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
                 CC|                            |ams at gcc dot gnu.org
         AssignedTo|unassigned at gcc dot       |ams at gcc dot gnu.org
                   |gnu.org                     |

--- Comment #2 from Andrew Stubbs <ams at gcc dot gnu.org> 2010-12-10 17:23:51 UTC ---
I've posted a patch to fix this issue (among others):

  http://gcc.gnu.org/ml/gcc-patches/2010-12/msg00652.html

[Bug target/46092] Improve constant handling of thumb2 instructions

[ams at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46092

--- Comment #3 from Andrew Stubbs <ams at gcc dot gnu.org> 2010-12-10 17:27:04 UTC ---
Oh, I should say, what it actually does is this:

         subhi   r0, r0, #65280   ; 0xff00
         subhi   r0, r0, #241     ; 0x00f1

[Bug target/46092] Improve constant handling of thumb2 instructions

[ramana at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46092

Ramana Radhakrishnan <ramana at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
                 CC|                            |ramana at gcc dot gnu.org
         Resolution|                            |FIXED
   Target Milestone|---                         |4.7.0
           Severity|normal                      |enhancement

--- Comment #4 from Ramana Radhakrishnan <ramana at gcc dot gnu.org> 2012-02-24 12:24:04 UTC ---
Trunk generates this today. Thus fixed. 

Ramana

tv:
    @ args = 0, pretend = 0, frame = 0
    @ frame_needed = 0, uses_anonymous_args = 0
    @ link register save eliminated.
    movw    r3, #65520
    cmp    r0, r3
    itt    hi
    subhi    r0, r0, #65280
    subhi    r0, r0, #241
    bx    lr
    .size    tv, .-tv
    .ident    "GCC: (GNU) 4.7.0 20120116 (experimental)"