[Bug target/43311] missed 'movw' optimization.

[Bug target/43311] missed 'movw' optimization.

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43311

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-01-24 22:52:54 UTC ---
This is dup of the merge loads from two adjacent memory location into one load
bug.  Which I don't remember the number of the bug right now.

[Bug target/43311] missed 'movw' optimization.

[zsojka at seznam dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43311

Zdenek Sojka <zsojka at seznam dot cz> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |zsojka at seznam dot cz

--- Comment #3 from Zdenek Sojka <zsojka at seznam dot cz> 2012-01-25 10:35:47 UTC ---
Does the following code have undefined behavior?

typedef struct { unsigned char b1, b2; } __attribute__((aligned(8))) S;
void f( S const* s, unsigned char* b1, unsigned char* b2 )
{
        *b1 = s->b1;
        *b2 = s->b2;
}

int main()
{
  S s = { 1, 2 };
  f(&s, &s.b2, &s.b2);
  return s.b2 != 1;
}

[Bug target/43311] missed 'movw' optimization.

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43311

--- Comment #4 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-01-25 11:13:40 UTC ---
(In reply to comment #3)
> Does the following code have undefined behavior?
> 
> typedef struct { unsigned char b1, b2; } __attribute__((aligned(8))) S;
> void f( S const* s, unsigned char* b1, unsigned char* b2 )
> {
>         *b1 = s->b1;
>         *b2 = s->b2;
> }
> 
> int main()
> {
>   S s = { 1, 2 };
>   f(&s, &s.b2, &s.b2);
>   return s.b2 != 1;
> }

No, s.b2 should be 1.

[Bug target/43311] missed 'movw' optimization.

[zsojka at seznam dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43311

--- Comment #5 from Zdenek Sojka <zsojka at seznam dot cz> 2012-01-25 12:56:30 UTC ---
(In reply to comment #4)
> 
> No, s.b2 should be 1.

Thank you for the answer. In that case, the optimisation in comment #0 can't be
done in a general case (unless I have overlooked something).

[Bug target/43311] missed 'movw' optimization.

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43311

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

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

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-01-25 13:16:10 UTC ---
For storing two small adjacent constants there is PR22141, that is always safe,
but the question is when it is desirable (that is the primary reason why the
patch isn't in the tree yet).

[Bug target/43311] missed 'movw' optimization.

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=43311

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
               Host|x86_64-gnu-linux            |
              Build|x86_64-gnu-linux            |

--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Zdenek Sojka from comment #5)
> (In reply to comment #4)
> > 
> > No, s.b2 should be 1.
> 
> Thank you for the answer. In that case, the optimisation in comment #0 can't
> be done in a general case (unless I have overlooked something).

Right this makes the code wrong if we do this, an aliasing issue. 

But this is well defined to do:
typedef struct { unsigned char b1, b2; } __attribute__((aligned(8))) S;
void f( S const* __restrict__ s, unsigned char* __restrict__ b1, unsigned char*
__restrict__ b2 )
{
        *b1 = s->b1;
        *b2 = s->b2;
}

[Bug target/43311] New: missed 'movw' optimization.

[pluto at agmk dot net]

reduced testcase:

typedef struct { unsigned char b1, b2; } __attribute__((aligned(8))) S;
void f( S const* s, unsigned char* b1, unsigned char* b2 )
{
        *b1 = s->b1;
        *b2 = s->b2;
}

generates at -Os and -O3:

f:
        movb    (%rdi), %al     # <variable>.b1, <variable>.b1
        movb    %al, (%rsi)     # <variable>.b1,* b1
        movb    1(%rdi), %al    # <variable>.b2, <variable>.b2
        movb    %al, (%rdx)     # <variable>.b2,* b2
        ret

gcc could (at least at -Os) reduce code size and memory accesses to:

        movw    (%rdi), %ax
        movb    %al, (%rsi)     # <variable>.b1,* b1
        movb    %ah, (%rdx)     # <variable>.b2,* b2
        ret


-- 
           Summary: missed 'movw' optimization.
           Product: gcc
           Version: 4.4.3
            Status: UNCONFIRMED
          Severity: enhancement
          Priority: P3
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: pluto at agmk dot net
 GCC build triplet: x86_64-gnu-linux
  GCC host triplet: x86_64-gnu-linux
GCC target triplet: x86_64-gnu-linux


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43311

[Bug target/43311] missed 'movw' optimization.

[rguenth at gcc dot gnu dot org]

------- Comment #1 from rguenth at gcc dot gnu dot org  2010-03-10 10:55 -------
Confirmed.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2010-03-10 10:55:26
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43311