[Bug middle-end/68027] New: conditional statement and unnecessary register assignment

[Bug middle-end/68027] New: conditional statement and unnecessary register assignment

[SztfG at yandex dot ru]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68027

            Bug ID: 68027
           Summary: conditional statement and unnecessary register
                    assignment
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: SztfG at yandex dot ru
  Target Milestone: ---

C source code:

int a1(int);
int a2(int);
int a3(int);

int test1(int a)
{
  if (a > 100) return a1(a);
  else if (a < 100) return a2(a);
  return a3(a);
}

-----------------------------------

Assembly output:


test1:
.LFB0:
        .cfi_startproc
        cmpl    $100, %edi
        jg      .L5
        jne     .L6
        movl    $100, %edi   # no need to do this, eax is equal $100 at this
point
        jmp     a3
        .p2align 4,,10
        .p2align 3
.L6:
        jmp     a2
        .p2align 4,,10
        .p2align 3
.L5:
        jmp     a1
        .cfi_endproc

-----------------------------------

Tested on http://gcc.godbolt.org/ : gcc versions 4.4.7 - 5.2.0 have this
problem.

[Bug middle-end/68027] conditional statement and unnecessary register assignment

[SztfG at yandex dot ru]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68027

--- Comment #1 from SztfG at yandex dot ru ---
># no need to do this, eax is equal $100 at this point
I mean edi. after this part:
        cmpl    $100, %edi
        jg      .L5
        jne     .L6
if it passed this jg jne instruction without conditional jump, edi shoud be 100

[Bug tree-optimization/68027] conditional statement and unnecessary register assignment

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68027

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2015-10-20
          Component|middle-end                  |tree-optimization
            Version|unknown                     |5.2.0
     Ever confirmed|0                           |1

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.  It's VRP doing this constant propagation and uncprop not undoing
it.
When we do not need to spill 'a' then re-using 'a' would be better but the
optimization is supposed to catch the more frequent case where we'd have to
spill 'a'.

[Bug tree-optimization/68027] conditional statement and unnecessary register assignment

[SztfG at yandex dot ru]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68027

--- Comment #3 from SztfG at yandex dot ru ---
btw why nobody confirm my old reported bug about stack-allocated array missed
optimization https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66152 ?

[Bug tree-optimization/68027] conditional statement and unnecessary register assignment

[SztfG at yandex dot ru]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68027

--- Comment #4 from SztfG at yandex dot ru ---
I think this can be optimized in this way:

        cmpl    $100, %edi
        jg      a1
        jne     a2
        jmp     a3

without any label jumps

[Bug tree-optimization/68027] uncprop should work on more than PHI nodes

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68027

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|conditional statement and   |uncprop should work on more
                   |unnecessary register        |than PHI nodes
                   |assignment                  |
           Severity|normal                      |enhancement

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
tree-ssa-uncprop.c does some of this but only handles PHI nodes currrently.
Note ICC and MSVC generate the code without the move while GCC and clang both
do the move.

[Bug tree-optimization/68027] uncprop should work on more than PHI nodes

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68027

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to SztfG from comment #4)
> I think this can be optimized in this way:
> 
>         cmpl    $100, %edi
>         jg      a1
>         jne     a2
>         jmp     a3

Note that is a different bug and should be filed seperately.  Something about
conditional tail calls.  I thought there was a bug filed for that case but I
can't seem to find it.