[Bug middle-end/41004] missed merge of basic blocks

[Bug middle-end/41004] missed merge of basic blocks

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41004

--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-02-04 03:00:33 UTC ---
First off, I think cselim should also be run at -Os.
And then we have the issue of tree-ssa-tail-merge not running at -Os.

[Bug middle-end/41004] missed merge of basic blocks

[steven at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41004

--- Comment #8 from Steven Bosscher <steven at gcc dot gnu.org> 2012-03-16 23:59:41 UTC ---
Ehm, why does tree tail-merge not run at -Os? It's a size optimization, after
all!

[Bug middle-end/41004] missed merge of basic blocks

[steven at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41004

--- Comment #9 from Steven Bosscher <steven at gcc dot gnu.org> 2012-11-08 23:07:36 UTC ---
May be fixed by the patch that was applied for PR54364.

[Bug middle-end/41004] missed merge of basic blocks

[steven at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41004

Steven Bosscher <steven at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.8.0
      Known to fail|                            |4.8.0

--- Comment #10 from Steven Bosscher <steven at gcc dot gnu.org> 2012-11-09 21:38:39 UTC ---
Resulting code from trunk r193358

primal_update_flow:
        push    {r4, r5, lr}
        mov     r3, #1
        mov     r4, #0
.L2:
        cmp     r0, r2
        beq     .L11
        ldr     r5, [r0]
        cmp     r5, #0
        beq     .L3
        str     r4, [r0, #8]
        b       .L4
.L3:
        str     r3, [r0, #8]
.L4:
        ldr     r0, [r0, #4]
        b       .L2
.L11:
        mov     r0, #1
.L6:
        cmp     r1, r2
        beq     .L12
        ldr     r3, [r1]
        cmp     r3, #0
        beq     .L7
        str     r0, [r1, #8]
        b       .L8
.L7:
        str     r3, [r1, #8]
.L8:
        ldr     r1, [r1, #4]
        b       .L6
.L12:   
        @ sp needed
        pop     {r4, r5, pc}

So the issue of comment #0 is fixed.

With "-march=armv7 -mthumb" the stores to "[r0,#8]" and "[r0,#8]" are
properly if-converted, I don't see anything that could be improved
further.  So, fixed.

[Bug middle-end/41004] missed merge of basic blocks

[steven at gcc dot gnu dot org]

------- Comment #6 from steven at gcc dot gnu dot org  2010-01-11 09:35 -------
Re. comment #5 -- rth, any suggestions what an algorithm would look like to
"minimize the number of branches, or the total size of all of the branch
instructions"?  And what do you mean with "some clever changes to the
shorten_branches code"?  Did you mean the calculation of insn sizes, or
avoiding padding, or something else completely?  Hope you can help a bit...


-- 

steven at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rth at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41004

[Bug middle-end/41004] missed merge of basic blocks

[rth at gcc dot gnu dot org]

------- Comment #5 from rth at gcc dot gnu dot org  2009-08-19 23:06 -------
Basic block reordering *should* be altered to work with -Os.
It's definitely required to clean up silly jump sequences that
exception handling creates.

There's no reason, in principal, that the ordering algorithm
can't be adjusted to minimize the number of branches, or the
total size of all of the branch instructions (though that could
require some clever changes to the shorten_branches code).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41004

[Bug middle-end/41004] missed merge of basic blocks

[carrot at google dot com]

------- Comment #4 from carrot at google dot com  2009-08-19 21:55 -------
(In reply to comment #2)
> Why does the basic block reordering pass also not handle this?
> 

Basic block reordering is disabled with options -Os. 

The basic block reordering algorithm is for performance only, it usually
increases code size. So it won't be called when do optimization for size. But
for this specific case, the extra branch can be removed when I compile it with
-O2.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41004

[Bug middle-end/41004] missed merge of basic blocks

[rguenth at gcc dot gnu dot org]

------- Comment #3 from rguenth at gcc dot gnu dot org  2009-08-08 10:36 -------
I would suggest to closely look at SSA expansion as well.  There is I think
still the regression from pre-SSA expand that it doesn't expand PHI nodes
optimally at -Os (pre-SSA expand split common PHI args into extra blocks to
minimize copy instructions which is something that can be done as a tree
level transform).  Like

BB5:
 # a_1 = PHI <a_2(2), a_2(3), b_3(4)>

to

BB6:
 # tmp_4 = PHI <a_2(2), a_2(3)>

BB5:
 # a_1 = PHI <tmp_4(6), b_3(4)>

which we usually don't want and in fact undo during mergephi.  But it needs
one less copy if a_2 and tmp_4 can be coalesced together.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41004

[Bug middle-end/41004] missed merge of basic blocks

[steven at gcc dot gnu dot org]

-- 

steven at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |steven at gcc dot gnu dot
                   |dot org                     |org
             Status|NEW                         |ASSIGNED
   Last reconfirmed|2009-08-08 09:07:04         |2009-08-08 10:06:54
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41004