[Bug target/19686] New: avr-gcc 4.0: loop performance decrease (extra conversion)

[Bug target/19686] New: avr-gcc 4.0: loop performance decrease (extra conversion)

[dmixm at marine dot febras dot ru]

Program: 
 
int foo (int x) 
{ 
    int i= x; 
    do { 
        x *= 2; 
    } while (--i); 
    return x; 
} 
 
avr-gcc 4.0 (20050116) with `-Os' option: 
 
foo: 
/* prologue: frame size=0 */ 
/* prologue end (size=0) */ 
        ldi r18,lo8(0) 
        ldi r19,hi8(0) 
        clr r20 
        clr r21 
        sub r20,r24 
        sbc r21,r25 
.L2: 
        add r24,r24 
        adc r25,r25 
        subi r18,lo8(-(-1)) 
        sbci r19,hi8(-(-1)) 
        cp r18,r20 
        cpc r19,r21 
        brne .L2 
/* epilogue: frame size=0 */ 
        ret 
 
avr-gcc 3.3.4: 
 
foo: 
/* prologue: frame size=0 */ 
/* prologue end (size=0) */ 
        mov r18,r24 
        mov r19,r25 
.L2: 
        add r24,r24 
        adc r25,r25 
        subi r18,lo8(-(-1)) 
        sbci r19,hi8(-(-1)) 
        brne .L2 
/* epilogue: frame size=0 */ 
        ret

-- 
           Summary: avr-gcc 4.0: loop performance decrease (extra
                    conversion)
           Product: gcc
           Version: 4.0.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: dmixm at marine dot febras dot ru
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19686

[Bug target/19686] avr-gcc 4.0: loop performance decrease (extra conversion)

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-29 12:45 -------
There is a conversion to unsigned int:
(ivtmp.2 != (unsigned int) (x - 1) * 65535 + 65535)

but isn't that free?

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19686

[Bug tree-optimization/19686] [4.0 Regression] loop performance decrease, not comparing against 0

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-30 19:52 -------
Actually the conversion to unsigned is free.  The problem is something different.

The problem is that we are no longer comparing against 0 instead we are comparing something totally 
different.

(ivtmp.2 != (unsigned int) (x - 1) * 65535 + 65535)



-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |minor
             Status|UNCONFIRMED                 |NEW
          Component|target                      |tree-optimization
     Ever Confirmed|                            |1
           Keywords|                            |missed-optimization
   Last reconfirmed|0000-00-00 00:00:00         |2005-01-30 19:53:00
               date|                            |
            Summary|avr-gcc 4.0: loop           |[4.0 Regression] loop
                   |performance decrease (extra |performance decrease, not
                   |conversion)                 |comparing against 0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19686

[Bug tree-optimization/19686] [4.0 Regression] loop performance decrease, not comparing against 0

[andrewhutchinson at cox dot net]

------- Additional Comments From andrewhutchinson at cox dot net  2005-02-06 23:06 -------
Taking X as the initial value of x on function entry.

The loop is defined as i=X to 0, step -1. Which is a simple do loop.

It gets "optimized" as i=0 to -X, step -1. (Which is something bizarre!)

The code increase is due to 1) Computation of -X and  2) compare said -X



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19686

[Bug tree-optimization/19686] [4.0 Regression] loop performance decrease, not comparing against 0

[schlie at comcast dot net]

------- Additional Comments From schlie at comcast dot net  2005-02-07 02:42 -------
(In reply to comment #2)

Might it be possible to change the severity to at least "normal"
and possibly reclassify it as a "mis-optimization", as it's very
typical for folks who know processors to intentionally write
code such that compairs are agaisnt zero, as they are typically
free; having the compiler de-optimize an otherwise more effecient
explicit code fragment would seem like something that would be fairly
critical to remedy?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19686

[Bug tree-optimization/19686] [4.0 Regression] loop performance decrease, not comparing against 0

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-02-07 02:46 -------
(In reply to comment #4)
> (In reply to comment #2)
It is minor because it is a missed optimization.  (this really is a target dependent bug in that different 
cost matrixes give different answers so it might be just avr's cost does not work at all which is a 
different bug too).

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19686

[Bug tree-optimization/19686] [4.0 Regression] loop performance decrease, not comparing against 0

[schlie at comcast dot net]

------- Additional Comments From schlie at comcast dot net  2005-02-07 03:01 -------
(In reply to comment #5)
> (In reply to comment #4)

Understood, Thanks (apparently it' becomming more important to get the costs more correct).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19686

[Bug tree-optimization/19686] [4.0 Regression] loop performance decrease, not comparing against 0

[andrewhutchinson at cox dot net]

------- Additional Comments From andrewhutchinson at cox dot net  2005-02-08 01:48 -------
I ran testcase with proposed avr_costs patch applied. The result is unchanged. 

The initially generated RTL is unfortunately beyond that which can be fixed by
backend. I dont think this problem is avr specific, it should appear on other
targets.

I have attached initially generated RTL. It is alarmingly complex given starting
point. (This is not so apparent in the assembler as the backend has done a
rather good job of tidying up what it can.)

Perhaps somebody could glance at this to see what exactly went off the rails. It
might just be a manifestation of a known problem.

Wish I could help more - but trees are beyond me at the moment.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19686

[Bug tree-optimization/19686] [4.0/4.1 Regression] loop performance decrease, not comparing against 0

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.1.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19686

[Bug tree-optimization/19686] [4.0/4.1 Regression] loop performance decrease, not comparing against 0

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-20 22:26 -------
Fixed on the mainline:
<L0>:;
  fooc ();
  i = i - 1;
  if (i != 0) goto <L0>; else goto <L1>;

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19686