[Bug target/39423] New: [

[Bug target/39423] New: [

[chrbr at gcc dot gnu dot org]

-- 
           Summary: [
           Product: gcc
           Version: 4.3.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: chrbr at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39423

[Bug target/39423] [

[paolo dot carlini at oracle dot com]

------- Comment #1 from paolo dot carlini at oracle dot com  2009-03-10 13:34 -------
.


-- 

paolo dot carlini at oracle dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39423

[Bug target/39423] [SH] performance regression: lost mov @(disp,Rn)

[chrbr at gcc dot gnu dot org]

------- Comment #2 from chrbr at gcc dot gnu dot org  2009-03-11 08:46 -------
I observed some large performance regressions in 4.3 and upwards for many
benchmarks for the superh targets, There are many causes but the main one is
reduced to the indirect+offset access :

int
foo (int tab[], int index)
{
  return tab [index+1];
} 

compiles (-O2 -fomit-frame-pointer) into 
        mov     r5,r0
        add     #1,r0
        shll2   r0
        rts     
        mov.l   @(r0,r4),r0

instead of
        shll2   r5
        add     r4,r5
        rts     
        mov.l   @(4,r5),r0

Note that in more complex code the problem is emphasized because only r0
register class can be used as indirect register index, putting extra pressure
on reload.

It seems to be that the problem is in the way that the constant index is now
hidden by gimple, so we now have

 return *(tab + ((unsigned int) index + 1) * 4)

instead of 

 return *(tab + 4B + (int *) ((unsigned int) index * 4))

It seems more easy to change gimple, but this is a target dependant
transformation. On the other hand the RTL code gen should be able to
redistribute the factorization, but that seems extra work to undo what was done
previously.


-- 

chrbr at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |UNCONFIRMED
         Resolution|INVALID                     |
            Summary|[                           |[SH]  performance
                   |                            |regression: lost mov
                   |                            |@(disp,Rn)


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39423

[Bug target/39423] [SH] performance regression: lost mov @(disp,Rn)

[chrbr at gcc dot gnu dot org]

-- 

chrbr at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |chrbr at gcc dot gnu dot org
                   |dot org                     |
             Status|UNCONFIRMED                 |ASSIGNED
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2009-03-11 08:52:58
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39423

[Bug target/39423] [SH] performance regression: lost mov @(disp,Rn)

[ebotcazou at gcc dot gnu dot org]

------- Comment #3 from ebotcazou at gcc dot gnu dot org  2009-03-11 09:06 -------
See http://gcc.gnu.org/ml/gcc-patches/2008-12/msg01134.html


-- 

ebotcazou at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39423

[Bug target/39423] [SH] performance regression: lost mov @(disp,Rn)

[chrbr at gcc dot gnu dot org]

------- Comment #4 from chrbr at gcc dot gnu dot org  2009-03-11 09:30 -------
(In reply to comment #3)
> See http://gcc.gnu.org/ml/gcc-patches/2008-12/msg01134.html
> 

Thanks, I tried your patch against a 4.3.3 base but it didn't fix the problem,
your patch canonicalizes while what I need is a distribution
(base + 1) * 4 => base * 4 + 4


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39423

[Bug target/39423] [SH] performance regression: lost mov @(disp,Rn)

[ebotcazou at gcc dot gnu dot org]

------- Comment #5 from ebotcazou at gcc dot gnu dot org  2009-03-11 09:36 -------
> Thanks, I tried your patch against a 4.3.3 base but it didn't fix the problem,
> your patch canonicalizes while what I need is a distribution

Read again the RTL expander hunk.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39423

[Bug target/39423] [SH] performance regression: lost mov @(disp,Rn)

[chrbr at gcc dot gnu dot org]

------- Comment #6 from chrbr at gcc dot gnu dot org  2009-03-11 09:46 -------
(In reply to comment #5)
> > Thanks, I tried your patch against a 4.3.3 base but it didn't fix the problem,
> > your patch canonicalizes while what I need is a distribution
> 
> Read again the RTL expander hunk.
> 

ah yes sorry, when I read
"
  base + 2*ind - 2

now it's

  base + (1-ind) * -2
"
I thought that the now what as with your patch. So it looks indeed like quite
similar to what I see. I'll try to see why it doesn't solve my case.

thanks,


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39423

[Bug target/39423] [SH] performance regression: lost mov @(disp,Rn)

[ebotcazou at gcc dot gnu dot org]

------- Comment #7 from ebotcazou at gcc dot gnu dot org  2009-03-11 10:25 -------
> I thought that the now what as with your patch. So it looks indeed like quite
> similar to what I see. I'll try to see why it doesn't solve my case.

Reverting to the old canonicalization at the tree level is not doable, that's
why the patch does it during RTL expansion and only for offsets.  It probably
would need to be extended if that's not sufficient for your testcase.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39423

[Bug target/39423] [SH] performance regression: lost mov @(disp,Rn)

[chrbr at gcc dot gnu dot org]

------- Comment #8 from chrbr at gcc dot gnu dot org  2009-03-11 14:07 -------
I have picky disabled the canonicalization in fold_plusminus_mult_expr for 
identical constants that are power of 2, so my mov @(disp, rn) is back :-(. For
some reason your patch let the base+index computation factorization thru 

This is experimental for now, because expand_expr needs to be extended to
repair expressions like return ((a * 4) + 4) that are not an indirect_ref.
(thanks we differ PLUS expr from POINTER_PLUS_EXPR)

+++ fold-const.c        2009-03-11 13:49:40.000000000 +0100
@@ -7431,7 +7431,10 @@
   same = NULL_TREE;

   if (operand_equal_p (arg01, arg11, 0))
-    same = arg01, alt0 = arg00, alt1 = arg10;
+    {
+      if (code != PLUS_EXPR || exact_log2 (TREE_INT_CST_LOW (arg01)) == -1)
+       same = arg01, alt0 = arg00, alt1 = arg10;
+    }
   else if (operand_equal_p (arg00, arg10, 0))
     same = arg00, alt0 = arg01, alt1 = arg11;
   else if (operand_equal_p (arg00, arg11, 0))


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39423

[Bug target/39423] [SH] performance regression: lost mov @(disp,Rn)

[chrbr at gcc dot gnu dot org]

------- Comment #9 from chrbr at gcc dot gnu dot org  2009-03-12 15:04 -------
The attached patch improves the SH generation, but I noticed a small regression
with the ARM that could make use before of a shifted constant addressing mode,
so not using the extra register for the value.

A target description check should be done while expanding the
canonicalization, that should not be done only when a base+cst addressing mode
exists and the cst must be shifter in a register. Any input welcome to how
to target conditionalize this transformation.

No performance impact was noticed on i686 however.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39423

[Bug target/39423] [SH] performance regression: lost mov @(disp,Rn)

[chrbr at gcc dot gnu dot org]

------- Comment #10 from chrbr at gcc dot gnu dot org  2009-03-12 15:10 -------
Created an attachment (id=17447)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=17447&action=view)
SH illustrative patch

for feedback only. win on SH. lost on ARM

2009-03-12  Christian Bruel  <christian.bruel@st.com>

        * fold-const.c (fold_plusminus_mult_expr): Move canonicalization of
         index+cst...
        * expr.c (expand_expr_real_1): ... here.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39423

[Bug target/39423] [4.3/4.4/4.5 Regression] [SH] performance regression: lost mov @(disp,Rn)

[jsm28 at gcc dot gnu dot org]

-- 

jsm28 at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[SH]  performance           |[4.3/4.4/4.5 Regression]
                   |regression: lost mov        |[SH]  performance
                   |@(disp,Rn)                  |regression: lost mov
                   |                            |@(disp,Rn)
   Target Milestone|---                         |4.3.4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39423

[Bug target/39423] [4.3/4.4/4.5 Regression] [SH] performance regression: lost mov @(disp,Rn)

[jakub at gcc dot gnu dot org]

-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39423

[Bug target/39423] [4.3/4.4/4.5 Regression] [SH] performance regression: lost mov @(disp,Rn)

[rguenth at gcc dot gnu dot org]

------- Comment #11 from rguenth at gcc dot gnu dot org  2009-08-04 12:29 -------
GCC 4.3.4 is being released, adjusting target milestone.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.3.4                       |4.3.5


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39423

[Bug target/39423] [4.3/4.4/4.5/4.6 Regression] [SH] performance regression: lost mov @(disp,Rn)

[rguenth at gcc dot gnu dot org]

------- Comment #12 from rguenth at gcc dot gnu dot org  2010-05-22 18:13 -------
GCC 4.3.5 is being released, adjusting target milestone.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.3.5                       |4.3.6


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39423