[Bug tree-optimization/18463] [4.0 Regression] suboptimal use of fancy x86 addressing modes

[Bug tree-optimization/18463] [4.0 Regression] suboptimal use of fancy x86 addressing modes

[mmitchel at gcc dot gnu dot org]

-- 

mmitchel at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.0.3                       |4.0.4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18463

[Bug tree-optimization/18463] [4.0 Regression] suboptimal use of fancy x86 addressing modes

[gdr at gcc dot gnu dot org]

------- Comment #30 from gdr at gcc dot gnu dot org  2007-01-18 03:09 -------
Fixed in GCC-4.1.0.
Not to be fixed in GCC-4.0.x


-- 

gdr at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|4.0.4                       |4.1.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18463

[Bug tree-optimization/18463] [4.0 Regression] suboptimal use of fancy x86 addressing modes

[mmitchel at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.0.2                       |4.0.3


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18463

[Bug tree-optimization/18463] [4.0 Regression] suboptimal use of fancy x86 addressing modes

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-09-20 07:09 -------
Subject: Bug 18463

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	rakdver@gcc.gnu.org	2005-09-20 07:09:22

Modified files:
	gcc            : ChangeLog tree-chrec.c tree-chrec.h 
	                 tree-scalar-evolution.c 

Log message:
	PR tree-optimization/18463
	* tree-chrec.c (chrec_convert): Return fold_converted chrec if
	converting it directly is not possible.
	(chrec_convert_aggressive): New function.
	* tree-chrec.h (chrec_convert_aggressive): Declare.
	* tree-scalar-evolution.c (instantiate_parameters_1, resolve_mixers):
	Fold chrec conversions aggressively if asked to.
	(instantiate_parameters): Modified because of changes in
	instantiate_parameters_1.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.9985&r2=2.9986
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-chrec.c.diff?cvsroot=gcc&r1=2.27&r2=2.28
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-chrec.h.diff?cvsroot=gcc&r1=2.10&r2=2.11
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-scalar-evolution.c.diff?cvsroot=gcc&r1=2.37&r2=2.38



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18463

[Bug tree-optimization/18463] [4.0 Regression] suboptimal use of fancy x86 addressing modes

[rakdver at gcc dot gnu dot org]

------- Additional Comments From rakdver at gcc dot gnu dot org  2005-09-19 15:19 -------
Another patch that improves the code on i686:  
  
http://gcc.gnu.org/ml/gcc-patches/2005-09/msg01159.html 
 
(with -O2 -fomit-frame-pointer:) 
 
.L4: 
        movl    (%ebp,%edx,4), %eax 
        movl    %eax, (%esi,%edx,4) 
        movl    (%edi,%edx,4), %eax 
        movl    %eax, (%ebx,%edx,4) 
        incl    %edx 
        cmpl    %edx, %ecx 
        jne     .L4 
 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18463

[Bug tree-optimization/18463] [4.0 Regression] suboptimal use of fancy x86 addressing modes

[rakdver at atrey dot karlin dot mff dot cuni dot cz]

------- Additional Comments From rakdver at atrey dot karlin dot mff dot cuni dot cz  2005-09-14 12:41 -------
Subject: Re:  [4.0 Regression] suboptimal use of fancy x86 addressing modes

> It looks like it is just following existing practice?

yes, I know.  The practice is just wrong, though.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18463

[Bug tree-optimization/18463] [4.0 Regression] suboptimal use of fancy x86 addressing modes

[rguenth at gcc dot gnu dot org]

------- Additional Comments From rguenth at gcc dot gnu dot org  2005-09-14 12:22 -------
It looks like it is just following existing practice?

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18463

[Bug tree-optimization/18463] [4.0 Regression] suboptimal use of fancy x86 addressing modes

[rakdver at gcc dot gnu dot org]

------- Additional Comments From rakdver at gcc dot gnu dot org  2005-09-14 11:04 -------
Auch; that patch is actually a very bad idea.  Pretending that complex
addressing modes are cheaper, when they are not, is just confusing.  If there
are optimizers that indeed want to prefer complex addressing modes (like cse and
combine, I suppose), they should reflect that in the cost function they are
using, rather than forcing this missconception to every other optimizer.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18463

[Bug tree-optimization/18463] [4.0 Regression] suboptimal use of fancy x86 addressing modes

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-09-14 08:21 -------
(From update of attachment 7541)
Already in mainline


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
Attachment #7541 is|0                           |1
           obsolete|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18463

[Bug tree-optimization/18463] [4.0 Regression] suboptimal use of fancy x86 addressing modes

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-13 18:51 -------
This is what we get one the mainline:
.L4:
        movl    (%ecx), %eax
        addl    $4, %ecx
        movl    %eax, (%edi,%edx,4)
        movl    (%ebp,%edx,4), %eax
        movl    %eax, (%esi,%edx,4)
        incl    %edx
        cmpl    %edx, %ebx
        jne     .L4

Note the code in comment #4  has a target patch which improves this a little further:
Index: i386.c
===============================================================
====
RCS file: /cvs/gcc/gcc/gcc/config/i386/i386.c,v
retrieving revision 1.858
diff -u -p -r1.858 i386.c
--- i386.c      6 Sep 2005 19:57:46 -0000       1.858
+++ i386.c      13 Sep 2005 18:49:44 -0000
@@ -5273,6 +5273,10 @@ ix86_address_cost (rtx x)
   /* More complex memory references are better.  */
   if (parts.disp && parts.disp != const0_rtx)
     cost--;
+
+  if (parts.scale != 1)
+    cost--;
+
   if (parts.seg != SEG_DEFAULT)
     cost--;
 

But since I don't have SPEC, I have not submitted the patch.  Steven could you test this patch and 
submit it for me?

ChangeLog (please make a better changelog): 
(ix86_address_cost): More complex is cheaper than anything else.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[4.0/4.1 Regression]        |[4.0 Regression] suboptimal
                   |suboptimal use of fancy x86 |use of fancy x86 addressing
                   |addressing modes            |modes


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18463

[Bug tree-optimization/18463] [4.0 Regression] suboptimal use of fancy x86 addressing modes

[tbptbp at gmail dot com]

------- Additional Comments From tbptbp at gmail dot com  2005-01-29 03:15 -------
Some recent discussion about related symptoms.

http://gcc.gnu.org/ml/gcc/2005-01/msg01667.html

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18463

[Bug tree-optimization/18463] [4.0 Regression] suboptimal use of fancy x86 addressing modes

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-01-29 02:37 -------
This problem is bigger than just the int<-->fp problem of the original 
bug report.  It's basically the same problem, namely a poor choice of 
addressing modes. 

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[4.0 Regression] Moving     |[4.0 Regression] suboptimal
                   |floating point through an   |use of fancy x86 addressing
                   |integer register            |modes


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18463