[Bug target/22497] New: A register is wasted in simple vectorised loops

[Bug target/22497] New: A register is wasted in simple vectorised loops

[uros at kss-loka dot si]

Hello!

Consider this simple testcase:

#define N 16

short ia[N];
short ic[N] = {0,3,6,9,12,15,18,21,24,27,30,33,36,39,42,45};
short ib[N] = {0,3,6,9,12,15,18,21,24,27,30,33,36,39,42,45};


int main ()
{
  int i;

  for (i = 0; i < N; i++)
    ia[i] = ib[i] + ic[i];

  return 0;
}

The loop in this testcase is compiled with 'gcc -O2 -ftree-vectorize -msse2' 
into:

.L2:
	movdqa	ib(%eax), %xmm0
	paddw	ic(%eax), %xmm0
	incl	%edx
	movdqa	%xmm0, ia(%eax)
	addl	$16, %eax
	cmpl	$2, %edx
	jne	.L2

There is no (,%reg,16) SIB mode available in i386, and it looks to me that loop 
optimizer fallbacks to the most simple addressing mode in this case. 
Unfortunatelly, %edx register is wasted in above code.

A better code would be:

.L2:
	movdqa	ib(,%eax,8), %xmm0
	paddw	ic(,%eax,8), %xmm0
	movdqa	%xmm0, ia(,%eax,8)
	addl	$2, %eax
	cmpl	$4, %eax
	jne	.L2

or with the simplest addressing scheme:

.L2:
	movdqa	ib(%eax), %xmm0
	paddw	ic(%eax), %xmm0
	movdqa	%xmm0, ia(%eax)
	addl	$16, %eax
	cmpl	$32, %eax
	jne	.L2

Uros.

-- 
           Summary: A register is wasted in simple vectorised loops
           Product: gcc
           Version: 4.1.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P2
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: uros at kss-loka dot si
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: i686-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22497

[Bug target/22497] A register is wasted in simple vectorised loops

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-07-15 22:18 -------
With last night's compiler I get:
.L2:
        movdqa  ib(,%eax,8), %xmm0
        paddw   ic(,%eax,8), %xmm0
        movdqa  %xmm0, ia(,%eax,8)
        addl    $2, %eax
        cmpl    $4, %eax
        jne     .L2

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22497

[Bug target/22497] A register is wasted in simple vectorised loops

[uros at kss-loka dot si]

------- Additional Comments From uros at kss-loka dot si  2005-07-18 06:36 -------
(In reply to comment #1)
> With last night's compiler I get:

Strange...

I have tested gcc snapshot 'gcc version 4.1.0 20050716 (experimental)', 
with 'gcc -O2 -ftree-vectorize -msse2' and it still produces code with both %
eax and %edx used.

Andrew, which gcc version and compile flags have you used to produce your asm 
code?

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22497

[Bug target/22497] A register is wasted in simple vectorised loops

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-07-18 06:49 -------
(In reply to comment #2)
> I have tested gcc snapshot 'gcc version 4.1.0 20050716 (experimental)', 
> with 'gcc -O2 -ftree-vectorize -msse2' and it still produces code with both %
> eax and %edx used.
> 
> Andrew, which gcc version and compile flags have you used to produce your asm 
> code?

Oh, I have a local patch which will cause this, woops:
Index: config/i386/i386.c
===============================================================
====
RCS file: /cvs/gcc/gcc/gcc/config/i386/i386.c,v
retrieving revision 1.842
diff -u -p -r1.842 i386.c
--- config/i386/i386.c  14 Jul 2005 07:46:16 -0000      1.842
+++ config/i386/i386.c  18 Jul 2005 06:48:33 -0000
@@ -5044,6 +5044,10 @@ 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--;
 

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2005-07-18 06:49:26
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22497

[Bug target/22497] A register is wasted in simple vectorised loops

[uros at kss-loka dot si]

------- Additional Comments From uros at kss-loka dot si  2005-07-22 09:41 -------
(In reply to comment #3)

> Oh, I have a local patch which will cause this, woops:

> +  if (parts.scale != 1)
> +    cost--;
> +
>    if (parts.seg != SEG_DEFAULT)
>      cost--;

Do you plan to submit this patch to gcc-patches? It sure optimizes some code, 
and the testcase from decription could be used to show benefits of this patch.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22497

[Bug target/22497] A register is wasted in simple vectorised loops

[uros at kss-loka dot si]

------- Additional Comments From uros at kss-loka dot si  2005-09-28 08:14 -------
The problem was fixed by the patch for PR 18463:
http://gcc.gnu.org/ml/gcc-patches/2005-09/msg01159.html

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22497

[Bug target/22497] A register is wasted in simple vectorised loops

[uros at kss-loka dot si]

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22497