[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[pinskia at gcc dot gnu dot org]

-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|SUSPENDED                   |NEW
   Last reconfirmed|2005-05-04 18:12:43         |2006-01-15 20:30:45
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[hjl at lucon dot org]

------- Comment #19 from hjl at lucon dot org  2006-02-27 23:40 -------
Even with the new patch

http://gcc.gnu.org/ml/gcc-patches/2006-02/msg01994.html

I still got the same result. The new see pass won't touch

(insn:HI 13 9 14 2 (set (reg/v:SI 73 [ t ])
        (mem/s:SI (symbol_ref:DI ("state") [flags 0x40] <var_decl 0x2a98541420
state>) [3 state+0 S4 A32])) 40 {*movsi_1} (nil)
    (nil))

(insn:HI 14 13 16 2 (parallel [
            (set (reg/v:SI 73 [ t ])
                (xor:SI (mem/s:SI (symbol_ref:DI ("S") [flags 0x40] <var_decl
0x2a985412c0 S>) [3 S+0 S4 A32])
                    (reg/v:SI 73 [ t ])))
            (clobber (reg:CC 17 flags))
        ]) 340 {*xorsi_1} (insn_list:REG_DEP_TRUE 13 (nil))
    (expr_list:REG_UNUSED (reg:CC 17 flags)
        (expr_list:REG_EQUAL (xor:SI (mem/s:SI (symbol_ref:DI ("S") [flags
0x40] <var_decl 0x2a985412c0 S>) [3 S+0 S4 A32])
                (mem/s:SI (symbol_ref:DI ("state") [flags 0x40] <var_decl
0x2a98541420 state>) [3 state+0 S4 A32]))
            (nil))))

(insn:HI 16 14 18 2 (set (mem/s:SI (symbol_ref:DI ("state") [flags 0x40]
<var_decl 0x2a98541420 state>) [3 state+0 S4 A32])
        (reg/v:SI 73 [ t ])) 40 {*movsi_1} (insn_list:REG_DEP_TRUE 14 (nil))
    (nil))

(insn:HI 18 16 21 2 (set (reg:DI 79 [ t ])
        (zero_extend:DI (reg/v:SI 73 [ t ]))) 111 {zero_extendsidi2_rex64}
(nil)
    (expr_list:REG_DEAD (reg/v:SI 73 [ t ])
        (nil)))


-- 

hjl at lucon dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Version|4.0.0                       |4.2.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[steven at gcc dot gnu dot org]

------- Comment #20 from steven at gcc dot gnu dot org  2007-04-20 21:58 -------
It is my intention to fix see.c to work on x86* hardware, so I'm taking this
bug.


-- 

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|2006-01-15 20:30:45         |2007-04-20 21:58:36
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[steven at gcc dot gnu dot org]

------- Comment #21 from steven at gcc dot gnu dot org  2007-04-20 22:10 -------
Collection of important related links:
http://gcc.gnu.org/ml/gcc-patches/2006-04/msg00766.html
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27437#c5


-- 

steven at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  BugsThisDependsOn|                            |27437


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[steven at gcc dot gnu dot org]

------- Comment #22 from steven at gcc dot gnu dot org  2007-05-25 20:32 -------
With the current implementation of SEE it is almost impossible to make it work
on x86.  You have to take into account the liveness of the flags register, and
there currently is no way to include that in the dataflow equations.  Maybe
someone else knows how to do this...


-- 

steven at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|steven at gcc dot gnu dot   |unassigned at gcc dot gnu
                   |org                         |dot org
             Status|ASSIGNED                    |NEW


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant zero extension instructions in loop optimization

[bonzini at gnu dot org]

------- Comment #23 from bonzini at gnu dot org  2009-02-18 13:02 -------
> Gcc doesn't know/remember
> 
> movl    S(,%rax,4), %eax
> 
> will zero extend to 64bit. I don't know you can touch only the lower
> 32bit bits.

This could be fixed by LOAD_EXTEND_OP, right?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant zero extension instructions in loop optimization

[hjl dot tools at gmail dot com]

------- Comment #24 from hjl dot tools at gmail dot com  2009-02-18 21:24 -------
I tried:

--- config/i386/i386.h.zero     2009-02-18 08:42:40.000000000 -0800
+++ config/i386/i386.h  2009-02-18 13:16:26.000000000 -0800
@@ -1940,6 +1940,11 @@ do {                                                    
\
    is done just by pretending it is already truncated.  */
 #define TRULY_NOOP_TRUNCATION(OUTPREC, INPREC) 1

+/* When in 64-bit mode, move insns will zero extend SImode.  All other
+   references are unknown.  */
+#define LOAD_EXTEND_OP(MODE) \
+  (TARGET_64BIT && (MODE) == SImode ? ZERO_EXTEND : UNKNOWN)
+
 /* A macro to update M and UNSIGNEDP when an object whose type is
    TYPE and which has the specified mode and signedness is to be
    stored in a register.  This macro is only called when TYPE is a

It makes no differences.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant zero extension instructions in loop optimization

[hjl dot tools at gmail dot com]

------- Comment #25 from hjl dot tools at gmail dot com  2009-02-18 21:31 -------
All 32bit load insns are zero extended to 64bit, not just move.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant zero extension instructions in loop optimization

[hjl dot tools at gmail dot com]

------- Comment #26 from hjl dot tools at gmail dot com  2009-09-17 13:47 -------
*** Bug 34653 has been marked as a duplicate of this bug. ***


-- 

hjl dot tools at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dean at arctic dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant zero extension instructions in loop optimization

[hjl dot tools at gmail dot com]

------- Comment #27 from hjl dot tools at gmail dot com  2010-05-22 20:48 -------
Fixed by revision 159342:

http://gcc.gnu.org/ml/gcc-cvs/2010-05/msg00394.html


-- 

hjl dot tools at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.6.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-29 03:44 -------
New patch was posted:
http://gcc.gnu.org/ml/gcc-patches/2005-09/msg01769.html

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
 GCC target triplet|x86_64-*-linux-gnu          |x86_64-*-*


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-27 13:36 -------
(In reply to comment #13)
> No, I didn't :-)
> 
> So you suggest closing this as SUSPEND?

Yes.  But note IBM haifa is going to submit a pass for this and has been outlined before.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |SUSPENDED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-01-27 10:24 -------
/me should read up on the amd64 instruction set first :-( 

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|steven at gcc dot gnu dot   |unassigned at gcc dot gnu
                   |org                         |dot org
             Status|REOPENED                    |NEW


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-01-27 10:23 -------
moron alert 

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-01-27 10:13 -------
More knowledgable sources than me say: 
"[mov     %eax, %eax ] is not nop.  32bit operations implicitly zero 
extend, so this is zero extension.  There is no movqzx." 
 
 

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[stevenb at suse dot de]

------- Additional Comments From stevenb at suse dot de  2005-01-27 07:14 -------
Subject: Re:  Redundant instructions in loop optimization

On Thursday 27 January 2005 07:05, rth at gcc dot gnu dot org wrote:
> ------- Additional Comments From rth at gcc dot gnu dot org  2005-01-27
> 06:05 ------- Steven, you do realize this is essentially unfixable without
> a new pass that optimially places widened operations, don't you?

No, I didn't :-)

So you suggest closing this as SUSPEND?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[rth at gcc dot gnu dot org]

------- Additional Comments From rth at gcc dot gnu dot org  2005-01-27 06:05 -------
Steven, you do realize this is essentially unfixable without a new pass that
optimially places widened operations, don't you?

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[hjl at lucon dot org]

------- Additional Comments From hjl at lucon dot org  2005-01-27 04:03 -------
Gcc doesn't know/remember

movl    S(,%rax,4), %eax

will zero extend to 64bit. I don't know you can touch only the lower
32bit bits.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-01-27 01:43 -------
GCC can't see that this is a NOP because of the zero_extend: 
 
#(insn 34 30 37 (set (reg:DI 0 ax [orig:76 D.1460 ] [76]) 
#        (zero_extend:DI (reg:SI 0 ax [orig:70 D.1460 ] [70]))) 111 
{zero_extendsidi2_rex64} (insn_list:REG_DEP_ANTI 30 (insn_list:REG_DEP_TRUE 29 
(nil))) 
#    (nil)) 
        mov     %eax, %eax      # D.1460, D.1460        # 34    
zero_extendsidi2_rex64/1        [length = 3] 
 
Perhaps we should have a peephole2 for this.  I'm curious why we can not 
eliminate the move earlier on, though. 
 
 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-01-27 01:36 -------
My first suspicion was the `register' keyword, but that's not it: 
 
-------------------------------------------------- 
extern unsigned int S[]; 
extern unsigned int state[]; 
 
unsigned int 
foo () 
{ 
  unsigned int t; 
  int j; 
 
  t=0; 
  for (j=0; j<16; j+=4) 
   { 
     t = state[j + 0] ^= S[t]; 
     t = state[j + 1] ^= S[t]; 
   } 
  return t; 
} 
-------------------------------------------------- 
 
--> 
 
-------------------------------------------------- 
foo: 
.LFB2: 
        xorl    %eax, %eax 
        movl    $4, %ecx 
        movl    $state, %edx 
        .p2align 4,,7 
.L2: 
        mov     %eax, %eax 
        movl    S(,%rax,4), %eax 
        xorl    (%rdx), %eax 
        movl    %eax, (%rdx) 
        mov     %eax, %eax        <-- bah! 
        movl    S(,%rax,4), %eax 
        xorl    4(%rdx), %eax 
        movl    %eax, 4(%rdx) 
        addq    $16, %rdx 
        decl    %ecx 
        jne     .L2 
        rep ; ret 
.LFE2: 
-------------------------------------------------- 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-01-27 01:27 -------
Uhm, right.  OK then, lemme have a look. 

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |steven at gcc dot gnu dot
                   |dot org                     |org
             Status|NEW                         |ASSIGNED
   Last reconfirmed|2005-01-24 18:52:13         |2005-01-27 01:27:34
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-24 18:52 -------
Confirmed again.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|2004-11-27 20:49:43         |2005-01-24 18:52:13
               date|                            |
   Target Milestone|4.0.0                       |---


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[hjl at lucon dot org]

------- Additional Comments From hjl at lucon dot org  2005-01-24 18:17 -------
Looking at the mainline result, I still see "mov %eax, %eax", which was
the original bug report for mainline.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-01-23 13:35 -------
Mainline today produces: 
 
foo: 
.LFB2: 
        xorl    %eax, %eax 
        movl    $4, %ecx 
        movl    $state, %edx 
        .p2align 4,,7 
.L2: 
        mov     %eax, %eax 
        movl    S(,%rax,4), %eax 
        xorl    (%rdx), %eax 
        movl    %eax, (%rdx) 
        mov     %eax, %eax 
        movl    S(,%rax,4), %eax 
        xorl    4(%rdx), %eax 
        movl    %eax, 4(%rdx) 
        mov     %eax, %eax 
        movl    S(,%rax,4), %eax 
        xorl    8(%rdx), %eax 
        movl    %eax, 8(%rdx) 
        mov     %eax, %eax 
        movl    S(,%rax,4), %eax 
        xorl    12(%rdx), %eax 
        movl    %eax, 12(%rdx) 
        addq    $16, %rdx 
        decl    %ecx 
        jne     .L2 
        andl    $255, %eax 
        ret 
 

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.0.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |enhancement
   Last reconfirmed|0000-00-00 00:00:00         |2004-11-27 20:49:43
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[rakdver at gcc dot gnu dot org]

------- Additional Comments From rakdver at gcc dot gnu dot org  2004-09-11 19:55 -------
ivopts gets rid of the extends once the record_bounds pass is enabled by the 
following patch.  The nonsensical "mov     %eax, %eax" type moves however
still remain in the code.

Zdenek

Index: tree-optimize.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/tree-optimize.c,v
retrieving revision 2.45
diff -c -3 -p -r2.45 tree-optimize.c
*** tree-optimize.c     9 Sep 2004 20:53:36 -0000       2.45
--- tree-optimize.c     11 Sep 2004 19:44:56 -0000
*************** init_tree_optimization_passes (void)
*** 392,397 ****
--- 392,398 ----
    NEXT_PASS (pass_loop_init);
    NEXT_PASS (pass_lim);
    NEXT_PASS (pass_iv_canon);
+   NEXT_PASS (pass_record_bounds);
    NEXT_PASS (pass_if_conversion);
    NEXT_PASS (pass_vectorize);
    NEXT_PASS (pass_linear_transform);
Index: tree-pass.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/tree-pass.h,v
retrieving revision 2.15
diff -c -3 -p -r2.15 tree-pass.h
*** tree-pass.h 9 Sep 2004 20:53:37 -0000       2.15
--- tree-pass.h 11 Sep 2004 19:44:56 -0000
*************** extern struct tree_opt_pass pass_loop;
*** 126,131 ****
--- 126,132 ----
  extern struct tree_opt_pass pass_loop_init;
  extern struct tree_opt_pass pass_lim;
  extern struct tree_opt_pass pass_iv_canon;
+ extern struct tree_opt_pass pass_record_bounds;
  extern struct tree_opt_pass pass_if_conversion;
  extern struct tree_opt_pass pass_vectorize;
  extern struct tree_opt_pass pass_complete_unroll;
Index: tree-ssa-loop.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/tree-ssa-loop.c,v
retrieving revision 2.17
diff -c -3 -p -r2.17 tree-ssa-loop.c
*** tree-ssa-loop.c     8 Sep 2004 15:28:56 -0000       2.17
--- tree-ssa-loop.c     11 Sep 2004 19:44:57 -0000
*************** struct tree_opt_pass pass_iv_canon =
*** 263,268 ****
--- 263,297 ----
    0                                   /* letter */
  };

+ /* Record bounds on numbers of iterations of loops.  */
+
+ static void
+ tree_ssa_loop_bounds (void)
+ {
+   if (!current_loops)
+     return;
+
+   estimate_numbers_of_iterations (current_loops);
+   scev_reset ();
+ }
+
+ struct tree_opt_pass pass_record_bounds =
+ {
+   "bounds",                           /* name */
+   NULL,                                       /* gate */
+   tree_ssa_loop_bounds,                       /* execute */
+   NULL,                                       /* sub */
+   NULL,                                       /* next */
+   0,                                  /* static_pass_number */
+   0,                                  /* tv_id */
+   PROP_cfg | PROP_ssa,                        /* properties_required */
+   0,                                  /* properties_provided */
+   0,                                  /* properties_destroyed */
+   0,                                  /* todo_flags_start */
+   0,                                  /* todo_flags_finish */
+   0                                   /* letter */
+ };
+
  /* Complete unrolling of loops.  */

  static void


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-09-10 23:21 -------
Actually this can be reproduce without IV.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|rakdver at atrey dot karlin |unassigned at gcc dot gnu
                   |dot mff dot cuni dot cz     |dot org
             Status|ASSIGNED                    |NEW
   Target Milestone|4.0.0                       |---


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[hubicka at gcc dot gnu dot org]

------- Additional Comments From hubicka at gcc dot gnu dot org  2004-09-10 09:53 -------
The funny instruction is a zero extension done via 32bit move.  tree-ssa-ivopts should promote the induction variable to 64bit counter for sure.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |rakdver at atrey dot karlin
                   |dot org                     |dot mff dot cuni dot cz
             Status|UNCONFIRMED                 |ASSIGNED
   Target Milestone|---                         |4.0.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-09-10 08:05 -------
The patch listed here might help: <http://gcc.gnu.org/ml/gcc/2004-09/msg00377.html>.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[law at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |law at redhat dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387

[Bug rtl-optimization/17387] Redundant instructions in loop optimization

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|c                           |rtl-optimization
           Keywords|                            |missed-optimization
            Summary| Redundant instructions in  |Redundant instructions in
                   |loop optimization           |loop optimization


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17387