[Bug rtl-optimization/42522] (zero_extract:SI (mem:QI) ...) misoptimized

[Bug rtl-optimization/42522] (zero_extract:SI (mem:QI) ...) misoptimized

[ami_stuff at o2 dot pl]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42522

ami_stuff <ami_stuff at o2 dot pl> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |

--- Comment #15 from ami_stuff <ami_stuff at o2 dot pl> 2011-05-03 12:34:47 UTC ---
Almost 1,5 year passed and still no one cares to fix a long-standing, critical
bug which generates a broken code for m68k target (ffmpeg, sox, ...).

[Bug rtl-optimization/42522] (zero_extract:SI (mem:QI) ...) misoptimized

[mikpe at it dot uu.se]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42522

Mikael Pettersson <mikpe at it dot uu.se> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mikpe at it dot uu.se

--- Comment #16 from Mikael Pettersson <mikpe at it dot uu.se> 2012-02-19 23:48:41 UTC ---
Since r171341 (a big complex optabs cleanup patch) trunk aka 4.7 generates

bar:
        move.l 4(%sp),%a0
        tst.b (%a0)
        jne .L3
        clr.l %d1
        move.b 1(%a0),%d1
        swap %d1
        clr.w %d1
        clr.l %d0
        move.b 2(%a0),%d0
        lsl.l #8,%d0
        or.l %d1,%d0
        or.b 3(%a0),%d0
        moveq #8,%d1
        cmp.l %d0,%d1
        scs %d0
        extb.l %d0
        neg.l %d0
        rts
.L3:
        clr.l %d0
        rts

for the m68k-elf target.  (The code is much better for m68k-linux.)

[Bug rtl-optimization/42522] (zero_extract:SI (mem:QI) ...) misoptimized

[law at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=42522

Jeffrey A. Law <law at redhat dot com> changed:

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

--- Comment #17 from Jeffrey A. Law <law at redhat dot com> ---
This bug has gone latent on the trunk; however, the problem still remains that
cse will incorrectly simplify a ZERO/SIGN_EXTRACT in some cases.

ZERO/SIGN_EXTRACT are somewhat special in that if they are extracting from a
memory operand, that memory operand will always have QImode regardless of the
size of the extracted field.

ie, the mode of the MEM in a ZERO/SIGN_EXTRACT does not really mean anything
and the ZERO/SIGN_EXTRACT can read bits beyond QImode.

So given a (mem:QI x) with an equivalence to (const_int 0) in the hash tables. 
If we have an extraction like

(zero_extract:SI (mem:QI x) (const_int 0) (const_int 24))

CSE will substitute (const_int 0) for the MEM in the extraction resulting in

(zero_extract:SI (const_int 0) (const_int 0) (const_int 24))

Which simplifies to (const_int 0)

For REGs, the mode is the same as the operand of the insv/extv pattern, and may
(or may not) more closely resemble reality depending on the target.

The safe thing to do in CSE is to lookup the ZERO/SIGN_EXTRACT as a whole.

Given the bug is latent and not currently a regression this will need to wait
for the next stage1 development cycle.

[Bug rtl-optimization/42522] (zero_extract:SI (mem:QI) ...) misoptimized

[law at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=42522

--- Comment #18 from Jeffrey A. Law <law at gcc dot gnu.org> ---
Author: law
Date: Wed Apr 15 12:24:28 2015
New Revision: 222125

URL: https://gcc.gnu.org/viewcvs?rev=222125&root=gcc&view=rev
Log:
        PR rtl-optimization/42522
        * cse.c (fold_rtx): Try to simplify a ZERO_EXTRACT or
        SIGN_EXTRACT as a whole object rather than simplifying
        its operand.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/cse.c

[Bug rtl-optimization/42522] (zero_extract:SI (mem:QI) ...) misoptimized

[law at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=42522

Jeffrey A. Law <law at redhat dot com> changed:

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

--- Comment #19 from Jeffrey A. Law <law at redhat dot com> ---
This bug was latent on the trunk.  I've committed a fix so that it shouldn't
ever come back again.

[Bug rtl-optimization/42522] (zero_extract:SI (mem:QI) ...) misoptimized

[ami_stuff at o2 dot pl]

------- Comment #14 from ami_stuff at o2 dot pl  2010-05-09 13:54 -------
Any chance to see this bug fixed in the near future?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42522

[Bug rtl-optimization/42522] (zero_extract:SI (mem:QI) ...) misoptimized

[rguenth at gcc dot gnu dot org]

-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
      Known to fail|4.3.4 4.4.3                 |4.3.4 4.4.3 4.5.0
   Target Milestone|4.4.4                       |---


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42522

[Bug rtl-optimization/42522] (zero_extract:SI (mem:QI) ...) misoptimized

[schwab at linux-m68k dot org]

-- 

schwab at linux-m68k dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |4.3.4 4.4.3
   Target Milestone|---                         |4.4.4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42522

[Bug rtl-optimization/42522] (zero_extract:SI (mem:QI) ...) misoptimized

[schwab at linux-m68k dot org]

------- Comment #13 from schwab at linux-m68k dot org  2010-01-17 22:27 -------
The bug exists since the dawn of time.  The problem is that cse thinks that
(zero_extract:SI (mem:QI ...) ...) is zero if (mem:QI ...) is known to be zero,
but the first argument of zero_extract is only a base address if it is a memory
address.  On m68k this is only generated when STRICT_ALIGNMENT, and the bug was
hidden until e2496245b8453f49bb4428f281af7686be09ddf8:

2000-07-17  Zack Weinberg  <zack@wolery.cumb.org>
        (simplify_ternary_operation):  Do not examine MODE_BITSIZE of
           a CONST_INT, it will always be zero.


-- 

schwab at linux-m68k dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
          Component|target                      |rtl-optimization
     Ever Confirmed|0                           |1
 GCC target triplet|m68k-*-*                    |m68k-*-elf
   Last reconfirmed|0000-00-00 00:00:00         |2010-01-17 22:27:47
               date|                            |
            Summary|[m68k] Wrong code generated |(zero_extract:SI (mem:QI)
                   |with -O2/-O3                |...) misoptimized


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42522