[Bug optimization/13373] New: optimization with -frerun-cse-after-loop -fexpensive-optimizations produces wrong code on mcore

[Bug optimization/13373] New: optimization with -frerun-cse-after-loop -fexpensive-optimizations produces wrong code on mcore

[olaf at bonorden dot de]

Compilation of a crc8 faild: 
 
source: 
unsigned short Calc_crc8(unsigned char data, unsigned short crc ) 
{ 
        unsigned char i,x16,carry; 
 
        for (i = 0; i < 8; i++) 
    { 
                x16 = (unsigned char)((data & 1) ^ ((unsigned char)crc & 1)); 
                data >>= 1; 
                if (x16 == 1) 
                { 
                   crc ^= 0x4002; 
                   carry = 1; 
                } 
                else carry = 0; 
                crc >>= 1; 
                if (carry) 
                   crc |= 0x8000; 
                else 
                   crc &= 0x7fff; 
    } 
        return crc; 
} 
----------- 
Compilation with "mcore-be-elf-gcc -mbig-endian -O0 -frerun-cse-after-loop 
crc.c -S" is ok, 
with "mcore-be-elf-gcc -mbig-endian -O0 -frerun-cse-after-loop 
-fexpensive-optimizations crc.c -S" the code is not correct. 
 
Part in the produced assembler code: 
first version: 
        ld.h    r7,(r8,2) 
        bgeni   r6,14 
        addi    r6,2    // 16386 0x4002 
        xor     r7,r6 
        st.h    r7,(r8,2) 
        ldw     r6,(r8,8) 
        st.b    r6,(r8,6) 
wrong second version: 
        ld.h    r7,(r8,2) 
        st.h    r7,(r8,2) 
        ldw     r7,(r8,8) 
        st.b    r7,(r8,6) 
The xor part is removed :-( 
 
Compiler without optimization is nearly useless.

-- 
           Summary: optimization with -frerun-cse-after-loop -fexpensive-
                    optimizations produces wrong code on mcore
           Product: gcc
           Version: 3.3.2
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: olaf at bonorden dot de
                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: mcore-be-elf


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13373

[Bug optimization/13373] optimization with -frerun-cse-after-loop -fexpensive-optimizations produces wrong code on mcore

[kazu at cs dot umass dot edu]

------- Additional Comments From kazu at cs dot umass dot edu  2003-12-24 05:49 -------
In my reduced case, xor is removed during the first cse pass.



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13373

[Bug optimization/13373] optimization with -frerun-cse-after-loop -fexpensive-optimizations produces wrong code on mcore

[kazu at cs dot umass dot edu]

------- Additional Comments From kazu at cs dot umass dot edu  2003-12-24 05:30 -------
Confirmed on 3.3.
Not the case on mainline.
Reduced to:

short
foo (short crc)
{
  if (crc)
    crc ^= 0x4002;

  return crc;
}

With -O1 instead of -O0, the difference is easier to see:

 foo:
        zexth   r2
        cmpnei  r2,0
        jbf     .L2
-       bgeni   r7,14
-       addi    r7,2    // 16386 0x4002
-       xor     r2,r7
+       zexth   r2
 .L2:
        sexth   r2
        jmp     r15


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |kazu at cs dot umass dot edu
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2003-12-24 05:30:51
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13373

[Bug optimization/13373] optimization with -frerun-cse-after-loop -fexpensive-optimizations produces wrong code on mcore

[kazu at cs dot umass dot edu]

------- Additional Comments From kazu at cs dot umass dot edu  2003-12-24 05:52 -------
Reduced to:

short
foo (short crc)
{
  return crc ^ 0x4002;
}

The problem still seems to be cse.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13373

[Bug target/13373] optimization with -frerun-cse-after-loop -fexpensive-optimizations produces wrong code on mcore

[kazu at cs dot umass dot edu]

------- Additional Comments From kazu at cs dot umass dot edu  2003-12-24 15:30 -------
Backporting the following patch fixes the problem.

http://gcc.gnu.org/ml/gcc-patches/2003-09/msg00044.html

I haven't run the testsuite though.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rsandifo at redhat dot com
          Component|optimization                |target


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13373

[Bug target/13373] [3.3 only] optimization with -frerun-cse-after-loop -fexpensive-optimizations produces wrong code on mcore

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
            Summary|optimization with -frerun-  |[3.3 only] optimization with
                   |cse-after-loop -fexpensive- |-frerun-cse-after-loop -
                   |optimizations produces wrong|fexpensive-optimizations
                   |code on mcore               |produces wrong code on mcore
   Target Milestone|---                         |3.3.3


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13373

[Bug target/13373] [3.3 only] optimization with -frerun-cse-after-loop -fexpensive-optimizations produces wrong code on mcore

[kazu at cs dot umass dot edu]

------- Additional Comments From kazu at cs dot umass dot edu  2003-12-24 21:20 -------
OK with 3.2 branch, so this is a regression from 3.2.x.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13373

[Bug target/13373] [3.3 regression] optimization with -frerun-cse-after-loop -fexpensive-optimizations produces wrong code on mcore

[kazu at cs dot umass dot edu]

------- Additional Comments From kazu at cs dot umass dot edu  2003-12-24 21:38 -------
Here is a patch:

http://gcc.gnu.org/ml/gcc-patches/2003-12/msg02057.html


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |patch


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13373

[Bug target/13373] [3.3 regression] optimization with -frerun-cse-after-loop -fexpensive-optimizations produces wrong code on mcore

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |critical


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13373

[Bug target/13373] [3.3 regression] optimization with -frerun-cse-after-loop -fexpensive-optimizations produces wrong code on mcore

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2004-01-01 00:24 -------
Subject: Bug 13373

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	gcc-3_3-branch
Changes by:	kazu@gcc.gnu.org	2004-01-01 00:24:24

Modified files:
	gcc            : ChangeLog 
	gcc/config/mcore: mcore-protos.h mcore.c mcore.h mcore.md 

Log message:
	Backport from mainline:
	
	2003-09-12  Richard Sandiford  <rsandifo@redhat.com>
	
	PR target/13373
	* config/mcore/mcore-protos.h (mcore_r15_operand_p): Declare.
	(mcore_secondary_reload_class): Declare.
	(mcore_output_inline_const_forced): Remove.
	* config/mcore/mcore.md (movsi): Remove the code that forced
	non-inlineable constants into a register if the target was r15
	or the stack pointer.  Remove constant restrictions from the main
	define_insn.  Remove r <- I, r <- M and r <- N alternatives in favor
	of an r <- P alternative.  Remove fallback define_insn for reload.
	(movhi, movqi): Use gen_lowpart rather than gen_SUBREG.  Remove reload
	define_insn.  Use mcore_output_move in the remaining define_insn.
	Adjust condition and constraints in the way as for movsi.
	(movdi): Always split unacceptable constants into two.  Use
	simplify_gen_subreg instead of operand_subword{,_force}.
	* config/mcore/mcore.c (mcore_output_inline_const_forced): Remove.
	(mcore_output_move): Support HImode and QImode moves as well.
	(mcore_m15_operand_p): New function.
	(mcore_reload_class): Use it to detect cases where LRW_REGS are better.
	(mcore_secondary_reload_class): New function.
	* config/mcore/mcore.h (SECONDARY_RELOAD_CLASS): Redefine in
	terms of mcore_secondary_reload_class.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-3_3-branch&r1=1.16114.2.863&r2=1.16114.2.864
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/mcore/mcore-protos.h.diff?cvsroot=gcc&only_with_tag=gcc-3_3-branch&r1=1.10&r2=1.10.20.1
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/mcore/mcore.c.diff?cvsroot=gcc&only_with_tag=gcc-3_3-branch&r1=1.40&r2=1.40.4.1
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/mcore/mcore.h.diff?cvsroot=gcc&only_with_tag=gcc-3_3-branch&r1=1.41.4.1&r2=1.41.4.2
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/mcore/mcore.md.diff?cvsroot=gcc&only_with_tag=gcc-3_3-branch&r1=1.7.4.1&r2=1.7.4.2



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13373

[Bug target/13373] [3.3 regression] optimization with -frerun-cse-after-loop -fexpensive-optimizations produces wrong code on mcore

[kazu at cs dot umass dot edu]

------- Additional Comments From kazu at cs dot umass dot edu  2004-01-01 00:28 -------
Just backported a fix.


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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13373