[Bug optimization/11708] New: Non-Optimal jump code generation.

[Bug optimization/11708] New: Non-Optimal jump code generation.

[naveens at noida dot hcltech dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11708

           Summary: Non-Optimal jump code generation.
           Product: gcc
           Version: 3.3.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: naveens at noida dot hcltech dot com
                CC: gcc-bugs at gcc dot gnu dot org,joern dot rennecke at
                    superh dot com
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: sh-unknown-elf

I am seeing some redundant jumps in code generated by GCC3.3 and 
the mainline. e.g. I see this assembly generated for 
one of my benchmark (sh-elf target, options: -ml -m4 -O2)

        ! basic block 1
        cmp/pl  r3
        mov.w   .L147,r1
        bt      .L3     <--- conditional branch C1
        mov     #0,r1
        bra     .L208   <--- Branch B1
        mov     r1,r3   <--- delay slot
.L3:
        ! basic block 2
        mov     r1,r3
.L208:
        mov.w   .L149,r2
        mov.w   .L150,r1
        cmp/gt  r1,r3
        bt      .L7

It seems it should be better generated as:

        ! basic block 1
        cmp/pl  r3
        mov.w   .L147,r1
        bt      .L208    <--- conditional branch C1
        mov     #0,r1
.L208:
       ! basic block 2
        mov     r1,r3
        mov.w   .L149,r2
        mov.w   .L150,r1
        cmp/gt  r1,r3
        bt      .L7

So, a jump(bra) instruction could be eliminated.

[Bug optimization/11708] Non-Optimal jump code generation.

[naveens at noida dot hcltech dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11708



------- Additional Comments From naveens at noida dot hcltech dot com  2003-07-29 13:46 -------
Created an attachment (id=4505)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=4505&action=view)
Preprocessed file to reproduce the bug.

compile with 
sh-elf-gcc -ml -m4 -S -O2 -dA BUG.c to get the assembler dump.

[Bug optimization/11708] Non-Optimal jump code generation.

[pinskia at physics dot uc dot edu]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11708


pinskia at physics dot uc dot edu changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |pessimizes-code

[Bug optimization/11708] Non-Optimal jump code generation.

[naveens at noida dot hcltech dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11708



------- Additional Comments From naveens at noida dot hcltech dot com  2003-07-29 14:10 -------
It seems, delay slot scheduling has the responsibility of generating 
optimal code here. 

In fact, it is able to do the required transformation in some cases. e.g.

Consider this part of the assembler 
with -fno-delayed-branch -O2 -ml -m4 -S -dA

        ! basic block 25
        cmp/pl  r3
        mov.w   .L147,r1
        bt      .L203
        bra     .L93
        nop
.L203:
.L36:
        ! basic block 26

   ..... snipped ...

.L93:
        ! basic block 69
        mov     #0,r1
        bra     .L36
        nop
        .align 5

With -fdelayed-branch, this is optimized to:

         ! basic block 25
        cmp/pl  r3
        mov.w   .L147,r1
        bt      .L36
        mov     #0,r1
.L36:
        ! basic block 26
        mov.w   .L136,r3
        mov.w   .L149,r2
        add     r1,r3
        mov.w   .L150,r1
        cmp/gt  r1,r3

So, it seems that delayed branch scheduling is able to 
optimize some cases, but misses some. I will try to 
confirm the reason.

[Bug optimization/11708] [sh4-elf-gcc] Non-Optimal jump code generation.

[naveens at noida dot hcltech dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11708


naveens at noida dot hcltech dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|Non-Optimal jump code       |[sh4-elf-gcc] Non-Optimal
                   |generation.                 |jump code generation.

[Bug optimization/11708] [sh4-elf-gcc] Non-Optimal jump code generation.

[pinskia at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11708


pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|3.4                         |---

[Bug optimization/11708] [sh4-elf-gcc] Non-Optimal jump code generation.

[pinskia at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11708


pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |enhancement

[Bug optimization/11708] [sh4-elf-gcc] Non-Optimal jump code generation.

[dhazeghi at yahoo dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11708


dhazeghi at yahoo dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2003-08-25 00:04:00
               date|                            |


------- Additional Comments From dhazeghi at yahoo dot com  2003-08-25 00:03 -------
Confirmed with mainline (20030824).

[Bug optimization/11708] [sh4-elf-gcc] Non-Optimal jump code generation.

[naveens at noida dot hcltech dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11708



------- Additional Comments From naveens at noida dot hcltech dot com  2003-09-17 07:02 -------
Something weird has happened on the mainline. 
I don't get any assembler ouput (-O2 -m4 -ml) for this test case, 
when I compile with snapshot of 12th Sep, 2003. 

Anybody else seeing this sort of a problem or there is something 
wrong in my environment ?

[Bug optimization/11708] [sh4-elf-gcc] Non-Optimal jump code generation.

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-01-03 19:45 -------
no -O2 enables -funit-at-a-time which removes static functions, remove the static from the 
file and it should produce code.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11708

[Bug optimization/11708] [sh4-elf-gcc] Non-Optimal jump code generation.

[naveens at noida dot hcltech dot com]

------- Additional Comments From naveens at noida dot hcltech dot com  2004-01-05 06:01 -------
Yes, thats right !! although it wasn't obvious immediately.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11708