GCC optimization problem

[Joel Sherrill]

Silverio Diquigiovanni wrote:
> 
> Hi to all,
> I'm aware this isn't a good place to post this question but I hope
> someone can help me in this hard moment....
> 
> Architecture : sh7045F
> RTOS         : RTEMS 4.5
> GCC          : sh-rtems-gcc ( 2.95.2 )
> 
> In my source code I must often use an assembler line which call an
> software trap called " trap #37 ". During that trap a function modify
> various variables and also the variable called " my_var " :
> 
> < -- start snip code --- >
> 
>     while ( my_var && 0x08 )
>     {
>         asm( " trapa #37 " : : );    /* some function modify my_var */
>     }
> 
> < -- end snip code --- >

A lot depends on precisely what gcc thinks is happening in the
asm.  My intuition tells me that it does not think my_var can
change.  I see a couple of options:

  + Make my_var volatile
  + add output constraints that say my_var could be modified.

I think your as should be "asm volatile" as well.

> In the not optimized code ( -O0 ) the result code is :
> 
> < -- start snip code --- >
> 
> L18:    mov.l   L102,r1         ! my_var address
>         mov.b   @r1,r2
>         mov     #8,r1
>         and     r2,r1
>         exts.b  r1,r2
>         tst     r2,r2
>         bt      L20
>         bra     L19
>         nop
> 
> L20:    trapa #37
>         bra     L18
>         nop
> 
> L19:    ........
> 
> < -- end snip code --- >
> 
> This work very fine, but when I try to optimize the code with -O2,
> -O3, -O.., a bad code result because the compiler isn't aware of
> assembler body matter .....
> 
> < -- start snip code --- >
> 
> L17:    mov     r10,r0
>         mov.b   @(r0,r8),r1
>         mov     r1,r0
>         and     #8,r0
>         tst     r0,r0
>         bf      L19
> 
> L20:    trapa #37
>         bra     L20
>         nop
> 
> L19:    ........
> 
> < -- end snip code --- >
> 
> how you can see in L20 don't read the possible new value of
> " my_var " variable. In this case this become an infinite bad loop.
> 
> I have tried to declare my_var " volatile " but the result is same.
> 
> Then,
> how I can teach to compiler that my assembler routine ( trapa #37 )
> can modify the value of " my_var " ?
> 
> Thank you very much !!!!!!!
> Best Regards,
> 
> Silverio Diquigiovanni

-- 
Joel Sherrill, Ph.D.             Director of Research & Development
joel at OARcorp.com                 On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
   Support Available             (256) 722-9985