topic Re: problem with optimization in ColdFire/68K Microcontrollers and Processors

problem with optimization

doom — Mon, 30 Mar 2009 15:03:44 GMT

I use cw6.4 and when I use optimization it comes out an annoying problem.

the optimization setting are:

optimization level 4;

IPA FILE level optimization;

codes as follows:

void Delay20(void)
{

    KEYCNT=0;
    while(KEYCNT<=19)
    {
//    asm{NOP};
    }

}

If I do not use asm{NOP}; the result of disassmble are :

; 2808: void Delay20(void)
; 2809: {
; 2810:
; 2811:     KEYCNT=0;
; 2812:     while(KEYCNT<=19)
; 2813:     {
; 2814:
; 2815: //    asm{NOP};
;
0x00000000                    _Delay20:
;                             Delay20:
0x00000000 0x4E560000               link     a6,#0
;
; 2816:     }
; 2817:
; 2818:
;
0x00000004 0x60FE                   bra.s    *+0                   ; 0x00000004
;
; 2819: }
0x00000006 0x4E71                   nop

And if use asm{NOP}; the result of disassmble are :

; 2807: //~~~~~~~delay20ms~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
; 2808: void Delay20(void)
; 2809: {
; 2810:
;
0x00000000                    _Delay20:
;                             Delay20:
0x00000000 0x4E560000               link     a6,#0
;
; 2811:    KEYCNT=0;
; 2812:    while(KEYCNT<=19)
; 2813:    {
; 2814:
;
0x00000004 0x42B900000000           clr.l    _KEYCNT
0x0000000A 0x7013                   moveq    #19,d0
0x0000000C 0x6002                   bra.s    *+4                   ; 0x00000010
;
; 2815:    asm{NOP};
;
0x0000000E 0x4E71                   nop
;
; 2816:    }
; 2817:
; 2818:
;
0x00000010 0xB0B900000000           cmp.l    _KEYCNT,d0
0x00000016 0x64F6                   bcc.s    *-8                   ; 0x0000000e
;
; 2819: }
0x00000018 0x4E5E                   unlk     a6
0x0000001A 0x4E75                   rts

why thy are so different ?It is quite annoying>...

Re: problem with optimization

jbezem — Mon, 30 Mar 2009 18:44:24 GMT

Well, they are not so different to me.

First, your loop seems to be an endless loop, since KEYCNT is never incremented after initialization.

So the first compilation without asm(NOP) simply implements this endless loop by a branch instruction jumping onto itself. 'unlk' and 'rts' are not necessary, since this loop will never exit.

And since the compiler doesn't know anything about the inline assembler code, adding the asm(NOP) will need to include this assembler instruction in the final code, then checking the value of KEYCNT once more just in case the assembler code has changed the value of KEYCNT without the C compiler knowing anything about it. It will remain endless, of course...

HTH,

Johan

Re: problem with optimization

ChrisJohns — Tue, 31 Mar 2009 05:52:52 GMT

Is KEYCNT global and updated else where, ie interrupt ?

If so declare it volatile and the compiler should do the right thing.

Re: problem with optimization

doom — Tue, 31 Mar 2009 08:29:37 GMT

Thanks for your reply.In fact the variable KEYCNT is really updated in a periodic interrupt.My project works well now.BUT I did not quite catch you.You mean Variables assigned to registers through compiler optimization may be corrupted during execution? If so ,where? Would you plz explain it some further more?

Thanks

Doom.

Message Edited by doom on 2009-03-31 02:31 AM

Re: problem with optimization

doom — Tue, 31 Mar 2009 08:42:33 GMT

Hi Johan

As a matter of fact, KEYCNT is updated in an interrupt.I take Chris’ advice and my project is working well.Tanks all the same.

regards

doom

Re: problem with optimization

jbezem — Tue, 31 Mar 2009 12:10:23 GMT

No, nothing will get 'corrupted', but a variable that is otherwise not used in a loop may get optimized away. Declaring it 'volatile' tells the compiler that the variable may get updated outside of the scope visible to the compiler, so it needs to re-read that variable every time it wants to use the variable.

In your case, the variable apparently was not declared 'volatile', so the compiler assumed it knew everything there was to know about KEYCNT, and since no modification was visible in the code (without the NOP), the variable was optimized away.

Inserting the NOP, the compiler didn't know at least a small part of the code, so it couldn't optimize as aggressively, even though no 'volatile' was used.

If you will, I wrote an article on 'volatile' (and 'const', but that's not relevant here; 'restrict' is mentioned, but not yet included...), which you may find at http://www.bezem.de/pdf/tqinc.pdf.

HTH,

Johan

Re: problem with optimization

doom — Wed, 01 Apr 2009 07:38:03 GMT

Thanks again!

I have read your article .The last part is a little complicated but it is very helpful.

Regards

doom