[avr-gcc-list] Question on optimizing behavior of GCC.

[Hudaidai]

Hi,

I have a question on optimizing behavior of GCC(tested with WinAVR-20030312).

Here is the C code:

#include <inttypes.h>

#define BUF_LEN     16

uint8_t aucBuf[BUF_LEN];

uint8_t * pucMsgWr;
uint8_t * volatile pucNewLine;

#define INC_PRN_MSG_P(pmsg) \
{                                   \
    pmsg++;                         \
    if (pmsg>=aucBuf+BUF_LEN)       \
        pmsg=aucBuf;                \
}

void fast_func(void)
{
uint8_t *pucTemp;

    pucTemp=pucMsgWr;
    *pucTemp='\0';
    INC_PRN_MSG_P(pucTemp);
    pucMsgWr=pucTemp;
    pucNewLine=pucTemp;
}

void slow_func(void)
{
    *pucMsgWr='\0';
    INC_PRN_MSG_P(pucMsgWr);
    pucNewLine=pucMsgWr;
}

int main(void)
{
    pucMsgWr=aucBuf;
    
    fast_func();
    slow_func();
}

Let's look at the list file:

000000a4 <fast_func>:
void fast_func(void)
{
uint8_t *pucTemp;

    pucTemp=pucMsgWr;
  a4:   e0 91 72 00     lds     r30, 0x0072
  a8:   f0 91 73 00     lds     r31, 0x0073
    *pucTemp='\0';
  ac:   11 92           st      Z+, r1
    INC_PRN_MSG_P(pucTemp);
  ae:   80 e0           ldi     r24, 0x00       ; 0
  b0:   e2 37           cpi     r30, 0x72       ; 114
  b2:   f8 07           cpc     r31, r24
  b4:   10 f0           brcs    .+4             ; 0xba
  b6:   e2 e6           ldi     r30, 0x62       ; 98
  b8:   f0 e0           ldi     r31, 0x00       ; 0
    pucMsgWr=pucTemp;
  ba:   f0 93 73 00     sts     0x0073, r31
  be:   e0 93 72 00     sts     0x0072, r30
    pucNewLine=pucTemp;
  c2:   f0 93 61 00     sts     0x0061, r31
  c6:   e0 93 60 00     sts     0x0060, r30
}
  ca:   08 95           ret

000000cc <slow_func>:

void slow_func(void)
{
    *pucMsgWr='\0';
  cc:   e0 91 72 00     lds     r30, 0x0072
  d0:   f0 91 73 00     lds     r31, 0x0073             ;*** first time 
accessing "pucMsgWr".
  d4:   10 82           st      Z, r1
    INC_PRN_MSG_P(pucMsgWr);
  d6:   80 91 72 00     lds     r24, 0x0072
  da:   90 91 73 00     lds     r25, 0x0073             ;*** why accessing 
"pucMsgWr" again?
  de:   01 96           adiw    r24, 0x01       ; 1
  e0:   90 93 73 00     sts     0x0073, r25
  e4:   80 93 72 00     sts     0x0072, r24             ;*** waste space(may be 
done after compare).
  e8:   82 57           subi    r24, 0x72       ; 114
  ea:   90 40           sbci    r25, 0x00       ; 0
  ec:   30 f0           brcs    .+12            ; 0xfa
  ee:   82 e6           ldi     r24, 0x62       ; 98
  f0:   90 e0           ldi     r25, 0x00       ; 0
  f2:   90 93 73 00     sts     0x0073, r25
  f6:   80 93 72 00     sts     0x0072, r24             ;*** this storeing is 
really needed.
    pucNewLine=pucMsgWr;
  fa:   80 91 72 00     lds     r24, 0x0072
  fe:   90 91 73 00     lds     r25, 0x0073             ;*** but why loading it 
again?
 102:   90 93 61 00     sts     0x0061, r25
 106:   80 93 60 00     sts     0x0060, r24
}
 10a:   08 95           ret

I am very surprised that "pucMsgWr" was accessed 4 times in the "slow_func". It 
is something like that "pucMsgWr" declared as "uint8_t * volatile". Nothing is 
diffrent when either "-O3" or "-Os" option is used.

Who can tell me why?

hudaidai
2003-04-16