Path: chuka.playstation.co.uk!news
From: raasch@fh-worms.de (Michael Raasch)
Newsgroups: scee.yaroze.programming.codewarrior
Subject: where are the differences?
Date: Tue, 12 Aug 1997 00:55:55 +0200
Organization: Michael Raasch Softwareentwicklung
Lines: 61
Message-ID: <MPG.e59b669f49a97a989684@news.playstation.co.uk>
NNTP-Posting-Host: port11.Fh-Worms.DE
X-Newsreader: Anawave Gravity v1.10.570

Hi there Codewarriors.

these are some lines of my fireroutine and I wonder why Codewarrior 
generates such a different code that the version described as "fastest" 
is 23 scanlines faster than the "faster"-version. As you can see, the 
only difference is the p++ written in an extra line compared to the other 
version where everything is compacted into one expression. 

--- snip ---

#define FIRE_WIDTH 80
#define FIRE_HEIGHT 50
  u_char imgFire[FIRE_WIDTH][FIRE_HEIGHT];
  u_char fadetable[256];
  register u_short i;
  register u_char *p;
  p = &imgFire[1][0];
  for (i = 0; i < FIRE_WIDTH * FIRE_HEIGHT; i++)
  {

// faster
//  *(p - FIRE_WIDTH) = fadetable[(*(p - 1) + *p + *(p + 1) + *(p + 
FIRE_WIDTH)) >> 2];
//  p++;

// fastest
  *(p++ - FIRE_WIDTH) = fadetable[(*(p - 1) + *p + *(p + 1) + *(p + 
FIRE_WIDTH)) >> 2];
}

--- snap ---

I also tried another trick with some more register-variables:

  register u_char left, middle, right;
  p = &imgFire[1][0];
  left=*(p-1);
  middle=*p;
  for (i = 0; i < FIRE_WIDTH * FIRE_HEIGHT; i++)
  {
    right=*(p+1);
    *(p - FIRE_WIDTH) = fadetable[(left+middle+right + *(p + FIRE_WIDTH)) 
>> 2];
    left=middle;
    middle=right;
    p++;
}

now it takes only 2 memoryreads but it takes much longer to execute. Even 
with the register-variables.

So, how many variables can CW fit into the 32 registers?

And now about the D-Cache. Does it make any difference when accessing it 
by BYTE, WORD or LONG?


-- 
hasta la pasta, .\\ichael

"Wenn das John Wayne noch erlebt haette..."