Thread: Multiplexing bitplanes into bytes

Yes, this should probably have gone in the Game Programming Forum.

Code:

// setup pointers
for (i=0;i<4;i++)
	lpPointers[i] = bmpBitmap->lpPlanes[i];

You could initialize those pointers to those values.

The thing is, it's not terribly fast. Running this on a 2 GHz Pentium 4, CPU usage of up to 98% is observed when bmpBitmap describes a set of 640x480 bitplanes and the code is executed 60 times a second.

Your code looks pretty good. What delay function are you using to keep the frame rate at 60 FPS? Most delay functions that I know of run the CPU at 100%. The DOS delay(), the SDL function, etc all run my processor up, too.

To see how fast the code really is, run it on a slower computer (like mine), or remove the syncronizing code (if you can) and see what the FPS is then.

Exactly. Don't worry about the CPU usage. Try running it on a slower computer an it should be fine.

(What I meant by my first point was this:

Code:

unsigned char *lpPointers[4];

// setup pointers
for (i=0;i<4;i++)
	lpPointers[i] = bmpBitmap->lpPlanes[i];

->

Code:

unsigned char *lpPointers[4] = {bmpBitmap->lpPlanes[0], ...};

which may be a bit faster.)

Code:

>unsigned char *lpPointers[4], ucColour, ucSourceMask = 0x80, ucDestMask = 0xF;
>
>// setup pointers
>for (i=0;i<4;i++)
>	lpPointers[i] = bmpBitmap->lpPlanes[i];

You could change this to:

Code:

unsigned char lpPointers[4], ucColour, ucSourceMask = 0x80, ucDestMask = 0xF;

// setup pointers
for (i=0;i<4;i++)
	lpPointers[i] = *(bmpBitmap->lpPlanes[i]);

And then later on it would just be lpPointers[j] without the dereference.

> lpPointers[j]++;
Actually you're incrementing your pointers here, so my last suggestion is no good.

Ok, here's another try.

Code:

int i, j, iMaskShift = 7, iShift = 4, iSize;
unsigned char *lpPointers[4], ucColour, ucSourceMask = 0x80, ucDestMask = 0xF;

// setup pointers
for (i=0;i<4;i++)
	lpPointers[i] = bmpBitmap->lpPlanes[i];

iSize = bmpBitmap->sWidth * bmpBitmap->sHeight;
for (i=0;i<iSize;i+=8)
{
   for (k=0; k<8; k++)
   {
      
      ucColour = 0;
      for (j=0;j<4;j++)
         ucColour |= ((*lpPointers[j] & ucSourceMask) >> iMaskShift) << (j + iShift);

      *lpBuffer = (*lpBuffer & ucDestMask) | ucColour;
      ucDestMask = ~ucDestMask;
      if (iShift)
         iShift = 0;
      else
      {
         iShift = 4;
         lpBuffer++;
      }

      iMaskShift--;
      ucSourceMask >>= 1;
   }

   iMaskShift = 7;
   ucSourceMask = 0x80;
   for (j=0;j<4;j++)
      lpPointers[j]++;

}

What about expanding the source and destination operands so that it processes a doubleword (four bytes) at a time instead of one, would that help?

That would probably help.

Actually from what I've read your best bet is to check the optimization flags of your compiler, and try out each different one, something you may have already done. Compilers these days are very good at optimizing code.