Oldskooler Ramblings

the unlikely child born of the home computer wars

Dev journal, day 6

Posted by Trixter on May 6, 2014


Dev journal, day 6

Advertisements

5 Responses to “Dev journal, day 6”

  1. reenigne said

    BIU-waiting-for-EU is 6 cycles per iteration for REP STOSW, not 9. I did some measurements with refresh enabled too – that works out at ~27.4 cycles per iteration total for REP STOSW and 38.4 cycles per iteration total for REP MOVSW with the default refresh period. That’s no so easy to break down into parts though, since the refresh IOs can overlap the BIU-waiting-for-EU cycles and the CGA wait state depends on everything else.

    • Trixter said

      A few hours before you wrote your comment, I had corrected the source to read:

        {
        Maximum cycle counts we cannot exceed based on Andrew Jenner's measurements:
        }
        REPMOVSWcycleCost=37+(1/3); {one iteration of REP MOVSW!}
        REPSTOSWcycleCost=26+(2/3); {one iteration of REP STOSW!}
        totalCyclesPerFrame=76*262*4; {total CPU cycles available in 1/60th sec}
      

      I will fine-tune it further based on your measurements with refresh enabled, thanks!

  2. reenigne said

    Also, why is maxCyclesPerFrame a thousand iterations of REP MOVSW rather than 79648? It seems like as written you’ll only be using about half of the available time in the frame.

    • Trixter said

      It’s intentional:

        totalCyclesPerFrame=76*262*4; {total CPU cycles available in 1/60th sec}
        CyclesPerFrame:real=totalCyclesPerFrame / 2;
        {this is calibrated for 60Hz output -- we know from experience that 60hz
        is not feasible if we exceed this.  Don't exceed this!}
      

      Otherwise, there’s no time left over for hard drive accesses or soundcard DMA.

      Because I’m keeping track of source FPS, and thanks to REP STOS, we will have several opportunities to burn more cycles than this. This is just the conservative value.

      I’ve said too much!

      • Trixter said

        I’ve since refined it to be total cycles per frame, /2, and then an additional 1% taken away so that I can inject the largest possible REP MOVSW including its setup cost.

        This number can get adjusted further if I specifically want to steal more time for framerates significantly below 60Hz, such as the common target of 24. Which is likely.

Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

 
%d bloggers like this: