August AGI Interpreter Update
Posted: Mon Aug 26, 2024 10:47 am
This post is regarding my Sierra AGI interpreter for the CommanderX16 on which I have previously posted.
The last big block of work focused on drawing the screen faster, reducing screen algorithm memory requirements, and drawing the priority screen.
I used an algorithm by the Scanline algorithm by CosmicR, which was already quite optimised.
The source is https://github.com/cosmicr/astral_body and viewtopic.php?p=33151&hilit=forgot#p33151
I optimised it further by mainly:
- Rewriting the C parts as assembler
- Using 8 bit rather than 16 bit math, for coordinates (0 - 167 is the range)
- Using VERA auto increment in the loops
There is probably more that can be done, I can't claim to be the world's best optimizer.
You can see the results here.
https://www.youtube.com/watch?v=GHegsdkEFvE
This complicated screen now takes 120 jiffies or 2 seconds.
The old algorithm took around 6 seconds.
The old algorithm required a queue which chewed up 9 whole banks!
The new algorithm queue takes only 256 bytes.
A future update will focus on reducing the file IO overhead at the very beginning. I don't think you can beat memory mapped IO for this job.
The next block of work will focus on:
- Finishing optimising and tidying the algorithm
- Using the new priority screen to:
-Make background object solid so the character can't walk through them
-Adding the 3D effect. Making the character able to walk behind background objects
The last big block of work focused on drawing the screen faster, reducing screen algorithm memory requirements, and drawing the priority screen.
I used an algorithm by the Scanline algorithm by CosmicR, which was already quite optimised.
The source is https://github.com/cosmicr/astral_body and viewtopic.php?p=33151&hilit=forgot#p33151
I optimised it further by mainly:
- Rewriting the C parts as assembler
- Using 8 bit rather than 16 bit math, for coordinates (0 - 167 is the range)
- Using VERA auto increment in the loops
There is probably more that can be done, I can't claim to be the world's best optimizer.
You can see the results here.
https://www.youtube.com/watch?v=GHegsdkEFvE
This complicated screen now takes 120 jiffies or 2 seconds.
The old algorithm took around 6 seconds.
The old algorithm required a queue which chewed up 9 whole banks!
The new algorithm queue takes only 256 bytes.
A future update will focus on reducing the file IO overhead at the very beginning. I don't think you can beat memory mapped IO for this job.
The next block of work will focus on:
- Finishing optimising and tidying the algorithm
- Using the new priority screen to:
-Make background object solid so the character can't walk through them
-Adding the 3D effect. Making the character able to walk behind background objects