Unlimited CPU Cycles (LowRes NX)

nathanielbabiak 2022-07-02 04:58 (Edited)

I found a console hack a while ago, thought it might be worthy of some discussion. It's interesting if nothing else! I have no interest in using this, personally it feels kind'a gross.

Two findings:

Strings are only constrained by the emulated clock cycle limit of per-byte operations, so long strings are feasible, on the order of thousands of characters long.
The instruction INSTR(d$,s$[,p]) does not have additional cost of CPU cycles.

Combining these two findings allows for a Turing-complete instruction pipeline in O(1). This means immediate math, rendering, etc is possible!

To use large strings:

initialize large strings by concatenating small ones together (a bunch of spaces works fine)
once the string is large, use a separate string variable to develop a string you want to use for overwriting
Overwrite the large string with the separate string using MID$(s$,p,n)=a$

To create an instruction pipeline for your super-slow code:

reserve one character to use as a header marker separation between pipeline entries
create a character encoding that's capable of storing all possible inputs
create a character encoding that's capable of storing all possible outputs
pre-calculate all possible outputs of all possible inputs
- use large strings to store both encodings to the pipeline (see above)
- track the next position in the pipeline that can take storage
- repeat!

To use your fast instruction pipeline:

within your main loop, take the actual inputs and use your character input encoding
use large strings (see above)
parse from the 'found' position of INSTR to the nearby output encoding
decode it with your character output decoder

Discussion

Unlimited CPU Cycles