Skip to content

Latest commit

 

History

History
35 lines (27 loc) · 2.24 KB

README.md

File metadata and controls

35 lines (27 loc) · 2.24 KB
Raw

BFWorker

run brainfsck code in a worker thread in your browser, see the output in realtime at full speed

this is mostly an expirement in learning a few things:

  • how to run code in a webworker
  • learning a bit of functional programming

runner.js has all the code that transpiles brainfuck code into javascript, and runs it directly in the webworker thread. the text output is buffered so on every newline, a line is sent back to the calling page (bfworker.html in this case) and it decides what to do with that output.

There are no input methods to get input into the worker thread, but the code is there to support it

some interesting points:

runner js has optimization code that does a few things:

  • it condenses repeated +/-/</> instructions into single ops with integer values equal to the number of repeated instructions
  • it tries to optimize balanced loops (aka copy or mul/add loops). loops like [-] become =0 instructions
  • uses a for loop for scans, aka [>] loops
  • after all optimizations are done, a final pass is done to eliminate as much pointer movement as possible, putting all the pointer movement into offsets for instructions. inside while loops, any unbalanced pointer movement is added to the end of the while loop

for some reason, unoptimized code runs faster in my browser than the optimized code does. Im guessing the v8 runtime in chrome is the culprit, it actually does a better job optimizing the raw code than I could do with all the optimization techniques explained above. Thats why theres 2 run buttons. im betting that wouldnt be the case if i could emit proper x86 machine code with the above techniques...something i'll try later ;)

i might try a web assembly version of this as well. grokking that spec is not trivial ;)

Edit: V8 is truly a beast now. It turn out it does 2 passes on js code. First pass does a simple js->native code conversion to get the code started...then it really kicks in and optimizes the bejesus out of it..again in native code. This new code runs as close to true cpu speed as you could get. I'll have to study the new optimizers and see if there is anything I could write that would work better. So far..not happening.