Mindboards

linusa wrote:There are possible performance improvements with this setbitarr and readbitarr code:

1.) Shift operations are slow in NXC (if anybody is interested, have a look at my NXC profiler and try for yourself, the results should be around here somewhere (or on nxtasy :/) ). You can pre-calc the 8 possible bitmasks:

Code: Select all

  byte mask[0]  = 1;
  byte mask[1]  = 2;
  byte mask[2]  = 4;
  byte mask[3]  = 8;
  byte mask[4]  = 16;
  //...

Code: Select all

  // this:
  byte mask = 1 << (idx % 8);
  // then becomes:
  mask[idx % 8];
  // use in both functions, of course!

You should also pre-calc ~mask to avoid the not-op.

Since array-lookups are (fairly) slow in NXC, you could try using a big SWITCH and see how fast it gets.

2.) In general, compiler like "const" keywords. Sophisticated compilers (gcc, MS VisualC++) will probably recognize many situations where you don't use "const" but don't modify the value, but it can't help to use it as much as possible in NXC, as it's probably not as optimizing as other compilers (just a guess).
Also, inlining the bittarr functions seems reasonable!

3.) Consider splitting the functions to setbitarrtrue, setbitarrfalse, and togglebitarr (with its own binary op). Then you save the if (?-op) and make the function smaller, if you know what you want to set (in Erasthothenes Sieve, you probably always want to set to true/false inside the loop).

4.) Try reducing calls to the bitarr functions and/or caching values...

void setbitarr( unsigned long idx, bool value ){
  byte mask = 1 << (idx % 8);
  mask[idx % 8]; ...? <------------------------? value....?
}

bool readbitarr( unsigned long idx ){
  byte mask = 1 << (idx % 8);
  mask[idx % 8]; 
  return ...? <------------------ ? what?
}