Sabrina profiling

[stegemma]

Here's the last profiling for Sabrina:

Code: Select all

Evaluate: 44.93%
MakeMoves: 25.17%
Executemove: 14.56%
IsInCheck: 1.94%

It is your engine giving similar result?

[cetormenter]

I am assuming you mean generate moves when you say "makeMoves",

In Nirvana,

Evaluate: 35%
Generate moves (includes sorting): 18%
Making and unmaking moves: 6%
Is checking: 5%

The other large time sinks are ttProbes and see, which take 7% and 5% respectively.

[stegemma]

With TT enabled, I'll get a result more similar to your:

Code: Select all

Evaluate: 36.35%
MakeMoves: 21.76%
ExecuteMove: 12.47%
IsInCheck: 1.61%

It seems that my ExecuteMove is slower than your but maybe is just because I haven't any kind of pruning (only "pure" alfabeta).

My IsInCheck requires less time but I've worked a lot to avoid this check-test any time that isn't required.

[cetormenter]

Remember that these numbers are only relative to the total CPU time of your program. If I wrote each method twice as efficiently I would get the same numbers as I do now.

[AndrewGrant]

Code: Select all

Each sample counts as 0.01 seconds.
  %   cumulative   self              self     total           
 time   seconds   seconds    calls   s/call   s/call  name    
 13.16      2.42     2.42 55557110     0.00     0.00  selectNextMove
  8.21      3.93     1.51 33204726     0.00     0.00  applyNormalMove
  7.37      5.29     1.36 36844190     0.00     0.00  evaluateRooks
  5.93      6.38     1.09 54841283     0.00     0.00  squareIsAttacked
  5.93      7.47     1.09 36844190     0.00     0.00  evaluateBishops
  5.38      8.46     0.99 12214495     0.00     0.00  genAllNoisyMoves
  4.49      9.28     0.83 36844190     0.00     0.00  evaluateQueens
  4.32     10.08     0.80      518     0.00     0.03  search
  4.21     10.85     0.78 63779288     0.00     0.00  getHistoryScore
  4.13     11.61     0.76 36844190     0.00     0.00  evaluateKnights
  3.54     12.26     0.65 236173712     0.00     0.00  popcount
  3.34     12.88     0.62 33574689     0.00     0.00  revertMove
  3.05     13.44     0.56 18422095     0.00     0.00  initializeEvalInfo
  2.28     13.86     0.42  6285531     0.00     0.00  qsearch
  1.96     14.22     0.36 36844190     0.00     0.00  evaluatePawns
  1.90     14.57     0.35  6235641     0.00     0.00  getTranspositionEntry
  1.79     14.90     0.33 36844190     0.00     0.00  evaluatePassedPawns
  1.77     15.22     0.33 18427499     0.00     0.00  evaluateBoard
  1.74     15.54     0.32 36844190     0.00     0.00  evaluateKings
  1.69     15.85     0.31 12214495     0.00     0.00  evaluateNoisyMoves
  1.69     16.16     0.31       36     0.01     0.01  clearTranspositionTable
  1.63     16.46     0.30 18422095     0.00     0.00  evaluatePieces
  1.63     16.76     0.30  1288059     0.00     0.00  genAllQuietMoves
  1.52     17.04     0.28 33574689     0.00     0.00  applyMove
  1.31     17.28     0.24 54771801     0.00     0.00  isNotInCheck
  1.31     17.52     0.24  1288059     0.00     0.00  evaluateQuietMoves
  1.03     17.71     0.19 38926968     0.00     0.00  moveIsTactical

Here are the big time users from Ethereal. Our ExecuteMove times are about the same.
squareIsAttacked is about the same as isInCheck, so 6% here.

[jdart]

I haven't done profiling for a while.

It is normal for eval to be the biggest component.

Your generate move/make move numbers look high to me. Do you do incremental make/unmake or copy/make?

--Jon

[stegemma]

I haven't done profiling for a while.

It is normal for eval to be the biggest component.

Your generate move/make move numbers look high to me. Do you do incremental make/unmake or copy/make?

--Jon

Actually I use copy/make but I don't use standard bitboards. I must investigate, because I get a very high nps but a very poor ELO.

[cdani]

Andscacs in big numbers: 42.4% eval, 57.6% the rest.
Some functions:
* make/unmake moves/nulls: 6%
* calc attacks: 4.4%
* see: 2.66%
* save/load hash: 5.3%
* move is check: 2.63%

[Henk]

but I don't use standard bitboards.

The more complexity the more misery

[AndrewGrant]

Still looking forward to seeing the simplicity you have added into Skipper.

Perhaps one day you'll share your knowledge with us.