Crafty-22.9 is available

[bob]

The new crafty seems to be somewhat faster than older versions in terms of NPS.

I saw up to 15M NPS in opening positions.
That's 7.5% of Deep Blue.

run it on a dual-chip I7.

So .. what's the speed on such a box ??

It can reach 40M nodes per second.

[bob]

crafty has started to use c++ features in variable declarations?

iterate.c

if (iteration_depth > 3) {
time_used = ReadClock() - start_time;
int time_left = time_limit - time_used;
int prev_ply = ReadClock() - prev_time;

Aha. That is something Jason changed. Peter Skinner was having problems compiling, probably with a compiler that didn't appreciate this. I will change it and re-copy the source so that no one will have difficulty...

[bob]

The new crafty seems to be somewhat faster than older versions in terms of NPS.

Only by about 1%, at least based on the following (admittedly scant) data:

Code: Select all

Crafty v22.8 (1 cpus)

White(1): bench
Running benchmark. . .
......
Total nodes: 111671982
Raw nodes per second: 2147538
Total elapsed time: 52.21

Crafty v22.9 (1 cpus)

White(1): bench
Running benchmark. . .
......
Total nodes: 114972412
Raw nodes per second: 2169290
Total elapsed time: 53.16

Perhaps Bob Hyatt can quantify the improvement more accurately.

It is very small and is mainly an issue of cache usage. HashProbe() now has a 2x loop rather than two separate blocks of probe code. Same thing happens a couple of places in the "NextMove/NextEvasion" etc procedures...

It is not a big improvement, but it is measurable.

[Denis P. Mendoza]

crafty has started to use c++ features in variable declarations?

iterate.c

if (iteration_depth > 3) {
time_used = ReadClock() - start_time;
int time_left = time_limit - time_used;
int prev_ply = ReadClock() - prev_time;

I thought I was the only one who was faced by this problem during my first trial.

PS. My 64-bit compiles have a problem as reported by werner and still fixing.

[jarkkop]

While at it

I had to add this #ifdef because the definition of pthread was not seen by the compiler and this not needed when CPUS=1, I suppose

utility.c

#if (CPUS > 1)

Code: Select all

pthread_t NumaStartThread(void *func, void *args)
{
  HANDLE hThread;
  ULONGLONG ullMask;

  WinNumaInit();
  if (fSystemIsNUMA) {
    ulNumaNode++;
    if (ulNumaNode > ulNumaNodes)
      ulNumaNode = 0;
    ullMask = ullProcessorMask[ulNumaNode];
    printf("Starting thread on node %d CPU mask %I64d\n", ulNumaNode, ullMask);
    SetThreadAffinityMask(GetCurrentThread(), (DWORD_PTR) ullMask);
    hThread = (HANDLE) _beginthreadex(0, 0, func, args, CREATE_SUSPENDED, 0);
    SetThreadAffinityMask(hThread, (DWORD_PTR) ullMask);
    ResumeThread(hThread);
    SetThreadAffinityMask(GetCurrentThread(), ullProcessorMask[0]);
  } else
    hThread = (HANDLE) _beginthreadex(0, 0, func, args, 0, 0);
  return hThread;
}

#endif

[Jim Ablett]

Windows Crafty 22.9 compiles are now available on my homepage.

These include the 'draw offer' bug fix.

Direct download link:
http://www.mediafire.com/?hz4ytvvn5mv

Jim.

[gerold]

Windows Crafty 22.9 compiles are now available on my homepage.

These include the 'draw offer' bug fix.

Direct download link:
http://www.mediafire.com/?hz4ytvvn5mv

Jim.

Thanks Jim. Working fine in Arena.
Thanks Bob for the new Crafty.

Best,

Gerold.

[bob]

Windows Crafty 22.9 compiles are now available on my homepage.

These include the 'draw offer' bug fix.

What is this about???

Direct download link:
http://www.mediafire.com/?hz4ytvvn5mv

Jim.

[DarioZ]

Playing Crafty 22.9 in an engine tournament with a - 40m/40' -repeating- time setup, in either of Arena 1.1 or 2.0.1, Crafty "thinks" for long periods after the 1st 40 moves and eventauly runs out of time. Anyone seen the same?

[Dirt]

Windows Crafty 22.9 compiles are now available on my homepage.

These include the 'draw offer' bug fix.

What is this about???

See this thread.