Designing an analysis friendly Stockfish?

[BB+]

I am seeing a slow down of ~15%. Is this to be expected?

If that is just NPS, then "yes" is the likely answer --- for instance, turning the hash table completely off will give you more NPS (it does more make/unmake operations as it spends less time in memory lookups), but the time to reach a given depth will go way up.

[gaard]

It would look like this:

Code: Select all

if (tte && (PvNode ? ok_to_use_TT_PV(tte, depth, alpha, beta, ply) : ok_to_use_TT(tte, depth, beta, ply))
 {
    TT.refresh(tte);
    ss->bestMove = ttMove; // Can be MOVE_NONE
    return value_from_tt(tte->value(), ply);
}

I made the changes both in the normal and qsearch parts of search.cpp. I am seeing a slow down of ~15%. Is this to be expected?

For anyone interested,

http://dl.dropbox.com/u/11904592/stockf ... _F_PGO.zip

Should you ever want to switch to the default behavior, there is now a UCI option to do so. PH is on by default. The bench feature uses different positions in this version also, so (bench)marks from this version should not be compared to earlier versions or JA's builds.

[Prima]

Thanks gaad for this Stockfish version. For some reason, your PGO doesn't work in my Q6700 (Win 7, 64-bit). Anyway, I've made Microsoft compiles (Ms VS2010) and a Microsoft PGO compile.

[gaard]

Thanks gaad for this Stockfish version. For some reason, your PGO doesn't work in my Q6700 (Win 7, 64-bit). Anyway, I've made Microsoft compiles (Ms VS2010) and a Microsoft PGO compile.

Was there a specific error you were getting? or was it just failing to start?

[BB+]

PH is on by default.

Is this really "Persistent Hash" (that is, it saves positions to disk, and then loads them back when desired), or is it just a more analysis-friendly hashing scheme?

[gaard]

PH is on by default.

Is this really "Persistent Hash" (that is, it saves positions to disk, and then loads them back when desired), or is it just a more analysis-friendly hashing scheme?

The latter. I guess preserve analysis would be more appropriate. I had forgotten that V. Rajlich had coined PH to mean "saved to disk"

[mcostalba]

It would look like this:

Code: Select all

if (tte && (PvNode ? ok_to_use_TT_PV(tte, depth, alpha, beta, ply) : ok_to_use_TT(tte, depth, beta, ply))
 {
    TT.refresh(tte);
    ss->bestMove = ttMove; // Can be MOVE_NONE
    return value_from_tt(tte->value(), ply);
}

I made the changes both in the normal and qsearch parts of search.cpp. I am seeing a slow down of ~15%. Is this to be expected?

The above code should have zero speed impact.

[Prima]

Thanks gaad for this Stockfish version. For some reason, your PGO doesn't work in my Q6700 (Win 7, 64-bit). Anyway, I've made Microsoft compiles (Ms VS2010) and a Microsoft PGO compile.

Was there a specific error you were getting? or was it just failing to start?

I've attached the error report with your PGO compile in my Core2 Quad Q6700, Win 7 64-bit OS.

EDIT: The file is 269KB and is considered too big to upload in the forum. Here's Rapid share link: http://rapidshare.com/files/445415842/S ... Report.PNG

[gaard]

Thanks gaad for this Stockfish version. For some reason, your PGO doesn't work in my Q6700 (Win 7, 64-bit). Anyway, I've made Microsoft compiles (Ms VS2010) and a Microsoft PGO compile.

Was there a specific error you were getting? or was it just failing to start?

I've attached the error report with your PGO compile in my Core2 Quad Q6700, Win 7 64-bit OS.

EDIT: The file is 269KB and is considered too big to upload in the forum. Here's Rapid share link: http://rapidshare.com/files/445415842/S ... Report.PNG

http://dl.dropbox.com/u/11904592/stockfish_201_PA_G.7z

Does this one fail for you too?

[Prima]

http://dl.dropbox.com/u/11904592/stockfish_201_PA_G.7z

Does this one fail for you too?

gaad, thanks for the build. The "Stockfish 2.0.1 PA G" now works in my Q6700. I was just pointing out the error, in case other Q6700 owners encountered the same problem.

Just out of curiosity, what does "PA_G" mean in the updated Stockfish 2.0.1 build?

Regards,
Prima.