MP Stockfish

[hcyrano]

Thanks, this is a possible patch (only slightly tested)


While there I have also done this one to remove an useless lock

Code: Select all

--- a/src/search.cpp
+++ b/src/search.cpp
@@ -2658,14 +2658,10 @@ namespace {
 
     // We have returned from the idle loop, which means that all threads are
     // finished. Update alpha and bestValue, and return.
-    lock_grab(&MPLock);
-
     *alpha = splitPoint.alpha;
     *bestValue = splitPoint.bestValue;
     masterThread.activeSplitPoints--;
     masterThread.splitPoint = splitPoint.parent;
-
-    lock_release(&MPLock);
   }

i think, masterThread.activeSplitPoints-- make a race condition

remain cautious of these patches

[bob]

Thanks, this is a possible patch (only slightly tested)


Regarding your proposal is not that we don't love, simply it doesn't seems to improve anything with a QUAD, perhas with an OCTAL we could see something better, but we don't have an OCTAL

New macpro can have 24 threads (12 + 12HT)

You _definitely_ do not want to use 24 therads if 1/2 of them are hyper-threading. That will cost way more than it gains. A significant net loss.

[hcyrano]

You _definitely_ do not want to use 24 therads if 1/2 of them are hyper-threading. That will cost way more than it gains. A significant net loss.

yes, but u know the end users

[bob]

You _definitely_ do not want to use 24 therads if 1/2 of them are hyper-threading. That will cost way more than it gains. A significant net loss.

yes, but u know the end users

Those same users can ignore the "do not put foot or hand under mower while it is cranked" also, if they want to.