Pre-Release McCain X4

[ChiefPushesWood]

Thanks Chief. I wonder if this is a "feature" or a bug in CB.

That's by design... All of my engines are like that and always have been.

Chief

[dbx12]

Dear Mike/all

Here is some Linux feedback from:

Ubuntu 14.04.5 LTS [x64]
gcc version 6.5.0 20181026

about latest McCain-X4

For "original src"

Code: Select all

git clone -b McCain https://github.com/MichaelB7/Stockfish.git McX

compilation fails with:

Code: Select all

In file included from main.cpp:31:0:
tt.h:89:3: error: static assertion failed: Cluster size incorrect
   static_assert(CacheLineSize % sizeof(Cluster) == 0, "Cluster size incorrect");

So it seems that line 89 in tt.h is responsible...
However, for windows compiler [mingw] everything runs smoothly!
Does anyone has a clue?

A temp workaround is to comment out the relevant static_assert line in tt.h
Then everything runs as it should.

By the way why do you revert native variable/flag in Makefile from yes --> no?

dbx

[Dann Corbit]

Probably needs an aligned allocation.
Objects can be different sizes because of the OS and compiler.
For instance, with GCC on a 64 bit Linux, long is 8 bytes but on 64 bit windows it is 4 bytes.
I didn't even look at the code yet so this is a guess.
Strange that an assert was used since it sounds like a performance warning.

[MikeB]

Dear Mike/all

Here is some Linux feedback from:

Ubuntu 14.04.5 LTS [x64]
gcc version 6.5.0 20181026

about latest McCain-X4

For "original src"

Code: Select all

git clone -b McCain https://github.com/MichaelB7/Stockfish.git McX

compilation fails with:

Code: Select all

In file included from main.cpp:31:0:
tt.h:89:3: error: static assertion failed: Cluster size incorrect
   static_assert(CacheLineSize % sizeof(Cluster) == 0, "Cluster size incorrect");

So it seems that line 89 in tt.h is responsible...
However, for windows compiler [mingw] everything runs smoothly!
Does anyone has a clue?

A temp workaround is to comment out the relevant static_assert line in tt.h
Then everything runs as it should.

By the way why do you revert native variable/flag in Makefile from yes --> no?

dbx

Only for those that are compiling a modern version on a bmi computer - but for most people, compiling for their own computer, it is best to set to "yes".

[Dann Corbit]

I am not a fan of the "LC0 style scoring."
Worse, you cannot turn it off when it sees a win.
I chose ce in the UCI options.

Analyzing this position:
[d]8/8/3R3K/P7/8/4ppk1/8/8 w - -

I get this:

Code: Select all

                      
    Avoid move: 
    Best move (McCain-X4-modern-64): Rd6-g6
    Not found in: 01:00
   Found 510 tablebases
     1/2	00:01	         108	10,800	+100.00	Rg6+ Kh3
     2/3	00:01	         317	31,700	+100.00	Rg6+ Kh2 Rg4
     3/4	00:01	         663	66,300	+100.00	Rg6+ Kh2 Rg4 e2
     4/5	00:01	       2,859	259,909	+100.00	Rg6+ Kh2 a6 f2 a7
     5/8	00:01	       4,633	421,181	+100.00	Rg6+ Kh2 a6 f2 a7
     6/16	00:01	       9,076	825,090	+100.00	Rg6+ Kh3 a6 f2 a7 e2 a8Q
     7/10	00:01	      14,279	1,189,916	+100.00	Rg6+ Kf2 a6 Ke1 a7 f2 a8Q
     8/15	00:01	      22,339	1,861,583	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q
     9/19	00:01	      73,597	4,329,235	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe5 Qa2+ Qe2 Qa5 Qe3+ Kh5
    10/34	00:01	     377,602	8,581,863	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh7 Qd3 Kg7 Qd4+ Kf7 Qc4+ Kf8 Qb4+ Ke8 Qc4
    11/23	00:01	     412,283	8,771,978	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh7 Qd3 Kg7 Qd4+ Kf7 Qc4+ Kf8 Qc5+ Ke8 Qc4 Qa5 Qe4+ Kf7 Qc4+ Ke7
    12/41	00:01	     899,697	10,223,829	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qe6 Qb7 Qe8+ Kg4 Qe6+ Rf5 Qg6+ Kf4
    13/43	00:01	   2,715,842	11,706,215	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qc1+ Rg5 Qf4 Kg6 Qd6+ Kh5 Qh2+ Kg4 Qg3+ Kf5 Qh3+ Ke5 Qd7 Qa2+ Ke3 Qb3+ Ke2 Qc4+ Qd3
    14/44	00:01	   3,532,824	11,935,216	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qc1+ Kh7 Qc7+ Rg7 Qc2+ Kh8 Qc3 Qa2+ Kf1 Qe6 f2 Kg8 Qd3 Rg6 Qf3 Qa6+ Ke1
    15/49	00:01	   4,119,658	12,081,108	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qc1+ Kh5 Qh1+ Kg4 Qg1+ Kf5 Qb1+ Kf6 Qb6+ Kg5 Qc5+ Kg4 Qd4+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg3+ Kf5 Qh3+ Ke5 Qd7 Qa5 Ke2 Kf6 Qd6+ Kf7 Qd7+ Kg8 Qe6+ Kf8 Qf6+ Ke8 Qe6+ Kd8 Qd6+ Kc8 Qe6+ Kb8 Qd6+ Kb7
    16/52	00:01	   5,266,836	12,191,750	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qc1+ Kh5 Qh1+ Kg4 Qg1+ Kf5 Qb1+ Kf6 Qb6+ Kg5 Qe3+ Kg4 Qd4+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg3+ Kf5 Qh3+ Ke5 Qd7 Rf5 Qe7+ Kd5 Qd7+ Ke4 Qe6+ Kf4 Qd6+ Kg5 Qg3+ Kf6 Qd6+ Kg7 Qd7+ Kg6 Qd6+ Rf6 Qg3+ Kf7 Qc7+ Ke6 Qc4+ Kd6 Qd4+ Ke7
    17/62	00:01	   5,404,553	12,172,416	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qc1+ Kh5 Qh1+ Kg4 Qg1+ Kf5 Qb1+ Kf6 Qb6+ Kg5 Qc5+ Kg4 Qd4+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg3+ Kf5 Qh3+ Ke5 Qd7 Rf5 Qe7+ Kd5 Qd7+ Ke4 Qe6+ Kf4 Qd6+ Kg5 Qg3+ Kf6 Qh4+ Ke6 Qc4+ Ke7 Qb4+ Ke8 Qb3 Kd7 Qd3+ Rd5 Qh7+ Kc6 Qg6+ Kb5 Qb1+ Kc6
    18/88	00:02	  11,652,977	11,818,435	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg1+ Kf5 Qb1+ Kf6 Qb2+ Kg6 Qb6+ Kh5 Qe6 Qb7 Qh3+ Kg6 Qe6+ Kg7 Ke3 Qc7 f2 Rg3+ Ke2 Qc2+ Kf1 Qd1+ Qe1
    19/60	00:02	  14,203,562	11,965,932	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg3+ Kf5 Qh3+ Ke5 Qd7 Qb8 Qe7+ Kf4 Qf6+ Kg4 Qe6+ Kh5 Qh3+ Kg6 Qe6+ Kg7 Qe7+ Kh6 Qe6+ Rg6 Qh3+ Kg7 Qd7+ Kf6 Qd3 Qg3+ Kf1 Kg5 Qd5+ Kf4 Ke2 Rf6 f2 Qe3+ Kf1 Kg3 Qg2+ Kh4
    20/65	00:02	  17,412,888	11,992,347	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qc1+ Rg5 Qf4 Kg6 Qd6+ Kh5 Qh2+ Kg4 Qg3+ Kf5 Qh3+ Ke5 Qd7 Qb8 Qd3 Ke6 Kf1 Qf4 Qb3+ Kf6 Qb2+ Kf7 Qb3+ Kg7 Qb2+ Kh7 Qc2+ Kh8 f2 Qf3 Qb2+ Kh7
    21/65	00:03	  25,734,065	12,236,835	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg2+ Kf5 Qh3+ Ke5 Qd7 Qb8 Ke2 Kf6 Qd4+ Kf7 f2 Qb5+ Kf3 Rf5+ Kg2 Qc6+ Kg1 Qg6+ Kf1 Qa6+ Ke1 Qb5 Qa7+ Ke6 Qa2+ Kf6 Qa1+ Re5+ Qxe5+ Qxe5+ Kf1 Qe4
    22/73	00:04	  34,825,362	12,168,190	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg2+ Kf5 Qh3+ Ke5 Qd7 Qb8 Qe7+ Kf5 Qf7+ Ke4 Qh7+ Rf5 Qe7+ Re5 Qh7+ Kd4 Qc2 Qb4 Kg2 Rg5+ Kf1 Qb5+ Kf2 Qa5 Qe2 Kd5 Qd1+ Kc6 Qd4 Qa2+ Kf1 Qb1+ Ke2 Rf5 f2 Qc2+ Ke1 Rf8
    23/67	00:04	  39,492,517	12,147,805	+100.00	Rg6+
    23/67	00:04	  39,931,236	12,140,844	+100.00	Rg6+
    23/67	00:04	  42,650,354	12,095,959	+100.00	Rg6+
    23/72	00:04	  44,386,402	12,068,081	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg1+ Kf5 Qb1+ Kf6 Qb6+ Kf7 Qb3+ Kg7 Qe6 Qb8 Qe7+ Kh6 Qe6+ Rg6 Qh3+ Kg7 Qd7+ Kf6 Qd5 Qb6+ Ke2 Ke7 f2 Re6+ Qxe6+ Qxe6+ Kf1 Kd6
    24/76	00:05	  47,075,641	12,076,870	+100.00	Rg6+ Kf2
    24/76	00:05	  47,842,972	12,084,610	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg1+ Kf5 Qb1+ Kf6 Qb6+ Kf7 Qb3+ Kg7 Qe6 Qa5 Qd7+ Kf6 Qd4+ Kg6 Qd6+ Kf5 Qd7+ Ke4 Qc6+ Ke5 Qc4 Rf5 Ke2 Rf4 Qg8 Qb5+ Kf2 Qc5+ Kg2 Rf5 Qg3+ Kf6 f2 Qd5+ Kg1 Qd4
    25/72	00:05	  53,733,548	12,085,818	+100.00	Rg6+
    25/87	00:06	  60,371,729	12,047,840	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg3+ Kf5 Qh3+ Ke5 Qd7 Qa5 Ke2 Rf5 Qe7+ Kd5 Qd7+ Ke4 Qd3+ Kf4 Qd4+ Kg5 Qg1+ Kh5 Qh2+ Kg6 Qd6+ Rf6 Qd3+ Kh5 Qh7+ Kg5 Qg7+ Kf5 f2 Qb5+ Ke3 Qb3+ Ke2 Qc4+ Kf3 Qf1
    26/82	00:07	  72,333,915	12,069,733	+100.00	Rg6+ Kf2
    26/82	00:07	  72,343,891	12,069,384	+100.00	Rg6+ Kf2
    26/82	00:07	  72,586,716	12,061,601	+100.00	Rg6+ Kf2
    26/82	00:07	  76,185,998	12,062,380	+100.00	Rg6+
    26/83	00:08	  81,968,326	12,068,363	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg3+ Kf5 Qh3+ Ke5 Qd7 Qa5 Ke2 Rf5 Qe7+ Kd5 Kf2 Qd2+ Qe2 Qd4+ Kf1 Qe4 Kf2 Re5 Qb5+ Kd6 Qb6+ Ke7 Qh6 Qd4+ Kf1 Rf5 Qh7+ Rf7 Qh5 Rf6 Qh7+ Kd6 Qh2+ Kd5 Qe2 Rh6 f2 Qa1+ Qe1 Rh1+ Ke2 Qe5+ Kd1 Qd4+ Ke2
    27/19	00:08	  82,158,335	12,067,910	+100.00	Rg6+ Kf2
    27/62	00:08	  83,125,339	12,062,884	+100.00	Rg6+ Kf2
    27/83	00:08	  83,880,859	12,053,579	+100.00	Rg6+ Kf2
    27/83	00:08	  86,562,125	12,096,440	+100.00	Rg6+
    27/87	00:10	 110,052,947	12,169,959	+100.00	Rg6+ Kf2
    27/87	00:10	 116,528,128	12,158,611	+100.00	Rg6+ Kf2
    27/87	00:11	 125,497,851	12,102,010	+100.00	Rg6+
    27/88	00:15	 174,658,359	11,986,710	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg2+ Kf5 Qh3+ Ke5 Qd7 Qa5 Ke2 Rf5 Qe7+ Kd5 Qd7+ Ke4 Qe6+ Kf4 Qe3+ Kg4 Qg1+ Kh5 Qh2+ Kg6 Qd6+ Rf6 Qd3+ Kh5 Qh7+ Kg5 Qg7+ Kf5 Qh7+ Ke5 Qh2+ Ke6 f2 Qb5+ Ke1 Qb1+ Ke2 Qe4+ Kd2 Qd5+ Ke2
    28/88	00:15	 175,348,543	11,988,824	+100.00	Rg6+ Kf2
    28/88	00:15	 175,391,192	11,989,281	+100.00	Rg6+ Kf2
    28/88	00:16	 178,521,591	11,992,583	+100.00	Rg6+ Kf2
    28/88	00:16	 180,654,417	12,003,615	+100.00	Rg6+
    28/93	00:16	 186,479,155	12,007,672	+100.00	Rg6+ Kf2
    28/93	00:16	 187,108,956	12,008,019	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg2+ Kf5 Qh3+ Ke5 Qd7 Qa5 Ke2 Rf5 Qe7+ Kd5 Qd7+ Ke4 Qe6+ Kf4 Qe3+ Kg4 Qg1+ Kh5 Qh2+ Kg5 Qg3+ Kf6 Qd6+ Kf7 Qd7+ Kg6 Qd6+ Rf6 Qd3+ Kh6 Qe3+ Kg7 Qe7+ Rf7 Qe4 Kf8 Kf1 Qa6+ Kf2 Rf6 Kg2 Rg6+ Kf2 Qd6 Qe3 Qc6 Qe5 Kf7 Qf5+ Kg8 Ke2 Qa6+ Ke3 Qa7+ Ke2 Qa2+ Ke3 Re6+ Kf4 Qc4+ Kg3 Re7 f2 Rg7+ Kf3 Qc3+ Ke2 Qb2+ Kf3 Qa3+ Kf4 Qc1+ Kf3
    29/79	00:16	 189,238,747	12,015,921	+100.00	Rg6+ Kf2
    29/79	00:17	 191,438,354	12,026,533	+100.00	Rg6+
    29/93	00:17	 199,179,700	12,043,759	+100.00	Rg6+
    29/93	00:19	 215,136,991	12,083,632	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg2+ Kf5 Qh3+ Ke5 Qd7 Qb8 Ke2 Rf5 Qd2 Qb5+ Kf2 Qb7 Qd3 Qe4 Qe3 Rf7 Qg5+ Ke6 Qg2 Kf5 Qh3+ Kg6 Qg3+ Kf6 Qg8 Qd4+ Kf1 Qd3+ Kg2 Qd5 Kf2 Qe4 Qh8+ Kg6 Qg8+ Rg7 Qb8 Qc2+ Kf1 Qc1+ Kf2 Qc5+ Ke2 Kf5 Qb1+ Kg5 Qb3 Re7+ Kf1 Kf4 Qg8 Kxf3 Qg2+ Kf4 Qh2+ Kg5 Qd2+ Kg6
    30/95	00:20	 228,999,689	12,113,821	+100.00	Rg6+ Kf2
    30/95	00:20	 229,030,398	12,114,164	+100.00	Rg6+ Kf2
    30/95	00:20	 231,473,311	12,113,313	+100.00	Rg6+ Kf2
    30/95	00:20	 232,713,866	12,111,052	+100.00	Rg6+
    30/95	00:20	 233,061,949	12,110,889	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg2+ Kf5 Qh3+ Ke5 Qd7 Qa5 Ke2 Rf5 Qe7+ Kd5 Qd7+ Ke4 Qe6+ Kf4 Qe3+ Kg4 Qg1+ Kh5 Qh2+ Kg6 Qd6+ Rf6 Qd3+ Kg5 Qe3+ Kh5 Qe8+ Kh4 Qh8+ Kg5 Qg8+ Rg6 Qc4 Qe5+ Kf2 Kh5 Qf7 Qb2+ Ke1 Qc3+ Ke2 Qc2+ Ke1 Kg4 f2 Qb1+ Ke2 Qb5+ Ke1
    31/73	00:20	 235,048,436	12,119,647	+100.00	Rg6+ Kf2
    31/73	00:20	 236,334,811	12,120,977	+100.00	Rg6+ Kf2
    31/73	00:20	 236,805,446	12,120,250	+100.00	Rg6+ Kf2
    31/73	00:21	 241,901,096	12,138,145	+100.00	Rg6+
    31/73	00:22	 252,613,303	12,141,951	+100.00	Rg6+
    31/98	00:22	 261,098,261	12,159,375	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg1+ Kf5 Qb1+ Kf6 Qb6+ Kf7 Qc7+ Ke8 Qf4 Qa2+ Kf1 Qa1+ Kf2 Qb2+ Kf1 Qb5+ Kf2 Rf5 Qe3+ Kf7 Qe4 Qb2+ Kf1 Qa1+ Kf2 Qf6 Qd3 Qb2+ Kf1 Qc1+ Kf2 Qc5+ Qe3 Qd5 Kg3 Rg5+ Kf2 Kg7 Qf4 Kh6 Ke2 Kh5 Qh2+ Kg6 Qf4 Qb5+ Kf2 Qb2+ Kf1 Kh5 Qf7+ Kg4 Qe6+ Kf4 f2 Qd4 Qe7
    32/97	00:23	 270,958,383	12,175,715	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg1+ Kf5 Qb1+ Kf6 Qb6+ Kf7 Qc7+ Ke8 Qf4 Qa2+ Kf1 Qa1+ Kf2 Qb2+ Kf1 Qb5+ Kf2 Rf5 Qe3+ Kf7 Qa7+ Kg6 Qd4 Kh5 Kg2 Rg5+ Kf2 Qf5 Ke3 Qe6+ Kf2 Qa2+ Kf1 Qa6+ Kf2 Qb5 Ke3 Rf5 Kf2 Qb8 Qg7 Qh2+ Qg2 Qh4+ Kf1 Qc4+ Kg1 Qc1+ Kf2 Qd2+ Kf1 Qd1+ Kf2 Qd5 Qg3 Qe4 Qh2+ Kg5 Qg3+ Qg4 Qg2 Kf4 Qh2+ Ke4 Qh7 Qxf3+ Ke1 Qf1+ Kd2
    33/103	00:23	 277,000,884	12,189,257	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg1+ Kf5 Qb1+ Kf6 Qb6+ Kf7 Qc7+ Ke8 Qf4 Qa2+ Kf1 Qa1+ Kf2 Qb2+ Kf1 Qb5+ Kf2 Rf5 Qe3+ Kf7 Qd4 Qc6 Qe3 Qd5 Kg3 Rg5+ Kf2 Re5 Qf4+ Rf5 Qe3 Kg6 Kg3 Rg5+ Kf2 Kh5 Qf4 Qc5+ Qe3 Qc2+ Qe2 Qc7 Qe8+ Rg6 Ke3 Qc1+ Kf2 Qd2+ Kf1 Qd3+ Kf2 Qd5 Qh8+ Kg5 Qc3 Kh6 Ke2 Re6+ Kf1 Kg5 Kf2 Re5 Qc1+ Kg6 Qf4 Rh5 Qg4+ Rg5 Qf4 Qa2+ Kf1 Kh5 Qxg5+ Kxg5 f2
    34/70	00:26	 314,813,377	12,263,863	+100.00	Rg6+
    34/99	00:28	 328,819,977	12,263,910	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg1+ Kf5 Qb1+ Kf6 Qb6+ Kf7 Qb3+ Kg7 Qc3+ Kg6 Qd3+ Rf5 Qd6+ Rf6 Qg3+ Kf7 Qc7+ Ke6 Qg3 Qa2+ Ke3 Qb3+ Ke2 Qc2+ Ke1 Kf7 Kf1 Qc4+ Kg1 Qc1+ Kf2 Rf5 Qd6 Qb2+ Kg3 Qb7 Qd3 Qc7+ Kh3 Qf4 Qb3+ Kg7 Qc3+ Kg6 Kg2 Qg4+ Kf2 Qh4+ Kg2 Rg5+ Kf1 Qh3+ Ke2 Qh2+ f2 Re5+ Qxe5 Qxe5+ Kf3
    35/108	00:28	 339,126,429	12,260,092	+100.00	Rg6+ Kf2
    35/108	00:29	 345,308,354	12,258,017	+100.00	Rg6+ Kf2
    35/108	00:30	 353,123,826	12,260,818	+100.00	Rg6+
    35/108	00:35	 425,866,054	12,350,029	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg1+ Kf5 Qb1+ Kf6 Qb6+ Kf7 Qb3+ Kg7 Qc3+ Kg6 Qd3+ Rf5 Qd6+ Rf6 Qg3+ Kf7 Qc7+ Ke6 Qc4+ Kf5 Qc5+ Kg6 Qe3 Qd5 Kg3 Kh5 Qe8+ Rg6+ Kf2 Kg5 Qe1 Rf6 Qc3 Kg6 Kg3 Rf5 Kg2 Rg5+ Kf2 Kh5 Qe1 Rf5 Qc3 Kh4 Kg2 Kg4 Qg7+ Rg5 Qf6 Qe4 Qc3 Kf4+ Kf2 Re5 Qc7 Qxf3+ Kg1 Qg3+ Kh1
    36/108	00:40	 481,400,044	12,393,801	+100.00	Rg6+
    36/108	00:40	 484,432,781	12,392,754	+100.00	Rg6+ Kf2 a6 e2 a7 e1Q a8Q Qe3+ Kh5 Qe5+ Rg5 Qh2+ Kg4 Qg1+ Kf5 Qb1+ Kf6 Qb6+ Kf7 Qc7+ Ke8 Qf4 Qd5 Ke2 Qb5+ Kf2 Rf5 Qd4 Kf7 Qa7+ Kg6 Qd4 Kh5 Kg2 Rg5+ Kf2 Qb8 Kf1 Qb1+ Ke2 Qa2+ Kf1 Qa6+ Ke1 Qe6+ Kf1 Qc6 Kf2 Qa8 Qf4 Qa2+ Ke3 Qb3+ Ke2 Qb5+ Kf2 Qc5+ Qe3 Qd6 Qe8+ Kg4 Qe4+ Qf4 Qe6+ Rf5 Qg6+ Qg5 Qe6 Qd2+ Kf1 Qd1+ Kf2 Qxf3+ Ke1 Qf1+ Kd2 Qf4+ Kd1
    37/83	00:44	 544,930,065	12,452,128	+100.00	Rg6+
    37/87	00:50	 611,056,801	12,507,815	+100.00	Rg6+
    37/103	00:55	 685,159,377	12,541,585	+100.00	Rg6+
   2019-06-13 5:12:36 PM, Time for this analysis: 00:01:00, Rated time: 05:00

Because of that, I can't use it for my analysis.

[Andrew]

Thanks for the updates! I posted this position in another thread recently which a recent version of McCain
couldn't get in 40 minutes

k1q4r/p7/P3P3/1N6/8/nQ2K3/6p1/2n5 w - - bm Qd5+; id "Arves.16248"; dm 34;

This version gets it in seconds (an actual # mate score) well done if it was from your changes or the Stockfish changes!

Andrew

[mar]

Probably needs an aligned allocation.
Objects can be different sizes because of the OS and compiler.
For instance, with GCC on a 64 bit Linux, long is 8 bytes but on 64 bit windows it is 4 bytes.
I didn't even look at the code yet so this is a guess.
Strange that an assert was used since it sounds like a performance warning.

The assert simply checks whether cacheline size is divisible by the size of a struct.
I don't see how this could happen unless sizeof(char)>1, which seems extremely unlikely especially on x64 Ubuntu.
Otherwise there's no way a sane compiler would mis-pack the struct.
The size should be 32, maybe the OP could tell us what the size actually was and whether padding with uint8_t would help.
Something seems broken and my bet is it's not the SF code.

[Joerg Oster]

Probably needs an aligned allocation.
Objects can be different sizes because of the OS and compiler.
For instance, with GCC on a 64 bit Linux, long is 8 bytes but on 64 bit windows it is 4 bytes.
I didn't even look at the code yet so this is a guess.
Strange that an assert was used since it sounds like a performance warning.

The assert simply checks whether cacheline size is divisible by the size of a struct.
I don't see how this could happen unless sizeof(char)>1, which seems extremely unlikely especially on x64 Ubuntu.
Otherwise there's no way a sane compiler would mis-pack the struct.
The size should be 32, maybe the OP could tell us what the size actually was and whether padding with uint8_t would help.
Something seems broken and my bet is it's not the SF code.

I already hinted to a possible cause in one of the former McCain threads: viewtopic.php?f=2&t=69849

From what I have googled, e. g.

Code: Select all

uint_fast8_t

must NOT necessarily be 8 bits, it can be larger ...
See https://stackoverflow.com/questions/350 ... t-least8-t
Why did you change this in tt.h?

[mar]

I already hinted to a possible cause in one of the former McCain threads: viewtopic.php?f=2&t=69849

From what I have googled, e. g.

Code: Select all

uint_fast8_t

must NOT necessarily be 8 bits, it can be larger ...
See https://stackoverflow.com/questions/350 ... t-least8-t
Why did you change this in tt.h?

Ouch, yes now I see it in the branch
You're right, it's a change that not only breaks alignment but also increases the overall memory usage.

(u)int_fast16_t is actually 8 bytes in 64-bit mode when using gcc or clang, so this "clever" change made the TT 4 times bigger

(some people really should not touch things they don't understand):D

[MikeB]

Thanks for the updates! I posted this position in another thread recently which a recent version of McCain
couldn't get in 40 minutes

k1q4r/p7/P3P3/1N6/8/nQ2K3/6p1/2n5 w - - bm Qd5+; id "Arves.16248"; dm 34;

This version gets it in seconds (an actual # mate score) well done if it was from your changes or the Stockfish changes!

Andrew

I believe, not 100% sure, this change can be directly attributed to the shuffle extension patch in SF. At least it appeared that way in some quick testing I performed.