Stockfish: Slowdown respect to Official Stockfish
Multi Variant Stockfish shows a 15% slowdown respect to Official Stockfish.
Is it this expected?
native build with MSYS2
make profile-build ARCH=x86-64-modern COMP=mingw -j 8
base = official stockfish (bench=5803228)
test = multi-v stockfish (bench=5803228)
run base test diff
1 2209070 1879892 -329178
2 2214972 1885389 -329583
3 2220906 1892768 -328138
4 2234589 1897099 -337490
5 2225164 1894622 -330542
6 2212439 1884164 -328275
7 2226017 1893385 -332632
8 2221756 1894003 -327753
9 2235449 1897099 -338350
10 2227726 1895241 -332485
base = 2222808 +/- 5765
test = 1891366 +/- 3917
diff = -331442 +/- 2490
speedup = -0.149110
ppigazzini
❤1
All 32 comments
It's been that way for a while. You can reduce the slowdown by just compiling for the variant version that interests you. Not sure if is possible to fix. The latest version of SugaR has the opposite dynamic , it's about 5% faster than official SF
Sent from my iPhone
On Mar 13, 2017, at 2:15 PM, ppigazzini notifications@github.com wrote:
Multi Variant Stockfish shows a 15% slowdown respect to Official Stockfish.
Is it this expected?native build with MSYS2
make profile-build ARCH=x86-64-modern COM=mingw -j 8base = official stockfish (bench=5803228)
test = multi-v stockfish (bench=5803228)run base test diff
1 2209070 1879892 -329178
2 2214972 1885389 -329583
3 2220906 1892768 -328138
4 2234589 1897099 -337490
5 2225164 1894622 -330542
6 2212439 1884164 -328275
7 2226017 1893385 -332632
8 2221756 1894003 -327753
9 2235449 1897099 -338350
10 2227726 1895241 -332485base = 2222808 +/- 5765
test = 1891366 +/- 3917
diff = -331442 +/- 2490
speedup = -0.149110—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub, or mute the thread.
MichaelB7
on 13 Mar 2017
Yes, this is expected and known. It is difficult to add the functionality for variants without introducing slowdowns for standard chess.
ianfab
on 13 Mar 2017
Following the Michael's suggestion, I played with the Makefile flags in 3.2.1 Debugging section.
Slowdown wrt flags kept:
15% All flags
7% -DANTI
4% -DATOMIC
4% -DCRAZYHOUSE
1% No flags
So in fishtest should be possible to have builds for the variant tested, gaining 10% of CPU power.
ppigazzini
on 14 Mar 2017
👍2
Thanks for the statistics @ppigazzini . I see (bench, all flags):
Total time (ms) : 8740
Nodes searched : 5803228
Nodes/second : 663984
versus without -DANTI:
Total time (ms) : 8263
Nodes searched : 5803228
Nodes/second : 702314
ddugovic
on 14 Mar 2017
@ddugovic if I'm not wrong without -DANTI you should get a build working for all versions except antichess: more interesting a build with only -DANTI (as my statistics)
here the bash script to bench two builds of stockfish (mind the ./ before the build name)
[EDITED: check below the script debugged by @ianfab ]
ppigazzini
on 14 Mar 2017
My main concern at the moment is what, if anything, can be done for Lichess given that 1% of players play antichess and the other 99% may appreciate a speedup. So I find my test more interesting; it may be possible to speed up Stockfish while leaving antichess intact.
(Currently I am looking at movegen.cpp to see if template functions improve performance.)
Here are my results with that script:
run 1 / 10
run 2 / 10
run 3 / 10
run 4 / 10
run 5 / 10
run 6 / 10
run 7 / 10
run 8 / 10
run 9 / 10
run 10 / 10
run base test diff
1 660734 0 -660734
2 624540 0 -624540
3 670350 0 -670350
4 675265 0 -675265
5 675579 0 -675579
6 664669 0 -664669
7 627511 0 -627511
8 614423 0 -614423
9 622931 0 -622931
10 635552 0 -635552
base = 647155 +/- 15880
test = 0 +/- 0
diff = -647155 +/- 15880
speedup = -1.000000
So in fishtest should be possible to have builds for the variant tested, gaining 10% of CPU power.
10% slow down means around -12 Elo points.
For an engines at top humans level, -12 Elo is not a big thing.
Vinvin20
on 14 Mar 2017
@Vinvin20 there are only 3 active contributors of CPU: ddugovic, ianfab and me. So a 10% of speedup in fishtest, IMO, is useful.
@ddugovic the third column is 0, I think that you have forgotten the ./ before the second build :)
$ bash bench-parallel.sh ./stockfish-base.exe ./stockfish-test.exe 10
base = all flags
test = without -DANTI -DSUICIDE
run base test diff
1 1894622 1964532 +69910
2 1890918 1965863 +74945
3 1899583 1968530 +68947
4 1902074 1972545 +70471
5 1889071 1959226 +70155
6 1894622 1970535 +75913
7 1904571 1969866 +65295
8 1890302 1961212 +70910
9 1901450 1972545 +71095
10 1899583 1972545 +72962
base = 1896679 +/- 3586
test = 1967739 +/- 3164
diff = 71060 +/- 1977
speedup = 0.037466
I had similar strange results like @ddugovic. I think that this is due to overwriting of the temporary files with pipes from the same files, which can, as far as I know, cause undefined behaviour. A slightly modified version of the script works for me.
ianfab
on 14 Mar 2017
@ianfab thank you :)
ppigazzini
on 14 Mar 2017
The time control for fishtest is fixed, though, no? So it would use the alloted time more efficiently, but not finish faster. In that case I am in favor of testing with all variants enabled, because that's the version most people will use.
@ianfab Is fishtest power a bottleneck for you currently? I think I can contribute a couple of instances if so (and if I can figure out how to set them up).
niklasf
on 14 Mar 2017
👍1
@niklasf Currently, the bottleneck is the creation of patches and tests, especially of those that are not simply parameter tweaks.
ianfab
on 14 Mar 2017
@niklasf the time control is scaled according to the bench time of the worker. So in this moment each test use 10% more time than strictly necessary.
@ianfab the project needs also more workers 1. the worker pool must have more variability to avoid ovefitting 2. the regression tests are performed with only 1000 games, the error bar is high and the gain of that latest patches is masked 3. a developer wants the test result asap :)
ppigazzini
on 14 Mar 2017
@ppigazzini Sure, it is nice to have more workers and I very much appreciate any contribution of CPU time, but the machines need meaningful tests to work on. Many of my ideas have already been tested (or at least certain implementations of them), and I currently have less time to write patches, so any contribution of patches and tests is very welcome.
ianfab
on 14 Mar 2017
👍1
Well... it seems 2c5969d1a04e9e41fb1ab8a10190854ebd769d46 and 1e57268132032a1ec4549ba8aeadd78cc858f9e1 are a 1.2% speedup on my machine in 40 iterations, and ~at least for standard chess~ the benchmark numbers are the same... test submitted.
ddugovic
on 15 Mar 2017
Apparently I don't know how to properly submit the test for branch versus master with speedup bounds (no functional change).
ddugovic
on 15 Mar 2017
@ddugovic I think that for patches with no functional change we only need to verify the speed-up on a few machines. A test would probably fail anyway if the speed-up is smaller than 4-5%. On my machine the speed-up is smaller, but still there:
diff = 5596 +/- 3375
speedup = 0.003547
Regarding the test submission: I think this problem with one branch tested against itself occurs when you select the test type "regression", so please use "standard" for any kind of test.
It would be nice to have the possibility of doing speed measurements on fishtest, but I think it will not be added upstream soon and I do not know whether or when I will be able to add it myself.
ianfab
on 15 Mar 2017
👍1
Thanks, that makes sense and all my tests with this change have been positive.
I suppose next I'll use callgrind or similar to look for other room for optimization.
ddugovic
on 15 Mar 2017
👍1
Callgrind annotation without/with -DANTI (excerpt):
------------------------------------------------------------- -------------------------------------------------------------
Profile data file 'callgrind.out.18478' (creator: callgrind-3 | Profile data file 'callgrind.out.18832' (creator: callgrind-3
------------------------------------------------------------- -------------------------------------------------------------
I1 cache: I1 cache:
D1 cache: D1 cache:
LL cache: LL cache:
Timerange: Basic block 0 - 3067735945 | Timerange: Basic block 0 - 3325791624
Trigger: Program termination Trigger: Program termination
Profiled target: ./stockfish_base bench (PID 18478, part 1) | Profiled target: ./stockfish bench (PID 18832, part 1)
Events recorded: Ir Events recorded: Ir
Events shown: Ir Events shown: Ir
Event sort order: Ir Event sort order: Ir
Thresholds: 99 Thresholds: 99
Include dirs: Include dirs:
User annotated: User annotated:
Auto-annotation: off Auto-annotation: off
------------------------------------------------------------- -------------------------------------------------------------
Ir Ir
------------------------------------------------------------- -------------------------------------------------------------
19,101,514,352 PROGRAM TOTALS | 20,222,342,852 PROGRAM TOTALS
------------------------------------------------------------- -------------------------------------------------------------
Ir file:function Ir file:function
------------------------------------------------------------- -------------------------------------------------------------
1,823,907,316 search.cpp:Value (anonymous namespace)::search | 2,047,873,044 search.cpp:Value (anonymous namespace)::search
1,467,929,891 movepick.cpp:MovePicker::next_move() [/home/li | 1,502,438,085 movepick.cpp:MovePicker::next_move() [/home/li
1,345,721,132 evaluate.cpp:Value Eval::evaluate<false>(Posit | 1,405,249,353 evaluate.cpp:Value Eval::evaluate<false>(Posit
1,028,670,253 evaluate.cpp:Score (anonymous namespace)::eval | 1,099,081,913 evaluate.cpp:Score (anonymous namespace)::eval
819,019,917 position.cpp:Position::see_ge(Move, Value) con | 974,217,760 position.cpp:Position::see_ge(Move, Value) con
677,130,431 position.cpp:Position::do_move(Move, StateInfo | 708,334,202 position.cpp:Position::do_move(Move, StateInfo
400,707,071 types.h:Value Eval::evaluate<false>(Position c | 404,208,276 types.h:Value (anonymous namespace)::search<((
388,975,638 types.h:Value (anonymous namespace)::search<(( | 388,129,706 types.h:Value Eval::evaluate<false>(Position c
362,780,583 position.cpp:Position::set_check_info(StateInf | 332,939,118 types.h:MovePicker::next_move()
273,765,607 position.h:Score (anonymous namespace)::evalua | 318,033,272 position.h:Score (anonymous namespace)::evalua
268,173,002 types.h:MovePicker::next_move() | 269,559,168 position.cpp:Position::set_check_info(StateInf
238,721,072 pawns.cpp:Pawns::probe(Position const&) [/home | 238,721,072 pawns.cpp:Pawns::probe(Position const&) [/home
231,781,071 bitboard.h:Score (anonymous namespace)::evalua | 237,370,744 bitboard.h:Score (anonymous namespace)::evalua
229,485,444 movepick.h:MovePicker::next_move() | 230,445,283 position.cpp:Position::legal(Move) const [/hom
224,727,989 types.h:Score (anonymous namespace)::evaluate_ | 226,069,023 types.h:Score (anonymous namespace)::evaluate_
222,662,625 tt.cpp:TranspositionTable::probe(unsigned long 222,662,625 tt.cpp:TranspositionTable::probe(unsigned long
221,360,877 types.h:Position::see_ge(Move, Value) const | 218,423,148 position.cpp:Position::slider_blockers(unsigne
211,631,064 types.h:Position::do_move(Move, StateInfo&, bo | 211,876,496 types.h:Position::do_move(Move, StateInfo&, bo
210,642,044 position.cpp:Position::undo_move(Move) [/home/ | 211,862,059 movepick.h:MovePicker::next_move()
202,386,454 position.cpp:Position::legal(Move) const [/hom | 210,773,064 position.cpp:Position::undo_move(Move) [/home/
199,485,927 search.cpp:Value (anonymous namespace)::qsearc | 209,014,768 search.cpp:Value (anonymous namespace)::qsearc
191,653,835 position.h:Position::is_variant_end() const [/ | 200,704,071 types.h:Position::see_ge(Move, Value) const
179,179,003 types.h:Position::moved_piece(Move) const [/ho | 191,653,835 position.h:Position::is_variant_end() const [/
167,353,122 position.h:Position::set_check_info(StateInfo* | 179,179,003 types.h:Position::moved_piece(Move) const [/ho
165,487,973 bitboard.h:Value Eval::evaluate<false>(Positio | 164,403,431 position.h:Position::capture_or_promotion(Move
164,403,431 position.h:Position::capture_or_promotion(Move | 160,712,147 position.cpp:Position::is_draw(int) const [/ho
160,712,147 position.cpp:Position::is_draw(int) const [/ho | 159,870,452 bitboard.h:Value Eval::evaluate<false>(Positio
155,776,287 types.h:Position::advanced_pawn_push(Move) con | 155,776,540 types.h:Position::advanced_pawn_push(Move) con
142,735,866 position.cpp:Position::attackers_to(Square, un | 146,522,445 movepick.cpp:void MovePicker::score<(GenType)0
> 142,527,868 position.cpp:Position::attackers_to(Square, un
142,409,336 /usr/include/c++/6/bits/stl_algo.h:MovePicker: 142,409,336 /usr/include/c++/6/bits/stl_algo.h:MovePicker:
136,230,679 search.cpp:Value (anonymous namespace)::qsearc | 141,976,445 search.cpp:Value (anonymous namespace)::qsearc
> 139,408,742 types.h:Position::undo_move(Move)
135,630,568 position.h:Position::undo_move(Move) 135,630,568 position.h:Position::undo_move(Move)
134,186,779 movepick.cpp:void MovePicker::score<(GenType)0 | 133,134,246 position.h:Position::do_move(Move, StateInfo&,
133,921,248 position.h:Position::do_move(Move, StateInfo&, <
130,625,208 position.cpp:Position::key_after(Move) const 130,625,208 position.cpp:Position::key_after(Move) const
125,698,385 types.h:Position::undo_move(Move) | 121,028,040 evaluate.cpp:void (anonymous namespace)::eval_
118,647,825 position.h:Position::advanced_pawn_push(Move) | 118,647,825 position.h:Position::advanced_pawn_push(Move)
115,494,897 pawns.cpp:Value Pawns::Entry::shelter_storm<(C | 118,146,420 evaluate.cpp:void (anonymous namespace)::eval_
> 115,494,897 pawns.cpp:Value Pawns::Entry::shelter_storm<(C
> 111,850,866 position.h:Square Position::square<(PieceType)
> 111,738,026 position.h:Square Position::square<(PieceType)
107,499,981 position.h:Position::moved_piece(Move) const 107,499,981 position.h:Position::moved_piece(Move) const
103,391,548 bitboard.h:Position::see_ge(Move, Value) const | 102,967,189 evaluate.cpp:(anonymous namespace)::evaluate_s
102,967,189 evaluate.cpp:(anonymous namespace)::evaluate_s | 101,805,620 position.h:Position::attackers_to(Square, unsi
101,954,190 position.h:Position::attackers_to(Square, unsi | 96,562,406 pawns.cpp:Value Pawns::Entry::shelter_storm<(C
98,215,047 types.h:Position::legal(Move) const | 96,308,567 types.h:Position::legal(Move) const
96,562,406 pawns.cpp:Value Pawns::Entry::shelter_storm<(C | 94,505,700 position.h:Value (anonymous namespace)::search
94,505,700 position.h:Value (anonymous namespace)::search | 94,150,528 search.cpp:Value (anonymous namespace)::search
90,787,890 search.cpp:Value (anonymous namespace)::search | 94,092,828 position.cpp:Position::pseudo_legal(Move) cons
90,134,132 position.cpp:Position::pseudo_legal(Move) cons | 93,655,034 search.cpp:Value (anonymous namespace)::search
87,982,702 position.h:Value Eval::evaluate<false>(Positio | 90,999,762 position.h:Square Position::square<(PieceType)
85,864,665 evaluate.cpp:(anonymous namespace)::evaluate_i | 88,746,285 evaluate.cpp:(anonymous namespace)::evaluate_i
85,763,351 search.cpp:Value (anonymous namespace)::search | 88,102,235 bitboard.h:Position::see_ge(Move, Value) const
81,771,687 movegen.cpp:ExtMove* generate<(GenType)0>(Posi | 87,828,849 movegen.cpp:ExtMove* generate<(GenType)0>(Posi
> 84,611,315 position.h:Value Eval::evaluate<false>(Positio
81,245,192 /usr/include/x86_64-linux-gnu/bits/string3.h:P 81,245,192 /usr/include/x86_64-linux-gnu/bits/string3.h:P
77,967,112 tt.h:Value (anonymous namespace)::search<((ano | 77,967,133 tt.h:Value (anonymous namespace)::search<((ano
73,788,450 material.cpp:Material::probe(Position const&) | 74,130,514 material.cpp:Material::probe(Position const&)
73,206,236 evaluate.cpp:void (anonymous namespace)::eval_ | 73,183,733 types.h:Position::capture_or_promotion(Move) c
73,179,804 types.h:Position::capture_or_promotion(Move) c <
73,127,178 position.h:Square Position::square<(PieceType) <
70,577,768 bitbase.cpp:main 70,577,768 bitbase.cpp:main
68,082,612 types.h:Position::key_after(Move) const [/home | 68,082,612 types.h:Position::key_after(Move) const [/home
67,508,258 evaluate.cpp:void (anonymous namespace)::eval_ | 66,150,829 pawns.h:Value Eval::evaluate<false>(Position c
67,477,093 pawns.h:Value Eval::evaluate<false>(Position c | 65,820,816 types.h:(anonymous namespace)::update_cm_stats
66,320,583 bitboard.h:Position::set_check_info(StateInfo* | 63,587,304 movegen.cpp:ExtMove* (anonymous namespace)::ge
65,820,120 types.h:(anonymous namespace)::update_cm_stats <
61,983,376 movegen.cpp:ExtMove* (anonymous namespace)::ge <
61,678,330 types.h:void MovePicker::score<(GenType)0>() 61,678,330 types.h:void MovePicker::score<(GenType)0>()
56,062,813 pawns.h:Score (anonymous namespace)::evaluate_ | 56,982,187 movegen.cpp:ExtMove* generate<(GenType)1>(Posi
55,635,734 movegen.cpp:ExtMove* generate<(GenType)1>(Posi | 56,062,813 pawns.h:Score (anonymous namespace)::evaluate_
52,538,111 pawns.cpp:Score Pawns::Entry::do_king_safety<( | 53,854,561 movegen.cpp:ExtMove* (anonymous namespace)::ge
52,429,908 movegen.cpp:ExtMove* (anonymous namespace)::ge | 52,538,111 pawns.cpp:Score Pawns::Entry::do_king_safety<(
52,282,384 /usr/include/c++/6/bits/move.h:MovePicker::nex 52,282,384 /usr/include/c++/6/bits/move.h:MovePicker::nex
50,020,384 search.cpp:Value (anonymous namespace)::qsearc | 51,753,377 position.h:Position::see_ge(Move, Value) const
> 50,952,588 position.h:Position::slider_blockers(unsigned
> 50,510,031 search.cpp:Value (anonymous namespace)::qsearc
> 50,020,384 search.cpp:Value (anonymous namespace)::qsearc
49,519,542 bitboard.h:Pawns::probe(Position const&) 49,519,542 bitboard.h:Pawns::probe(Position const&)
49,235,881 search.cpp:Value (anonymous namespace)::qsearc | 48,987,540 position.h:(anonymous namespace)::evaluate_ini
47,059,357 types.h:(anonymous namespace)::update_stats(Po | 47,858,245 movegen.cpp:ExtMove* generate<(GenType)3>(Posi
46,451,730 bitboard.h:Value (anonymous namespace)::search | 47,059,464 types.h:(anonymous namespace)::update_stats(Po
46,354,221 movegen.cpp:ExtMove* generate<(GenType)3>(Posi | 46,451,730 bitboard.h:Value (anonymous namespace)::search
46,105,920 position.h:(anonymous namespace)::evaluate_ini | 45,239,908 movegen.cpp:ExtMove* (anonymous namespace)::ge
45,239,908 movegen.cpp:ExtMove* (anonymous namespace)::ge | 44,542,936 types.h:Value (anonymous namespace)::qsearch<(
44,542,036 types.h:Value (anonymous namespace)::qsearch<( | 43,693,450 pawns.cpp:Score Pawns::Entry::do_king_safety<(
43,693,450 pawns.cpp:Score Pawns::Entry::do_king_safety<( | 43,234,964 movepick.cpp:Value (anonymous namespace)::sear
42,970,005 movepick.cpp:Value (anonymous namespace)::sear | 42,884,394 position.h:Position::set_check_info(StateInfo*
41,779,159 movegen.cpp:ExtMove* (anonymous namespace)::ge | 41,779,159 movegen.cpp:ExtMove* (anonymous namespace)::ge
40,781,676 bitboard.h:Position::attackers_to(Square, unsi | 41,288,400 types.h:Position::pseudo_legal(Move) const
40,752,954 position.h:Position::remove_piece(Piece, Squar 40,752,954 position.h:Position::remove_piece(Piece, Squar
40,486,359 movepick.h:Value (anonymous namespace)::search | 40,722,248 bitboard.h:Position::attackers_to(Square, unsi
40,428,896 search.cpp:Value (anonymous namespace)::qsearc | 40,479,377 movepick.h:Value (anonymous namespace)::search
39,162,615 movepick.cpp:MovePicker::MovePicker(Position c | 40,428,896 search.cpp:Value (anonymous namespace)::qsearc
38,930,480 evaluate.cpp:Score (anonymous namespace)::eval | 39,099,783 movepick.cpp:MovePicker::MovePicker(Position c
37,957,218 evaluate.cpp:Score (anonymous namespace)::eval | 38,930,480 evaluate.cpp:Score (anonymous namespace)::eval
37,497,188 movegen.cpp:ExtMove* (anonymous namespace)::ge | 37,957,218 evaluate.cpp:Score (anonymous namespace)::eval
35,179,240 types.h:Position::pseudo_legal(Move) const | 37,497,188 movegen.cpp:ExtMove* (anonymous namespace)::ge
34,660,420 movegen.cpp:ExtMove* (anonymous namespace)::ge | 36,758,052 bitboard.h:Position::set_check_info(StateInfo*
34,533,043 movegen.cpp:ExtMove* (anonymous namespace)::ge | 36,378,876 movegen.cpp:MovePicker::next_move()
34,145,599 movegen.cpp:ExtMove* (anonymous namespace)::ge | 34,660,420 movegen.cpp:ExtMove* (anonymous namespace)::ge
33,588,948 movegen.cpp:MovePicker::next_move() | 34,533,043 movegen.cpp:ExtMove* (anonymous namespace)::ge
33,190,128 movegen.cpp:ExtMove* (anonymous namespace)::ge | 34,145,599 movegen.cpp:ExtMove* (anonymous namespace)::ge
33,060,065 bitboard.h:Position::undo_move(Move) | 33,190,128 movegen.cpp:ExtMove* (anonymous namespace)::ge
32,511,553 position.h:Position::legal(Move) const 32,511,553 position.h:Position::legal(Move) const
31,821,730 tt.h:TranspositionTable::probe(unsigned long, | 31,821,730 tt.h:TranspositionTable::probe(unsigned long,
31,736,690 movepick.h:(anonymous namespace)::update_stats | 31,736,690 movepick.h:(anonymous namespace)::update_stats
31,266,079 position.h:Position::see_ge(Move, Value) const | 31,697,820 position.h:void (anonymous namespace)::eval_in
30,440,346 movegen.cpp:ExtMove* (anonymous namespace)::ge | 30,440,346 movegen.cpp:ExtMove* (anonymous namespace)::ge
29,972,110 search.cpp:(anonymous namespace)::update_stats | 29,972,110 search.cpp:(anonymous namespace)::update_stats
29,484,143 types.h:Value (anonymous namespace)::qsearch<( | 29,838,203 position.h:Position::capture(Move) const [/hom
28,479,823 tt.h:Value (anonymous namespace)::qsearch<((an | 29,562,531 bitboard.h:Position::slider_blockers(unsigned
27,739,596 position.h:Square Position::square<(PieceType) | 29,484,272 types.h:Value (anonymous namespace)::qsearch<(
27,479,364 position.h:Square Position::square<(PieceType) | 28,816,200 position.h:void (anonymous namespace)::eval_in
26,599,135 position.h:Position::capture(Move) const [/hom | 28,816,200 types.h:(anonymous namespace)::evaluate_initia
25,934,580 types.h:(anonymous namespace)::evaluate_initia | 27,994,762 /build/glibc-t3gR2i/glibc-2.23/string/../sysde
25,934,580 position.h:void (anonymous namespace)::eval_in | 27,963,954 tt.h:Value (anonymous namespace)::qsearch<((an
25,758,793 types.h:Pawns::probe(Position const&) 25,758,793 types.h:Pawns::probe(Position const&)
25,635,402 /build/glibc-t3gR2i/glibc-2.23/string/../sysde <
24,750,344 position.h:Position::key_after(Move) const 24,750,344 position.h:Position::key_after(Move) const
24,535,358 bitboard.h:Value Pawns::Entry::shelter_storm<( | 24,535,358 bitboard.h:Value Pawns::Entry::shelter_storm<(
24,505,368 /usr/lib/gcc/x86_64-linux-gnu/6/include/bmi2in 24,505,368 /usr/lib/gcc/x86_64-linux-gnu/6/include/bmi2in
23,289,270 movepick.h:(anonymous namespace)::update_cm_st | 23,289,270 movepick.h:(anonymous namespace)::update_cm_st
23,052,960 position.h:void (anonymous namespace)::eval_in | 23,020,240 bitboard.h:void (anonymous namespace)::eval_in
22,751,543 types.h:Value (anonymous namespace)::search<(( | 22,640,530 bitboard.h:Position::remove_piece(Piece, Squar
22,640,530 bitboard.h:Position::remove_piece(Piece, Squar | 22,438,545 types.h:Value (anonymous namespace)::search<((
22,262,730 types.h:Position::set_check_info(StateInfo*) c | 22,235,217 position.cpp:Score (anonymous namespace)::eval
22,235,217 position.cpp:Score (anonymous namespace)::eval | 21,650,746 tt.h:Value (anonymous namespace)::qsearch<((an
22,079,282 tt.h:Value (anonymous namespace)::qsearch<((an | 21,546,053 bitboard.h:Position::undo_move(Move)
20,977,004 material.h:(anonymous namespace)::evaluate_sca | 20,977,004 material.h:(anonymous namespace)::evaluate_sca
20,745,747 position.h:(anonymous namespace)::evaluate_sca | 20,745,747 position.h:(anonymous namespace)::evaluate_sca
20,726,525 types.h:Value (anonymous namespace)::search<(( | 20,670,999 movegen.cpp:ExtMove* (anonymous namespace)::ge
20,670,999 movegen.cpp:ExtMove* (anonymous namespace)::ge | 20,460,860 types.h:Value (anonymous namespace)::search<((
20,390,838 /usr/lib/gcc/x86_64-linux-gnu/6/include/bmi2in | 20,361,124 /usr/lib/gcc/x86_64-linux-gnu/6/include/bmi2in
20,138,620 bitboard.h:void (anonymous namespace)::eval_in | 19,975,465 bitboard.h:void (anonymous namespace)::eval_in
19,975,465 bitboard.h:void (anonymous namespace)::eval_in | 19,775,653 /usr/lib/gcc/x86_64-linux-gnu/6/include/bmi2in
19,775,653 /usr/lib/gcc/x86_64-linux-gnu/6/include/bmi2in | 18,726,148 types.h:Value Pawns::Entry::shelter_storm<(Col
18,726,148 types.h:Value Pawns::Entry::shelter_storm<(Col <
18,687,822 /usr/include/c++/6/bits/stl_vector.h:Pawns::pr 18,687,822 /usr/include/c++/6/bits/stl_vector.h:Pawns::pr
18,669,909 /build/glibc-t3gR2i/glibc-2.23/math/../sysdeps 18,669,909 /build/glibc-t3gR2i/glibc-2.23/math/../sysdeps
18,545,722 position.h:Value (anonymous namespace)::qsearc | 18,545,722 position.h:Value (anonymous namespace)::qsearc
17,048,198 movegen.cpp:ExtMove* (anonymous namespace)::ge | 18,379,026 types.h:Position::set_check_info(StateInfo*) c
> 17,048,198 movegen.cpp:ExtMove* (anonymous namespace)::ge
> 16,979,270 movegen.cpp:ExtMove* (anonymous namespace)::ge
16,700,565 /usr/include/c++/6/bits/stl_algobase.h:Value E 16,700,565 /usr/include/c++/6/bits/stl_algobase.h:Value E
16,077,483 bitboard.h:ExtMove* (anonymous namespace)::gen 16,077,483 bitboard.h:ExtMove* (anonymous namespace)::gen
> 15,448,444 position.cpp:Position::gives_check(Move) const
15,260,698 bitboard.h:ExtMove* (anonymous namespace)::gen 15,260,698 bitboard.h:ExtMove* (anonymous namespace)::gen
14,757,600 /usr/include/x86_64-linux-gnu/bits/string3.h:V | 14,757,600 /usr/include/x86_64-linux-gnu/bits/string3.h:V
14,321,462 position.cpp:Position::gives_check(Move) const | 14,754,108 types.h:Position::capture(Move) const
14,143,098 bitboard.h:ExtMove* (anonymous namespace)::gen 14,143,098 bitboard.h:ExtMove* (anonymous namespace)::gen
> 14,142,865 position.h:ExtMove* (anonymous namespace)::gen
> 13,799,313 position.h:Square Position::square<(PieceType)
13,584,318 types.h:Position::remove_piece(Piece, Square) 13,584,318 types.h:Position::remove_piece(Piece, Square)
13,448,799 types.h:Position::capture(Move) const | 13,454,304 movegen.cpp:ExtMove* (anonymous namespace)::ge
13,431,808 search.cpp:Value (anonymous namespace)::qsearc | 13,431,808 search.cpp:Value (anonymous namespace)::qsearc
13,221,465 bitboard.h:ExtMove* generate<(GenType)1>(Posit 13,221,465 bitboard.h:ExtMove* generate<(GenType)1>(Posit
13,221,327 movegen.cpp:ExtMove* (anonymous namespace)::ge | 13,221,327 movegen.cpp:ExtMove* (anonymous namespace)::ge
13,094,646 bitboard.h:ExtMove* generate<(GenType)3>(Posit 13,094,646 bitboard.h:ExtMove* generate<(GenType)3>(Posit
12,920,501 movegen.cpp:ExtMove* generate<(GenType)5>(Posi | 13,001,173 movegen.cpp:ExtMove* generate<(GenType)5>(Posi
12,775,706 movegen.cpp:ExtMove* (anonymous namespace)::ge <
12,665,928 /usr/include/c++/6/bits/stl_vector.h:Material: 12,665,928 /usr/include/c++/6/bits/stl_vector.h:Material:
12,538,937 position.h:ExtMove* (anonymous namespace)::gen | 12,398,366 position.h:ExtMove* (anonymous namespace)::gen
12,273,001 movegen.cpp:ExtMove* (anonymous namespace)::ge | 12,273,001 movegen.cpp:ExtMove* (anonymous namespace)::ge
12,229,352 movegen.cpp:ExtMove* (anonymous namespace)::ge | 12,229,352 movegen.cpp:ExtMove* (anonymous namespace)::ge
12,192,996 movegen.cpp:ExtMove* (anonymous namespace)::ge | 12,192,996 movegen.cpp:ExtMove* (anonymous namespace)::ge
12,178,850 bitboard.h:ExtMove* (anonymous namespace)::gen 12,178,850 bitboard.h:ExtMove* (anonymous namespace)::gen
12,074,535 /usr/include/c++/6/bits/stl_vector.h:Position: 12,074,535 /usr/include/c++/6/bits/stl_vector.h:Position:
11,954,583 position.h:Position::pseudo_legal(Move) const | 11,875,692 /usr/include/c++/6/bits/stl_algobase.h:Value (
11,816,571 position.h:Value (anonymous namespace)::qsearc | 11,816,571 position.h:Value (anonymous namespace)::qsearc
11,432,069 types.h:Value Pawns::Entry::shelter_storm<(Col | 11,432,069 types.h:Value Pawns::Entry::shelter_storm<(Col
11,384,551 types.h:Value (anonymous namespace)::qsearch<( | 11,371,533 types.h:Value (anonymous namespace)::qsearch<(
11,287,953 position.h:Square Position::square<(PieceType) <
11,143,558 bitboard.h:main 11,143,558 bitboard.h:main
11,078,223 bitboard.h:ExtMove* (anonymous namespace)::gen 11,078,223 bitboard.h:ExtMove* (anonymous namespace)::gen
10,984,637 position.h:void MovePicker::score<(GenType)0>( 10,984,637 position.h:void MovePicker::score<(GenType)0>(
10,973,713 position.h:ExtMove* (anonymous namespace)::gen <
10,752,897 bitboard.h:ExtMove* (anonymous namespace)::gen 10,752,897 bitboard.h:ExtMove* (anonymous namespace)::gen
10,707,811 bitboard.h:attacks_bb(Piece, Square, unsigned 10,707,811 bitboard.h:attacks_bb(Piece, Square, unsigned
10,342,218 movegen.cpp:ExtMove* (anonymous namespace)::ge | 10,146,348 search.cpp:(anonymous namespace)::update_cm_st
10,146,348 search.cpp:(anonymous namespace)::update_cm_st | 10,131,676 movegen.cpp:ExtMove* (anonymous namespace)::ge
10,131,676 movegen.cpp:ExtMove* (anonymous namespace)::ge | 10,066,751 search.cpp:Value (anonymous namespace)::qsearc
Callgrind annotation using --inclusive=yes without/with -DANTI (complete):
------------------------------------------------------------- -------------------------------------------------------------
Profile data file 'callgrind.out.18478' (creator: callgrind-3 | Profile data file 'callgrind.out.18832' (creator: callgrind-3
------------------------------------------------------------- -------------------------------------------------------------
I1 cache: I1 cache:
D1 cache: D1 cache:
LL cache: LL cache:
Timerange: Basic block 0 - 3067735945 | Timerange: Basic block 0 - 3325791624
Trigger: Program termination Trigger: Program termination
Profiled target: ./stockfish_base bench (PID 18478, part 1) | Profiled target: ./stockfish bench (PID 18832, part 1)
Events recorded: Ir Events recorded: Ir
Events shown: Ir Events shown: Ir
Event sort order: Ir Event sort order: Ir
Thresholds: 99 Thresholds: 99
Include dirs: Include dirs:
User annotated: User annotated:
Auto-annotation: off Auto-annotation: off
------------------------------------------------------------- -------------------------------------------------------------
Ir Ir
------------------------------------------------------------- -------------------------------------------------------------
19,101,514,352 PROGRAM TOTALS | 20,222,342,852 PROGRAM TOTALS
------------------------------------------------------------- -------------------------------------------------------------
Ir file:function Ir file:function
------------------------------------------------------------- -------------------------------------------------------------
83,536,425,237 search.cpp:Value (anonymous namespace)::searc | 88,442,985,321 search.cpp:Value (anonymous namespace)::searc
82,243,617,426 search.cpp:Value (anonymous namespace)::searc | 87,091,990,499 search.cpp:Value (anonymous namespace)::searc
18,967,523,598 /build/glibc-t3gR2i/glibc-2.23/nptl/pthread_c | 20,085,885,299 /build/glibc-t3gR2i/glibc-2.23/nptl/pthread_c
18,967,516,050 /home/lila/fishnet/Stockfish/src/thread.cpp:r | 20,085,877,751 /home/lila/fishnet/Stockfish/src/thread.cpp:r
18,967,512,372 thread.cpp:run_idle_loop(void*) [/home/lila/f | 20,085,874,098 thread.cpp:run_idle_loop(void*) [/home/lila/f
18,967,500,641 /home/lila/fishnet/Stockfish/src/search.cpp:M | 20,085,862,378 /home/lila/fishnet/Stockfish/src/search.cpp:M
18,921,866,756 search.cpp:MainThread::search() [/home/lila/f | 20,040,232,737 search.cpp:MainThread::search() [/home/lila/f
18,921,749,845 search.cpp:Thread::search() [/home/lila/fishn | 20,040,119,161 search.cpp:Thread::search() [/home/lila/fishn
18,899,976,702 search.cpp:Value (anonymous namespace)::searc | 20,018,021,858 search.cpp:Value (anonymous namespace)::searc
18,353,502,835 search.cpp:Value (anonymous namespace)::searc | 19,445,750,061 search.cpp:Value (anonymous namespace)::searc
6,930,460,002 search.cpp:Value (anonymous namespace)::qsear | 7,304,268,360 search.cpp:Value (anonymous namespace)::qsear
6,853,866,289 search.cpp:Value (anonymous namespace)::qsear | 7,227,674,647 search.cpp:Value (anonymous namespace)::qsear
5,598,954,686 /home/lila/fishnet/Stockfish/src/evaluate.cpp | 5,949,327,779 /home/lila/fishnet/Stockfish/src/evaluate.cpp
5,078,325,439 search.cpp:Value (anonymous namespace)::qsear | 5,344,881,244 search.cpp:Value (anonymous namespace)::qsear
5,016,418,692 search.cpp:Value (anonymous namespace)::qsear | 5,282,974,497 search.cpp:Value (anonymous namespace)::qsear
4,426,260,340 evaluate.cpp:Value Eval::evaluate<false>(Posi | 4,798,007,064 evaluate.cpp:Value Eval::evaluate<false>(Posi
4,066,481,658 /home/lila/fishnet/Stockfish/src/movepick.cpp | 4,225,879,132 /home/lila/fishnet/Stockfish/src/movepick.cpp
3,214,026,252 movepick.cpp:MovePicker::next_move() [/home/l | 3,313,023,950 movepick.cpp:MovePicker::next_move() [/home/l
2,235,801,714 search.cpp:Value (anonymous namespace)::qsear | 2,345,899,898 search.cpp:Value (anonymous namespace)::qsear
2,215,234,258 search.cpp:Value (anonymous namespace)::qsear | 2,325,332,442 search.cpp:Value (anonymous namespace)::qsear
1,890,006,027 /home/lila/fishnet/Stockfish/src/position.cpp | 2,075,694,511 /home/lila/fishnet/Stockfish/src/position.cpp
1,857,018,603 /home/lila/fishnet/Stockfish/src/position.h:S | 1,978,628,635 /home/lila/fishnet/Stockfish/src/position.h:S
1,420,088,153 position.cpp:Position::do_move(Move, StateInf | 1,606,303,356 position.cpp:Position::do_move(Move, StateInf
1,403,087,215 /home/lila/fishnet/Stockfish/src/position.cpp | 1,541,865,342 /home/lila/fishnet/Stockfish/src/position.cpp
1,039,362,837 position.cpp:Position::see_ge(Move, Value) co | 1,194,560,680 position.cpp:Position::see_ge(Move, Value) co
1,028,670,253 evaluate.cpp:Score (anonymous namespace)::eva | 1,099,081,913 evaluate.cpp:Score (anonymous namespace)::eva
698,359,488 position.cpp:Position::set_check_info(StateIn | 860,319,261 position.cpp:Position::set_check_info(StateIn
554,323,098 /home/lila/fishnet/Stockfish/src/movegen.cpp: | 566,437,422 /home/lila/fishnet/Stockfish/src/movegen.cpp:
539,180,193 movegen.cpp:ExtMove* generate<(GenType)0>(Pos | 551,294,517 movegen.cpp:ExtMove* generate<(GenType)0>(Pos
508,174,158 /home/lila/fishnet/Stockfish/src/position.cpp | 510,501,523 /home/lila/fishnet/Stockfish/src/position.cpp
470,740,759 pawns.h:Value Eval::evaluate<false>(Position | 469,414,495 pawns.h:Value Eval::evaluate<false>(Position
400,707,071 types.h:Value Eval::evaluate<false>(Position | 417,705,235 /home/lila/fishnet/Stockfish/src/position.cpp
391,351,977 /home/lila/fishnet/Stockfish/src/position.cpp | 404,208,276 types.h:Value (anonymous namespace)::search<(
388,975,638 types.h:Value (anonymous namespace)::search<( | 388,129,706 types.h:Value Eval::evaluate<false>(Position
> 364,696,940 search.cpp:Value (anonymous namespace)::qsear
> 361,888,015 search.cpp:Value (anonymous namespace)::qsear
348,601,201 /home/lila/fishnet/Stockfish/src/pawns.cpp:Pa 348,601,201 /home/lila/fishnet/Stockfish/src/pawns.cpp:Pa
347,510,303 search.cpp:Value (anonymous namespace)::qsear | 332,939,118 types.h:MovePicker::next_move()
344,701,378 search.cpp:Value (anonymous namespace)::qsear | 318,033,272 position.h:Score (anonymous namespace)::evalu
> 302,821,971 position.cpp:Position::slider_blockers(unsign
286,678,984 /home/lila/fishnet/Stockfish/src/types.h:Posi 286,678,984 /home/lila/fishnet/Stockfish/src/types.h:Posi
274,424,112 /home/lila/fishnet/Stockfish/src/position.h:P | 274,424,365 /home/lila/fishnet/Stockfish/src/position.h:P
273,765,607 position.h:Score (anonymous namespace)::evalu <
268,173,002 types.h:MovePicker::next_move() <
254,484,355 /home/lila/fishnet/Stockfish/src/tt.h:Transpo 254,484,355 /home/lila/fishnet/Stockfish/src/tt.h:Transpo
244,932,812 /home/lila/fishnet/Stockfish/src/movegen.cpp: | 247,555,471 /home/lila/fishnet/Stockfish/src/movegen.cpp:
244,153,930 movegen.cpp:ExtMove* (anonymous namespace)::g | 247,361,786 movegen.cpp:ExtMove* (anonymous namespace)::g
238,721,072 pawns.cpp:Pawns::probe(Position const&) [/hom | 246,711,680 /home/lila/fishnet/Stockfish/src/position.cpp
238,404,868 /home/lila/fishnet/Stockfish/src/position.cpp | 238,721,072 pawns.cpp:Pawns::probe(Position const&) [/hom
237,583,235 /home/lila/fishnet/Stockfish/src/types.h:Posi | 237,587,164 /home/lila/fishnet/Stockfish/src/position.h:P
231,781,071 bitboard.h:Score (anonymous namespace)::evalu | 237,370,744 bitboard.h:Score (anonymous namespace)::evalu
229,485,444 movepick.h:MovePicker::next_move() | 231,194,185 position.cpp:Position::legal(Move) const [/ho
224,727,989 types.h:Score (anonymous namespace)::evaluate | 226,069,023 types.h:Score (anonymous namespace)::evaluate
223,458,164 /home/lila/fishnet/Stockfish/src/types.h:Posi 223,458,164 /home/lila/fishnet/Stockfish/src/types.h:Posi
> 223,236,718 movegen.cpp:ExtMove* generate<(GenType)1>(Pos
222,888,960 /home/lila/fishnet/Stockfish/src/position.h:P 222,888,960 /home/lila/fishnet/Stockfish/src/position.h:P
222,662,625 tt.cpp:TranspositionTable::probe(unsigned lon 222,662,625 tt.cpp:TranspositionTable::probe(unsigned lon
221,803,031 /home/lila/fishnet/Stockfish/src/pawns.cpp:Sc 221,803,031 /home/lila/fishnet/Stockfish/src/pawns.cpp:Sc
221,360,877 types.h:Position::see_ge(Move, Value) const | 219,185,412 movepick.cpp:void MovePicker::score<(GenType)
220,614,059 movegen.cpp:ExtMove* generate<(GenType)1>(Pos | 216,103,882 movegen.cpp:ExtMove* (anonymous namespace)::g
213,785,140 position.cpp:Position::undo_move(Move) [/home | 213,916,160 position.cpp:Position::undo_move(Move) [/home
213,254,576 movegen.cpp:ExtMove* (anonymous namespace)::g | 211,876,496 types.h:Position::do_move(Move, StateInfo&, b
211,631,064 types.h:Position::do_move(Move, StateInfo&, b | 211,862,059 movepick.h:MovePicker::next_move()
210,262,336 pawns.cpp:Score Pawns::Entry::do_king_safety< | 210,262,336 pawns.cpp:Score Pawns::Entry::do_king_safety<
206,849,746 movepick.cpp:void MovePicker::score<(GenType) | 200,704,071 types.h:Position::see_ge(Move, Value) const
202,999,192 position.cpp:Position::legal(Move) const [/ho <
191,653,835 position.h:Position::is_variant_end() const [ <
191,653,835 /home/lila/fishnet/Stockfish/src/position.h:P <
181,460,635 /home/lila/fishnet/Stockfish/src/pawns.cpp:Sc <
Callgrind annotation using --threshold=50 without/with -DANTI (complete):
------------------------------------------------------------- -------------------------------------------------------------
Profile data file 'callgrind.out.18478' (creator: callgrind-3 | Profile data file 'callgrind.out.18832' (creator: callgrind-3
------------------------------------------------------------- -------------------------------------------------------------
I1 cache: I1 cache:
D1 cache: D1 cache:
LL cache: LL cache:
Timerange: Basic block 0 - 3067735945 | Timerange: Basic block 0 - 3325791624
Trigger: Program termination Trigger: Program termination
Profiled target: ./stockfish_base bench (PID 18478, part 1) | Profiled target: ./stockfish bench (PID 18832, part 1)
Events recorded: Ir Events recorded: Ir
Events shown: Ir Events shown: Ir
Event sort order: Ir Event sort order: Ir
Thresholds: 50 Thresholds: 50
Include dirs: Include dirs:
User annotated: User annotated:
Auto-annotation: off Auto-annotation: off
------------------------------------------------------------- -------------------------------------------------------------
Ir Ir
------------------------------------------------------------- -------------------------------------------------------------
19,101,514,352 PROGRAM TOTALS | 20,222,342,852 PROGRAM TOTALS
------------------------------------------------------------- -------------------------------------------------------------
Ir file:function Ir file:function
------------------------------------------------------------- -------------------------------------------------------------
1,823,907,316 search.cpp:Value (anonymous namespace)::search | 2,047,873,044 search.cpp:Value (anonymous namespace)::search
1,467,929,891 movepick.cpp:MovePicker::next_move() [/home/li | 1,502,438,085 movepick.cpp:MovePicker::next_move() [/home/li
1,345,721,132 evaluate.cpp:Value Eval::evaluate<false>(Posit | 1,405,249,353 evaluate.cpp:Value Eval::evaluate<false>(Posit
1,028,670,253 evaluate.cpp:Score (anonymous namespace)::eval | 1,099,081,913 evaluate.cpp:Score (anonymous namespace)::eval
819,019,917 position.cpp:Position::see_ge(Move, Value) con | 974,217,760 position.cpp:Position::see_ge(Move, Value) con
677,130,431 position.cpp:Position::do_move(Move, StateInfo | 708,334,202 position.cpp:Position::do_move(Move, StateInfo
400,707,071 types.h:Value Eval::evaluate<false>(Position c | 404,208,276 types.h:Value (anonymous namespace)::search<((
388,975,638 types.h:Value (anonymous namespace)::search<(( | 388,129,706 types.h:Value Eval::evaluate<false>(Position c
362,780,583 position.cpp:Position::set_check_info(StateInf | 332,939,118 types.h:MovePicker::next_move()
273,765,607 position.h:Score (anonymous namespace)::evalua | 318,033,272 position.h:Score (anonymous namespace)::evalua
268,173,002 types.h:MovePicker::next_move() | 269,559,168 position.cpp:Position::set_check_info(StateInf
238,721,072 pawns.cpp:Pawns::probe(Position const&) [/home | 238,721,072 pawns.cpp:Pawns::probe(Position const&) [/home
231,781,071 bitboard.h:Score (anonymous namespace)::evalua | 237,370,744 bitboard.h:Score (anonymous namespace)::evalua
229,485,444 movepick.h:MovePicker::next_move() | 230,445,283 position.cpp:Position::legal(Move) const [/hom
My last couple commits both seem to be about a 1% speedup. Now for some more callgrind testing!
ddugovic
on 17 Mar 2017
👍1
My last commit is on a new branch because it changes functionality. I will submit a test.
EDIT: Nevermind, cutechess-cli and/or fishtest currently do not support antichess.
ddugovic
on 17 Mar 2017
Now removing -DANTI is only a 4% speedup (previously was a 6% speedup):
base = 630162 +/- 8076
test = 655207 +/- 10425
diff = 25045 +/- 7135
speedup = 0.039744
@ddugovic 3 day ago removing -DANTI on my pc I got speedup = 0.037466 (view my previous post). If I'm not wrong the 6% speedup is with only 1 measure.
ppigazzini
on 17 Mar 2017
Ah, you are correct. Allow me to rerun that 3-day-old test:
base = 626633 +/- 8032
test = 657890 +/- 11225
diff = 31257 +/- 12764
speedup = 0.049881
I am curious, what is the performance penalty when using make profile-build instead of make build? Perhaps there isn't a large performance penalty after all?
ddugovic
on 18 Mar 2017
make profile-build is a 11% speedup
bash bench-parallel.sh ./stockfish ./stockfish-profile 10
run base test diff
1 1775428 1983423 +207995
2 1711471 1924263 +212792
3 1759121 1968137 +209016
4 1698193 1879993 +181800
5 1701021 1912852 +211831
6 1700077 1878840 +178763
7 1757607 1964352 +206745
8 1790993 1998949 +207956
9 1767747 1968137 +200390
10 1782658 1978301 +195643
base = 1744431 +/- 24444
test = 1945724 +/- 28396
diff = 201293 +/- 7978
speedup = 0.115392
Ah, thanks... and now that I reread your initial comment (and understand the nuance of profile-build being used for official releases) there's still an issue.
ddugovic
on 18 Mar 2017
I think this is fixed. What do you think?
$ bash bench-parallel.sh ./stockfish_base ./stockfish_test 10
run 1 / 10
run 2 / 10
run 3 / 10
run 4 / 10
run 5 / 10
run 6 / 10
run 7 / 10
run 8 / 10
run 9 / 10
run 10 / 10
run base test diff
1 695424 661569 -33855
2 875414 741982 -133432
3 686928 672679 -14249
4 673122 624111 -49011
5 682188 632876 -49312
6 697959 622400 -75559
7 624875 650475 +25600
8 652483 638347 -14136
9 646086 622147 -23939
10 651997 625640 -26357
base = 688647 +/- 45592
test = 649222 +/- 24185
diff = -39425 +/- 27783
speedup = -0.057250
PGO (profile-build):
$ bash bench-parallel.sh ./stockfish_base ./stockfish_test 10
run 1 / 10
run 2 / 10
run 3 / 10
run 4 / 10
run 5 / 10
run 6 / 10
run 7 / 10
run 8 / 10
run 9 / 10
run 10 / 10
run base test diff
1 727106 735394 +8288
2 778356 868837 +90481
3 838639 903949 +65310
4 802826 884766 +81940
5 697561 691813 -5748
6 684015 652205 -31810
7 704052 700111 -3941
8 679089 653387 -25702
9 649992 667550 +17558
10 684168 671867 -12301
base = 724580 +/- 40211
test = 742987 +/- 66549
diff = 18407 +/- 29240
speedup = 0.025404
On my PC there is always a 13-14% slowdown respect to Official Stockfish
MSYS2 & MinGW-w64
$ bash bench-parallel.sh ./stockfish-off.exe ./stockfish-mv.exe 10
run base test diff
1 2167177 1857768 -309409
2 2191981 1860588 -331393
3 2161826 1863417 -298409
4 2181836 1858895 -322941
5 2164881 1857768 -307113
6 2174096 1848802 -325294
7 2187287 1870240 -317047
8 2182613 1831673 -350940
9 2158020 1852714 -305306
10 2190414 1861719 -328695
base = 2176013 +/- 8141
test = 1856358 +/- 6815
diff = -319654 +/- 10138
speedup = -0.146899
LSW
~# bash bench-parallel.sh ./stockfish-off ./stockfish-mv 10
run base test diff
1 2264048 1963722 -300326
2 2156502 1885198 -271304
3 2171016 1897454 -273562
4 2122153 1865118 -257035
5 2163352 1877114 -286238
6 2247443 1963093 -284350
7 2197482 1888101 -309381
8 2263212 1949977 -313235
9 2205390 1916441 -288949
10 2146682 1848244 -298438
base = 2193728 +/- 33005
test = 1905446 +/- 26936
diff = -288281 +/- 11559
speedup = -0.131412
👍1
Ah, you are correct. Sorry after much effort debugging I'd been so focused on "all variants plus/minus antichess" that I'd forgotten the original issue.
ddugovic
on 18 Mar 2017
Another benchmark test (no variant flags in Makefile versus all variant flags in Makefile):
$ bash bench-parallel.sh ./stockfish-base ./stockfish 10
run 1 / 10
run 2 / 10
run 3 / 10
run 4 / 10
run 5 / 10
run 6 / 10
run 7 / 10
run 8 / 10
run 9 / 10
run 10 / 10
run base test diff
1 1324023 1169922 -154101
2 1398909 1241048 -157861
3 1417876 1231815 -186061
4 1434597 1258605 -175992
5 1434597 1261535 -173062
6 1407830 1243637 -164193
7 1255159 1116873 -138286
8 1214250 1083970 -130280
9 1425676 1263139 -162537
10 1430466 1262871 -167595
base = 1374338 +/- 52912
test = 1213341 +/- 43150
diff = -160996 +/- 11043
speedup = -0.117145
Today I ran another test:
run base test diff
1 1517140 1358765 -158375
2 1473646 1320027 -153619
3 1501368 1330159 -171209
4 1512223 1363346 -148877
5 1514110 1364573 -149537
6 1308918 1185367 -123551
7 1488473 1329868 -158605
8 1488107 1321753 -166354
9 1512600 1349394 -163206
10 1519420 1360593 -158827
base = 1483600 +/- 41304
test = 1328384 +/- 34780
diff = -155216 +/- 8590
speedup = -0.104621
Was this page helpful?
0 / 5 - 0 ratings
Related issues
ppigazzini
·
8Comments
ddugovic
·
10Comments
ppigazzini
·
14Comments
ianfab
·
6Comments
isaacl
·
9Comments
Most helpful comment
Following the Michael's suggestion, I played with the Makefile flags in 3.2.1 Debugging section.
Slowdown wrt flags kept:
15% All flags
7% -DANTI
4% -DATOMIC
4% -DCRAZYHOUSE
1% No flags
So in fishtest should be possible to have builds for the variant tested, gaining 10% of CPU power.