Fast bitboard chess move generation based on magic bitboards, optional bulking or nullmovenewcount, transposition table & multithreading.
mperft <args> Enumerate moves. The following options are available:
--bulk|-b Do fast bulk counting at the last ply.
--nullmove|-n Do fast nullmove counting at the penultimate ply.
[--depth|-d] <depth> Test up to this depth (default = 6).
--div Print a node count for each move.
--fast Automatically set highest settings.
--fen|-f <fen> Use the position indicated in FEN format (default = starting position).
--hash|-h <size> Use a hashtable with <size> Megabytes (default = 0, no hashtable).
--help|-? Print this message.
--kiwipete|-k Use the kiwipete position.
--moves|-m <moves> Play a serie of moves to build the position to use.
--loop|-l Loop from depth 1 to <depth>.
--quiet|-q Disable verbose output.
--repeat|-r <n> Repeat the test <n> time (default = 1).
--seed|-s <seed> Change the seed of the pseudo move generator to <seed>.
--test Run an internal test to check the move generator.
--threads|-t <threads> Use <threads> threads for parallel processing (default = 1).
Mperft is limited to a maximal depth of 64 plies. The hash table is limited to the available memory or 2⁶⁴ bytes. The number of threads is limited to the number of logical cores available. On 64 bits versions, the leaf counter is limited to 2⁵⁸ = 288,230,376,151,711,744 leaves, which is enough for executing mperft up to depth 13. On 128 bits versions, the leaf counter is limited to 2¹²² = 5,316,911,983,139,663,491,615,228,241,121,378,304 leaves.
You can download the binary release for your platform from the releases page. Executable are given for Linux & Windows for intel64/amd64 cpu in 3 flavours:
- x86-64
- x86-64-v2 support AVX extension, including the popcount instruction.
- x86-64-v3 support AVX2 extension, including the PEX instruction.
- x86-64-v3-128 support AVX2 extension, including the PEX instruction and use 128 bits counter & Zobrist's hash keys.
You can compile mperft for your own CPU using the makefile or the .bat file provided for clang & icx (intel C compiler). For example under Linux, using the clang compiler:
CC=clang make
or under MS-Windows, after opening a terminal windows:
build-clang.bat native
$ mperft --fast -l 12
Magic Perft version 5.3 (c) Richard Delorme 2020 - 2026
Bitboard move generation based on magic (pext) bitboards.
Perft setting: hashtable size: 32768 Mbytes (2147483652 entries); with 32 threads; with nullmove counting.
a b c d e f g h
8 r n b q k b n r 8
7 p p p p p p p p 7
6 . . . . . . . . 6
5 . . . . . . . . 5
4 . . . . . . . . 4
3 . . . . . . . . 3
2 P P P P P P P P 2
1 R N B Q K B N R 1
a b c d e f g h
w, KQkq
perft 1: 20 positions in 0:00.000 3.995 Mpos/s
perft 2: 400 positions in 0:00.000 186.414 Mpos/s
perft 3: 8,902 positions in 0:00.000 321.877 Mpos/s
perft 4: 197,281 positions in 0:00.000 898.433 Mpos/s
perft 5: 4,865,609 positions in 0:00.000 8.360 Gpos/s
perft 6: 119,060,324 positions in 0:00.003 30.156 Gpos/s
perft 7: 3,195,901,860 positions in 0:00.030 103.482 Gpos/s
perft 8: 84,998,978,956 positions in 0:00.354 239.668 Gpos/s
perft 9: 2,439,530,234,167 positions in 0:04.046 602.825 Gpos/s
perft 10: 69,352,859,712,417 positions in 0:44.580 1.556 Tpos/s
perft 11: 2,097,651,003,696,806 positions in 10:00.076 3.496 Tpos/s
perft 12: 62,854,969,236,701,747 positions in 3:28:33.121 5.023 Tpos/s
total : 65,024,500,949,358,489 positions in 3:39:22.215 4.940 Tpos/s
full time: 3:39:57.540 s
- Version 1.0
- Version 1.1 More compact hash table using bitfields to store the counter & the depth.
- Version 2.0 Use copy/make instead of update/restore after each move. Replace bitfields with shift/mask.
- Version 3.0 Add support for multithreading.
- Version 3.1 Remove some OS/platform dependencies. Alphabetically sort the move with the --div command.
- Version 3.2 Remove more OS/platform dependencies. Fix a bug (free() incompatibility with _aligned_malloc()). Faster executable for Windows.
- Version 4.0 Code fix & optimisation. Add --fast & --moves option. remove --capture option.
- Version 4.1 Code fix & optimisation. Add optionally compiled 128 bit support for crunching big numbers.
- Version 5.0 Add nullmove counting at the penultimate remaining depth (=2). Significantly faster.
- Version 5.1 Add Gigantua's benchmark, more readable output & some small enhancements. Slightly faster.
- Version 5.2 Fix the 128 bits version.
- Version 5.3 Code fix. Faster lockless hashtable.
Time to compute perft 7 with bulk counting on a ryzen 9 5950x @ 4.2 Ghz under Linux :
| Version | Time | Options | Progress |
|---|---|---|---|
| 1.1 | 3.865 s | -b 7 | |
| 2.0 | 3.787 s | -b 7 | +2.1% |
| 3.2 | 3.811 s | -b 7 | -0.6% |
| 4.1 | 3.006 s | -b 7 | +26.8% |
| 5.0 | 3.314 s | -b 7 | -9.3% |
| 5.0 | 2.070 s | -n 7 | +60.1% |
| 5.1 | 1.993 s | -n 7 | +3.8% |
| 5.3 | 2.045 s | -n 7 | -3.6% |
Time to compute perft 8 with bulk counting and 256 MB of hash table:
| Version | Time | Options | Progress |
|---|---|---|---|
| 1.1 | 9.985 s | -b 8 -H 24 | |
| 2.0 | 9.709 s | -b 8 -h 256 | +2.8% |
| 3.2 | 10.138 s | -b 8 -h 256 | -4.2% |
| 4.1 | 8.420 s | -b 8 -h 256 | +20.4% |
| 5.0 | 6.655 s | -n 8 -h 256 | +26.5% |
| 5.1 | 6.522 s | -n 8 -h 256 | +2.0% |
| 5.1 | 6.522 s | -n 8 -h 256 | +2.0% |
| 5.3 | 6.429 s | -n 8 -h 256 | +1.6% |
Time to compute perft 9 with bulk counting, 32 threads and 1 GB of hash table:
| Version | Time | Options | Progress |
|---|---|---|---|
| 2.0 | 118.197 s | -b 9 -h 1024 | |
| 3.2 | 6.249 s | -b 9 -t 32 -h 1024 | +1791.4% |
| 4.1 | 5.386 s | -b 9 -t 32 -h 1024 | +16.0% |
| 5.0 | 4.708 s | -n 9 -t 32 -h 1024 | +14.4% |
| 5.1 | 4.422 s | -n 9 -t 32 -h 1024 | +5.3% |
| 5.3 | 4.160 s | -n 9 -t 32 -h 1024 | +5.9% |