Welcome to the project of translating Stockfish into assembly language. This project
now uses the new assembler engine fasmg from Tomasz Grysztar. The includes in
arm/includes/ or x86/include/ contain instruction and formatting macros for
the four popular targets in the Building section. The hello world examples in these
directories should provide enough to grasp the syntax.
All versions of the executables may be built using the fasmg executable. However,
fasmg is currently only available as an x86 executable. fasmg is a generic assembler
which relies on the particual flavor of the assembly language to be supplied by macros.
This slows down the processing of the source by a few orders of magnitute. The location
of these macros (relative to the current source) is harded coded into the fish source in
the variable FASMG_INC. The -e 100 switch tells fasmg to display the last 100 errors
when processing the source. The -i switch inserts lines at the beginning at the
source. The fish source expect that VERSION_OS is defined this way.
This allows multiple versions to be assembled from the same source. Your working
directory should be the root directory of this repository (the one that contains the
fasmg executables)
The x86-64 linux version links against nothing and should work with any 64 bit x86 linux kernel.
~/asm$ ./fasmg "x86/fish.asm" "asmfish" -e 100 -i "VERSION_OS='L'" -i "VERSION_POST = 'popcnt'"
flat assembler version g.hwx32
4 passes, 18.5 seconds, 112326 bytes.
~/asm$ chmod 755 ./asmfish
~/asm$ ./asmfish bench
The x86-64 windows version links against only kernel32.dll and should work even on XP.
C:\asm>fasmg.exe "x86\fish.asm" "asmfish.exe" -e 100 -i "VERSION_OS='W'" -i "VERSION_POST = 'popcnt'"
flat assembler version g.hwx32
5 passes, 23.2 seconds, 115200 bytes.
C:\asm>asmfish.exe bench
The x86-64 macOS version links against /usr/lib/libSystem.B.dylib and works on version 10.12.16.
~/asm$ ./fasmg "x86/fish.asm" "asmfish" -e 100 -i "VERSION_OS='X'" -i "VERSION_POST = 'popcnt'"
flat assembler version g.hwx32
4 passes, 17.7 seconds, 119495 bytes.
~/asm$ chmod 755 ./asmfish
~/asm$ ./asmfish bench
The aarch64 linux version links against nothing should work with any 64 bit arm linux kernel. Of course it can currently only be built on x86 machines.
~/asm$ ./fasmg "arm/fish.arm" "armfish" -e 100 -i "VERSION_OS='L'" -i "VERSION_POST = 'v8'"
flat assembler version g.hwx32
3 passes, 8.1 seconds, 128018 bytes.
~/asm$ chmod 755 ./armfish
~/asm$ qemu-aarch64 ./armfish bench
You can feed in commands to the engine as (":" -> ";" on windows):
./asmfish setoption name hash value 256: go depth 5: wait
The engine quits after this if assemble flag USE_CMDLINEQUIT=1 is set.
Besides the usual uci commands there are the following:
| included by default | |
|---|---|
| perft | Usual move generation verification. Use like perft 7. |
| bench | Usual bench command. Use like stockfish or the more readable form bench hash 16 threads 1 depth 13. These are the defaults. |
| wait | Waits for the main search thread to finish. Use with caution (esp. on an infinite search). This is useful when feeding commands via the command line. The command wait can be used after go to ensure that engine doesn't quit before finishing. |
USE_BOOK=1 default assemble option |
|
| bookprobe | Displays book entries from the current position. Use like bookprobe 3. |
VERBOSE=1 assemble option |
|
| show | Prints out the internal rep of the position. |
| moves | Makes the succeeding moves then does show. |
| undo | Undoes one or a certain number of moves |
| donull | Does a null move. |
| eval | Displays evaluation. |
| included by default | |
|---|---|
| Priority | Try to set the priority of the process. The default is 'none', which runs the engine at whichever priority it was started. |
| LogFile | Set the location to write all communication. Useful for buggy gui's. A value of <empty> means the logger is off. |
| TTFile | Set the location of the file for TTSave and TTLoad. A value of <empty> means that the following two command will fail. |
| TTSave | Saves the current hash table to TTFile. |
| TTLoad | Loads the current hash table from TTFile while possibily changing the size. |
| LargePages | Try to use large pages when allocating the hash. Hash and threads are only allocated when receiving isready or go. |
| NodeAffinity | The default is all. The command setoption name nodeaffinity value all will show the detected cores/nodes in your machine. Here is the general behavior: |
all pin threads to all nodes your machine in a uniform way
none disable pinning threads to nodes
0 1 2 3 only use nodes 0, 1, 2 and 3
2 only use node 2
0.1 2.3 use nodes 0, 1, 2 and 3
but node 1 shares per-node memory with node 0
node 3 shares per-node memory with node 2
0.1.2.3 use nodes 0, 1, 2 and 3
but nodes 1, 2 and 3 share per-node memory with node 0
USE_SYZYGY=1 default assemble option |
|
|---|---|
| SyzygyProbeDepth | Don't probe if plies from root is less than this. |
| SyzygyProbeLimit | Don't probe if number of board pieces is bigger than this. |
| Syzygy50MoveRule | Consider 50 move rule when probing. |
| SyzygyPath | Path to syzygy tablebases. |
USE_BOOK=1 default assemble option |
|
|---|---|
| OwnBook | Lookup position in book if possible. Ponder moves are also selected from the book when possible |
| BookFile | Loads polyglot book into engine. |
| BestBookMove | Use only the best moves from the book (highest weight) |
| BookDepth | Tricky setting works as follows: |
BookDepth <= 0:
suppose the lines the book from the current position are
T0: h2h3(30) c5d4(10) e3d4(14) g4h5(10) g2g4(11)
h2h3(30) g4h5(5)
d4c5(17) d6c5(17) b1c3(7)
the moves g2g4(11), g4h5(5) and b1c3(7) are leaves and don't lead
to a position in the book. Triming off these leaves three times,
T1: h2h3(30) c5d4(10) e3d4(14) g4h5(10)
d4c5(17) d6c5(17)
T2: h2h3(30) c5d4(10) e3d4(14)
d4c5(17)
T3: h2h3(30) c5d4(10)
If BookDepth = 0, probe as if it were in T0 (unchanged)
If BookDepth =-1, probe as if it were in T1 (leaves off)
If BookDepth =-2, probe as if it were in T2 (trim twice)
If BookDepth =-3, probe as if it were in T3 (trim trice)
So with BookDepth <= -3, the move d4c5 is not considered.
With BookDepth <= -5, the move h2h3 is also not considered.
BookDepth >= 1:
Book is not probed if gameply >= BookDepth
The option BestBookMove is also slightly complicated for compatibility with the
cerebellum book in polyglot format. In the following pseudocode, movelist is the
list of legal book moves after processing the BookDepth option. After this pseudocode
if movelist is non-empty, then a random move is chosen from movelist according to
the weights of these moves.
if BestBookMove = true
if length(movelist) > 1
filter out moves from movelist that lead to repetitions
filter out moves from movelist without highest weight
end if
else
filter out moves from movelist that lead to repetitions
end if
This means that if BestBookMove=true and the first move is encoded with a higher weight
than the second move (both having non-zero weight in the book), then the behaviour should
match that of brainfish. If BestBookMove=false, then the only differences are
- if the first move leads to repetition and there is no second move, then the engine starts calculating
- if there are two moves, neither of which lead to a repition, then both of these moves will be considered.
USE_WEAKNESS=1 assembly option |
|
|---|---|
| UCI_LimitStrength | make the engine play at certain level |
| UCI_Elo | level at which to play |
USE_VARIETY=1 assemble option |
|
| Variety | In QSearch, if the score is in the range [-4v, 1000+4v), add a pseudorandom integer from (-4v, 4v) to the score. |
| The distribution of this integer is roughly triangular with mean 0. |
X
X X X
X X X X X
X X X X X X X
X X X X X X X
X X X X X X X X X
X X X X X X X X X X X
X X X X X X X X X X X X X
-4v 0 +4v
If you observe a crash/misbehaviour in asmfish, please raise an issue here and give the following information:
- name of the executable that crashed/misbehaved
- exception code and exception offset in the case of a crash
- a log of the commands that were sent to asmFish by your gui before the crash Simply stating that asmFish crashed in your gui is useless information by itself.
asmfish is known to have problems in the fritz15 gui, while it plays much better in the fritz11 gui.
Windows might throw a "The system cannot execute the specified program." or "Insufficient system resources exist to complete the requested service.". The likely source of this problem is your virus software. Rest assured that the sources here do not produce any behaviour that is even remotely virus-like (unless you run analysis for a long time with syzygy6 installed).
Q: Why not just start with the compiler output and speed up the critical functions? or write critical functions in asm and include them in cpp code?
A: With this approach the critical functions would still need to conform to the standards set in place by the ABI. All of the critical functions in asmFish do not conform to these standards. Plus, asmFish would be dependent on a compiler in this case, which introduces many unnecessary compilcations. Both asmFish and its assembler are around 100KB; lets keep it simple. Note that compiler output was used in the case of Ronald de Man's syzygy probing code, as this is not speed critical but cumbersome to write by hand.
Q: Is asmFish search the same as official stockfish?
A: It does now that PEDANTIC = 1 is the default! The changes previously thought to be inconsequential lose about 2 Elo in a head-to-head matchup. The functionality when using syzygy is not 100% identical because asmFish uses Ronald's original alpha-beta search while official stockfish does not. This causes minor inconsequential differences due to the piece lists.
Q: Where can I find the executable files of the old versions?
A: All older versions of asmFish/pedantFish are in the branch https://github.com/lantonov/asmFish/tree/executables
For a change log, see the Wiki https://github.com/lantonov/asmFish/wiki/Change-log