Scam is a chess engine written in Rust by Fabian von der Warth and Terje Kirstihagen. Scam is uci compliant and supports FRC.
Install the rust compiler and cargo, the recommended way to do this is via rustup. Then run
git clone https://github.com/fabianvdW/Scam
cd Scam
set RUSTFLAGS=-C target-cpu=native
cargo run --release
If you're on an AMD processor with bmi2 instruction available but microcoded, please disable them manually:
set RUSTFLAGS=-C target-cpu=native -C target-feature=-bmi2
cargo run --release
Install Rust nightly:
rustup toolchain install nightly
Add .cargo folder, .cargo/config file in the root folder of the repo. Content:
[target.x86_64-unknown-linux-musl]
linker="rust-lld"
Finally we are ready to install our cross-compilation target and use it:
rustup default nightly
rustup target add x86_64-unknown-linux-musl
set RUSTFLAGS=-C target-cpu=<your-target-machine-cpu>
cargo rustc --release --bin scam --target x86_64-unknown-linux-musl
To find a suitable target-cpu, choose one from
rustc --target=x86_64-unknown-linux-musl --print target-cpus
Please note the comment in our build.rs file:
In the case that the Host does not have BMI2, while the target-cpu wants BMI2 instructions, this build script will fail.
To run the bench command:
cargo run --release -- bench
To run the built-in benchmarks made by us to evaluate Scam's performance, run
cargo bench --bench movegen -- --verbose
, but make sure to close any other running processes first. On Fabi's Intel i5-6400 this yields:
Gnuplot not found, using plotters backend
Benchmarking pseudolegal
Benchmarking pseudolegal: Warming up for 3.0000 s
Benchmarking pseudolegal: Collecting 100 samples in estimated 5.0011 s (1398850 iterations)
Benchmarking pseudolegal: Analyzing
pseudolegal time: [3.5781 us 3.5850 us 3.5936 us]
change: [-1.0866% -0.4203% +0.1589%] (p = 0.20 > 0.05)
No change in performance detected.
Found 12 outliers among 100 measurements (12.00%)
4 (4.00%) high mild
8 (8.00%) high severe
slope [3.5781 us 3.5936 us] R^2 [0.9884607 0.9881747]
mean [3.5830 us 3.6027 us] std. dev. [25.700 ns 74.070 ns]
median [3.5733 us 3.5815 us] med. abs. dev. [10.029 ns 17.670 ns]
Benchmarking makemove
Benchmarking makemove: Warming up for 3.0000 s
Benchmarking makemove: Collecting 100 samples in estimated 5.0665 s (131300 iterations)
Benchmarking makemove: Analyzing
makemove time: [38.570 us 38.651 us 38.751 us]
change: [+23.488% +24.265% +25.203%] (p = 0.00 < 0.05)
Performance has regressed.
Found 11 outliers among 100 measurements (11.00%)
1 (1.00%) high mild
10 (10.00%) high severe
slope [38.570 us 38.751 us] R^2 [0.9859528 0.9856462]
mean [38.590 us 38.968 us] std. dev. [300.02 ns 1.6091 us]
median [38.502 us 38.562 us] med. abs. dev. [77.118 ns 144.32 ns]
Benchmarking perft1
Benchmarking perft1: Warming up for 3.0000 s
Benchmarking perft1: Collecting 100 samples in estimated 5.1409 s (106050 iterations)
Benchmarking perft1: Analyzing
perft1 time: [48.164 us 48.280 us 48.424 us]
change: [+18.173% +18.934% +19.780%] (p = 0.00 < 0.05)
Performance has regressed.
Found 12 outliers among 100 measurements (12.00%)
2 (2.00%) high mild
10 (10.00%) high severe
slope [48.164 us 48.424 us] R^2 [0.9814246 0.9810094]
mean [48.246 us 48.770 us] std. dev. [480.66 ns 2.1541 us]
median [48.081 us 48.174 us] med. abs. dev. [100.26 ns 189.77 ns]
Benchmarking perft2
Benchmarking perft2: Warming up for 3.0000 s
Warning: Unable to complete 100 samples in 5.0s. You may wish to increase target time to 8.5s, enable flat sampling, or reduce sample count to 50.
Benchmarking perft2: Collecting 100 samples in estimated 8.5363 s (5050 iterations)
Benchmarking perft2: Analyzing
perft2 time: [1.6921 ms 1.6950 ms 1.6984 ms]
change: [+17.217% +17.657% +18.088%] (p = 0.00 < 0.05)
Performance has regressed.
Found 12 outliers among 100 measurements (12.00%)
6 (6.00%) high mild
6 (6.00%) high severe
slope [1.6921 ms 1.6984 ms] R^2 [0.9916868 0.9915719]
mean [1.6930 ms 1.7014 ms] std. dev. [11.811 us 29.448 us]
median [1.6884 ms 1.6913 ms] med. abs. dev. [4.0494 us 7.5664 us]