JMH performance benchmark for Java's native call APIs/libraries: JNI, FFM, JNA, JNADirect, JNR, JFFI, Multiffi/FFI.
5 functions to measure:
- no arguments, return void
- no arguments, return int
- 1 argument: int, return int
- no arguments, return double
- 1 argument: double, return double
- JDK 22+
- Gradle 8.10
- C compiler with C99 support
- CMake 3.24+ with a valid backend
- JAVA_HOME environment variable
First, build the native shared library:
cd ${PROJECT_ROOT}/src/main/c
cmake -S . -B ./build -DCMAKE_BUILD_TYPE=MinSizeRel
cd build
make
cp ${LIBRARY_NAME} ${PROJECT_ROOT}/Then, run the jmh main class:
cd ${PROJECT_ROOT}
./gradlew clean jmhCPU: Intel® Core™ i7-1165G7 @ 2.80GHz
OS: GNU/Linux
JDK: Eclipse Temurin JDK 23.0.1+11
VM: OpenJDK 64-Bit Server VM, 23.0.1+11
JMH version: 1.37
Blackhole mode: compiler (auto-detected)
Warmup: 3 iterations, 10 s each
Measurement: 5 iterations, 10 s each
Benchmark mode: Average time, time/op (smaller is better)
Benchmark Mode Cnt Score Error Units
FFIBenchmark.measureFFMCriticalReturnDouble avgt 5 4.194 ± 5.300 ns/op
FFIBenchmark.measureFFMCriticalReturnDoubleD avgt 5 4.012 ± 1.472 ns/op
FFIBenchmark.measureFFMCriticalReturnInt avgt 5 3.451 ± 2.118 ns/op
FFIBenchmark.measureFFMCriticalReturnIntI avgt 5 4.161 ± 3.261 ns/op
FFIBenchmark.measureFFMCriticalReturnVoid avgt 5 4.142 ± 2.442 ns/op
FFIBenchmark.measureFFMReturnDouble avgt 5 9.895 ± 3.708 ns/op
FFIBenchmark.measureFFMReturnDoubleD avgt 5 10.758 ± 4.867 ns/op
FFIBenchmark.measureFFMReturnInt avgt 5 10.593 ± 2.854 ns/op
FFIBenchmark.measureFFMReturnIntI avgt 5 9.901 ± 3.693 ns/op
FFIBenchmark.measureFFMReturnVoid avgt 5 9.908 ± 4.729 ns/op
FFIBenchmark.measureJFFIReturnDouble avgt 5 24.674 ± 12.044 ns/op
FFIBenchmark.measureJFFIReturnDoubleD avgt 5 42.278 ± 31.962 ns/op
FFIBenchmark.measureJFFIReturnInt avgt 5 12.857 ± 4.925 ns/op
FFIBenchmark.measureJFFIReturnIntI avgt 5 13.017 ± 5.100 ns/op
FFIBenchmark.measureJFFIReturnVoid avgt 5 12.865 ± 5.120 ns/op
FFIBenchmark.measureJNADirectReturnDouble avgt 5 86.693 ± 48.477 ns/op
FFIBenchmark.measureJNADirectReturnDoubleD avgt 5 112.694 ± 41.022 ns/op
FFIBenchmark.measureJNADirectReturnInt avgt 5 82.319 ± 36.637 ns/op
FFIBenchmark.measureJNADirectReturnIntI avgt 5 107.550 ± 64.214 ns/op
FFIBenchmark.measureJNADirectReturnVoid avgt 5 88.388 ± 41.813 ns/op
FFIBenchmark.measureJNAReturnDouble avgt 5 100.716 ± 58.079 ns/op
FFIBenchmark.measureJNAReturnDoubleD avgt 5 336.410 ± 246.009 ns/op
FFIBenchmark.measureJNAReturnInt avgt 5 100.289 ± 32.735 ns/op
FFIBenchmark.measureJNAReturnIntI avgt 5 284.400 ± 125.287 ns/op
FFIBenchmark.measureJNAReturnVoid avgt 5 90.989 ± 30.362 ns/op
FFIBenchmark.measureJNIReturnDouble avgt 5 10.550 ± 5.542 ns/op
FFIBenchmark.measureJNIReturnDoubleD avgt 5 10.235 ± 5.418 ns/op
FFIBenchmark.measureJNIReturnInt avgt 5 10.466 ± 2.353 ns/op
FFIBenchmark.measureJNIReturnIntI avgt 5 10.515 ± 4.926 ns/op
FFIBenchmark.measureJNIReturnVoid avgt 5 10.712 ± 5.547 ns/op
FFIBenchmark.measureJNRReturnDouble avgt 5 11.334 ± 2.811 ns/op
FFIBenchmark.measureJNRReturnDoubleD avgt 5 10.772 ± 3.878 ns/op
FFIBenchmark.measureJNRReturnInt avgt 5 10.894 ± 4.452 ns/op
FFIBenchmark.measureJNRReturnIntI avgt 5 11.371 ± 3.534 ns/op
FFIBenchmark.measureJNRReturnVoid avgt 5 11.372 ± 3.911 ns/op
FFIBenchmark.measureMultiffiFFMCriticalProxyReturnDouble avgt 5 3.789 ± 1.959 ns/op
FFIBenchmark.measureMultiffiFFMCriticalProxyReturnDoubleD avgt 5 3.965 ± 2.024 ns/op
FFIBenchmark.measureMultiffiFFMCriticalProxyReturnInt avgt 5 3.567 ± 1.961 ns/op
FFIBenchmark.measureMultiffiFFMCriticalProxyReturnIntI avgt 5 3.789 ± 2.077 ns/op
FFIBenchmark.measureMultiffiFFMCriticalProxyReturnVoid avgt 5 3.820 ± 0.849 ns/op
FFIBenchmark.measureMultiffiFFMCriticalReturnDouble avgt 5 12.155 ± 12.412 ns/op
FFIBenchmark.measureMultiffiFFMCriticalReturnDoubleD avgt 5 14.578 ± 8.389 ns/op
FFIBenchmark.measureMultiffiFFMCriticalReturnInt avgt 5 9.433 ± 4.139 ns/op
FFIBenchmark.measureMultiffiFFMCriticalReturnIntI avgt 5 10.839 ± 5.821 ns/op
FFIBenchmark.measureMultiffiFFMCriticalReturnVoid avgt 5 14.600 ± 7.638 ns/op
FFIBenchmark.measureMultiffiFFMProxyReturnDouble avgt 5 10.409 ± 5.587 ns/op
FFIBenchmark.measureMultiffiFFMProxyReturnDoubleD avgt 5 9.611 ± 2.304 ns/op
FFIBenchmark.measureMultiffiFFMProxyReturnInt avgt 5 9.703 ± 3.488 ns/op
FFIBenchmark.measureMultiffiFFMProxyReturnIntI avgt 5 12.952 ± 5.469 ns/op
FFIBenchmark.measureMultiffiFFMProxyReturnVoid avgt 5 10.175 ± 5.208 ns/op
FFIBenchmark.measureMultiffiFFMReturnDouble avgt 5 16.730 ± 7.672 ns/op
FFIBenchmark.measureMultiffiFFMReturnDoubleD avgt 5 18.294 ± 6.760 ns/op
FFIBenchmark.measureMultiffiFFMReturnInt avgt 5 14.094 ± 5.861 ns/op
FFIBenchmark.measureMultiffiFFMReturnIntI avgt 5 15.242 ± 8.650 ns/op
FFIBenchmark.measureMultiffiFFMReturnVoid avgt 5 14.006 ± 5.996 ns/op
FFIBenchmark.measureMultiffiJNANoASMProxyReturnDouble avgt 5 95.944 ± 70.058 ns/op
FFIBenchmark.measureMultiffiJNANoASMProxyReturnDoubleD avgt 5 343.547 ± 261.444 ns/op
FFIBenchmark.measureMultiffiJNANoASMProxyReturnInt avgt 5 82.089 ± 37.464 ns/op
FFIBenchmark.measureMultiffiJNANoASMProxyReturnIntI avgt 5 259.277 ± 152.047 ns/op
FFIBenchmark.measureMultiffiJNANoASMProxyReturnVoid avgt 5 81.552 ± 35.034 ns/op
FFIBenchmark.measureMultiffiJNAProxyReturnDouble avgt 5 77.852 ± 35.094 ns/op
FFIBenchmark.measureMultiffiJNAProxyReturnDoubleD avgt 5 109.233 ± 68.433 ns/op
FFIBenchmark.measureMultiffiJNAProxyReturnInt avgt 5 80.248 ± 31.446 ns/op
FFIBenchmark.measureMultiffiJNAProxyReturnIntI avgt 5 105.182 ± 44.533 ns/op
FFIBenchmark.measureMultiffiJNAProxyReturnVoid avgt 5 87.607 ± 66.138 ns/op
FFIBenchmark.measureMultiffiJNAReturnDouble avgt 5 76.634 ± 49.938 ns/op
FFIBenchmark.measureMultiffiJNAReturnDoubleD avgt 5 302.456 ± 136.001 ns/op
FFIBenchmark.measureMultiffiJNAReturnInt avgt 5 75.469 ± 52.398 ns/op
FFIBenchmark.measureMultiffiJNAReturnIntI avgt 5 242.772 ± 172.470 ns/op
FFIBenchmark.measureMultiffiJNAReturnVoid avgt 5 73.695 ± 35.541 ns/op
FFIBenchmark.measureMultiffiJNRNoASMProxyReturnDouble avgt 5 44.013 ± 20.332 ns/op
FFIBenchmark.measureMultiffiJNRNoASMProxyReturnDoubleD avgt 5 81.801 ± 50.209 ns/op
FFIBenchmark.measureMultiffiJNRNoASMProxyReturnInt avgt 5 24.891 ± 9.386 ns/op
FFIBenchmark.measureMultiffiJNRNoASMProxyReturnIntI avgt 5 50.352 ± 42.205 ns/op
FFIBenchmark.measureMultiffiJNRNoASMProxyReturnVoid avgt 5 24.866 ± 10.057 ns/op
FFIBenchmark.measureMultiffiJNRProxyReturnDouble avgt 5 11.387 ± 5.635 ns/op
FFIBenchmark.measureMultiffiJNRProxyReturnDoubleD avgt 5 11.679 ± 7.289 ns/op
FFIBenchmark.measureMultiffiJNRProxyReturnInt avgt 5 11.323 ± 7.082 ns/op
FFIBenchmark.measureMultiffiJNRProxyReturnIntI avgt 5 10.858 ± 4.306 ns/op
FFIBenchmark.measureMultiffiJNRProxyReturnVoid avgt 5 11.022 ± 3.553 ns/op
FFIBenchmark.measureMultiffiJNRReturnDouble avgt 5 27.093 ± 18.061 ns/op
FFIBenchmark.measureMultiffiJNRReturnDoubleD avgt 5 42.993 ± 29.516 ns/op
FFIBenchmark.measureMultiffiJNRReturnInt avgt 5 12.751 ± 5.317 ns/op
FFIBenchmark.measureMultiffiJNRReturnIntI avgt 5 13.555 ± 7.328 ns/op
FFIBenchmark.measureMultiffiJNRReturnVoid avgt 5 19.049 ± 7.520 ns/op