A small project which performs a comparative performance test between the original parsii library, and the BigDecimal variant bdparsii.
The executed code is
Scope scope = new Scope();
scope.setMathContext(mathcontext); // only for bdparsii
Variable var = scope.create("x");
Expression expr = Parser.parse("2 + (7 - 5) * 3.14159 * x^(12-10) + sin(-3.141)", scope);
for (int x = 1_000_000-iterations; x < 1_000_000; ++x) {
var.setValue(x);
expr.evaluate();
}This will test both parsing/compilation of the expression as execution of the given expression. This test is based on the original performance test of the parsii library.
Parsii will simplify the above expression as much as possible. The resulting expression which will be executed during expr.evaluate() is 1.99941 + 6.28318 * x^2 as the rest has already been evaluated to constants.
Tested on JDK 1.8.0_102, VM 25.102-b14
DbParsii
Parsii
Full Results
Benchmark (iterations) (mathContext) Mode Cnt Score Error Units
BdParsii.run 1 DECIMAL32 thrpt 50 79460.994 ± 672.434 ops/s
BdParsii.run 1 DECIMAL64 thrpt 50 47553.672 ± 351.739 ops/s
BdParsii.run 1 DECIMAL128 thrpt 50 29329.106 ± 197.162 ops/s
BdParsii.run 1 UNLIMITED thrpt 50 29772.296 ± 177.296 ops/s
BdParsii.run 10 DECIMAL32 thrpt 50 68319.088 ± 696.285 ops/s
BdParsii.run 10 DECIMAL64 thrpt 50 41858.520 ± 336.723 ops/s
BdParsii.run 10 DECIMAL128 thrpt 50 26645.651 ± 218.029 ops/s
BdParsii.run 10 UNLIMITED thrpt 50 28486.879 ± 237.865 ops/s
BdParsii.run 100 DECIMAL32 thrpt 50 29826.022 ± 218.464 ops/s
BdParsii.run 100 DECIMAL64 thrpt 50 18410.439 ± 159.023 ops/s
BdParsii.run 100 DECIMAL128 thrpt 50 14246.796 ± 107.520 ops/s
BdParsii.run 100 UNLIMITED thrpt 50 21217.938 ± 152.603 ops/s
BdParsii.run 1000 DECIMAL32 thrpt 50 4588.383 ± 69.692 ops/s
BdParsii.run 1000 DECIMAL64 thrpt 50 2808.223 ± 27.695 ops/s
BdParsii.run 1000 DECIMAL128 thrpt 50 2474.903 ± 22.553 ops/s
BdParsii.run 1000 UNLIMITED thrpt 50 6102.212 ± 41.085 ops/s
BdParsii.run 10000 DECIMAL32 thrpt 50 481.395 ± 3.481 ops/s
BdParsii.run 10000 DECIMAL64 thrpt 50 296.237 ± 2.207 ops/s
BdParsii.run 10000 DECIMAL128 thrpt 50 265.686 ± 2.415 ops/s
BdParsii.run 10000 UNLIMITED thrpt 50 796.020 ± 9.266 ops/s
BdParsii.run 100000 DECIMAL32 thrpt 50 47.832 ± 0.395 ops/s
BdParsii.run 100000 DECIMAL64 thrpt 50 29.998 ± 0.202 ops/s
BdParsii.run 100000 DECIMAL128 thrpt 50 27.432 ± 0.206 ops/s
BdParsii.run 100000 UNLIMITED thrpt 50 81.581 ± 1.035 ops/s
BdParsii.run 1000000 DECIMAL32 thrpt 50 5.454 ± 0.053 ops/s
BdParsii.run 1000000 DECIMAL64 thrpt 50 2.921 ± 0.034 ops/s
BdParsii.run 1000000 DECIMAL128 thrpt 50 2.859 ± 0.050 ops/s
BdParsii.run 1000000 UNLIMITED thrpt 50 7.948 ± 0.045 ops/s
Parsii.run 1 N/A thrpt 50 83044.987 ± 386.330 ops/s
Parsii.run 10 N/A thrpt 50 80499.569 ± 704.917 ops/s
Parsii.run 100 N/A thrpt 50 75333.864 ± 654.469 ops/s
Parsii.run 1000 N/A thrpt 50 41177.726 ± 432.268 ops/s
Parsii.run 10000 N/A thrpt 50 7710.984 ± 175.995 ops/s
Parsii.run 100000 N/A thrpt 50 834.514 ± 6.545 ops/s
Parsii.run 1000000 N/A thrpt 50 83.238 ± 1.582 ops/s
As these results show, performing BigDecimal calculations are slower than double based calculations.
The performance depends a lot on the chosen MathContext, and used mathematical functions.
High precisions with complex math perform worse.
If only basic math (*/+-%) is used the unlimited precision will actually perform better because no time is spend on rounding the results to the correct precision.