Benchmarks for various decimal programs.

General notes

• Times are measured in seconds, unless otherwise noted
• Measured on a MacBook Pro, 2.9 GHz Intel Core i5, 8 GB 2133 MHz LPDDR3
• Some benchmarks are adapted from www.bytereef.org/mpdecimal/benchmarks.html

The benchmarks aim to be as fair as possible, but ultimately they do compare different libraries with different feature sets. For example, go-inf/inf boasts the fastest overall runtime of any library, but it also doesn't implement the GDA spec: it lacks contexts, non-finite NaN/Inf/±zero values, conditions, etc. Further, programs like cocroachdb/apd sacrifice speed to ensure strict compatibility with the GDA spec.

In general, libraries that cannot fully complete a challenge will be unranked. For example, Go's float64 type cannot provide 19 or more digits of precision, so it's unranked in the Pi test. Similarly so with apmckinlay/dnum.

Pi

Program (version)	9 digits	19 digits	38 digits	100 digits	average
go-inf/inf (Go 1.9)	0.10	0.23	0.53	1.43	0.572
JDK BigDecimal (Java 1.8, warm)	0.049	0.19	0.60	3.29	1.05
ericlagergren/decimal (Go 1.9, mode Go)	0.046	0.29	0.81	3.09	1.10
ericlagergren/decimal (Go 1.9, mode GDA)	0.048	0.34	0.97	3.70	1.26
Python decimal (Python 3.6.2)	0.27	0.58	1.32	4.52	1.67
JDK BigDecimal (Java 1.8)	0.29	0.96	1.79	3.99	1.76
shopspring/decimal decimal (Go 1.9)	0.38	0.94	1.95	5.26	2.13
cockroachdb/apd (Go 1.9)	0.52	2.14	9.01	71.62	20.81
Python decimal (Python 2.7.10)	12.93	28.91	64.96	192.58	74.84
float64 (Go 1.9)	0.057	-	-	-	-
double (C LLVM 9.0.0 -O3)	0.057	-	-	-	-
apmckinlay/dnum (Go 1.9)	0.091	-	-	-	-
float (Python 2.7.10)	0.59	-	-	-	-

Mandelbrot

Program (version)	9 digits	16 digits	19 digits	34 digits	38 digits	average
ericlagergren/decimal (Go 1.9, mode GDA)	2.73	9.07	14.54	24.95	25.09	15.27
ericlagergren/decimal (Go 1.9, mode Go)	2.73	9.70	15.02	26.13	26.62	16.04
float64 (Go 1.9)	0.0034	-	-	-	-	-