fennecdjay / jitverse

Comparison of all the JITs in the universe

JITverse

All the JITs in the universe compared based on:

• Performance (WIP)
• Microprocessor architecture support (WIP)
• Maximum number of available registers (TODO)
• Ease of integration (TODO)
• Compilation speed (TODO)

Performance

Algorithm	LibJIT	libgccjit	MCJIT	AsmJit	DynASM	QBE	MIR	GNU Lightning	MyJit	GCC	NASM
fib(42)	1.541 s ± 0.012 s	538.0 ms ± 5.3 ms	-	-	-	-	-	1.344 s ± 0.007 s	1.809 s ± 0.067 s	769.5 ms ± 22.1 ms	803.6 ms ± 12.0 ms

Mean elapsed time against 10 runs. Lower is better.

fib(42) is the recursive Fibonacci sequence algorithm.

C compilers and assemblers added as reference.

Summary:

Summary
  './libgccjit/toyvm ./libgccjit/fibonacci.toy 42' ran
    1.43 ± 0.04 times faster than './gcc/fib'
    1.49 ± 0.03 times faster than './nasm/fib'
    2.50 ± 0.03 times faster than './gnu_lightning/fib'
    2.87 ± 0.04 times faster than './libjit/fib'
    3.36 ± 0.13 times faster than './myjit/fib'

Peformed using:

• gcc (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0
• NASM version 2.14.02
• hyperfine 1.11.0

See the related directories for more details.

Microprocessor Architecture Support

Architecture	LibJIT	libgccjit	MCJIT	AsmJit	DynASM	QBE	MIR	GNU Lightning	MyJit
x86	-	Yes	-	Yes	Yes	-	-	Yes	Yes
x86-64	-	Yes	-	Yes	Yes	-	-	Yes	Yes
ARM32	-	Yes	-	No	Yes	-	-	Yes	Yes
AArch64	-	Yes	-	Yes	Yes	-	-	Yes	No
IA-64	-	Yes	-	No	No	-	-	Yes	No
RISC-V	-	Yes	-	No	No	-	-	Yes	No
MIPS	-	Yes	-	No	Yes	-	-	Yes	No
HPPA	-	Yes	-	No	No	-	-	Yes	No
SPARC	-	Yes	-	No	No	-	-	Yes	Yes
PowerPC	-	Yes	-	No	Yes	-	-	Yes	No
Alpha	-	Yes	-	No	No	-	-	Yes	No
S390	-	Yes	-	No	No	-	-	Yes	No