Bytecode compiler benchmark

Introduction

Here we provide results of performance tests of new Unigine v0.32 bytecode compiler. We had compared GLSH (Unigine-based application) with PERL, Python and Lua.

Hardware

Tests were performed at AMD and Intel platforms under Linux 2.6 (GCC 3.3.5, optimization flags: -fomit-frame-pointer -march=i686 -O3).
Intel-based PC: Intel Pentium 4 2,8 GHz (800 MHz FSB) / 1024 Mb / Intel BLKD915PSYL (i915P)
AMD-based PC: AMD Barton 2500+ / 512 Mb / ASUS A7N8X-VM/400 (nForce2)

Results

Full tests archive could be downloaded here (212 Kb). All of the tests were performed 3 times (to minimize errors), average values of execution times (in seconds) are shown in diagrams (lower values are better).

Diagrams legend:

Array elements access

Simple math operations and array elements access (1000000 iterations)

Intel	AMD

Fibbonachi numbers

Recursive Fibbonachi numbers generator fib(32)

Intel	AMD

Loop

Loop with variable inrement inside (10000000 iterations)

Intel	AMD

Nested loops

Six nested loops with 16 iterations per each loop, variable increment inside

Intel	AMD

Matrix multiplication

128x128 matrix multiplication

Intel	AMD

Optimizator

Loop with simple math redundant expression inside (1000000 iterations)

Intel	AMD

Vectors math

Simple math for 3D vectors (500000 iterations)

Intel	AMD

3D vectors normalization

Normalization of 100000 3D vectors

Intel	AMD

3D vectors dot product

Dot product of 100000 3D vectors

Intel	AMD

3D vectors cross product

Cross product of 100000 3D vectors

Intel	AMD

Random numbers generator

Simple random number generator, loop with 20000000 iterations and function call

Intel	AMD

Gaussian distribution

Simple Gaussian random number generator and calculation of Gaussian distribution

Intel	AMD

Quick sort

Quick sort algorithm of 1980 elements array (no external C functions, pure bytecode). Bigger array causes stack overflow in qsort for python

Intel	AMD

Bubble sort

Bubble sort algorithm of 1980 elements array

Intel	AMD

Diagrams legend:

Conclusion

Performance of Unigine bytecode compiler is comparable with PERL at common tasks. Python looks like an outsider in these tests. Lua has better performance at simple tasks (it can be caused by the fact that list of built-in types is very short for this language). Unigine bytecode compiler is especially optimized for game development and similar 3D applications - thus, it has built-in data types for vector and matrix operations. Therefore Unigine bytecode compiler has decided superiority in such tasks as sorting and vectors/matrixes manipulations.

There should be mentioned another advantage of Unigine scripts - they have C/C++ like syntax. Therefore you can prototype something using Unigine bytecode compiler and then obtain great speed increase by compiling this code, if necessary. Moreover, there are a lot of good C/C++ developers who can start building their own Unigine-based applications without learning a new language for scripting.