voldien / hpm

hpm - High Performance Matrices, 3D math library

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

HPM

Hpm Linux License: LGPL v3 GitHub release

HPM is a matrix library that supports multiple SIMD implementations that can be loaded in run-time. Furthermore, the program can load the most appropriated implementation on its own and, it can as well explicitly load a specific SIMD extension with the available SIMD enumerators.

The number of features is relatively low as a result of the same function as to be implemented multiple times for multiple SIMD extensions.

Features

  • Matrix4x4, Matrix3x3 - Common matrix operation for a 4x4 and 3x3 is supported.
  • Vector4, Vector3 - Common vector operation for vector3 and vector4 is supported.
  • Quaternion - About all common quaternion operation is supported.
  • Compare - Compare matrices, vectors, and quaternion by the object or by component-wise.
  • Math - Simple maths function as max, min, sqrt, and random are supported.
  • Utility - Support functions that utilities the other features for creating specific types of matrices.
  • Print - Easy function for printing vector primitive type values.

Remark

The matrix2x2 and the vector2 are not supported as a result of not being able to take advantages of the SIMD features at all.

Motivation

The main motivation of this project is to create a single library header interface for performing matrix operations, commonly used in 3D programming. Where the appropriated implementation, based on the available CPU extensions, will load the most efficient implementation. Where it can be done either automatically or manually.

Installation

The software can be easily installed with invoking the following commands.

mkdir build && cd build
cmake ..
cmake --build .
make install

Examples

  1. The following code initializes the hpm library to use the SSE2 SIMD extension and, create an identity matrix. The release function will close the loaded library which in succession will make each function pointer undefined, which means all the matrices functions will be undefined.
#include<hpm/hpm.h>

int main(int argc, const char** argv){
	hpm_init(HPM_SSE2);
	hpmvec4x4f_t mat4;
	hpm_mat4x4_identitiy(mat4);
	hpm_mat4x4_print(mat4);
	hpm_release();
}

The following line demonstrates how to compile it and link the program in a Unix environment:

gcc *.c -o hpmexample -lhpm

Assert/Performance Tools

In order to evaluate if the library has any performance differences on the system. The following test program for hpm can be used.

Performance

It will perform a performance test for all SIMD feature if supported by the machine when no SIMD extension argument has been set.

hpmassert --performance

The performance gain can be rather subtle for high-end computer specification, whereas low ends are more distinguishable. What should remark about the hpmassert program is that everything is executed in consecutive order. This does not represent what the performance differences would be when each function is called in random order.

The following table is an example of a subset of the result that is presented with the benchmark program. To see some tested CPU performance, Check the benchmarks.

Function NOSIMD SSE SSE2 SSE3 SSSE3 SSE41 SSE42 AVX
hpm_mat4x4_copyfv 1.000 1.375 1.375 1.375 1.375 1.375 1.376 1.032
hpm_mat4x4_multiply_mat4x4fv 1.000 2.949 2.936 2.925 2.949 2.899 2.939 3.476
hpm_mat4x4_multiply_mat1x4fv 1.000 2.352 2.352 2.352 2.352 2.353 2.354 2.537
hpm_mat4x4_identityfv 1.143 1.143 1.143 1.143 1.143 1.143 1.143 1.000
hpm_mat4x4_transposefv 1.111 1.000 1.000 1.000 1.000 1.000 1.000 1.213
hpm_mat4x4_determinantfv 1.111 1.000 1.000 1.000 1.000 1.000 1.000 1.212
....

Assertion - Integrity

The following command will perform a set of unit tests that will check and determine that all of the implementations for each of the SIMD extensions work.

hpmassert --assert

hpmglbunny

The hpmglbunny program is a simple test program to determine if the matrices and quaternion function implementation work, which is commonly used in rendering 3D models. This program will render multiple rotating bunnies. See the following for an example of how to execute the test tool:

hpmglbunny --simd=avx

The hpmglbunny will by default use the HPM_SIMD_DEFAULT enumerator as the flag parameter for loading function pointers through the hpm_init function. This will attempt to load the latest SIMD extension supported on the system. However, the simd= argument can override it to a specific SIMD extension.

The primary purpose of this small program is to check if the hpm implementations work as accordingly in a 3D environment.

Important

Because the library is still in development. That means that the symbol table has yet not been finalized. Thus it can be expected that it might change in till the first stable release.

Dependencies

The hpm project does not have any additional dependency other than the libraries provided with the standard c libraries. However, it is recommended that the following Debian packages have to be installed prior, see the following. If a higher version is already installed this is not required.

apt-get install gcc-6

The following command will display current version of gcc.

gcc --version

License

This project is licensed under the LGPL+3 License - see the LICENSE file for details

About

hpm - High Performance Matrices, 3D math library

License:GNU Lesser General Public License v3.0


Languages

Language:C 91.2%Language:CMake 3.4%Language:Java 2.8%Language:C# 2.6%Language:Lua 0.0%