Questions, functionality requests, bug reports, features - contact us at support@gpusystems.com

For pricing, licensing and purchase information, contact sales@gpusystems.com

Documentation located in SDK /doc folder. See also web - www.gpusystems.com/doc
Libra API functions follow Octave/Matlab syntax closely.

---------------------------------------------------------------------------------------------------------------------

Libra SDK Requirements:

Windows x86/x64:
* Windows Xp, Vista, WIn7 x86/x64 + service packs.
* Windows x64 requires x86 Libra.dlls ro run in x86 mode. (for example, run matlab x86 instead of matlab x64)

Mac x86/x64
* Mac required system libraries -> at command prompt : type 'otool -L libLibra.dylib' 

Linux x86/x64
* Linux required system libraries -> at command prompt : type 'ldd libLibra.so' 
* To build Libra samples, gcc-4.5, g++-4.5 32 bit and 64 bit libraries and libstdc++ standard libraries are required

NVIDIA:
* NVIDIA hardware CUDA 1.0 or higher (for GPU acceleration of computations)
* NVIDIA Driver for Microsoft Windows XP, Vista, Win7 with CUDA >= 3.0 Support

AMD:
* AMD hardware > 5xxx series and OpenCL 1.1 or higher (for GPU acceleration of computations)
* AMD driver for Microsoft Windows XP, Vista, Win7 with AMD App SDK >= 2.5 Support (OpenCL runtime)

---------------------------------------------------------------------------------------------------------------------

Installing Libra with your product/application:

* Link with Libra.lib and LibraCpp.lib for C++ bindings.
* Deploy all Libra binaries from Libra SDK/bin/ under your application binary PATH.

Developing & Debugging:

* Run/browse Libra SDK sample programs.
* Getting started: Run initial tests & computations using the EmptySample project.

* Use libra_GetLastErrorMessage() to check Libra execution errors and to control program flow.
* Use myVar.print() to flush your variables to std::out and IDE output window.
* Use myVar.getData() to download and view Libra variables data in your debugger.
* Use render(myVar) to view your variables graphically. Enable cmdLine -c ClientDebug for visual debugging.
  - Using the debugger, break into your application code after calling Libra function render(), then freeze your application thread and release breakpoint and continue execution.
* Try -b CPU_BACKEND for comparison of results.

Compute states:

* Preferred compute backend can be selected by passing  "-b CUDA_BACKEND", "-b OPENCL_BACKEND", "-b CPU_BACKEND" or "-b OPENGL_BACKEND" to libra_Init() 
  or switched during runtime execution calling libra_SetCurrentBackend().
* Note that default array indexing is 0 indexed and default compute precision is 32-bit floating point (GFLOAT32) unless other stated. See libra_SetDefaultIndexing() and libra_SetDefaultDataType().
* Libra runtime supports GFLOAT32, GFLOAT64, GINT32 precision computations using real or complex number arithmetics. See www.gpusystems.com/doc for more details.