edusporto / ingpu Goto Github PK

------------------------------------------------------------
ingpu - an experimental GPU-based interaction nets evaluator
------------------------------------------------------------
author: Eugen Jiresch
for any comments, suggestions, feedback, please send an email to: jiresch (at) logic . at

NOTE: If running the new version on Linux:
- Requires CMake and CUDA
- Run `build.sh`

Requirements:
-------------
CUDA 4.0 or later
Boost 1.47 or later
ingpu uses some C++ 0x features

Build:
------
Project files for Visual Studio 2010 are provided.

Usage:
------
Currently, rulesets are coded as C++ CUDA kernels. To add your own, implement a header file similar to ackermann.h or fibonacci.h .
sample kernels provided:
ackermann.h
fibonacci.h
algae.h (basic L-System)
cantor_dust.h (Cantor dust fractal set)

Running ingpu will evaluate the input nets in input.txt. 
The following options are available in options.txt:
verbose: print debug information during reduction 
print_final: print the final set of rules
pause: if verbose is enabled, pause after each interaction/communication loop
ruleset: 1 = ackermann, 2 = L-System, 3 = fibonacci. to add your own ruleset, modify kernel.cu at line 796.
file_output: output some statistics/profiling to out.csv
soa: use "structure of arrays" input data instead of "array of structures" (default). work in progress, may crash for some rules/inputs
eval_all: evaluate all (uncommented) inputs in input.txt. If set to 0, only the first input is evaluated.
pause_between_inputs: pause after evaluating each input net.


Current known issues:
---------------
- sometimes crashes for large inputs
- soa mode is broken for ackermann
- VS2010 Debug mode is broken on some machines