/CuCodeML

GPU version CodeML

Primary LanguageC

CuCodeML Manual

  1. Introduction
  2. Getting Started
  3. License and Warranty

1. Introduction

CuCodeML is a GPU version of CodeML for the branch-site model.

2. Getting Started

The following instructions take you through a sequence of steps to get the default configuration of CuCodeML up and running.

(a) Prerequisites

  • Hardware Requirements
    • GPU: NVIDIA graphics card(s) with CUDA compute capability 2.0 and above
  • Software Requirements
    • OS: we have tested on Centos6.4
    • C compiler: gcc is sufficient
    • CUDA suites: CUDA Driver, CUDA Toolkit

(b) Configuration

  • Edit Makefile with your favorite editor and edit the following options in configuration part to fit your need:

    • CUDA_INSTALL_PATH: your cuda installation path
    • MULTICARD_ONLY_GPU: yes if you want to use multiple GPU cards and not cooperate with CPU
    • HYBRID: yes if you want to use multiply GPU cards or cooperate with CPU(single GPU card or multiple cards) (only available if MULTICARD_ONLY_GPU is no)
    • SINGLE_GPU_ID: the single GPU card id which you want to use (from 0 to #GPUs-1)(only available if MULTICARD_ONLY_GPU and HYBRID are no)
    • DIVIDE_EQUALLY: yes if you want to divide the sites equally among all the GPU cards (yes is suggested if your GPU cards are of the same type) (only avaliable if MULTICARD_ONLY_GPU is yes)
    • SSE: yes if your CPU support SSE
    • MAXGPU: how many GPU cards to use
    • MAXCPU: how many CPU cores to use (zero if you want to use GPU only)

    WARNING: DO NOT EDIT ANY OTHER PART OF THE MAKEFILE

    Summary: If you want a quick configuration, just modifying the CUDA_INSTALL_PATH and SINGLE_GPU_ID configuration is usually enough.

    If you just want to use one GPU card and not cooperate with CPU, edit the configuration just like this:

         MULTICARD_ONLY_GPU = no
     HYBRID = no
     SINGLE_GPU_ID = #GPUID you want to use
     MAXGPU = 1

    If you want to use one GPU card and cooperate with CPU, edit like this:

         MULTICARD_ONLY_GPU = no
     HYBRID = yes
     MAXGPU = 1

    If you want to use multiple GPU cards and not cooperate with CPU, edit like this:

         MULTICARD_ONLY_GPU = yes
     HYBRID = no
     MAXGPU = #GPUs you want to use ( bigger than 1)

    If you want to use multiple GPU cards and cooperate with CPU, edit like this:

         MULTICARD_ONLY_GPU = no
     HYBRID = yes
     MAXGPU = #GPUs you want to use( bigger than 1)

(c) Build

  • For all shells: make

(d) Installation

  • CuCodeML doesn't need to be installed.

(e) Execution

  • Run the program with the following command: your-CuCodeML-path/CuCodeML
  • The arguments it takes is exactly the same as that codeml takes.

3. License and Warranty

CuCodeML is a free, open-source software under GNU GPL v3 license.