Using the Jetson TK1 as a remote development environment for CUDA

since: 2015/08/01
update: 2015/08/01
reference:
1. How to use the Jetson TK1 as a remote development environment for CUDA
2. Remote application development using NVIDIA® Nsight™ Eclipse Edition
3. NVIDIA CUDA GETTING STARTED GUIDE FOR LINUX v6.5
4. CUDA 6.0 in Ubuntu 14.04.1 LTS - NVIDIA Developer Forums
5. CUDA Toolkit 6.5
6. CUDA Toolkit Documentation v7.0

A. 在 Local (Mac) 上安裝 CUDA toolkit
     1. 安裝 Xcode

     2. 安裝 Xcode Command Line Tools:
         $ xcode-select --install

     3. 到 CUDA Toolkit 6.5 下載 CUDA Toolkit 6.5 for Mac OSX
         版本: cuda_6.5.14_mac_64.pkg  (latest 64-bit CUDA 6.5 package)

     4. 點二下安裝, 其中 CUDA Driver 在 Local 處可以不需要安裝, 除非要執行 CUDA-Z.

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

B. 在 Remote (Jetson TK1) 上安裝 CUDA toolkit
     1. Verify You Have a CUDA-Capable GPU
         $ lspci | grep -i nvidia
         00:00.0 PCI bridge: NVIDIA Corporation Device 0e13 (rev a1)

     2. 到 CUDA Toolkit 6.5 下載 CUDA Toolkit 6.5 for Linux ARM
         版本: ONLY to be used for Jetson TK1
         cuda-repo-l4t-r21.2-6-5-prod_6.5-34_armhf.debf (L4T 21.2 DEB)
         (將檔案傳送到 remote 的 /home/ubuntu/RD/software 目錄下)

     3. 安裝:
          $ cd /home/ubuntu/RD/software
          $ sudo dpkg -i cuda-repo-l4t-r21.2-6-5-prod_6.5-34_armhf.deb
          $ sudo apt-get update
          $ sudo apt-get install cuda-toolkit-6-5

     4. Add the user to the video group
          $ sudo usermod -a -G video ubuntu

     5. Environment Setup ...
          To change the environment variables for 32-bit ARM operating systems:
          $ export PATH=/usr/local/cuda-6.5/bin:$PATH
          $ export LD_LIBRARY_PATH=/usr/local/cuda-6.5/lib:$LD_LIBRARY_PATH

     6. check the L4T version with the following command
         $ head -1 /etc/nv_tegra_release
            # R21 (release), REVISION: 4.0, GCID: 5650832, BOARD: ardbeg, EABI: hard, DATE: Thu Jun 25 22:38:59 UTC 2015

     7. install g++-4.6
         $ sudo apt-get install g++-4.6

     8. (Optional) Install Writable Samples
          $ cd /usr/local/cuda-6.5/bin
          $ cuda-install-samples-6.5.sh /home/ubuntu/RD/projects/
          $ cd NVIDIA_CUDA-6.5_Samples
          $ make

          The resulting binaries will be placed under:
          /home/ubuntu/RD/projects/NVIDIA_CUDA-6.5_Samples/bin

     9. After restart the system, Finally start one of the samples! 
         $ cd /home/ubuntu/RD/projects/NVIDIA_CUDA-6.5_Samples/bin
         $ cd armv7l/linux/release/gnueabihf
         $ ./deviceQuery

         $ ./bandwidthTest

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

C. Configure Git
       To synchronize CUDA projects between local(Mac) and target(Jetson TK1) systems,
       you need to configure git on both the local and the target systems using these
       commands:

       $ git config --global user.email "lanli0210@gmail.com"
       $ git config --global user.name "lanli"

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

D.Create nsight project on local(Mac)
     1. 開啟 nsight:
         YourMac > Developer/NVIDIA/CUDA-6.5/libnsight/nsight.app

     2. Select a workspace


     3. Create a new project
         File > New > CUDA C/C++ Project

     4. Project type: Empty Project

     5. Basic settings:

     6. Target system > Manage... > Remote Connections

     7. Configure CUDA Toolkit

     8. Target System(Project Path)

     9. remove the Local System

    結果:

     10. 完成:

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

E. Adding the source code for CUDA project
     1. File > New > Source File

     2. Source File Configure (just use a template)

     3. 完成:

     4. 按下 "編譯"

       p.s. Note that the build command is /usr/local/cuda-6.5/bin/nvcc,
              which is the compiler on the Jetson, not on the Macbook.

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

F. Debugging the code on CUDA project
     1. Debug Configurations

     2. Debugging ...

         p.s. By default the debugger will break in main() – this can be switched off
                in the debug configuration.