Cross compiling in the board itself

one of the steps to be done to work in the jetson tegra TK1 is to setup Tegra K1 cross compiles to run on x86 Ubuntu 14.04.
however , if I want to work on the board all alone what should I do?
should I setup the ARM CPU to be able to cross-compile and than install the Jetpack on the board OS (L4T) itself ?
if that’s the case than anyone know the steps to do so ?

I’m uncertain as to what your goal is, but the following may be related and help…

JetPack is never installed to the Jetson, and always runs on the x86_64 host. JetPack is essentially a front end to package installs and the flash program. The flash program is an x86_64 binary executable which understands a Jetson when Jetson is in recovery mode. The flash program (a.k.a., driver package plus sample rootfs) and the packages can be installed separately without JetPack. Some of those packages are infrastructure for CUDA or programs running under GPU…some of those packages are for developing programs which run under CUDA or GPU. When you use an Ubuntu host, JetPack saves time because it uses the Ubuntu packaging system…this also means JetPack doesn’t work with other Linux host distributions, e.g., JetPack does not work with Fedora or CentOS, although the flash program itself works with any Linux x86_64 host.

So far as developing and compiling code to run on Jetson, the Jetson itself is a full Ubuntu distribution. You can use standard package tools (e.g., apt or apt-get) to install native compilers and libraries…no cross compile is required in that situation. Jetson TK1 code is very often built directly on the Jetson without cross compile.

Cross compile is possible if you want to develop from the x86_64 host (or any host architecture with a cross compiler available, e.g., a Mac). In that case you need a cross tool chain. The tool chain runs on your desktop but produces code to execute in the ARMv7 architecture. Typically the particular cross tool chain used is a Linaro tool chain.

Sometimes people set up emulators as an aid, e.g., QEMU. This would not be practical for GPU development, but it would work for “ordinary” ARMv7 applications.