We are running two Tesla K20m on a Linux machine (Ubuntu precise). Now I would like to use features only supported by Nsight Visual Studio Edition (especially the ‘Assembly Code Correlation’). My local Windows machine has no CUDA-capable device.
Is it possible to configure Visual Studio for remote debugging on a non-Windows target?
on a Linux target from another Linux host with NSIGHT Eclipse
on a Windows target from another Windows host with NSIGHT Visual Studio?
Does this mean that the NVIDIA Nsight Monitor is only available under Windows and not for example as a service for Linux machine? Furthermore it seems that there is no cuda-gdb client available for Windows machines.
The only way to debug from a Windows host to a Linux target I see would then be to
copy files from the Windows host to the Linux target,
ssh to the Linux machine and start cuda-gdb (not the cuda-gdbserver) there “locally”.
Could I use a virtual machine Linux on top of a Windows CPU to install Eclipse and NSIGHT Eclipse? Would I be able to virtual debug targeting a remote Linux machine from the VM Linux host then?
Hi, sorry for bringing old topic back to live, but I have to ask.
Microsoft introduced support for linux development in visual studio 2017 (or vs2015 update 3 - I am not sure). I have never used this feature before, but from what I can tell, the main idea is to connect to a linux target machine and perform; remote build, remote debugging, etc.
I know that at the time of writing this message, there is no nsight edition for visual studio 2017. But I believe it is currently under development… Therefor, is there Anyone who can tell me if it will be possible to configure Nsight (VS Edition) for remote building / debugging on a linux target?
Thank you in advance.
Edit:
Ok since yesterday I tested this VS’s linux development feature, in its most basic form (makefile project). And it turns out it is possible to build cuda apps on remote linux host, from widnows machine…
I checked this configuration is able to debug host part of the cuda program. Therefore i tried to force VS to call cuda-dbg instead of dbg on the linux side. But it turned out to be not possible - VS does not allow to change debugger.
Well it is possible to build VS plugin, and custom debug-visualizers on VS marketplace are kind of popular, therefor I believe (I am not sure) it could be possible to build cuda-dbg-visualizer, that would look and behave just like visual studio nsight edition.
And because VS can call debugger directly on the target machine, there is no need for cuda-dbg on host, just something that can interpret targets response through ssh.