I haven’t attempted it yet but has anyone dared to try the NSight VS plugin with the newest version of Visual Studio? Now that VS supports CMake, I’m finding it hard to want to migrate back to a supported VS version as CMake is the most sane-way I’ve found of managing cross-platform builds of relatively small software projects (like mine is).
My apologies, I should’ve googled this :P
Apparently, CUDA 8 doesn’t support the the newest MSVC so it’s also understandable that the IDE wouldn’t be supported either.
My brain encountered an overflow error when I saw the words “CMake” and “sane” in the same sentence :-)
To be fair, I’ve heard quite a few complaints about it as well. Which as we all know, means it’s being used. It’s not the worst thing in the world.
I’ve also heard good things about Boost.Build. Apparently Boost does everything :P
CMake is pretty black magic only in the sense that you can generate Ninja files, Makefiles and MSBuild solutions all out of the same file.
CMAKE as a ‘meta-build tool’ has its rough corners and its programming language is primitive (but works for small stuff), but you can make your own wrappers around that stuff. It works also finee for bigger projects (e.g. OpenCV with its large set of modules uses CMAKE). Furthermore, platform-specific differences (standard GCC / VS compiler flags you want to set etc.) can be hidden also in your wrappers. And the combination of its diverse buld targets (from VS, Xcode, Unix MakeFile, Code::Blocks project files) and the large set of high-quality Find scripts for common thirdparty libraries (Boost, CUDA, …) makes it very powerful. E.g. for a computer-vision application I was developing and debugging the whole (command-line) application with Visual Studio (which has maybe the best IDE/Debugger which is around) and then compiling it natively and deploying the application on a Raspberry Pi 2 (using the Code::Blocks IDE which is light-weight enough to run natively on the Raspi).
But back to the topic: It would be nice if the next CUDA toolkit would support VS 2017 …
msvc 2017 has been out for what, like a month already? I mean, come on, Nvidia :P
I bet they’re actually very hard at work, transitioning the CUDA toolkit to the newest VS.
they caught vs2015 in a year, why it should be faster this time? taking into account Volta that may have new architecture, i expect that they will release cuda9 preview at the time of Volta announce