Getting started - CUDA-x86 - Windows compilation

Hi!

I’m finding ample documentation/samples in getting started with the Fortran build and compiling under linux, meanwhile while trying to compile C for CUDA under windows with visual studio I’m not finding any good documentation/samples.


From here: [1] I gathered that I should be using the PGCPP compiler, here is a command line test:

C:\Program Files\PGI\win64\14.7\bin>pgcpp -c -o C:\tmp\HellowWorldTest.cu
pgcpp-Warning-No files to process

C:\Program Files\PGI\win64\14.7\bin>

As stated here [2] the packaged I downloaded ( PGI accelerator C\C++ workstation) should contain a “CUDA X86 SDK” folder, which it does not.

For example I should have (for my 64 bit build):
C:\Program Files\PGI\win64\2011\cuda\CUDA X86 SDK

But I simply have:
C:\Program Files\PGI\win64\2014\cuda\5.5
C:\Program Files\PGI\win64\2014\cuda\6.0


It would almost seem like I have the wrong package?

How to get started with a simple application under windows?
How to compile a simple application from the command-line?
How to setup visual studio to use PGI compiler for compilation? (ie are there any available build customizations?)

Are there actually any code samples for CUDA x86 to get started with as stated in[2]?

Thank you!
Jimy

[1] http://www.pgroup.com/lit/articles/insider/v3n2a1.htm
[2] https://www.pgroup.com/lit/articles/insider/v3n2a3.htm

Hi Jimy,

This particular error is because you’re using the lower case “-o” instead of the upper case “-O”. “-O” stands for the optimization level to use while "-o " says to output to a file. So in this case you’re telling the compiler to output to a file named “C:\tmp\HellowWorldTest.cu” but have no input file.

However, we don’t support CUDA-X86 on Windows and why you can’t find documentation. Some of our documentation said we did, but that was in error.

NVIDIA’s CUDA C/C++ compiler, nvcc, is supported on Windows as well as PGI’s CUDA Fortran Compiler. Though both target NVIDIA GPUs and not x86.

We do have a Visual Studio product, PVF, but it’s Fortran only.

  • Mat

Ah right, silly mistake on my part!

However, we don’t support CUDA-X86 on Windows and why you can’t find documentation. Some of our documentation said we did, but that was in error.

NVIDIA’s CUDA C/C++ compiler, nvcc, is supported on Windows as well as PGI’s CUDA Fortran Compiler. Though both target NVIDIA GPUs and not x86.

We do have a Visual Studio product, PVF, but it’s Fortran only.

  • Mat

That’s very strange Mat. Just by looking at your main introduction page [1] FAQ:


Q: Which operating systems does PGI support with its CUDA-x86 compilers?

A: Linux, Windows and OS X.

Also downloading the PGI accelerator thats being done specifically for 64-bit windows.

This is all very misleading if indeed you are corecct.

Regards
Jim

[1]https://www.pgroup.com/resources/cuda-x86.htm

Hi Jim,

We do apologize for the wrong information on the web page. It was supposed to have been updated but was missed. Windows had been supported in previous releases but we’ve had issues with it in the current release.

Also downloading the PGI accelerator thats being done specifically for 64-bit windows

Sorry, I’m not clear on what you mean here. The Accelerator products all work on Win64, it’s just the CUDA-X86 product (i.e. CUDA targeting a multi-core CPU) that we’re having problems with. We do hope to have the issues resolved in the near future.

Thanks,
Mat

Ok, so you do plan on supporting win64 in the future?

Are any of the older releases available for download on trial/purchase?

Ok, so you do plan on supporting win64 in the future?

I’m not sure.

Are any of the older releases available for download on trial/purchase?

Archived releases can be found at: http://www.pgroup.com/support/download_releases.php

Purchasing a current license entitles you to all preceding releases.

  • Mat