I have .cu and their generated .ptx files under source control. And I build these .cu files from within Visual Studio using a Custom Build Rule. I don’t want to build cuda files that have a read-only generated .ptx file ( or somehow skip read-only cuda files ).
In the Cuda 2.3 toolkit, when nvcc would compile a .cu file and the generated output file ( in my case a .ptx ) already existed and was read-only, there would be an error printed but it would not report the error to Visual
Studio. So even though the compilation would technically fail, it would still succeed in Visual Studio, which is what I want. But it seems as with the Cuda 3.0 toolkit, this error is reported, and the build fails.
Is there any way to skip read-only files that are build with a custom build rule or skip cuda files that have a read-only output file? Or somehow suppress the “nvopencc INTERNAL ERROR: cannot unlink temp file output.ptx” error from nvopencc?