Where can I find the code for sucpp in unifrac version 1.0.0

[tillea]

Hi,
I intend to upgrade the Debian packaged unifrac to version 1.0.0. Unfortunately the code that used to be in dir sucpp in version 0.10.0 vanished. Where can I find this code?
Kind regards, Andreas.

[wasade]

Hi @tillea, the code was pulled into github.com/biocore/unifrac-binaries/.

[tillea]

Am Thu, Feb 17, 2022 at 12:35:13PM -0800 schrieb Daniel McDonald:

Hi @tillea, the code was pulled into github.com/biocore/unifrac-binaries/.

Thanks a lot for the extraordinary fast and helpful response, Andreas.

[sfiligoi]

Hi @tillea .
If at all possible, please follow the build instructions in
https://github.com/biocore/unifrac-binaries/blob/main/.github/workflows/main.yml

It relies on the NVIDIA HPC SDK compiler (which internally relies on gcc) to produce an executable that will use any NVIDIA GPU it finds on the system.
Note that there is no runtime dependency on the NVIDIA HPC SDK, just a compile one.

While compiling with gcc still works (like it did for 0.10.0), the GPU code path will only be compiled if the NVIDIA HPC SDK is used as a compiler.

[tillea]

Hi Igor, Am Thu, Feb 17, 2022 at 01:00:44PM -0800 schrieb Igor Sfiligoi:

If at all possible, please follow the build instructions in https://github.com/biocore/unifrac-binaries/blob/main/.github/workflows/main.yml It relies on the NVIDIA HPC SDK compiler (which internally relies on gcc) to produce an executable that will use any NVIDIA GPU it finds on the system. Note that there is no runtime dependency on the NVIDIA HPC SDK, just a compile one.

Unfortunately that's not possible in Debian main. Since cuda is considered non-free we need to build without it (but we might consider an extra package unifrac-binaries-cuda which can end up in contrib. This is used in some other packages as well.

While compiling with gcc still works (like it did for 0.10.0), the GPU code path will only be compiled if the NVIDIA HPC SDK is used as a compiler.

I think NVIDIA HPC SDK is not available in Debian so this is no option unfortunately. Thanks a lot for your hints anyway. I'll add this to the docs to inform users. Kind regards, Andreas.

[sfiligoi]

Hi @tillea .
You don't need cuda or the NVIDIA HPC SDK from debian.
You download it during compile time and delete it immediately after the compilation is completed.

The NVIDIA license permits such use; we verified with them.
This is the same process we use with bioconda.

PS: Indeed, you are not allowed to redistribute either CUDA or the NVIDIA HPC SDK themselves, but you are allowed to distribute any binary produced by those tools.

[tillea]

Hi Igor, Am Thu, Feb 17, 2022 at 01:58:04PM -0800 schrieb Igor Sfiligoi:

You don't need cuda or the NVIDIA HPC SDK from debian. You download it during compile time and delete it immediately after the compilation is completed.

Exactly this is not possible. Debian package builds need to be possible "on an isolated island with a Debian mirror only". No network access is permitted.

The NVIDIA license permits such use; we verified with them. This is the same process we use with bioconda.

The rules for conda builds are less strict in that aspect. Kind regards, Andreas.

[wasade]

@tillea is this still an issue? If so I wonder then if the debian package should be limited to a CPU build which does not utilize the NVIDIA toolkit?