Allow more recent blas versions in GATK

[edmundmiller]

Getting this build failure https://wave.seqera.io/view/builds/bd-8fb12ec35653a923_12

error    libmamba Could not solve for environment specs
    The following package could not be installed
    └─ gcnvkernel 0.9*  is not installable because it requires
       └─ blas 1.0 mkl, which does not exist (perhaps a missing channel).

So, I'm hoping we can go against the template and loosen the 1.0 version requirement.

---

Please read the guidelines for Bioconda recipes before opening a pull request (PR).

General instructions

• If this PR adds or updates a recipe, use "Add" or "Update" appropriately as the first word in its title.
• New recipes not directly relevant to the biological sciences need to be submitted to the conda-forge channel instead of Bioconda.
• PRs require reviews prior to being merged. Once your PR is passing tests and ready to be merged, please issue the @BiocondaBot please add label command.
• Please post questions on Gitter or ping @bioconda/core in a comment.

Instructions for avoiding API, ABI, and CLI breakage issues

Conda is able to record and lock (a.k.a. pin) dependency versions used at build time of other recipes.
This way, one can avoid that expectations of a downstream recipe with regards to API, ABI, or CLI are violated by later changes in the recipe.
If not already present in the meta.yaml, make sure to specify run_exports (see here for the rationale and comprehensive explanation).
Add a run_exports section like this:

build:
  run_exports:
    - ...

with ... being one of:

Case	run_exports statement
semantic versioning	{{ pin_subpackage("myrecipe", max_pin="x") }}
semantic versioning (0.x.x)	{{ pin_subpackage("myrecipe", max_pin="x.x") }}
known breakage in minor versions	{{ pin_subpackage("myrecipe", max_pin="x.x") }} (in such a case, please add a note that shortly mentions your evidence for that)
known breakage in patch versions	{{ pin_subpackage("myrecipe", max_pin="x.x.x") }} (in such a case, please add a note that shortly mentions your evidence for that)
calendar versioning	{{ pin_subpackage("myrecipe", max_pin=None) }}

while replacing "myrecipe" with either name if a name|lower variable is defined in your recipe or with the lowercase name of the package in quotes.

Bot commands for PR management

Please use the following BiocondaBot commands:

Everyone has access to the following BiocondaBot commands, which can be given in a comment:

@BiocondaBot please update	Merge the master branch into a PR.
@BiocondaBot please add label	Add the please review & merge label.
@BiocondaBot please fetch artifacts	Post links to CI-built packages/containers. You can use this to test packages locally.

Note that the @BiocondaBot please merge command is now depreciated. Please just squash and merge instead.

Also, the bot watches for comments from non-members that include @bioconda/<team> and will automatically re-post them to notify the addressed <team>.

[dslarm]

For linux-aarch64 this needs to point to either Arm Performance Libraries (armpl) or openblas for the blas implementation -as MKL is an x86_64 only library.

I believe something like:

- conda-forge:blas >=1.0=mkl  # [x86_64]
- conda-forge:blas >=1.0=openblas  # [aarch64]

will allow this.
(amended to use openblas - armpl binary is rather out of date, but I need to dig out how this is updated in conda)

[edmundmiller]

@dslarm We're gonna need to bump pytorch as well then right?

[mfansler]

Was just commenting on this elsewhere: #52392 (comment)

Unless the package actually uses MKL-specific accelerations, the dependency should simply be libblas and liblapack. That lets users choose the implementation when creating the environment.

[dslarm]

@dslarm We're gonna need to bump pytorch as well then right?

I'll ask here - it installs OK - but with nomkl conda-forge/noarch::nomkl-1.0-... in the slot where I'd expect MKL to be.

I'm not entirely sure that this matters, I'll ask some colleagues tomorrow whether this is getting the right pytorch libraries as a whole - I believe most of the performance comes from other base math libraries not part of the BLAS part of MKL.