py-minkowskiengine: new package (sparse tensor autodiff by Nvidia) (#35302)

* py-minkowskiengine: new package (sparse tensor autodiff by Nvidia)

This python package (with cuda support) provides torch support for sparse
tensors. The `pybind11` headers are not found without the patch to `setup.py`.

* [@spackbot] updating style on behalf of wdconinc

* py-minkowskiengine: depends_on numpy, pybind11 type=link; no patch

* [@spackbot] updating style on behalf of wdconinc

---------

Co-authored-by: wdconinc <wdconinc@users.noreply.github.com>

var/spack/repos/builtin/packages/py-minkowskiengine/package.py

@@ -0,0 +1,32 @@
+# Copyright 2013-2023 Lawrence Livermore National Security, LLC and other
+# Spack Project Developers. See the top-level COPYRIGHT file for details.
+#
+# SPDX-License-Identifier: (Apache-2.0 OR MIT)
+
+from spack.package import *
+
+
+class PyMinkowskiengine(PythonPackage, CudaPackage):
+    """The MinkowskiEngine is an auto-differentiation
+    library for sparse tensors."""
+
+    homepage = "https://nvidia.github.io/MinkowskiEngine/"
+    pypi = "MinkowskiEngine/MinkowskiEngine-0.5.4.tar.gz"
+
+    maintainers = ["wdconinc"]
+
+    version("0.5.4", sha256="b1879c00d0b0b1d30ba622cce239886a7e3c78ee9da1064cdfe2f64c2ab15f94")
+
+    depends_on("py-setuptools", type="build")
+    depends_on("py-pybind11", type="link")
+
+    depends_on("py-numpy", type=("build", "run"))
+    depends_on("py-torch", type=("build", "run"))
+
+    def install_options(self, spec, prefix):
+        options = []
+        if spec.satisfies("+cuda"):
+            options.append("--force_cuda")
+        else:
+            options.append("--cpu_only")
+        return options