The big package clean up: Part 1

[pearce8]

Thanks @scheibelp on starting the taxonomy!

• "local" = benchpark's package repo
• "upstream" / "remote" = the builtin package repo
• "remote definition, not inherited" = we should try to start inheriting the upstream
• "submodules=False" = we have to keep these per version we want to use, upstream the rest.

High priority

• @rfhaque amg2023 please upstream. We have a local definition, not inherited. Is there an upstreamed package? local definition looks "mostly" definitive.
• @rfhaque branson will need to upstream. We do not care about the CPU version, leave CPU portion of the package as is. We have remote definition, not inherited. this one is tricky because e.g. it removes a dep on boost. boost isn't used in the newer versions and that potentially simplifies things for benchpark. OTOH, if benchpark only uses later versions of branson (OP: we only need the newest version), we could just reinsert the upstream constraint with the knowledge that we will never have to contend with the complexities it introduces. it's possible that the primary semantic difference is the addition of caliper/adiak deps

Second priority

• affinity: remote definition, not inherited. significant logic difference. all local changes appear unnecessary (this could potentially be tested by copying the upstream into local).
• caliper: inherited. small changes. BUT. one change is related to cray-mpich-gtl: cannot be upstreamed. One change is a papi constraint that can be upstreamed

Third priority

• babelstream: remote definition, not inherited. these are too large for me to diff quickly, the remote does not have a caliper dependency. that may be the primary difference and we are actually close to being able to use upstream

Leave as is for now

• ad: local definition, no upstream. could be upstreamed 100% no issue.
• commbench: local definition, no upstream. definition has some chaff (commented out bits that need cleanup) but the package could generally be upstreamed with no issue. Low priority.

Done

• amg2023 cuda+allow-unsupported-compilers
• remove the local copy. blt: inherited. we only add a version here. we can delete local instance. the spack-packages commit we use has this version now. Removing local copies of BLT, Camp, Chai packages #1373
• branson cuda+allow-unsupported-compilers
• camp change to using develop, will need to switch to using the next release. Right now we inherit with small changes. new version can be removed if we update to spack-packages >= Feb. 25th. I think we could potentially delete local camp definition. Removing local copies of BLT, Camp, Chai packages #1373
• camp +allow-unsupported-compilers
• move to develop. chai: inherited. small changes. same unsupported-compilers issue as amg2023 and camp. new version: we can remove this if we use spack-packages >= Feb. 5th. with these two changes, we can just delete local version Removing local copies of BLT, Camp, Chai packages #1373
• chai +allow-unsupported-compilers