coll: Collective Selection Summary Printout

[yfguo]

Pull Request Description

This is the first PR for the collective selection refactoring. This PR focus on providing better way to print the tree and summary of the loaded collective selection logic.

This PR merges the collective algorithm enums for MPIR, CH4, POSIX and OFI to once place. This allows us to print algorithm names using the string constants added in this PR. This change enables us to use the existing tree printing function to output meaningful info about the loaded tree.

This PR creates a new function to print the collective selection summarization. This format organizes the conditions by individual algorithm (see examples in the next comment). Both tree and summary report format are controlled by a new CVAR MPIR_CVAR_COLLECTIVE_SELECTION_REPORT={none, tree, summary, all}.

Author Checklist

• Provide Description
  Particularly focus on why, not what. Reference background, issues, test failures, xfail entries, etc.
• Commits Follow Good Practice
  Commits are self-contained and do not do two things at once.
  Commit message is of the form: module: short description
  Commit message explains what's in the commit.
• Passes All Tests
  Whitespace checker. Warnings test. Additional tests via comments.
• Contribution Agreement
  For non-Argonne authors, check contribution agreement.
  If necessary, request an explicit comment from your companies PR approval manager.

[yfguo]

Here is an example of the output.

Loaded Collective Selection Tree from /Users/yguo/tuning/MPIR_Coll_tuning.json:
====================================
Processed Collective Selection Tree:
collective: bcast
  intra_comm
    comm_size <= 1
      avg_msg_size <= 0
        Algorithm: MPIR_Bcast_intra_binomial

      avg_msg_size <= 128
        Algorithm: MPIR_Bcast_intra_tree

      avg_msg_size <= 8192
        Algorithm: MPIR_Bcast_intra_tree

      avg_msg_size <= 32768
        Algorithm: MPIR_Bcast_intra_tree

      avg_msg_size <= 262144
        Algorithm: MPIR_Bcast_intra_pipelined_tree

      any
        Algorithm: MPIR_Bcast_intra_scatter_recursive_doubling_allgather

    comm_size <= 8
      avg_msg_size <= 0
        Algorithm: MPIR_Bcast_intra_binomial

      avg_msg_size <= 8192
        Algorithm: MPIR_Bcast_intra_tree

      avg_msg_size <= 32768
        Algorithm: MPIR_Bcast_intra_tree

      avg_msg_size <= 65536
        Algorithm: MPIR_Bcast_intra_tree

      avg_msg_size <= 131072
        Algorithm: MPIR_Bcast_intra_binomial

      any
        Algorithm: MPIR_Bcast_intra_scatter_recursive_doubling_allgather

  
......

==========================================
Summary of rules per collective algorithm:
MPIR_Allgather_intra_brucks
   >>> intra_comm >>> any >>> total_msg_size < 81920
MPIR_Allgather_intra_recursive_doubling
   >>> intra_comm >>> comm_size is power-of-two >>> total_msg_size < 524288
MPIR_Allgather_intra_ring
   >>> intra_comm >>> comm_size is power-of-two >>> any
   >>> intra_comm >>> any >>> any
MPIR_Allgather_inter_local_gather_remote_bcast
   >>> inter_comm
MPIR_Allgatherv_intra_brucks
   >>> intra_comm >>> any >>> total_msg_size < 81920
MPIR_Allgatherv_intra_recursive_doubling
   >>> intra_comm >>> comm_size is power-of-two >>> total_msg_size < 524288
MPIR_Allgatherv_intra_ring
   >>> intra_comm >>> comm_size is power-of-two >>> any
   >>> intra_comm >>> any >>> any
MPIR_Allgatherv_inter_remote_gather_local_bcast
   >>> inter_comm
MPIR_Allreduce_intra_recursive_doubling
   >>> intra_comm >>> comm_size <= 1 >>> avg_msg_size <= 16 >>> operation is not commutative
   >>> intra_comm >>> comm_size <= 1 >>> avg_msg_size <= 64 >>> operation is not commutative
   >>> intra_comm >>> comm_size <= 1 >>> avg_msg_size <= 2048 >>> operation is not commutative
   >>> intra_comm >>> comm_size <= 1 >>> avg_msg_size <= 16384 >>> operation is not commutative
   >>> intra_comm >>> comm_size <= 1 >>> any >>> built-in operators
   >>> intra_comm >>> comm_size <= 8 >>> avg_msg_size <= 32 >>> operation is not commutative
   >>> intra_comm >>> comm_size <= 8 >>> avg_msg_size <= 64 >>> operation is not commutative
......

[hzhou]

Maybe a separate CVAR? It is too much output to fit into debug summary.

[mjwilkins18]

Commenting to record in-person discussion; It would be great to have value ranges in the per-algorithm summary to see when each rule actually applies

[yfguo]

test: mpich/ch4/ofi

[yfguo]

test: mpich/ch4/ofi

[yfguo]

test: mpich/ch4/ofi

[yfguo]

test:mpich/ch4/ofi

[hzhou]

Please add commit messages

[yfguo]

From offline discussion, I need to split the optimization changes to a separate PR and only keep the print out one here.

[yfguo]

PR updates with only the changes to print of the collective selection.