Add compiler output to query_config

[rljacob]

Currently, query_config --machine gives this output:

[jacob@chrlogin2 scripts]$ ./query_config --machines frontier
Machine(s)

  frontier : Frontier exascale supercomputer at ORNL. 9408 nodes, Node: 4 AMD MI250X GPUs (2 GCDs) ~ 8 GPUs, 512 GB HDB2E, AMD EPYC 64 cores, 512GB DDR4
      os              Linux
      compilers       gnu,amdclang,crayclang,gnugpu,amdclanggpu,crayclanggpu
      mpilibs         ['mpich']
      pes/node        56
      max_tasks/node  56
      max_gpus/node  0

We would like to add an ability to git more info on the compiler. something like
./query_config --machines frontier --compiler gnugpu

The info to print out would be the info contained in these xml blocks (for the case of "gnugpu" on E3SM's "frontier" machine):

      <modules compiler="gnu.*">
        <command name="reset"></command>
        <command name="switch">Core Core/24.00</command>
        <command name="switch">PrgEnv-cray PrgEnv-gnu/8.3.3</command>
        <command name="switch">gcc gcc/12.2.0</command>
      </modules>
      <modules compiler="gnugpu">
        <command name="load">craype-accel-amd-gfx90a</command>
        <command name="load">rocm/5.4.0</command>
      </modules>
    <environment_variables compiler=".*gpu.*">
      <env name="NTASKS_PER_GPU">--ntasks-per-gpu=$SHELL{echo "`./xmlquery --value MAX_MPITASKS_PER_NODE`/8"|bc}</env>
      <env name="GPU_BIND_ARGS">--gpu-bind=closest</env>
      <env name="PNETCDF_HINTS">romio_cb_read=disable</env>
      <env name="MPICH_GPU_SUPPORT_ENABLED">0</env>
      <env name="MPICH_CXX">$SHELL{which hipcc}</env>
    </environment_variables>
    <environment_variables compiler=".*gpu.*" DEBUG="TRUE">
      <env name="AMD_LOG_LEVEL">10</env>
      <env name="CRAY_ACC_DEBUG">3</env>
    </environment_variables>
    <environment_variables compiler="gnu.*" mpilib="mpich">
      <env name="ADIOS2_ROOT">$SHELL{if [ -z "$ADIOS2_ROOT" ]; then echo /lustre/orion/cli115/world-shared/frontier/3rdparty/adios2/2.9.1/cray-mpich-8.1.23/gcc-11.2.0; else echo "$ADIOS2_ROOT"; fi}</env>
    </environment_variables>

The printout might look like:

  frontier : compiler: gnugpu
Module commands:
  reset
  switch Core Core/24.00
  switch PrgEnv-cray PrgEnv-gnu/8.3.3
  switch gcc gcc/12.2.0
  load craype-accel-amd-gfx90a
  load rocm/5.4.0
Environment variables:
  NTASKS_PER_GPU:  -ntasks-per-gpu=$SHELL{echo "`./xmlquery --value MAX_MPITASKS_PER_NODE`/8"|bc}
  GPU_BIND_ARGS: --gpu-bind=closest
  PNETCDF_HINTS: romio_cb_read=disable
  MPICH_GPU_SUPPORT_ENABLED:
  MPICH_CXX:  $SHELL{which hipcc}
  (with DEBUG="TRUE") : 
      AMD_LOG_LEVEL: 10
      CRAY_ACC_DEBUG: 3
  (with mpilib="mpich"):
     ADIOS2_ROOT:  $SHELL{if [ -z "$ADIOS2_ROOT" ]; then echo /lustre/orion/cli115/world-shared/frontier/3rdparty/adios2/2.9.1/cray-mpich-8.1.23/gcc-11.2.0; else echo "$ADIOS2_ROOT"; fi}

The reason is that we are having a hard time getting all the relevant info in to the compiler name. See E3SM-Project/E3SM#6773 for background.

[rljacob]

@bartgol what do you think of the above?

[rljacob]

For a setting like NTASKS_PER_GPU, instead of literally printing out "$SHELL{echo "./xmlquery --value MAX_MPITASKS_PER_NODE/8"|bc}", print the resulting value of that shell command.

[rljacob]

The "--compiler" argument must be accompanied by a "--machine" argument. Its an error to just use "--compiler".

[bartgol]

So you would print the content of the modules and environment_variables sections that match the compiler? Seems reasonable, but there may be a few details to clarify:

• what do we do with sections that also have other filters, such as "mpilib" or "BUILD_THREADED"? One option is to print them all (but highlight that they depend on inner options). Not printing them altogether may be too restrictive. Another option is to allow the user to specify additional filters, such as --filters mpilib=mpich BUILD_THREADED=false. It should not be hard to parse these.
• so doing ./query_config --machine frontier will print LESS stuff than ./query_config --machine frontier --compiler crayclang? This is somewhat the opposite of what one could expect, since less filters should give "more matches". However, it makes sense to not print lots of configs without a compiler. That's why, if the compiler is not specified, I would print a line like

Modules commands:
     UNKNOWN
     requires '--compiler <compiler_name>' option
Environment variables:
    UNKNOWN
    requires '--compiler <compiler_name>' option

[rljacob]

My example above shows how the DEBUG=TRUE options, for example, would be indicated in the output.

Its a command line argument, not a filter. So more arguments equals more detail. I don't think that's counter-intuitive.

[bartgol]

Ah, I somehow missed the last few lines. So the options that depend on further filters are printed along with what option is needed to get them. Yes, I think that's a good format.