Update minimum compiler versions + clean up

[masterleinad]

Fixes #5285. This excludes ROCm/HIPCC and icpx. We'll update them later.

[masterleinad]

This corresponds to #5311.

[PhilMiller]

I'll check the Intel version number stuff

[PhilMiller]

Intel 19.0.5:

[pbmille@kokkos-dev-2 ~]$ module load sems-intel/19.0.5

$ which icpc
/projects/sems/install/rhel7-x86_64/sems-compilers/tpl/intel/19.0.5/gcc/8.3.0/base/u4jh4al/compilers_and_libraries_2019.5.281/linux/bin/intel64/icpc

$ icpc --version
icpc (ICC) 19.0.5.281 20190815
Copyright (C) 1985-2019 Intel Corporation.  All rights reserved.

$ cmake -DCMAKE_CXX_COMPILER=icpc ~/repos/kokkos 
-- Setting default Kokkos CXX standard to 14
-- The CXX compiler identification is Intel 19.0.5.20190815
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: /projects/sems/install/rhel7-x86_64/sems-compilers/tpl/intel/19.0.5/gcc/8.3.0/base/u4jh4al/compilers_and_libraries_2019.5.281/linux/bin/intel64/icpc - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Setting build type to 'RelWithDebInfo' as none was specified.
-- The project name is: Kokkos
-- SERIAL backend is being turned on to ensure there is at least one Host space. To change this, you must enable another host execution space and configure with -DKokkos_ENABLE_SERIAL=OFF or change CMakeCache.txt
-- Using -std=gnu++14 for C++14 extensions as feature
-- Built-in Execution Spaces:
--     Device Parallel: NoTypeDefined
--     Host Parallel: NoTypeDefined
--       Host Serial: SERIAL
-- 
-- Architectures:
-- Found TPLLIBDL: /usr/include  
-- Using internal desul_atomics copy
-- Kokkos Devices: SERIAL, Kokkos Backends: SERIAL
-- Configuring done
-- Generating done
-- Build files have been written to: /ascldap/users/pbmille/build/2022-08-11-icpc19

$ cmake -DCMAKE_CXX_STANDARD=17 -DCMAKE_CXX_COMPILER=icpc ~/repos/kokkos 
-- Setting default Kokkos CXX standard to 17
-- The CXX compiler identification is Intel 19.0.5.20190815
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: /projects/sems/install/rhel7-x86_64/sems-compilers/tpl/intel/19.0.5/gcc/8.3.0/base/u4jh4al/compilers_and_libraries_2019.5.281/linux/bin/intel64/icpc - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Setting build type to 'RelWithDebInfo' as none was specified.
-- The project name is: Kokkos
-- SERIAL backend is being turned on to ensure there is at least one Host space. To change this, you must enable another host execution space and configure with -DKokkos_ENABLE_SERIAL=OFF or change CMakeCache.txt
-- Using -std=gnu++17 for C++17 extensions as feature
-- Built-in Execution Spaces:
--     Device Parallel: NoTypeDefined
--     Host Parallel: NoTypeDefined
--       Host Serial: SERIAL
-- 
-- Architectures:
-- Found TPLLIBDL: /usr/include  
-- Using internal desul_atomics copy
-- Kokkos Devices: SERIAL, Kokkos Backends: SERIAL
-- Configuring done
-- Generating done
-- Build files have been written to: /ascldap/users/pbmille/build/2022-08-11-icpc19

$ cat intel.cc 
#include <iostream>

int main() {
  std::cout << '"' << __INTEL_COMPILER << '"' << std::endl;
  return 0;
}

$ make CXX=icpc intel
icpc     intel.cc   -o intel

$ ./intel 
"1900"

[masterleinad]

Intel 19.0.5:

Thanks!

[masterleinad]

The only failure is

: [ RUN      ] default_exec.overlap_mdrange_policy
24: 
24: Kokkos::Cuda::initialize WARNING: Cuda is allocating into UVMSpace by default
24:                                   without setting CUDA_MANAGED_FORCE_DEVICE_ALLOC=1 or
24:                                   setting CUDA_VISIBLE_DEVICES.
24:                                   This could on multi GPU systems lead to severe performance"
24:                                   penalties.
24: 
24: Kokkos::Cuda::initialize WARNING: Cuda is allocating into UVMSpace by default
24:                                   without setting CUDA_MANAGED_FORCE_DEVICE_ALLOC=1 or
24:                                   setting CUDA_VISIBLE_DEVICES.
24:                                   This could on multi GPU systems lead to severe performance"
24:                                   penalties.
24: /var/jenkins/workspace/Kokkos/core/perf_test/PerfTest_ExecSpacePartitioning.cpp:395: Failure
24: Expected: (time_end) > (1.5 * time_overlap), actual: 0.00477586 vs 0.00496419
24: [  FAILED  ] default_exec.overlap_mdrange_policy (451 ms)

in CUDA-11.0-NVCC-C++17-RDC.

[dalg24]

The list does not explicitly mention IBM XL
Should we say it is not supported?
Also should we have a flag to turn the error into a warning and continue processing?

[masterleinad]

See 30fc3d6.

[PhilMiller]

Other than Damien's remark about explicitly reporting that IBM XL is no longer supported, this all looks good to me.

We may want to audit the remaining #if compiler version checks throughout the code for similar mistakes of boolean logic, since they seemed to be quite frequent. We may even want to introduce helper macros to encapsulate the correct comparisons, to avoid those mistakes in the future.

[cz4rs]

Looks good to me.

[masterleinad]

Dropping XL is now part of #5349.

[dalg24]

please format this way

ifeq (COND1)
  ...
else ifeq (COND2)
  ...
else ifeq (COND3)
...
else
  ...
endif

to improve readability

[dalg24]

Looks good besides the format fix I asked for