Name all parallel kernels and regions

[crtrott]

I think we should add labels to all our parallel calls and also add regions for top level calls.
Consider our unit tests, right now simple kernel timer will print this:

Regions: 

                                                               KokkosBlas::Test::gemm_complex_double (Region)          12.53954            1        12.53954  53.671  15.326
                                                                       KokkosBlas::Test::gemm_double (Region)           2.87859            1         2.87859  12.321   3.518
                                                                               KokkosBlas::gemm[ETI] (Region)           1.83371           64         0.02865   7.849   2.241

-------------------------------------------------------------------------
Kernels: 

                                                                       KokkosBlas::Test::VanillaGEMM (ParFor)          13.31410           80         0.16643  56.986  16.273
                                                                        Kokkos::View::initialization (ParFor)           2.19648        37758         0.00006   9.401   2.685
                                                                       KokkosBlas::gemv[SingleLevel] (ParFor)           0.70460           48         0.01468   3.016   0.861
                                                                                KokkosBlas::gemm[NN] (ParFor)           0.45747           16         0.02859   1.958   0.559
                                                                                KokkosBlas::gemm[NC] (ParFor)           0.38384            8         0.04798   1.643   0.469
                                                                                KokkosBlas::gemm[CN] (ParFor)           0.38346            8         0.04793   1.641   0.469
                                                                                KokkosBlas::gemm[CC] (ParFor)           0.37958            8         0.04745   1.625   0.464
N12KokkosSparse4Impl11GaussSeidelIN13KokkosKernels12Experimental19KokkosKernelsHandleIKiS5_KN6Kokkos7complexIdEENS6_6OpenMPENS6_9HostSpaceESB_EENS6_4ViewIPS5_JNS6_10LayoutLeftENS6_6DeviceISA_SB_EENS6_12MemoryTraitsILj1EEEEEESK_NSD_IPS9_JSF_SH_SJ_EEEE4PSGSE (ParFor)           0.33012        29000         0.00001   1.413   0.403
N12KokkosSparse4Impl11GaussSeidelIN13KokkosKernels12Experimental19KokkosKernelsHandleIKmKiKN6Kokkos7complexIdEENS7_6OpenMPENS7_9HostSpaceESC_EENS7_4ViewIPS5_JNS7_10LayoutLeftENS7_6DeviceISB_SC_EENS7_12MemoryTraitsILj1EEEEEENSE_IPS6_JSG_SI_SK_EEENSE_IPSA_JSG_SI_SK_EEEE4PSGSE (ParFor)           0.32352        28800         0.00001   1.385   0.395
N12KokkosSparse4Impl11GaussSeidelIN13KokkosKernels12Experimental19KokkosKernelsHandleIKiKlKN6Kokkos7complexIdEENS7_6OpenMPENS7_9HostSpaceESC_EENS7_4ViewIPS5_JNS7_10LayoutLeftENS7_6DeviceISB_SC_EENS7_12MemoryTraitsILj1EEEEEENSE_IPS6_JSG_SI_SK_EEENSE_IPSA_JSG_SI_SK_EEEE4PSGSE (ParFor)           0.32184        28800         0.00001   1.378   0.393

Then another 820 lines of mangled CType names and then this:

                                                              17ArithTraitsTesterIdN6Kokkos6OpenMPEE (ParRed)           0.00001            1         0.00001   0.000   0.000
                                               17ArithTraitsTesterIN6Kokkos7complexIdEENS0_6OpenMPEE (ParRed)           0.00001            1         0.00001   0.000   0.000
                                                              17ArithTraitsTesterIxN6Kokkos6OpenMPEE (ParRed)           0.00001            1         0.00001   0.000   0.000
                                                              17ArithTraitsTesterIyN6Kokkos6OpenMPEE (ParRed)           0.00001            1         0.00001   0.000   0.000
                                                              17ArithTraitsTesterIfN6Kokkos6OpenMPEE (ParRed)           0.00001            1         0.00001   0.000   0.000
                                               17ArithTraitsTesterIN6Kokkos7complexIfEENS0_6OpenMPEE (ParRed)           0.00001            1         0.00001   0.000   0.000

-------------------------------------------------------------------------
Summary:

Total Execution Time (incl. Kokkos + Non-Kokkos:                   81.81700 seconds
Total Time in Kokkos kernels:                                      23.36379 seconds
   -> Time outside Kokkos kernels:                                 58.45320 seconds
   -> Percentage in Kokkos kernels:                                   28.56 %
Total Calls to Kokkos Kernels:                                       298588

-------------------------------------------------------------------------

What I propose is the following:
Mark Kernels with something like

Namespace::TopLevelFunction::Subfunction[Options]

For example (there is no subfunction for gemm):

KokkosBlas::gemm[NN]

Also we should mark regions and those should add the info whether its a TPL call, ETI call or noETI call:

KokkosBlas::gemm[TPL_BLAS]
KokkosBlas::gemm[ETI]
KokkosBlas::gemm[noETI]

Any thoughts?

[ndellingwood]

Cross-reference #239

[kyungjoo-kim]

See PR #359

[ndellingwood]

PR #359 merged with added labels.