Profiling can't be disabled in CMake, and a parallel_for is missing for tasks (reported by Kyungjoo Kim)

[kyungjoo-kim]

When I turn on both Kokkos task dag (a task use team interface inside) and profiling, I got the following error.

/ascldap/users/kyukim/Work/lib/trilinos/master/packages/kokkos/core/src/Kokkos_Parallel.hpp(182):
error: class "Kokkos::Impl::ThreadVectorRangeBoundariesStruct<int, Kokkos::Impl::TaskExec<Kokkos::Cuda>>" has no member "work_tag"

This is guarded by profiling macro. It seems that something is not defined with scheduler

#if defined(KOKKOS_ENABLE_PROFILING)
  uint64_t kpID = 0;
  if(Kokkos::Profiling::profileLibraryLoaded()) {
    Kokkos::Impl::ParallelConstructName<FunctorType, typename ExecPolicy::work_tag> name(str);
    Kokkos::Profiling::beginParallelFor(name.get(), 0, &kpID);
  }
#endif

[kyungjoo-kim]

How can I disable profiling ?

I tried to give -D KOKKOS_ENABLE_PROFILING:BOOL=OFF. It does not filter out the profiling code block. In the KokkosCore_Config.h the macro is still defined.

[kyungjoo-kim]

I disable

-D KOKKOS_ENABLE_PROFILING_INTERNAL:BOOL=OFF
-D KOKKOS_ENABLE_PROFILING:BOOL=OFF

I still see the following. Any clue ?

/* ---------------------------------------------
Makefile constructed configuration:
Thu Jan 18 14:35:43 MST 2018
----------------------------------------------*/
#if !defined(KOKKOS_MACROS_HPP) || defined(KOKKOS_CORE_CONFIG_H)
#error "Do not include KokkosCore_config.h directly; include Kokkos_Macros.hpp instead."
#else
#define KOKKOS_CORE_CONFIG_H
#endif
/* Execution Spaces */
#define KOKKOS_HAVE_CUDA
#define KOKKOS_HAVE_OPENMP
#define KOKKOS_HAVE_SERIAL
#ifndef __CUDA_ARCH__
#define KOKKOS_USE_ISA_POWERPCLE
#endif
/* General Settings */
#define KOKKOS_HAVE_CXX11
#define KOKKOS_ENABLE_PROFILING
/* Optimization Settings */
/* Cuda Settings */
#define KOKKOS_CUDA_USE_RELOCATABLE_DEVICE_CODE
#define KOKKOS_CUDA_USE_LAMBDA
#define KOKKOS_ARCH_POWER8
#define KOKKOS_ARCH_PASCAL
#define KOKKOS_ARCH_PASCAL60

[ibaned]

sub-task: figure out why profiling is not properly disabled through TriBITS

[ibaned]

sub-task 2: compile failure with RDC + task thread vector policy, add unit test

[ibaned]

Pull request #1352 should get -D KOKKOS_ENABLE_PROFILING:BOOL=OFF to work again

[ibaned]

Pull request #1357 will fix the missing parallel_for API, with is the only missing piece to resolve this issue.