# ---------------------------------------------------------------------------
#  Copyright (c) 2013 Kyle Lutz <kyle.r.lutz@gmail.com>
#
#  Distributed under the Boost Software License, Version 1.0
#  See accompanying file LICENSE_1_0.txt or copy at
#  http://www.boost.org/LICENSE_1_0.txt
#
# ---------------------------------------------------------------------------

include_directories(../include)

set(PERF_BOOST_COMPONENTS system timer chrono program_options)

if (${BOOST_COMPUTE_USE_OFFLINE_CACHE})
  set(PERF_BOOST_COMPONENTS ${PERF_BOOST_COMPONENTS} filesystem)
endif()

if(${BOOST_COMPUTE_THREAD_SAFE} AND NOT ${BOOST_COMPUTE_USE_CPP11})
  set(PERF_BOOST_COMPONENTS ${PERF_BOOST_COMPONENTS} thread)
elseif(${BOOST_COMPUTE_HAVE_BOLT} AND ${BOOST_COMPUTE_USE_CPP11})
  set(PERF_BOOST_COMPONENTS ${PERF_BOOST_COMPONENTS} thread)
endif()

if(${BOOST_COMPUTE_HAVE_BOLT} AND ${BOOST_COMPUTE_USE_CPP11})
  set(PERF_BOOST_COMPONENTS ${PERF_BOOST_COMPONENTS} date_time)
endif()

if(PERF_BOOST_COMPONENTS)
  list(REMOVE_DUPLICATES PERF_BOOST_COMPONENTS)
endif()
find_package(Boost 1.54 REQUIRED COMPONENTS ${PERF_BOOST_COMPONENTS})
include_directories(SYSTEM ${Boost_INCLUDE_DIRS})

set(BENCHMARKS
  accumulate
  bernoulli_distribution
  binary_find
  cart_to_polar
  comparison_sort
  copy_if
  copy_to_device
  count
  discrete_distribution
  erase_remove
  exclusive_scan
  fill
  find
  find_end
  includes
  inner_product
  is_permutation
  is_sorted
  max_element
  merge
  next_permutation
  nth_element
  partial_sum
  partition
  partition_point
  prev_permutation
  reverse
  reverse_copy
  rotate
  rotate_copy
  host_sort
  random_number_engine
  reduce_by_key
  saxpy
  search
  search_n
  set_difference
  set_intersection
  set_symmetric_difference
  set_union
  sort
  sort_by_key
  sort_float
  stable_partition
  uniform_int_distribution
  unique
  unique_copy
)

foreach(BENCHMARK ${BENCHMARKS})
  set(PERF_TARGET perf_${BENCHMARK})
  add_executable(${PERF_TARGET} perf_${BENCHMARK}.cpp)
  target_link_libraries(${PERF_TARGET} ${OpenCL_LIBRARIES} ${Boost_LIBRARIES})
endforeach()

# stl benchmarks (for comparison)
set(STL_BENCHMARKS
  stl_accumulate
  stl_count
  stl_find
  stl_find_end
  stl_includes
  stl_inner_product
  stl_max_element
  stl_merge
  stl_next_permutation
  stl_partial_sum
  stl_partition
  stl_prev_permutation
  stl_reverse
  stl_reverse_copy
  stl_rotate
  stl_rotate_copy
  stl_saxpy
  stl_search
  stl_search_n
  stl_set_difference
  stl_set_intersection
  stl_set_symmetric_difference
  stl_set_union
  stl_sort
  stl_stable_partition
  stl_unique
  stl_unique_copy
)

# stl benchmarks which require c++11
if(${BOOST_COMPUTE_USE_CPP11})
  list(APPEND
    STL_BENCHMARKS
    stl_is_permutation
    stl_partition_point
  )
endif()

foreach(BENCHMARK ${STL_BENCHMARKS})
  set(PERF_TARGET perf_${BENCHMARK})
  add_executable(${PERF_TARGET} perf_${BENCHMARK}.cpp)
  target_link_libraries(${PERF_TARGET} ${Boost_LIBRARIES})
endforeach()

# cuda/thrust benchmarks (for comparison)
if(${BOOST_COMPUTE_HAVE_CUDA})
  find_package(CUDA 5.0 REQUIRED)

  set(CUDA_BENCHMARKS
    thrust_accumulate
    thrust_count
    thrust_exclusive_scan
    thrust_find
    thrust_inner_product
    thrust_merge
    thrust_partial_sum
    thrust_partition
    thrust_reduce_by_key
    thrust_reverse
    thrust_reverse_copy
    thrust_rotate
    thrust_saxpy
    thrust_set_difference
    thrust_sort
    thrust_unique
  )

  foreach(BENCHMARK ${CUDA_BENCHMARKS})
    set(PERF_TARGET perf_${BENCHMARK})
    cuda_add_executable(${PERF_TARGET} perf_${BENCHMARK}.cu)
    target_link_libraries(${PERF_TARGET} ${CUDA_LIBRARIES} ${Boost_LIBRARIES})
  endforeach()
endif()

# intel tbb benchmarks (for comparison)
if(${BOOST_COMPUTE_HAVE_TBB})
  find_package(TBB REQUIRED)
  include_directories(SYSTEM ${TBB_INCLUDE_DIRS})

  set(TBB_BENCHMARKS
    tbb_accumulate
    tbb_merge
    tbb_sort
  )

  foreach(BENCHMARK ${TBB_BENCHMARKS})
    set(PERF_TARGET perf_${BENCHMARK})
    add_executable(${PERF_TARGET} perf_${BENCHMARK}.cpp)
    target_link_libraries(${PERF_TARGET} ${TBB_LIBRARIES} ${Boost_LIBRARIES})
  endforeach()
endif()

# bolt c++ template lib benchmarks (for comparison)
if(${BOOST_COMPUTE_HAVE_BOLT} AND ${BOOST_COMPUTE_USE_CPP11})
  find_package(Bolt REQUIRED)
  include_directories(SYSTEM ${BOLT_INCLUDE_DIRS})

  set(BOLT_BENCHMARKS
    bolt_accumulate
    bolt_count
    bolt_exclusive_scan
    bolt_fill
    bolt_inner_product
    bolt_max_element
    bolt_merge
    bolt_partial_sum
    bolt_reduce_by_key
    bolt_saxpy
    bolt_sort
  )

  foreach(BENCHMARK ${BOLT_BENCHMARKS})
    set(PERF_TARGET perf_${BENCHMARK})
    add_executable(${PERF_TARGET} perf_${BENCHMARK}.cpp)
    target_link_libraries(${PERF_TARGET} ${OpenCL_LIBRARIES} ${BOLT_LIBRARIES} ${Boost_LIBRARIES})
  endforeach()
elseif(${BOOST_COMPUTE_HAVE_BOLT} AND NOT ${BOOST_COMPUTE_USE_CPP11})
  message(WARNING "BOOST_COMPUTE_USE_CPP11 must be ON for building Bolt C++ Template Library performance tests.")
endif()