Set gen code by default.

dmlc · Apr 10, 2019 · 720d947 · 720d947
1 parent db75a19
commit 720d947
Show file tree

Hide file tree

Showing 5 changed files with 26 additions and 35 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -8,7 +8,8 @@ set_default_configuration_release()
 msvc_use_static_runtime()
 
 #-- Options
-option(BUILD_C_DOC "Build documentation for C APIs using Doxygen.")
+option(BUILD_C_DOC "Build documentation for C APIs using Doxygen." OFF)
+option(USE_OPENMP "Build with OpenMP support." ON)
 ## Bindings
 option(JVM_BINDINGS "Build JVM bindings" OFF)
 option(R_LIB "Build shared library for R package" OFF)
@@ -23,7 +24,7 @@ option(GENERATE_COMPILATION_DATABASE
 option(USE_CUDA  "Build with GPU acceleration" OFF)
 option(USE_NCCL  "Build with NCCL to enable multi-GPU support." OFF)
 set(GPU_COMPUTE_VER "" CACHE STRING
-  "Space separated list of compute versions to be built against, e.g. '35 61'")
+  "Semicolon separated list of compute versions to be built against, e.g. '35;61'")
 ## Sanitizers
 option(USE_SANITIZER "Use santizer flags" OFF)
 option(SANITIZER_PATH "Path to sanitizes.")
@@ -47,20 +48,14 @@ endif (USE_SANITIZER)
 
 if (USE_CUDA)
   cmake_minimum_required(VERSION 3.12)
-  # `export CXX=' is ignored by CMake CUDA, `CMAKE_CUDA_HOST_COMPILER'
-  # is not available before 3.10.
-  if (CMAKE_VERSION VERSION_LESS 3.10)
-    set(CCBIN "-ccbin=${CMAKE_CXX_COMPILER}")
-    message(STATUS "set -ccbin=${CMAKE_CXX_COMPILER}")
-  else ()
-    set(CMAKE_CUDA_HOST_COMPILER ${CMAKE_CXX_COMPILER})
-    message(STATUS "Configured CUDA host compiler: ${CMAKE_CUDA_HOST_COMPILER}")
-  endif (CMAKE_VERSION VERSION_LESS 3.10)
+  # `export CXX=' is ignored by CMake CUDA.
+  set(CMAKE_CUDA_HOST_COMPILER ${CMAKE_CXX_COMPILER})
+  message(STATUS "Configured CUDA host compiler: ${CMAKE_CUDA_HOST_COMPILER}")
 
   enable_language(CUDA)
-  if (GPU_COMPUTE_VER)
-    format_gencode_flags(${GPU_COMPUTE_VER} GEN_CODE)
-  endif (GPU_COMPUTE_VER)
+  set(GEN_CODE "")
+  format_gencode_flags("${GPU_COMPUTE_VER}" GEN_CODE)
+  message(STATUS "CUDA GEN_CODE: ${GEN_CODE}")
 endif (USE_CUDA)
 
 # dmlc-core

diff --git a/cmake/Utils.cmake b/cmake/Utils.cmake
@@ -68,11 +68,12 @@ function(format_gencode_flags flags out)
   endif()
   # Generate SASS
   foreach(ver ${flags})
-    set(${out} "${${out}}--gpu-architecture=compute_${ver};--gpu-code=sm_${ver};")
+    set(${out} "${${out}}--generate-code=arch=compute_${ver},code=sm_${ver};")
   endforeach()
   # Generate PTX for last architecture
   list(GET flags -1 ver)
-  set(${out} "${${out}}--gpu-architecture=compute_${ver};--gpu-code=compute_${ver};")
+  set(${out} "${${out}}--generate-code=arch=compute_${ver},code=compute_${ver};")
+  # set(${out} "${${out}}--gpu-architecture=compute_${ver};--gpu-code=compute_${ver};")
 
   set(${out} "${${out}}" PARENT_SCOPE)
 endfunction(format_gencode_flags flags)

diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
@@ -43,20 +43,16 @@ if (USE_CUDA)
     $<$<COMPILE_LANGUAGE:CUDA>:--expt-relaxed-constexpr>
     $<$<COMPILE_LANGUAGE:CUDA>:-lineinfo>
     $<$<COMPILE_LANGUAGE:CUDA>:--std=c++11>)
+  target_compile_options(objxgboost
+    PRIVATE
+    $<$<COMPILE_LANGUAGE:CUDA>:${GEN_CODE}>)
   if (CMAKE_VERSION VERSION_LESS 3.10)
     target_compile_options(objxgboost PRIVATE
       $<$<COMPILE_LANGUAGE:CUDA>:${CCBIN}>
       # PIC flag doesn't work for OBJECT target.
       $<$<COMPILE_LANGUAGE:CUDA>:-Xcompiler=-fPIC>)
   endif (CMAKE_VERSION VERSION_LESS 3.10)
 
-  if (GEN_CODE)
-    message(STATUS "GEN_CODE: ${GEN_CODE}")
-    target_compile_options(objxgboost
-      PRIVATE
-      $<$<COMPILE_LANGUAGE:CUDA>:${GEN_CODE}>)
-  endif (GEN_CODE)
-
   if (USE_NCCL)
     find_package(Nccl REQUIRED)
     target_include_directories(objxgboost PRIVATE ${NCCL_INCLUDE_DIR})
@@ -107,11 +103,12 @@ if (XGBOOST_BUILTIN_PREFETCH_PRESENT)
     -DXGBOOST_BUILTIN_PREFETCH_PRESENT=1)
 endif (XGBOOST_BUILTIN_PREFETCH_PRESENT)
 
-find_package(OpenMP)
-if (OpenMP_CXX_FOUND OR OPENMP_FOUND)
-  target_compile_options(objxgboost PRIVATE  $<$<COMPILE_LANGUAGE:CXX>:${OpenMP_CXX_FLAGS}>)
-  list(APPEND SRC_LIBS ${OpenMP_CXX_LIBRARIES})
-  set(LINKED_LIBRARIES_PRIVATE "${LINKED_LIBRARIES_PRIVATE};${SRC_LIBS}" PARENT_SCOPE)
-endif (OpenMP_CXX_FOUND OR OPENMP_FOUND)
-
+if (USE_OPENMP)
+  find_package(OpenMP REQUIRED)
+  if (OpenMP_CXX_FOUND OR OPENMP_FOUND)
+    target_compile_options(objxgboost PRIVATE  $<$<COMPILE_LANGUAGE:CXX>:${OpenMP_CXX_FLAGS}>)
+    list(APPEND SRC_LIBS ${OpenMP_CXX_LIBRARIES})
+    set(LINKED_LIBRARIES_PRIVATE "${LINKED_LIBRARIES_PRIVATE};${SRC_LIBS}" PARENT_SCOPE)
+  endif (OpenMP_CXX_FOUND OR OPENMP_FOUND)
+endif (USE_OPENMP)
 #-- End object library
diff --git a/src/tree/updater_gpu_hist.cu b/src/tree/updater_gpu_hist.cu
@@ -470,7 +470,7 @@ struct CalcWeightTrainParam {
 };
 
 // Bin each input data entry, store the bin indices in compressed form.
-inline
+template<typename std::enable_if<true,  int>::type = 0>
 __global__ void CompressBinEllpackKernel(
     common::CompressedBufferWriter wr,
     common::CompressedByteT* __restrict__ buffer,  // gidx_buffer

diff --git a/tests/cpp/CMakeLists.txt b/tests/cpp/CMakeLists.txt
@@ -16,15 +16,13 @@ if (USE_CUDA)
     $<$<COMPILE_LANGUAGE:CUDA>:--expt-relaxed-constexpr>
     $<$<COMPILE_LANGUAGE:CUDA>:-lineinfo>
     $<$<COMPILE_LANGUAGE:CUDA>:--std=c++11>)
+  target_compile_options(testxgboost PRIVATE
+    $<$<COMPILE_LANGUAGE:CUDA>:${GEN_CODE}>)
   if (CMAKE_VERSION VERSION_LESS 3.10)
     target_compile_options(testxgboost PRIVATE
       $<$<COMPILE_LANGUAGE:CUDA>:${CCBIN}>
       $<$<COMPILE_LANGUAGE:CUDA>:-Xcompiler=-fPIC>)
   endif (CMAKE_VERSION VERSION_LESS 3.10)
-  if (GEN_CODE)
-    target_compile_options(testxgboost PRIVATE
-      $<$<COMPILE_LANGUAGE:CUDA>:${GEN_CODE}>)
-  endif()
 
   target_compile_definitions(testxgboost
     PRIVATE -DXGBOOST_USE_CUDA=1)