CPU optimizations

.gitignore

@@ -96,3 +96,6 @@ plugin/updater_gpu/test/cpp/data
 # files from R-package source install
 **/config.status
 R-package/src/Makevars
+
+# Python install
+python-package/xgboost/tracker.py

CMakeLists.txt

@@ -1,5 +1,5 @@
 cmake_minimum_required(VERSION 3.3)
-project(xgboost LANGUAGES CXX C VERSION 0.82)
+project(xgboost LANGUAGES CXX C VERSION 0.90)
 include(cmake/Utils.cmake)
 list(APPEND CMAKE_MODULE_PATH "${PROJECT_SOURCE_DIR}/cmake/modules")
 cmake_policy(SET CMP0022 NEW)
@@ -10,7 +10,6 @@ if (MSVC)
 endif (MSVC)
 
 set_default_configuration_release()
-msvc_use_static_runtime()
 
 #-- Options
 option(BUILD_C_DOC "Build documentation for C APIs using Doxygen." OFF)
@@ -20,13 +19,18 @@ option(JVM_BINDINGS "Build JVM bindings" OFF)
 option(R_LIB "Build shared library for R package" OFF)
 ## Dev
 option(GOOGLE_TEST "Build google tests" OFF)
+option(USE_DMLC_GTEST "Use google tests bundled with dmlc-core submodule (EXPERIMENTAL)" OFF)
 option(USE_NVTX "Build with cuda profiling annotations. Developers only." OFF)
 set(NVTX_HEADER_DIR "" CACHE PATH "Path to the stand-alone nvtx header")
 ## CUDA
 option(USE_CUDA  "Build with GPU acceleration" OFF)
 option(USE_NCCL  "Build with NCCL to enable multi-GPU support." OFF)
+option(BUILD_WITH_SHARED_NCCL "Build with shared NCCL library." OFF)
 set(GPU_COMPUTE_VER "" CACHE STRING
   "Semicolon separated list of compute versions to be built against, e.g. '35;61'")
+if (BUILD_WITH_SHARED_NCCL AND (NOT USE_NCCL))
+  message(SEND_ERROR "Build XGBoost with -DUSE_NCCL=ON to enable BUILD_WITH_SHARED_NCCL.")
+endif (BUILD_WITH_SHARED_NCCL AND (NOT USE_NCCL))
 ## Sanitizers
 option(USE_SANITIZER "Use santizer flags" OFF)
 option(SANITIZER_PATH "Path to sanitizes.")
@@ -42,6 +46,26 @@ if (USE_AVX)
   message(WARNING "The option 'USE_AVX' is deprecated as experimental AVX features have been removed from xgboost.")
 endif (USE_AVX)
 
+# Check existence of software pre-fetching
+include(CheckCXXSourceCompiles)
+check_cxx_source_compiles("
+#include <xmmintrin.h>
+int main() {
+  char data = 0;
+  const char* address = &data;
+  _mm_prefetch(address, _MM_HINT_NTA);
+  return 0;
+}
+" XGBOOST_MM_PREFETCH_PRESENT)
+check_cxx_source_compiles("
+int main() {
+  char data = 0;
+  const char* address = &data;
+  __builtin_prefetch(address, 0, 0);
+  return 0;
+}
+" XGBOOST_BUILTIN_PREFETCH_PRESENT)
+
 # Sanitizer
 if (USE_SANITIZER)
   # Older CMake versions have had troubles with Sanitizer
@@ -52,6 +76,7 @@ endif (USE_SANITIZER)
 
 if (USE_CUDA)
   cmake_minimum_required(VERSION 3.12)
+  SET(USE_OPENMP ON CACHE BOOL "CUDA requires OpenMP" FORCE)
   # `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}")
@@ -63,6 +88,7 @@ if (USE_CUDA)
 endif (USE_CUDA)
 
 # dmlc-core
+msvc_use_static_runtime()
 add_subdirectory(${PROJECT_SOURCE_DIR}/dmlc-core)
 set_target_properties(dmlc PROPERTIES
   CXX_STANDARD 11
@@ -120,6 +146,10 @@ endif (JVM_BINDINGS)
 #-- CLI for xgboost
 add_executable(runxgboost ${PROJECT_SOURCE_DIR}/src/cli_main.cc ${XGBOOST_OBJ_SOURCES})
 # For cli_main.cc only
+if (USE_OPENMP)
+  find_package(OpenMP REQUIRED)
+  target_compile_options(runxgboost PRIVATE ${OpenMP_CXX_FLAGS})
+endif (USE_OPENMP)
 target_include_directories(runxgboost
   PRIVATE
   ${PROJECT_SOURCE_DIR}/include
@@ -135,6 +165,8 @@ set_target_properties(
 
 set_output_directory(runxgboost ${PROJECT_SOURCE_DIR})
 set_output_directory(xgboost ${PROJECT_SOURCE_DIR}/lib)
+# Ensure these two targets do not build simultaneously, as they produce outputs with conflicting names
+add_dependencies(xgboost runxgboost)
 
 #-- Installing XGBoost
 if (R_LIB)
@@ -210,3 +242,8 @@ if (GOOGLE_TEST)
     PROPERTIES
     PASS_REGULAR_EXPRESSION ".*test-rmse:0.087.*")
 endif (GOOGLE_TEST)
+
+# For MSVC: Call msvc_use_static_runtime() once again to completely
+# replace /MD with /MT. See https://github.com/dmlc/xgboost/issues/4462
+# for issues caused by mixing of /MD and /MT flags
+msvc_use_static_runtime()

CONTRIBUTORS.md

@@ -89,3 +89,6 @@ List of Contributors
 * [Sam Wilkinson](https://samwilkinson.io)
 * [Matthew Jones](https://github.com/mt-jones)
 * [Jiaxiang Li](https://github.com/JiaxiangBU)
+* [Bryan Woods](https://github.com/bryan-woods)
+  - Bryan added support for cross-validation for the ranking objective
+* [Haoda Fu](https://github.com/fuhaoda)

Jenkinsfile

@@ -25,7 +25,7 @@ pipeline {
 
   // Build stages
   stages {
-    stage('Get sources') {
+    stage('Jenkins Linux: Get sources') {
       agent { label 'linux && cpu' }
       steps {
         script {
@@ -35,7 +35,7 @@ pipeline {
         milestone ordinal: 1
       }
     }
-    stage('Formatting Check') {
+    stage('Jenkins Linux: Formatting Check') {
       agent none
       steps {
         script {
@@ -49,35 +49,37 @@ pipeline {
         milestone ordinal: 2
       }
     }
-    stage('Build') {
+    stage('Jenkins Linux: Build') {
       agent none
       steps {
         script {
           parallel ([
             'build-cpu': { BuildCPU() },
             'build-gpu-cuda8.0': { BuildCUDA(cuda_version: '8.0') },
-            'build-gpu-cuda9.2': { BuildCUDA(cuda_version: '9.2') },
+            'build-gpu-cuda9.0': { BuildCUDA(cuda_version: '9.0') },
             'build-gpu-cuda10.0': { BuildCUDA(cuda_version: '10.0') },
-            'build-jvm-packages': { BuildJVMPackages(spark_version: '2.4.1') },
+            'build-gpu-cuda10.1': { BuildCUDA(cuda_version: '10.1') },
+            'build-jvm-packages': { BuildJVMPackages(spark_version: '2.4.3') },
             'build-jvm-doc': { BuildJVMDoc() }
           ])
         }
         milestone ordinal: 3
       }
     }
-    stage('Test') {
+    stage('Jenkins Linux: Test') {
       agent none
       steps {
         script {
           parallel ([
             'test-python-cpu': { TestPythonCPU() },
             'test-python-gpu-cuda8.0': { TestPythonGPU(cuda_version: '8.0') },
-            'test-python-gpu-cuda9.2': { TestPythonGPU(cuda_version: '9.2') },
+            'test-python-gpu-cuda9.0': { TestPythonGPU(cuda_version: '9.0') },
             'test-python-gpu-cuda10.0': { TestPythonGPU(cuda_version: '10.0') },
-            'test-python-mgpu-cuda10.0': { TestPythonGPU(cuda_version: '10.0', multi_gpu: true) },
-            'test-cpp-gpu': { TestCppGPU(cuda_version: '10.0') },
-            'test-cpp-mgpu': { TestCppGPU(cuda_version: '10.0', multi_gpu: true) },
-            'test-jvm-jdk8': { CrossTestJVMwithJDK(jdk_version: '8') },
+            'test-python-gpu-cuda10.1': { TestPythonGPU(cuda_version: '10.1') },
+            'test-python-mgpu-cuda10.1': { TestPythonGPU(cuda_version: '10.1', multi_gpu: true) },
+            'test-cpp-gpu': { TestCppGPU(cuda_version: '10.1') },
+            'test-cpp-mgpu': { TestCppGPU(cuda_version: '10.1', multi_gpu: true) },
+            'test-jvm-jdk8': { CrossTestJVMwithJDK(jdk_version: '8', spark_version: '2.4.3') },
             'test-jvm-jdk11': { CrossTestJVMwithJDK(jdk_version: '11') },
             'test-jvm-jdk12': { CrossTestJVMwithJDK(jdk_version: '12') },
             'test-r-3.4.4': { TestR(use_r35: false) },
@@ -153,8 +155,11 @@ def Doxygen() {
     def container_type = "cpu"
     def docker_binary = "docker"
     sh """
-    ${dockerRun} ${container_type} ${docker_binary} tests/ci_build/doxygen.sh
+    ${dockerRun} ${container_type} ${docker_binary} tests/ci_build/doxygen.sh ${BRANCH_NAME}
     """
+    archiveArtifacts artifacts: "build/${BRANCH_NAME}.tar.bz2", allowEmptyArchive: true
+    echo 'Uploading doc...'
+    s3Upload file: "build/${BRANCH_NAME}.tar.bz2", bucket: 'xgboost-docs', acl: 'PublicRead', path: "doxygen/${BRANCH_NAME}.tar.bz2"
     deleteDir()
   }
 }
@@ -192,10 +197,13 @@ def BuildCUDA(args) {
     ${dockerRun} ${container_type} ${docker_binary} ${docker_args} tests/ci_build/build_via_cmake.sh -DUSE_CUDA=ON -DUSE_NCCL=ON -DOPEN_MP:BOOL=ON
     ${dockerRun} ${container_type} ${docker_binary} ${docker_args} bash -c "cd python-package && rm -rf dist/* && python setup.py bdist_wheel --universal"
     """
-    // Only stash wheel for CUDA 8.0 target
+    // Stash wheel for CUDA 8.0 / 9.0 target
     if (args.cuda_version == '8.0') {
       echo 'Stashing Python wheel...'
-      stash name: 'xgboost_whl', includes: 'python-package/dist/*.whl'
+      stash name: 'xgboost_whl_cuda8', includes: 'python-package/dist/*.whl'
+    } else if (args.cuda_version == '9.0') {
+      echo 'Stashing Python wheel...'
+      stash name: 'xgboost_whl_cuda9', includes: 'python-package/dist/*.whl'
       archiveArtifacts artifacts: "python-package/dist/*.whl", allowEmptyArchive: true
       echo 'Stashing C++ test executable (testxgboost)...'
       stash name: 'xgboost_cpp_tests', includes: 'build/testxgboost'
@@ -213,7 +221,7 @@ def BuildJVMPackages(args) {
     // Use only 4 CPU cores
     def docker_extra_params = "CI_DOCKER_EXTRA_PARAMS_INIT='--cpuset-cpus 0-3'"
     sh """
-    ${docker_extra_params} ${dockerRun} ${container_type} ${docker_binary} tests/ci_build/build_jvm_packages.sh
+    ${docker_extra_params} ${dockerRun} ${container_type} ${docker_binary} tests/ci_build/build_jvm_packages.sh ${args.spark_version}
     """
     echo 'Stashing XGBoost4J JAR...'
     stash name: 'xgboost4j_jar', includes: 'jvm-packages/xgboost4j/target/*.jar,jvm-packages/xgboost4j-spark/target/*.jar,jvm-packages/xgboost4j-example/target/*.jar'
@@ -239,7 +247,7 @@ def BuildJVMDoc() {
 
 def TestPythonCPU() {
   node('linux && cpu') {
-    unstash name: 'xgboost_whl'
+    unstash name: 'xgboost_whl_cuda9'
     unstash name: 'srcs'
     echo "Test Python CPU"
     def container_type = "cpu"
@@ -254,7 +262,11 @@ def TestPythonCPU() {
 def TestPythonGPU(args) {
   nodeReq = (args.multi_gpu) ? 'linux && mgpu' : 'linux && gpu'
   node(nodeReq) {
-    unstash name: 'xgboost_whl'
+    if (args.cuda_version == '8.0') {
+      unstash name: 'xgboost_whl_cuda8'
+    } else {
+      unstash name: 'xgboost_whl_cuda9'
+    }
     unstash name: 'srcs'
     echo "Test Python GPU: CUDA ${args.cuda_version}"
     def container_type = "gpu"
@@ -299,12 +311,17 @@ def CrossTestJVMwithJDK(args) {
   node('linux && cpu') {
     unstash name: 'xgboost4j_jar'
     unstash name: 'srcs'
-    echo "Test XGBoost4J on a machine with JDK ${args.jdk_version}"
+    if (args.spark_version != null) {
+      echo "Test XGBoost4J on a machine with JDK ${args.jdk_version}, Spark ${args.spark_version}"
+    } else {
+      echo "Test XGBoost4J on a machine with JDK ${args.jdk_version}"
+    }
     def container_type = "jvm_cross"
     def docker_binary = "docker"
-    def docker_args = "--build-arg JDK_VERSION=${args.jdk_version}"
-    // Only run integration tests for JDK 8, as Spark doesn't support later JDKs yet
-    def docker_extra_params = (args.jdk_version == '8') ? "CI_DOCKER_EXTRA_PARAMS_INIT='-e RUN_INTEGRATION_TEST=1'" : ""
+    def spark_arg = (args.spark_version != null) ? "--build-arg SPARK_VERSION=${args.spark_version}" : ""
+    def docker_args = "--build-arg JDK_VERSION=${args.jdk_version} ${spark_arg}"
+    // Run integration tests only when spark_version is given
+    def docker_extra_params = (args.spark_version != null) ? "CI_DOCKER_EXTRA_PARAMS_INIT='-e RUN_INTEGRATION_TEST=1'" : ""
     sh """
     ${docker_extra_params} ${dockerRun} ${container_type} ${docker_binary} ${docker_args} tests/ci_build/test_jvm_cross.sh
     """
@@ -323,6 +340,8 @@ def TestR(args) {
     sh """
     ${dockerRun} ${container_type} ${docker_binary} ${docker_args} tests/ci_build/build_test_rpkg.sh
     """
+    // Save error log, if any
+    archiveArtifacts artifacts: "xgboost.Rcheck/00install.out", allowEmptyArchive: true
     deleteDir()
   }
 }