Make RC2

* Fix build on big endian CPUs

* Clang-tidy

.travis.yml

@@ -6,7 +6,7 @@ os:
   - linux
   - osx
 
-osx_image: xcode10.3
+osx_image: xcode10.1
 dist: bionic
 
 # Use Build Matrix to do lint and build seperately
@@ -21,6 +21,10 @@ env:
     # cmake test
     - TASK=cmake_test
 
+  global:
+    - secure: "PR16i9F8QtNwn99C5NDp8nptAS+97xwDtXEJJfEiEVhxPaaRkOp0MPWhogCaK0Eclxk1TqkgWbdXFknwGycX620AzZWa/A1K3gAs+GrpzqhnPMuoBJ0Z9qxXTbSJvCyvMbYwVrjaxc/zWqdMU8waWz8A7iqKGKs/SqbQ3rO6v7c="
+    - secure: "dAGAjBokqm/0nVoLMofQni/fWIBcYSmdq4XvCBX1ZAMDsWnuOfz/4XCY6h2lEI1rVHZQ+UdZkc9PioOHGPZh5BnvE49/xVVWr9c4/61lrDOlkD01ZjSAeoV0fAZq+93V/wPl4QV+MM+Sem9hNNzFSbN5VsQLAiWCSapWsLdKzqA="
+
 matrix:
   exclude:
     - os: linux

R-package/vignettes/xgboostPresentation.Rmd

@@ -410,7 +410,7 @@ In some very specific cases, like when you want to pilot **XGBoost** from `caret
 
 ```{r saveLoadRBinVectorModel, message=F, warning=F}
 # save model to R's raw vector
-rawVec <- xgb.save.raw(bst)
+rawVec <- xgb.serialize(bst)
 
 # print class
 print(class(rawVec))

cmake/Python_version.in

@@ -1 +1 @@
-@xgboost_VERSION_MAJOR@.@xgboost_VERSION_MINOR@.@xgboost_VERSION_PATCH@-SNAPSHOT
+@xgboost_VERSION_MAJOR@.@xgboost_VERSION_MINOR@.@xgboost_VERSION_PATCH@rc2

include/xgboost/c_api.h

@@ -418,7 +418,14 @@ XGB_DLL int XGBoosterEvalOneIter(BoosterHandle handle,
  *          4:output feature contributions to individual predictions
  * \param ntree_limit limit number of trees used for prediction, this is only valid for boosted trees
  *    when the parameter is set to 0, we will use all the trees
- * \param training Whether the prediction value is used for training.
+ * \param training Whether the prediction function is used as part of a training loop.
+ *    Prediction can be run in 2 scenarios:
+ *    1. Given data matrix X, obtain prediction y_pred from the model.
+ *    2. Obtain the prediction for computing gradients. For example, DART booster performs dropout
+ *       during training, and the prediction result will be different from the one obtained by normal
+ *       inference step due to dropped trees.
+ *    Set training=false for the first scenario. Set training=true for the second scenario.
+ *    The second scenario applies when you are defining a custom objective function.
  * \param out_len used to store length of returning result
  * \param out_result used to set a pointer to array
  * \return 0 when success, -1 when failure happens

include/xgboost/data.h

@@ -9,6 +9,7 @@
 
 #include <dmlc/base.h>
 #include <dmlc/data.h>
+#include <dmlc/serializer.h>
 #include <rabit/rabit.h>
 #include <xgboost/base.h>
 #include <xgboost/span.h>
@@ -102,7 +103,7 @@ class MetaInfo {
   /*!
    * \brief Validate all metainfo.
    */
-  void Validate() const;
+  void Validate(int32_t device) const;
 
   MetaInfo Slice(common::Span<int32_t const> ridxs) const;
   /*!
@@ -554,5 +555,21 @@ inline BatchSet<EllpackPage> DMatrix::GetBatches(const BatchParam& param) {
 
 namespace dmlc {
 DMLC_DECLARE_TRAITS(is_pod, xgboost::Entry, true);
-}
+
+namespace serializer {
+
+template <>
+struct Handler<xgboost::Entry> {
+  inline static void Write(Stream* strm, const xgboost::Entry& data) {
+    strm->Write(data.index);
+    strm->Write(data.fvalue);
+  }
+
+  inline static bool Read(Stream* strm, xgboost::Entry* data) {
+    return strm->Read(&data->index) && strm->Read(&data->fvalue);
+  }
+};
+
+}  // namespace serializer
+}  // namespace dmlc
 #endif  // XGBOOST_DATA_H_

include/xgboost/span.h

@@ -30,6 +30,7 @@
 #define XGBOOST_SPAN_H_
 
 #include <xgboost/base.h>
+#include <xgboost/logging.h>
 
 #include <cinttypes>          // size_t
 #include <limits>             // numeric_limits
@@ -85,9 +86,11 @@ namespace common {
     }                                                                          \
   } while (0);
 
-#ifdef __CUDA_ARCH__
+#if defined(__CUDA_ARCH__)
 #define SPAN_CHECK KERNEL_CHECK
-#else
+#elif defined(XGBOOST_STRICT_R_MODE) && XGBOOST_STRICT_R_MODE == 1  // R package
+#define SPAN_CHECK CHECK  // check from dmlc
+#else  // not CUDA, not R
 #define SPAN_CHECK(cond)                                                       \
   do {                                                                         \
     if (XGBOOST_EXPECT(!(cond), false)) {                                      \

jvm-packages/pom.xml

@@ -6,7 +6,7 @@
 
     <groupId>ml.dmlc</groupId>
     <artifactId>xgboost-jvm_2.12</artifactId>
-    <version>1.1.0-SNAPSHOT</version>
+    <version>1.1.0-RC2</version>
     <packaging>pom</packaging>
     <name>XGBoost JVM Package</name>
     <description>JVM Package for XGBoost</description>

jvm-packages/xgboost4j-example/pom.xml

@@ -6,10 +6,10 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm_2.12</artifactId>
-        <version>1.1.0-SNAPSHOT</version>
+        <version>1.1.0-RC2</version>
     </parent>
     <artifactId>xgboost4j-example_2.12</artifactId>
-    <version>1.1.0-SNAPSHOT</version>
+    <version>1.1.0-RC2</version>
     <packaging>jar</packaging>
     <build>
         <plugins>
@@ -26,7 +26,7 @@
         <dependency>
             <groupId>ml.dmlc</groupId>
             <artifactId>xgboost4j-spark_${scala.binary.version}</artifactId>
-            <version>1.1.0-SNAPSHOT</version>
+            <version>1.1.0-RC2</version>
         </dependency>
         <dependency>
             <groupId>org.apache.spark</groupId>
@@ -37,7 +37,7 @@
         <dependency>
             <groupId>ml.dmlc</groupId>
             <artifactId>xgboost4j-flink_${scala.binary.version}</artifactId>
-            <version>1.1.0-SNAPSHOT</version>
+            <version>1.1.0-RC2</version>
         </dependency>
         <dependency>
             <groupId>org.apache.commons</groupId>

jvm-packages/xgboost4j-flink/pom.xml

@@ -6,10 +6,10 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm_2.12</artifactId>
-        <version>1.1.0-SNAPSHOT</version>
+        <version>1.1.0-RC2</version>
     </parent>
     <artifactId>xgboost4j-flink_2.12</artifactId>
-    <version>1.1.0-SNAPSHOT</version>
+    <version>1.1.0-RC2</version>
     <build>
         <plugins>
             <plugin>
@@ -26,7 +26,7 @@
         <dependency>
             <groupId>ml.dmlc</groupId>
             <artifactId>xgboost4j_${scala.binary.version}</artifactId>
-            <version>1.1.0-SNAPSHOT</version>
+            <version>1.1.0-RC2</version>
         </dependency>
         <dependency>
             <groupId>org.apache.commons</groupId>

jvm-packages/xgboost4j-spark/pom.xml

@@ -6,7 +6,7 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm_2.12</artifactId>
-        <version>1.1.0-SNAPSHOT</version>
+        <version>1.1.0-RC2</version>
     </parent>
     <artifactId>xgboost4j-spark_2.12</artifactId>
     <build>
@@ -24,7 +24,7 @@
         <dependency>
             <groupId>ml.dmlc</groupId>
             <artifactId>xgboost4j_${scala.binary.version}</artifactId>
-            <version>1.1.0-SNAPSHOT</version>
+            <version>1.1.0-RC2</version>
         </dependency>
         <dependency>
             <groupId>org.apache.spark</groupId>

jvm-packages/xgboost4j/pom.xml

@@ -6,10 +6,10 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm_2.12</artifactId>
-        <version>1.1.0-SNAPSHOT</version>
+        <version>1.1.0-RC2</version>
     </parent>
     <artifactId>xgboost4j_2.12</artifactId>
-    <version>1.1.0-SNAPSHOT</version>
+    <version>1.1.0-RC2</version>
     <packaging>jar</packaging>
 
     <dependencies>

python-package/xgboost/VERSION

@@ -1 +1 @@
-1.1.0-SNAPSHOT
+1.1.0rc2

python-package/xgboost/core.py

@@ -159,8 +159,9 @@ def _load_lib():
             'XGBoost Library ({}) could not be loaded.\n'.format(libname) +
             'Likely causes:\n' +
             '  * OpenMP runtime is not installed ' +
-            '(vcomp140.dll or libgomp-1.dll for Windows, ' +
-            'libgomp.so for UNIX-like OSes)\n' +
+            '(vcomp140.dll or libgomp-1.dll for Windows, libomp.dylib for Mac OSX, ' +
+            'libgomp.so for Linux and other UNIX-like OSes). Mac OSX users: Run ' +
+            '`brew install libomp` to install OpenMP runtime.\n' +
             '  * You are running 32-bit Python on a 64-bit OS\n' +
             'Error message(s): {}\n'.format(os_error_list))
     lib.XGBGetLastError.restype = ctypes.c_char_p

src/c_api/c_api.cc

@@ -426,7 +426,7 @@ XGB_DLL int XGBoosterPredict(BoosterHandle handle,
                              DMatrixHandle dmat,
                              int option_mask,
                              unsigned ntree_limit,
-                             int32_t training,
+                             int training,
                              xgboost::bst_ulong *len,
                              const bst_float **out_result) {
   API_BEGIN();

src/data/data.cc

@@ -37,7 +37,7 @@ template <typename T>
 void SaveScalarField(dmlc::Stream *strm, const std::string &name,
                      xgboost::DataType type, const T &field) {
   strm->Write(name);
-  strm->Write(type);
+  strm->Write(static_cast<uint8_t>(type));
   strm->Write(true);  // is_scalar=True
   strm->Write(field);
 }
@@ -47,7 +47,7 @@ void SaveVectorField(dmlc::Stream *strm, const std::string &name,
                      xgboost::DataType type, std::pair<uint64_t, uint64_t> shape,
                      const std::vector<T>& field) {
   strm->Write(name);
-  strm->Write(type);
+  strm->Write(static_cast<uint8_t>(type));
   strm->Write(false);  // is_scalar=False
   strm->Write(shape.first);
   strm->Write(shape.second);
@@ -71,7 +71,9 @@ void LoadScalarField(dmlc::Stream* strm, const std::string& expected_name,
   CHECK(strm->Read(&name)) << invalid;
   CHECK_EQ(name, expected_name)
       << invalid << " Expected field: " << expected_name << ", got: " << name;
-  CHECK(strm->Read(&type)) << invalid;
+  uint8_t type_val;
+  CHECK(strm->Read(&type_val)) << invalid;
+  type = static_cast<xgboost::DataType>(type_val);
   CHECK(type == expected_type)
       << invalid << "Expected field of type: " << static_cast<int>(expected_type) << ", "
       << "got field type: " << static_cast<int>(type);
@@ -91,7 +93,9 @@ void LoadVectorField(dmlc::Stream* strm, const std::string& expected_name,
   CHECK(strm->Read(&name)) << invalid;
   CHECK_EQ(name, expected_name)
     << invalid << " Expected field: " << expected_name << ", got: " << name;
-  CHECK(strm->Read(&type)) << invalid;
+  uint8_t type_val;
+  CHECK(strm->Read(&type_val)) << invalid;
+  type = static_cast<xgboost::DataType>(type_val);
   CHECK(type == expected_type)
     << invalid << "Expected field of type: " << static_cast<int>(expected_type) << ", "
     << "got field type: " << static_cast<int>(type);
@@ -338,7 +342,7 @@ void MetaInfo::SetInfo(const char* key, const void* dptr, DataType dtype, size_t
   }
 }
 
-void MetaInfo::Validate() const {
+void MetaInfo::Validate(int32_t device) const {
   if (group_ptr_.size() != 0 && weights_.Size() != 0) {
     CHECK_EQ(group_ptr_.size(), weights_.Size() + 1)
         << "Size of weights must equal to number of groups when ranking "
@@ -350,30 +354,44 @@ void MetaInfo::Validate() const {
         << "Invalid group structure.  Number of rows obtained from groups "
            "doesn't equal to actual number of rows given by data.";
   }
+  auto check_device = [device](HostDeviceVector<float> const &v) {
+    CHECK(v.DeviceIdx() == GenericParameter::kCpuId ||
+          device  == GenericParameter::kCpuId ||
+          v.DeviceIdx() == device)
+        << "Data is resided on a different device than `gpu_id`. "
+        << "Device that data is on: " << v.DeviceIdx() << ", "
+        << "`gpu_id` for XGBoost: " << device;
+  };
+
   if (weights_.Size() != 0) {
     CHECK_EQ(weights_.Size(), num_row_)
         << "Size of weights must equal to number of rows.";
+    check_device(weights_);
     return;
   }
   if (labels_.Size() != 0) {
     CHECK_EQ(labels_.Size(), num_row_)
         << "Size of labels must equal to number of rows.";
+    check_device(labels_);
     return;
   }
   if (labels_lower_bound_.Size() != 0) {
     CHECK_EQ(labels_lower_bound_.Size(), num_row_)
         << "Size of label_lower_bound must equal to number of rows.";
+    check_device(labels_lower_bound_);
     return;
   }
   if (labels_upper_bound_.Size() != 0) {
     CHECK_EQ(labels_upper_bound_.Size(), num_row_)
         << "Size of label_upper_bound must equal to number of rows.";
+    check_device(labels_upper_bound_);
     return;
   }
   CHECK_LE(num_nonzero_, num_col_ * num_row_);
   if (base_margin_.Size() != 0) {
     CHECK_EQ(base_margin_.Size() % num_row_, 0)
         << "Size of base margin must be a multiple of number of rows.";
+    check_device(base_margin_);
   }
 }
 

src/data/device_dmatrix.cu

@@ -201,6 +201,7 @@ template <typename AdapterT>
 DeviceDMatrix::DeviceDMatrix(AdapterT* adapter, float missing, int nthread, int max_bin) {
   common::HistogramCuts cuts =
       common::AdapterDeviceSketch(adapter, max_bin, missing);
+  dh::safe_cuda(cudaSetDevice(adapter->DeviceIdx()));
   auto& batch = adapter->Value();
   // Work out how many valid entries we have in each row
   dh::caching_device_vector<size_t> row_counts(adapter->NumRows() + 1, 0);

src/data/simple_dmatrix.cu

@@ -99,6 +99,7 @@ void CopyDataRowMajor(AdapterT* adapter, common::Span<Entry> data,
 // be supported in future. Does not currently support inferring row/column size
 template <typename AdapterT>
 SimpleDMatrix::SimpleDMatrix(AdapterT* adapter, float missing, int nthread) {
+  dh::safe_cuda(cudaSetDevice(adapter->DeviceIdx()));
   CHECK(adapter->NumRows() != kAdapterUnknownSize);
   CHECK(adapter->NumColumns() != kAdapterUnknownSize);
 

src/learner.cc

@@ -1052,7 +1052,7 @@ class LearnerImpl : public LearnerIO {
 
   void ValidateDMatrix(DMatrix* p_fmat) const {
     MetaInfo const& info = p_fmat->Info();
-    info.Validate();
+    info.Validate(generic_parameters_.gpu_id);
 
     auto const row_based_split = [this]() {
       return tparam_.dsplit == DataSplitMode::kRow ||

src/metric/survival_metric.cc

@@ -70,8 +70,7 @@ struct EvalAFT : public Metric {
 
     double nloglik_sum = 0.0;
     double weight_sum = 0.0;
-    #pragma omp parallel for default(none) \
-     firstprivate(nsize, is_null_weight, aft_loss_distribution_scale) \
+    #pragma omp parallel for \
      shared(weights, y_lower, y_upper, yhat) reduction(+:nloglik_sum, weight_sum)
     for (omp_ulong i = 0; i < nsize; ++i) {
       // If weights are empty, data is unweighted so we use 1.0 everywhere

src/objective/aft_obj.cc

@@ -56,8 +56,7 @@ class AFTObj : public ObjFunction {
     const omp_ulong nsize = static_cast<omp_ulong>(yhat.size());
     const float aft_loss_distribution_scale = param_.aft_loss_distribution_scale;
 
-    #pragma omp parallel for default(none) \
-      firstprivate(nsize, is_null_weight, aft_loss_distribution_scale) \
+    #pragma omp parallel for \
       shared(weights, y_lower, y_upper, yhat, gpair)
     for (omp_ulong i = 0; i < nsize; ++i) {
       // If weights are empty, data is unweighted so we use 1.0 everywhere
@@ -74,7 +73,7 @@ class AFTObj : public ObjFunction {
     // Trees give us a prediction in log scale, so exponentiate
     std::vector<bst_float> &preds = io_preds->HostVector();
     const long ndata = static_cast<long>(preds.size()); // NOLINT(*)
-    #pragma omp parallel for default(none) firstprivate(ndata) shared(preds)
+    #pragma omp parallel for shared(preds)
     for (long j = 0; j < ndata; ++j) {  // NOLINT(*)
       preds[j] = std::exp(preds[j]);
     }

tests/ci_build/rename_whl.py

@@ -21,8 +21,6 @@ whl_path = sys.argv[1]
 commit_id = sys.argv[2]
 platform_tag = sys.argv[3]
 
-assert platform_tag in ['manylinux1_x86_64', 'manylinux2010_x86_64', 'win_amd64']
-
 dirname, basename = os.path.dirname(whl_path), os.path.basename(whl_path)
 
 with cd(dirname):

tests/cpp/data/test_metainfo.cc

@@ -149,9 +149,17 @@ TEST(MetaInfo, Validate) {
   info.num_col_ = 3;
   std::vector<xgboost::bst_group_t> groups (11);
   info.SetInfo("group", groups.data(), xgboost::DataType::kUInt32, 11);
-  EXPECT_THROW(info.Validate(), dmlc::Error);
+  EXPECT_THROW(info.Validate(0), dmlc::Error);
 
   std::vector<float> labels(info.num_row_ + 1);
   info.SetInfo("label", labels.data(), xgboost::DataType::kFloat32, info.num_row_ + 1);
-  EXPECT_THROW(info.Validate(), dmlc::Error);
+  EXPECT_THROW(info.Validate(0), dmlc::Error);
+
+#if defined(XGBOOST_USE_CUDA)
+  info.group_ptr_.clear();
+  labels.resize(info.num_row_);
+  info.SetInfo("label", labels.data(), xgboost::DataType::kFloat32, info.num_row_);
+  info.labels_.SetDevice(0);
+  EXPECT_THROW(info.Validate(1), dmlc::Error);
+#endif  // defined(XGBOOST_USE_CUDA)
 }

tests/python-gpu/test_from_cupy.py

@@ -136,3 +136,14 @@ Arrow specification.'''
         n = 100
         X = cp.random.random((n, 2))
         xgb.DeviceQuantileDMatrix(X.toDlpack())
+
+    @pytest.mark.skipif(**tm.no_cupy())
+    @pytest.mark.mgpu
+    def test_specified_device(self):
+        import cupy as cp
+        cp.cuda.runtime.setDevice(0)
+        dtrain = dmatrix_from_cupy(
+            np.float32, xgb.DeviceQuantileDMatrix, np.nan)
+        with pytest.raises(xgb.core.XGBoostError):
+            xgb.train({'tree_method': 'gpu_hist', 'gpu_id': 1},
+                      dtrain, num_boost_round=10)

tests/python-gpu/test_gpu_prediction.py

@@ -121,6 +121,7 @@ class TestGPUPredict(unittest.TestCase):
     @pytest.mark.skipif(**tm.no_cupy())
     def test_inplace_predict_cupy(self):
         import cupy as cp
+        cp.cuda.runtime.setDevice(0)
         rows = 1000
         cols = 10
         cp_rng = cp.random.RandomState(1994)

tests/travis/run_test.sh

@@ -23,18 +23,42 @@ if [ ${TASK} == "python_test" ]; then
     mkdir build && cd build
     cmake .. -DUSE_OPENMP=ON -DCMAKE_VERBOSE_MAKEFILE=ON
     make -j$(nproc)
-    cd ..
 
     echo "-------------------------------"
     conda activate python3
     conda --version
     python --version
 
+    # Build binary wheel
+    cd ../python-package
+    python setup.py bdist_wheel
+    TAG=macosx_10_13_x86_64.macosx_10_14_x86_64.macosx_10_15_x86_64
+    python ../tests/ci_build/rename_whl.py dist/*.whl ${TRAVIS_COMMIT} ${TAG}
+    python -m pip install ./dist/xgboost-*-py3-none-${TAG}.whl
+
+    # Run unit tests
+    cd ..
     python -m pip install graphviz pytest pytest-cov codecov
     python -m pip install datatable
     python -m pip install numpy scipy pandas matplotlib scikit-learn dask[complete]
     python -m pytest -v --fulltrace -s tests/python --cov=python-package/xgboost || exit -1
     codecov
+
+    # Deploy binary wheel to S3
+    python -m pip install awscli
+    if [ "${TRAVIS_PULL_REQUEST}" != "false" ]
+    then
+        S3_DEST="s3://xgboost-nightly-builds/PR-${TRAVIS_PULL_REQUEST}/"
+    else
+        if [ "${TRAVIS_BRANCH}" == "master" ]
+        then
+            S3_DEST="s3://xgboost-nightly-builds/"
+        elif [ -z "${TRAVIS_TAG}" ]
+        then
+            S3_DEST="s3://xgboost-nightly-builds/${TRAVIS_BRANCH}/"
+        fi
+    fi
+    python -m awscli s3 cp python-package/dist/*.whl "${S3_DEST}" --acl public-read || true
 fi
 
 if [ ${TASK} == "java_test" ]; then