Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix non-openmp build. (#5566)

Browse Source 
* Add test to Jenkins.
* Fix threading utils tests.
* Require thread library.
This commit is contained in:
Jiaming Yuan
2020-04-20 12:16:38 +08:00
committed by GitHub
parent b2827a80e1
commit ccd30e4491
9 changed files with 52 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/CMakeLists.txt
View File

@@ -88,12 +88,15 @@ if (XGBOOST_BUILTIN_PREFETCH_PRESENT)
-DXGBOOST_BUILTIN_PREFETCH_PRESENT=1)
endif (XGBOOST_BUILTIN_PREFETCH_PRESENT)
find_package(Threads REQUIRED)
list(APPEND SRC_LIBS Threads::Threads ${CMAKE_THREAD_LIBS_INIT})
if (USE_OPENMP OR USE_CUDA) # CUDA requires OpenMP
find_package(OpenMP REQUIRED)
list(APPEND SRC_LIBS OpenMP::OpenMP_CXX)
set(LINKED_LIBRARIES_PRIVATE "${LINKED_LIBRARIES_PRIVATE};${SRC_LIBS}" PARENT_SCOPE)
target_link_libraries(objxgboost PRIVATE OpenMP::OpenMP_CXX)
endif (USE_OPENMP OR USE_CUDA)
set(LINKED_LIBRARIES_PRIVATE "${LINKED_LIBRARIES_PRIVATE};${SRC_LIBS}" PARENT_SCOPE)
# For MSVC: Call msvc_use_static_runtime() once again to completely
# replace /MD with /MT. See https://github.com/dmlc/xgboost/issues/4462

3
src/common/column_matrix.h
View File

@@ -227,8 +227,7 @@ class ColumnMatrix {
/* missing values make sense only for column with type kDenseColumn,
and if no missing values were observed it could be handled much faster. */
if (noMissingValues) {
const int32_t nthread = omp_get_max_threads(); // NOLINT
#pragma omp parallel for num_threads(nthread)
#pragma omp parallel for num_threads(omp_get_max_threads())
for (omp_ulong rid = 0; rid < nrow; ++rid) {
const size_t ibegin = rid*nfeature;
const size_t iend = (rid+1)*nfeature;

6
src/common/threading_utils.h
View File

@@ -110,10 +110,12 @@ class BlockedSpace2d {
// Wrapper to implement nested parallelism with simple omp parallel for
template<typename Func>
void ParallelFor2d(const BlockedSpace2d& space, const int nthreads, Func func) {
void ParallelFor2d(const BlockedSpace2d& space, int nthreads, Func func) {
const size_t num_blocks_in_space = space.Size();
nthreads = std::min(nthreads, omp_get_max_threads());
nthreads = std::max(nthreads, 1);
#pragma omp parallel num_threads(nthreads)
#pragma omp parallel num_threads(nthreads)
{
size_t tid = omp_get_thread_num();
size_t chunck_size = num_blocks_in_space / nthreads + !!(num_blocks_in_space % nthreads);

13
src/metric/rank_metric.cc
View File

@@ -189,12 +189,7 @@ struct EvalAuc : public Metric {
rec[j - gptr[group_id]] = {h_preds[j], j};
}
if (omp_in_parallel()) {
std::stable_sort(rec.begin(), rec.end(), common::CmpFirst);
} else {
XGBOOST_PARALLEL_SORT(rec.begin(), rec.end(), common::CmpFirst);
}
XGBOOST_PARALLEL_SORT(rec.begin(), rec.end(), common::CmpFirst);
// calculate AUC
double sum_pospair = 0.0;
double sum_npos = 0.0, sum_nneg = 0.0, buf_pos = 0.0, buf_neg = 0.0;
@@ -557,11 +552,7 @@ struct EvalAucPR : public Metric {
continue;
}
if (omp_in_parallel()) {
std::stable_sort(rec.begin(), rec.end(), common::CmpFirst);
} else {
XGBOOST_PARALLEL_SORT(rec.begin(), rec.end(), common::CmpFirst);
}
XGBOOST_PARALLEL_SORT(rec.begin(), rec.end(), common::CmpFirst);
// calculate AUC
double tp = 0.0, prevtp = 0.0, fp = 0.0, prevfp = 0.0, h = 0.0, a = 0.0, b = 0.0;