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

Consistent use of context to specify number of threads. (#8733)

Browse Source 
- Use context in all tests.
- Use context in R.
- Use context in C API DMatrix initialization. (0 threads is used as dft).
This commit is contained in:
Jiaming Yuan
2023-01-30 15:25:31 +08:00
committed by GitHub
parent 21a28f2cc5
commit 3760cede0f
24 changed files with 212 additions and 152 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
src/common/threading_utils.cc
View File

@@ -1,5 +1,5 @@
/*!
* Copyright 2022 by XGBoost Contributors
/**
* Copyright 2022-2023 by XGBoost Contributors
*/
#include "threading_utils.h"
@@ -10,14 +10,6 @@
namespace xgboost {
namespace common {
/**
* \brief Get thread limit from CFS
*
* Modified from
* github.com/psiha/sweater/blob/master/include/boost/sweater/hardware_concurrency.hpp
*
* MIT License: Copyright (c) 2016 Domagoj Šarić
*/
int32_t GetCfsCPUCount() noexcept {
#if defined(__linux__)
// https://bugs.openjdk.java.net/browse/JDK-8146115
@@ -47,5 +39,20 @@ int32_t GetCfsCPUCount() noexcept {
#endif // defined(__linux__)
return -1;
}
std::int32_t OmpGetNumThreads(std::int32_t n_threads) {
// Don't use parallel if we are in a parallel region.
if (omp_in_parallel()) {
return 1;
}
// If -1 or 0 is specified by the user, we default to maximum number of threads.
if (n_threads <= 0) {
n_threads = std::min(omp_get_num_procs(), omp_get_max_threads());
}
// Honor the openmp thread limit, which can be set via environment variable.
n_threads = std::min(n_threads, OmpGetThreadLimit());
n_threads = std::max(n_threads, 1);
return n_threads;
}
} // namespace common
} // namespace xgboost

33
src/common/threading_utils.h
View File

@@ -1,5 +1,5 @@
/*!
* Copyright 2019-2022 by XGBoost Contributors
/**
* Copyright 2019-2023 by XGBoost Contributors
*/
#ifndef XGBOOST_COMMON_THREADING_UTILS_H_
#define XGBOOST_COMMON_THREADING_UTILS_H_
@@ -231,23 +231,28 @@ void ParallelFor(Index size, int32_t n_threads, Func fn) {
ParallelFor(size, n_threads, Sched::Static(), fn);
}
inline int32_t OmpGetThreadLimit() {
int32_t limit = omp_get_thread_limit();
inline std::int32_t OmpGetThreadLimit() {
std::int32_t limit = omp_get_thread_limit();
CHECK_GE(limit, 1) << "Invalid thread limit for OpenMP.";
return limit;
}
int32_t GetCfsCPUCount() noexcept;
inline int32_t OmpGetNumThreads(int32_t n_threads) {
if (n_threads <= 0) {
n_threads = std::min(omp_get_num_procs(), omp_get_max_threads());
}
n_threads = std::min(n_threads, OmpGetThreadLimit());
n_threads = std::max(n_threads, 1);
return n_threads;
}
/**
* \brief Get thread limit from CFS.
*
* This function has non-trivial overhead and should not be called repeatly.
*
* Modified from
* github.com/psiha/sweater/blob/master/include/boost/sweater/hardware_concurrency.hpp
*
* MIT License: Copyright (c) 2016 Domagoj Šarić
*/
std::int32_t GetCfsCPUCount() noexcept;
/**
* \brief Get the number of available threads based on n_threads specified by users.
*/
std::int32_t OmpGetNumThreads(std::int32_t n_threads);
/*!
* \brief A C-style array with in-stack allocation. As long as the array is smaller than