Skip to content

GitLab

"vscode:/vscode.git/clone" did not exist on "9654f16528d2d1e3462a298106cb1fc6ed43cbac"
Commit 20996c92 authored by Jeff Daily's avatar Jeff Daily
Browse files

partial revert of 61ec4f1a 

Instead of replacing all #ifdef USE_CUDA, just add USE_CUDA define to ROCm build.
parent 1b3deb5f
Hide whitespace changes
Inline Side-by-side
Showing with 92 additions and 92 deletions
src/boosting/cuda/cuda_score_updater.cpp
View file @ 20996c92
......@@ -5,7 +5,7 @@
#include "cuda_score_updater.hpp"
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
namespace LightGBM {
......@@ -91,4 +91,4 @@ inline void CUDAScoreUpdater::MultiplyScore(double val, int cur_tree_id) {
} // namespace LightGBM
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
src/boosting/cuda/cuda_score_updater.cu
View file @ 20996c92
......@@ -5,7 +5,7 @@
#include "cuda_score_updater.hpp"
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
namespace LightGBM {
......@@ -42,4 +42,4 @@ void CUDAScoreUpdater::LaunchMultiplyScoreConstantKernel(const double val, const
} // namespace LightGBM
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
src/boosting/cuda/cuda_score_updater.hpp
View file @ 20996c92
......@@ -6,7 +6,7 @@
#ifndef LIGHTGBM_BOOSTING_CUDA_CUDA_SCORE_UPDATER_HPP_
#define LIGHTGBM_BOOSTING_CUDA_CUDA_SCORE_UPDATER_HPP_
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
#include <LightGBM/cuda/cuda_utils.hu>
......@@ -60,6 +60,6 @@ class CUDAScoreUpdater: public ScoreUpdater {
} // namespace LightGBM
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
#endif // LIGHTGBM_BOOSTING_CUDA_CUDA_SCORE_UPDATER_HPP_
src/boosting/gbdt.cpp
View file @ 20996c92
......@@ -72,12 +72,12 @@ void GBDT::Init(const Config* config, const Dataset* train_data, const Objective
if (config_->device_type == std::string("cuda")) {
LGBM_config_::current_learner = use_cuda_learner;
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
if (config_->device_type == std::string("cuda")) {
const int gpu_device_id = config_->gpu_device_id >= 0 ? config_->gpu_device_id : 0;
CUDASUCCESS_OR_FATAL(cudaSetDevice(gpu_device_id));
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
}
// load forced_splits file
......@@ -118,15 +118,15 @@ void GBDT::Init(const Config* config, const Dataset* train_data, const Objective
}
training_metrics_.shrink_to_fit();
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
if (config_->device_type == std::string("cuda")) {
train_score_updater_.reset(new CUDAScoreUpdater(train_data_, num_tree_per_iteration_, boosting_on_gpu_));
} else {
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
train_score_updater_.reset(new ScoreUpdater(train_data_, num_tree_per_iteration_));
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
num_data_ = train_data_->num_data();
......@@ -188,11 +188,11 @@ void GBDT::AddValidDataset(const Dataset* valid_data,
}
// for a validation dataset, we need its score and metric
auto new_score_updater =
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
config_->device_type == std::string("cuda") ?
std::unique_ptr<CUDAScoreUpdater>(new CUDAScoreUpdater(valid_data, num_tree_per_iteration_,
objective_function_ != nullptr && objective_function_->IsCUDAObjective())) :
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
std::unique_ptr<ScoreUpdater>(new ScoreUpdater(valid_data, num_tree_per_iteration_));
// update score
for (int i = 0; i < iter_; ++i) {
......@@ -501,15 +501,15 @@ void GBDT::UpdateScore(const Tree* tree, const int cur_tree_id) {
const data_size_t bag_data_cnt = data_sample_strategy_->bag_data_cnt();
// we need to predict out-of-bag scores of data for boosting
if (num_data_ - bag_data_cnt > 0) {
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
if (config_->device_type == std::string("cuda")) {
train_score_updater_->AddScore(tree, data_sample_strategy_->cuda_bag_data_indices().RawData() + bag_data_cnt, num_data_ - bag_data_cnt, cur_tree_id);
} else {
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
train_score_updater_->AddScore(tree, data_sample_strategy_->bag_data_indices().data() + bag_data_cnt, num_data_ - bag_data_cnt, cur_tree_id);
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
}
} else {
......@@ -523,17 +523,17 @@ void GBDT::UpdateScore(const Tree* tree, const int cur_tree_id) {
}
}
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
std::vector<double> GBDT::EvalOneMetric(const Metric* metric, const double* score, const data_size_t num_data) const {
#else
std::vector<double> GBDT::EvalOneMetric(const Metric* metric, const double* score, const data_size_t /*num_data*/) const {
#endif // USE_CUDA || USE_ROCM
#if defined(USE_CUDA) || defined(USE_ROCM)
#endif // USE_CUDA
#ifdef USE_CUDA
const bool evaluation_on_cuda = metric->IsCUDAMetric();
if ((boosting_on_gpu_ && evaluation_on_cuda) || (!boosting_on_gpu_ && !evaluation_on_cuda)) {
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
return metric->Eval(score, objective_function_);
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
} else if (boosting_on_gpu_ && !evaluation_on_cuda) {
const size_t total_size = static_cast<size_t>(num_data) * static_cast<size_t>(num_tree_per_iteration_);
if (total_size > host_score_.size()) {
......@@ -549,7 +549,7 @@ std::vector<double> GBDT::EvalOneMetric(const Metric* metric, const double* scor
CopyFromHostToCUDADevice<double>(cuda_score_.RawData(), score, total_size, __FILE__, __LINE__);
return metric->Eval(cuda_score_.RawData(), objective_function_);
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
}
std::string GBDT::OutputMetric(int iter) {
......@@ -684,14 +684,14 @@ void GBDT::GetPredictAt(int data_idx, double* out_result, int64_t* out_len) {
num_data = valid_score_updater_[used_idx]->num_data();
*out_len = static_cast<int64_t>(num_data) * num_class_;
}
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
std::vector<double> host_raw_scores;
if (boosting_on_gpu_) {
host_raw_scores.resize(static_cast<size_t>(*out_len), 0.0);
CopyFromCUDADeviceToHost<double>(host_raw_scores.data(), raw_scores, static_cast<size_t>(*out_len), __FILE__, __LINE__);
raw_scores = host_raw_scores.data();
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
if (objective_function_ != nullptr) {
#pragma omp parallel for num_threads(OMP_NUM_THREADS()) schedule(static)
for (data_size_t i = 0; i < num_data; ++i) {
......@@ -754,26 +754,26 @@ void GBDT::ResetTrainingData(const Dataset* train_data, const ObjectiveFunction*
}
training_metrics_.shrink_to_fit();
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
boosting_on_gpu_ = objective_function_ != nullptr && objective_function_->IsCUDAObjective() &&
!data_sample_strategy_->IsHessianChange(); // for sample strategy with Hessian change, fall back to boosting on CPU
tree_learner_->ResetBoostingOnGPU(boosting_on_gpu_);
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
if (train_data != train_data_) {
train_data_ = train_data;
data_sample_strategy_->UpdateTrainingData(train_data);
// not same training data, need reset score and others
// create score tracker
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
if (config_->device_type == std::string("cuda")) {
train_score_updater_.reset(new CUDAScoreUpdater(train_data_, num_tree_per_iteration_, boosting_on_gpu_));
} else {
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
train_score_updater_.reset(new ScoreUpdater(train_data_, num_tree_per_iteration_));
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
// update score
for (int i = 0; i < iter_; ++i) {
......@@ -851,7 +851,7 @@ void GBDT::ResetGradientBuffers() {
const bool is_use_subset = data_sample_strategy_->is_use_subset();
const data_size_t bag_data_cnt = data_sample_strategy_->bag_data_cnt();
if (objective_function_ != nullptr) {
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
if (config_->device_type == std::string("cuda") && boosting_on_gpu_) {
if (cuda_gradients_.Size() < total_size) {
cuda_gradients_.Resize(total_size);
......@@ -860,16 +860,16 @@ void GBDT::ResetGradientBuffers() {
gradients_pointer_ = cuda_gradients_.RawData();
hessians_pointer_ = cuda_hessians_.RawData();
} else {
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
if (gradients_.size() < total_size) {
gradients_.resize(total_size);
hessians_.resize(total_size);
}
gradients_pointer_ = gradients_.data();
hessians_pointer_ = hessians_.data();
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
} else if (data_sample_strategy_->IsHessianChange() || (is_use_subset && bag_data_cnt < num_data_ && !boosting_on_gpu_)) {
if (gradients_.size() < total_size) {
gradients_.resize(total_size);
......
src/boosting/gbdt.h
View file @ 20996c92
......@@ -560,7 +560,7 @@ class GBDT : public GBDTBase {
/*! \brief Mutex for exclusive models initialization */
std::mutex instance_mutex_;
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
/*! \brief First order derivative of training data */
std::vector<score_t, CHAllocator<score_t>> gradients_;
/*! \brief Second order derivative of training data */
......@@ -577,7 +577,7 @@ class GBDT : public GBDTBase {
score_t* hessians_pointer_;
/*! \brief Whether boosting is done on GPU, used for device_type=cuda */
bool boosting_on_gpu_;
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
/*! \brief Gradient vector on GPU */
CUDAVector<score_t> cuda_gradients_;
/*! \brief Hessian vector on GPU */
......@@ -586,7 +586,7 @@ class GBDT : public GBDTBase {
mutable std::vector<double> host_score_;
/*! \brief Buffer for scores when boosting is not on GPU but evaluation is, used only with device_type=cuda */
mutable CUDAVector<double> cuda_score_;
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
/*! \brief Number of training data */
data_size_t num_data_;
......
src/boosting/goss.hpp
View file @ 20996c92
......@@ -45,33 +45,33 @@ class GOSSStrategy : public SampleStrategy {
bag_data_cnt_ = left_cnt;
// set bagging data to tree learner
if (!is_use_subset_) {
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
if (config_->device_type == std::string("cuda")) {
CopyFromHostToCUDADevice<data_size_t>(cuda_bag_data_indices_.RawData(), bag_data_indices_.data(), static_cast<size_t>(num_data_), __FILE__, __LINE__);
tree_learner->SetBaggingData(nullptr, cuda_bag_data_indices_.RawData(), bag_data_cnt_);
} else {
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
tree_learner->SetBaggingData(nullptr, bag_data_indices_.data(), bag_data_cnt_);
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
} else {
// get subset
tmp_subset_->ReSize(bag_data_cnt_);
tmp_subset_->CopySubrow(train_data_, bag_data_indices_.data(),
bag_data_cnt_, false);
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
if (config_->device_type == std::string("cuda")) {
CopyFromHostToCUDADevice<data_size_t>(cuda_bag_data_indices_.RawData(), bag_data_indices_.data(), static_cast<size_t>(num_data_), __FILE__, __LINE__);
tree_learner->SetBaggingData(tmp_subset_.get(), cuda_bag_data_indices_.RawData(),
bag_data_cnt_);
} else {
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
tree_learner->SetBaggingData(tmp_subset_.get(), bag_data_indices_.data(),
bag_data_cnt_);
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
}
}
......
src/cuda/cuda_algorithms.cu
View file @ 20996c92
......@@ -4,7 +4,7 @@
* Modifications Copyright(C) 2023 Advanced Micro Devices, Inc. All rights reserved.
*/
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
#include <LightGBM/cuda/cuda_algorithms.hpp>
#include <LightGBM/cuda/cuda_rocm_interop.h>
......@@ -445,4 +445,4 @@ void BitonicArgSortGlobal<data_size_t, int, true>(const data_size_t* values, int
} // namespace LightGBM
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
src/cuda/cuda_utils.cpp
View file @ 20996c92
......@@ -3,7 +3,7 @@
* Licensed under the MIT License. See LICENSE file in the project root for license information.
*/
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
#include <LightGBM/cuda/cuda_rocm_interop.h>
#include <LightGBM/cuda/cuda_utils.hu>
......@@ -35,4 +35,4 @@ int GetCUDADevice(const char* file, int line) {
} // namespace LightGBM
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
src/io/bin.cpp
View file @ 20996c92
......@@ -874,7 +874,7 @@ namespace LightGBM {
return nullptr;
}
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
template <>
const void* MultiValDenseBin<uint8_t>::GetRowWiseData(uint8_t* bit_type,
size_t* total_size,
......@@ -1069,6 +1069,6 @@ namespace LightGBM {
return to_return;
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
} // namespace LightGBM
src/io/cuda/cuda_column_data.cpp
View file @ 20996c92
......@@ -3,7 +3,7 @@
* Licensed under the MIT License. See LICENSE file in the project root for license information.
*/
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
#include <LightGBM/cuda/cuda_column_data.hpp>
......@@ -319,4 +319,4 @@ void CUDAColumnData::InitColumnMetaInfo() {
} // namespace LightGBM
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
src/io/cuda/cuda_column_data.cu
View file @ 20996c92
......@@ -4,7 +4,7 @@
*/
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
#include <LightGBM/cuda/cuda_column_data.hpp>
......@@ -58,4 +58,4 @@ void CUDAColumnData::LaunchCopySubrowKernel(void* const* in_cuda_data_by_column)
} // namespace LightGBM
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
src/io/cuda/cuda_metadata.cpp
View file @ 20996c92
......@@ -3,7 +3,7 @@
* Licensed under the MIT License. See LICENSE file in the project root for license information.
*/
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
#include <LightGBM/cuda/cuda_metadata.hpp>
......@@ -89,4 +89,4 @@ void CUDAMetadata::SetInitScore(const double* init_score, data_size_t len) {
} // namespace LightGBM
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
src/io/cuda/cuda_row_data.cpp
View file @ 20996c92
......@@ -3,7 +3,7 @@
* Licensed under the MIT License. See LICENSE file in the project root for license information.
*/
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
#include <LightGBM/cuda/cuda_row_data.hpp>
......@@ -474,4 +474,4 @@ template const uint64_t* CUDARowData::GetPartitionPtr<uint64_t>() const;
} // namespace LightGBM
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
src/io/cuda/cuda_tree.cpp
View file @ 20996c92
......@@ -3,7 +3,7 @@
* Licensed under the MIT License. See LICENSE file in the project root for license information.
*/
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
#include <LightGBM/cuda/cuda_tree.hpp>
......@@ -338,4 +338,4 @@ void CUDATree::AsConstantTree(double val, int count) {
} // namespace LightGBM
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
src/io/cuda/cuda_tree.cu
View file @ 20996c92
......@@ -4,7 +4,7 @@
*/
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
#include <LightGBM/cuda/cuda_tree.hpp>
......@@ -456,4 +456,4 @@ void CUDATree::LaunchAddPredictionToScoreKernel(
} // namespace LightGBM
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
src/io/dataset.cpp
View file @ 20996c92
......@@ -451,14 +451,14 @@ void Dataset::FinishLoad() {
}
metadata_.FinishLoad();
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
if (device_type_ == std::string("cuda")) {
CreateCUDAColumnData();
metadata_.CreateCUDAMetadata(gpu_device_id_);
} else {
cuda_column_data_.reset(nullptr);
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
is_finish_load_ = true;
}
......@@ -886,7 +886,7 @@ void Dataset::CopySubrow(const Dataset* fullset,
device_type_ = fullset->device_type_;
gpu_device_id_ = fullset->gpu_device_id_;
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
if (device_type_ == std::string("cuda")) {
if (cuda_column_data_ == nullptr) {
cuda_column_data_.reset(new CUDAColumnData(fullset->num_data(), gpu_device_id_));
......@@ -894,7 +894,7 @@ void Dataset::CopySubrow(const Dataset* fullset,
}
cuda_column_data_->CopySubrow(fullset->cuda_column_data(), used_indices, num_used_indices);
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
}
bool Dataset::SetFieldFromArrow(const char* field_name, const ArrowChunkedArray &ca) {
......@@ -1735,13 +1735,13 @@ void Dataset::AddFeaturesFrom(Dataset* other) {
raw_data_.push_back(other->raw_data_[i]);
}
}
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
if (device_type_ == std::string("cuda")) {
CreateCUDAColumnData();
} else {
cuda_column_data_ = nullptr;
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
}
const void* Dataset::GetColWiseData(
......@@ -1763,7 +1763,7 @@ const void* Dataset::GetColWiseData(
return feature_groups_[feature_group_index]->GetColWiseData(sub_feature_index, bit_type, is_sparse, bin_iterator);
}
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
void Dataset::CreateCUDAColumnData() {
cuda_column_data_.reset(new CUDAColumnData(num_data_, gpu_device_id_));
int num_columns = 0;
......@@ -1898,6 +1898,6 @@ void Dataset::CreateCUDAColumnData() {
feature_to_column);
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
} // namespace LightGBM
src/io/dataset_loader.cpp
View file @ 20996c92
......@@ -279,14 +279,14 @@ Dataset* DatasetLoader::LoadFromFile(const char* filename, int rank, int num_mac
dataset->device_type_ = config_.device_type;
dataset->gpu_device_id_ = config_.gpu_device_id;
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
if (config_.device_type == std::string("cuda")) {
dataset->CreateCUDAColumnData();
dataset->metadata_.CreateCUDAMetadata(dataset->gpu_device_id_);
} else {
dataset->cuda_column_data_ = nullptr;
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
}
// check meta data
dataset->metadata_.CheckOrPartition(num_global_data, used_data_indices);
......
src/io/dense_bin.hpp
View file @ 20996c92
......@@ -607,7 +607,7 @@ class DenseBin : public Bin {
private:
data_size_t num_data_;
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
std::vector<VAL_T, CHAllocator<VAL_T>> data_;
#else
std::vector<VAL_T, Common::AlignmentAllocator<VAL_T, kAlignedSize>> data_;
......
src/io/metadata.cpp
View file @ 20996c92
......@@ -21,9 +21,9 @@ Metadata::Metadata() {
position_load_from_file_ = false;
query_load_from_file_ = false;
init_score_load_from_file_ = false;
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
cuda_metadata_ = nullptr;
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
}
void Metadata::Init(const char* data_filename) {
......@@ -380,11 +380,11 @@ void Metadata::SetInitScoresFromIterator(It first, It last) {
}
init_score_load_from_file_ = false;
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
if (cuda_metadata_ != nullptr) {
cuda_metadata_->SetInitScore(init_score_.data(), init_score_.size());
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
}
void Metadata::SetInitScore(const double* init_score, data_size_t len) {
......@@ -434,11 +434,11 @@ void Metadata::SetLabelsFromIterator(It first, It last) {
label_[i] = Common::AvoidInf(first[i]);
}
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
if (cuda_metadata_ != nullptr) {
cuda_metadata_->SetLabel(label_.data(), label_.size());
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
}
void Metadata::SetLabel(const label_t* label, data_size_t len) {
......@@ -492,11 +492,11 @@ void Metadata::SetWeightsFromIterator(It first, It last) {
CalculateQueryWeights();
weight_load_from_file_ = false;
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
if (cuda_metadata_ != nullptr) {
cuda_metadata_->SetWeights(weights_.data(), weights_.size());
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
}
void Metadata::SetWeights(const label_t* weights, data_size_t len) {
......@@ -555,7 +555,7 @@ void Metadata::SetQueriesFromIterator(It first, It last) {
CalculateQueryWeights();
query_load_from_file_ = false;
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
if (cuda_metadata_ != nullptr) {
if (query_weights_.size() > 0) {
CHECK_EQ(query_weights_.size(), static_cast<size_t>(num_queries_));
......@@ -564,7 +564,7 @@ void Metadata::SetQueriesFromIterator(It first, It last) {
cuda_metadata_->SetQuery(query_boundaries_.data(), nullptr, num_queries_);
}
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
}
void Metadata::SetQuery(const data_size_t* query, data_size_t len) {
......@@ -583,9 +583,9 @@ void Metadata::SetPosition(const data_size_t* positions, data_size_t len) {
num_positions_ = 0;
return;
}
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
Log::Fatal("Positions in learning to rank is not supported in CUDA version yet.");
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
if (num_data_ != len) {
Log::Fatal("Positions size (%i) doesn't match data size (%i)", len, num_data_);
}
......@@ -788,12 +788,12 @@ void Metadata::FinishLoad() {
CalculateQueryBoundaries();
}
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
void Metadata::CreateCUDAMetadata(const int gpu_device_id) {
cuda_metadata_.reset(new CUDAMetadata(gpu_device_id));
cuda_metadata_->Init(label_, weights_, query_boundaries_, query_weights_, init_score_);
}
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
void Metadata::LoadFromMemory(const void* memory) {
const char* mem_ptr = reinterpret_cast<const char*>(memory);
......
src/io/multi_val_dense_bin.hpp
View file @ 20996c92
......@@ -328,13 +328,13 @@ class MultiValDenseBin : public MultiValBin {
MultiValDenseBin<VAL_T>* Clone() override;
#if defined(USE_CUDA) || defined(USE_ROCM)
#ifdef USE_CUDA
const void* GetRowWiseData(uint8_t* bit_type,
size_t* total_size,
bool* is_sparse,
const void** out_data_ptr,
uint8_t* data_ptr_bit_type) const override;
#endif // USE_CUDA || USE_ROCM
#endif // USE_CUDA
private:
data_size_t num_data_;
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment