[CUDA] consolidate CUDA versions (#5677)

* [ci] speed up if-else, swig, and lint conda setup

* add 'source activate'

* python constraint

* start removing cuda v1

* comment out CI

* remove more references

* revert some unnecessaary changes

* revert a few more mistakes

* revert another change that ignored params

* sigh

* remove CUDATreeLearner

* fix tests, docs

* fix quoting in setup.py

* restore all CI

* Apply suggestions from code review
Co-authored-by: shiyu1994 <shiyu_k1994@qq.com>

* Apply suggestions from code review

* completely remove cuda_exp, update docs

---------
Co-authored-by: shiyu1994 <shiyu_k1994@qq.com>

src/objective/cuda/cuda_binary_objective.cpp

@@ -4,7 +4,7 @@
  * license information.
  */
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #include "cuda_binary_objective.hpp"
 
@@ -61,4 +61,4 @@ void CUDABinaryLogloss::Init(const Metadata& metadata, data_size_t num_data) {
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA

src/objective/cuda/cuda_binary_objective.cu

@@ -4,7 +4,7 @@
  * license information.
  */
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #include <algorithm>
 
@@ -206,4 +206,4 @@ void CUDABinaryLogloss::LaunchResetOVACUDALabelKernel() const {
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA

src/objective/cuda/cuda_binary_objective.hpp

@@ -7,7 +7,7 @@
 #ifndef LIGHTGBM_OBJECTIVE_CUDA_CUDA_BINARY_OBJECTIVE_HPP_
 #define LIGHTGBM_OBJECTIVE_CUDA_CUDA_BINARY_OBJECTIVE_HPP_
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #define GET_GRADIENTS_BLOCK_SIZE_BINARY (1024)
 #define CALC_INIT_SCORE_BLOCK_SIZE_BINARY (1024)
@@ -58,6 +58,6 @@ class CUDABinaryLogloss : public CUDAObjectiveInterface<BinaryLogloss> {
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA
 
 #endif  // LIGHTGBM_OBJECTIVE_CUDA_CUDA_BINARY_OBJECTIVE_HPP_

src/objective/cuda/cuda_multiclass_objective.cpp

@@ -3,7 +3,7 @@
  * Licensed under the MIT License. See LICENSE file in the project root for license information.
  */
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #include "cuda_multiclass_objective.hpp"
 
@@ -59,4 +59,4 @@ const double* CUDAMulticlassOVA::ConvertOutputCUDA(const data_size_t num_data, c
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA

src/objective/cuda/cuda_multiclass_objective.cu

@@ -3,7 +3,7 @@
  * Licensed under the MIT License. See LICENSE file in the project root for license information.
  */
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #include <algorithm>
 
@@ -105,4 +105,4 @@ const double* CUDAMulticlassSoftmax::LaunchConvertOutputCUDAKernel(
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA

src/objective/cuda/cuda_multiclass_objective.hpp

@@ -5,7 +5,7 @@
 #ifndef LIGHTGBM_OBJECTIVE_CUDA_CUDA_MULTICLASS_OBJECTIVE_HPP_
 #define LIGHTGBM_OBJECTIVE_CUDA_CUDA_MULTICLASS_OBJECTIVE_HPP_
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #include <LightGBM/cuda/cuda_objective_function.hpp>
 
@@ -74,5 +74,5 @@ class CUDAMulticlassOVA: public CUDAObjectiveInterface<MulticlassOVA> {
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA
 #endif  // LIGHTGBM_OBJECTIVE_CUDA_CUDA_MULTICLASS_OBJECTIVE_HPP_

src/objective/cuda/cuda_rank_objective.cpp

@@ -4,7 +4,7 @@
  * license information.
  */
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #include <string>
 #include <vector>
@@ -64,4 +64,4 @@ void CUDARankXENDCG::GenerateItemRands() const {
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA

src/objective/cuda/cuda_rank_objective.cu

@@ -4,7 +4,7 @@
  * license information.
  */
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #include "cuda_rank_objective.hpp"
 
@@ -658,4 +658,4 @@ void CUDARankXENDCG::LaunchGetGradientsKernel(const double* score, score_t* grad
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA

src/objective/cuda/cuda_rank_objective.hpp

@@ -7,7 +7,7 @@
 #ifndef LIGHTGBM_OBJECTIVE_CUDA_CUDA_RANK_OBJECTIVE_HPP_
 #define LIGHTGBM_OBJECTIVE_CUDA_CUDA_RANK_OBJECTIVE_HPP_
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #define NUM_QUERY_PER_BLOCK (10)
 
@@ -118,5 +118,5 @@ class CUDARankXENDCG : public CUDALambdaRankObjectiveInterface<RankXENDCG> {
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA
 #endif  // LIGHTGBM_OBJECTIVE_CUDA_CUDA_RANK_OBJECTIVE_HPP_

src/objective/cuda/cuda_regression_objective.cpp

@@ -4,7 +4,7 @@
  * license information.
  */
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #include "cuda_regression_objective.hpp"
 
@@ -85,4 +85,4 @@ double CUDARegressionPoissonLoss::LaunchCalcInitScoreKernel(const int class_id)
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA

src/objective/cuda/cuda_regression_objective.cu

@@ -4,7 +4,7 @@
  * license information.
  */
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #include "cuda_regression_objective.hpp"
 #include <LightGBM/cuda/cuda_algorithms.hpp>
@@ -353,4 +353,4 @@ const double* CUDARegressionPoissonLoss::LaunchConvertOutputCUDAKernel(const dat
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA

src/objective/cuda/cuda_regression_objective.hpp

@@ -7,7 +7,7 @@
 #ifndef LIGHTGBM_OBJECTIVE_CUDA_CUDA_REGRESSION_OBJECTIVE_HPP_
 #define LIGHTGBM_OBJECTIVE_CUDA_CUDA_REGRESSION_OBJECTIVE_HPP_
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #define GET_GRADIENTS_BLOCK_SIZE_REGRESSION (1024)
 
@@ -135,5 +135,5 @@ class CUDARegressionPoissonLoss : public CUDARegressionObjectiveInterface<Regres
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA
 #endif  // LIGHTGBM_OBJECTIVE_CUDA_CUDA_REGRESSION_OBJECTIVE_HPP_

src/objective/objective_function.cpp

@@ -18,8 +18,8 @@
 namespace LightGBM {
 
 ObjectiveFunction* ObjectiveFunction::CreateObjectiveFunction(const std::string& type, const Config& config) {
-  #ifdef USE_CUDA_EXP
-  if (config.device_type == std::string("cuda_exp") &&
+  #ifdef USE_CUDA
+  if (config.device_type == std::string("cuda") &&
       config.data_sample_strategy != std::string("goss") &&
       config.boosting != std::string("rf")) {
     if (type == std::string("regression")) {
@@ -27,7 +27,7 @@ ObjectiveFunction* ObjectiveFunction::CreateObjectiveFunction(const std::string&
     } else if (type == std::string("regression_l1")) {
       return new CUDARegressionL1loss(config);
     } else if (type == std::string("quantile")) {
-      Log::Warning("Objective quantile is not implemented in cuda_exp version. Fall back to boosting on CPU.");
+      Log::Warning("Objective quantile is not implemented in cuda version. Fall back to boosting on CPU.");
       return new RegressionQuantileloss(config);
     } else if (type == std::string("huber")) {
       return new CUDARegressionHuberLoss(config);
@@ -46,26 +46,26 @@ ObjectiveFunction* ObjectiveFunction::CreateObjectiveFunction(const std::string&
     } else if (type == std::string("multiclassova")) {
       return new CUDAMulticlassOVA(config);
     } else if (type == std::string("cross_entropy")) {
-      Log::Warning("Objective cross_entropy is not implemented in cuda_exp version. Fall back to boosting on CPU.");
+      Log::Warning("Objective cross_entropy is not implemented in cuda version. Fall back to boosting on CPU.");
       return new CrossEntropy(config);
     } else if (type == std::string("cross_entropy_lambda")) {
-      Log::Warning("Objective cross_entropy_lambda is not implemented in cuda_exp version. Fall back to boosting on CPU.");
+      Log::Warning("Objective cross_entropy_lambda is not implemented in cuda version. Fall back to boosting on CPU.");
       return new CrossEntropyLambda(config);
     } else if (type == std::string("mape")) {
-      Log::Warning("Objective mape is not implemented in cuda_exp version. Fall back to boosting on CPU.");
+      Log::Warning("Objective mape is not implemented in cuda version. Fall back to boosting on CPU.");
       return new RegressionMAPELOSS(config);
     } else if (type == std::string("gamma")) {
-      Log::Warning("Objective gamma is not implemented in cuda_exp version. Fall back to boosting on CPU.");
+      Log::Warning("Objective gamma is not implemented in cuda version. Fall back to boosting on CPU.");
       return new RegressionGammaLoss(config);
     } else if (type == std::string("tweedie")) {
-      Log::Warning("Objective tweedie is not implemented in cuda_exp version. Fall back to boosting on CPU.");
+      Log::Warning("Objective tweedie is not implemented in cuda version. Fall back to boosting on CPU.");
       return new RegressionTweedieLoss(config);
     } else if (type == std::string("custom")) {
-      Log::Warning("Using customized objective with cuda_exp. This requires copying gradients from CPU to GPU, which can be slow.");
+      Log::Warning("Using customized objective with cuda. This requires copying gradients from CPU to GPU, which can be slow.");
       return nullptr;
     }
   } else {
-  #endif  // USE_CUDA_EXP
+  #endif  // USE_CUDA
     if (type == std::string("regression")) {
       return new RegressionL2loss(config);
     } else if (type == std::string("regression_l1")) {
@@ -101,9 +101,9 @@ ObjectiveFunction* ObjectiveFunction::CreateObjectiveFunction(const std::string&
     } else if (type == std::string("custom")) {
       return nullptr;
     }
-  #ifdef USE_CUDA_EXP
+  #ifdef USE_CUDA
   }
-  #endif  // USE_CUDA_EXP
+  #endif  // USE_CUDA
   Log::Fatal("Unknown objective type name: %s", type.c_str());
   return nullptr;
 }

src/treelearner/cuda/cuda_best_split_finder.cpp

@@ -4,7 +4,7 @@
  * license information.
  */
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #include <algorithm>
 
@@ -383,4 +383,4 @@ void CUDABestSplitFinder::SetUsedFeatureByNode(const std::vector<int8_t>& is_fea
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA

src/treelearner/cuda/cuda_best_split_finder.cu

@@ -4,7 +4,7 @@
  * license information.
  */
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #include <algorithm>
 
@@ -1802,4 +1802,4 @@ void CUDABestSplitFinder::LaunchInitCUDARandomKernel() {
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA

src/treelearner/cuda/cuda_best_split_finder.hpp

@@ -7,7 +7,7 @@
 #ifndef LIGHTGBM_TREELEARNER_CUDA_CUDA_BEST_SPLIT_FINDER_HPP_
 #define LIGHTGBM_TREELEARNER_CUDA_CUDA_BEST_SPLIT_FINDER_HPP_
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #include <LightGBM/bin.h>
 #include <LightGBM/dataset.h>
@@ -211,5 +211,5 @@ class CUDABestSplitFinder {
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA
 #endif  // LIGHTGBM_TREELEARNER_CUDA_CUDA_BEST_SPLIT_FINDER_HPP_

src/treelearner/cuda/cuda_data_partition.cpp

@@ -4,7 +4,7 @@
  * license information.
  */
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #include <algorithm>
 #include <memory>
@@ -370,4 +370,4 @@ void CUDADataPartition::ResetByLeafPred(const std::vector<int>& leaf_pred, int n
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA

src/treelearner/cuda/cuda_data_partition.cu

@@ -4,7 +4,7 @@
  * license information.
  */
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #include "cuda_data_partition.hpp"
 
@@ -1071,4 +1071,4 @@ void CUDADataPartition::LaunchAddPredictionToScoreKernel(const double* leaf_valu
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA

src/treelearner/cuda/cuda_data_partition.hpp

@@ -6,7 +6,7 @@
 #ifndef LIGHTGBM_TREELEARNER_CUDA_CUDA_DATA_PARTITION_HPP_
 #define LIGHTGBM_TREELEARNER_CUDA_CUDA_DATA_PARTITION_HPP_
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #include <LightGBM/bin.h>
 #include <LightGBM/meta.h>
@@ -384,5 +384,5 @@ class CUDADataPartition {
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA
 #endif  // LIGHTGBM_TREELEARNER_CUDA_CUDA_DATA_PARTITION_HPP_

src/treelearner/cuda/cuda_histogram_constructor.cpp

@@ -4,7 +4,7 @@
  * license information.
  */
 
-#ifdef USE_CUDA_EXP
+#ifdef USE_CUDA
 
 #include "cuda_histogram_constructor.hpp"
 
@@ -193,4 +193,4 @@ void CUDAHistogramConstructor::ResetConfig(const Config* config) {
 
 }  // namespace LightGBM
 
-#endif  // USE_CUDA_EXP
+#endif  // USE_CUDA