[backport] Fix device dispatch for linear updater. (dmlc#9507) (dmlc#9532)

Author: trivialfis

doc/parameter.rst

@@ -329,7 +329,7 @@ Parameters for Linear Booster (``booster=gblinear``)
   - Choice of algorithm to fit linear model
 
     - ``shotgun``: Parallel coordinate descent algorithm based on shotgun algorithm. Uses 'hogwild' parallelism and therefore produces a nondeterministic solution on each run.
-    - ``coord_descent``: Ordinary coordinate descent algorithm. Also multithreaded but still produces a deterministic solution.
+    - ``coord_descent``: Ordinary coordinate descent algorithm. Also multithreaded but still produces a deterministic solution. When the ``device`` parameter is set to ``cuda`` or ``gpu``, a GPU variant would be used.
 
 * ``feature_selector`` [default= ``cyclic``]
 

include/xgboost/global_config.h

@@ -1,30 +1,28 @@
-/*!
- * Copyright 2020 by Contributors
+/**
+ * Copyright 2020-2023, XGBoost Contributors
  * \file global_config.h
  * \brief Global configuration for XGBoost
  * \author Hyunsu Cho
  */
 #ifndef XGBOOST_GLOBAL_CONFIG_H_
 #define XGBOOST_GLOBAL_CONFIG_H_
 
-#include <xgboost/parameter.h>
-#include <vector>
-#include <string>
+#include <dmlc/thread_local.h>  // for ThreadLocalStore
+#include <xgboost/parameter.h>  // for XGBoostParameter
 
-namespace xgboost {
-class Json;
+#include <cstdint>  // for int32_t
 
+namespace xgboost {
 struct GlobalConfiguration : public XGBoostParameter<GlobalConfiguration> {
-  int verbosity { 1 };
-  bool use_rmm { false };
+  std::int32_t verbosity{1};
+  bool use_rmm{false};
   DMLC_DECLARE_PARAMETER(GlobalConfiguration) {
     DMLC_DECLARE_FIELD(verbosity)
         .set_range(0, 3)
         .set_default(1)  // shows only warning
         .describe("Flag to print out detailed breakdown of runtime.");
-    DMLC_DECLARE_FIELD(use_rmm)
-        .set_default(false)
-        .describe("Whether to use RAPIDS Memory Manager to allocate GPU memory in XGBoost");
+    DMLC_DECLARE_FIELD(use_rmm).set_default(false).describe(
+        "Whether to use RAPIDS Memory Manager to allocate GPU memory in XGBoost");
   }
 };
 

src/gbm/gblinear.cc

@@ -1,5 +1,5 @@
-/*!
- * Copyright 2014-2022 by XGBoost Contributors
+/**
+ * Copyright 2014-2023, XGBoost Contributors
  * \file gblinear.cc
  * \brief Implementation of Linear booster, with L1/L2 regularization: Elastic Net
  *        the update rule is parallel coordinate descent (shotgun)
@@ -26,9 +26,9 @@
 #include "../common/timer.h"
 #include "../common/common.h"
 #include "../common/threading_utils.h"
+#include "../common/error_msg.h"
 
-namespace xgboost {
-namespace gbm {
+namespace xgboost::gbm {
 
 DMLC_REGISTRY_FILE_TAG(gblinear);
 
@@ -83,7 +83,16 @@ class GBLinear : public GradientBooster {
     }
     param_.UpdateAllowUnknown(cfg);
     param_.CheckGPUSupport();
-    updater_.reset(LinearUpdater::Create(param_.updater, ctx_));
+    if (param_.updater == "gpu_coord_descent") {
+      LOG(WARNING) << error::DeprecatedFunc("gpu_coord_descent", "2.0.0",
+                                            R"(device="cuda", updater="coord_descent")");
+    }
+
+    if (param_.updater == "coord_descent" && ctx_->IsCUDA()) {
+      updater_.reset(LinearUpdater::Create("gpu_coord_descent", ctx_));
+    } else {
+      updater_.reset(LinearUpdater::Create(param_.updater, ctx_));
+    }
     updater_->Configure(cfg);
     monitor_.Init("GBLinear");
   }
@@ -354,5 +363,4 @@ XGBOOST_REGISTER_GBM(GBLinear, "gblinear")
     .set_body([](LearnerModelParam const* booster_config, Context const* ctx) {
       return new GBLinear(booster_config, ctx);
     });
-}  // namespace gbm
-}  // namespace xgboost
+}  // namespace xgboost::gbm

src/linear/updater_coordinate.cc

@@ -9,8 +9,7 @@
 #include "coordinate_common.h"
 #include "xgboost/json.h"
 
-namespace xgboost {
-namespace linear {
+namespace xgboost::linear {
 
 DMLC_REGISTER_PARAMETER(CoordinateParam);
 DMLC_REGISTRY_FILE_TAG(updater_coordinate);
@@ -39,8 +38,9 @@ class CoordinateUpdater : public LinearUpdater {
     FromJson(config.at("linear_train_param"), &tparam_);
     FromJson(config.at("coordinate_param"), &cparam_);
   }
-  void SaveConfig(Json* p_out) const override {
-    auto& out = *p_out;
+  void SaveConfig(Json *p_out) const override {
+    LOG(DEBUG) << "Save config for CPU updater.";
+    auto &out = *p_out;
     out["linear_train_param"] = ToJson(tparam_);
     out["coordinate_param"] = ToJson(cparam_);
   }
@@ -99,5 +99,4 @@ class CoordinateUpdater : public LinearUpdater {
 XGBOOST_REGISTER_LINEAR_UPDATER(CoordinateUpdater, "coord_descent")
     .describe("Update linear model according to coordinate descent algorithm.")
     .set_body([]() { return new CoordinateUpdater(); });
-}  // namespace linear
-}  // namespace xgboost
+}  // namespace xgboost::linear

src/linear/updater_gpu_coordinate.cu

@@ -15,8 +15,7 @@
 #include "../common/timer.h"
 #include "./param.h"
 
-namespace xgboost {
-namespace linear {
+namespace xgboost::linear {
 
 DMLC_REGISTRY_FILE_TAG(updater_gpu_coordinate);
 
@@ -29,7 +28,7 @@ DMLC_REGISTRY_FILE_TAG(updater_gpu_coordinate);
 class GPUCoordinateUpdater : public LinearUpdater {  // NOLINT
  public:
   // set training parameter
-  void Configure(Args const& args) override {
+  void Configure(Args const &args) override {
     tparam_.UpdateAllowUnknown(args);
     coord_param_.UpdateAllowUnknown(args);
     selector_.reset(FeatureSelector::Create(tparam_.feature_selector));
@@ -41,8 +40,9 @@ class GPUCoordinateUpdater : public LinearUpdater {  // NOLINT
     FromJson(config.at("linear_train_param"), &tparam_);
     FromJson(config.at("coordinate_param"), &coord_param_);
   }
-  void SaveConfig(Json* p_out) const override {
-    auto& out = *p_out;
+  void SaveConfig(Json *p_out) const override {
+    LOG(DEBUG) << "Save config for GPU updater.";
+    auto &out = *p_out;
     out["linear_train_param"] = ToJson(tparam_);
     out["coordinate_param"] = ToJson(coord_param_);
   }
@@ -101,10 +101,9 @@ class GPUCoordinateUpdater : public LinearUpdater {  // NOLINT
     monitor_.Stop("LazyInitDevice");
 
     monitor_.Start("UpdateGpair");
-    auto &in_gpair_host = in_gpair->ConstHostVector();
     // Update gpair
     if (ctx_->gpu_id >= 0) {
-      this->UpdateGpair(in_gpair_host);
+      this->UpdateGpair(in_gpair->ConstHostVector());
     }
     monitor_.Stop("UpdateGpair");
 
@@ -249,5 +248,4 @@ XGBOOST_REGISTER_LINEAR_UPDATER(GPUCoordinateUpdater, "gpu_coord_descent")
         "Update linear model according to coordinate descent algorithm. GPU "
         "accelerated.")
     .set_body([]() { return new GPUCoordinateUpdater(); });
-}  // namespace linear
-}  // namespace xgboost
+}  // namespace xgboost::linear

tests/cpp/gbm/test_gblinear.cu

@@ -0,0 +1,42 @@
+/**
+ * Copyright 2023, XGBoost Contributors
+ */
+#include <gtest/gtest.h>
+#include <xgboost/global_config.h>  // for GlobalConfigThreadLocalStore
+#include <xgboost/json.h>           // for Json, Object
+#include <xgboost/learner.h>        // for Learner
+
+#include <algorithm>  // for transform
+#include <string>     // for string
+#include <utility>    // for swap
+
+#include "../helpers.h"  // for RandomDataGenerator
+
+namespace xgboost {
+TEST(GBlinear, DispatchUpdater) {
+  auto verbosity = 3;
+  std::swap(GlobalConfigThreadLocalStore::Get()->verbosity, verbosity);
+
+  auto test = [](std::string device) {
+    auto p_fmat = RandomDataGenerator{10, 10, 0.0f}.GenerateDMatrix(true);
+    std::unique_ptr<Learner> learner{Learner::Create({p_fmat})};
+    learner->SetParams(
+        Args{{"booster", "gblinear"}, {"updater", "coord_descent"}, {"device", device}});
+    learner->Configure();
+    for (std::int32_t iter = 0; iter < 3; ++iter) {
+      learner->UpdateOneIter(iter, p_fmat);
+    }
+    Json config{Object{}};
+    ::testing::internal::CaptureStderr();
+    learner->SaveConfig(&config);
+    auto str = ::testing::internal::GetCapturedStderr();
+    std::transform(device.cbegin(), device.cend(), device.begin(),
+                   [](char c) { return std::toupper(c); });
+    ASSERT_NE(str.find(device), std::string::npos);
+  };
+  test("cpu");
+  test("gpu");
+
+  std::swap(GlobalConfigThreadLocalStore::Get()->verbosity, verbosity);
+}
+}  // namespace xgboost