Objective function evaluation on GPU with minimal PCIe transfers (#2935)

* Added GPU objective function and no-copy interface.

- xgboost::HostDeviceVector<T> syncs automatically between host and device
- no-copy interfaces have been added
- default implementations just sync the data to host
  and call the implementations with std::vector
- GPU objective function, predictor, histogram updater process data
  directly on GPU

src/objective/regression_loss.h

@@ -0,0 +1,110 @@
+/*!
+ * Copyright 2017 XGBoost contributors
+ */
+#ifndef XGBOOST_OBJECTIVE_REGRESSION_LOSS_H_
+#define XGBOOST_OBJECTIVE_REGRESSION_LOSS_H_
+
+#include <dmlc/omp.h>
+#include <xgboost/logging.h>
+#include <algorithm>
+#include "../common/math.h"
+
+namespace xgboost {
+namespace obj {
+
+// common regressions
+// linear regression
+struct LinearSquareLoss {
+  // duplication is necessary, as __device__ specifier
+  // cannot be made conditional on template parameter
+  XGBOOST_DEVICE static bst_float PredTransform(bst_float x) { return x; }
+  XGBOOST_DEVICE static bool CheckLabel(bst_float x) { return true; }
+  XGBOOST_DEVICE static bst_float FirstOrderGradient(bst_float predt, bst_float label) {
+    return predt - label;
+  }
+  XGBOOST_DEVICE static bst_float SecondOrderGradient(bst_float predt, bst_float label) {
+    return 1.0f;
+  }
+  template <typename T>
+  static T PredTransform(T x) { return x; }
+  template <typename T>
+  static T FirstOrderGradient(T predt, T label) { return predt - label; }
+  template <typename T>
+  static T SecondOrderGradient(T predt, T label) { return T(1.0f); }
+  static bst_float ProbToMargin(bst_float base_score) { return base_score; }
+  static const char* LabelErrorMsg() { return ""; }
+  static const char* DefaultEvalMetric() { return "rmse"; }
+};
+
+// logistic loss for probability regression task
+struct LogisticRegression {
+  // duplication is necessary, as __device__ specifier
+  // cannot be made conditional on template parameter
+  XGBOOST_DEVICE static bst_float PredTransform(bst_float x) { return common::Sigmoid(x); }
+  XGBOOST_DEVICE static bool CheckLabel(bst_float x) { return x >= 0.0f && x <= 1.0f; }
+  XGBOOST_DEVICE static bst_float FirstOrderGradient(bst_float predt, bst_float label) {
+    return predt - label;
+  }
+  XGBOOST_DEVICE static bst_float SecondOrderGradient(bst_float predt, bst_float label) {
+    const float eps = 1e-16f;
+    return fmaxf(predt * (1.0f - predt), eps);
+  }
+  template <typename T>
+  static T PredTransform(T x) { return common::Sigmoid(x); }
+  template <typename T>
+  static T FirstOrderGradient(T predt, T label) { return predt - label; }
+  template <typename T>
+  static T SecondOrderGradient(T predt, T label) {
+    const T eps = T(1e-16f);
+    return std::max(predt * (T(1.0f) - predt), eps);
+  }
+  static bst_float ProbToMargin(bst_float base_score) {
+    CHECK(base_score > 0.0f && base_score < 1.0f)
+      << "base_score must be in (0,1) for logistic loss";
+    return -logf(1.0f / base_score - 1.0f);
+  }
+  static const char* LabelErrorMsg() {
+    return "label must be in [0,1] for logistic regression";
+  }
+  static const char* DefaultEvalMetric() { return "rmse"; }
+};
+
+// logistic loss for binary classification task
+struct LogisticClassification : public LogisticRegression {
+  static const char* DefaultEvalMetric() { return "error"; }
+};
+
+// logistic loss, but predict un-transformed margin
+struct LogisticRaw : public LogisticRegression {
+  // duplication is necessary, as __device__ specifier
+  // cannot be made conditional on template parameter
+  XGBOOST_DEVICE static bst_float PredTransform(bst_float x) { return x; }
+  XGBOOST_DEVICE static bst_float FirstOrderGradient(bst_float predt, bst_float label) {
+    predt = common::Sigmoid(predt);
+    return predt - label;
+  }
+  XGBOOST_DEVICE static bst_float SecondOrderGradient(bst_float predt, bst_float label) {
+    const float eps = 1e-16f;
+    predt = common::Sigmoid(predt);
+    return fmaxf(predt * (1.0f - predt), eps);
+  }
+  template <typename T>
+    static T PredTransform(T x) { return x; }
+  template <typename T>
+    static T FirstOrderGradient(T predt, T label) {
+    predt = common::Sigmoid(predt);
+    return predt - label;
+  }
+  template <typename T>
+    static T SecondOrderGradient(T predt, T label) {
+    const T eps = T(1e-16f);
+    predt = common::Sigmoid(predt);
+    return std::max(predt * (T(1.0f) - predt), eps);
+  }
+  static const char* DefaultEvalMetric() { return "auc"; }
+};
+
+}  // namespace obj
+}  // namespace xgboost
+
+#endif  // XGBOOST_OBJECTIVE_REGRESSION_LOSS_H_