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

Reorder if-else statements to allow using of cpu branches for sycl-devices (#9682)

Browse Source
This commit is contained in:
Dmitry Razdoburdin
2023-10-18 04:55:33 +02:00
committed by GitHub
parent 4c0e4422d0
commit ea9f09716b
14 changed files with 99 additions and 98 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
src/objective/adaptive.h
View File

@@ -96,13 +96,13 @@ void UpdateTreeLeafHost(Context const* ctx, std::vector<bst_node_t> const& posit
inline void UpdateTreeLeaf(Context const* ctx, HostDeviceVector<bst_node_t> const& position,
std::int32_t group_idx, MetaInfo const& info, float learning_rate,
HostDeviceVector<float> const& predt, float alpha, RegTree* p_tree) {
if (ctx->IsCPU()) {
detail::UpdateTreeLeafHost(ctx, position.ConstHostVector(), group_idx, info, learning_rate,
predt, alpha, p_tree);
} else {
if (ctx->IsCUDA()) {
position.SetDevice(ctx->Device());
detail::UpdateTreeLeafDevice(ctx, position.ConstDeviceSpan(), group_idx, info, learning_rate,
predt, alpha, p_tree);
} else {
detail::UpdateTreeLeafHost(ctx, position.ConstHostVector(), group_idx, info, learning_rate,
predt, alpha, p_tree);
}
}
} // namespace obj

10
src/objective/lambdarank_obj.cc
View File

@@ -108,14 +108,14 @@ class LambdaRankObj : public FitIntercept {
li_.SetDevice(ctx_->Device());
lj_.SetDevice(ctx_->Device());
if (ctx_->IsCPU()) {
cpu_impl::LambdaRankUpdatePositionBias(ctx_, li_full_.View(ctx_->Device()),
lj_full_.View(ctx_->Device()), &ti_plus_, &tj_minus_,
&li_, &lj_, p_cache_);
} else {
if (ctx_->IsCUDA()) {
cuda_impl::LambdaRankUpdatePositionBias(ctx_, li_full_.View(ctx_->Device()),
lj_full_.View(ctx_->Device()), &ti_plus_, &tj_minus_,
&li_, &lj_, p_cache_);
} else {
cpu_impl::LambdaRankUpdatePositionBias(ctx_, li_full_.View(ctx_->Device()),
lj_full_.View(ctx_->Device()), &ti_plus_, &tj_minus_,
&li_, &lj_, p_cache_);
}
li_full_.Data()->Fill(0.0);

48
src/objective/quantile_obj.cu
View File

@@ -71,15 +71,15 @@ class QuantileRegression : public ObjFunction {
auto gpair = out_gpair->View(ctx_->Device());
info.weights_.SetDevice(ctx_->Device());
common::OptionalWeights weight{ctx_->IsCPU() ? info.weights_.ConstHostSpan()
: info.weights_.ConstDeviceSpan()};
common::OptionalWeights weight{ctx_->IsCUDA() ? info.weights_.ConstDeviceSpan()
: info.weights_.ConstHostSpan()};
preds.SetDevice(ctx_->Device());
auto predt = linalg::MakeVec(&preds);
auto n_samples = info.num_row_;
alpha_.SetDevice(ctx_->Device());
auto alpha = ctx_->IsCPU() ? alpha_.ConstHostSpan() : alpha_.ConstDeviceSpan();
auto alpha = ctx_->IsCUDA() ? alpha_.ConstDeviceSpan() : alpha_.ConstHostSpan();
linalg::ElementWiseKernel(
ctx_, gpair, [=] XGBOOST_DEVICE(std::size_t i, GradientPair const&) mutable {
@@ -107,27 +107,7 @@ class QuantileRegression : public ObjFunction {
base_score->Reshape(n_targets);
double sw{0};
if (ctx_->IsCPU()) {
auto quantiles = base_score->HostView();
auto h_weights = info.weights_.ConstHostVector();
if (info.weights_.Empty()) {
sw = info.num_row_;
} else {
sw = std::accumulate(std::cbegin(h_weights), std::cend(h_weights), 0.0);
}
for (bst_target_t t{0}; t < n_targets; ++t) {
auto alpha = param_.quantile_alpha[t];
auto h_labels = info.labels.HostView();
if (h_weights.empty()) {
quantiles(t) =
common::Quantile(ctx_, alpha, linalg::cbegin(h_labels), linalg::cend(h_labels));
} else {
CHECK_EQ(h_weights.size(), h_labels.Size());
quantiles(t) = common::WeightedQuantile(ctx_, alpha, linalg::cbegin(h_labels),
linalg::cend(h_labels), std::cbegin(h_weights));
}
}
} else {
if (ctx_->IsCUDA()) {
#if defined(XGBOOST_USE_CUDA)
alpha_.SetDevice(ctx_->Device());
auto d_alpha = alpha_.ConstDeviceSpan();
@@ -164,6 +144,26 @@ class QuantileRegression : public ObjFunction {
#else
common::AssertGPUSupport();
#endif // defined(XGBOOST_USE_CUDA)
} else {
auto quantiles = base_score->HostView();
auto h_weights = info.weights_.ConstHostVector();
if (info.weights_.Empty()) {
sw = info.num_row_;
} else {
sw = std::accumulate(std::cbegin(h_weights), std::cend(h_weights), 0.0);
}
for (bst_target_t t{0}; t < n_targets; ++t) {
auto alpha = param_.quantile_alpha[t];
auto h_labels = info.labels.HostView();
if (h_weights.empty()) {
quantiles(t) =
common::Quantile(ctx_, alpha, linalg::cbegin(h_labels), linalg::cend(h_labels));
} else {
CHECK_EQ(h_weights.size(), h_labels.Size());
quantiles(t) = common::WeightedQuantile(ctx_, alpha, linalg::cbegin(h_labels),
linalg::cend(h_labels), std::cbegin(h_weights));
}
}
}
// For multiple quantiles, we should extend the base score to a vector instead of

8
src/objective/regression_obj.cu
View File

@@ -254,8 +254,8 @@ class PseudoHuberRegression : public FitIntercept {
auto predt = linalg::MakeVec(&preds);
info.weights_.SetDevice(ctx_->Device());
common::OptionalWeights weight{ctx_->IsCPU() ? info.weights_.ConstHostSpan()
: info.weights_.ConstDeviceSpan()};
common::OptionalWeights weight{ctx_->IsCUDA() ? info.weights_.ConstDeviceSpan()
: info.weights_.ConstHostSpan()};
linalg::ElementWiseKernel(ctx_, labels, [=] XGBOOST_DEVICE(size_t i, float const y) mutable {
auto sample_id = std::get<0>(linalg::UnravelIndex(i, labels.Shape()));
@@ -714,8 +714,8 @@ class MeanAbsoluteError : public ObjFunction {
preds.SetDevice(ctx_->Device());
auto predt = linalg::MakeVec(&preds);
info.weights_.SetDevice(ctx_->Device());
common::OptionalWeights weight{ctx_->IsCPU() ? info.weights_.ConstHostSpan()
: info.weights_.ConstDeviceSpan()};
common::OptionalWeights weight{ctx_->IsCUDA() ? info.weights_.ConstDeviceSpan()
: info.weights_.ConstHostSpan()};
linalg::ElementWiseKernel(ctx_, labels, [=] XGBOOST_DEVICE(std::size_t i, float y) mutable {
auto sign = [](auto x) {