xgboost/src/common/quantile.cc
Jiaming Yuan 29b7fea572
Optimize cpu sketch allreduce for sparse data. (#6009)
* Bypass RABIT serialization reducer and use custom allgather based merging.
2020-08-19 10:03:45 +08:00

327 lines
12 KiB
C++

/*!
* Copyright 2020 by XGBoost Contributors
*/
#include <limits>
#include <utility>
#include "quantile.h"
#include "hist_util.h"
namespace xgboost {
namespace common {
HostSketchContainer::HostSketchContainer(std::vector<bst_row_t> columns_size,
int32_t max_bins, bool use_group)
: columns_size_{std::move(columns_size)}, max_bins_{max_bins},
use_group_ind_{use_group} {
monitor_.Init(__func__);
CHECK_NE(columns_size_.size(), 0);
sketches_.resize(columns_size_.size());
for (size_t i = 0; i < sketches_.size(); ++i) {
auto n_bins = std::min(static_cast<size_t>(max_bins_), columns_size_[i]);
n_bins = std::max(n_bins, static_cast<decltype(n_bins)>(1));
auto eps = 1.0 / (static_cast<float>(n_bins) * WQSketch::kFactor);
sketches_[i].Init(columns_size_[i], eps);
sketches_[i].inqueue.queue.resize(sketches_[i].limit_size * 2);
}
}
std::vector<bst_row_t>
HostSketchContainer::CalcColumnSize(SparsePage const &batch,
bst_feature_t const n_columns,
size_t const nthreads) {
auto page = batch.GetView();
std::vector<std::vector<bst_row_t>> column_sizes(nthreads);
for (auto &column : column_sizes) {
column.resize(n_columns, 0);
}
ParallelFor(page.Size(), nthreads, [&](size_t i) {
auto &local_column_sizes = column_sizes.at(omp_get_thread_num());
auto row = page[i];
auto const *p_row = row.data();
for (size_t j = 0; j < row.size(); ++j) {
local_column_sizes.at(p_row[j].index)++;
}
});
std::vector<bst_row_t> entries_per_columns(n_columns, 0);
ParallelFor(n_columns, nthreads, [&](size_t i) {
for (auto const &thread : column_sizes) {
entries_per_columns[i] += thread[i];
}
});
return entries_per_columns;
}
std::vector<bst_feature_t> HostSketchContainer::LoadBalance(
SparsePage const &batch, bst_feature_t n_columns, size_t const nthreads) {
/* Some sparse datasets have their mass concentrating on small number of features. To
* avoid wating for a few threads running forever, we here distirbute different number
* of columns to different threads according to number of entries.
*/
auto page = batch.GetView();
size_t const total_entries = page.data.size();
size_t const entries_per_thread = common::DivRoundUp(total_entries, nthreads);
std::vector<std::vector<bst_row_t>> column_sizes(nthreads);
for (auto& column : column_sizes) {
column.resize(n_columns, 0);
}
std::vector<bst_row_t> entries_per_columns =
CalcColumnSize(batch, n_columns, nthreads);
std::vector<bst_feature_t> cols_ptr(nthreads + 1, 0);
size_t count {0};
size_t current_thread {1};
for (auto col : entries_per_columns) {
cols_ptr.at(current_thread)++; // add one column to thread
count += col;
CHECK_LE(count, total_entries);
if (count > entries_per_thread) {
current_thread++;
count = 0;
cols_ptr.at(current_thread) = cols_ptr[current_thread-1];
}
}
// Idle threads.
for (; current_thread < cols_ptr.size() - 1; ++current_thread) {
cols_ptr[current_thread+1] = cols_ptr[current_thread];
}
return cols_ptr;
}
void HostSketchContainer::PushRowPage(SparsePage const &page,
MetaInfo const &info) {
monitor_.Start(__func__);
int nthread = omp_get_max_threads();
CHECK_EQ(sketches_.size(), info.num_col_);
// Data groups, used in ranking.
std::vector<bst_uint> const &group_ptr = info.group_ptr_;
// Use group index for weights?
auto batch = page.GetView();
dmlc::OMPException exec;
// Parallel over columns. Each thread owns a set of consecutive columns.
auto const ncol = static_cast<uint32_t>(info.num_col_);
auto const is_dense = info.num_nonzero_ == info.num_col_ * info.num_row_;
auto thread_columns_ptr = LoadBalance(page, info.num_col_, nthread);
#pragma omp parallel num_threads(nthread)
{
exec.Run([&]() {
auto tid = static_cast<uint32_t>(omp_get_thread_num());
auto const begin = thread_columns_ptr[tid];
auto const end = thread_columns_ptr[tid + 1];
size_t group_ind = 0;
// do not iterate if no columns are assigned to the thread
if (begin < end && end <= ncol) {
for (size_t i = 0; i < batch.Size(); ++i) {
size_t const ridx = page.base_rowid + i;
SparsePage::Inst const inst = batch[i];
if (use_group_ind_) {
group_ind = this->SearchGroupIndFromRow(group_ptr, i + page.base_rowid);
}
size_t w_idx = use_group_ind_ ? group_ind : ridx;
auto w = info.GetWeight(w_idx);
auto p_inst = inst.data();
if (is_dense) {
for (size_t ii = begin; ii < end; ii++) {
sketches_[ii].Push(p_inst[ii].fvalue, w);
}
} else {
for (size_t i = 0; i < inst.size(); ++i) {
auto const& entry = p_inst[i];
if (entry.index >= begin && entry.index < end) {
sketches_[entry.index].Push(entry.fvalue, w);
}
}
}
}
}
});
}
exec.Rethrow();
monitor_.Stop(__func__);
}
void HostSketchContainer::GatherSketchInfo(
std::vector<WQSketch::SummaryContainer> const &reduced,
std::vector<size_t> *p_worker_segments,
std::vector<bst_row_t> *p_sketches_scan,
std::vector<WQSketch::Entry> *p_global_sketches) {
auto& worker_segments = *p_worker_segments;
worker_segments.resize(1, 0);
auto world = rabit::GetWorldSize();
auto rank = rabit::GetRank();
auto n_columns = sketches_.size();
std::vector<bst_row_t> sketch_size;
for (auto const& sketch : reduced) {
sketch_size.push_back(sketch.size);
}
std::vector<bst_row_t>& sketches_scan = *p_sketches_scan;
sketches_scan.resize((n_columns + 1) * world, 0);
size_t beg_scan = rank * (n_columns + 1);
std::partial_sum(sketch_size.cbegin(), sketch_size.cend(),
sketches_scan.begin() + beg_scan + 1);
// Gather all column pointers
rabit::Allreduce<rabit::op::Sum>(sketches_scan.data(), sketches_scan.size());
for (int32_t i = 0; i < world; ++i) {
size_t back = (i + 1) * (n_columns + 1) - 1;
auto n_entries = sketches_scan.at(back);
worker_segments.push_back(n_entries);
}
// Offset of sketch from each worker.
std::partial_sum(worker_segments.begin(), worker_segments.end(),
worker_segments.begin());
CHECK_GE(worker_segments.size(), 1);
auto total = worker_segments.back();
auto& global_sketches = *p_global_sketches;
global_sketches.resize(total, WQSketch::Entry{0, 0, 0, 0});
auto worker_sketch = Span<WQSketch::Entry>{global_sketches}.subspan(
worker_segments[rank], worker_segments[rank + 1] - worker_segments[rank]);
size_t cursor = 0;
for (auto const &sketch : reduced) {
std::copy(sketch.data, sketch.data + sketch.size,
worker_sketch.begin() + cursor);
cursor += sketch.size;
}
static_assert(sizeof(WQSketch::Entry) / 4 == sizeof(float), "");
rabit::Allreduce<rabit::op::Sum>(
reinterpret_cast<float *>(global_sketches.data()),
global_sketches.size() * sizeof(WQSketch::Entry) / sizeof(float));
}
void HostSketchContainer::AllReduce(
std::vector<WQSketch::SummaryContainer> *p_reduced,
std::vector<int32_t>* p_num_cuts) {
monitor_.Start(__func__);
auto& num_cuts = *p_num_cuts;
CHECK_EQ(num_cuts.size(), 0);
auto &reduced = *p_reduced;
reduced.resize(sketches_.size());
size_t n_columns = sketches_.size();
rabit::Allreduce<rabit::op::Max>(&n_columns, 1);
CHECK_EQ(n_columns, sketches_.size()) << "Number of columns differs across workers";
// Prune the intermediate num cuts for synchronization.
std::vector<bst_row_t> global_column_size(columns_size_);
rabit::Allreduce<rabit::op::Sum>(global_column_size.data(), global_column_size.size());
size_t nbytes = 0;
for (size_t i = 0; i < sketches_.size(); ++i) {
int32_t intermediate_num_cuts = static_cast<int32_t>(std::min(
global_column_size[i], static_cast<size_t>(max_bins_ * WQSketch::kFactor)));
if (global_column_size[i] != 0) {
WQSketch::SummaryContainer out;
sketches_[i].GetSummary(&out);
reduced[i].Reserve(intermediate_num_cuts);
CHECK(reduced[i].data);
reduced[i].SetPrune(out, intermediate_num_cuts);
nbytes = std::max(
WQSketch::SummaryContainer::CalcMemCost(intermediate_num_cuts),
nbytes);
}
num_cuts.push_back(intermediate_num_cuts);
}
auto world = rabit::GetWorldSize();
if (world == 1) {
return;
}
std::vector<size_t> worker_segments(1, 0); // CSC pointer to sketches.
std::vector<bst_row_t> sketches_scan((n_columns + 1) * world, 0);
std::vector<WQSketch::Entry> global_sketches;
this->GatherSketchInfo(reduced, &worker_segments, &sketches_scan,
&global_sketches);
std::vector<WQSketch::SummaryContainer> final_sketches(n_columns);
ParallelFor(n_columns, omp_get_max_threads(), [&](size_t fidx) {
int32_t intermediate_num_cuts = num_cuts[fidx];
auto nbytes =
WQSketch::SummaryContainer::CalcMemCost(intermediate_num_cuts);
for (int32_t i = 1; i < world + 1; ++i) {
auto size = worker_segments.at(i) - worker_segments[i - 1];
auto worker_sketches = Span<WQSketch::Entry>{global_sketches}.subspan(
worker_segments[i - 1], size);
auto worker_scan =
Span<bst_row_t>(sketches_scan)
.subspan((i - 1) * (n_columns + 1), (n_columns + 1));
auto worker_feature = worker_sketches.subspan(
worker_scan[fidx], worker_scan[fidx + 1] - worker_scan[fidx]);
CHECK(worker_feature.data());
WQSummary<float, float> summary(worker_feature.data(),
worker_feature.size());
auto &out = final_sketches.at(fidx);
out.Reduce(summary, nbytes);
}
reduced.at(fidx).Reserve(intermediate_num_cuts);
reduced.at(fidx).SetPrune(final_sketches.at(fidx), intermediate_num_cuts);
});
monitor_.Stop(__func__);
}
void AddCutPoint(WQuantileSketch<float, float>::SummaryContainer const &summary,
int max_bin, HistogramCuts *cuts) {
size_t required_cuts = std::min(summary.size, static_cast<size_t>(max_bin));
auto& cut_values = cuts->cut_values_.HostVector();
for (size_t i = 1; i < required_cuts; ++i) {
bst_float cpt = summary.data[i].value;
if (i == 1 || cpt > cuts->cut_values_.ConstHostVector().back()) {
cut_values.push_back(cpt);
}
}
}
void HostSketchContainer::MakeCuts(HistogramCuts* cuts) {
monitor_.Start(__func__);
std::vector<WQSketch::SummaryContainer> reduced;
std::vector<int32_t> num_cuts;
this->AllReduce(&reduced, &num_cuts);
cuts->min_vals_.HostVector().resize(sketches_.size(), 0.0f);
for (size_t fid = 0; fid < reduced.size(); ++fid) {
WQSketch::SummaryContainer a;
size_t max_num_bins = std::min(num_cuts[fid], max_bins_);
a.Reserve(max_num_bins + 1);
CHECK(a.data);
if (num_cuts[fid] != 0) {
a.SetPrune(reduced[fid], max_num_bins + 1);
CHECK(a.data && reduced[fid].data);
const bst_float mval = a.data[0].value;
cuts->min_vals_.HostVector()[fid] = mval - fabs(mval) - 1e-5f;
} else {
// Empty column.
const float mval = 1e-5f;
cuts->min_vals_.HostVector()[fid] = mval;
}
AddCutPoint(a, max_num_bins, cuts);
// push a value that is greater than anything
const bst_float cpt
= (a.size > 0) ? a.data[a.size - 1].value : cuts->min_vals_.HostVector()[fid];
// this must be bigger than last value in a scale
const bst_float last = cpt + (fabs(cpt) + 1e-5f);
cuts->cut_values_.HostVector().push_back(last);
// Ensure that every feature gets at least one quantile point
CHECK_LE(cuts->cut_values_.HostVector().size(), std::numeric_limits<uint32_t>::max());
auto cut_size = static_cast<uint32_t>(cuts->cut_values_.HostVector().size());
CHECK_GT(cut_size, cuts->cut_ptrs_.HostVector().back());
cuts->cut_ptrs_.HostVector().push_back(cut_size);
}
monitor_.Stop(__func__);
}
} // namespace common
} // namespace xgboost