This is an automated email from the ASF dual-hosted git repository.
wangbo pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-doris.git
The following commit(s) were added to refs/heads/master by this push:
new c2520c8 [Improvement](Vectorized) optimize SegmentIterator
predication evaluate (#7795)
c2520c8 is described below
commit c2520c878c3d827e27c32f1dab5e5953a25b52e7
Author: Zeno Yang <[email protected]>
AuthorDate: Sat Jan 22 15:31:07 2022 +0800
[Improvement](Vectorized) optimize SegmentIterator predication evaluate
(#7795)
* [Improvement](Vectorized) optimize SegmentIterator predication evaluate
* fix bug
* move bytes32_mask_to_bits32_mask to util/simd/bits.h
---
be/src/olap/rowset/segment_v2/segment_iterator.cpp | 67 ++++++++++++++--------
be/src/olap/rowset/segment_v2/segment_iterator.h | 2 +-
be/src/util/simd/bits.h | 60 +++++++++++++++++++
be/src/vec/columns/column_decimal.cpp | 4 +-
be/src/vec/columns/column_vector.cpp | 4 +-
be/src/vec/columns/columns_common.cpp | 3 +-
be/src/vec/columns/columns_common.h | 30 ----------
7 files changed, 109 insertions(+), 61 deletions(-)
diff --git a/be/src/olap/rowset/segment_v2/segment_iterator.cpp
b/be/src/olap/rowset/segment_v2/segment_iterator.cpp
index 86a89ac..bbf5005 100644
--- a/be/src/olap/rowset/segment_v2/segment_iterator.cpp
+++ b/be/src/olap/rowset/segment_v2/segment_iterator.cpp
@@ -17,6 +17,7 @@
#include "olap/rowset/segment_v2/segment_iterator.h"
+#include <memory>
#include <set>
#include <utility>
@@ -29,8 +30,9 @@
#include "olap/rowset/segment_v2/column_reader.h"
#include "olap/rowset/segment_v2/segment.h"
#include "olap/short_key_index.h"
-#include "util/doris_metrics.h"
#include "olap/in_list_predicate.h"
+#include "util/doris_metrics.h"
+#include "util/simd/bits.h"
using strings::Substitute;
@@ -197,8 +199,8 @@ Status SegmentIterator::_prepare_seek(const
StorageReadOptions::KeyRange& key_ra
}
}
}
- _seek_schema.reset(new Schema(key_fields, key_fields.size()));
- _seek_block.reset(new RowBlockV2(*_seek_schema, 1, _mem_tracker));
+ _seek_schema = std::make_unique<Schema>(key_fields, key_fields.size());
+ _seek_block = std::make_unique<RowBlockV2>(*_seek_schema, 1, _mem_tracker);
// create used column iterator
for (auto cid : _seek_schema->column_ids()) {
@@ -453,7 +455,7 @@ void SegmentIterator::_init_lazy_materialization() {
for (auto predicate : _col_predicates) {
predicate_columns.insert(predicate->column_id());
}
-
_opts.delete_condition_predicates.get()->get_all_column_ids(predicate_columns);
+
_opts.delete_condition_predicates->get_all_column_ids(predicate_columns);
// when all return columns have predicates, disable lazy
materialization to avoid its overhead
if (_schema.column_ids().size() > predicate_columns.size()) {
@@ -542,7 +544,7 @@ Status SegmentIterator::next_batch(RowBlockV2* block) {
// phase 2: run vectorized evaluation on remaining predicates to prune
rows.
// block's selection vector will be set to indicate which rows have passed
predicates.
// TODO(hkp): optimize column predicate to check column block once for one
column
- if (!_col_predicates.empty() || _opts.delete_condition_predicates.get() !=
nullptr) {
+ if (!_col_predicates.empty() || _opts.delete_condition_predicates !=
nullptr) {
// init selection position index
uint16_t selected_size = block->selected_size();
uint16_t original_size = selected_size;
@@ -596,13 +598,13 @@ void SegmentIterator::_vec_init_lazy_materialization() {
_is_all_column_basic_type = true;
bool is_predicate_column_exists = false;
bool is_non_predicate_column_exists = false;
-
+
if (!_col_predicates.empty()) {
is_predicate_column_exists = true;
std::set<ColumnId> short_cir_pred_col_id_set; // using set for
distinct cid
std::set<ColumnId> vec_pred_col_id_set;
-
+
for (auto predicate : _col_predicates) {
auto cid = predicate->column_id();
FieldType type = _schema.column(cid)->type();
@@ -649,7 +651,7 @@ void SegmentIterator::_vec_init_lazy_materialization() {
}
// note(wb) in following cases we disable lazy materialization
- // case 1: when all column is basic type(is_all_column_basic_type = true)
+ // case 1: when all column is basic type(is_all_column_basic_type = true)
// because we think `seek and read` cost > read page cost, lazy
materialize may cause more `seek and read`, so disable it
// case 2: all column is predicate column
// case 3: all column is not predicate column
@@ -662,7 +664,7 @@ void SegmentIterator::_vec_init_lazy_materialization() {
for (int i = 0; i < _schema.num_column_ids(); i++) {
auto cid = _schema.column_ids()[i];
if (pred_set.find(cid) != pred_set.end()) {
- _first_read_column_ids.push_back(cid);
+ _first_read_column_ids.push_back(cid);
} else if (non_pred_set.find(cid) != non_pred_set.end()) {
_first_read_column_ids.push_back(cid);
_is_pred_column[cid] = true; // in this case, non-predicate
column should also be filtered by sel idx, so we regard it as pred columns
@@ -752,15 +754,15 @@ void
SegmentIterator::_output_non_pred_columns(vectorized::Block* block, bool is
}
}
-void SegmentIterator::_output_column_by_sel_idx(vectorized::Block* block,
std::vector<ColumnId> columnIds,
+void SegmentIterator::_output_column_by_sel_idx(vectorized::Block* block,
const std::vector<ColumnId>& columnIds,
uint16_t* sel_rowid_idx, uint16_t select_size, bool
is_block_mem_reuse) {
for (auto cid : columnIds) {
auto &column_ptr = _current_return_columns[cid];
if (is_block_mem_reuse) {
- column_ptr->filter_by_selector(sel_rowid_idx, select_size,
+ column_ptr->filter_by_selector(sel_rowid_idx, select_size,
&block->get_by_position(_schema_block_id_map[cid]).column);
} else {
- block->replace_by_position(_schema_block_id_map[cid],
+ block->replace_by_position(_schema_block_id_map[cid],
(*column_ptr).get_ptr()->filter_by_selector(sel_rowid_idx,
select_size));
}
}
@@ -807,10 +809,25 @@ void
SegmentIterator::_evaluate_vectorization_predicate(uint16_t* sel_rowid_idx,
bool ret_flags[selected_size];
memset(ret_flags, 1, selected_size);
_pre_eval_block_predicate->evaluate_vec(_current_return_columns,
selected_size, ret_flags);
-
- for (uint32_t i = 0; i < selected_size; ++i) {
- if (ret_flags[i]) {
- sel_rowid_idx[new_size++] = i;
+
+ uint32_t sel_pos = 0;
+ const uint32_t sel_end = sel_pos + selected_size;
+ static constexpr size_t SIMD_BYTES = 32;
+ const uint32_t sel_end_simd = sel_pos + selected_size / SIMD_BYTES *
SIMD_BYTES;
+
+ while (sel_pos < sel_end_simd) {
+ auto mask = simd::bytes32_mask_to_bits32_mask(ret_flags + sel_pos);
+ while (mask) {
+ const size_t bit_pos = __builtin_ctzll(mask);
+ sel_rowid_idx[new_size++] = sel_pos + bit_pos;
+ mask = mask & (mask - 1);
+ }
+ sel_pos += SIMD_BYTES;
+ }
+
+ for (; sel_pos < sel_end; sel_pos++) {
+ if (ret_flags[sel_pos]) {
+ sel_rowid_idx[new_size++] = sel_pos;
}
}
@@ -822,7 +839,7 @@ void
SegmentIterator::_evaluate_short_circuit_predicate(uint16_t* vec_sel_rowid_
if (_short_cir_pred_column_ids.size() == 0) {
return;
}
-
+
for (auto column_predicate : _short_cir_eval_predicate) {
auto column_id = column_predicate->column_id();
auto& short_cir_column = _current_return_columns[column_id];
@@ -854,10 +871,10 @@ Status SegmentIterator::next_batch(vectorized::Block*
block) {
if (UNLIKELY(!_inited)) {
RETURN_IF_ERROR(_init(true));
_inited = true;
- if (_vec_pred_column_ids.size() > 0 ||
_short_cir_pred_column_ids.size() > 0) {
+ if (!_vec_pred_column_ids.empty() ||
!_short_cir_pred_column_ids.empty()) {
_block_rowids.reserve(_opts.block_row_max);
}
- _current_return_columns.resize(_schema.columns().size());
+ _current_return_columns.resize(_schema.columns().size());
for (size_t i = 0; i < _schema.num_column_ids(); i++) {
auto cid = _schema.column_ids()[i];
if (_is_pred_column[cid]) {
@@ -869,7 +886,7 @@ Status SegmentIterator::next_batch(vectorized::Block*
block) {
}
_init_current_block(block, _current_return_columns);
-
+
uint32_t nrows_read = 0;
uint32_t nrows_read_limit = _opts.block_row_max;
_read_columns_by_index(nrows_read_limit, nrows_read,
_col_predicates.size() > 0);
@@ -890,10 +907,10 @@ Status SegmentIterator::next_batch(vectorized::Block*
block) {
block->replace_by_position(i, data_type->create_column());
}
}
- // not sure whether block is clear before enter segmentIter, so
clear it here.
- if (is_mem_reuse) {
- block->clear_column_data();
- }
+ }
+ // not sure whether block is clear before enter segmentIter, so clear
it here.
+ if (is_mem_reuse) {
+ block->clear_column_data();
}
return Status::EndOfFile("no more data in segment");
}
@@ -920,7 +937,7 @@ Status SegmentIterator::next_batch(vectorized::Block*
block) {
// to reduce cost of read short circuit columns.
// In SSB test, it make no difference; So need more scenarios
to test
_evaluate_short_circuit_predicate(sel_rowid_idx, &selected_size);
-
+
// step3: read non_predicate column
if (_non_predicate_columns.size() != 0) {
_read_columns_by_rowids(_non_predicate_columns, _block_rowids,
sel_rowid_idx, selected_size, &_current_return_columns);
diff --git a/be/src/olap/rowset/segment_v2/segment_iterator.h
b/be/src/olap/rowset/segment_v2/segment_iterator.h
index c06077e..d7778ac 100644
--- a/be/src/olap/rowset/segment_v2/segment_iterator.h
+++ b/be/src/olap/rowset/segment_v2/segment_iterator.h
@@ -95,7 +95,7 @@ private:
void _evaluate_vectorization_predicate(uint16_t* sel_rowid_idx, uint16_t&
selected_size);
void _evaluate_short_circuit_predicate(uint16_t* sel_rowid_idx, uint16_t*
selected_size);
void _output_non_pred_columns(vectorized::Block* block, bool
is_block_mem_reuse);
- void _output_column_by_sel_idx(vectorized::Block* block,
std::vector<ColumnId> columnids, uint16_t* sel_rowid_idx, uint16_t select_size,
bool is_block_mem_reuse);
+ void _output_column_by_sel_idx(vectorized::Block* block, const
std::vector<ColumnId>& columnids, uint16_t* sel_rowid_idx, uint16_t
select_size, bool is_block_mem_reuse);
void _read_columns_by_rowids(std::vector<ColumnId>& read_column_ids,
std::vector<rowid_t>& rowid_vector,
uint16_t* sel_rowid_idx, size_t select_size,
vectorized::MutableColumns* mutable_columns);
diff --git a/be/src/util/simd/bits.h b/be/src/util/simd/bits.h
new file mode 100644
index 0000000..d2e9423
--- /dev/null
+++ b/be/src/util/simd/bits.h
@@ -0,0 +1,60 @@
+// Licensed to the Apache Software Foundation (ASF) under one
+// or more contributor license agreements. See the NOTICE file
+// distributed with this work for additional information
+// regarding copyright ownership. The ASF licenses this file
+// to you under the Apache License, Version 2.0 (the
+// "License"); you may not use this file except in compliance
+// with the License. You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing,
+// software distributed under the License is distributed on an
+// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+// KIND, either express or implied. See the License for the
+// specific language governing permissions and limitations
+// under the License.
+
+#include <cstdint>
+
+#ifdef __AVX2__
+#include <immintrin.h>
+#elif __SSE2__
+#include <emmintrin.h>
+#endif
+
+namespace doris {
+namespace simd {
+
+/// todo(zeno) Compile add avx512 parameter, modify it to
bytes64_mask_to_bits64_mask
+/// Transform 32-byte mask to 32-bit mask
+inline uint32_t bytes32_mask_to_bits32_mask(const uint8_t* data) {
+#ifdef __AVX2__
+ auto zero32 = _mm256_setzero_si256();
+ uint32_t mask =
static_cast<uint32_t>(_mm256_movemask_epi8(_mm256_cmpgt_epi8(
+ _mm256_loadu_si256(reinterpret_cast<const __m256i*>(data)),
zero32)));
+#elif __SSE2__
+ auto zero16 = _mm_setzero_si128();
+ uint32_t mask =
+ (static_cast<uint32_t>(_mm_movemask_epi8(_mm_cmpgt_epi8(
+ _mm_loadu_si128(reinterpret_cast<const __m128i*>(data)),
zero16)))) |
+ ((static_cast<uint32_t>(_mm_movemask_epi8(_mm_cmpgt_epi8(
+ _mm_loadu_si128(reinterpret_cast<const __m128i*>(data +
16)), zero16)))
+ << 16) &
+ 0xffff0000);
+#else
+ uint32_t mask = 0;
+ for (size_t i = 0; i < 32; ++i) {
+ mask |= static_cast<uint32_t>(1 == *(data + i)) << i;
+ }
+#endif
+ return mask;
+}
+
+inline uint32_t bytes32_mask_to_bits32_mask(const bool * data) {
+ return bytes32_mask_to_bits32_mask(reinterpret_cast<const uint8_t*>(data));
+}
+
+
+} // namespace simd
+} // namespace doris
\ No newline at end of file
diff --git a/be/src/vec/columns/column_decimal.cpp
b/be/src/vec/columns/column_decimal.cpp
index 5cc5853..fbfc1b4 100644
--- a/be/src/vec/columns/column_decimal.cpp
+++ b/be/src/vec/columns/column_decimal.cpp
@@ -20,12 +20,12 @@
#include "vec/columns/column_decimal.h"
-#include "vec/columns/columns_common.h"
#include "vec/common/arena.h"
#include "vec/common/assert_cast.h"
#include "vec/common/exception.h"
#include "vec/common/sip_hash.h"
#include "vec/common/unaligned.h"
+#include "util/simd/bits.h"
template <typename T>
bool decimal_less(T x, T y, doris::vectorized::UInt32 x_scale,
doris::vectorized::UInt32 y_scale);
@@ -171,7 +171,7 @@ ColumnPtr ColumnDecimal<T>::filter(const IColumn::Filter&
filt, ssize_t result_s
const UInt8* filt_end_sse = filt_pos + size / SIMD_BYTES * SIMD_BYTES;
while (filt_pos < filt_end_sse) {
- uint32_t mask = bytes32_mask_to_bits32_mask(filt_pos);
+ uint32_t mask = simd::bytes32_mask_to_bits32_mask(filt_pos);
if (0xFFFFFFFF == mask) {
res_data.insert(data_pos, data_pos + SIMD_BYTES);
diff --git a/be/src/vec/columns/column_vector.cpp
b/be/src/vec/columns/column_vector.cpp
index 75ff144..f640a60 100644
--- a/be/src/vec/columns/column_vector.cpp
+++ b/be/src/vec/columns/column_vector.cpp
@@ -27,7 +27,6 @@
#include <cstring>
#include "runtime/datetime_value.h"
-#include "vec/columns/columns_common.h"
#include "vec/common/arena.h"
#include "vec/common/assert_cast.h"
#include "vec/common/bit_cast.h"
@@ -35,6 +34,7 @@
#include "vec/common/nan_utils.h"
#include "vec/common/sip_hash.h"
#include "vec/common/unaligned.h"
+#include "util/simd/bits.h"
namespace doris::vectorized {
@@ -252,7 +252,7 @@ ColumnPtr ColumnVector<T>::filter(const IColumn::Filter&
filt, ssize_t result_si
const UInt8* filt_end_sse = filt_pos + size / SIMD_BYTES * SIMD_BYTES;
while (filt_pos < filt_end_sse) {
- uint32_t mask = bytes32_mask_to_bits32_mask(filt_pos);
+ uint32_t mask = simd::bytes32_mask_to_bits32_mask(filt_pos);
if (0xFFFFFFFF == mask) {
res_data.insert(data_pos, data_pos + SIMD_BYTES);
diff --git a/be/src/vec/columns/columns_common.cpp
b/be/src/vec/columns/columns_common.cpp
index 3045c5b..455e563 100644
--- a/be/src/vec/columns/columns_common.cpp
+++ b/be/src/vec/columns/columns_common.cpp
@@ -26,6 +26,7 @@
#include "vec/columns/column_vector.h"
#include "vec/columns/columns_common.h"
#include "vec/common/typeid_cast.h"
+#include "util/simd/bits.h"
namespace doris::vectorized {
@@ -178,7 +179,7 @@ void filter_arrays_impl_generic(const PaddedPODArray<T>&
src_elems,
const auto filt_end_aligned = filt_pos + size / SIMD_BYTES * SIMD_BYTES;
while (filt_pos < filt_end_aligned) {
- auto mask = bytes32_mask_to_bits32_mask(filt_pos);
+ auto mask = simd::bytes32_mask_to_bits32_mask(filt_pos);
if (mask == 0xffffffff) {
/// SIMD_BYTES consecutive rows pass the filter
diff --git a/be/src/vec/columns/columns_common.h
b/be/src/vec/columns/columns_common.h
index 83c38c4..ef9c00c 100644
--- a/be/src/vec/columns/columns_common.h
+++ b/be/src/vec/columns/columns_common.h
@@ -22,40 +22,10 @@
#include "vec/columns/column.h"
-#ifdef __AVX2__
-#include <immintrin.h>
-#elif __SSE2__
-#include <emmintrin.h>
-#endif
-
/// Common helper methods for implementation of different columns.
namespace doris::vectorized {
-/// Transform 32-byte mask to 32-bit mask
-inline uint32_t bytes32_mask_to_bits32_mask(const uint8_t* filt_pos) {
-#ifdef __AVX2__
- auto zero32 = _mm256_setzero_si256();
- uint32_t mask =
static_cast<uint32_t>(_mm256_movemask_epi8(_mm256_cmpgt_epi8(
- _mm256_loadu_si256(reinterpret_cast<const __m256i*>(filt_pos)),
zero32)));
-#elif __SSE2__
- auto zero16 = _mm_setzero_si128();
- uint32_t mask =
- (static_cast<uint32_t>(_mm_movemask_epi8(_mm_cmpgt_epi8(
- _mm_loadu_si128(reinterpret_cast<const
__m128i*>(filt_pos)), zero16)))) |
- ((static_cast<uint32_t>(_mm_movemask_epi8(_mm_cmpgt_epi8(
- _mm_loadu_si128(reinterpret_cast<const
__m128i*>(filt_pos + 16)), zero16)))
- << 16) &
- 0xffff0000);
-#else
- uint32_t mask = 0;
- for (size_t i = 0; i < 32; ++i) {
- mask |= static_cast<uint32_t>(1 == *(filt_pos + i)) << i;
- }
-#endif
- return mask;
-}
-
/// Counts how many bytes of `filt` are greater than zero.
size_t count_bytes_in_filter(const IColumn::Filter& filt);
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
