This is an automated email from the ASF dual-hosted git repository.
kxiao pushed a commit to branch branch-2.0-beta
in repository https://gitbox.apache.org/repos/asf/doris.git
commit 46e2277a304399a2ff77a41adcd82c26bb1eae94
Author: Gabriel <gabrielleeb...@gmail.com>
AuthorDate: Fri Jun 9 13:10:57 2023 +0800
[Improvement](column) reduce cache miss for data copy (#20583)
---
be/src/vec/columns/column.h | 1 +
be/src/vec/columns/column_decimal.h | 3 +++
be/src/vec/columns/column_string.cpp | 3 +++
be/src/vec/columns/column_string.h | 5 +++++
be/src/vec/columns/column_vector.cpp | 6 ++++++
be/src/vec/exec/join/vhash_join_node.cpp | 4 ++--
6 files changed, 20 insertions(+), 2 deletions(-)
diff --git a/be/src/vec/columns/column.h b/be/src/vec/columns/column.h
index 8bf83b533c..00f1824b99 100644
--- a/be/src/vec/columns/column.h
+++ b/be/src/vec/columns/column.h
@@ -98,6 +98,7 @@ protected:
using Offsets64 = PaddedPODArray<Offset64>;
public:
+ static constexpr int PREFETCH_STEP = 64;
// 32bit offsets for string
using Offset = UInt32;
using Offsets = PaddedPODArray<Offset>;
diff --git a/be/src/vec/columns/column_decimal.h
b/be/src/vec/columns/column_decimal.h
index 26a9abde9a..f935c0c826 100644
--- a/be/src/vec/columns/column_decimal.h
+++ b/be/src/vec/columns/column_decimal.h
@@ -129,6 +129,9 @@ public:
const T* src_data = reinterpret_cast<const
T*>(src.get_raw_data().data);
for (int i = 0; i < new_size; ++i) {
+ if (i + IColumn::PREFETCH_STEP < new_size) {
+ __builtin_prefetch(&src_data[indices_begin[i +
IColumn::PREFETCH_STEP]], 0, 1);
+ }
data[origin_size + i] = src_data[indices_begin[i]];
}
}
diff --git a/be/src/vec/columns/column_string.cpp
b/be/src/vec/columns/column_string.cpp
index 66fec57787..81ee29a70f 100644
--- a/be/src/vec/columns/column_string.cpp
+++ b/be/src/vec/columns/column_string.cpp
@@ -113,6 +113,9 @@ void ColumnString::insert_indices_from(const IColumn& src,
const int* indices_be
if (*x == -1) {
ColumnString::insert_default();
} else {
+ if (x + IColumn::PREFETCH_STEP < indices_end) {
+ ColumnString::prefetch(src, *(x + IColumn::PREFETCH_STEP));
+ }
ColumnString::insert_from(src, *x);
}
}
diff --git a/be/src/vec/columns/column_string.h
b/be/src/vec/columns/column_string.h
index 06e8914907..66d0a77e5d 100644
--- a/be/src/vec/columns/column_string.h
+++ b/be/src/vec/columns/column_string.h
@@ -149,6 +149,11 @@ public:
offsets.push_back(new_size);
}
+ void prefetch(const IColumn& src_, size_t n) {
+ const ColumnString& src = assert_cast<const ColumnString&>(src_);
+ __builtin_prefetch(&src.chars[src.offsets[n - 1]], 0, 1);
+ }
+
void insert_from(const IColumn& src_, size_t n) override {
const ColumnString& src = assert_cast<const ColumnString&>(src_);
const size_t size_to_append =
diff --git a/be/src/vec/columns/column_vector.cpp
b/be/src/vec/columns/column_vector.cpp
index 388b436bc5..35534d95e7 100644
--- a/be/src/vec/columns/column_vector.cpp
+++ b/be/src/vec/columns/column_vector.cpp
@@ -376,12 +376,18 @@ void ColumnVector<T>::insert_indices_from(const IColumn&
src, const int* indices
if constexpr (std::is_same_v<T, UInt8>) {
// nullmap : indices_begin[i] == -1 means is null at the here, set
true here
for (int i = 0; i < new_size; ++i) {
+ if (i + IColumn::PREFETCH_STEP < new_size) {
+ __builtin_prefetch(&src_data[indices_begin[i +
IColumn::PREFETCH_STEP]], 0, 1);
+ }
data[origin_size + i] = (indices_begin[i] == -1) +
(indices_begin[i] != -1) *
src_data[indices_begin[i]];
}
} else {
// real data : indices_begin[i] == -1 what at is meaningless
for (int i = 0; i < new_size; ++i) {
+ if (i + IColumn::PREFETCH_STEP < new_size) {
+ __builtin_prefetch(&src_data[indices_begin[i +
IColumn::PREFETCH_STEP]], 0, 1);
+ }
data[origin_size + i] = src_data[indices_begin[i]];
}
}
diff --git a/be/src/vec/exec/join/vhash_join_node.cpp
b/be/src/vec/exec/join/vhash_join_node.cpp
index c2b9d915ec..e47cda7d1d 100644
--- a/be/src/vec/exec/join/vhash_join_node.cpp
+++ b/be/src/vec/exec/join/vhash_join_node.cpp
@@ -469,6 +469,8 @@ Status HashJoinNode::prepare(RuntimeState* state) {
_build_rows_counter = ADD_COUNTER(record_profile, "BuildRows",
TUnit::UNIT);
_build_side_compute_hash_timer = ADD_TIMER(record_profile,
"BuildSideHashComputingTime");
_build_runtime_filter_timer = ADD_TIMER(record_profile,
"BuildRuntimeFilterTime");
+ _push_down_timer = ADD_TIMER(record_profile, "PublishRuntimeFilterTime");
+ _push_compute_timer = ADD_TIMER(record_profile, "PushDownComputeTime");
// Probe phase
auto probe_phase_profile = runtime_profile()->create_child("ProbePhase",
true, true);
@@ -484,8 +486,6 @@ Status HashJoinNode::prepare(RuntimeState* state) {
_open_timer = ADD_TIMER(runtime_profile(), "OpenTime");
_allocate_resource_timer = ADD_TIMER(runtime_profile(),
"AllocateResourceTime");
- _push_down_timer = ADD_TIMER(runtime_profile(),
"PublishRuntimeFilterTime");
- _push_compute_timer = ADD_TIMER(runtime_profile(), "PushDownComputeTime");
_build_buckets_counter = ADD_COUNTER(runtime_profile(), "BuildBuckets",
TUnit::UNIT);
_build_buckets_fill_counter = ADD_COUNTER(runtime_profile(),
"FilledBuckets", TUnit::UNIT);
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@doris.apache.org
For additional commands, e-mail: commits-h...@doris.apache.org
