This is an automated email from the ASF dual-hosted git repository.
yiguolei pushed a commit to branch branch-4.1
in repository https://gitbox.apache.org/repos/asf/doris.git
The following commit(s) were added to refs/heads/branch-4.1 by this push:
new 64ba650e670 [opt](exec) optimize fixed key packing (#65572) (#66220)
64ba650e670 is described below
commit 64ba650e6704363acfa687493f20471fdead62d1
Author: Mryange <[email protected]>
AuthorDate: Thu Jul 30 09:22:52 2026 +0800
[opt](exec) optimize fixed key packing (#65572) (#66220)
### What problem does this PR solve?
Fixed-key packing could lose `restrict` alias information when the
source data pointer was captured by reference in an outlined lambda,
producing a less efficient copy loop. This change obtains the restricted
pointer inside the lambda after nullable key replacement, preserving
alias information without changing the key encoding or query behavior.
(cherry picked from commit 0127314e7a50fd43fa9a8b959bb73ba600fb304b)
### What problem does this PR solve?
Issue Number: close #xxx
Related PR: #xxx
Problem Summary:
### Release note
None
### Check List (For Author)
- Test <!-- At least one of them must be included. -->
- [ ] Regression test
- [ ] Unit Test
- [ ] Manual test (add detailed scripts or steps below)
- [ ] No need to test or manual test. Explain why:
- [ ] This is a refactor/code format and no logic has been changed.
- [ ] Previous test can cover this change.
- [ ] No code files have been changed.
- [ ] Other reason <!-- Add your reason? -->
- Behavior changed:
- [ ] No.
- [ ] Yes. <!-- Explain the behavior change -->
- Does this need documentation?
- [ ] No.
- [ ] Yes. <!-- Add document PR link here. eg:
https://github.com/apache/doris-website/pull/1214 -->
### Check List (For Reviewer who merge this PR)
- [ ] Confirm the release note
- [ ] Confirm test cases
- [ ] Confirm document
- [ ] Add branch pick label <!-- Add branch pick label that this PR
should merge into -->
---
be/src/exec/common/hash_table/hash_map_context.h | 9 +++++++--
1 file changed, 7 insertions(+), 2 deletions(-)
diff --git a/be/src/exec/common/hash_table/hash_map_context.h
b/be/src/exec/common/hash_table/hash_map_context.h
index 121008cd830..49f3050dc9b 100644
--- a/be/src/exec/common/hash_table/hash_map_context.h
+++ b/be/src/exec/common/hash_table/hash_map_context.h
@@ -943,17 +943,21 @@ struct MethodKeysFixed : public MethodBase<TData> {
}
for (size_t j = 0; j < key_columns.size(); ++j) {
- const char* __restrict data = key_columns[j]->get_raw_data().data;
-
auto goo = [&]<typename Fixed, bool aligned>(Fixed zero) {
CHECK_EQ(sizeof(Fixed), key_sizes[j]);
if (has_null_column.size() && has_null_column[j]) {
const auto* nullmap =
assert_cast<const
ColumnUInt8&>(*nullmap_columns[j]).get_data().data();
// make sure null cell is filled by 0x0
+ // This mutates the same underlying buffer returned by
get_raw_data(), so get
+ // the restricted data pointer only after the replacement
finishes.
const_cast<IColumn*>(key_columns[j])->replace_column_null_data(nullmap);
}
auto* __restrict current = result_data + offset;
+ // Do not hoist data out of goo. A reference capture is
lowered to an ordinary
+ // closure field, so restrict/noalias information is lost when
goo is not inlined,
+ // which prevents the copy loop from being optimized.
+ const char* __restrict data =
key_columns[j]->get_raw_data().data;
for (size_t i = 0; i < row_numbers; ++i) {
memcpy_fixed<Fixed, aligned>(current, data);
current += sizeof(T);
@@ -965,6 +969,7 @@ struct MethodKeysFixed : public MethodBase<TData> {
// Also verify that the stride sizeof(T) is a multiple of
alignof(Fixed),
// otherwise alignment will be lost on subsequent loop
iterations
// (e.g. UInt96 has sizeof=12, stride 12 is not a multiple of
alignof(uint64_t)=8).
+ const char* data = key_columns[j]->get_raw_data().data;
if (sizeof(T) % alignof(Fixed) == 0 &&
reinterpret_cast<uintptr_t>(result_data + offset) %
alignof(Fixed) == 0 &&
reinterpret_cast<uintptr_t>(data) % alignof(Fixed) == 0) {
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
