adonis0147 commented on code in PR #10386:
URL: https://github.com/apache/doris/pull/10386#discussion_r906043333
##########
be/src/olap/rowset/segment_v2/segment_iterator.cpp:
##########
@@ -872,9 +872,11 @@ Status SegmentIterator::_read_columns_by_index(uint32_t
nrows_read_limit, uint32
_read_columns(_first_read_column_ids, _current_return_columns,
rows_to_read));
_cur_rowid += rows_to_read;
if (set_block_rowid) {
- for (uint32_t rid = range_from; rid < range_to; rid++) {
- _block_rowids[nrows_read++] = rid;
- }
+ // Here use std::iota is better performance than for-loop, maybe
for-loop is not vectorized
Review Comment:
> There are two reasons why this loop can not be vectorized:
>
> 1. `nrows_read` is reference
> 2. `nrows_read` is unsigned
These reasons are not the root causes.
```cpp
#include <string>
#include <vector>
std::vector<uint32_t> _block_rowids;
void func(uint32_t range_from, uint32_t range_to, uint32_t& nrows_read) {
uint32_t* data = _block_rowids.data();
for (uint32_t rid = range_from; rid < range_to; rid++) {
data[nrows_read++] = rid;
}
}
int main(int argc, char* argv[]) {
uint32_t nrows_read = 0;
_block_rowids.resize(4096);
func(100, 2000, nrows_read);
return 0;
}
```
In above snippet, if you change the type of `nrows_read` from `uint32_t` to
`uint64_t`, you will find compiler vectorizes the loop.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
