zanmato1984 commented on code in PR #43832:
URL: https://github.com/apache/arrow/pull/43832#discussion_r1743894384
##########
cpp/src/arrow/acero/swiss_join_avx2.cc:
##########
@@ -233,27 +243,270 @@ int RowArrayAccessor::VisitNulls_avx2(const
RowTableImpl& rows, int column_id,
const uint8_t* null_masks = rows.null_masks();
__m256i null_bits_per_row =
_mm256_set1_epi32(8 * rows.metadata().null_masks_bytes_per_row);
+ __m256i pos_after_encoding =
+ _mm256_set1_epi32(rows.metadata().pos_after_encoding(column_id));
for (int i = 0; i < num_rows / unroll; ++i) {
__m256i row_id = _mm256_loadu_si256(reinterpret_cast<const
__m256i*>(row_ids) + i);
__m256i bit_id = _mm256_mullo_epi32(row_id, null_bits_per_row);
- bit_id = _mm256_add_epi32(bit_id, _mm256_set1_epi32(column_id));
+ bit_id = _mm256_add_epi32(bit_id, pos_after_encoding);
__m256i bytes = _mm256_i32gather_epi32(reinterpret_cast<const
int*>(null_masks),
Review Comment:
Right. Each row has a sequence of bits (occupying
`null_masks_bytes_per_rows` bytes), each for one column in this particular row.
And the bytes for these bits of all rows are stored contiguously in a buffer
pointed by `rows.null_masks()`.
--
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]
