viirya commented on code in PR #6288:
URL: https://github.com/apache/arrow-rs/pull/6288#discussion_r1742700576
##########
arrow-buffer/src/util/bit_mask.rs:
##########
@@ -32,33 +31,126 @@ pub fn set_bits(
) -> usize {
let mut null_count = 0;
- let mut bits_to_align = offset_write % 8;
- if bits_to_align > 0 {
- bits_to_align = std::cmp::min(len, 8 - bits_to_align);
+ let mut acc = 0;
+ while len > acc {
+ let (n, l) = set_upto_64bits(
+ write_data,
+ data,
+ offset_write + acc,
+ offset_read + acc,
+ len - acc,
+ );
+ null_count += n;
+ acc += l;
}
- let mut write_byte_index = ceil(offset_write + bits_to_align, 8);
-
- // Set full bytes provided by bit chunk iterator (which iterates in 64
bits at a time)
- let chunks = BitChunks::new(data, offset_read + bits_to_align, len -
bits_to_align);
- chunks.iter().for_each(|chunk| {
- null_count += chunk.count_zeros();
- write_data[write_byte_index..write_byte_index +
8].copy_from_slice(&chunk.to_le_bytes());
- write_byte_index += 8;
- });
-
- // Set individual bits both to align write_data to a byte offset and the
remainder bits not covered by the bit chunk iterator
- let remainder_offset = len - chunks.remainder_len();
- (0..bits_to_align)
- .chain(remainder_offset..len)
- .for_each(|i| {
- if get_bit(data, offset_read + i) {
- set_bit(write_data, offset_write + i);
+
+ null_count
+}
+
+#[inline]
+fn set_upto_64bits(
+ write_data: &mut [u8],
+ data: &[u8],
+ offset_write: usize,
+ offset_read: usize,
+ len: usize,
+) -> (usize, usize) {
+ let read_byte = offset_read / 8;
+ let read_shift = offset_read % 8;
+ let write_byte = offset_write / 8;
+ let write_shift = offset_write % 8;
+
+ if len >= 64 {
+ // SAFETY: chunk gets masked when necessary, so it is safe
+ let chunk = unsafe { read_bytes_to_u64(data, read_byte, 8) };
Review Comment:
Oh, I see. Even `len` is more than 64 bits, it reads/writes at most 64 bites.
--
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]
