github/redis
1
1
Fork 0
mirror of https://github.com/redis/redis.git synced 2026-04-21 03:01:35 -04:00
Code Issues Packages Projects Releases Wiki Activity

add comments about little endian

Browse Source
This commit is contained in:
Nugine
2024-10-31 00:38:15 +08:00
parent 2a5018981c
commit 02e09f85ac
1 changed files with 14 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
src/hyperloglog.c
View File

@@ -1091,11 +1091,14 @@ void hllMergeDenseAVX2(uint8_t *reg_raw, const uint8_t *reg_dense) {
* {AAA0|BBB0|CCC0|DDD0|EEE0|FFF0|GGG0|HHH0}
* {bbaaaaaa|ccccbbbb|ddddddcc|00000000} x8
*
* AVX2 is little endian, each of the 8 groups is a little-endian int32.
* A group (int32) contains 3 valid bytes (4 registers) and a zero byte.
*
* extract registers in each group with AND and SHIFT:
* {00aaaaaa|00000000|00000000|00000000} x8
* {00000000|00bbbbbb|00000000|00000000} x8
* {00000000|00000000|00cccccc|00000000} x8
* {00000000|00000000|00000000|00dddddd} x8
* {00aaaaaa|00000000|00000000|00000000} x8 (<<0)
* {00000000|00bbbbbb|00000000|00000000} x8 (<<2)
* {00000000|00000000|00cccccc|00000000} x8 (<<4)
* {00000000|00000000|00000000|00dddddd} x8 (<<6)
*
* merge the extracted registers with OR:
* {00aaaaaa|00bbbbbb|00cccccc|00dddddd} x8
@@ -1243,11 +1246,14 @@ void hllDenseCompressAVX2(uint8_t *reg_dense, const uint8_t *reg_raw) {
* LOAD 32 bytes (32 registers) per iteration:
* {00aaaaaa|00bbbbbb|00cccccc|00dddddd} x8
*
* AVX2 is little endian, each of the 8 groups is a little-endian int32.
* A group (int32) contains 4 registers.
*
* move the registers to correct positions with AND and SHIFT:
* {00aaaaaa|00000000|00000000|00000000} x8
* {bb000000|0000bbbb|00000000|00000000} x8
* {00000000|cccc0000|000000cc|00000000} x8
* {00000000|00000000|dddddd00|00000000} x8
* {00aaaaaa|00000000|00000000|00000000} x8 (>>0)
* {bb000000|0000bbbb|00000000|00000000} x8 (>>2)
* {00000000|cccc0000|000000cc|00000000} x8 (>>4)
* {00000000|00000000|dddddd00|00000000} x8 (>>6)
*
* merge the registers with OR:
* {bbaaaaaa|ccccbbbb|ddddddcc|00000000} x8