f4e10eee06
Improve the performance of crc64 for large batches by processing large number of bytes in parallel and combining the results. ## Performance * 53-73% faster on Xeon 2670 v0 @ 2.6ghz * 2-2.5x faster on Core i3 8130U @ 2.2 ghz * 1.6-2.46 bytes/cycle on i3 8130U * likely >2x faster than crcspeed on newer CPUs with more resources than a 2012-era Xeon 2670 * crc64 combine function runs in <50 nanoseconds typical with vector + cache optimizations (~8 *microseconds* without vector optimizations, ~80 *microseconds without cache, the combination is extra effective) * still single-threaded * valkey-server test crc64 --help (requires `make distclean && make SERVER_TEST=yes`) --------- Signed-off-by: Josiah Carlson <josiah.carlson@gmail.com> Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech> Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
11 lines
321 B
C
11 lines
321 B
C
|
|
#include <stdint.h>
|
|
|
|
|
|
/* mask types */
|
|
typedef unsigned long long v2uq __attribute__ ((vector_size (16)));
|
|
|
|
uint64_t gf2_matrix_times_vec2(uint64_t *mat, uint64_t vec);
|
|
void init_combine_cache(uint64_t poly, uint8_t dim);
|
|
uint64_t crc64_combine(uint64_t crc1, uint64_t crc2, uintmax_t len2, uint64_t poly, uint8_t dim);
|