From bc5a4751be5b535b3414fc12cdde4a7cb5c325e0 Mon Sep 17 00:00:00 2001 From: ross39 Date: Thu, 3 Jul 2025 12:21:38 +0100 Subject: [PATCH 2/2] update to use rational hashing. Still working out some performance details --- src/backend/lib/bloomfilter.c | 26 ++++++++++++++++++++++++-- 1 file changed, 24 insertions(+), 2 deletions(-) diff --git a/src/backend/lib/bloomfilter.c b/src/backend/lib/bloomfilter.c index 9875ec9aed8..0e79a07e284 100644 --- a/src/backend/lib/bloomfilter.c +++ b/src/backend/lib/bloomfilter.c @@ -400,8 +400,8 @@ lossless_bloom_create(bloom_config *config) filter->blocks[0].size = filter->total_bits; filter->blocks[0].offset = 0; filter->blocks[0].k_rational = log(2.0) * filter->total_bits / config->total_elems; - filter->blocks[0].k_floor = Max(1, Min((int)rint(filter->blocks[0].k_rational), MAX_HASH_FUNCS)); - filter->blocks[0].k_fractional = 0.0; /* Disable fractional hashing for speed */ + filter->blocks[0].k_floor = Max(1, Min((int)floor(filter->blocks[0].k_rational), MAX_HASH_FUNCS)); + filter->blocks[0].k_fractional = filter->blocks[0].k_rational - filter->blocks[0].k_floor; /* Enable fractional hashing */ /* Allocate bitset */ bitset_bytes = (filter->total_bits + 7) / 8; @@ -498,6 +498,17 @@ lossless_bloom_add_element(lossless_bloom_filter *filter, unsigned char *elem, s hashes[i] = x; } + /* RATIONAL HASHING: Apply fractional hash function probabilistically */ + if (filter->blocks[0].k_fractional > 0.0 && k_hash_funcs < MAX_HASH_FUNCS) + { + if (apply_rational_hashing(filter->blocks[0].k_fractional, base_hash)) + { + x += y; + hashes[k_hash_funcs] = x; + k_hash_funcs++; + } + } + /* OPTIMIZATION: Fast bit setting with bitwise operations */ for (i = 0; i < k_hash_funcs; i++) { @@ -551,6 +562,17 @@ lossless_bloom_contains_element(lossless_bloom_filter *filter, unsigned char *el hashes[i] = x; } + /* RATIONAL HASHING: Apply fractional hash function probabilistically */ + if (filter->blocks[0].k_fractional > 0.0 && k_hash_funcs < MAX_HASH_FUNCS) + { + if (apply_rational_hashing(filter->blocks[0].k_fractional, base_hash)) + { + x += y; + hashes[k_hash_funcs] = x; + k_hash_funcs++; + } + } + /* OPTIMIZATION: Fast bit checking with bitwise operations */ for (i = 0; i < k_hash_funcs; i++) { -- 2.39.5 (Apple Git-154)