From 93be53e34e2d15df8c43b37d79c6c349b434b5e0 Mon Sep 17 00:00:00 2001 From: "Wladimir J. van der Laan" Date: Mon, 9 May 2016 08:31:14 +0200 Subject: [PATCH] Merge #7934: Improve rolling bloom filter performance and benchmark 1953c40 More efficient bitsliced rolling Bloom filter (Pieter Wuille) aa62b68 Benchmark rolling bloom filter (Pieter Wuille) --- src/Makefile.bench.include | 3 ++- src/bench/rollingbloom.cpp | 43 ++++++++++++++++++++++++++++++++++++++ src/bloom.cpp | 40 +++++++++++++++++++++++------------ src/bloom.h | 13 +----------- src/test/bloom_tests.cpp | 5 ++++- 5 files changed, 77 insertions(+), 27 deletions(-) create mode 100644 src/bench/rollingbloom.cpp diff --git a/src/Makefile.bench.include b/src/Makefile.bench.include index 7b4a27815..3e45dd0aa 100644 --- a/src/Makefile.bench.include +++ b/src/Makefile.bench.include @@ -7,7 +7,8 @@ bench_bench_dash_SOURCES = \ bench/bench_dash.cpp \ bench/bench.cpp \ bench/bench.h \ - bench/Examples.cpp + bench/Examples.cpp \ + bench/rollingbloom.cpp bench_bench_dash_CPPFLAGS = $(AM_CPPFLAGS) $(BITCOIN_INCLUDES) $(EVENT_CLFAGS) $(EVENT_PTHREADS_CFLAGS) -I$(builddir)/bench/ bench_bench_dash_CXXFLAGS = $(AM_CXXFLAGS) $(PIE_FLAGS) diff --git a/src/bench/rollingbloom.cpp b/src/bench/rollingbloom.cpp new file mode 100644 index 000000000..73c02cf71 --- /dev/null +++ b/src/bench/rollingbloom.cpp @@ -0,0 +1,43 @@ +// Copyright (c) 2016 The Bitcoin Core developers +// Distributed under the MIT software license, see the accompanying +// file COPYING or http://www.opensource.org/licenses/mit-license.php. + +#include + +#include "bench.h" +#include "bloom.h" +#include "utiltime.h" + +static void RollingBloom(benchmark::State& state) +{ + CRollingBloomFilter filter(120000, 0.000001); + std::vector data(32); + uint32_t count = 0; + uint32_t nEntriesPerGeneration = (120000 + 1) / 2; + uint32_t countnow = 0; + uint64_t match = 0; + while (state.KeepRunning()) { + count++; + data[0] = count; + data[1] = count >> 8; + data[2] = count >> 16; + data[3] = count >> 24; + if (countnow == nEntriesPerGeneration) { + int64_t b = GetTimeMicros(); + filter.insert(data); + int64_t e = GetTimeMicros(); + std::cout << "RollingBloom-refresh,1," << (e-b)*0.000001 << "," << (e-b)*0.000001 << "," << (e-b)*0.000001 << "\n"; + countnow = 0; + } else { + filter.insert(data); + } + countnow++; + data[0] = count >> 24; + data[1] = count >> 16; + data[2] = count >> 8; + data[3] = count; + match += filter.contains(data); + } +} + +BENCHMARK(RollingBloom); diff --git a/src/bloom.cpp b/src/bloom.cpp index 34104b6e0..5285684d7 100644 --- a/src/bloom.cpp +++ b/src/bloom.cpp @@ -234,14 +234,18 @@ CRollingBloomFilter::CRollingBloomFilter(unsigned int nElements, double fpRate) */ uint32_t nFilterBits = (uint32_t)ceil(-1.0 * nHashFuncs * nMaxElements / log(1.0 - exp(logFpRate / nHashFuncs))); data.clear(); - /* We store up to 16 'bits' per data element. */ - data.resize((nFilterBits + 15) / 16); + /* For each data element we need to store 2 bits. If both bits are 0, the + * bit is treated as unset. If the bits are (01), (10), or (11), the bit is + * treated as set in generation 1, 2, or 3 respectively. + * These bits are stored in separate integers: position P corresponds to bit + * (P & 63) of the integers data[(P >> 6) * 2] and data[(P >> 6) * 2 + 1]. */ + data.resize(((nFilterBits + 63) / 64) << 1); reset(); } /* Similar to CBloomFilter::Hash */ -inline unsigned int CRollingBloomFilter::Hash(unsigned int nHashNum, const std::vector& vDataToHash) const { - return MurmurHash3(nHashNum * 0xFBA4C795 + nTweak, vDataToHash) % (data.size() * 16); +static inline uint32_t RollingBloomHash(unsigned int nHashNum, uint32_t nTweak, const std::vector& vDataToHash) { + return MurmurHash3(nHashNum * 0xFBA4C795 + nTweak, vDataToHash); } void CRollingBloomFilter::insert(const std::vector& vKey) @@ -252,18 +256,25 @@ void CRollingBloomFilter::insert(const std::vector& vKey) if (nGeneration == 4) { nGeneration = 1; } + uint64_t nGenerationMask1 = -(uint64_t)(nGeneration & 1); + uint64_t nGenerationMask2 = -(uint64_t)(nGeneration >> 1); /* Wipe old entries that used this generation number. */ - for (uint32_t p = 0; p < data.size() * 16; p++) { - if (get(p) == nGeneration) { - put(p, 0); - } + for (uint32_t p = 0; p < data.size(); p += 2) { + uint64_t p1 = data[p], p2 = data[p + 1]; + uint64_t mask = (p1 ^ nGenerationMask1) | (p2 ^ nGenerationMask2); + data[p] = p1 & mask; + data[p + 1] = p2 & mask; } } nEntriesThisGeneration++; for (int n = 0; n < nHashFuncs; n++) { - uint32_t h = Hash(n, vKey); - put(h, nGeneration); + uint32_t h = RollingBloomHash(n, nTweak, vKey); + int bit = h & 0x3F; + uint32_t pos = (h >> 6) % data.size(); + /* The lowest bit of pos is ignored, and set to zero for the first bit, and to one for the second. */ + data[pos & ~1] = (data[pos & ~1] & ~(((uint64_t)1) << bit)) | ((uint64_t)(nGeneration & 1)) << bit; + data[pos | 1] = (data[pos | 1] & ~(((uint64_t)1) << bit)) | ((uint64_t)(nGeneration >> 1)) << bit; } } @@ -276,8 +287,11 @@ void CRollingBloomFilter::insert(const uint256& hash) bool CRollingBloomFilter::contains(const std::vector& vKey) const { for (int n = 0; n < nHashFuncs; n++) { - uint32_t h = Hash(n, vKey); - if (get(h) == 0) { + uint32_t h = RollingBloomHash(n, nTweak, vKey); + int bit = h & 0x3F; + uint32_t pos = (h >> 6) % data.size(); + /* If the relevant bit is not set in either data[pos & ~1] or data[pos | 1], the filter does not contain vKey */ + if (!(((data[pos & ~1] | data[pos | 1]) >> bit) & 1)) { return false; } } @@ -295,7 +309,7 @@ void CRollingBloomFilter::reset() nTweak = GetRand(std::numeric_limits::max()); nEntriesThisGeneration = 0; nGeneration = 1; - for (std::vector::iterator it = data.begin(); it != data.end(); it++) { + for (std::vector::iterator it = data.begin(); it != data.end(); it++) { *it = 0; } } diff --git a/src/bloom.h b/src/bloom.h index b0ad8b875..ad6de625d 100644 --- a/src/bloom.h +++ b/src/bloom.h @@ -135,20 +135,9 @@ private: int nEntriesPerGeneration; int nEntriesThisGeneration; int nGeneration; - std::vector data; + std::vector data; unsigned int nTweak; int nHashFuncs; - - unsigned int Hash(unsigned int nHashNum, const std::vector& vDataToHash) const; - - inline int get(uint32_t position) const { - return (data[(position >> 4) % data.size()] >> (2 * (position & 0xF))) & 0x3; - } - - inline void put(uint32_t position, uint32_t val) { - uint32_t& cell = data[(position >> 4) % data.size()]; - cell = (cell & ~(((uint32_t)3) << (2 * (position & 0xF)))) | (val << (2 * (position & 0xF))); - } }; #endif // BITCOIN_BLOOM_H diff --git a/src/test/bloom_tests.cpp b/src/test/bloom_tests.cpp index b5503152e..fc43fc6e1 100644 --- a/src/test/bloom_tests.cpp +++ b/src/test/bloom_tests.cpp @@ -514,11 +514,14 @@ BOOST_AUTO_TEST_CASE(rolling_bloom) if (i >= 100) BOOST_CHECK(rb1.contains(data[i-100])); rb1.insert(data[i]); + BOOST_CHECK(rb1.contains(data[i])); } // Insert 999 more random entries: for (int i = 0; i < 999; i++) { - rb1.insert(RandomData()); + std::vector d = RandomData(); + rb1.insert(d); + BOOST_CHECK(rb1.contains(d)); } // Sanity check to make sure the filter isn't just filling up: nHits = 0;