diff options
Diffstat (limited to 'src/bench')
-rw-r--r-- | src/bench/Examples.cpp | 4 | ||||
-rw-r--r-- | src/bench/base58.cpp | 2 | ||||
-rw-r--r-- | src/bench/bench.cpp | 38 | ||||
-rw-r--r-- | src/bench/bench.h | 12 | ||||
-rw-r--r-- | src/bench/bench_bitcoin.cpp | 4 | ||||
-rw-r--r-- | src/bench/checkblock.cpp | 3 | ||||
-rw-r--r-- | src/bench/checkqueue.cpp | 103 | ||||
-rw-r--r-- | src/bench/coin_selection.cpp | 14 | ||||
-rw-r--r-- | src/bench/crypto_hash.cpp | 10 | ||||
-rw-r--r-- | src/bench/mempool_eviction.cpp | 4 | ||||
-rw-r--r-- | src/bench/perf.cpp | 53 | ||||
-rw-r--r-- | src/bench/perf.h | 37 | ||||
-rw-r--r-- | src/bench/verify_script.cpp | 9 |
13 files changed, 253 insertions, 40 deletions
diff --git a/src/bench/Examples.cpp b/src/bench/Examples.cpp index b6b020a971..314947d48c 100644 --- a/src/bench/Examples.cpp +++ b/src/bench/Examples.cpp @@ -1,9 +1,9 @@ -// Copyright (c) 2015 The Bitcoin Core developers +// Copyright (c) 2015-2016 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 "bench.h" -#include "main.h" +#include "validation.h" #include "utiltime.h" // Sanity test: this should loop ten times, and diff --git a/src/bench/base58.cpp b/src/bench/base58.cpp index a791b5b7fa..3319c179bf 100644 --- a/src/bench/base58.cpp +++ b/src/bench/base58.cpp @@ -4,7 +4,7 @@ #include "bench.h" -#include "main.h" +#include "validation.h" #include "base58.h" #include <vector> diff --git a/src/bench/bench.cpp b/src/bench/bench.cpp index 8942da8c74..1bd9d06b80 100644 --- a/src/bench/bench.cpp +++ b/src/bench/bench.cpp @@ -1,16 +1,15 @@ -// Copyright (c) 2015 The Bitcoin Core developers +// Copyright (c) 2015-2016 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 "bench.h" +#include "perf.h" #include <iostream> #include <iomanip> #include <sys/time.h> -using namespace benchmark; - -std::map<std::string, BenchFunction> BenchRunner::benchmarks; +std::map<std::string, benchmark::BenchFunction> benchmark::BenchRunner::benchmarks; static double gettimedouble(void) { struct timeval tv; @@ -18,34 +17,39 @@ static double gettimedouble(void) { return tv.tv_usec * 0.000001 + tv.tv_sec; } -BenchRunner::BenchRunner(std::string name, BenchFunction func) +benchmark::BenchRunner::BenchRunner(std::string name, benchmark::BenchFunction func) { benchmarks.insert(std::make_pair(name, func)); } void -BenchRunner::RunAll(double elapsedTimeForOne) +benchmark::BenchRunner::RunAll(double elapsedTimeForOne) { - std::cout << "#Benchmark" << "," << "count" << "," << "min" << "," << "max" << "," << "average" << "\n"; + perf_init(); + std::cout << "#Benchmark" << "," << "count" << "," << "min" << "," << "max" << "," << "average" << "," + << "min_cycles" << "," << "max_cycles" << "," << "average_cycles" << "\n"; - for (std::map<std::string,BenchFunction>::iterator it = benchmarks.begin(); + for (std::map<std::string,benchmark::BenchFunction>::iterator it = benchmarks.begin(); it != benchmarks.end(); ++it) { State state(it->first, elapsedTimeForOne); - BenchFunction& func = it->second; + benchmark::BenchFunction& func = it->second; func(state); } + perf_fini(); } -bool State::KeepRunning() +bool benchmark::State::KeepRunning() { if (count & countMask) { ++count; return true; } double now; + uint64_t nowCycles; if (count == 0) { lastTime = beginTime = now = gettimedouble(); + lastCycles = beginCycles = nowCycles = perf_cpucycles(); } else { now = gettimedouble(); @@ -53,6 +57,13 @@ bool State::KeepRunning() double elapsedOne = elapsed * countMaskInv; if (elapsedOne < minTime) minTime = elapsedOne; if (elapsedOne > maxTime) maxTime = elapsedOne; + + // We only use relative values, so don't have to handle 64-bit wrap-around specially + nowCycles = perf_cpucycles(); + uint64_t elapsedOneCycles = (nowCycles - lastCycles) * countMaskInv; + if (elapsedOneCycles < minCycles) minCycles = elapsedOneCycles; + if (elapsedOneCycles > maxCycles) maxCycles = elapsedOneCycles; + if (elapsed*128 < maxElapsed) { // If the execution was much too fast (1/128th of maxElapsed), increase the count mask by 8x and restart timing. // The restart avoids including the overhead of this code in the measurement. @@ -61,6 +72,8 @@ bool State::KeepRunning() count = 0; minTime = std::numeric_limits<double>::max(); maxTime = std::numeric_limits<double>::min(); + minCycles = std::numeric_limits<uint64_t>::max(); + maxCycles = std::numeric_limits<uint64_t>::min(); return true; } if (elapsed*16 < maxElapsed) { @@ -72,6 +85,7 @@ bool State::KeepRunning() } } lastTime = now; + lastCycles = nowCycles; ++count; if (now - beginTime < maxElapsed) return true; // Keep going @@ -80,7 +94,9 @@ bool State::KeepRunning() // Output results double average = (now-beginTime)/count; - std::cout << std::fixed << std::setprecision(15) << name << "," << count << "," << minTime << "," << maxTime << "," << average << "\n"; + int64_t averageCycles = (nowCycles-beginCycles)/count; + std::cout << std::fixed << std::setprecision(15) << name << "," << count << "," << minTime << "," << maxTime << "," << average << "," + << minCycles << "," << maxCycles << "," << averageCycles << "\n"; return false; } diff --git a/src/bench/bench.h b/src/bench/bench.h index f13b145aaf..80dad6a8ef 100644 --- a/src/bench/bench.h +++ b/src/bench/bench.h @@ -1,4 +1,4 @@ -// Copyright (c) 2015 The Bitcoin Core developers +// Copyright (c) 2015-2016 The Bitcoin Core developers // Distributed under the MIT software license, see the accompanying // file COPYING or http://www.opensource.org/licenses/mit-license.php. @@ -41,12 +41,18 @@ namespace benchmark { double maxElapsed; double beginTime; double lastTime, minTime, maxTime, countMaskInv; - int64_t count; - int64_t countMask; + uint64_t count; + uint64_t countMask; + uint64_t beginCycles; + uint64_t lastCycles; + uint64_t minCycles; + uint64_t maxCycles; public: State(std::string _name, double _maxElapsed) : name(_name), maxElapsed(_maxElapsed), count(0) { minTime = std::numeric_limits<double>::max(); maxTime = std::numeric_limits<double>::min(); + minCycles = std::numeric_limits<uint64_t>::max(); + maxCycles = std::numeric_limits<uint64_t>::min(); countMask = 1; countMaskInv = 1./(countMask + 1); } diff --git a/src/bench/bench_bitcoin.cpp b/src/bench/bench_bitcoin.cpp index db1402216d..61a0b31aed 100644 --- a/src/bench/bench_bitcoin.cpp +++ b/src/bench/bench_bitcoin.cpp @@ -1,11 +1,11 @@ -// Copyright (c) 2015 The Bitcoin Core developers +// Copyright (c) 2015-2016 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 "bench.h" #include "key.h" -#include "main.h" +#include "validation.h" #include "util.h" int diff --git a/src/bench/checkblock.cpp b/src/bench/checkblock.cpp index 4a564d3fc8..230e4ca773 100644 --- a/src/bench/checkblock.cpp +++ b/src/bench/checkblock.cpp @@ -5,7 +5,8 @@ #include "bench.h" #include "chainparams.h" -#include "main.h" +#include "validation.h" +#include "streams.h" #include "consensus/validation.h" namespace block_bench { diff --git a/src/bench/checkqueue.cpp b/src/bench/checkqueue.cpp new file mode 100644 index 0000000000..6fa9fe4fe8 --- /dev/null +++ b/src/bench/checkqueue.cpp @@ -0,0 +1,103 @@ +// Copyright (c) 2015 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 "bench.h" +#include "util.h" +#include "validation.h" +#include "checkqueue.h" +#include "prevector.h" +#include <vector> +#include <boost/thread/thread.hpp> +#include "random.h" + + +// This Benchmark tests the CheckQueue with the lightest +// weight Checks, so it should make any lock contention +// particularly visible +static const int MIN_CORES = 2; +static const size_t BATCHES = 101; +static const size_t BATCH_SIZE = 30; +static const int PREVECTOR_SIZE = 28; +static const int QUEUE_BATCH_SIZE = 128; +static void CCheckQueueSpeed(benchmark::State& state) +{ + struct FakeJobNoWork { + bool operator()() + { + return true; + } + void swap(FakeJobNoWork& x){}; + }; + CCheckQueue<FakeJobNoWork> queue {QUEUE_BATCH_SIZE}; + boost::thread_group tg; + for (auto x = 0; x < std::max(MIN_CORES, GetNumCores()); ++x) { + tg.create_thread([&]{queue.Thread();}); + } + while (state.KeepRunning()) { + CCheckQueueControl<FakeJobNoWork> control(&queue); + + // We call Add a number of times to simulate the behavior of adding + // a block of transactions at once. + + std::vector<std::vector<FakeJobNoWork>> vBatches(BATCHES); + for (auto& vChecks : vBatches) { + vChecks.resize(BATCH_SIZE); + } + for (auto& vChecks : vBatches) { + // We can't make vChecks in the inner loop because we want to measure + // the cost of getting the memory to each thread and we might get the same + // memory + control.Add(vChecks); + } + // control waits for completion by RAII, but + // it is done explicitly here for clarity + control.Wait(); + } + tg.interrupt_all(); + tg.join_all(); +} + +// This Benchmark tests the CheckQueue with a slightly realistic workload, +// where checks all contain a prevector that is indirect 50% of the time +// and there is a little bit of work done between calls to Add. +static void CCheckQueueSpeedPrevectorJob(benchmark::State& state) +{ + struct PrevectorJob { + prevector<PREVECTOR_SIZE, uint8_t> p; + PrevectorJob(){ + } + PrevectorJob(FastRandomContext& insecure_rand){ + p.resize(insecure_rand.rand32() % (PREVECTOR_SIZE*2)); + } + bool operator()() + { + return true; + } + void swap(PrevectorJob& x){p.swap(x.p);}; + }; + CCheckQueue<PrevectorJob> queue {QUEUE_BATCH_SIZE}; + boost::thread_group tg; + for (auto x = 0; x < std::max(MIN_CORES, GetNumCores()); ++x) { + tg.create_thread([&]{queue.Thread();}); + } + while (state.KeepRunning()) { + // Make insecure_rand here so that each iteration is identical. + FastRandomContext insecure_rand(true); + CCheckQueueControl<PrevectorJob> control(&queue); + std::vector<std::vector<PrevectorJob>> vBatches(BATCHES); + for (auto& vChecks : vBatches) { + vChecks.reserve(BATCH_SIZE); + for (size_t x = 0; x < BATCH_SIZE; ++x) + vChecks.emplace_back(insecure_rand); + control.Add(vChecks); + } + // control waits for completion by RAII, but + // it is done explicitly here for clarity + control.Wait(); + } + tg.interrupt_all(); + tg.join_all(); +} +BENCHMARK(CCheckQueueSpeed); +BENCHMARK(CCheckQueueSpeedPrevectorJob); diff --git a/src/bench/coin_selection.cpp b/src/bench/coin_selection.cpp index 7091ee3e11..29fbd34631 100644 --- a/src/bench/coin_selection.cpp +++ b/src/bench/coin_selection.cpp @@ -1,4 +1,4 @@ -// Copyright (c) 2012-2015 The Bitcoin Core developers +// Copyright (c) 2012-2016 The Bitcoin Core developers // Distributed under the MIT software license, see the accompanying // file COPYING or http://www.opensource.org/licenses/mit-license.php. @@ -8,9 +8,7 @@ #include <boost/foreach.hpp> #include <set> -using namespace std; - -static void addCoin(const CAmount& nValue, const CWallet& wallet, vector<COutput>& vCoins) +static void addCoin(const CAmount& nValue, const CWallet& wallet, std::vector<COutput>& vCoins) { int nInput = 0; @@ -19,7 +17,7 @@ static void addCoin(const CAmount& nValue, const CWallet& wallet, vector<COutput tx.nLockTime = nextLockTime++; // so all transactions get different hashes tx.vout.resize(nInput + 1); tx.vout[nInput].nValue = nValue; - CWalletTx* wtx = new CWalletTx(&wallet, tx); + CWalletTx* wtx = new CWalletTx(&wallet, MakeTransactionRef(std::move(tx))); int nAge = 6 * 24; COutput output(wtx, nInput, nAge, true, true); @@ -36,7 +34,7 @@ static void addCoin(const CAmount& nValue, const CWallet& wallet, vector<COutput static void CoinSelection(benchmark::State& state) { const CWallet wallet; - vector<COutput> vCoins; + std::vector<COutput> vCoins; LOCK(wallet.cs_wallet); while (state.KeepRunning()) { @@ -50,9 +48,9 @@ static void CoinSelection(benchmark::State& state) addCoin(1000 * COIN, wallet, vCoins); addCoin(3 * COIN, wallet, vCoins); - set<pair<const CWalletTx*, unsigned int> > setCoinsRet; + std::set<std::pair<const CWalletTx*, unsigned int> > setCoinsRet; CAmount nValueRet; - bool success = wallet.SelectCoinsMinConf(1003 * COIN, 1, 6, vCoins, setCoinsRet, nValueRet); + bool success = wallet.SelectCoinsMinConf(1003 * COIN, 1, 6, 0, vCoins, setCoinsRet, nValueRet); assert(success); assert(nValueRet == 1003 * COIN); assert(setCoinsRet.size() == 2); diff --git a/src/bench/crypto_hash.cpp b/src/bench/crypto_hash.cpp index 168006154f..737d3572ae 100644 --- a/src/bench/crypto_hash.cpp +++ b/src/bench/crypto_hash.cpp @@ -22,7 +22,7 @@ static void RIPEMD160(benchmark::State& state) uint8_t hash[CRIPEMD160::OUTPUT_SIZE]; std::vector<uint8_t> in(BUFFER_SIZE,0); while (state.KeepRunning()) - CRIPEMD160().Write(begin_ptr(in), in.size()).Finalize(hash); + CRIPEMD160().Write(in.data(), in.size()).Finalize(hash); } static void SHA1(benchmark::State& state) @@ -30,7 +30,7 @@ static void SHA1(benchmark::State& state) uint8_t hash[CSHA1::OUTPUT_SIZE]; std::vector<uint8_t> in(BUFFER_SIZE,0); while (state.KeepRunning()) - CSHA1().Write(begin_ptr(in), in.size()).Finalize(hash); + CSHA1().Write(in.data(), in.size()).Finalize(hash); } static void SHA256(benchmark::State& state) @@ -38,7 +38,7 @@ static void SHA256(benchmark::State& state) uint8_t hash[CSHA256::OUTPUT_SIZE]; std::vector<uint8_t> in(BUFFER_SIZE,0); while (state.KeepRunning()) - CSHA256().Write(begin_ptr(in), in.size()).Finalize(hash); + CSHA256().Write(in.data(), in.size()).Finalize(hash); } static void SHA256_32b(benchmark::State& state) @@ -46,7 +46,7 @@ static void SHA256_32b(benchmark::State& state) std::vector<uint8_t> in(32,0); while (state.KeepRunning()) { for (int i = 0; i < 1000000; i++) { - CSHA256().Write(begin_ptr(in), in.size()).Finalize(&in[0]); + CSHA256().Write(in.data(), in.size()).Finalize(&in[0]); } } } @@ -56,7 +56,7 @@ static void SHA512(benchmark::State& state) uint8_t hash[CSHA512::OUTPUT_SIZE]; std::vector<uint8_t> in(BUFFER_SIZE,0); while (state.KeepRunning()) - CSHA512().Write(begin_ptr(in), in.size()).Finalize(hash); + CSHA512().Write(in.data(), in.size()).Finalize(hash); } static void SipHash_32b(benchmark::State& state) diff --git a/src/bench/mempool_eviction.cpp b/src/bench/mempool_eviction.cpp index 0ae69c75fc..5790d51a82 100644 --- a/src/bench/mempool_eviction.cpp +++ b/src/bench/mempool_eviction.cpp @@ -1,4 +1,4 @@ -// Copyright (c) 2011-2015 The Bitcoin Core developers +// Copyright (c) 2011-2016 The Bitcoin Core developers // Distributed under the MIT software license, see the accompanying // file COPYING or http://www.opensource.org/licenses/mit-license.php. @@ -18,7 +18,7 @@ static void AddTx(const CTransaction& tx, const CAmount& nFee, CTxMemPool& pool) unsigned int sigOpCost = 4; LockPoints lp; pool.addUnchecked(tx.GetHash(), CTxMemPoolEntry( - tx, nFee, nTime, dPriority, nHeight, pool.HasNoInputsOf(tx), + MakeTransactionRef(tx), nFee, nTime, dPriority, nHeight, tx.GetValueOut(), spendsCoinbase, sigOpCost, lp)); } diff --git a/src/bench/perf.cpp b/src/bench/perf.cpp new file mode 100644 index 0000000000..a549ec29ea --- /dev/null +++ b/src/bench/perf.cpp @@ -0,0 +1,53 @@ +// 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 "perf.h" + +#if defined(__i386__) || defined(__x86_64__) + +/* These architectures support querying the cycle counter + * from user space, no need for any syscall overhead. + */ +void perf_init(void) { } +void perf_fini(void) { } + +#elif defined(__linux__) + +#include <unistd.h> +#include <sys/syscall.h> +#include <linux/perf_event.h> + +static int fd = -1; +static struct perf_event_attr attr; + +void perf_init(void) +{ + attr.type = PERF_TYPE_HARDWARE; + attr.config = PERF_COUNT_HW_CPU_CYCLES; + fd = syscall(__NR_perf_event_open, &attr, 0, -1, -1, 0); +} + +void perf_fini(void) +{ + if (fd != -1) { + close(fd); + } +} + +uint64_t perf_cpucycles(void) +{ + uint64_t result = 0; + if (fd == -1 || read(fd, &result, sizeof(result)) < (ssize_t)sizeof(result)) { + return 0; + } + return result; +} + +#else /* Unhandled platform */ + +void perf_init(void) { } +void perf_fini(void) { } +uint64_t perf_cpucycles(void) { return 0; } + +#endif diff --git a/src/bench/perf.h b/src/bench/perf.h new file mode 100644 index 0000000000..681bd0c8a2 --- /dev/null +++ b/src/bench/perf.h @@ -0,0 +1,37 @@ +// 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. + +/** Functions for measurement of CPU cycles */ +#ifndef H_PERF +#define H_PERF + +#include <stdint.h> + +#if defined(__i386__) + +static inline uint64_t perf_cpucycles(void) +{ + uint64_t x; + __asm__ volatile (".byte 0x0f, 0x31" : "=A" (x)); + return x; +} + +#elif defined(__x86_64__) + +static inline uint64_t perf_cpucycles(void) +{ + uint32_t hi, lo; + __asm__ __volatile__ ("rdtsc" : "=a"(lo), "=d"(hi)); + return ((uint64_t)lo)|(((uint64_t)hi)<<32); +} +#else + +uint64_t perf_cpucycles(void); + +#endif + +void perf_init(void); +void perf_fini(void); + +#endif // H_PERF diff --git a/src/bench/verify_script.cpp b/src/bench/verify_script.cpp index dc3940cdbd..23bbadc88d 100644 --- a/src/bench/verify_script.cpp +++ b/src/bench/verify_script.cpp @@ -36,7 +36,6 @@ static CMutableTransaction BuildSpendingTransaction(const CScript& scriptSig, co txSpend.nLockTime = 0; txSpend.vin.resize(1); txSpend.vout.resize(1); - txSpend.wit.vtxinwit.resize(1); txSpend.vin[0].prevout.hash = txCredit.GetHash(); txSpend.vin[0].prevout.n = 0; txSpend.vin[0].scriptSig = scriptSig; @@ -68,7 +67,7 @@ static void VerifyScriptBench(benchmark::State& state) CScript witScriptPubkey = CScript() << OP_DUP << OP_HASH160 << ToByteVector(pubkeyHash) << OP_EQUALVERIFY << OP_CHECKSIG; CTransaction txCredit = BuildCreditingTransaction(scriptPubKey); CMutableTransaction txSpend = BuildSpendingTransaction(scriptSig, txCredit); - CScriptWitness& witness = txSpend.wit.vtxinwit[0].scriptWitness; + CScriptWitness& witness = txSpend.vin[0].scriptWitness; witness.stack.emplace_back(); key.Sign(SignatureHash(witScriptPubkey, txSpend, 0, SIGHASH_ALL, txCredit.vout[0].nValue, SIGVERSION_WITNESS_V0), witness.stack.back(), 0); witness.stack.back().push_back(static_cast<unsigned char>(SIGHASH_ALL)); @@ -80,7 +79,7 @@ static void VerifyScriptBench(benchmark::State& state) bool success = VerifyScript( txSpend.vin[0].scriptSig, txCredit.vout[0].scriptPubKey, - &txSpend.wit.vtxinwit[0].scriptWitness, + &txSpend.vin[0].scriptWitness, flags, MutableTransactionSignatureChecker(&txSpend, 0, txCredit.vout[0].nValue), &err); @@ -91,10 +90,10 @@ static void VerifyScriptBench(benchmark::State& state) CDataStream stream(SER_NETWORK, PROTOCOL_VERSION); stream << txSpend; int csuccess = bitcoinconsensus_verify_script_with_amount( - begin_ptr(txCredit.vout[0].scriptPubKey), + txCredit.vout[0].scriptPubKey.data(), txCredit.vout[0].scriptPubKey.size(), txCredit.vout[0].nValue, - (const unsigned char*)&stream[0], stream.size(), 0, flags, nullptr); + (const unsigned char*)stream.data(), stream.size(), 0, flags, nullptr); assert(csuccess == 1); #endif } |