bsc/core/state/state_object.go

578 lines
17 KiB
Go
Raw Normal View History

2015-07-07 03:54:22 +03:00
// Copyright 2014 The go-ethereum Authors
// This file is part of the go-ethereum library.
2015-07-07 03:54:22 +03:00
//
// The go-ethereum library is free software: you can redistribute it and/or modify
2015-07-07 03:54:22 +03:00
// it under the terms of the GNU Lesser General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// The go-ethereum library is distributed in the hope that it will be useful,
2015-07-07 03:54:22 +03:00
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
2015-07-07 03:54:22 +03:00
// GNU Lesser General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public License
// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>.
2015-07-07 03:54:22 +03:00
2014-10-31 15:43:14 +02:00
package state
import (
"bytes"
"fmt"
"io"
"math/big"
"time"
2015-03-16 12:27:38 +02:00
"github.com/ethereum/go-ethereum/common"
2015-03-16 12:59:52 +02:00
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/metrics"
"github.com/ethereum/go-ethereum/rlp"
)
var emptyCodeHash = crypto.Keccak256(nil)
type Code []byte
func (c Code) String() string {
return string(c) //strings.Join(Disassemble(c), " ")
}
type Storage map[common.Hash]common.Hash
func (s Storage) String() (str string) {
for key, value := range s {
str += fmt.Sprintf("%X : %X\n", key, value)
}
return
}
func (s Storage) Copy() Storage {
cpy := make(Storage)
for key, value := range s {
cpy[key] = value
}
return cpy
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
// StateObject represents an Ethereum account which is being modified.
//
// The usage pattern is as follows:
// First you need to obtain a state object.
// Account values can be accessed and modified through the object.
// Finally, call CommitTrie to write the modified storage trie into a database.
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
type StateObject struct {
address common.Address
addrHash common.Hash // hash of ethereum address of the account
data Account
db *StateDB
// DB error.
// State objects are used by the consensus core and VM which are
// unable to deal with database-level errors. Any error that occurs
// during a database read is memoized here and will eventually be returned
// by StateDB.Commit.
dbErr error
// Write caches.
trie Trie // storage trie, which becomes non-nil on first access
code Code // contract bytecode, which gets set when code is loaded
originStorage Storage // Storage cache of original entries to dedup rewrites, reset for every transaction
pendingStorage Storage // Storage entries that need to be flushed to disk, at the end of an entire block
dirtyStorage Storage // Storage entries that have been modified in the current transaction execution
fakeStorage Storage // Fake storage which constructed by caller for debugging purpose.
// Cache flags.
// When an object is marked suicided it will be delete from the trie
// during the "update" phase of the state transition.
dirtyCode bool // true if the code was updated
suicided bool
deleted bool
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
//encode
encodeData []byte
}
// empty returns whether the account is considered empty.
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) empty() bool {
return s.data.Nonce == 0 && s.data.Balance.Sign() == 0 && bytes.Equal(s.data.CodeHash, emptyCodeHash)
}
// Account is the Ethereum consensus representation of accounts.
// These objects are stored in the main account trie.
type Account struct {
Nonce uint64
Balance *big.Int
Root common.Hash // merkle root of the storage trie
CodeHash []byte
}
2015-04-04 13:40:11 +03:00
// newObject creates a state object.
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func newObject(db *StateDB, address common.Address, data Account) *StateObject {
if data.Balance == nil {
data.Balance = new(big.Int)
}
if data.CodeHash == nil {
data.CodeHash = emptyCodeHash
2015-04-04 13:40:11 +03:00
}
if data.Root == (common.Hash{}) {
data.Root = emptyRoot
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
return &StateObject{
db: db,
address: address,
addrHash: crypto.Keccak256Hash(address[:]),
data: data,
originStorage: make(Storage),
pendingStorage: make(Storage),
dirtyStorage: make(Storage),
}
}
// EncodeRLP implements rlp.Encoder.
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) EncodeRLP(w io.Writer) error {
return rlp.Encode(w, s.data)
}
// setError remembers the first non-nil error it is called with.
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) setError(err error) {
if s.dbErr == nil {
s.dbErr = err
}
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) markSuicided() {
s.suicided = true
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) touch() {
s.db.journal.append(touchChange{
account: &s.address,
})
if s.address == ripemd {
// Explicitly put it in the dirty-cache, which is otherwise generated from
// flattened journals.
s.db.journal.dirty(s.address)
}
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) getTrie(db Database) Trie {
if s.trie == nil {
// Try fetching from prefetcher first
// We don't prefetch empty tries
if s.data.Root != emptyRoot && s.db.prefetcher != nil {
// When the miner is creating the pending state, there is no
// prefetcher
s.trie = s.db.prefetcher.trie(s.data.Root)
}
if s.trie == nil {
var err error
s.trie, err = db.OpenStorageTrie(s.addrHash, s.data.Root)
if err != nil {
s.trie, _ = db.OpenStorageTrie(s.addrHash, common.Hash{})
s.setError(fmt.Errorf("can't create storage trie: %v", err))
}
}
}
return s.trie
}
// GetState retrieves a value from the account storage trie.
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) GetState(db Database, key common.Hash) common.Hash {
// If the fake storage is set, only lookup the state here(in the debugging mode)
if s.fakeStorage != nil {
return s.fakeStorage[key]
}
// If we have a dirty value for this state entry, return it
value, dirty := s.dirtyStorage[key]
if dirty {
return value
}
// Otherwise return the entry's original value
return s.GetCommittedState(db, key)
}
// GetCommittedState retrieves a value from the committed account storage trie.
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) GetCommittedState(db Database, key common.Hash) common.Hash {
// If the fake storage is set, only lookup the state here(in the debugging mode)
if s.fakeStorage != nil {
return s.fakeStorage[key]
}
// If we have a pending write or clean cached, return that
if value, pending := s.pendingStorage[key]; pending {
return value
}
if value, cached := s.originStorage[key]; cached {
return value
}
// If no live objects are available, attempt to use snapshots
var (
enc []byte
err error
meter *time.Duration
)
readStart := time.Now()
if metrics.EnabledExpensive {
// If the snap is 'under construction', the first lookup may fail. If that
// happens, we don't want to double-count the time elapsed. Thus this
// dance with the metering.
defer func() {
if meter != nil {
*meter += time.Since(readStart)
}
}()
}
if s.db.snap != nil {
if metrics.EnabledExpensive {
meter = &s.db.SnapshotStorageReads
}
// If the object was destructed in *this* block (and potentially resurrected),
// the storage has been cleared out, and we should *not* consult the previous
// snapshot about any storage values. The only possible alternatives are:
// 1) resurrect happened, and new slot values were set -- those should
// have been handles via pendingStorage above.
// 2) we don't have new values, and can deliver empty response back
if _, destructed := s.db.snapDestructs[s.addrHash]; destructed {
return common.Hash{}
}
enc, err = s.db.snap.Storage(s.addrHash, crypto.Keccak256Hash(key.Bytes()))
}
// If snapshot unavailable or reading from it failed, load from the database
if s.db.snap == nil || err != nil {
if meter != nil {
// If we already spent time checking the snapshot, account for it
// and reset the readStart
*meter += time.Since(readStart)
readStart = time.Now()
}
if metrics.EnabledExpensive {
meter = &s.db.StorageReads
}
if enc, err = s.getTrie(db).TryGet(key.Bytes()); err != nil {
s.setError(err)
return common.Hash{}
}
}
var value common.Hash
if len(enc) > 0 {
_, content, _, err := rlp.Split(enc)
if err != nil {
s.setError(err)
}
value.SetBytes(content)
}
s.originStorage[key] = value
return value
}
// SetState updates a value in account storage.
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) SetState(db Database, key, value common.Hash) {
// If the fake storage is set, put the temporary state update here.
if s.fakeStorage != nil {
s.fakeStorage[key] = value
return
}
// If the new value is the same as old, don't set
prev := s.GetState(db, key)
if prev == value {
return
}
// New value is different, update and journal the change
s.db.journal.append(storageChange{
account: &s.address,
key: key,
prevalue: prev,
})
s.setState(key, value)
}
// SetStorage replaces the entire state storage with the given one.
//
// After this function is called, all original state will be ignored and state
// lookup only happens in the fake state storage.
//
// Note this function should only be used for debugging purpose.
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) SetStorage(storage map[common.Hash]common.Hash) {
// Allocate fake storage if it's nil.
if s.fakeStorage == nil {
s.fakeStorage = make(Storage)
}
for key, value := range storage {
s.fakeStorage[key] = value
}
// Don't bother journal since this function should only be used for
// debugging and the `fake` storage won't be committed to database.
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) setState(key, value common.Hash) {
s.dirtyStorage[key] = value
}
// finalise moves all dirty storage slots into the pending area to be hashed or
// committed later. It is invoked at the end of every transaction.
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) finalise(prefetch bool) {
slotsToPrefetch := make([][]byte, 0, len(s.dirtyStorage))
for key, value := range s.dirtyStorage {
s.pendingStorage[key] = value
if value != s.originStorage[key] {
slotsToPrefetch = append(slotsToPrefetch, common.CopyBytes(key[:])) // Copy needed for closure
}
}
if s.db.prefetcher != nil && prefetch && len(slotsToPrefetch) > 0 && s.data.Root != emptyRoot {
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
s.db.prefetcher.prefetch(s.data.Root, slotsToPrefetch, s.addrHash)
}
if len(s.dirtyStorage) > 0 {
s.dirtyStorage = make(Storage)
}
}
// updateTrie writes cached storage modifications into the object's storage trie.
// It will return nil if the trie has not been loaded and no changes have been made
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) updateTrie(db Database) Trie {
// Make sure all dirty slots are finalized into the pending storage area
s.finalise(false) // Don't prefetch any more, pull directly if need be
if len(s.pendingStorage) == 0 {
return s.trie
}
// Track the amount of time wasted on updating the storage trie
if metrics.EnabledExpensive {
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
defer func(start time.Time) {
s.db.MetricsMux.Lock()
s.db.StorageUpdates += time.Since(start)
s.db.MetricsMux.Unlock()
}(time.Now())
}
// The snapshot storage map for the object
var storage map[common.Hash][]byte
// Insert all the pending updates into the trie
tr := s.getTrie(db)
hasher := s.db.hasher
usedStorage := make([][]byte, 0, len(s.pendingStorage))
for key, value := range s.pendingStorage {
// Skip noop changes, persist actual changes
if value == s.originStorage[key] {
continue
}
s.originStorage[key] = value
var v []byte
if (value == common.Hash{}) {
s.setError(tr.TryDelete(key[:]))
} else {
// Encoding []byte cannot fail, ok to ignore the error.
v, _ = rlp.EncodeToBytes(common.TrimLeftZeroes(value[:]))
s.setError(tr.TryUpdate(key[:], v))
}
// If state snapshotting is active, cache the data til commit
if s.db.snap != nil {
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
s.db.snapMux.Lock()
if storage == nil {
// Retrieve the old storage map, if available, create a new one otherwise
if storage = s.db.snapStorage[s.addrHash]; storage == nil {
storage = make(map[common.Hash][]byte)
s.db.snapStorage[s.addrHash] = storage
}
}
storage[crypto.HashData(hasher, key[:])] = v // v will be nil if value is 0x00
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
s.db.snapMux.Unlock()
}
usedStorage = append(usedStorage, common.CopyBytes(key[:])) // Copy needed for closure
}
if s.db.prefetcher != nil {
s.db.prefetcher.used(s.data.Root, usedStorage)
}
if len(s.pendingStorage) > 0 {
s.pendingStorage = make(Storage)
}
2017-04-19 13:09:04 +03:00
return tr
}
// UpdateRoot sets the trie root to the current root hash of
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) updateRoot(db Database) {
// If nothing changed, don't bother with hashing anything
if s.updateTrie(db) == nil {
return
}
// Track the amount of time wasted on hashing the storage trie
if metrics.EnabledExpensive {
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
defer func(start time.Time) {
s.db.MetricsMux.Lock()
s.db.StorageHashes += time.Since(start)
s.db.MetricsMux.Unlock()
}(time.Now())
}
s.data.Root = s.trie.Hash()
}
2018-06-05 05:56:45 +03:00
// CommitTrie the storage trie of the object to db.
// This updates the trie root.
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) CommitTrie(db Database) error {
// If nothing changed, don't bother with hashing anything
if s.updateTrie(db) == nil {
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
if s.trie != nil && s.data.Root != emptyRoot {
db.CacheStorage(s.addrHash, s.data.Root, s.trie)
}
return nil
}
if s.dbErr != nil {
return s.dbErr
}
// Track the amount of time wasted on committing the storage trie
if metrics.EnabledExpensive {
defer func(start time.Time) { s.db.StorageCommits += time.Since(start) }(time.Now())
}
root, err := s.trie.Commit(nil)
if err == nil {
s.data.Root = root
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
if s.data.Root != emptyRoot {
db.CacheStorage(s.addrHash, s.data.Root, s.trie)
}
return err
}
// AddBalance adds amount to s's balance.
// It is used to add funds to the destination account of a transfer.
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) AddBalance(amount *big.Int) {
// EIP161: We must check emptiness for the objects such that the account
// clearing (0,0,0 objects) can take effect.
if amount.Sign() == 0 {
if s.empty() {
s.touch()
}
return
}
s.SetBalance(new(big.Int).Add(s.Balance(), amount))
}
// SubBalance removes amount from s's balance.
// It is used to remove funds from the origin account of a transfer.
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) SubBalance(amount *big.Int) {
if amount.Sign() == 0 {
return
}
s.SetBalance(new(big.Int).Sub(s.Balance(), amount))
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) SetBalance(amount *big.Int) {
s.db.journal.append(balanceChange{
account: &s.address,
prev: new(big.Int).Set(s.data.Balance),
})
s.setBalance(amount)
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) setBalance(amount *big.Int) {
s.data.Balance = amount
}
// Return the gas back to the origin. Used by the Virtual machine or Closures
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) ReturnGas(gas *big.Int) {}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) deepCopy(db *StateDB) *StateObject {
stateObject := newObject(db, s.address, s.data)
if s.trie != nil {
stateObject.trie = db.db.CopyTrie(s.trie)
2017-04-19 13:09:04 +03:00
}
stateObject.code = s.code
stateObject.dirtyStorage = s.dirtyStorage.Copy()
stateObject.originStorage = s.originStorage.Copy()
stateObject.pendingStorage = s.pendingStorage.Copy()
stateObject.suicided = s.suicided
stateObject.dirtyCode = s.dirtyCode
stateObject.deleted = s.deleted
return stateObject
}
//
// Attribute accessors
//
// Returns the address of the contract/account
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) Address() common.Address {
return s.address
}
// Code returns the contract code associated with this object, if any.
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) Code(db Database) []byte {
if s.code != nil {
return s.code
}
if bytes.Equal(s.CodeHash(), emptyCodeHash) {
return nil
}
code, err := db.ContractCode(s.addrHash, common.BytesToHash(s.CodeHash()))
if err != nil {
s.setError(fmt.Errorf("can't load code hash %x: %v", s.CodeHash(), err))
}
s.code = code
return code
2014-11-04 00:45:44 +02:00
}
// CodeSize returns the size of the contract code associated with this object,
// or zero if none. This method is an almost mirror of Code, but uses a cache
// inside the database to avoid loading codes seen recently.
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) CodeSize(db Database) int {
if s.code != nil {
return len(s.code)
}
if bytes.Equal(s.CodeHash(), emptyCodeHash) {
return 0
}
size, err := db.ContractCodeSize(s.addrHash, common.BytesToHash(s.CodeHash()))
if err != nil {
s.setError(fmt.Errorf("can't load code size %x: %v", s.CodeHash(), err))
}
return size
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) SetCode(codeHash common.Hash, code []byte) {
prevcode := s.Code(s.db.db)
s.db.journal.append(codeChange{
account: &s.address,
prevhash: s.CodeHash(),
prevcode: prevcode,
})
s.setCode(codeHash, code)
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) setCode(codeHash common.Hash, code []byte) {
s.code = code
s.data.CodeHash = codeHash[:]
s.dirtyCode = true
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) SetNonce(nonce uint64) {
s.db.journal.append(nonceChange{
account: &s.address,
prev: s.data.Nonce,
})
s.setNonce(nonce)
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) setNonce(nonce uint64) {
s.data.Nonce = nonce
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) CodeHash() []byte {
return s.data.CodeHash
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) Balance() *big.Int {
return s.data.Balance
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) Nonce() uint64 {
return s.data.Nonce
}
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
// Never called, but must be present to allow StateObject to be used
// as a vm.Account interface that also satisfies the vm.ContractRef
// interface. Interfaces are awesome.
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
func (s *StateObject) Value() *big.Int {
panic("Value on StateObject should never be called")
}