core/vm: optimize push2 opcode #31267
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
MariusVanDerWijden
requested review from
holiman and
rjl493456442
as code owners
Member
Author
|
Back of the envelope calculation, this will save us between 0.2 and 0.5 ms per block, which is quite significant for such a small change |
holiman
reviewed
Feb 26, 2025
core/vm/instructions.go
Outdated
| integer = new(uint256.Int) | ||
| ) | ||
| if *pc+2 < codeLen { | ||
| scope.Stack.push(integer.SetUint64(uint64(binary.BigEndian.Uint16(scope.Contract.Code[*pc+1 : *pc+3])))) |
Contributor
There was a problem hiding this comment.
I'm pretty certain there is a way for uint256 to parse the bytes from scope.Contract.Code which does not go via binary.BigEndian. (If there isn't, I feel like there should be)
Contributor
There was a problem hiding this comment.
Ah, well there's func (z *Int) SetBytes2(in []byte) *Int , but performance-wise it's about the same as what you do here. It would maybe look a little bit cleaner:
scope.Stack.push(integer.SetBytes2(scope.contract.Code[*pc+1:]))
Member
Author
There was a problem hiding this comment.
Applied, yep thats a cleaner way
jwasinger
reviewed
Feb 26, 2025
Co-authored-by: jwasinger <[email protected]>
Member
Author
|
Did another benchmark, a full block of PUSH2 POP instructions will go from |
Contributor
|
It's pretty unfortunate that we have to resort to workarounds like this because the Go compiler is too restrictive to allow disabling the inliner for certain functions.. |
fjl
approved these changes
Apr 8, 2025
9 tasks
sivaratrisrinivas
pushed a commit
to sivaratrisrinivas/go-ethereum
that referenced
this pull request
Apr 21, 2025
During my benchmarks on Holesky, around 10% of all CPU time was spent in
PUSH2
```
ROUTINE ======================== github.com/ethereum/go-ethereum/core/vm.newFrontierInstructionSet.makePush.func1 in github.com/ethereum/go-ethereum/core/vm/instructions.go
16.38s 20.35s (flat, cum) 10.31% of Total
740ms 740ms 976: return func(pc *uint64, interpreter *EVMInterpreter, scope *ScopeContext) ([]byte, error) {
. . 977: var (
40ms 40ms 978: codeLen = len(scope.Contract.Code)
970ms 970ms 979: start = min(codeLen, int(*pc+1))
200ms 200ms 980: end = min(codeLen, start+pushByteSize)
. . 981: )
670ms 2.39s 982: a := new(uint256.Int).SetBytes(scope.Contract.Code[start:end])
. . 983:
. . 984: // Missing bytes: pushByteSize - len(pushData)
410ms 410ms 985: if missing := pushByteSize - (end - start); missing > 0 {
. . 986: a.Lsh(a, uint(8*missing))
. . 987: }
12.69s 14.94s 988: scope.Stack.push2(*a)
10ms 10ms 989: *pc += size
650ms 650ms 990: return nil, nil
. . 991: }
. . 992:}
```
Which is quite crazy. We have a handwritten encoder for PUSH1 already,
this PR adds one for PUSH2.
PUSH2 is the second most used opcode as shown here:
https://gist.github.com/shemnon/fb9b292a103abb02d98d64df6fbd35c8 since
it is used by solidity quite significantly. Its used ~20 times as much
as PUSH20 and PUSH32.
# Benchmarks
```
BenchmarkPush/makePush-14 94196547 12.27 ns/op 0 B/op 0 allocs/op
BenchmarkPush/push-14 429976924 2.829 ns/op 0 B/op 0 allocs/op
```
---------
Co-authored-by: jwasinger <[email protected]>
sduchesneau
pushed a commit
to streamingfast/go-ethereum
that referenced
this pull request
May 22, 2025
During my benchmarks on Holesky, around 10% of all CPU time was spent in
PUSH2
```
ROUTINE ======================== github.com/ethereum/go-ethereum/core/vm.newFrontierInstructionSet.makePush.func1 in github.com/ethereum/go-ethereum/core/vm/instructions.go
16.38s 20.35s (flat, cum) 10.31% of Total
740ms 740ms 976: return func(pc *uint64, interpreter *EVMInterpreter, scope *ScopeContext) ([]byte, error) {
. . 977: var (
40ms 40ms 978: codeLen = len(scope.Contract.Code)
970ms 970ms 979: start = min(codeLen, int(*pc+1))
200ms 200ms 980: end = min(codeLen, start+pushByteSize)
. . 981: )
670ms 2.39s 982: a := new(uint256.Int).SetBytes(scope.Contract.Code[start:end])
. . 983:
. . 984: // Missing bytes: pushByteSize - len(pushData)
410ms 410ms 985: if missing := pushByteSize - (end - start); missing > 0 {
. . 986: a.Lsh(a, uint(8*missing))
. . 987: }
12.69s 14.94s 988: scope.Stack.push2(*a)
10ms 10ms 989: *pc += size
650ms 650ms 990: return nil, nil
. . 991: }
. . 992:}
```
Which is quite crazy. We have a handwritten encoder for PUSH1 already,
this PR adds one for PUSH2.
PUSH2 is the second most used opcode as shown here:
https://gist.github.com/shemnon/fb9b292a103abb02d98d64df6fbd35c8 since
it is used by solidity quite significantly. Its used ~20 times as much
as PUSH20 and PUSH32.
# Benchmarks
```
BenchmarkPush/makePush-14 94196547 12.27 ns/op 0 B/op 0 allocs/op
BenchmarkPush/push-14 429976924 2.829 ns/op 0 B/op 0 allocs/op
```
---------
Co-authored-by: jwasinger <[email protected]>
jakub-freebit
pushed a commit
to fblch/go-ethereum
that referenced
this pull request
Jul 3, 2025
During my benchmarks on Holesky, around 10% of all CPU time was spent in
PUSH2
```
ROUTINE ======================== github.com/ethereum/go-ethereum/core/vm.newFrontierInstructionSet.makePush.func1 in github.com/ethereum/go-ethereum/core/vm/instructions.go
16.38s 20.35s (flat, cum) 10.31% of Total
740ms 740ms 976: return func(pc *uint64, interpreter *EVMInterpreter, scope *ScopeContext) ([]byte, error) {
. . 977: var (
40ms 40ms 978: codeLen = len(scope.Contract.Code)
970ms 970ms 979: start = min(codeLen, int(*pc+1))
200ms 200ms 980: end = min(codeLen, start+pushByteSize)
. . 981: )
670ms 2.39s 982: a := new(uint256.Int).SetBytes(scope.Contract.Code[start:end])
. . 983:
. . 984: // Missing bytes: pushByteSize - len(pushData)
410ms 410ms 985: if missing := pushByteSize - (end - start); missing > 0 {
. . 986: a.Lsh(a, uint(8*missing))
. . 987: }
12.69s 14.94s 988: scope.Stack.push2(*a)
10ms 10ms 989: *pc += size
650ms 650ms 990: return nil, nil
. . 991: }
. . 992:}
```
Which is quite crazy. We have a handwritten encoder for PUSH1 already,
this PR adds one for PUSH2.
PUSH2 is the second most used opcode as shown here:
https://gist.github.com/shemnon/fb9b292a103abb02d98d64df6fbd35c8 since
it is used by solidity quite significantly. Its used ~20 times as much
as PUSH20 and PUSH32.
# Benchmarks
```
BenchmarkPush/makePush-14 94196547 12.27 ns/op 0 B/op 0 allocs/op
BenchmarkPush/push-14 429976924 2.829 ns/op 0 B/op 0 allocs/op
```
---------
Co-authored-by: jwasinger <[email protected]>
howjmay
pushed a commit
to iotaledger/go-ethereum
that referenced
this pull request
Aug 27, 2025
During my benchmarks on Holesky, around 10% of all CPU time was spent in
PUSH2
```
ROUTINE ======================== github.com/ethereum/go-ethereum/core/vm.newFrontierInstructionSet.makePush.func1 in github.com/ethereum/go-ethereum/core/vm/instructions.go
16.38s 20.35s (flat, cum) 10.31% of Total
740ms 740ms 976: return func(pc *uint64, interpreter *EVMInterpreter, scope *ScopeContext) ([]byte, error) {
. . 977: var (
40ms 40ms 978: codeLen = len(scope.Contract.Code)
970ms 970ms 979: start = min(codeLen, int(*pc+1))
200ms 200ms 980: end = min(codeLen, start+pushByteSize)
. . 981: )
670ms 2.39s 982: a := new(uint256.Int).SetBytes(scope.Contract.Code[start:end])
. . 983:
. . 984: // Missing bytes: pushByteSize - len(pushData)
410ms 410ms 985: if missing := pushByteSize - (end - start); missing > 0 {
. . 986: a.Lsh(a, uint(8*missing))
. . 987: }
12.69s 14.94s 988: scope.Stack.push2(*a)
10ms 10ms 989: *pc += size
650ms 650ms 990: return nil, nil
. . 991: }
. . 992:}
```
Which is quite crazy. We have a handwritten encoder for PUSH1 already,
this PR adds one for PUSH2.
PUSH2 is the second most used opcode as shown here:
https://gist.github.com/shemnon/fb9b292a103abb02d98d64df6fbd35c8 since
it is used by solidity quite significantly. Its used ~20 times as much
as PUSH20 and PUSH32.
# Benchmarks
```
BenchmarkPush/makePush-14 94196547 12.27 ns/op 0 B/op 0 allocs/op
BenchmarkPush/push-14 429976924 2.829 ns/op 0 B/op 0 allocs/op
```
---------
Co-authored-by: jwasinger <[email protected]>
gballet
pushed a commit
to gballet/go-ethereum
that referenced
this pull request
Sep 11, 2025
During my benchmarks on Holesky, around 10% of all CPU time was spent in
PUSH2
```
ROUTINE ======================== github.com/ethereum/go-ethereum/core/vm.newFrontierInstructionSet.makePush.func1 in github.com/ethereum/go-ethereum/core/vm/instructions.go
16.38s 20.35s (flat, cum) 10.31% of Total
740ms 740ms 976: return func(pc *uint64, interpreter *EVMInterpreter, scope *ScopeContext) ([]byte, error) {
. . 977: var (
40ms 40ms 978: codeLen = len(scope.Contract.Code)
970ms 970ms 979: start = min(codeLen, int(*pc+1))
200ms 200ms 980: end = min(codeLen, start+pushByteSize)
. . 981: )
670ms 2.39s 982: a := new(uint256.Int).SetBytes(scope.Contract.Code[start:end])
. . 983:
. . 984: // Missing bytes: pushByteSize - len(pushData)
410ms 410ms 985: if missing := pushByteSize - (end - start); missing > 0 {
. . 986: a.Lsh(a, uint(8*missing))
. . 987: }
12.69s 14.94s 988: scope.Stack.push2(*a)
10ms 10ms 989: *pc += size
650ms 650ms 990: return nil, nil
. . 991: }
. . 992:}
```
Which is quite crazy. We have a handwritten encoder for PUSH1 already,
this PR adds one for PUSH2.
PUSH2 is the second most used opcode as shown here:
https://gist.github.com/shemnon/fb9b292a103abb02d98d64df6fbd35c8 since
it is used by solidity quite significantly. Its used ~20 times as much
as PUSH20 and PUSH32.
# Benchmarks
```
BenchmarkPush/makePush-14 94196547 12.27 ns/op 0 B/op 0 allocs/op
BenchmarkPush/push-14 429976924 2.829 ns/op 0 B/op 0 allocs/op
```
---------
Co-authored-by: jwasinger <[email protected]>
yperbasis
pushed a commit
to erigontech/erigon
that referenced
this pull request
Dec 22, 2025
- Optimize `opPush2` Refrence: [Geth #31267](ethereum/go-ethereum#31267) - make types consistent in `makeDup` ``` goos: linux goarch: amd64 pkg: github.com/erigontech/erigon/execution/vm cpu: Intel(R) Core(TM) i7-8565U CPU @ 1.80GHz │ new_bench.txt │ old_bench.txt │ │ sec/op │ sec/op vs base │ Push/makePush-8 13.23n ± 6% 18.72n ± 12% +41.53% (p=0.000 n=10) Push/push-8 8.531n ± 4% 11.890n ± 44% +39.37% (p=0.000 n=10) geomean 10.62n 14.92n +40.45% │ new_bench.txt │ old_bench.txt │ │ B/op │ B/op vs base │ Push/makePush-8 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=10) ¹ Push/push-8 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=10) ¹ geomean ² +0.00% ² ¹ all samples are equal ² summaries must be >0 to compute geomean │ new_bench.txt │ old_bench.txt │ │ allocs/op │ allocs/op vs base │ Push/makePush-8 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=10) ¹ Push/push-8 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=10) ¹ geomean ² +0.00% ² ¹ all samples are equal ² summaries must be >0 to compute geomean ``` Benchmark Test ``` func BenchmarkPush(b *testing.B) { var ( code = common.FromHex("0011223344556677889900aabbccddeeff0102030405060708090a0b0c0d0e0ff1e1d1c1b1a19181716151413121") push2 = makePush(2, 2) callContext = &CallContext{Contract: Contract{ Code: code, }} pc = new(uint64) ) b.Run("makePush", func(b *testing.B) { for i := 0; i < b.N; i++ { push2(*pc, nil, callContext) callContext.Stack.pop() } }) b.Run("push", func(b *testing.B) { for i := 0; i < b.N; i++ { opPush2(*pc, nil, callContext) callContext.Stack.pop() } }) } ```
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
6 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
During my benchmarks on Holesky, around 10% of all CPU time was spent in PUSH2
Which is quite crazy. We have a handwritten encoder for PUSH1 already, this PR adds one for PUSH2.
PUSH2 is the second most used opcode as shown here: https://gist.github.com/shemnon/fb9b292a103abb02d98d64df6fbd35c8 since it is used by solidity quite significantly. Its used ~20 times as much as PUSH20 and PUSH32.
Benchmarks
This PR also adds a fuzzer to sanity check the handwritten opcode