Optimize by pre allocate slices to avoid Runtime.Growslice
#15261
Conversation
terencechain
force-pushed
the
avoid-grow-slice
branch
from
13893c7 to
a663ac7
Compare
terencechain
force-pushed
the
avoid-grow-slice
branch
from
a663ac7 to
e1497c9
Compare
| @@ -265,7 +265,9 @@ func (s *Service) hasSeenAggregatorIndexEpoch(epoch primitives.Epoch, aggregator | |||
| func (s *Service) setAggregatorIndexEpochSeen(epoch primitives.Epoch, aggregatorIndex primitives.ValidatorIndex) { | |||
| s.seenAggregatedAttestationLock.Lock() | |||
| defer s.seenAggregatedAttestationLock.Unlock() | |||
| b := append(bytesutil.Bytes32(uint64(epoch)), bytesutil.Bytes32(uint64(aggregatorIndex))...) | |||
| b := make([]byte, 64) | |||
| copy(b[0:32], bytesutil.Bytes32(uint64(epoch))) | |||
There was a problem hiding this comment.
I assume this code was previously calling Bytes32 for padding, but since you are already pre-allocating the slice, couldn't you just call copy directly (knowing that the pre-allocated bytes are already zero)? That should save you a couple more allocations right?
terencechain
force-pushed
the
avoid-grow-slice
branch
from
0e87da0 to
e3217b9
Compare
terencechain
force-pushed
the
avoid-grow-slice
branch
from
e3217b9 to
e9807b2
Compare
|
|
||
| b := make([]byte, 64) | ||
| binary.BigEndian.PutUint64(b[24:], uint64(epoch)) | ||
| binary.BigEndian.PutUint64(b[56:], uint64(aggregatorIndex)) |
There was a problem hiding this comment.
These slices offsets look wrong, and it seems like this is changing the encoding from LittleEndian to BigEndian? I think it should be:
b := make([]byte, 64)
binary.LittleEndian.PutUint64(b, uint64(epoch))
binary.LittleEndian.PutUint64(b[32:], uint64(aggregatorIndex))
crypto/bls/signature_batch.go
Outdated
|
|
||
| total := len(s.Signatures) + len(set.Signatures) | ||
|
|
||
| if cap(s.Signatures) < total { |
There was a problem hiding this comment.
re this updated version - to get the amortization benefit you want to straight up append, which will let the go runtime do its default behavior of allocating cap()*2. So if you want to go this way, you can just get rid of lines 40-61.
I've fixed the top 5 offenders given the following profile.
(s *SignatureBatch) Joinis the worst one.