This is an archive of the discontinued Mercurial Phabricator instance.

Differential D31

revlog: micro-optimize the computation of hashes
ClosedPublic

Authored by alex_gaynor on Jul 10 2017, 4:59 PM.

Details

Reviewers
durin42
Commits
rHG943b8c37f49d: revlog: micro-optimize the computation of hashes
Summary

Diff Detail

Repository
rHG Mercurial
Lint
Lint Skipped
Unit
Unit Tests Skipped

Event Timeline

alex_gaynor created this revision.Jul 10 2017, 4:59 PM
durin42 accepted this revision.Jul 11 2017, 11:21 AM

I'm grumpy at Python for making stupid things like this measurable. Queued.

This revision is now accepted and ready to land.Jul 11 2017, 11:21 AM
dsp added a subscriber: dsp.Jul 11 2017, 11:24 AM

LGTM.

Do we have any data if this speeds up hash computation? The patch seems small enough to justify either way.

alex_gaynor added a comment.Jul 11 2017, 11:25 AM

I forget which workload it was (might have been hg up <bookmark>), I measured this stack as being a 7% improvement.

durin42 added a comment.Jul 11 2017, 11:30 AM

I've been trying to come up with a perf* command that shows off the win here, but at the end of the day it's producing fewer tuples, and that'll be a win.

Sigh.

alex_gaynor added a comment.Jul 11 2017, 11:32 AM

Do you mean lists? The original version of this code doesn't allocate any tuples.

durin42 added a comment.Jul 11 2017, 11:32 AM

Yes, listen to what I mean, not what I say. ;)

quark retitled this revision from revlog: Micro-optimize the computation of hashes to revlog: micro-optimize the computation of hashes.Jul 11 2017, 12:14 PM
quark added a subscriber: quark.

I have un-capitalized the title so it looks more consistent with existing commits.

Closed by commit rHG943b8c37f49d: revlog: micro-optimize the computation of hashes (authored by alex_gaynor). · Explain WhyJul 14 2017, 12:33 PM
This revision was automatically updated to reflect the committed changes.

Revision Contents
Changeset List

PathPackages
Mmercurial/revlog.py (12 lines)

Diff 47

mercurial/revlog.py

""" """
# As of now, if one of the parent node is null, p2 is null # As of now, if one of the parent node is null, p2 is null
if p2 == nullid: if p2 == nullid:
# deep copy of a hash is faster than creating one # deep copy of a hash is faster than creating one
s = _nullhash.copy() s = _nullhash.copy()
s.update(p1) s.update(p1)
else: else:
# none of the parent nodes are nullid # none of the parent nodes are nullid
l = [p1, p2] if p1 < p2:
l.sort() a = p1
s = hashlib.sha1(l[0]) b = p2
s.update(l[1]) else:
a = p2
b = p1
s = hashlib.sha1(a)
s.update(b)
s.update(text) s.update(text)
return s.digest() return s.digest()
# index v0: # index v0:
# 4 bytes: offset # 4 bytes: offset
# 4 bytes: compressed length # 4 bytes: compressed length
# 4 bytes: base rev # 4 bytes: base rev
# 4 bytes: link rev # 4 bytes: link rev