Skip to content

specialize hexadecimal sequence decoding#489

Merged
dtolnay merged 1 commit intoserde-rs:masterfrom
yjh0502:master
Sep 26, 2018
Merged

specialize hexadecimal sequence decoding#489
dtolnay merged 1 commit intoserde-rs:masterfrom
yjh0502:master

Conversation

@yjh0502
Copy link
Copy Markdown
Contributor

@yjh0502 yjh0502 commented Sep 26, 2018

serde_json is slower than expected when parsing long string filled with hexadecimal escape sequences. I discovered the problem while testing with following benchmark case:

https://raw.githubusercontent.com/devinus/poison/master/bench/data/utf-8-escaped.json

serde_json is two times slower than moderately optimized c++ implementation, and here's performance profile of serde_json for the testcase, captured with perf record.

# Overhead  Command      Shared Object       Symbol
# ........  ...........  ..................  ............................................................................
#
    25.81%  6_scheduler  librjs.so           [.] serde_json::read::parse_escape
    25.19%  6_scheduler  librjs.so           [.] serde_json::read::decode_hex_escape
    16.59%  6_scheduler  librjs.so           [.] serde_json::read::next_or_eof
    11.78%  6_scheduler  libc-2.27.so        [.] __memcpy_ssse3
    11.27%  6_scheduler  librjs.so           [.] <serde_json::read::StrRead<'a> as serde_json::read::Read<'a>>::parse_str
     4.46%  6_scheduler  librjs.so           [.] <alloc::raw_vec::RawVec<T, A>>::reserve
     3.08%  6_scheduler  librjs.so           [.] core::slice::<impl [T]>::copy_from_slice

This PR makes hex decoding faster with decode_hex_escape specialization, and it brings ~40% performance improvement for the case. Here's a benchmark result which shows number of iterations that decoding the testcase, total time, and time per each iteration.

# without the patch
iterations=10000, 1.558 sec, 155.820 usec/iter
# with the patch
iterations=10000, 1.067 sec, 106.709 usec/iter

Here's a profile after the PR

# Overhead  Command      Shared Object       Symbol
# ........  ...........  ..................  ..........................................................................................
#
    31.30%  4_scheduler  librjs.so           [.] serde_json::read::parse_escape
    20.64%  4_scheduler  libc-2.27.so        [.] __memcpy_ssse3
    16.11%  4_scheduler  librjs.so           [.] <serde_json::read::StrRead<'a> as serde_json::read::Read<'a>>::parse_str
    11.80%  4_scheduler  librjs.so           [.] serde_json::read::next_or_eof
     6.88%  4_scheduler  librjs.so           [.] <serde_json::read::SliceRead<'a> as serde_json::read::Read<'a>>::decode_hex_escape
     5.93%  4_scheduler  librjs.so           [.] <alloc::raw_vec::RawVec<T, A>>::reserve
     4.56%  4_scheduler  librjs.so           [.] core::slice::<impl [T]>::copy_from_slice

dtolnay
dtolnay approved these changes Sep 26, 2018
View reviewed changes
Copy link
Copy Markdown
Member

@dtolnay dtolnay left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks!

@dtolnay dtolnay merged commit 1ef3da9 into serde-rs:master Sep 26, 2018
@dtolnay
Copy link
Copy Markdown
Member

dtolnay commented Sep 26, 2018

I published 1.0.31 containing this fix.

takumi-earth pushed a commit to earthlings-dev/json that referenced this pull request Jan 27, 2026
@dtolnay
Merge pull request serde-rs#489 from yjh0502/master
41ebbdb 
specialize hexadecimal sequence decoding
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@dtolnay dtolnay dtolnay approved these changes

Assignees

No one assigned

Labels

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@yjh0502 @dtolnay