jstedfast / MimeKit

95%

master: 94%

LAST BUILD ON BRANCH mimereader-span2

branch: mimereader-span2

CHANGE BRANCH

x

Reset

• mimereader-span2
• HEAD
• binary-quoted-printable-decoder
• bouncycastle2.0
• core
• crypto-abstractions
• dependabot/nuget/System.Memory-4.5.5
• dependabot/nuget/nunit-4.0.0
• hwaccel-base64
• issue682
• master
• mime-compliance
• mime-compliance-violations
• mime-compliance2
• mimereader
• mimereader-span
• nullable-api
• nullable-api2
• parser-options-create
• refs/pull/1072/merge
• refs/pull/1162/merge
• refs/pull/1163/merge
• refs/pull/1176/merge
• refs/pull/1205/merge
• refs/tags/2.12.0
• refs/tags/3.3.0
• refs/tags/3.4.1
• refs/tags/3.4.3
• rfc2047-perf
• rfc2047-tokens
• smime-revocation-checks
• stream-spans
• unix2dos-indexof
• unsafe-scancontent
• vnext

Build # 4.12.0.1624

Build Type

push

coveralls.net

Committed by jstedfast

Commit Message

Update MimeReader to use Span<T> for built-in intrinsics

It's gotten to a point where .NET Core's built-in SIMD optimizations
in Span<T>.IndexOf() have gotten as good or better than my current
optimization for scanning for a '\n' in ScanContent().

It also allows me to now use Span<T>.IndexOf() in all other locations
that were using `while (*inptr != '\n') inptr++`, so we get improved
performance in those locations for free.

As an added benefit, the code is cleaner when using Span<T> instead of
pointers. The Async code, especially, is a lot cleaner not having to
used unsafe/fixed statements everywhere.

The downside is that Span<T>.IndexOf() is *slower* than my current
optimizations on .NET Framework and possibly also even .NET Core where
the platform architecture does not have support for SIMD.

That's why this PR is a Work-In-Progress and not yet merged to master.

Run Details

32059 of 33812 relevant lines covered (94.82%)

0.95 hits per line