starred/sharpcompress
1
0
Fork 0
mirror of https://github.com/adamhathcock/sharpcompress.git synced 2026-02-03 21:23:38 +00:00
Code Issues 975 Packages Projects Releases 20 Wiki Activity

[PR #910] Optimize LZMA range decoder #1341

New Issue
Closed
opened 2026-01-29 22:20:04 +00:00 by claunia · 0 comments
claunia commented 2026-01-29 22:20:04 +00:00
Owner
Copy Link

Original Pull Request: https://github.com/adamhathcock/sharpcompress/pull/910

State: closed
Merged: Yes

I noticed this duplicated bit of logic inside the range decoder. Turns out, there's a helper function called Normalize2 that does the exact same thing and was never even called. This quirk came all the way from the original LZMA C# SDK it seems. I can see why, as with .NET Framework 4.8, attempting to use the helper function results in performance degradation, so someone manually inlined it. But that's not even necessary, as [MethodImpl(MethodImplOptions.AggressiveInlining)] can be used to achieve the same performance.

But the more interesting part, hence this PR, is that newer JITs (tested with .NET 8.0) don't like the manually inlined version of the code as much; when calling Normalize2 instead, it seems to get inlined either way (even without the attribute that we put as a hint for .NET 4.8), and the performance is better.

Core i7-6700k (3.2% reduction):

Method Mean Error StdDev
Before 1.128 s 0.0018 s 0.0015 s
After 1.092 s 0.0011 s 0.0009 s

Apple M3 (11.6% reduction):

Method Mean Error StdDev
Before 888.9 ms 16.78 ms 14.01 ms
After 785.8 ms 1.98 ms 1.85 ms

That's reduction in overall time to extract the whole archive. Tested with a smaller archive in BenchmarkDotNet and I was honestly in disbelief at first on the M3, but it's real. I also manually benchmarked using the same setup/archive as my first PR, it was a 412MB Qt 7z taking 26+ seconds, and it indeed shaved an entire 3 seconds off the extraction time.

**Original Pull Request:** https://github.com/adamhathcock/sharpcompress/pull/910 **State:** closed **Merged:** Yes --- I noticed this duplicated bit of logic inside the range decoder. Turns out, there's a helper function called `Normalize2` that does the exact same thing and was never even called. This quirk came all the way from the original LZMA C# SDK it seems. I can see why, as with .NET Framework 4.8, attempting to use the helper function results in performance degradation, so someone manually inlined it. But that's not even necessary, as `[MethodImpl(MethodImplOptions.AggressiveInlining)]` can be used to achieve the same performance. But the more interesting part, hence this PR, is that newer JITs (tested with .NET 8.0) don't like the manually inlined version of the code as much; when calling `Normalize2` instead, it seems to get inlined either way (even without the attribute that we put as a hint for .NET 4.8), and the performance is better. Core i7-6700k (3.2% reduction): | Method | Mean | Error | StdDev | |------- |--------:|---------:|---------:| | Before | 1.128 s | 0.0018 s | 0.0015 s | | After | 1.092 s | 0.0011 s | 0.0009 s | Apple M3 (11.6% reduction): | Method | Mean | Error | StdDev | |------- |---------:|---------:|---------:| | Before | 888.9 ms | 16.78 ms | 14.01 ms | | After | 785.8 ms | 1.98 ms | 1.85 ms | That's reduction in overall time to extract the whole archive. Tested with a smaller archive in BenchmarkDotNet and I was honestly in disbelief at first on the M3, but it's real. I also manually benchmarked using the same setup/archive as my first PR, it was a 412MB Qt 7z taking 26+ seconds, and it indeed shaved an entire 3 seconds off the extraction time.
claunia added the pull-request label 2026-01-29 22:20:04 +00:00
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
bug
duplicate
enhancement
feedback requested
not bug
pull-request
Mirrored from GitHub Pull Request
 question
up for grabs
wontfix
No Label pull-request
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
claunia
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: starred/sharpcompress#1341
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?