Webp: Use byte arrays instead of Dictionary's for lookups #1800
Conversation
Codecov Report
@@ Coverage Diff @@
## master #1800 +/- ##
==========================================
+ Coverage 87.05% 87.30% +0.25%
==========================================
Files 936 936
Lines 47638 47832 +194
Branches 6017 6009 -8
==========================================
+ Hits 41469 41761 +292
+ Misses 5177 5082 -95
+ Partials 992 989 -3
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
brianpopow
changed the title
| p[offset - step3] = WebpLookupTables.Clip1[p2 + a3 + 255]; | ||
| p[offset - step2] = WebpLookupTables.Clip1[p1 + a2 + 255]; | ||
| p[offsetMinusStep] = WebpLookupTables.Clip1[p0 + a1 + 255]; | ||
| p[offset] = WebpLookupTables.Clip1[q0 - a1 + 255]; | ||
| p[offset + step] = WebpLookupTables.Clip1[q1 - a2 + 255]; | ||
| p[offset + step2] = WebpLookupTables.Clip1[q2 - a3 + 255]; |
There was a problem hiding this comment.
Bit late to the party but:
Do we always shift with the same value for the same lookup? Shouldn't we define a helper method then or at least add an explanatory comment?
There was a problem hiding this comment.
Bit late to the party but
Maybe I was a bit to quick in merging it.
Do we always shift with the same value for the same lookup?
yes, in C it is defined like this:
const uint8_t* const VP8kclip1 = &clip1[255];
See dec_clip_tables.c in the libwebp source.
I thought there might be a way to do the same in C#, but i could not come up with a better solution then this.
There was a problem hiding this comment.
I will create a follow up PR with helper methods
There was a problem hiding this comment.
Maybe you could use a Span that starts at the 255 offset?
Prerequisites
Description
During profiling I noticed alot of time with lossy decoding is spend in System.Collections. Turns out using Dictionary's for lookup tables was not the best idea. I have changed that now to byte array's. it reduces the lossy decoding time from 645.7 ms to 280.8 ms.
Before:
After: