Improve type stability in conversion methods #427
Conversation
| fy = (c.l + 16) / 116 | ||
| fx = c.a / 500 + fy | ||
| fx = fy + c.a / 500 | ||
| fz = fy - c.b / 200 |
There was a problem hiding this comment.
If we change these divisions into reciprocal multiplications, this method can be inlined, i.e. successfully vectorized. However, even if only Lab-->XYZ conversion is speeded up, Lab-->RGB conversion is not speeded up.
Codecov Report
@@ Coverage Diff @@
## master #427 +/- ##
==========================================
- Coverage 91.21% 91.08% -0.13%
==========================================
Files 9 9
Lines 956 965 +9
==========================================
+ Hits 872 879 +7
- Misses 84 86 +2
Continue to review full report at Codecov.
|
kimikage
changed the title
find_maximum_chroma
|
I separated part of this PR to #442. |
kimikage
force-pushed
the
stability
branch
2 times, most recently
from
518d4ea to
5f23315
Compare
This also optimizes conversion methods and changes the results slightly.
|
Now that we have added the support for ColorTypes v0.11, we will start developing Colors v0.13 series. This means that breaking changes can be merged. Also, Colors v0.13.0 is scheduled to be released at the same time as ColorTypes v0.12.0, so there will be some time before the release. In other words, even if bugs are introduced, that is not a problem as long as they are fixed before the release. |
This also optimizes conversion methods and changes the results slightly.
This also optimizes conversion methods and changes the results slightly.
I found some type instability issues while tackling JuliaLang/julia#35972 or #425. Unfortunately, this doesn't solve JuliaLang/julia#35972, and doesn't provide much performance improvement.
This is a successor of PR #424.(Merged. However, I'm still looking for a way to solve #425.)This should be the next v0.13 material since this changes the conversion results slightly.