Age | Commit message (Collapse) | Author | Files | Lines |
|
Replaces redundant code with one function. Oddly this fixes rendering
with earlyclip – not sure why. Drop transparency and channel settings
(always transparent, always four channels).
|
|
Internal color representation is [0,1] now for every channel.
|
|
Not much functionality lost here. Move on.
|
|
Introduced by 15f89a8390af9338227c2ee3f603b32a1a206a8f
|
|
Fixes segfault with -march=native
|
|
Also vectorizes some color functions and switches to double accumulation
buffer. Does not seem to be slower.
|
|
First of all, it does not look too pretty. But more importantly now we
can reduce the bucket size to four (instead of five) and vectorize that
stuff.
|
|
|
|
Finally drop all isaac references, do not use system rng any more. Drop
rng state from flam3_frame – it was not used anyway.
|
|
|
|
Atomic ops are slower than locks, probably going to switch buckets to
uint64_t later.
|
|
|
|
No difference in performance. rdrand instruction alone is too slow.
|
|
For some reason filter is overwritten by pthread_join, removing its
second argument fixes the problem. It is unused anyway.
|
|
|