Search
Saturday, 21st of September 2019, 13:58:24 UTC
14:42:28
Shinmera
Sloooooowly crawling out under the mountain of broken shit I made
15:07:05
Shinmera
Alright, seems to work now.
15:07:41
Shinmera
Still got ugly stuff that I need to work out, like style overrides vanishing if a component is realised twice
15:08:17
|3b|
ACTION is trying to decide if it is worth calling compiler at runtime to save 2 cycles :p
15:08:27
Shinmera
https://filebox.tymoon.eu//file/TVRjNU9BPT0=
15:08:27
Colleen
filebox.tymoon.eu//file/TVR... Image (PNG)
15:09:16
Shinmera
Do you need to call it just once to memoize?
15:09:56
|3b|
once per attribute per vertex format theoretically
15:10:45
|3b|
hmm, maybe i have another alternative to test
15:12:56
|3b|
nope that's even slower :p
15:14:13
|3b|
ACTION is building functions to write vertex attributes into a buffer, and compiling fixed offsets is a bit faster than adding to a closed-over offset
15:14:38
Shinmera
Ah, I've been musing about that kind of thing myself
15:14:59
|3b|
for extra fun, safety 0 is slower than safety 1 half the time in my tests :/
15:15:32
|3b|
which probably means my measurements are entirely within the noise of changes of compiler versions
15:15:35
Shinmera
50/50 means no difference
15:16:08
|3b|
no, i mean consistently slower by a good bit for about half the configurations i'm testing
15:16:37
|3b|
(and very vague 'half' estimate)
15:18:09
|3b|
ACTION supposes if i'm going to call that an advantage of CL, i should probably just use it :)
15:18:42
|3b|
(though probably keep other version around in case it ends up a problem in practice)
15:42:53
|3b|
ah, maybe just needed some more type declarations on the other version, times are closer now
19:58:24
Shinmera
ACTION does the "everything I make is broken" dance
Sunday, 22nd of September 2019, 1:58:24 UTC