Improve performance of shader lighting code in Forward renderers. #99136
Conversation
DarioSamo
force-pushed
the
light-compute-attenuation-skip
branch
from
e480716 to
831f763
Compare
- Skip sampling shadows if attenuation is very small. - Skip computing diffuse and specular light if attenuation and shadow are very small.
DarioSamo
force-pushed
the
light-compute-attenuation-skip
branch
from
831f763 to
0ce4c6d
Compare
|
Marking as ready for review, benchmarking and testing for regressions is welcome. Differences in projects with only directional lights are unlikely to show any difference whatsoever. |
|
Finally tested on an older device (Pixel 4 Adreno 640). This PR roughly reduces the frame time from 285 ms to about 260 ms which lines up with the other mobile devices you tested on. |
|
Thanks! |
| #endif | ||
| float Dr = D_GGX(ccNdotH, mix(0.001, 0.1, clearcoat_roughness)); | ||
| float Gr = 0.25 / (cLdotH * cLdotH); | ||
| float Fr = mix(.04, 1.0, cLdotH5); |
There was a problem hiding this comment.
Pretty sure this is the source of the errors I'm getting on a fresh compile. Opening certain scenes straight up crashes the editor.
servers\rendering\renderer_rd\shader_rd.cpp:336 - Error compiling Fragment shader, variant #16 ().
servers\rendering\renderer_rd\shader_rd.cpp:337 - Failed parse:
WARNING: 0:60: '' : all default precisions are highp; use precision statements to quiet warning, e.g.:
"precision mediump int; precision highp float;"
ERROR: 0:1086: 'cLdotH5' : undeclared identifier
ERROR: 0:1086: '' : missing #endif
ERROR: 0:1086: '' : compilation terminated
ERROR: 3 compilation errors. No code generated.
cLdotH5 is only defined when !defined(SPECULAR_SCHLICK_GGX) is true. Should this section be moved into the same conditional block, or should cLdotH5 be defined outside of it?
There was a problem hiding this comment.
We should remove the !defined(SPECULAR_SCHLICK_GGX) check in the clearcoat section. and then in the specular GGX section add a !defined(LIGHT_CLEARCOAT_USED) and only define cLdotH5 when clearcoat hasn't already defined it
Using the TPS demo from this PR with the Mobile renderer.
Approximate GPU times.
Ryzen 7950 Integrated Graphics
master: ~32.5 mslight-compute-attenuation-skip: ~27.0 msMali G-78 (Samsung S21)
master: ~27.4 mslight-compute-attenuation-skip: ~24.6 msMali G-715 (Google Pixel 8)
master: ~19.4 mslight-compute-attenuation-skip: ~17.3 msBehavior should be identical in theory although verification is welcome. Performance improvements are also expected in Forward+ but will be less significant as there's light clustering involved.
Opening as draft for now as I intend to do more research into improving some of the other features by reorganizing the code further.
Contributed by W4 Games. 🍀