diff options
author | EricAnholt <EricAnholt@web> | 2015-10-23 10:15:50 -0700 |
---|---|---|
committer | dri <iki-dri@freedesktop.org> | 2015-10-23 10:15:50 -0700 |
commit | f7a10c9e3ff34493a931c7bba4678b3eed2122e1 (patch) | |
tree | 2fe8055fb8925fa6ab6a0d5b180d729752ba7d1a /VC4.mdwn | |
parent | dd485064209047de55983fb89f11f5d61afa33d2 (diff) |
couple of performance tasks done!
Diffstat (limited to 'VC4.mdwn')
-rw-r--r-- | VC4.mdwn | 12 |
1 files changed, 0 insertions, 12 deletions
@@ -57,18 +57,6 @@ If you use half the register space(ra0-15, rb0-15), you can do a hyperthreading- Especially if we order texture results collection after other texture setup, this could be a big win (which would be hard to achieve post-register-allocation, when there are more dependencies). See vc4-qir-sched. -* Improve the pairing instruction scheduler. - -We could pair some cases with the PM bit set in only one of the instructions. - -* Blend in ARGB8888 instead of vec4. - -This would take about 1/4 the instructions. However, we need to retain the float path because you want the improved precision for sRGB rendering. See vc4-blend for WIP. - -* Handle vertex attribute size overrides in NIR. - -emit_vertex_input() currently handles things like making the .w coordinate of a vertex attribute be 1 when only xyz are passed in, and QIR optimizations aren't doing as well on that as they could. Making a NIR pass to do this transformation would fix some silly math that occurs. - * Copy propagate VPM reads into their usage if only used once. * Reorder fragment shader varyings so we can interpolate directly into texture coordinate setup? |