[Bug 109216] 4-27% performance drop in Vulkan benchmarks

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=109216

--- Comment #6 from Lionel Landwerlin <lionel.g.landwerlin at linux.intel.com> ---
(In reply to Lionel Landwerlin from comment #4)
> Bisected to :
> 
> commit 8fb8ebfbb05d3323481c8ba6d320b3a3580bad99
> Author: Ian Romanick <ian.d.romanick at intel.com>
> Date:   Tue May 22 18:56:41 2018 -0700
> 
>     intel/compiler: More peephole select
>     
>     Shader-db results:
>     
>     The one shader hurt for instructions is a compute shader that had both
>     spills and fills hurt.

This significantly impacts one a couple of the shaders of the multisampling
demo :

First one :
Before: SIMD16 shader: 356 instructions. 0 loops. 1812 cycles. 0:0
spills:fills. Promoted 2 constants. Compacted 5696 to 3488 bytes (39%)
After: SIMD16 shader: 157 instructions. 3 loops. 99970 cycles. 0:0
spills:fills. Promoted 2 constants. Compacted 2512 to 1536 bytes (39%)

Second one:
Before: SIMD16 shader: 249 instructions. 0 loops. 1170 cycles. 0:0
spills:fills. Promoted 0 constants. Compacted 3984 to 2560 bytes (36%)
After: SIMD16 shader: 102 instructions. 3 loops. 4580 cycles. 0:0 spills:fills.
Promoted 0 constants. Compacted 1632 to 1088 bytes (33%)

This is one of the shader :
https://github.com/SaschaWillems/Vulkan/blob/master/data/shaders/deferredmultisampling/deferred.frag

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20190107/778dbbca/attachment-0001.html>