[Mesa-dev] [PATCH 0/3] [RFC] mesa/st: glsl_to_tgsi: improved temp-reg lifetime estimation

[Nicolai Hähnle]

On 12.06.2017 11:32, Gert Wollny wrote:
> Am Montag, den 12.06.2017, 15:44 +0900 schrieb Michel Dänzer:
>> On 10/06/17 08:15 AM, Gert Wollny wrote:
>>>
>>> Piglit shows 7 fixes and 6 regressions compared to git 8fac894f,
>>> but they don't
>>> seem to be related to shaders.
>>
>> Which tests regressed (maybe you can put up a piglit HTML summary
>> somewhere generated from a run with and without your patches)? Do
>> they consistently pass without your patches and fail with them?
> 
> 
> I had to redo the results, because I realized that I had compared the
> system mesa version (with EGL support) versus the test version (without
>    EGL support).
> 
> Now both tested versions were configure with the same options and I run
> both versions two times. The result diff of the quick test are:
> 
> piglit summary console -d results/o1 results/o2 results/n1 results/n2
> 
> glx/glx-multithread-texture: pass pass fail fail
> 
> glx/glx-visuals-stencil: fail fail pass pass
> 
> glx/glx_arb_sync_control/timing -fullscreen -divisor 1: pass fail pass
> fail
> 
> glx/glx_arb_sync_control/timing -fullscreen -divisor 2: pass fail fail
> warn
> 
> glx/glx_arb_sync_control/timing -fullscreen -msc-delta 1: fail fail
> warn fail
> 
> glx/glx_arb_sync_control/timing -fullscreen -msc-delta 2: fail fail
> fail pass
> 
> glx/glx_arb_sync_control/timing -msc-delta 2: warn fail pass fail
> 
> glx/glx_arb_sync_control/timing -waitformsc -msc-delta 2: fail pass
> pass fail

The above are probably noise.


> spec/arb_shader_bit_encoding/execution/and-clamp: fail fail pass fail
> 
> spec/arb_shading_language_420pack/active sampler conflict: fail fail
> pass fail
> 
> spec/glsl-1.50/execution/variable-indexing/gs-input-array-vec2-index-
> rd: fail fail pass pass
> 
> spec/nv_conditional_render/drawpixels: fail pass fail pass
> 
> spec/nv_conditional_render/vertex_array: fail pass pass pass 

It's disconcerting that you have tests here whose pass status is 
unstable. Those tests really should be deterministic.

You should definitely investigate 
spec/arb_shader_bit_encoding/execution/and-clamp to see if it's related 
to your patches.

Cheers,
Nicolai


> 
> 
> summary:
>         name:     o1     o2     n1     n2
>         ----  ------ ------ ------ ------
>         pass:  31583  31584  31588  31585
>         fail:   1454   1454   1449   1452
>        crash:      5      5      5      5
>         skip:  17356  17356  17356  17356
>      timeout:      0      0      0      0
>         warn:     14     13     14     14
>   incomplete:      0      0      0      0
>   dmesg-warn:      0      0      0      0
>   dmesg-fail:      0      0      0      0
>      changes:      0      6      9      9
>        fixes:      0      3      7      3
> regressions:      0      3      2      6
>        total:  50412  50412  50412  50412
> 
> 
> Best,
> Gert
> 
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
> 


-- 
Lerne, wie die Welt wirklich ist,
Aber vergiss niemals, wie sie sein sollte.