[Mesa-dev] [PATCH v5 5/7] glsl: Add precision information to ir_variable

[Ilia Mirkin]

On Mon, Nov 16, 2015 at 11:42 AM, Samuel Iglesias Gonsálvez
<siglesias at igalia.com> wrote:
>
>
> On 16/11/15 17:34, Ilia Mirkin wrote:
>> On Mon, Nov 16, 2015 at 11:29 AM, Samuel Iglesias Gonsálvez
>> <siglesias at igalia.com> wrote:
>>>
>>>
>>> On 16/11/15 13:07, Tapani Pälli wrote:
>>>>
>>>> On 11/16/2015 01:35 PM, Tapani Pälli wrote:
>>>>>
>>>>>
>>>>> On 11/16/2015 01:29 PM, Samuel Iglesias Gonsálvez wrote:
>>>>>> Hello Ilia, Tapani:
>>>>>>
>>>>>> I have reproduced the issue with a piglit test but not with the trace
>>>>>> uploaded in the bug report :-(
>>>>>>
>>>>>> The piglit test was: bin/arb_shader_storage_buffer_object-maxblocks
>>>>>>
>>>>>> I have upload a branch with some fixes at Igalia's mesa repo:
>>>>>>
>>>>>> Git repo: https://github.com/Igalia/mesa.git
>>>>>> Branch: wip/siglesias/precision-fixes
>>>>>>
>>>>>> But as this error might come from other initializations that I might
>>>>>> overlook:
>>>>>> * Ilia: Could you test if this issue is still happening to you? As I
>>>>>> cannot reproduce it locally, I might be forgetting something.
>>>>>> * Tapani: Could you do a quick run on CTS to check I have not broken
>>>>>> anything?
>>>>>
>>>>> Sure thing, I'll run testing. FWIW one of the patches was identical to
>>>>> my fix sent for fixing tessellation shader problems:
>>>>>
>>>>> http://lists.freedesktop.org/archives/mesa-dev/2015-November/100396.html
>>>>
>>>> No CTS regressions with these patches, I've gone through these and
>>>> changes look good to me!
>>>>
>>>>
>>>
>>> OK, once Ilia replies that the issue is fixed with those patches, I will
>>> send them for review to the mailing list :-)
>>
>> I won't have time to look until tonight. However the repro steps were
>> pretty simple... download the trace and run through valgrind. Probably
>> tons of other ways to trigger it too, of course... I'd esp look for
>> piglits that have uniform structs.
>>
>
> The problem is that I could not reproduce it with the trace. That's why
> I am asking.
>
> I reproduce it with a piglit tests, but maybe precision is uninitialized
> in other cases. Tomorrow I will do some more testing, just in case.

Well, irrespective of other cases, if the things you're fixing are
real fixes, no need to wait on me. I'll be sure to complain again if I
still see problems. FWIW I did see them with nouveau, not i965. I
suspect llvmpipe would take the same paths.

  -ilia