[Mesa-stable] [Mesa-dev] [PATCH] [RFC] gallivm: Use new LLVM fast-math-flags API

[Emil Velikov]

On 5 March 2018 at 15:13, Emil Velikov <emil.l.velikov at gmail.com> wrote:
> On 3 March 2018 at 15:40, Marek Olšák <maraeo at gmail.com> wrote:
>> On Fri, Mar 2, 2018 at 1:38 PM, Emil Velikov <emil.l.velikov at gmail.com> wrote:
>>> Hi Alex,
>>>
>>> On 28 February 2018 at 15:25, Alex Smith <asmith at feralinteractive.com> wrote:
>>>> Hi,
>>>>
>>>> Could this (commit 5d61fa4e68b7eb6d481a37efdbb35fdce675a6ad on master) be
>>>> backported to the 17.3 branch to allow it to build with LLVM 6?
>>>>
>>> Normally we don't aim to support LLVM versions released after the .0
>>> Mesa release is out.
>>> Not that we don't want to - there is simply not enough testing happening.
>>>
>>> Sometimes picking the odd build fix is enough, but not always.
>>>
>>> As a matter of fact, the only feedback for the AMD drivers status
>>> (brokenness) is the LunarG testing rig.
>>>
>>> Michel, usually you are usually more realistic/conservative on with
>>> this kind of changes.
>>
>> Are you saying that I'm less realistic? :)
>>
> You're right - my wording was bad. I should have only said conservative.
>
> I would love to see agreement within the AMD team - one way or another.
> If the decision is to go with these kind of changes, testing will also
> be appreciated.
>
> Be that independent individuals, teams, other. Let me ask if the
> Lunarg team can add LLVM version to the test matrix.
>
Have some good news - the Lunarg team will add LLVM 6.0 in the list.
So as soon as we get that + there's no glaring regressions I think
we'll be in decent shape.

For anyone wondering why I tend towards the conservative side:
 - making it build, hence having partial LLVM X support is not enough
The 'partial' word will be missed and you'll get plenty of unhappy
users as regressions happen

 - missing a wide/popular test base
The odd report of game X working fine is _greatly_ appreciated, yet
quite limited

 - test results are not [easily] accessible by many people
We want something to refer to as we decide to allow (or forbid) LLVM X

HTH
Emil