[Mesa-dev] [PATCH 4/6] configure.ac: Set and use HAVE_GALLIUM_LLVM define

[Jose Fonseca]

On 20/01/17 02:48, Emil Velikov wrote:
> On 19 January 2017 at 19:26, Tobias Droste <tdroste at gmx.de> wrote:
>> Am Mittwoch, 18. Januar 2017, 18:45:04 CET schrieb Emil Velikov:
>>> On 18 January 2017 at 18:12, Jose Fonseca <jfonseca at vmware.com> wrote:
>>>>>> In order to untangle things we want to have a distinction between the
>>>>>> gallium (gallivm afaict) and other users - RADV presently.
>>>>>> So how about we update the RADV instances and ensure that the we set
>>>>>> the HAVE_{RADV,}_LLVM lot appropriately. Latter will be picky but
>>>>>> overall things should work w/o annoyances that HAVE_GALLIUM_LLVM
>>>>>> brings ?
>>>>
>>>> I honestly don't even understand why we'd want to build parts of the tree
>>>> with LLVM while hiding LLVM from other components.  We can't we just build
>>>> everything with LLVM and avoid this combinatorial explosion of wierd
>>>> options that are nothing more than yet another way the build can break!!?
>>>
>>> Sadly the combinatoric explosion has been there for a while. Based on
>>> how well my previous attempts to resolve similar issues (see the
>>> "platforms" topic) I doubt we'll even get to fix that.
>>>
>>>> But if a separate option is truly necessary, have the newcomer pick a
>>>> different name, or something.
>>>
>>> That's pretty much what I suggested above. Tobias can you please give it a
>>> try ?
>>
>> I would rather "fix" the other build systems. (As in just define
>> HAVE_GALLIUM_LLVM if HAVE_LLVM is defined).
>>
>> I think there is still a misunderstanding on Joses side on what this really
>> means. No file in gallivm or llvmpipe will be touched. It's really just
>> auxilliary/draw and there it's exactly 8 lines that will change.
>>
>> That's it.
>>
>> I really fail to see how this will break everything that is being worked on
>> and cause merge conflicts everywhere.
>>
>> If you still want the other way, I can do that to, but this will of course
>> need the same fix in the other build system or we have the same situation we
>> have now, but with other drivers.
>>
> Afaict one point is that the use of HAVE_GALLIUM_LLVM vs HAVE_LLVM is
> too subtle. Let's not forget that barring the WIP(?) branches, VMWare
> has closed source components. Guess how much fun it will be as
> suddenly things fail to build/work properly as they re-sync the code
> base. No idea how likely the latter is, but considering Jose (and a
> few other VMWare guys) wrote sizeable hunk of that code (and Mesa as a
> whole) I'd go with his instinct.
>
> Emil
>


The HAVE_LLVM->HAVE_GALLIUM_LLVM rename is indeed not as invasive as I 
thought.

But I still don't understand why HAVE_LLVM->HAVE_GALLIUM_LLVM is 
necessary in draw and not on gallivm/llvmpipe.

People want to build draw with LLVM support, but without 
gallivm/llvmpipe? That's impossible.

Or is this because the draw files are the only .c files that are 
compiled even when HAVE_LLVM is undefined, so these are the only ones 
that get to receive the renaming treatment?  That's crazy confusing. 
There's no away I can accept that.


Let me make this crystal clear to avoid making this discussion even more 
protracted: I will not accept any HAVE_LLVM change in 
draw/gallivm/llvmpipe .C/.H source code.  Period.


HAVE_LLVM used to mean, "whole Mesa being built with LLVM".  Now people 
want to build something (no idea what yet to be honest) with LLVM, but 
not build draw/gallivm/llvmpipe.

If you want to build some other component with LLVM but not 
draw/gallivm/llvmpipe, then add a new HAVE_LLVM_FOR_FOOBAR define and 
use it where you need it.


And honestly, I think people should reconsider enabling building parts 
with Mesa with LLVM and others without it.  It sounds an awful idea that 
will cause nothing more than grief.  But that's not for me to decide.


Jose