[Mesa-dev] [PATCH 1/1] automake: r600 should only depend on libamd_common if opencl is enabled

[Emil Velikov]

On 5 June 2017 at 12:00, Jan Vesely <jan.vesely at rutgers.edu> wrote:
> On Mon, 2017-06-05 at 01:39 +0100, Emil Velikov wrote:
>> On 2 June 2017 at 17:34, Jan Vesely <jan.vesely at rutgers.edu> wrote:
>> > On Fri, 2017-06-02 at 17:22 +0100, Emil Velikov wrote:
>> > > On 2 June 2017 at 16:34, Jan Vesely <jan.vesely at rutgers.edu> wrote:
>> > > > On Fri, 2017-06-02 at 12:19 +0100, Emil Velikov wrote:
>> > > > > On 1 June 2017 at 21:28, Jan Vesely <jan.vesely at rutgers.edu> wrote:
>> > > > > > Signed-off-by: Jan Vesely <jan.vesely at rutgers.edu>
>> > > > > > ---
>> > > > > > Hi guys,
>> > > > > >
>> > > > > > this is the first step towards dropping libamd_common dependency.
>> > > > > > It's based on Emil's patches 3/5 and 4/5.
>> > > > > > Enabling opencl still falls back to the old way of requiring libamd_common.
>> > > > > > I'll try to address that in the next step (no time estimate, feel
>> > > > > > free to beat me to it). I think we can drop part of those functions
>> > > > > > rather than just copying them.
>> > > > > >
>> > > > >
>> > > > > AFAICT one still need the rest of my series, correct?
>> > > >
>> > > > kind of, 1/5 is mostly unrelated. 2,3/5 should be replaced by this one.
>> > > > 4/5 is needed,
>> > > > and 5/5 seems to be not applicable since ac_gpu_info.c
>> > > > still needs the header.
>> > > >
>> > >
>> > > Hmm indeed. I seems to have misread your patch.
>> > > Which makes me wonder if you've tested the patch as I mentioned earlier:
>> > >  - (re)move amdgpu.h
>> > >  - apply mesa patches (be that any of my, your and other mix)
>> > >  - build r300 and/or r600, w/o radeonsi
>> > >  - the the above combo a try with and w/o opencl
>> > >
>> > > I'm leaning that things will fail to build?
>> >
>> > This patch only removes dependence on libamd_common unless you enable
>> > OpenCL.
>>
>> Right - this thing here needs to be fixed. Rather than
>> pressuring/annoying you to "fix it now", I'll kindly suggest opting
>> for the neutering series as an intermediate step. We already had five
>> (iirc) separate reports about this :-(
>
> I'm once again puzzled by this. the posted patches remove dependence
> and fix the problem for the common (no OpenCL) configuration. I don't
> see how 'removed r600g+clover' is better than 'r600g+clover needs
> libdrm_amdgpu'.
>
Former builds, admittedly slightly neutered, while the latter does not.
Having software fail to build for a day or two is ~OK, but approx. 2
weeks is less so.

Another option is be error out at configure time. Either one is fine,
as long as you get a descriptive message why things are as-is.

-Emil