2011/9/28 Maarten Bosmans <mkbosmans at gmail.com>:
> 2011/9/28 Arun Raghavan <arun.raghavan at collabora.co.uk>:
>> On Wed, 2011-09-28 at 09:50 +0200, Maarten Bosmans wrote:
>>> This patch was sent to the list before, but was deemend to riscy pre-1.0.
>>> So now I'll resend. It does touch de same code as David fixed with 42bcb418aed SSE/MMX: Fix problem with highpitched
>>> noise on i386, but I leaves his solution intact, so it should be safe.
>>>
>>> The problem this patch is trying to solve is that the (pa_reg_x86) is not meaningfull with the "m" modifier. GCC ignores
>>> this, but clang (correctly) whines about it.
>>
>> Do we have an easy way to make sure the problem David fixed continues to
>> not be there with this patch?
>
> There's no regression test for this. I have some tests here locally
> and will try it.
OK, so I have verified it, the patch is safe.
This is tested with a new test: svolume-test, which I added in order
to be able to verify the further orcification of the svolume functions
I did. Basically it is just the testing code that was already present
in the svolume_{arm,mmx,sse,orc}.c files beefed up a bit.:
https://github.com/mkbosmans/pulseaudio/commit/cf0c5c9ad47ba0434b0518ca79ca802d0e62153a
Running svolume-test with David's patch reversed exits with an error
status and both with current git master and with my patch on top of
that succeeds.
Maarten
