amd64 arch and sse optimization

[Reinhard Tartler]

On Mon, Feb 3, 2014 at 10:43 AM, Felipe Sateler <fsateler at debian.org> wrote:
> On Mon, Feb 3, 2014 at 12:36 PM, Reinhard Tartler <siretart at gmail.com> wrote:
>> On Mon, Feb 3, 2014 at 9:02 AM, Fabian Greffrath <fabian at greffrath.com> wrote:
>>> Am Montag, den 03.02.2014, 10:41 -0300 schrieb Felipe Sateler:
>>>> I have some packages that also add (on amd64 and i386) -mtune=generic.
>>>> This causes the compiler to optimize for the most common cpus, but
>>>> still be compatible with older ones.
>>>
>>> Isn't GCC smart enough to know that it can use SSE optimizations on
>>> amd64?
>>
>> I strongly believe it is, if GCC was not, then it should be fixed.
>
> Indeed, that should be the correct solution.
>
>>
>> Please be careful with adding those flags. They instruct gcc to emit
>> code that does not work on some machines. The admitedlyonly way to use
>> them correctly is to ensure that the emitted code is only used on
>> machines that actually supports that.
>
> The mtune flag instructs the compiler to optimize for a certain
> instruction set, but still provide a fallback for when the instrucions
> are not available. I don't know if this includes the use of SSE or
> other coprocessor.
>
> http://stackoverflow.com/questions/10559275/gcc-how-is-march-different-from-mtune

OK, then we will only see performance impact on machines not matching
the optimization target. Which seems OK to me, if the benefit for the
targeted architectures justify this. Sorry for the noise.

>> We've been burned with that previously. In Libav, we're installing the
>> libraries into special "magic" paths that the dynamic linker only
>> includes if the hardware is capable. This works only on libraries, but
>> not for programs (and there are still open bugs for that, for instance
>> with mplayer/mplayer2 and possibly mpv, all for powerpc, admittedly).
>>
>> If you are unsure on those flags, better not enable them.
>
> I think sse is required by the amd64 spec, so no point in not using it
> (wikipedia agrees).

In that case, there is no point in explicitly requesting it in each
package, because the baseline optimization flags should already catch
it.

-- 
regards,
    Reinhard