[MPlayer-dev-eng] [PATCH] configure to detect prescott and nocona

[Corey Hickey]

Zuxy wrote:
> The x86-64 case is more complicated than I thought. Therefore I've
> made a regression and only brought in prescott and pentium-m support
> under ia32. Sorry for the trouble.

It's ok. This should be dealt with, somehow. I've written up a patch
partly based on your two patches.

I've successfully tested this patch on the following:
x86_64:
  Athlon64           --> k8
x86:
  K6 3D              --> k6-2
  Athlon Thunderbird --> athlon-tbird
  Pentium 4m         --> pentium4

Some testing on prescott and nocona machines would be prudent.


Notes:

-----
I moved a part of the x86 section wherein configure sets pname, pvendor,
etc.. This is now its own section, and is run 'if (x86 || x86_64)'

-----
The gcc manpage indicates that "nocona" is like "prescott" but with
64-bit extensions. From what I've read off of google, some prescotts
actually do have 64-bit extensions, and the actual differences between
prescott and nocona are cache size and bus speed.
http://news.zdnet.com/2100-9584_22-5160169.html
Since the instruction sets are the same (except for 64-bit) I think it
should be safe to assign proc=nocona for intel chips wherever host_arch
is x86_64. Conversely, there isn't any reason to assign proc=nocona when
host_arch is x86, so we should use proc=prescott.

This was implicit in Zuxy's patches, and I've put relevant comments in mine.

-----
I've decided that the cleanest default case for x86_64 optimization is
to simply disable optimization by setting proc=error. The optimization
tests catch this and set march and mtune/mcpu to empty strings.

As Zuxy pointed out, it would be possible to set up "generic" x86_64
optimization by disabling extensions that either AMD or Intel processors
don't support. I don't think this is worth it, however, since the
default case is only there to catch a hypothetical future processor made
by some other manufacturer. In order to properly support that processor
when it manifests, configure will have to be adjusted anyway. There
isn't any reason to guess at this juncture.


-Corey
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: configure-new-intel.diff
URL: <http://lists.mplayerhq.hu/pipermail/mplayer-dev-eng/attachments/20051017/9f3ffd30/attachment.asc>