Hi,

On Tue, Jul 31, 2012 at 12:25 PM, Luca Barbato <[email protected]> wrote:
> On 07/31/2012 08:33 PM, Ronald S. Bultje wrote:
>> Hi.
>>
>> On Tue, Jul 31, 2012 at 11:14 AM, Ronald S. Bultje <[email protected]> 
>> wrote:
>>> From: "Ronald S. Bultje" <[email protected]>
>>>
>>> ---
>>>  libavcodec/x86/Makefile    |    2 -
>>>  libavcodec/x86/fft_3dn.c   |   23 ------
>>>  libavcodec/x86/fft_3dn2.c  |  173 -----------------------------------------
>>>  libavcodec/x86/fft_mmx.asm |  185 
>>> +++++++++++++++++++++++++++++++++++++-------
>>>  4 files changed, 158 insertions(+), 225 deletions(-)
>>>  delete mode 100644 libavcodec/x86/fft_3dn.c
>>>  delete mode 100644 libavcodec/x86/fft_3dn2.c
>>
>> This one passes make fate-fft CPUFLAGS=3dnow and make fate-fft
>> CPUFLAGS=3dnowext on x86-32. 64-bit support probably doesn't work but
>> is dropped in the next commit so I don't care.
>
> We might invert the commits, but I guess there is enough agreement in
> dropping 3dnow since is deprecated by amd as well.

This patch doesn't drop it, it converts it to yasm. We're disabling it
for x86-64 for the simple reason that all 64bit CPUs have SSE, which
is always faster than 3dnow/ext, simply because SSE acts on 16-byte
XMM registers, whereas 3dnow/ext acts on 8-byte MMX registers.

I can merge the patches if that makes people happier. OK to commit then?

Ronald
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to