[Ffmpeg-devel] [RFC] port cabac asm to AMD64

Loren Merritt lorenm
Sun Oct 22 00:18:14 CEST 2006


On Sat, 21 Oct 2006, Guillaume POIRIER wrote:
> On 10/21/06, Reimar D?ffinger <Reimar.Doeffinger at stud.uni-karlsruhe.de> 
>> On Sat, Oct 21, 2006 at 03:33:00PM -0600, Loren Merritt wrote:
>>> On Sat, 21 Oct 2006, Reimar D?ffinger wrote:
>>>
>>>> though it should be possible to optimize further, mixing use of 
>>>> 64 and 32 bit registers is said to be quite slow.
>>>
>>> How so? My AMD optimization manual says that you _should_ mix 32bit and
>>> 64bit registers, because there's no penalty and the 32bit opcodes are
>>> smaller.
>>> (http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/25112.PDF
>>> page 68)
>> 
>> Maybe I confused that with access to 32 bit and 64 bit stuff at the same
>> memory address, that definitely had some problems.
>
> Or also partial register stall problems that happen when you do a
> write operation to a partial register and then a read operation from
> its larger version of this register.

There is no partial register stall, because all 32bit operations zero the 
high half of the register (unlike 8bit and 16bit ops, where it is a 
problem).

--Loren Merritt



More information about the ffmpeg-devel mailing list