On 02/01/2012 02:37 PM, Christophe Gisquet wrote:

> Hi,
> 
> 2012/2/1 Ronald S. Bultje <[email protected]>:
>> some of the failures we see after bswap32 was SSE'ified.
> 
> I guess I'm responsible for those failures (assuming destination was
> always aligned, while it is not guaranteed).
> 
> This is off-topic, but if this goes through, then I suggest the following:
> 1) Mention somewhere the allowed assumptions for bswap_buf;
> 2) Rewrite (and in fact simplify) a bit the SSE implementation, as I
> guess the padding will now allow working on batches of 4 values at a
> minimum.


I don't think we can assume padding on the input in all cases currently.
If it will simplify the asm, we could of course define that size is a
multiple of 4 (or even 8 if needed) and check all uses to make sure they
comply.

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

Reply via email to