"Ronald S. Bultje" <[email protected]> writes:

> Hi,
>
> On Sat, May 28, 2011 at 3:36 PM, Yuriy Kaminskiy <[email protected]> wrote:
>> Ronald S. Bultje wrote:
>>> Hi,
>>>
>>> On Fri, May 27, 2011 at 6:30 AM, Gil Pedersen <[email protected]> wrote:
>>>> On 25/05/2011, at 20.38, Ronald S. Bultje wrote:
>>>>
>>>>> Preparatory step to implement support for sizes > VOFW.
>>>>> ---
>>>>> libswscale/swscale.c              |   68 ++++---
>>>>> libswscale/swscale_internal.h     |   27 ++-
>>>>> libswscale/swscale_template.c     |  130 +++++++------
>>>>> libswscale/utils.c                |   19 ++-
>>>>> libswscale/x86/swscale_template.c |  380 
>>>>> +++++++++++++++++++++----------------
>>>>> 5 files changed, 359 insertions(+), 265 deletions(-)
>>>> The b4a224c5e4109cb2cca8bac38628673d685fe763 commit breaks libswscale on 
>>>> my setup. A 64-bit Intel(R) Xeon(R) CPU X3370.
>>>
>>> Thanks for the ping, should be fixed now
>>> (9f5d45025e8df9d5f39832caad16b94cb6ac11c5).
>>
>> No, it is not fixed. After replace REG_a with %3, %3 is modified in asm code,
>> but not marked as early clobber, and not even marked as modified.
>> It may accidentally work in some configuration, but still wrong and broken.
>
> I could mark it as clobber, although it doesn't matter much ATM since
> uv_off is not used afterwards. I'll accept a patch that does this, of
> course.
>
> It doesn't have to be marked as modified (this is called "output", not
> "modified", btw) because I don't care about the end value of the loop,
> I discard it anyway.

You are not allowed to modify input-only parameters.  The compiler may
be relying on these being preserved.

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

Reply via email to