[FFmpeg-devel] [PATCH] Revert "Merge commit '0a39c9ac0bfd7345fe676b4e2707d9cec3cbb553'"

[James Almer]

On 1/31/2017 7:12 PM, Michael Niedermayer wrote:
> On Tue, Jan 31, 2017 at 05:39:56PM -0300, James Almer wrote:
>> On 1/31/2017 4:34 PM, Michael Niedermayer wrote:
>>> The assumtation this is based on is wrong, the code is not always run with bitexact flags

Assumption not assumtation. Missed it the first time.

>>>
>>> This reverts commit a956164e1eb3418922cae949f02ad4035f013213, reversing
>>> changes made to f6005907fdeb9e4de37568ed5c1a8e7b869126f6.
>>> ---
>>>  libavcodec/x86/hpeldsp.h          |  2 +-
>>>  libavcodec/x86/hpeldsp_init.c     |  2 +-
>>>  libavcodec/x86/hpeldsp_vp3_init.c | 14 +++++++++-----
>>>  3 files changed, 11 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/libavcodec/x86/hpeldsp.h b/libavcodec/x86/hpeldsp.h
>>> index 0ecc97a83a..bf97029b57 100644
>>> --- a/libavcodec/x86/hpeldsp.h
>>> +++ b/libavcodec/x86/hpeldsp.h
>>> @@ -52,6 +52,6 @@ void ff_put_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels,
>>>  void ff_put_pixels16_xy2_ssse3(uint8_t *block, const uint8_t *pixels,
>>>                                 ptrdiff_t line_size, int h);
>>>  
>>> -void ff_hpeldsp_vp3_init_x86(HpelDSPContext *c, int cpu_flags);
>>> +void ff_hpeldsp_vp3_init_x86(HpelDSPContext *c, int cpu_flags, int flags);
>>>  
>>>  #endif /* AVCODEC_X86_HPELDSP_H */
>>> diff --git a/libavcodec/x86/hpeldsp_init.c b/libavcodec/x86/hpeldsp_init.c
>>> index e583bd9ffe..58e27e3542 100644
>>> --- a/libavcodec/x86/hpeldsp_init.c
>>> +++ b/libavcodec/x86/hpeldsp_init.c
>>> @@ -309,5 +309,5 @@ av_cold void ff_hpeldsp_init_x86(HpelDSPContext *c, int flags)
>>>          hpeldsp_init_ssse3(c, flags);
>>>  
>>>      if (CONFIG_VP3_DECODER)
>>
>> How about checking for AV_CODEC_FLAG_BITEXACT here instead of reverting the
>> function signature? Keeps differences as minimal as possible while having
>> the same effect.
> 
> technically yes thats possible it makes the code confusing though and
> may cause bugs in the future, for example then the whole set of
> vp3 optimizations here would depend on AV_CODEC_FLAG_BITEXACT being set
> and thats unexpected. Unexpected things / surprises can be bad for code
> quality.
> 
> Anyway i can change it to that test it and resubmit if preferred ?

No, it was mostly a nit. The patch is ok as is. The differences with libav
with this change will be minimal in any case.