[Ffmpeg-devel] ffmpeg now 2x slower: sse16 vs nsse16

[g.]

I recently updated my copy of ffmpeg with the latest CVS release from the 
February version I had previously been using and found that with the same 
settings it is now twice as slow to encode!

Doing a profile of the two versions I found that with the latest version 47% 
of the time was spent in nsse16_mmx.

In the previous version nsse16_mmx doesn't seem to have run at all.

In the previous version sse16_mmx took 4% and in the latest version it was 
replaced by sse16_sse2 which took 1.6%

Any ideas as to what has changed over the last few months that could have 
caused this?

In both cases I had:

    m_codec_context->me_sub_cmp = FF_CMP_NSSE;
    m_codec_context->mb_cmp = FF_CMP_NSSE;
    m_codec_context->nsse_weight = 0;
    m_codec_context->mb_decision = FF_MB_DECISION_SIMPLE;

I'm using a P4 with Hyperthreading.

g.