[FFmpeg-devel] [PATCH] ac3enc: Add x86-optimized function to speed up log2_tab().

[Justin Ruggles]

On 02/13/2011 05:49 PM, Loren Merritt wrote:

>> +cglobal ac3_max_msb_abs_int16_%1, 2,2,5, src, len
>> +    pxor        m2, m2
>> +    pxor        m3, m3
>> +.loop:
>> +%ifidn %2, min_max
>> +    mova        m0, [srcq]
>> +    mova        m1, [srcq+mmsize]
>> +    pminsw      m2, m0
>> +    pminsw      m2, m1
>> +    pmaxsw      m3, m0
>> +    pmaxsw      m3, m1
>> +%else ; or_abs
>> +%ifidn %1, mmx
>> +    mova        m0, [srcq]
>> +    mova        m1, [srcq+mmsize]
>> +    ABS2        m0, m1, m3, m4
>> +%else ; ssse3
>> +    ; using memory args is faster for ssse3
>> +    pabsw       m0, [srcq]
>> +    pabsw       m1, [srcq+mmsize]
>> +%endif
>> +    por         m2, m0
>> +    por         m2, m1
>> +%endif
>> +    add       srcq, mmsize*2
>> +    sub       lend, mmsize
>> +    ja .loop
>> +%ifidn %2, min_max
>> +    ABS2        m2, m3, m0, m1
>> +    por         m2, m3
>> +%endif
>> +%ifidn mmsize, 16
>> +    mova        m0, m2
>> +    punpckhqdq  m0, m0
> 
> movhlps


Ah, I thought there was some instruction like that, but I must have
missed it when I searched for it.  I'll send a new patch to change this
line since the original patch was already committed.

Thanks,
Justin