x86: hpeldsp: implement SSSE3 version of _xy2
authorChristophe Gisquet <christophe.gisquet@gmail.com>
Thu, 22 May 2014 21:47:06 +0000 (23:47 +0200)
committerMichael Niedermayer <michaelni@gmx.at>
Sat, 24 May 2014 13:15:56 +0000 (15:15 +0200)
commit81aa0f4604f98da692f2689c84968f90354a92ea
tree243197f645e40a7dd5219df7992e4f8452fc73a9
parent726316240bcc41cef6053dd6d1e46a3c57328498
x86: hpeldsp: implement SSSE3 version of _xy2

Loading pb_1 rather than pw_8192 was benchmarked to be more efficient.
Loading of the 2 yields no advantage. Loading of one saves ~11 cycles.

decicycles count:
put8:  3223(mmx)    -> 2387
avg8:  2863(mmxext) -> 2125
put16: 4356(sse2)   -> 3553
avg16: 4481(sse2)   -> 3513

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
libavcodec/x86/hpeldsp.asm
libavcodec/x86/hpeldsp_init.c