FIR Filter on MMX Pentium

loop1:

movq mm0, [esi] ; load four samples

pmaddwd mm0, COEFaddr[edi] ; 4 multiplies, 2 adds

/* two cycle stall happens here */

paddd mm7, mm0 ; accumulate intermed results

add edi, 8 ; update coefficient index

add esi, 8 ; update delay line pointer

dec ecx ; decrement loop count

jnz loop1