psllw xmm7,BYTE_BIT
por xmm5,xmm7 ; xmm5=Cb
- movdqa XMMWORD [ebx], xmm5 ; Save Cb
+ movdqa XMMWORD [rbx], xmm5 ; Save Cb
movdqa xmm0, XMMWORD [wk(3)] ; xmm0=BO
movdqa xmm6, XMMWORD [wk(2)] ; xmm6=BE
pslldq xmm1,(SIZEOF_XMMWORD-2)
movdqa xmm2,xmm1
- pand xmm1, XMMWORD [edx+1*SIZEOF_XMMWORD]
- pand xmm2, XMMWORD [edi+1*SIZEOF_XMMWORD]
+ pand xmm1, XMMWORD [rdx+1*SIZEOF_XMMWORD]
+ pand xmm2, XMMWORD [rdi+1*SIZEOF_XMMWORD]
movdqa XMMWORD [wk(2)], xmm1 ; xmm1=(-- -- -- -- -- -- -- 15)
movdqa XMMWORD [wk(3)], xmm2 ; xmm2=(-- -- -- -- -- -- -- 15)