&test ("ecx",1<<20); # check for P4
&jnz (&label("loop"));
&test ("edx",1<<11); # check for XOP
- &jnz (&label("XOP"));
+ &jnz (&label("XOP")) if ($ymm);
&and ("ecx",1<<30); # mask "Intel CPU" bit
&and ("edx",1<<28); # mask AVX bit
&or ("ecx","edx");