I5110401%DBSTU1.BITNET@cunyvm.cuny.edu (01/02/90)
HBO043@DJUKFA1 writes: >- But you can avoid the 32bit multiplication generated by proc_addr > if you enlarge a proc table entry by some junk to 128 bytes. Any > reasonable compiler will generate a shift instruction to perform the > multiplication. This should be slightly faster than the additional > memory reference due to a pointer array proposed here some time ago > and it involves a minimal source code change. >(This should concern PC-users, too]) >Any comments? I do not really believe that shifting around for 7 bits is such a big win against shifting twice and reading 2 words. Since the 68k cpu does not have a barrel shifter shifting costs 2 cycles *per bit*, hence 14 cycles plus the time for the instruction evaluation are needed for the former. The latter costs 4 cycles for the shifts and 12 cycles for the memory reference if i am not wrong here. Someone should take a look at the 68k manuals here. However, the possible win is not that big that i am willing to waste some memory for that. --- Kai-Uwe Bloem, I5110401@DBSTU1.BITNET, TU Braunschweig, w/germany ---