I am trying to figure out the best way to implement the PowerPC equivalent of the x32/x64 RDTSC instruction on Windows NT V4.00 SP2 and it seems to be to use the PowerPC TB register. I have written some assembler and would be interested to know if others think it's optimal/correct. It gives values what seem correct when I run it on my 133 Mhz PowerPC 604 Motorola PowerStack system, but I need to multiply the value by 8.

Looking at http://www.cap-lore.com/code/TB/ it mentions x 16 is needed, it also has the http://www.cap-lore.com/code/TB/tb.s code that seems very strange to me, my code is:

1        .globl      ..ppc_rdtsc
2
3        .section    .text
4
5..ppc_rdtsc:
6        mftbu       r4                          # read TB upper
7        mftb        r3                          # read TB lower
8        mftbu       r5                          # read TB upper for check
9        cmpl        0,0,r4,r5                   # TB upper changed ?
10        bne         ..ppc_rdtsc                 # if NE yes, try again
11        blr                                     # done

How can I know what to multiply by? Will I need a table for all the different CPU models (PVR values)?

Note: I am interested in solutions, not posts that simply imply it's tricky without links to justify the opinions.