Re: RS6000 to G4 assembly changes

Thank you for the information. Most likely I won't be able to get it
working then.

Bob


On Sat, 22 Jan 2000, Gabriel Paubert wrote:

On Sat, 22 Jan 2000, Robert Neugebauer wrote:

quoted

I am not trying to build a cross compiler, I am not trying to build GCC.
My course at university uses its own compiler called uC++ (Micro C++).

The compiler thus far only has the following targets that work:
        dmake sun-sparc-sunos
        dmake sun-sparc-svr4
        dmake sun-m68k-sunos
        dmake dec-ultrix-mips
        dmake dec-alpha
        dmake sgi-mips-r4000
        dmake sgi-mips-r3000
        dmake ibm-rs6000
        dmake hp-hppa
        dmake pc-i386-linux
        dmake gizmo

By my knowledge rs6000 is a RISC based architecture, thus I tried to build
uC++ using the target ibm-rs6000. This resulted in an unrecognized opcode
movl during compilation.

Then your compiler is seriously buggy since there is no such instruction
in the RS6000 or PPC assembly. There significant differences between
the architectures (for application level code) are:

a) multiply and divides are different (not even the same opcode for most
of them), the RS6000 has an MQ register for remainders and widening
multiplies. The PPC does not have this register.

b) all the extended shifts in the RS6000 which use the MQ register do not
exist in the PPC architecture, so you have to synthesize multiword shifts
(how to do this is documented in an appendix of the programmer's
environment manual, see below).

c) a few other oddball RS6000 instructions do not exist in the PPC:
lscbx (that one was indeed very odd), doz, dozi, abs, nabs, rlmi, maskir,
maskg, and I certainly forget a few more...

d) the PPC has a full complement of single precision floating point
operations, the RS6000 does not

For a C compiler, a) is very important, b) affects only compilers which
support 64 bit operations, c) the only critical instruction is abs which
can be replaced with a simple sequence and d) can safely be ignored by
performing all FP operations in double precision.

For more information you can download the programming environments
manuals from IBM or Motorola website:

http://www.chips.ibm.com/techlib/products/powerpc/manuals/
http://www.motorola.com/SPS/PowerPC/teksupport/teklibrary/index.html

and if you are interested in improving your compiler, you might find the
compiler writer's guide (link on IBM's website) interesting.

	Gabriel.

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/