On Mon, Apr 04, 2016 at 05:51:10PM -0400, Lennart Sorensen wrote:
> On Mon, Apr 04, 2016 at 05:39:48PM -0300, Breno Leitao wrote:
> > Ok, since we will not even have VMX, I understand that we might not need a
> > POWER8, i.e, we can pursue initially with a POWER7. Does it make sense?
> >
> > Other than that, I will create a entry for you on one machine. But I am
> > wondering if we should go with powerpc arhc other than ppc64, since I am not
> > sure how stable is ppc64 right now. What is your recommendation?
> >
> > Also, feel free to send me your SSH pub key, so I can put on the VM.
>
> I am trying to think of any benefit 64bit powerpc has over 32bit other
> than the memory space for the process and I don't know if there is any.
> After all the architecture was designed from the begining with both 32
> and 64 bit in mind and hence I suspect they were meant to be as similar
> as possible in terms of register count and such. So I suspect you gain
> an increase the the process size along with the doubling of pointer size
> which is usually considered a disadvantage of course (although how much
> that matters depends on the application of course).
What is really missing in ppc-land is a variant which uses 32 bit
addresses but otherwise knows that general purposes registers are
64 bit wide and can be used as such, i.e., an alternative 32 bit ABI
which can only run on 64 bit processors and use 64 bit instructions
when beneficial while not needing over 4GB of address space.
(Truth be said, PPC code density is not great but much worse in
64 bit mode than in 32 bit mode).
I don't know for sure, but MIPS and/or Sparc may have something along
these lines.
Gabriel
