[Info-vax] If I wanted to get there, I certainly wouldn't start from here (was: Re: VMS on a PC)

[John Reagan]

"Richard Maher" <maher_rj at hotspamnotmail.com> wrote in message 
news:gl2up1$dc6$1 at news-01.bur.connect.com.au...
> Hi John,
>
>> As I've said before, Macro-32 is the real bottleneck.  Besides redoing
> about
>> 1/3 of the compiler (the same 1/3 redone from Alpha to I64), you have to
>> come up with some mapping of the registers used by the current code base
>> (code currently assumes registers R0-R31 are available in some fashion).
>> Try mapping that onto the 16 registers in the X86-64 architecture and get
>> that to work with linkages to C/BLISS; exception handling; unwinding; 
>> etc.
>
> Again just idle curiosity, but would a lot of these problems go away if 
> one
> were porting VAX/VMS to X86 rather that Alpha or I64? I guess there's just
> so much software and new versions that just never made it to VAX, but 
> would
> CISC to CISC have been a more straight forward exercise? Surely after
> anything-to-EPIC, everything else has to be easy?
>

Well, you'd have a 32-bit operating system with absolutely no 64-bit 
support, etc.  Plus the VAX code base needed extensive changes for the 
original port to Alpha.  There were VAX-isms buried everywhere.  For VAX to 
Alpha, every single .MAR file needed some changes (some minor, some major). 
For Alpha to I64, I doubt that 5% needed touching.  All of that work to 
isolate target dependencies would have to be replayed.  As you mentioned, 
all the new features across the board would have to be re-ported/back-ported 
to the old code base.  And you'd still have an architecture that didn't have 
IPL-based exceptions & ASTs, interlocked queue instructions, etc. that would 
need some sort of SWIS layer to implement (you'd have to redo this 
regardless of which code base you started from).

The fact that Itanium is EPIC doesn't really matter to most code other than 
compilers or those brave few who want to write Itanium assembly for 
performance reasons (as opposed to SWIS which is written in assembly to have 
absolute direct access to the hardware without any compiler getting in the 
way).


>
> PS. Presumably after BLISS on VMS for I64, a "supported" BLISS for Windows
> I64 would not be out of the question? Does Microsoft not bother with a
> Windows version for Itanium any more? Does Rdb no longer induldge their
> fantasies with that Rdb for NT Workbench thing? Did anyone (apart from Rdb
> Engineering) ever really want Rdb on Windows anyway? And who really cares?
>

There is Windows for I64.  Can't speak for Oracle/Rdb.  However, most BLISS 
code by definition is chocked full of OpenVMS-isms (system service calls, 
RMS calls, LIB$ calls).  Compiling for Windows I64 (which doesn't use ELF by 
the way), would just give you objects that probably wouldn't link.  Doesn't 
seem to make sense to support something that would not have any users.

John