Re: [gforth] IA64 asm?

[Jeffrey Chimene]

On 04/26/2013 04:49 AM, Bernd Paysan wrote:
> Am Donnerstag, 25. April 2013, 18:17:04 schrieb Jeffrey Chimene:
>> On 04/25/2013 05:26 PM, Bernd Paysan wrote:
>>> Am Donnerstag, 25. April 2013, 12:17:15 schrieb Jeffrey Chimene:
>>>> Hi Folks,
>>>>
>>>> I didn't see anything about this in the archives.
>>>>
>>>> Can anyone tell me why there's no IA64 assembler?
>>>
>>> Probably because nobody had the time to write one?
>>
>> HI Bernd,
>>
>> Thanks for the reply!
>>
>> I've been following GForth for a few years. I finally have a use case.
>>
>> Background
>> =======
>>     I'm interested in porting Node.js to OpenVMS. While I'd prefer
>> starting w/ Alpha, it's probably better to start w/ IA64.
>>
>>     To port Node.js, one must also port the V8 _javascript_ engine. V8
>> generates machine code at run time along two paths: a sort of "fast
>> first" path, and a dynamically optimized path. Both paths converge on a
>> set of common routines all of which converge on a common code generator.
>>
>>     To bootstrap the porting process to Mips from an existing
>> architecture (e.g. AMD64), somebody wrote a Mips simulator (C++). This
>> simulator exports the same function signature as the code generator above.
>>
>> Today
>> ===
>> The simulator is a fine piece of work, but it's written for a RISC
>> machine. I cannot see converting the simulator out of the box to IA64. I
>> really think I'm going to need a workbench to figure out the intricacies
>> of IA64. Fortunately, I don't need the /entire/ IA64 instruction set.
>> Some instructions are forbidden to user-space code, and other
>> instructions don't make sense in the context of V8 (e.g. MMX & XMM
>> instructions). Because this is OpenVMS, the alternate IA32e mode does
>> not apply. Once I have that subset sort-of, kind-of working, I can
>> direct attention to the simulator.
>>
>> And, what better workbench than Forth?
>>
>> Are you interested in the results? Or, is the subsetting a no-go for you?
>
> Several assemblers in Gforth (including x86/x64 and ARM) are subsets, because
> there are so many additions to the ISAs, so subsets are not a problem at all. 
> Go for it and share the results!
>
Great! Thanks for the encouraging words!

Be seeing you,
jec