On 16.04.19 10:43, Richard Henderson wrote:
> On 4/15/19 10:33 PM, David Hildenbrand wrote:
>>>>
>>>> Indeed, I didn't really explore vector operations yet. This is more
>>>> compact than I expected :)
>>>
>>> :-)
>>>
>>> That said, in implementing vector variable shifts today,
>>> I've come up with a representational problem here.
>>> You may want to hold off on these until I can address them.
>>
>> I assume you mean vector helpers *in general* in this file. Yes, we can
>> add them later.
>>
>> What exact problem are you dealing with?
> 
> The .opc field lets you only specify one opcode which is optional in the
> backend on which you depend.  In writing support for AArch64 USHL, I find that
> I needed 3 optional opcodes.

I was asking myself this exact thing when looking at the opc field in
the example you gave ("which instruction is one supposed to indicate
here") :)

> 
> I'm thinking of a mass change whereby .opc becomes a pointer to an array with
> terminator.  But then, for debugging purposes, I think I need to validate that
> array, so that it's not missing things that ought to be specified, but which
> happen to be supported by the current host.

Makes sense!

-- 

Thanks,

David / dhildenb

Reply via email to