Re: New assembly language instructions to support OO languages?

>>> Having a fast hash instruction might help: It would take two numbers
>>> as arguments and produce a single number that is a hash of the pair.
>>> It would not need to be cryptographically strong, just good enough for
>>> hash tables.
>>
>> if that's the only criterion xor is as good as anything else.
>
> Xor is not really good enough for hash tables.
>
>> otoh there are some MMX op-codes that could be abused for this purpose
>> and may give better results when used on strings.
>
> Possibly. I haven't studied MMX that closely.


it has some shift-and multiply operations I've no looked closely either.
I think these were 8-bit operations


>>> A good hash can take many "normal" instructions, but you can make a
>>> good hash efficiently in hardware.
>>
>> processors with strong hashes are available (Via Eden has AES in hardware)
>>
>> good hashes (like CRC-32?) do lend themselves to very simple hardware
>> implementation (if you can do them in serial or byte-wise)
>> how fast does it need to be to be worth-while?
>
> I was thinking about as fast as 1-3 additions.


I guess it can be done a word at a time for 32-bit
words I think that'd require a bunch of multi-input XOR
gates, it seems poteintially almost as slow as a multiply operation


doing it fast will be almost as hard as doing fast multiplies.

Return to the comp.compilers page.
Search the comp.compilers archives again.