Numeric access to char[]

[Chad J]

nobody wrote:
> Chad J wrote:
> 
>> nobody wrote:
>> ....
>>
>>>
>>> So you will have to do it manually. I would like to suggest that if 
>>> you can pad the char[] to ensure its .length % 8 == 0 then you can 
>>> cast it to a ulong and your shifting will be faster.
>>
>>
>> Sure about ulong?  In my spare time I made my own minimal Bignum 
>> implementation.  I'm not sure about shifts, but for addition with 
>> carrying it was faster to use size_t (uint in this case) rather than 
>> ulong.  I wonder if maybe the compiler could optimize it better if it 
>> didn't have to emulate 64 bit integers.  Or my benchmark was borked.
> 
> 
> I certainly could be wrong but I would be quite surprised. I would be 
> interested to see what tests you ran that suggest addition with carry 
> was faster.

This BigNum I made is very incomplete and probably slow.  Changing the 
size of it's digits shouldn't do anything besides just that, though.

I have attached 2 .d files, BigNum.d is the implementation, main.d is 
the benchmark.  I compiled it with dmd -O -inline -release.  At the top 
of BigNum.d there is an alias that I use to switch between size_t and 
ulong variables used in the computation.

Here are my results.  See main.d for details of the setup.

Microseconds to complete 1000000 additions using size_t:
786779
776801
793653
783896
787937
789150

Microseconds to complete 1000000 additions using ulong:
866767
860499
866921
865791
860874
858154
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: main.d
Url: http://lists.puremagic.com/pipermail/digitalmars-d-learn/attachments/20060825/0c0da3cd/attachment.diff 
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: BigNum.d
Url: http://lists.puremagic.com/pipermail/digitalmars-d-learn/attachments/20060825/0c0da3cd/attachment-0001.diff