value range propagation for _bitwise_ OR
"Jérôme M. Berger"
jeberger at free.fr
Mon Apr 12 10:39:57 PDT 2010
Ali Çehreli wrote:
> Jérôme M. Berger wrote:
>> Ali Çehreli wrote:
>>> � wrote:
>>>
>>>> The idea is to build a value that is between minA and maxA and will
>>>> set as many bits as possible when or'ed with maxB:
>>> The assumption that maxB would be the value that produces the maximum
>>> a|b is not correct. A lower valued b may fill more gaps in the bit
>>> representation of what is calculated from min_a and max_a.
>>>
>>> Your function failed for me with the following values:
>>>
>>> min_a 00000000000000000000000011001000 000000c8 200
>>> max_a 00000000000000000000001100001111 0000030f 783
>>> min_b 00000000000000000000000001000101 00000045 69
>>> max_b 00000000000000000000001001100001 00000261 609
>>> calculated 00000000000000000000001001100101 00000265 613
>>> WRONG! empirical 00000000000000000000001111111111 000003ff 1023
>>> emp_max_a 00000000000000000000000110011110 0000019e 414
>>> emp_max_b 00000000000000000000001001100001 00000261 609
>>>
>>> Please see my test code elsewhere in the same thread: :)
>>>
>>>
> http://www.digitalmars.com/webnews/newsgroups.php?art_group=digitalmars.D&article_id=108851
>
>>>
>> The "calculated" value above obviously was not computed with my
>> function!
>
> My mistake: It was your function but the highbit() that I used was not
> correct. The one I used was returning the *value* of the highest bit.
>
>> Since the return value from my function includes maxA and
>> maxB, at least all bits that are set in either of those should be
>> set in the output.
>>
>> I've run my code with those input and the result is 3ff as
>> expected... (See attached source file).
>
> Perhaps my test code is wrong; but I can't find my error. :/
>
> Your function reports 0b_111 for these set of values:
>
> min_a = 5;
> max_a = 6;
> min_b = 4;
> max_b = 4;
>
> But the maximum value of a|b is 6.
>
Yes, like I said with my code, it is conservative. It will give the
optimal result for over 99% of the cases but give a slightly higher
value for the rest. I'll post a new version that's 100% precise in a
few minutes.
Jerome
--
mailto:jeberger at free.fr
http://jeberger.free.fr
Jabber: jeberger at jabber.fr
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: OpenPGP digital signature
URL: <http://lists.puremagic.com/pipermail/digitalmars-d/attachments/20100412/eff3e6e9/attachment.pgp>
More information about the Digitalmars-d
mailing list