Interval arithmetic functions

[Todd_R_Intel]

Because we have not seen much adoption of our interval arithmetic functions, we are considering removingthem from Intel MKL in an upcoming release. For those that might be using them, we're interested to hear from you.

Update: By "interval arithmetic functions" I meant the interval linear solvers documented in the reference manual as well as any associated arithmetic functions and operators. Sorry for the confusion.

Todd

[g_f_thomas]

This is the only feature of MKL that I use and I would be more than annoyed were it to be dropped. All other features of MKL are readily available elsewhere and for free.Interval arithmetic isnot so readily accessibleunless from Matlab or Sun. I have an outstanding query on the implementation of interval arithmetic in MKL that remains unanswered from months ago. Perhaps if the monitors of this forum were more attentive to user's requests for assistance then the entire product and not just its interval arithmetic features would benefit. Why has the rationale for its inclusion in the first instance changed? I doubt that its maintenance amounts to much. If your going to chip away at what you perceive as 'unadopted features' what's next? The documentation, with its entrenched obscurantism and mesmerism that renders MKL almost unusable, would be a good place to start. Piff.

Gerry

[yoann_fabre]

Hi,

I'd be a heavy user of such functions, had I known where to find them...
I'm using 10.0.1 for just a few days but I've browsed all the docs and
I was /not/ aware of them !
I know about the interval /solvers/ but I cannot find any documentation
about arithmetic functions (that should exist anyway for the solver...)

I actually use my own implementation of interval arith. and I was considering
rewriting it to use the SSE trick allowing to rely only on the round toward +inf
mode... but if you guys can make them 'clearly' public I'd more than happy to
use them... I strongly suspect that I'm not the only one : at my shop we are
quite a lot to do interval computations !

Please do NOT remove these functions. If I may : small adoption maybe the
consequence of "under-documentation" :-)

NB : To reply to the previous poster : I agree that before 10.0 release
documentation was really poor but you guys have done a really great
job about that with the 10.0 ! Congrats (love the .chm)

PS : The only reference I can find about these functions is in the faq :

Limitations to the interval arithmetic functions in Intel MKL 10.0 Update 1:
* The interval libraries will require the libifcore library from Intel Fortran compiler.
* Interval arithmetic functions require a processor, which supports SSE instructions.

A full search into the REFMAN in PDF has not given any result !
Please WHERE (and HOW) can I find/use these functions with MS VC++ (2005) ?

[g_f_thomas]

The OP presumably was referring to the chapter Linear Interval Solvers in the .chm help.

BTW my beef with the documentation relates to the userguide.pdf.

Gerry

[Tyler_T_Intel]

Gerry, Others,

Would you mind sharing specific issues that you see with the documentation (User's Guide or other)? We would love to furtherimprove our documentation, and specific pointers (gripes) are appreciated.

--tyler

[Todd_R_Intel]

I'm sorry about the confusion. We are considering removing the interval linear solvers as well as any associated interval arithmetic functionality.I updated my original post to reflect this

As far as arithmeticis concerned,there are a few operators defined in mkl_interval.f90.

[g_f_thomas]

The mkl_interval.f90 module is the essence of the interval linear solvers package wherein the interval type is defined and interval operators which overload point data type operators are also defined.

Where in mkl does 'associated interval arithmetic functionality' reside and what is this?

If interval arithmetic is too much of a bother to the mkl maintainers then pass its custodianship on to the Intel Visual Fortran folks, they won't drop it.

Gerry

[g_f_thomas]

Tyler,

Not a day goes by that there isn't a user query that stems from the ambiguity of the userguide.pdf. A perusal of the record of recent forum postings, most of them fielded by tim18,will reveal that this is so. Perhaps the ug is as clear as glass to its author but to me and others who have to use it it's as clear as mud.

As to pointing to specific instances of confusion in the current version of the ug, sorry, no, I don't have the time to do QA for Intel. Besides, as a subscriber to mkl I expect Intel to do this.

Gerry

[Tyler_T_Intel]

Gerry,

The Intel Compiler team argued against our inclusion of IA into MKL from the start - I'm not so sure they would be happy custodians of IA going forward.

As for the User's Guide, I am certainly not asking you to do our QA/Test for us - we do indeed do that ourselves. However, despite our efforts, we seem to have fallen short regarding the effectiveness of our User's Guide. Indeed, we do see many vague comments that our documentation is "bad", but very little in the way of specifics that we could easily resolve. We willcontinue to stumble through trying to guess which aspects of the User's Guide are problematic, and/or, if users point out specific issues, we can address those (and similar) more quickly.

--tyler

[yoann_fabre]

Well... You could, for exemple, clarify the issue of my forum post [MKL 10.0.1 - PB : Sequential Version using both cores !] related to the documentation of thread safety in MKL.

See point (2) and (4) of my last post. See also the issue of the seemingly non existant file [mkl_ms_thread_dll.lib] at then end of my first post.

Both point are covered by the doc but I still have questions, so they might be unclear :-)

Moreover it seems I cannot get a clear answer to these questions. There's seem to be some confusion about "thread safety" in general and w.r.t OpenMP in particular.

A while back I've evaluated MKL 8.1 and from the doc I've gathread that the library was NOT thread safe. I was pleased to read that 10.0 is now thread safe but (afaik) only for the so called sequantial version. Now I read in replies that all versions of 10.0 are thread safe : the only point of the "sequential" one being to use internally only one thread. Since it is provably not so, well, I'm confused :-)

[g_f_thomas]

MAD thessin:

The Intel Compiler team argued against our inclusion of IA into MKL from the start - I'm not so sure they would be happy custodians of IA going forward.

That's only because they're too busytoadying to standards committees instead of catering to customers. After much effort by its proponents, IA was withdrawn from f2003but picked up by Sun Microsystems to the chagrin of the naysayers. How many subscriptions to MKL do these standards committees hold?

MAD thessin:

...we do see many vague comments that our documentation is "bad", but very little in the way of specifics that we could easily resolve. We willcontinue to stumble through trying to guess which aspects of the User's Guide are problematic, and/or, if users point out specific issues, we can address those (and similar) more quickly.

Luckilyyour technical writersdon'tproduce training material for pilots, operators of medical devices, space shuttle launchers, etc.but confine themselves to innocous ug's and the likes whose shortcomingscan be readily attributed to the user's failure in not bringing them to their attention. Piff. You'll know that the documentation is improving when the frequency of appeals for its clarification drops on the forum. Let these postings be your guide, second guessing is not required.

Gerry

[Tyler_T_Intel]

Gerry,

I offered my assistance toward helping improve our product to better meet your needs. At any time that you may have constructive feedback, I would welcome such, but your derogatory remarks don't do either of us much good.

--tyler

[Tyler_T_Intel]

Yoann Fabre,

I suspect Todd is gathering clear answers for your threading posts. Regarding your pointers re documentation posted here, these are certainly improvements we can explore.

Thanks. And, ourapologies for the confusion.

--tyler

[Todd_R_Intel]

Yoann,

I made a suggestion regarding the threading difficulties you were seeing over on the other thread. I'll also be suggesting some documentation improvements as I mentioned.

Thank you,
Todd