Intel Fortran Math Kernel Library for g77 on OS/2

Intel has written and maintains a mathematical library, the Math Kernal Library (MKL), which is basically a BLAS tuned for Intel x86 CPUs. So far it is available only for Windoze NT (and supposedly works on Windoze '95). More information can be found at the Intel web site:

http://developer.intel.com/design/perftool/perflibst/mkl/index.htm .

The MKL version 2.0 is specifically optimized for the Pentium Pro/Pentium II processors and is designed as multithreaded to support the advantages of the symmetrical multiprocessing, so far on Windoze NT.

Given that the floating point performance of Intel x86 CPUs traditionally can't hold a candle to many of the RISC CPUs from the same area I was interested in getting more out of my Pentium-Pro 150 (and Pentium 120) based systems. I am running OS/2 Warp 4 and am not planning to change to any of the Microsoft offerings.

My initial attempt to get an OS/2 version of the Intel MKL dates back to May 1997:

Date: Wed, 14 May 97 23:16:14 -0400

From: Stefan A. Deutscher <stefand@ibm.net>

Subject: Intel math library question

To: developer_support@intel.com

Reply-To: sad@utk.edu

X-Mailer: ELM [version 2.3 PL11] for OS/2

Hallo,

I just ran accross your math library and I was wondering whether you'd

also consider makeing available versions for IBM OS/2 Warp 4?

A Watcom f77 compiler exists for that operating system. Also, an a.out

(or *.a) style version for use with g77 (the gnu fortran compiler) would

be very interesting, especially since that compiler rund on a wide

variety of x86 operating systems including OS/2, Linux, FreeBSD,

Solaris, SCO Unix, and even the Microsoft offerings.

OS/2 Warp in its server incarnation (and supposedly soon in version 4.1

also in its client incarnation) also supports SMP and scales reportedly

more efficiently with CPU number than Microsoft Windows NT.

Kind regards! Stefan A. Deutscher

and resulted in the following reply:

Date: Thu, 15 May 1997 11:36:18 -0400 (EDT)

From: Corey Alsamriae <ialsupp@mailbox.intel.com>

Subject: Re: Intel math library question

You wrote:

>I just ran accross your math library and I was wondering whether you'd

>also consider makeing available versions for IBM OS/2 Warp 4?

At this time we have no plans.

Please let me know if you need further assistance.

Best regards,

Corey Alsamriae

Intel Corp.

That was not very much to fuel hope. Now I ran accross a link to the Intel MKL again a few days back and decided to try my luck again, this time my complete post has been quoted in the reply of the gentleman from Intel, so I repoduce it this way:

Date: Tue, 17 Mar 1998 15:06:23 -0500 (EST)

From: "bob h" <devsupp@mailbox.intel.com>

Subject: RE: Math Kernel Library for OS/2 ?

To: devsupp@mailbox.intel.com

Stefan

Thanks for you comments on our library, There is no offering of MKL

on OS/2 nor are there any plans for making MKL available on OS/2.

There has been almost no request for support of MKL on OS/2. MKL is

currently supported on Windows NT* and will run successfully on

Windows 95.

Bob H

Intel Customer Support

* All other brands and names are property of their respective owners.

>Dear Sir,

>

> I have found on the web a reference to an OS/2 version of the Intel

>Math Kernel library. Since I had placed a request for an OS/2 version a

>while ago with your support group and got a reply along the lines "not

>currently available but we will consider it" I was very excited about

>this information.

>

>It has been my experience that the floating point performance of Intel

>based systems lags even for newer CPU types (Pentium and Pentium Pro)

>behind that of the RISC based machines I have access to as well (Sun,

>DEC, IBM, HP, and SGI currently). While I have had no difficulties

>compiling the BLAS sources using GNU Fortran g77, the performance of the

>resulting library is bound to stay below the capabilities of the CPUs,

>and of a library version tuned by Intel specialists.

>

>Looking up your web site I see only Microsoft oerating environments and

>operating systems mentioned.

>

>However, at least one of the supported Fortran compilers (Watcom f77) is

>truely cross-platform and capable of producing highly optimized OS/2

>object files. This lets me assume that indeed Intel is able to cater

>also to the needs of those of their customers who chose to combine Intel

>CPUs with IBM OS/2 rather than with one of the Microsoft offerings.

>

>I hope I just overlooked the appropriate link. Would you kindly point me

>to the OS/2 version of your performance library?

>

>Should there be none, would you, please, pass this on as a request to

>the appropriate decision maker and also furnish me with the information

>needed to contact this person directly?

>

>A native OS/2 version (for Watcom f77, and as there is a large group of

>GNU Fortran g77 users on OS/2, also as an .a library) of your library

>would be very much appreciated, and could enhance the image of Intel as

>an open, cross-platform vendor at a very small cost in terms of time and

>maintenance.

>

>

> Kind regards,

>

> Stefan A. Deutscher

This response didn't fully resolve my question, and since I also wasn't furnished with a contact for a decision maker I sent a follow-up (again, quoted in full) and got the following reply:

To: stefand@ibm.net

From: "bob h" <devsupp@mailbox.intel.com>

Date: Wed, 18 Mar 98 11:42:47 -0800

Subject: RE: RE: Math Kernel Library for OS/2 ?

Stefan,

Thanks for the comments you sent back. It lets me know you have

received my message. We do not oppose you using the answer I sent you,

we are interested in what our customers needs are and the feedback we

receive will be helpful. I ask, that you ask the readers of the news

groups after posting their comments. also sent them to Intel at

developer_support@intel.com, or use the Intel news group at

http://newsgroups.intel.com/ngi.exe?group=ov02 to make comments. I

would like to be able to read all the feedback I can.

Bob H

Intel Customer Support

>Dear Bob,

>

> this is quite a pity. Sometimes one has to create a market / demand

>rather than hang back and see. This is how Sony had success with

>walkmen (*), Phillips with the CD etc. Relying on demand is important

>but can also lead to self-fulfilling prophecies.

>

>For me, this means that Intel will be a dead end for high performance

>computing, since experience has taught me not to consider anything

>except OS/2 and UNIX in its various incarnations (especially FreeBSD on

>the PC and AIX and Digital UNIX on the workstation side).

>

>As it appears, I will probably have to shift my own budget to

>appropriately supported systems by IBM and Digital/Compaq in the long

>run, and I will also have to suggest that to that to the researchers I

>advise in the US and in Europe. This is rather sad as I do quite enjoy

>the convenience that the use of the mainstream CPU by Intel has offered

>so far.

>

>How difficult or time consuming would it be to put out an OS/2

>recompile? After all, it is the same compiler, the same CPU, and a very

>similar OS.

>

>Unless you oppose I will take the liberty to post a copy of this in the

>relevant OS/2 news groups as well as in comp.lang.fortran. This may help

>to establish a more reliable idea for the need, as the Intel MKL is one

>of the better kept secrets of Intel CPUs which for known reasons are not

>normally considered as heavy duty floating point number crunchers.

>

>Let me reiterate: Please, forward this to the appropritate decision

>maker and be so kind as to furnish me with the relevant contact

>information.

>

> Kind regards! Stefan A. Deutscher

To sum it up: In the last year nothing has really changed in the availability of the library. If you are interested in getting a bit more

performance out of your Intel CPUs using Watcom f77 or g77 on OS/2, please send email to: developer_support@intel.com .

Should there be a clear enough demand I'd hope that Intel lives up to the expectations one can have from the major x86 CPU maker, even more so since it appears to involve not a lot of recoding.

Cheers! Stefan

---

============================================================================

Stefan A. Deutscher | (+33-(0)1) voice fax

Laboratoire des Collisions Atomiques et | LCAM : 6915-7699 6915-7671

Mol\'{e}culaires (LCAM), B\^{a}timent 351 | home : 5624-0992 call first

Universit\'{e} de Paris-Sud | email: sad@utk.edu

91405 Orsay Cedex, France (Europe) | (forwarded to France)

============================================================================

.. do you know what they call a quarter-pounder with cheese in Paris?