Re: [BLACKBOX] ACML- LAPACK - Mersenne Twister

From: [at]} <Georgy>
Date: Sat, 22 Nov 2008 10:24:26 +0100


Hello Robert,
 
thank you very much for sending the interfaces to Acml!
 
My experience with AMD started during pre Athlon era with K6, K6-2 and so on. As some versions of Acml are compiled with Intel compiler (!), I think there should be no problem using the library with Intel CPUs.

As for the use of Lib, I have Lib version from Cpc installed. And according to names the files you've send should also be installed in Lib Folder. So what would be the best way to go: should I remove the Cpc version of Lib, overwrite the existing Lib files, or do something else?
 
Best regards,
Georgy

On Fri, Nov 21, 2008 at 4:03 PM, Robert <robert.campbell_{([at]})nowhere.xy


Georgy

I attach below the files that interface to Acml.

The two files you want to look at closely are the Module called AcmlDll; these provide (and document) the actual DLL interface. It does not interface to all the Acml functions; just those I have used the most. It serves as a template for the other functions if you wish to extend it.

I have included Lib misc for the simple (and rather trivial) reason that it includes the procedure required to make the hyperlinks in the Docu files work.

The DLL uses FORTRAN style TYPEs for things such as Complex Matrices. Lib defines these types differently. The Module Acml provides the interface between the two, and my application code always calls Acml, not AcmlDll directly.

The Lib type definitions emphasise safety and convenience, rather than speed. You may well want to use different definitions for, for example, Complex Matrix. In this case you will want to write something different to Acml, but maybe using it as a template.

The majority of the attached files are there simply to explain the Types as I have defined them in Lib.


The attached files are later versions than those published on Cpc. I don't guarantee that they can be mixed with other Lib files taken from Cpc. (In fact,
I don't guarantee anything!).


Let me know if you decide to use Acml.

Regards

Robert


PS - I have not tried the latest version yet. The version I have used is defined
in the Acml Docu file.


Robert wrote:



Georgy Jikia wrote:



Hello Robert,
 I was also thinking about using some maths library with BlackBox, e.g. GSL.
If AMD library works also on Intel CPU's, which hopefully should be the case, I would be glad to have a look at your CP interfaces with ACML.
 Regards
Georgy



The version of ACML I have used is 3.6. I think the latest version is 4.2,
which I shall try when I have some time.


ACML may be 'optimised' for AMD chips, but I use it on both Amd and Intel
and it works fine. There is no obvious loss of speed on Pentiums, but maybe
the last few % are missing.

When I looked (a few years ago) Intel also did a similar Maths library, but it
cost money! (And I am a long term AMD supported; does anyone remember the great
'bit-slice' chips in the 70's, or the Intel floating point bug fiasco discovered
by Professor Nicely?)

The linear algebra library LAPACK is excellent.

ACML includes several random number generators. The only one I have used is the
Mersenne Twister. As I say, it is now my random number generator of choice. (I
previously preferred Knuth's Lagged Fibonacci Sequence generator.)

ACML also includes FFTs. These work ok, but are neither the fastest or most accurate implementations that I have.


I will collect together the interface material, and send it out in the next few days.

Cheers

Robert



---- To unsubscribe, send a message with body "SIGNOFF BLACKBOX" to LISTSERV{([at]})nowhere.xy
Received on Sat Nov 22 2008 - 10:24:26 UTC

This archive was generated by hypermail 2.3.0 : Thu Sep 26 2013 - 06:30:50 UTC