Dear Sirs ,

(A)

My ideas about dear Boian Mitov's views are
the following:

I do NOT think that my view is a narrow vision .

If you wonder how I am working I tell it :
Over my Desktop there are some html pages I prepared myself filled a large number of company/organization web page links .

Some of my web pages are open in front of me and when I want I am clicking any one of the company or organization link and I am studying its contexts .
I regularly use Yahoo or Google searches .

You are thinking in terms of Intel .

Is Intel the ONLY processor producer ?
There are a large number
of main frame producers and parallel processing is NOT a new subject of a few years .

Intel processors are used mainly by Microsoft ( DOS , Windows , etc. ) products since 1980 with DOS mainly on PCs .
Later Linux since ( around ) 1990 , later on
some other Unix operating systems , etc .
Apple a short time ago started to use Intel processors .

When quality of Intel processors are concerned , their potential is not exploited by Microsoft up to 1995 when Windows 95 appeared , ten years later because first Intel 386 capable of 32 bits appeared around 1985 .
In short , please do NOT think only in terms of Microsoft views and its operating systems .

Intel compilers and other programs is in my PC and I know them very well but I am not using them currently because I am working on my DBMS program . My Fortran programs are working on Windows . I do not have commercial Intel compilers in Windows but Linux free personal edition suite .
I can say that Intel is trying to promote its products and helping to people to utilize its capabilities . Obviously very good efforts .
This does NOT rescue the operating system developers to implement necessary facilities into their operating systems .

As I said trying to fill operating systems deficiencies is NOT a good practice .

Also parallel processing and distributed processing may be achieved distinct programs by message interchanging . This means that there is NO only ONE way .

In my opinion there are much more useful improvement points for FPC before multithreading and/or parallel processing .

This is a scheduling problem :
Which task should be performed first
when time and resources are concerned to
achieve greatest contribution ?

(B)

to Dear Florian Klaempfl ,

I am not saying that such facilities are useless or unnecessary . Our operating system views seems to be different . Scheduling of tasks on machine resources is a responsibility of operating system . When this is not done by a widely used operating system on PCs the people are trying to do it themselves .

When there is no such an operating system , what a compiler can do ? Tries to do everything by itself .

I do not know whether you studied IBM VM or not .
History of IBM VM and its features are very interesting .

Assume that there is a common data base for compiling a program . A cooperating process ( different program ) can write symbol table , etc . to that data base and other process ( different program ) can read from this data base and outputs its results into this data base or waits to completion of previous tasks . OR all of the compilation output may be shared through message passing with facilities supplied by the operating system
such as DDE ( Dynamic Data Exchange ) , etc.

It is just a transaction processing problem by multiple processors where processors being different computers or
different programs/CPUs within a computer .

Therefore , in my opinion there are many ways to achieve cooperative processing .

Thank you very much ,

Mehmet Erol Sanliturk

Hi Mehmet,

Actually you are missing the point. The problem is not the operating system. Lets assume for the moment the simple case of a 32 core system that apparently will be available sometime next year (Dual octal cores with HT). In case you are running multiple intensive processes you are utilizing probably well. If you are working on a workstation however and let say process video, you have 2 options. You can do what we do, and have the video processing split among multiple threads, or you can write a single processing thread, and have the compiler distribute the execution between the cores. Each approach has advantages and disadvantages. When dealing with video or audio processing you can spread the task among threads relatively easily, although you still will have a partial utilization, if you have the compiler to do the distribution you probably will achieve a better result. This is the same old RISC versus CISC argument. In modern RISC processors the compiler is responsible to handle the pipelining, in the classical CISC processor the pipeline is controlled by the hardware. Now the RISC processors ware not such a success due to other design problems, but Itanium as example uses the same compiler driven pipelining fairly successfully. You can argue that the pipeline has to be controlled by the processor the same way you can argue that the processor utilization should be OS driven, but there is always alternative approach with its own advantages and disadvantages. Compiler multicore utilization means that a single threaded application can utilize a multicore system. And with hundreds of cores already available from SUN and probably coming from Intel soon, we may have to consider this.

With best regards,
Boian Mitov

-------------------------------------------------------------------- Mitov Software
http://www.mitov.com
--------------------------------------------------------------------