Okay, I already published my question on a mailing list but it seems that all
MorphOS coders are busy so that no one did reply till now ( shame on you for
being busy, by the way ).
In order to implement "clock()" I need to know if it is possible to use the task
entries tc_Launch and tc_Switch under MorphOS 1.x and 2.x.
Unfortunately, I cannot lay my hands on the MorphOS programming
documentation (anyone with a clue where it can be found?) so I only have
the include files and there is stated that both entries are obsolete. If they are
obsolete, I guess that there is a new mechanism in order to get informed
when a task switch occurs? Unfortunately, because of the lacking
documentation, I don't know where to look. Can some one point me in the
right direction?

Thanks in advance.

We have special functions for that. No need to poke system structures.

NewGetTaskAttr() ... check <exec/task.h> or/and <exec/system.h>

I guess TASKINFOTYPE_AGETICKS or TASKINFOTIME_CPUTIME is what you want.
Not sure what clock() is doing exactly... or what do you want to use it
for; f.ex. for measuring time distances there are other ways too.




[ Edited by tokai on 2009/9/30 20:25 ]

Quote:

I have already feared that.


Quote:

Already checked before I did ask. Unfortunately, with no documentation
on my side it's not clear what they are good for.


Quote:

Thanks. Can you give me some more details about their functions? What
amount is returned and especially, what will be counted?
I already stumbled over TASKINFO_LAUNCHTIMETICKS and
SYSTEMINFO_LAUNCHTIMETICKS but I don't know what they are good
for.
I can only guess.


Quote:

clock_t clock ( void);

Returns the number of clock ticks elapsed since the program was launched.


With that, for example, a Delay(50) call would have no influence on the
value returned by clock(), because it really only takes into account the
time the task is up, running and especially performing something. In case
an asynchronous IO is made, it won't affect the returned value either.
With that one is able to determine the time consumed by the task by
ignoring the time which will be spent in devices (unless IOF_QUICK is set)
or in the DOS, where the task is sent asleep and at the earliest awakes
when the IO/packet is completed.

Normally clock() returns 50 ticks for an entire second but that value differs
from implementation to implementation. Using SAS/C for example,
TICKS_PER_SEC came to 1000.

So, my question is what TAG I should choose?


Thanks in advance.

TASKINFOTYPE_CPUTIME you want then. It's bascially the amount of cpu ticks your application has used. While AGETICKS would be the amount of cpu ticks since program start.


you get it like this:

UQUAD tmp;
NewGetTaskAttr(FindTask(NULL), &tmp, sizeof(tmp), TASKINFOTYPE_CPUTIME, TAG_DONE);


and to get the amount of CPU ticks per second you do (e.g. at start of program):

ULONG tmp2;
NewGetSystemAttr(&tmp2, sizeof(tmp2), SYSTEMINFOTYPE_TBCLOCKFREQUENCY, TAG_DONE);


the rest is simple math then ;)

UQUAD cputimeusedinseconds = (tmp / ((UQUAD)tmp2));


if you need a different scale you can divide tmp2 for your needs, i guess.


Hope that's what you need and it helps. ;)

Quote:

Okay, then I will use TASKINFOTYPE_CPUTIME.


Quote:

Thanks, examples are always welcome.


Quote:

Ouch, I am a computer, hence I cannot add one and one...


Quote:

Indeed. It will help me a lot! Thanks for your answers!

Hi,

one more question.

I need to count how many interruptions the Task-scheduler enacted for the
task in question.
That is, on other platforms I increase a counter everytime "tc_Switch()" is
called.

Is TASKINFOTYPE_HITCOUNT designed for it?

Thanks in advance.

No, you want this:

either total:
TASKINFOTYPE_VOLUNTARYCSW
TASKINFOTYPE_INVOLUNTARYCSW

or for the last second:
TASKINFOTYPE_LASTSECVOLUNTARYCSW
TASKINFOTYPE_LASTSECINVOLUNTARYCSW

CSW == context switch

Voluntary Context Switch?

May I asked who thinks so laterally?
Not in hundreds of years I would have associated VOLUNTARY_CSW with
performed task switches!
Even on Linux (I hate Linux, by the way) they use "context_swtch" or
something that sounds like this (can't remember well).
In addition, if I am right, voluntary and involuntary context switches mean
something different than a simple task switch, they are meant for the
outsourced work (giving up the processor explicitly (Wait() = voluntarily) and
by handing over work (IO requests = involuntarily)).
I can be mistaken but that's unlikely since I have a good memory (okay, it's an
oxymoron... ( ). :-}

Never mind, now I know what to do and that I urgently need the MorphOS
SDK-documentation for Exec!

Thanks for your answer and explanation.

Sending an I/O request is a voluntary context switch too, as Wait() is called in the context of your process. Unvoluntary switch happens when the task scheduler decides the time for your task is over for current period.

Quote:

My confusion started because I never encountered before the statement
'voluntary context switch' in relation with operating systems which don't
offer threads in the kernel.
In contrast to you and others, I did interpret 'voluntary' till now as a thread
that put itself explicitly to sleep, i.e. on MorphOS the closest I can think of
would be calling out of the program flow Wait(), but not by a device
driver, which calls Wait() for me in the context of my process. This I would
have equated with 'involuntary'.

Quote:

Round-robin.

But now I understand how they have been implemented for MorphOS.

Thanks.

Fascinating read... What are you cooking, Jolo?

It is merely a simple mini-project, "clock()".
Because I also need a special add-on which "clock()" doesn't offer, in order to
be able to do special measurements, I had to ask again. That's all.

Sorry to disappoint you.

Quote:

I'm not dissapointed (sorry to dissapoint you in turn ). It's very nice to read such in-depth discussions. Even if I don't understand them completely, it's indeed very nice to see this kind of feedback between a developer and a MorphOS team member.
When the question is interesting, the Team always answers.