[ECOS] # ticks per second?

[ECOS] # ticks per second?

[Andreas.Karlsson]

Hi,



Is this rows correct to obtain the number of ticks per second? I want some

timers to go off after some seconds.

/Andreas



cyg_resolution_t clock_res;

cyg_handle_t clock_handle;

cyg_uint32 sec;



clock_handle=cyg_real_time_clock();

clock_res=cyg_clock_get_resolution(clock_handle); 

sec=(clock_res.divisor/clock_res.dividend)*1000000000; //dividend/divisor?

Re: [ECOS] # ticks per second?

[Jonathan Larmour]

Andreas.Karlsson@combitechsystems.com wrote:
> 
> Is this rows correct to obtain the number of ticks per second? I want some
> 
> timers to go off after some seconds.
> 
> /Andreas
> 
> cyg_resolution_t clock_res;
> 
> cyg_handle_t clock_handle;
> 
> cyg_uint32 sec;
> 
> clock_handle=cyg_real_time_clock();
> 
> clock_res=cyg_clock_get_resolution(clock_handle);
> 
> sec=(clock_res.divisor/clock_res.dividend)*1000000000; //dividend/divisor?

Yes, although you run the risk of underflow: divisor would typically be
something like 100 whereas dividend would be 1E9. So try using:

unsigned long long lltmp;
tmp = (clock_res.divisor*1000000000)/clock_res.dividend;
sec = tmp;

Jifl
-- 
Red Hat, 35 Cambridge Place, Cambridge, UK. CB2 1NS  Tel: +44 (1223) 728762
"Plan to be spontaneous tomorrow."  ||  These opinions are all my own fault

Re: [ECOS] # ticks per second?

[Hugo 'NOx' Tyson]

Jonathan Larmour <jlarmour@redhat.com> writes:

> Andreas.Karlsson@combitechsystems.com wrote:
> > 
> > Is this rows correct to obtain the number of ticks per second? I want some
> > 
> > timers to go off after some seconds.
> > 
> > /Andreas
> > 
> > cyg_resolution_t clock_res;
> > 
> > cyg_handle_t clock_handle;
> > 
> > cyg_uint32 sec;
> > 
> > clock_handle=cyg_real_time_clock();
> > 
> > clock_res=cyg_clock_get_resolution(clock_handle);
> > 
> > sec=(clock_res.divisor/clock_res.dividend)*1000000000; //dividend/divisor?
> 
> Yes, although you run the risk of underflow: divisor would typically be
> something like 100 whereas dividend would be 1E9. So try using:
> 
> unsigned long long lltmp;
> tmp = (clock_res.divisor*1000000000)/clock_res.dividend;
> sec = tmp;

And there is a C++ solution to this within the kernel.  See the test
clockcnv.cxx in the kernel tests directory for how to use it, and clock.cxx
in the kernel source for its implementation.

It reduces and factorizes the big numbers so that instead of doing

	K * DIVISOR * 10^N / DIVIDEND

or whatever, it does

	K * A / C * B / D

where ABCD are chosen such that

	A * B    DIVISOR * 10^N
	----- == --------------
	C * D       DIVIDEND

thus making it all far less likely to underflow or lose precision.

It doesn't (yet) have a KAPI (C) API.  It's used within the uITRON
implementation to get milliSecond clocks, which is bare C++.

Feel free to design and contribute a KAPI API?

HTH
	- Huge

Re: [ECOS] # ticks per second?

[Grant Edwards]

On Wed, Aug 30, 2000 at 06:41:35PM +0100, Hugo 'NOx' Tyson wrote:

> And there is a C++ solution to this within the kernel.  See the test
> clockcnv.cxx in the kernel tests directory for how to use it, and clock.cxx
> in the kernel source for its implementation.
[...]
> thus making it all far less likely to underflow or lose precision.
> 
> It doesn't (yet) have a KAPI (C) API.  It's used within the uITRON
> implementation to get milliSecond clocks, which is bare C++.
> 
> Feel free to design and contribute a KAPI API?

Since it's not possible to change the clock tick period at
run-time (right?), I'd try for a pre-processor constant.  But
that might be pretty hard to do in an accurate, general way.

-- 
Grant Edwards
grante@visi.com

Re: [ECOS] # ticks per second?

[Hugo 'NOx' Tyson]

Grant Edwards <grante@visi.com> writes:
> On Wed, Aug 30, 2000 at 06:41:35PM +0100, Hugo 'NOx' Tyson wrote:
> > And there is a C++ solution to this within the kernel.  See the test
> > clockcnv.cxx in the kernel tests directory for how to use it, and clock.cxx
> > in the kernel source for its implementation.
> [...]
> > thus making it all far less likely to underflow or lose precision.
> > 
> > It doesn't (yet) have a KAPI (C) API.  It's used within the uITRON
> > implementation to get milliSecond clocks, which is bare C++.
> > 
> > Feel free to design and contribute a KAPI API?
> 
> Since it's not possible to change the clock tick period at
> run-time (right?), I'd try for a pre-processor constant.  But
> that might be pretty hard to do in an accurate, general way.

Precisely!

The thing within the kernel will create for you a "converter" object to or
from "system clocks" from or to any unit of your choice - you then use the
converter object to change eg. N seconds into M system ticks.  Or vice
versa, whatever...

You're right, any application would only create a very small number of
these of these during startup, but the calculation involved is way beyond
the preprocessor and so forth.  So to generalize fully, it's done at
runtime.  (This is one of the very very few examples of stuff I feel it's
OK to do at runtime, even though the result is the same every time - and it
also makes it easier to test properly)

	- Huge