Prev: A Jfet to use for switching?
Next: nexgen T-3 Fibs ProTrader Torrent - www.top-tradesoft.com
From: krw on 13 Feb 2010 11:24
On Fri, 12 Feb 2010 18:30:50 -0800 (PST), MooseFET
<kensmith(a)rahul.net> wrote:

>On Feb 12, 2:12�pm, krw <k...(a)att.bizzzzzzzzzzz> wrote:
>> On Thu, 11 Feb 2010 11:06:13 -0000, "Nial Stewart"
>>
>> <nial*REMOVE_TH...(a)nialstewartdevelopments.co.uk> wrote:
>> >> Quartus is worse than Maxplus-2 for the crashes per day.
>>
>> >0.0000002 rather than 0.00000015 ?
>>
>> >:-)
>>
>> >> It also doesn't produce the same resulting fuse map for the
>> >> same input file. This means that you can't recompile your old
>> >> code to make the same compiled results.
>>
>> >AFAIR it _should_ give repeatable results with the same input
>> >files and constraints (might be wrong here).
>>
>> I don't know why you would expect two versions of software, much less
>> two completely different packages, to create exactly the same bit map?
>> I wouldn't expect two passes of the *same* installation to create
>> exactly the same design. �There is too much magic going on in PAR.
>
>If the desired operation of the device is fully described by the input
>and the fitter gets the optimum fit, I would expect that 2 versions of
>the
>fitter would most likely make the same results. There is no reason
>for
>it to make different results.

Because there *is* more than one way correct to build the logic, there
is no expectation that there will only *be* only one way that the
logic gets built. Add in any optimization and forget the whole idea.
From: MooseFET on 13 Feb 2010 13:23
On Feb 13, 8:24 am, krw <k...(a)att.bizzzzzzzzzzz> wrote:
> On Fri, 12 Feb 2010 18:30:50 -0800 (PST), MooseFET
>
>
>
> <kensm...(a)rahul.net> wrote:
> >On Feb 12, 2:12 pm, krw <k...(a)att.bizzzzzzzzzzz> wrote:
> >> On Thu, 11 Feb 2010 11:06:13 -0000, "Nial Stewart"
>
> >> <nial*REMOVE_TH...(a)nialstewartdevelopments.co.uk> wrote:
> >> >> Quartus is worse than Maxplus-2 for the crashes per day.
>
> >> >0.0000002 rather than 0.00000015 ?
>
> >> >:-)
>
> >> >> It also doesn't produce the same resulting fuse map for the
> >> >> same input file. This means that you can't recompile your old
> >> >> code to make the same compiled results.
>
> >> >AFAIR it _should_ give repeatable results with the same input
> >> >files and constraints (might be wrong here).
>
> >> I don't know why you would expect two versions of software, much less
> >> two completely different packages, to create exactly the same bit map?
> >> I wouldn't expect two passes of the *same* installation to create
> >> exactly the same design.  There is too much magic going on in PAR.
>
> >If the desired operation of the device is fully described by the input
> >and the fitter gets the optimum fit, I would expect that 2 versions of
> >the
> >fitter would most likely make the same results.  There is no reason
> >for
> >it to make different results.
>
> Because there *is* more than one way correct to build the logic, there
> is no expectation that there will only *be* only one way that the
> logic gets built.  Add in any optimization and forget the whole idea.

Just because there could be more than one way for the fitter to make
the
gates to the described function, it doesn't mean that the fitter will
invent a different way from a previous version. MaxPlus-II went
through
several versions making the same results for the same inputs.
From: krw on 13 Feb 2010 13:42
On Sat, 13 Feb 2010 10:23:26 -0800 (PST), MooseFET
<kensmith(a)rahul.net> wrote:

>On Feb 13, 8:24�am, krw <k...(a)att.bizzzzzzzzzzz> wrote:
>> On Fri, 12 Feb 2010 18:30:50 -0800 (PST), MooseFET
>>
>>
>>
>> <kensm...(a)rahul.net> wrote:
>> >On Feb 12, 2:12�pm, krw <k...(a)att.bizzzzzzzzzzz> wrote:
>> >> On Thu, 11 Feb 2010 11:06:13 -0000, "Nial Stewart"
>>
>> >> <nial*REMOVE_TH...(a)nialstewartdevelopments.co.uk> wrote:
>> >> >> Quartus is worse than Maxplus-2 for the crashes per day.
>>
>> >> >0.0000002 rather than 0.00000015 ?
>>
>> >> >:-)
>>
>> >> >> It also doesn't produce the same resulting fuse map for the
>> >> >> same input file. This means that you can't recompile your old
>> >> >> code to make the same compiled results.
>>
>> >> >AFAIR it _should_ give repeatable results with the same input
>> >> >files and constraints (might be wrong here).
>>
>> >> I don't know why you would expect two versions of software, much less
>> >> two completely different packages, to create exactly the same bit map?
>> >> I wouldn't expect two passes of the *same* installation to create
>> >> exactly the same design. �There is too much magic going on in PAR.
>>
>> >If the desired operation of the device is fully described by the input
>> >and the fitter gets the optimum fit, I would expect that 2 versions of
>> >the
>> >fitter would most likely make the same results. �There is no reason
>> >for
>> >it to make different results.
>>
>> Because there *is* more than one way correct to build the logic, there
>> is no expectation that there will only *be* only one way that the
>> logic gets built. �Add in any optimization and forget the whole idea.
>
>Just because there could be more than one way for the fitter to make
>the
>gates to the described function, it doesn't mean that the fitter will
>invent a different way from a previous version. MaxPlus-II went
>through
>several versions making the same results for the same inputs.

Try reading. If it can be made different ways there is NO EXPECTATION
that it will always be made the same way. Algorithms change. Got it?
From: JosephKK on 14 Feb 2010 23:34
On Sat, 13 Feb 2010 10:23:26 -0800 (PST), MooseFET <kensmith(a)rahul.net> wrote:

>On Feb 13, 8:24 am, krw <k...(a)att.bizzzzzzzzzzz> wrote:
>> On Fri, 12 Feb 2010 18:30:50 -0800 (PST), MooseFET
>>
>>
>>
>> <kensm...(a)rahul.net> wrote:
>> >On Feb 12, 2:12 pm, krw <k...(a)att.bizzzzzzzzzzz> wrote:
>> >> On Thu, 11 Feb 2010 11:06:13 -0000, "Nial Stewart"
>>
>> >> <nial*REMOVE_TH...(a)nialstewartdevelopments.co.uk> wrote:
>> >> >> Quartus is worse than Maxplus-2 for the crashes per day.
>>
>> >> >0.0000002 rather than 0.00000015 ?
>>
>> >> >:-)
>>
>> >> >> It also doesn't produce the same resulting fuse map for the
>> >> >> same input file. This means that you can't recompile your old
>> >> >> code to make the same compiled results.
>>
>> >> >AFAIR it _should_ give repeatable results with the same input
>> >> >files and constraints (might be wrong here).
>>
>> >> I don't know why you would expect two versions of software, much less
>> >> two completely different packages, to create exactly the same bit map?
>> >> I wouldn't expect two passes of the *same* installation to create
>> >> exactly the same design.  There is too much magic going on in PAR.
>>
>> >If the desired operation of the device is fully described by the input
>> >and the fitter gets the optimum fit, I would expect that 2 versions of
>> >the
>> >fitter would most likely make the same results.  There is no reason
>> >for
>> >it to make different results.
>>
>> Because there *is* more than one way correct to build the logic, there
>> is no expectation that there will only *be* only one way that the
>> logic gets built.  Add in any optimization and forget the whole idea.
>
>Just because there could be more than one way for the fitter to make
>the
>gates to the described function, it doesn't mean that the fitter will
>invent a different way from a previous version. MaxPlus-II went
>through
>several versions making the same results for the same inputs.

The results of fitter processes is intrinsically probabilistic,
precisely because of the stochastic processes used to reach a
locally maximally good fit. Thus getting the same result is definitely
a maybe proposition.
From: MooseFET on 15 Feb 2010 11:37
On Feb 14, 8:34 pm, "JosephKK"<quiettechb...(a)yahoo.com> wrote:
> On Sat, 13 Feb 2010 10:23:26 -0800 (PST), MooseFET <kensm...(a)rahul.net> wrote:
> >On Feb 13, 8:24 am, krw <k...(a)att.bizzzzzzzzzzz> wrote:
> >> On Fri, 12 Feb 2010 18:30:50 -0800 (PST), MooseFET
>
> >> <kensm...(a)rahul.net> wrote:
> >> >On Feb 12, 2:12 pm, krw <k...(a)att.bizzzzzzzzzzz> wrote:
> >> >> On Thu, 11 Feb 2010 11:06:13 -0000, "Nial Stewart"
>
> >> >> <nial*REMOVE_TH...(a)nialstewartdevelopments.co.uk> wrote:
> >> >> >> Quartus is worse than Maxplus-2 for the crashes per day.
>
> >> >> >0.0000002 rather than 0.00000015 ?
>
> >> >> >:-)
>
> >> >> >> It also doesn't produce the same resulting fuse map for the
> >> >> >> same input file. This means that you can't recompile your old
> >> >> >> code to make the same compiled results.
>
> >> >> >AFAIR it _should_ give repeatable results with the same input
> >> >> >files and constraints (might be wrong here).
>
> >> >> I don't know why you would expect two versions of software, much less
> >> >> two completely different packages, to create exactly the same bit map?
> >> >> I wouldn't expect two passes of the *same* installation to create
> >> >> exactly the same design.  There is too much magic going on in PAR..
>
> >> >If the desired operation of the device is fully described by the input
> >> >and the fitter gets the optimum fit, I would expect that 2 versions of
> >> >the
> >> >fitter would most likely make the same results.  There is no reason
> >> >for
> >> >it to make different results.
>
> >> Because there *is* more than one way correct to build the logic, there
> >> is no expectation that there will only *be* only one way that the
> >> logic gets built.  Add in any optimization and forget the whole idea..
>
> >Just because there could be more than one way for the fitter to make
> >the
> >gates to the described function, it doesn't mean that the fitter will
> >invent a different way from a previous version.  MaxPlus-II went
> >through
> >several versions making the same results for the same inputs.
>
> The results of fitter processes is intrinsically probabilistic,
> precisely because of the stochastic processes used to reach a
> locally maximally good fit.  Thus getting the same result is definitely
> a maybe proposition.

The "random" function at work is a pseudo-random generator or the like
that gets seeded as the program starts. Chances are the "random"
values
really are 1,2,3,4 because with the internals of the Altera part, the
resources are in a nearly uniform pattern. The "random" just needs to
be such that it doesn't try equivalent solutions over and over before
stepping over to a different one.

The Maxplus-II fitter makes the same output from run to run and from
version to version. I did quite a few projects with the MAX7000
series
and had to maintain them over periods of decades. Altera still makes
the
5V parts as a MAX7xxxx-S part but they don't even show them as options
for designing in. These products will use those parts to the day they
are dead.

My latest project has a large number of micro controllers in it but no
CPLDs. Just about everything is done with standard logic connected to
a micro and a multiple DAC chip.
First  |  Prev  |  Next  |  Last
Pages: 1 2 3 4
Prev: A Jfet to use for switching?
Next: nexgen T-3 Fibs ProTrader Torrent - www.top-tradesoft.com