Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > sci.physics > #532800 > unrolled thread

Floating Point Enigma - Please Solve

Back to article view | Back to sci.physics

Contents

  Floating Point Enigma - Please Solve Fabian Russell <root@localhost.localdomain> - 2015-11-13 19:43 +0000
    Re: Floating Point Enigma - Please Solve jimp@specsol.spam.sux.com - 2015-11-13 19:59 +0000
      Re: Floating Point Enigma - Please Solve Fabian Russell <root@localhost.localdomain> - 2015-11-13 20:16 +0000
        Re: Floating Point Enigma - Please Solve jimp@specsol.spam.sux.com - 2015-11-13 21:02 +0000
          Re: Floating Point Enigma - Please Solve Fabian Russell <root@localhost.localdomain> - 2015-11-13 22:32 +0000
            Re: Floating Point Enigma - Please Solve jimp@specsol.spam.sux.com - 2015-11-13 23:32 +0000
              Re: Floating Point Enigma - Please Solve Fabian Russell <root@localhost.localdomain> - 2015-11-13 23:52 +0000
                Re: Floating Point Enigma - Please Solve jimp@specsol.spam.sux.com - 2015-11-14 00:06 +0000
    Re: Floating Point Enigma - Please Solve Fabian Russell <root@localhost.localdomain> - 2015-11-13 21:14 +0000
      Re: Floating Point Enigma - Please Solve noTthaTguY <abu.kuanysh05@gmail.com> - 2015-11-13 13:18 -0800
      Re: Floating Point Enigma - Please Solve jimp@specsol.spam.sux.com - 2015-11-13 23:33 +0000
    Re: Floating Point Enigma - Please Solve Double-A <double-a3@hush.com> - 2015-11-13 13:33 -0800
      Re: Floating Point Enigma - Please Solve Fabian Russell <root@localhost.localdomain> - 2015-11-13 22:04 +0000
        Re: Floating Point Enigma - Please Solve jimp@specsol.spam.sux.com - 2015-11-13 23:35 +0000
          Re: Floating Point Enigma - Please Solve Fabian Russell <root@localhost.localdomain> - 2015-11-13 23:54 +0000
            Re: Floating Point Enigma - Please Solve jimp@specsol.spam.sux.com - 2015-11-14 00:15 +0000
    Re: IEEE754 compliance is the _default_. Fabian Russell <root@localhost.localdomain> - 2015-11-14 02:08 +0000
      Re: IEEE754 compliance is the _default_. GreyCloud <cumulus@mist.com> - 2015-11-13 19:30 -0700

#532800 — Floating Point Enigma - Please Solve

Greetings physics enthusiasts,

To all those wise-ass dumb-fscks who adamantly claim that a knowledge
of floating point operations is superfluous, this one is for youse (you all).

You, and a lab down the hall, are both investigating a new algorithm
for simulating an important phenomenon.  You both have identical
machines and systems and are both compiling the software with GCC
(but any compiler will behave the same).

You are compiling with the fast-math option, because you are a wise-ass
dumb-fsck and you believe that only computational speed is important.
The lab down the hall is using options for IEEE754 compliance.

You are experiencing a lot of divide-by-zero exceptions and other strange
problems that is halting progress.  The lab down the hall is doing very
well with no issues.

Someone suggests to contact F. Russell, the world renowned authority, for
his advice.  F. Russell, however, tells you all to go and pound sand because
he does not accept wise-ass dumb-fscks as clients.  (You can file an
anti-discrimination suit but wise-ass dumb-fscks are not protected.)

So what do you do?  What is the source of the problem and how do you
fix it?

You all got big fat mouths, so use 'em.

Ha, ha, ha, ha, ha, ha!  There ain't no one gonna get the EASY EASY EASY
solution!

Note: the fictitious scenario depicted above is not all outlandish
or far-fetched and does represent a possibly very common occurrence.

[toc] | [next] | [standalone]

#532804

Fabian Russell <root@localhost.localdomain> wrote:
> Greetings physics enthusiasts,
> 
> To all those wise-ass dumb-fscks who adamantly claim that a knowledge
> of floating point operations is superfluous, this one is for youse (you all).

I don't think anyone other than you has said any such thing.

Rather, most people have said that your "interest" in floating point
operations is obsessive to say the least.

<snip>

> You are experiencing a lot of divide-by-zero exceptions and other strange
> problems that is halting progress.  The lab down the hall is doing very
> well with no issues.

Never in 43 years of programming have I seen such a problem due to lack
of understanding of IEEE754.

I have seen a LOT of such problems due to a lack of understanding of
memory management and use of pointers.

I have seen exactly one problem due to the lack of understanding of
the resolution limits for floating point arithmatic.

<snip>
 
> Note: the fictitious scenario depicted above is not all outlandish
> or far-fetched and does represent a possibly very common occurrence.

No, it does not by any stretch of the imagination represent a possibly
very common occurrence.


-- 
Jim Pennino

[toc] | [prev] | [next] | [standalone]

#532808

On Fri, 13 Nov 2015 19:59:52 +0000, jimp wrote:

> 
> I don't think anyone other than you has said any such thing.
> 

Judas Priest!  You need serious help and I'm not kidding.

This statement is an example of what psychoneurolgists would
call "confabulation."  It means, for one thing, remembering things
that did not happen.

I am not kidding.  GET HELP FAST.


> 
> Never in 43 years of programming have I seen such a problem due to lack
> of understanding of IEEE754.
> 

IEEE754 has not been around for 43 years.

This probably contributes heavily to you opposition.  You acquired
your formative experience well before the standard was established
and consequently have had a difficult time adapting.


> 
> No, it does not by any stretch of the imagination represent a possibly
> very common occurrence.

Yes, it does.

There are two important clues given, however.  Anyone who has remained
reasonably -- not obsessively -- up to date should be able to pick them
up.

[toc] | [prev] | [next] | [standalone]

#532828

Fabian Russell <root@localhost.localdomain> wrote:
> On Fri, 13 Nov 2015 19:59:52 +0000, jimp wrote:
> 
>> 
>> I don't think anyone other than you has said any such thing.
>> 
> 
> Judas Priest!  You need serious help and I'm not kidding.
> 
> This statement is an example of what psychoneurolgists would
> call "confabulation."  It means, for one thing, remembering things
> that did not happen.
> 
> I am not kidding.  GET HELP FAST.

It seems the problem is you have become obsessed with IEEE754 and can
not understand why the rest of the world is not obsessed with it also.

In addition, you seem to be stuck in a 1970's mindset.
 
>> Never in 43 years of programming have I seen such a problem due to lack
>> of understanding of IEEE754.
>> 
> 
> IEEE754 has not been around for 43 years.

Irrelevant.

> This probably contributes heavily to you opposition.  You acquired
> your formative experience well before the standard was established
> and consequently have had a difficult time adapting.
> 
> 
>> 
>> No, it does not by any stretch of the imagination represent a possibly
>> very common occurrence.
> 
> Yes, it does.

No, it does not.

In 2015 anyone analyzing data is most likely going to use a canned
application, such as a spreadsheet to do statistical analysis or
Mathcad to do computational analysis.

It is highly unlikely anyone is going to write their own software in
any general purpose programming language.

I can remember spending weeks in the 70's writting my own statistical
analyisis programs and days getting the 132 column line printer
graphing routines to work using a BASIC like language.

Today the whole thing would have been done and on the laser printer
in a few hours with a spreadsheet.

For people other than those writting applications such as Mathcad, the
computer is an appliance and knowledge of IEEE754 is not required nor
is it of any importance to them.

-- 
Jim Pennino

[toc] | [prev] | [next] | [standalone]

#532867

On Fri, 13 Nov 2015 21:02:28 +0000, jimp wrote:

> 
> the
> computer is an appliance and knowledge of IEEE754 is not required nor
> is it of any importance to them.
>

Translation (not courtesy of BabelFish):

I am just another lazy-fsck that would rather watch weenie videos
on YouTube and not be bothered by silly details the mastery of
which would certainly further my professional development.

[toc] | [prev] | [next] | [standalone]

#532890

Fabian Russell <root@localhost.localdomain> wrote:
> On Fri, 13 Nov 2015 21:02:28 +0000, jimp wrote:
> 
>> 
>> the
>> computer is an appliance and knowledge of IEEE754 is not required nor
>> is it of any importance to them.
>>
> 
> Translation (not courtesy of BabelFish):
> 
> I am just another lazy-fsck that would rather watch weenie videos
> on YouTube and not be bothered by silly details the mastery of
> which would certainly further my professional development.

While you are stuck in a 1970's mindset, I get things done in the most
efficient method possible, which usually means using applications, which
is an attribute that all my clients value highly.

If I stuck to a 1970's mindset like yours, I would have been out of
business decades ago.

The use of applications no more requires an understanding of IEEE754
than fueling your car requires an understanding of ASTM D4814-15a.

 

-- 
Jim Pennino

[toc] | [prev] | [next] | [standalone]

#532895

On Fri, 13 Nov 2015 23:32:00 +0000, jimp wrote:

> 
>  ASTM D4814-15a.
>

Hey wow!  That sounds interesting.  Where can I look it up?

(No, I'm not jesting.)

Ha, ha, ha, ha, ha, ha!

Some minds are _________ than others.

You fill in the blank.

[toc] | [prev] | [next] | [standalone]

#532902

Fabian Russell <root@localhost.localdomain> wrote:
> On Fri, 13 Nov 2015 23:32:00 +0000, jimp wrote:
> 
>> 
>>  ASTM D4814-15a.
>>
> 
> Hey wow!  That sounds interesting.  Where can I look it up?
> 
> (No, I'm not jesting.)

If you weren't stuck in the 1970's it would be obvious.


-- 
Jim Pennino

[toc] | [prev] | [next] | [standalone]

#532826

On Fri, 13 Nov 2015 19:43:32 +0000, Fabian Russell wrote:

> 
> To all those wise-ass dumb-fscks who adamantly claim that a knowledge
> of floating point operations is superfluous, this one is for youse (you all).
> 

In a million billion years, none of the physics "experts" on this
group will be able to solve this EASY EASY EASY problem.

Therefore, I will reveal the SIMPLE SIMPLE SIMPLE answer.

Using the fast-math option for speed, which ALL compilers will provide,
has the result of disabling DENORMAL floating point numbers and
invoking the FLUSH-TO-ZERO option.

What is a denormal floating point?  If you don't already know (and you
should know), then look it up.  I don't have the time to explain.

Denormal FP's are a part of the IEEE754 standard and without them
a relatively large gap, or interval, is created around zero.  Within
this gap, every non-zero representable quantity becomes equal
to zero. 

Without denormals, any algorithm that produces a small non-zero
numerical value will have that value automatically "flushed" to zero
by the FP processor.  As a consequence, divide-by-zero exceptions
can occur, and depending on the nature of the algorithm they can
occur frequently.

Aside: this is but one example of some of the many problems
that could arise.

Ha, ha, ha, ha, ha, ha!  The seasoned "experts" on this group,
when it comes to modern computational physics, are like blind
men stumbling around with their white canes and sunglasses.

Now go back to your smart phones, YouTube, and physorg,org
and pretend to be enlightened.

Ha, ha, ha, ha, ha, ha, ha, ha!

[toc] | [prev] | [next] | [standalone]

#532830

that would be minus or/and plus a certain fraction of one

> Denormal FP's are a part of the IEEE754 standard and without them
> a relatively large gap, or interval, is created around zero.  Within
> this gap, every non-zero representable quantity becomes equal
> to zero. 
> 
> Without denormals, any algorithm that produces a small non-zero
> numerical value will have that value automatically "flushed" to zero
> by the FP processor.  As a consequence, divide-by-zero exceptions
> can occur, and depending on the nature of the algorithm they can
> occur frequently.
> 
> Aside: this is but one example of some of the many problems
> that could arise.
> 
> Ha, ha, ha, ha, ha, ha!  The seasoned "experts" on this group,
> when it comes to modern computational physics, are like blind
> men stumbling around with their white canes and sunglasses.
> 
> Now go back to your smart phones, YouTube, and physorg,org
> and pretend to be enlightened.
> 
> Ha, ha, ha, ha, ha, ha, ha, ha!

[toc] | [prev] | [next] | [standalone]

#532891

Fabian Russell <root@localhost.localdomain> wrote:
> On Fri, 13 Nov 2015 19:43:32 +0000, Fabian Russell wrote:
> 
>> 
>> To all those wise-ass dumb-fscks who adamantly claim that a knowledge
>> of floating point operations is superfluous, this one is for youse (you all).
>> 
> 
> In a million billion years, none of the physics "experts" on this
> group will be able to solve this EASY EASY EASY problem.

Nor would they have any need to do so as they would be using a canned
application to solve their problems.


-- 
Jim Pennino

[toc] | [prev] | [next] | [standalone]

#532837

On Friday, November 13, 2015 at 11:44:29 AM UTC-8, Fabian Russl wrote:
> Greetings physics enthusiasts,
> 
> To all those wise-ass dumb-fscks who adamantly claim that a knowledge
> of floating point operations is superfluous, this one is for youse (you all).
> 
> You, and a lab down the hall, are both investigating a new algorithm
> for simulating an important phenomenon.  You both have identical
> machines and systems and are both compiling the software with GCC
> (but any compiler will behave the same).
> 
> You are compiling with the fast-math option, because you are a wise-ass
> dumb-fsck and you believe that only computational speed is important.
> The lab down the hall is using options for IEEE754 compliance.
> 
> You are experiencing a lot of divide-by-zero exceptions and other strange
> problems that is halting progress.  The lab down the hall is doing very
> well with no issues.
> 
> Someone suggests to contact F. Russell, the world renowned authority, for
> his advice.  F. Russell, however, tells you all to go and pound sand because
> he does not accept wise-ass dumb-fscks as clients.  (You can file an
> anti-discrimination suit but wise-ass dumb-fscks are not protected.)
> 
> So what do you do?  What is the source of the problem and how do you
> fix it?
> 
> You all got big fat mouths, so use 'em.
> 
> Ha, ha, ha, ha, ha, ha!  There ain't no one gonna get the EASY EASY EASY
> solution!
> 
> Note: the fictitious scenario depicted above is not all outlandish
> or far-fetched and does represent a possibly very common occurrence.



I wrote an amortization application in Quick Basic one time.  I used their floating point feature to get the precision I required.  However I ended up having to write a lot of editing subroutines to clean up underflow and overflow results and other results simply not acceptable for accounting applications.  Had to do my own rounding as well.  Also I noticed other amortization schedule applications being sold would occasionally output one penny errors in their schedules.  These could add up larger errors in the totals.  This was because of not screening the floating point results for rounding errors.  Some bought my application just for that reason.  

Double-A

[toc] | [prev] | [next] | [standalone]

#532854

On Fri, 13 Nov 2015 13:33:45 -0800, Double-A wrote:

> 
> 
> I wrote an amortization application in Quick Basic ...
> Also I noticed other amortization schedule applications
> being sold would occasionally output one penny errors
> in their schedules.  These could add up larger errors
> in the totals.
>

Amortization is a good example because it involves interest
rate calculations, and these calculations are notorious
for CATASTROPHIC CANCELLATION.

Ha, ha!  The wise-ass dumb-fscks on this group wouldn't
know catastrophic cancellation from a hole in the ground.

Usually, the precision of double floating point is sufficient
to nullify somewhat the errors resulting from catastrophic
cancellation but certainly not always.  Caution is always
advised.

The C programming language has a special function, called
log1p(), that is specifically designed for safe financial
interest, and other, calculations.

But don't ask the wise-ass dumb-fscks on this group.  They'll
claim that it's all nonsense and those lost pennies will keep
adding up.

[toc] | [prev] | [next] | [standalone]

#532892

Fabian Russell <root@localhost.localdomain> wrote:
> On Fri, 13 Nov 2015 13:33:45 -0800, Double-A wrote:
> 
>> 
>> 
>> I wrote an amortization application in Quick Basic ...
>> Also I noticed other amortization schedule applications
>> being sold would occasionally output one penny errors
>> in their schedules.  These could add up larger errors
>> in the totals.
>>
> 
> Amortization is a good example because it involves interest
> rate calculations, and these calculations are notorious
> for CATASTROPHIC CANCELLATION.

There is a standard for doing financial calculations and it is not
an IEEE standard.


-- 
Jim Pennino

[toc] | [prev] | [next] | [standalone]

#532896

On Fri, 13 Nov 2015 23:35:57 +0000, jimp wrote:

> 
> There is a standard for doing financial calculations and it is not
> an IEEE standard.
>

Give it a rest and go bed.  You are so befuddled you don't know
west from up.

[toc] | [prev] | [next] | [standalone]

#532903

Fabian Russell <root@localhost.localdomain> wrote:
> On Fri, 13 Nov 2015 23:35:57 +0000, jimp wrote:
> 
>> 
>> There is a standard for doing financial calculations and it is not
>> an IEEE standard.
>>
> 
> Give it a rest and go bed.  You are so befuddled you don't know
> west from up.

Says the babbling idiot that believes one needs to understand IEEE
standards to run canned applications and that real financial
applications use floating point.

 

-- 
Jim Pennino

[toc] | [prev] | [next] | [standalone]

#532944 — Re: IEEE754 compliance is the _default_.

On Fri, 13 Nov 2015 16:30:20 -0800, Jeff-Relf.Me wrote:

>
> [snip supreme shit]
>

Go take a long walk in the nearest ghetto slum, Jeffie boy.

Nature will surely take its course.

Ha, ha, ha, ha, ha, ha!

 

[toc] | [prev] | [next] | [standalone]

#532951 — Re: IEEE754 compliance is the _default_.

On 11/13/15 19:08, Fabian Russell wrote:
> On Fri, 13 Nov 2015 16:30:20 -0800, Jeff-Relf.Me wrote:
>
>>
>> [snip supreme shit]
>>
>
> Go take a long walk in the nearest ghetto slum, Jeffie boy.
>
> Nature will surely take its course.
>
> Ha, ha, ha, ha, ha, ha!
>
>
Guffaw!!!  Seems that you are way behind.  When was the last job you 
held?  Back around 1975 or so?


-- 
My problem is that I don't have enough middle fingers.

[toc] | [prev] | [standalone]

Back to top | Article view | sci.physics

csiph-web