Board index » delphi » MATH387 speeds up code

MATH387 speeds up code

If you ever write any code in Borland Delphi which does
trig calculations  - typically scientific, engineering or
statistical work  - and you need exponentiation (x^y), or any of
the trig functions sin, cos, tan, atan, ln  - then try the MATH387
unit. It dramatically speeds up code by using the built-in
features of 80387 or better co-processors (or 486DX/Pentium)
- which Borland's code does not.

To get your copy of MATH387, check out the "Number Crunching
Page" at:

  http://www.geocities.com/SiliconValley/2787

Try the demo program there  - it will graphically show you what
kind of speed improvement you can expect on your particular
machine.

The Number Crunching Page also gives you a delightful walk down
memory lane by tracing the history of floating point performance
on desk-top computers over the last 20 years. It shows how a
problem that took 12 hours to run in 1976 on a Z80 machine will
now run in about 1 SECOND on a Pentium machine - at a lower
price!

 

Re:MATH387 speeds up code


In article <4et9s8$...@news.asu.edu>,
   dmit...@phyast.la.asu.edu (Dmitry Streblechenko) wrote:
]-Are you guys serious? $20 for less than a dozen
]-inline routines which everyone can write in
]-20 minutes using any textbook on 80386 assembler???

I dunno, I've made a lot of money doing easy stuff that
other people were either too lazy or too ignorant to do
for themselves.  Hey, look at the success of the Jiffy
Lube franchise.  Changing the oil in your automobile
is not rocket science, but plenty of people gladly fork
over the $20 (or whatever it is) for the pleasure of
not having to do it themselves.

the problem with this particular offering is that there's
a freeware package available that does pretty much the
same stuff.  in fact it probably does more, as the
author also includes units for complex numbers and for
hyperbolic functions.  all code is included too.  look for
MathLib2.zip on the Delphi Super Page.  good stuff.

Mark Vaughan

]-
]-In article <4erojj$...@hawk.pix.za>,
]-   chris.willi...@pixie.co.za (Chris Williams) wrote:
]->If you ever write any code in Borland Delphi which does
]->trig calculations  - typically scientific, engineering or
]->statistical work  - and you need exponentiation (x^y), or any of
]->the trig functions sin, cos, tan, atan, ln  - then try the MATH387
]->unit. It dramatically speeds up code by using the built-in
]->features of 80387 or better co-processors (or 486DX/Pentium)
]->- which Borland's code does not.
]->
]->To get your copy of MATH387, check out the "Number Crunching
]->Page" at:
]->
]->  http://www.geocities.com/SiliconValley/2787
]->
]->Try the demo program there  - it will graphically show you what
]->kind of speed improvement you can expect on your particular
]->machine.
]->
]->The Number Crunching Page also gives you a delightful walk down
]->memory lane by tracing the history of floating point performance
]->on desk-top computers over the last 20 years. It shows how a
]->problem that took 12 hours to run in 1976 on a Z80 machine will
]->now run in about 1 SECOND on a Pentium machine - at a lower
]->price!
]->
]->

Re:MATH387 speeds up code


Are you guys serious? $20 for less than a dozen
inline routines which everyone can write in
20 minutes using any textbook on 80386 assembler???

In article <4erojj$...@hawk.pix.za>,
   chris.willi...@pixie.co.za (Chris Williams) wrote:

Quote
>If you ever write any code in Borland Delphi which does
>trig calculations  - typically scientific, engineering or
>statistical work  - and you need exponentiation (x^y), or any of
>the trig functions sin, cos, tan, atan, ln  - then try the MATH387
>unit. It dramatically speeds up code by using the built-in
>features of 80387 or better co-processors (or 486DX/Pentium)
>- which Borland's code does not.

>To get your copy of MATH387, check out the "Number Crunching
>Page" at:

>  http://www.geocities.com/SiliconValley/2787

>Try the demo program there  - it will graphically show you what
>kind of speed improvement you can expect on your particular
>machine.

>The Number Crunching Page also gives you a delightful walk down
>memory lane by tracing the history of floating point performance
>on desk-top computers over the last 20 years. It shows how a
>problem that took 12 hours to run in 1976 on a Z80 machine will
>now run in about 1 SECOND on a Pentium machine - at a lower
>price!

Re:MATH387 speeds up code


On 2 Feb 1996, Dmitry Streblechenko wrote:

Quote
>Are you guys serious? $20 for less than a dozen
>inline routines which everyone can write in
>20 minutes using any textbook on 80386 assembler???

Everyone can write???? In 20 minutes???? I'm sure there are hardly any
Pascal programmers out there who write mathematical programs who know,
or even want to know, anything about assembler - let alone embedding
inline 387 code. MATH387 is there to make their lives easier - it is
just as simple to code trig functions using MATH387 as without it, but
it is MUCH faster (and significantly faster than some of the other
"competition", e.g. MATHLIB2).

About the price: point taken. I don't live in the US, so I have
difficulty conceptualising a "value" in dollars. $20 just seems to be
the norm for many, many shareware programs out there - small and
large. Besides, for most people, the value has more to do with
usefulness than how many routines are provided. My gage of usefulness
for MATH387 is the absolute fastest possible performance - bar
nothing.

What price do you think would be reasonable?

(BTW, apologies for the rather late posting. Our @#$% news server is
giving lots of trouble)

Quote
>In article <4erojj$...@hawk.pix.za>,
>   chris.willi...@pixie.co.za (Chris Williams) wrote:
>>If you ever write any code in Borland Delphi which does
>>trig calculations  - typically scientific, engineering or
>>statistical work  - and you need exponentiation (x^y), or any of
>>the trig functions sin, cos, tan, atan, ln  - then try the MATH387
>>unit. It dramatically speeds up code by using the built-in
>>features of 80387 or better co-processors (or 486DX/Pentium)
>>- which Borland's code does not.

>>To get your copy of MATH387, check out the "Number Crunching
>>Page" at:

>>  http://www.geocities.com/SiliconValley/2787

>>Try the demo program there  - it will graphically show you what
>>kind of speed improvement you can expect on your particular
>>machine.

>>The Number Crunching Page also gives you a delightful walk down
>>memory lane by tracing the history of floating point performance
>>on desk-top computers over the last 20 years. It shows how a
>>problem that took 12 hours to run in 1976 on a Z80 machine will
>>now run in about 1 SECOND on a Pentium machine - at a lower
>>price!

Chris Williams       chris.willi...@pixie.co.za
-----------------------------------------------
Want to speed up trig functions in your Borland Pascal code?
See the MATH387 unit available from the Number Crunching Page:
http://www.geocities.com/SiliconValley/2787

Re:MATH387 speeds up code


On 2 Feb 1996, Mark Vaughan wrote:

Quote
>In article <4et9s8$...@news.asu.edu>,
>   dmit...@phyast.la.asu.edu (Dmitry Streblechenko) wrote:
>]-Are you guys serious? $20 for less than a dozen
>]-inline routines which everyone can write in
>]-20 minutes using any textbook on 80386 assembler???

>I dunno, I've made a lot of money doing easy stuff that
>other people were either too lazy or too ignorant to do
>for themselves.  Hey, look at the success of the Jiffy
>Lube franchise.  Changing the oil in your automobile
>is not rocket science, but plenty of people gladly fork
>over the $20 (or whatever it is) for the pleasure of
>not having to do it themselves.

>the problem with this particular offering is that there's
>a freeware package available that does pretty much the
>same stuff.  in fact it probably does more, as the
>author also includes units for complex numbers and for
>hyperbolic functions.  all code is included too.  look for
>MathLib2.zip on the Delphi Super Page.  good stuff.

Actually, MATH387 has one big advantage over MATHLIB2 - it uses INLINE
code, whereas MATHLIB2 does its work through function calls. Now, the
overhead of a function call isn't much, but when you call a function
to execute a single machine code instruction, the overheads become
quite significant. This is very much the case with processors like the
Pentium, where a machine code sine or cosine intruction takes very few
clock cycles, and the overhead of executing call and return
instructions become significant.

The bottom line? MATH387 is SIGNIFICANTLY faster than MATHLIB2! Just
try the SQWAVE demo program using MATHLIB2 instead of MATH387 to see
what I mean.

(BTW, apologies for the rather late posting. Our @#$% news server is
giving lots of trouble)

- Show quoted text -

Quote

>Mark Vaughan

>]-
>]-In article <4erojj$...@hawk.pix.za>,
>]-   chris.willi...@pixie.co.za (Chris Williams) wrote:
>]->If you ever write any code in Borland Delphi which does
>]->trig calculations  - typically scientific, engineering or
>]->statistical work  - and you need exponentiation (x^y), or any of
>]->the trig functions sin, cos, tan, atan, ln  - then try the MATH387
>]->unit. It dramatically speeds up code by using the built-in
>]->features of 80387 or better co-processors (or 486DX/Pentium)
>]->- which Borland's code does not.
>]->
>]->To get your copy of MATH387, check out the "Number Crunching
>]->Page" at:
>]->
>]->  http://www.geocities.com/SiliconValley/2787
>]->
>]->Try the demo program there  - it will graphically show you what
>]->kind of speed improvement you can expect on your particular
>]->machine.
>]->
>]->The Number Crunching Page also gives you a delightful walk down
>]->memory lane by tracing the history of floating point performance
>]->on desk-top computers over the last 20 years. It shows how a
>]->problem that took 12 hours to run in 1976 on a Z80 machine will
>]->now run in about 1 SECOND on a Pentium machine - at a lower
>]->price!
>]->
>]->

Chris Williams       chris.willi...@pixie.co.za
-----------------------------------------------
Want to speed up trig functions in your Borland Pascal code?
See the MATH387 unit available from the Number Crunching Page:
http://www.geocities.com/SiliconValley/2787

Other Threads