flashforth-devel

[Flashforth-devel] Floating point math for FF 5.x

[om1zz <om...@vo...>]

Hi, long time back I created a single precidsion floating point lib which was run on dspic33/pic24.
The lib adds following math words:

  facos   fasin   fatan   flog10  flog    fexp
  fsqrt   ftan    fcos    fsin    fpow    f/      f*      f-
  f+

plus words for printing fp numbers in eng and sci notations, etc.

You may find it at:

https://github.com/igor-m/FlashForth-FP

br
Igor

Re: [Flashforth-devel] Floating point math for FF 5.x

[om1zz <om...@vo...>]

..and what we need is a recogniser for floats.. It is already done for Amforth, so maybe it would be doable with FF too.
Currently you can convert integers to floats or define float constants via their IEEE values in hex (see the demos).
Would be great to be able to enter the floats directly, however :)
I.

Re: [Flashforth-devel] Floating point math for FF 5.x

[Mikael N. <mik...@fl...>]

I loaded your float number conversion routines just to check out the FP 
binary format.

I noticed that you can skip DO..LOOP in favour of FOR NEXT and it even 
simplifies  the code since 0 FOR NEXT just skips the loop completely.

So for example:

: flshiftn ( d-significand n-exponent n-times -- d n )
   ?dup 0> if 0 do flshift loop then ;

becomes:

: flshiftn ( d-significand n-exponent n-times -- d n )
   for flshift next ;

All the loops in the FP code simply loop a fixed amount of times, which 
nicely suits the FOR NEXT.

Anyway, I started to port Leons >float routine to FlashForth.

Have you seen any other >float routines than Leons, that could be 
checked out ?

BR Mike

On 10.05.2015 21:02, om1zz wrote:
> ..and what we need is a recogniser for floats.. It is already done for Amforth, so maybe it would be doable with FF too.
> Currently you can convert integers to floats or define float constants via their IEEE values in hex (see the demos).
> Would be great to be able to enter the floats directly, however :)
> I.
>
> ------------------------------------------------------------------------------
> One dashboard for servers and applications across Physical-Virtual-Cloud
> Widest out-of-the-box monitoring support with 50+ applications
> Performance metrics, stats and reports that give you Actionable Insights
> Deep dive visibility with transaction tracing using APM Insight.
> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y
> _______________________________________________
> Flashforth-devel mailing list
> Fla...@li...
> https://lists.sourceforge.net/lists/listinfo/flashforth-devel
>

Re: [Flashforth-devel] Floating point math for FF 5.x

[om1zz <om...@vo...>]

>Have you seen any other >float routines than Leons, that could be 
>checked out ?
Nope, I just worked with Leon at that time on what you can see there. The >float there requires set-recognizer and get-recognizer which come from amforth..

Re: [Flashforth-devel] Floating point math for FF 5.x

[Mikael N. <mik...@fl...>]

I think I will skip the recognizer stuff and just make >digit a deferred 
word inside interpret.

On 12.05.2015 00:00, om1zz wrote:
>> Have you seen any other >float routines than Leons, that could be
>> checked out ?
> Nope, I just worked with Leon at that time on what you can see there. The >float there requires set-recognizer and get-recognizer which come from amforth..
>

Re: [Flashforth-devel] Floating point math for FF 5.x

[om1zz <om...@vo...>]

What about a >float word done via a C-function which does strtod() or atof()?
i.

Re: [Flashforth-devel] Floating point math for FF 5.x

[om1zz <om...@vo...>]

I have found in 4y old post re amforth - the quick workaround when the recognisers were not working in amforth, afaik:

: # bl word count >float not abort" NaN"
okstate @ if postpone fliteral then ; immediate
ok
# 3.444e6 fs.
3.4439998E6 ok
# 1.234e6 fs.
1.23399997E6 ok
# 123456789e10 fs.
1.23456776E18 ok
# -1.3343e-12 # 6.222E22 f* fs.
-8.3020134E10 ok
# 3.44W-1
NaN

I.



______________________________________________________________
> Od: om1zz <om...@vo...>
> Komu: Mikael Nordman <mik...@fl...>, <fla...@li...>
> Datum: 11.05.2015 23:16
> Předmět: Re: [Flashforth-devel] Floating point math for FF 5.x
>
>What about a >float word done via a C-function which does strtod() or atof()?
>i.
>
>------------------------------------------------------------------------------
>One dashboard for servers and applications across Physical-Virtual-Cloud 
>Widest out-of-the-box monitoring support with 50+ applications
>Performance metrics, stats and reports that give you Actionable Insights
>Deep dive visibility with transaction tracing using APM Insight.
>http://ad.doubleclick.net/ddm/clk/290420510;117567292;y
>_______________________________________________
>Flashforth-devel mailing list
>Fla...@li...
>https://lists.sourceforge.net/lists/listinfo/flashforth-devel
>

Re: [Flashforth-devel] Floating point math for FF 5.x

[om1zz <om...@vo...>]

better formatted:

: # bl word count >float not abort" NaN" ok
state @ if postpone fliteral then ; immediate ok
# 3.444e6 fs.
3.4439998E6 ok
# 1.234e6 fs.
1.23399997E6 ok
# 123456789e10 fs.
1.23456776E18 ok
# -1.3343e-12 # 6.222E22 f* fs.
-8.3020134E10 ok
# 3.44W-1
NaN


i.

Re: [Flashforth-devel] Floating point math for FF 5.x

[Mikael N. <mik...@fl...>]

A first version that can parse and print floating point is available in 
git. Tested on dspic33fj128gp802. The CFloat.c file s not needed.
FF calls the Microchip FP lib routines directly without any C layer.

It is based on Leon Maurers FP work for Amforth. And on Igor Mokos 
adaptations.

Just send the float.txt
With ff-shell.py it looks like this:

cd ff
sudo shell/ff-shell.py -p /dev/ttyACM0 --xonxoff -s 38400

#send pic24/forth/float.txt

   ok<#,ram>
f# 3.14e-9 f. 0.00000000314000  ok<#,ram>
f# 3.14e-9 fs. 3.14000E-9  ok<#,ram>
f# 3.14e-8 fs. 3.14000E-8  ok<#,ram>

Excerpt from WORDS below:


facos	fasin
fatan	flog10	flog	fexp	fsqrt	ftan	fcos	fsin
fpow	f/	f*	f-	f+

f#	string>float	extract	fs.	fe.	f.	f.no-space	round2prc
precision	emitnzeros	fprintdigits	femitdigit	emitdigit	tkcofsign 
smlrpow10	f10^n
fround	fmod1	ceil	floor	f~	fprep4div	fmin	fmax
f<=	f>=	f>	f<	f2*	f2/	f=	fabs
negateiftrue	fnegateifneg	fnegate	f0<	f0=	f>s	f>d	s>f
d>f	faddtoexponent	f>sigexp	sigexp>f	fshifttoexp	flshiftn	flshift	frshiftn
frshift	fmksgnd	fsetsign	fsetexponent	fsignificand	fsign	fexponent 
frawexponent
dsplit	dnegateifneg	d10*	>=	f0.5	f10	f1	f0
fliteral	fvariable	fconstant	f!	f@	fnover	nfover	fnswap
nfswap	ftuck	fnip	frot	fswap	fover	fdup	fdrop
d>s	0>	-fpmath

Re: [Flashforth-devel] Floating point math for FF 5.x

[om1zz <om...@vo...>]

A test for your floats parsing experiments:

\ FLOATS PARSING TEST
\ IgorM 2012-15

\ TEST START

f#  0e0  fs.
f# -0e0  fs.
f# -0e-0 fs.
f# -0e-0 fs.
f# +0e-0 fs.
f# +0e+0 fs.

f#  0.0e0 fs.
f# -0.0e0 fs.
f# -0.0e-0 fs.
f# -0.0e-0 fs.
f# +0.0e-0 fs.
f# +0.0e+0 fs.

f#  1e0 fs.
f# -1e0 fs.
f# -1e-0 fs.
f# -1e-0 fs.
f# +1e-0 fs.
f# +1e+0 fs.

f#  1.0e0 fs.
f# -1.0e0 fs.
f# -1.0e-0 fs.
f# -1.0e-0 fs.
f# +1.0e-0 fs.
f# +1.0e+0 fs.

f#  1.0e09 fs.
f# -1.0e09 fs.
f# -1.0e-09 fs.
f# -1.0e-09 fs.
f# +1.0e-09 fs.
f# +1.0e+09 fs.

f#  9e0 fs.
f# -9e0 fs.
f#  9e-0 fs.
f# -9e+0 fs.

f#  9e09 fs.
f# -9e09 fs.
f#  9e-09 fs.
f# -9e+09 fs.

f# +9e+0 fs.
f# -9e+0 fs.
f# +9e-0 fs.
f# +9e+0 fs.

f# +9e+09 fs.
f# -9e+09 fs.
f# +9e-09 fs.
f# +9e+09 fs.

f# +9.e+0 fs.
f# -9.e+0 fs.
f# +9.e-0 fs.
f# +9.e+0 fs.

f# +9.9e0 fs.
f# -9.9e0 fs.
f# +9.9e0 fs.
f# +9.9e0 fs.

f# +9.9e+0 fs.
f# -9.9e+0 fs.
f# +9.9e-0 fs.
f# +9.9e+0 fs.


f# +9.9e+09 fs.
f# -9.9e+09 fs.
f# +9.9e-09 fs.
f# +9.9e+09 fs.

f#  0.000000e0 fs.
f#  0.000000e-0 fs.
f# -0.000000e0 fs.
f# -0.000000e-0 fs.

f# +0.000000e+0 fs.
f# +0.000000e-0 fs.
f# -0.000000e+0 fs.
f# -0.000000e-0 fs.

f#  9.000000e0 fs.
f#  9.000000e-0 fs.
f# -9.000000e0 fs.
f# -9.000000e-0 fs.

f# +9.000000e+0 fs.
f# +9.000000e-0 fs.
f# -9.000000e+0 fs.
f# -9.000000e-0 fs.

f#  9.000000e09 fs. 
f#  9.000000e-09 fs.
f# -9.000000e09 fs.
f# -9.000000e-09 fs.

f# +9.000000e+09 fs.
f# +9.000000e-09 fs.
f# -9.000000e+09 fs.
f# -9.000000e-09 fs.

f#  9.999000e0 fs.
f# -9.999000e0 fs.
f#  9.999000e9 fs.
f# -9.999000e9 fs.

f#  9.999000e+0 fs.
f# -9.999000e+0 fs.
f#  9.999000e-0 fs.
f# -9.999000e-0 fs.

f#  9.999000e+00 fs.
f# -9.999000e+00 fs.
f#  9.999000e-00 fs.
f# -9.999000e-00 fs.

f#  9.999999e+30 fs.
f# -9.999999e+30 fs.
f#  9.999999e-30 fs.
f# -9.999999e-30 fs.

f#  999999.9e+30 fs.
f# -999999.9e+30 fs.
f#  999999.9e-30 fs.
f# -999999.9e-30 fs.

f#  999999.9e30 fs.
f# -999999.9e30 fs.
f#  999999.9e-30 fs.
f# -999999.9e-30 fs.
f# +999999.9e30 fs.
f# +999999.9e-30 fs.

f#  0.000000e0 fs.
f#  000000.0e-0 fs.
f#  000000.0e+0 fs.
f#  +000000.0e-0 fs.
f#  +000000.0e+0 fs.
f#  -000000.0e-0 fs.
f#  -000000.0e+0 fs.

f#  000000.1e-0 fs.
f#  000000.1e+0 fs.
f#  +000000.1e-0 fs.
f#  +000000.1e+0 fs.
f#  -000000.1e-0 fs.
f#  -000000.1e+0 fs.

f#  000000.1e-01 fs.
f#  000000.1e+01 fs.
f#  +000000.1e-01 fs.
f#  +000000.1e+01 fs.
f#  -000000.1e-01 fs.
f#  -000000.1e+01 fs.

f#  0.000099e+31 fs.
f# -0.000099e+31 fs.
f#  0.000099e-31 fs.
f# -0.000099e-31 fs.

f#  1.000077e+24 fs.
f# -1.000077e+24 fs.
f#  1.000077e-24 fs.
f# -1.000077e-24 fs.

f#  1.234567e0 fs.
f# -1.234567e0 fs.
f#  1.234567e-0 fs.
f# -1.234567e-0 fs.
f#  1.234567e+0 fs.
f# -1.234567e+0 fs.
f#  1.234567e-00 fs.
f# -1.234567e+00 fs.

f#  1.234567e30 fs.
f# -1.234567e30 fs.
f#  1.234567e+30 fs.
f# -1.234567e+30 fs.
f#  1.234567e-30 fs.
f# -1.234567e-30 fs.

f#  9.999999e+30 fs.
f# -9.999999e+30 fs.
f#  9.999999e-30 fs.
f# -9.999999e-30 fs.

f# +9.999999e+30 fs.
f# -9.999999e+30 fs.
f# +9.999999e-30 fs.
f# -9.999999e-30 fs.
f# +9.999999e30 fs.
f# -9999999e30 fs.

\ END OF TEST

Re: [Flashforth-devel] Floating point math for FF 5.x

[om1zz <om...@vo...>]

Trying to compile for pic33 the latest trunk under MPLAB (not X) and getting linker errors - undefined reference to all those mchp floating point stuff like "___addsf3" etc. So we are missing something..

Re: [Flashforth-devel] Floating point math for FF 5.x

[Mikael N. <mik...@fl...>]

You need to add the library directory to the linker path.

I did that on my mplabx/xc16 setup to make it work.

I dont use old MPLAB anymore.

It could also be that your lib files have different symbol names than 
the XC16 libraries.

Re: [Flashforth-devel] Floating point math for FF 5.x

[om1zz <om...@vo...>]

It compiles fine under MPLABX..

The parse test I've send you passes ok, it seems, but I've found a small bug:
  ok<#,ram>
f# -1000000.0e10 fs. -1.00000E16  ok<#,ram>
f# -10000000.0e10 fs. -1.00000E17  ok<#,ram>
f# -100000000.0e10 fs. -1.00000E18  ok<#,ram>
f# -1000000000.0e10 fs. ďAS FlashForth PIC24 15.05.2015

i.

______________________________________________________________
> Od: Mikael Nordman <mik...@fl...>
> Komu: <fla...@li...>
> Datum: 15.05.2015 17:16
> Předmět: Re: [Flashforth-devel] Floating point math for FF 5.x
>
>You need to add the library directory to the linker path.
>
>I did that on my mplabx/xc16 setup to make it work.
>
>I dont use old MPLAB anymore.
>
>It could also be that your lib files have different symbol names than 
>the XC16 libraries.
>
>
>------------------------------------------------------------------------------
>One dashboard for servers and applications across Physical-Virtual-Cloud 
>Widest out-of-the-box monitoring support with 50+ applications
>Performance metrics, stats and reports that give you Actionable Insights
>Deep dive visibility with transaction tracing using APM Insight.
>http://ad.doubleclick.net/ddm/clk/290420510;117567292;y
>_______________________________________________
>Flashforth-devel mailing list
>Fla...@li...
>https://lists.sourceforge.net/lists/listinfo/flashforth-devel
>

Re: [Flashforth-devel] Floating point math for FF 5.x

[om1zz <om...@vo...>]

f# 180 fs.  float?
f# 180. fs.  float?
f# 180.0 fs.  float?
f# 180e fs. 1.80000E2  ok<#,ram>

180 is the integer in FF
180. is the long integer in FF
180.0 - this has to be recognised as float, not sure how it is handled today in the lib..

i.

Re: [Flashforth-devel] Floating point math for FF 5.x

[Mikael N. <mik...@fl...>]

Floats need always the trailing 'e'.
I thank that is the forth custom.

BR

Re: [Flashforth-devel] Floating point math for FF 5.x - trigonm. test

[om1zz <om...@vo...>]

\ TRIGONOMETRIC FUNCTIONS TEST - 9 degree test
\ Based on "9 DEGREE" calculator test
\ asin(acos(atan(tan(cos(sin(9.0)))))) == 9.0
\ other results at:
\ http://www.rskey.org/~mwsebastian/miscprj/results.htm
\ IgorM 2015

-trig9
marker -trig9
: Pi f# 3.14159265e ;
: d2r Pi f# 180e f/ ;
: r2d f# 180e Pi f/ ;
: TRIG9
		d2r f* fsin
		d2r f* fcos
		d2r f* ftan
		fatan r2d f*
		facos r2d f*
		fasin r2d f*
;
\   9.0 degree must return 9.0 degree
\   not with single precision, however :)
f# 9.0e TRIG9 fs.

f# 9.0e TRIG9 fs. 8.88809E0  ok<#,ram>

Re: [Flashforth-devel] Floating point math for FF 5.x - Edu - trigo

[om1zz <om...@vo...>]

For educational purposes :)

\ Library - SIN COS TAN
\ for FlashForth 5.x
\ IgorM 25-9-2010 and 5/2015
\ v 3.0.
\ based on 4tH library by J.L. Bezemer

-fsincostan
marker -fsincostan

\ the original Bezemer's taylor does not work here
\ as it needs a separate floating point stack!!

: >taylor fdup f* fover ;        \ setup for Taylor series
: _taylor fover f* frot fover ;
: +taylor f/ f+ frot frot ;      \ add Taylor iteration
: -taylor f/ f- frot frot ;      \ subtract Taylor iteration

\ put x in RADIANS within 2pi range
: >range                              
$0fdb $4049 fdup f+                  ( x pi2 )
fover fover f/                       ( x pi2 x/pi2 )
floor fover f*                      ( x pi2 mod )
frot fswap f-                        ( pi2 mod )
$0fdb $4049 fover                    ( pi2 mod pi mod )
f< if fswap f- else fnip then ;

: _fsin 
fdup >taylor                            ( x x2 x )
_taylor $0000 $40c0 -taylor                 ( x-3 x2 x3 )
_taylor $0000 $42f0 +taylor                 ( x+5 x2 x5 )
_taylor $8000 $459d -taylor                 ( x-7 x2 x7 )
_taylor $3000 $48b1 +taylor                 ( x+9 x2 x9 )
_taylor $4540 $4c18 -taylor                 ( x-11 x2 x11 )
_taylor $9466 $4fb9 +taylor                 ( x+13 x2 x13 )
_taylor $3bbc $5398 -taylor                 ( x-15 x2 x15 )
_taylor $bf77 $57a1 +taylor                 ( x+17 x2 x17 )
_taylor $15ca $5bd8 -taylor                 ( x-19 x2 x19 )
fdrop fdrop ;                                   ( x-19 ) 

\ calculate fsin
: fsinf  ( RAD -- sinus )
fdup f0< >r  fabs
>range 
fdup $0fdb $4049 f> if $0fdb $4049 f- 1 >r else 0 >r then 
fdup  $0fdb $3fc9 f> if $0fdb $4049 fswap f- then  
_fsin 
r> if fnegate then 
r> if fnegate then ;

\ calculate fcos        
: fcosf   ( RAD -- cosinus )        
$0fdb $3fc9 f+ fsinf ;

\ calculate ftan
: ftanf ( RAD -- tangens )  
fdup fsinf fswap fcosf 
fdup f0= if abort else f/  then ;   

f# 0.5e fsin fs. 4.79426E-1  ok<#,ram>
f# 0.5e fsinf fs. 4.79426E-1  ok<#,ram>
f# 0.5e fcos fs. 8.77583E-1  ok<#,ram>
f# 0.5e fcosf fs. 8.77583E-1  ok<#,ram>
f# 0.5e ftan fs. 5.46303E-1  ok<#,ram>
f# 0.5e ftanf fs. 5.46303E-1  ok<#,ram>

Re: [Flashforth-devel] Floating point math for FF 5.x

[Mikael N. <mik...@fl...>]

Fixed.

On 15.05.2015 18:28, om1zz wrote:
> It compiles fine under MPLABX..
>
> The parse test I've send you passes ok, it seems, but I've found a small bug:
>    ok<#,ram>
> f# -1000000000.0e10 fs. ďAS FlashForth PIC24 15.05.2015
>

Re: [Flashforth-devel] Floating point math for FF 5.x

[om1zz <om...@vo...>]

Could you set the printer precision to 7 (line 480) plz?
So we can see potential issues better. Also mchp uses 6dec places in the xc16 guide, thus we can compare the values.
PS: here in float.txt the precision is the number of digits to be printed in total, while xc16 uses it for number of digits after decimal point.
I.

[Flashforth-devel] Free memory size

[om1zz <om...@vo...>]

How we can see the volume of free ram and flash plz?
I.

Re: [Flashforth-devel] Free memory size

[Mikael N. <mik...@fl...>]

#send forth/free.txt
.free
Flash:29607 bytes
Ram:14151 bytes ok<#,ram>

Re: [Flashforth-devel] Free memory size

[om1zz <om...@vo...>]

Thanks!,
:) 
I.

______________________________________________________________
> Od: Mikael Nordman <mik...@fl...>
> Komu: <fla...@li...>
> Datum: 15.05.2015 20:26
> Předmět: Re: [Flashforth-devel] Free memory size
>
>#send forth/free.txt
>.free
>Flash:29607 bytes
>Ram:14151 bytes ok<#,ram>
>
>------------------------------------------------------------------------------
>One dashboard for servers and applications across Physical-Virtual-Cloud 
>Widest out-of-the-box monitoring support with 50+ applications
>Performance metrics, stats and reports that give you Actionable Insights
>Deep dive visibility with transaction tracing using APM Insight.
>http://ad.doubleclick.net/ddm/clk/290420510;117567292;y
>_______________________________________________
>Flashforth-devel mailing list
>Fla...@li...
>https://lists.sourceforge.net/lists/listinfo/flashforth-devel
>

Re: [Flashforth-devel] Floating point math for FF 5.x

[om1zz <om...@vo...>]

Not a critical bug, rather a feature :)

f# 0.0000000000000000158e fs. 1.580000E-17  ok<#,ram>
f# 9.999999999e0 fs. overflow float?
f# 3.141592653589793e0 fs. overflow float?

More than 7 digits to enter does not improve single precision, but maybe it could be handled somehow..
Nice to have requirement :)

PS: plz remove .s from f#  :)
i.

______________________________________________________________
> Od: Mikael Nordman <mik...@fl...>
> Komu: <fla...@li...>
> Datum: 15.05.2015 19:17
> Předmět: Re: [Flashforth-devel] Floating point math for FF 5.x
>
>Fixed.
>
>On 15.05.2015 18:28, om1zz wrote:
>> It compiles fine under MPLABX..
>>
>> The parse test I've send you passes ok, it seems, but I've found a small bug:
>>    ok<#,ram>
>> f# -1000000000.0e10 fs. ďAS FlashForth PIC24 15.05.2015
>>
>
>------------------------------------------------------------------------------
>One dashboard for servers and applications across Physical-Virtual-Cloud 
>Widest out-of-the-box monitoring support with 50+ applications
>Performance metrics, stats and reports that give you Actionable Insights
>Deep dive visibility with transaction tracing using APM Insight.
>http://ad.doubleclick.net/ddm/clk/290420510;117567292;y
>_______________________________________________
>Flashforth-devel mailing list
>Fla...@li...
>https://lists.sourceforge.net/lists/listinfo/flashforth-devel
>

Re: [Flashforth-devel] Floating point math for FF 5.x - special values

[om1zz <om...@vo...>]

We have problems with large numbers and alike results, it seems.

This test which worked with amforth crashes fs. :

\ large positive and negative
$ffff $ff7f fs. ." shall be -3.402823466e+38 " cr
$ffff $7f7f fs. ." shall be 3.402823466e+38 " cr

Moreover all tests where the result is undefined, and usualy returned above numbers (with my amforth asm float lib), returns now 0xffff 0x7fff from xc16 float, for example:

f-1.0 flog  ok<#,ram> 65535 32767
fpi/2 fasin  ok<#,ram> 65535 32767
fpi/2 facos  ok<#,ram> 65535 32767

what crashes fs. as well. So we must rethink how to print these states.
i.