DNEGATE doesn't in v2.2

[RigTig]

\ I am having great fun using eForth
\ thank you for a great piece of work
\ In v2.2
3 0 dnegate .s \ results in
FFFD 0 <sp ok
\ when it should be FFFD FFFF <sp ok

\ regards, RigTig

[TG9541]

Hi RigTig,
I can conform he problem. It has nothing to do with my meddling with Dr. C.H. Tings code: it's already present in STM8EF V2.1. So far I didn't use double arithmetics much, and it's a good thing that you're testing it.

[TG9541]

Hi RigTig,
it's fixed in the develop branch.
In case it's needed in the master branch before release of STM8EF V2.2.7 please use the following definition as a work-around:

: DNEGATE NOT >R NOT 1 UM+ R> + ;

Thanks again for reporting this issue :-)

Regards & Greetings from Germany,
Thomas

[RigTig]

Thomas, Many thanks for fix. All I'd done was figure the high byte was too high by 1, so my temporary fix was just : DNEGATE DNEGATE 1- ; anyway, just so you have an integer square root when you need it, here is one algorithm coded in eForth \ some maths utility words \ eForth + is u+ : u+ ( n n -- u ) + ; : dnegate ( d -- -d ) \ fix for bug in v2.2 dnegate 1- ; : ud+ ( ud ud -- ud ) ( l h l h ) r rot um+ ( h d ) rot u+ ( d ) r> u+ ; : d- ( d d -- d ) \ 2do: ensure works on full range dnegate ud+ ; : d> ( d d -- f ) d- swap drop 0< not ; : isqrt ( d -- n ) $8000 ( d c ) $8000 ( d c g ) begin dup dup um* ( d c g g^2) 6 pick 6 pick ( d c g g^2 d ) d> if ( d c g ) over xor then ( d c g ) swap 2/ $7fff and ( d g c ) dup 0= if drop rot rot 2drop -1 ( g true ) else swap over ( d c g c ) or 0 then until ; The only trick here is to do a right bit shift by one bit, I've used the 2/ (which accepts positive and negative integers) and co-opted it by just dropping the sign bit ( $7fff and ). During my testing, I had the following line just after the 'begin' cr 2dup . . \ g in col1 and c in col 2 Interesting to watch how the result gets refined. For the record, here is the C program I used as the source of the algorithm unsigned int sqrt32(unsigned long n) { unsigned int c = 0x8000; unsigned int g = 0x8000; for (;;) { if (g*g > n) { g ^= c; # Bitwise exclusive OR } c >>= 1; # Right bit shift if (c == 0) { return g; } g |= c; # Bitwise inclusive OR } } Please keep up the good work Regards, RigTig

…

On Mon, Feb 13, 2017 at 4:54 AM, Thomas ***@***.***> wrote: Closed #13 <#13>. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#13 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AWVRezIkQan1xwjHG3pxhqKmRQtclvDjks5rb0c4gaJpZM4L-d__> .

[TG9541]

Hej RigTig,

I tested the code you posted: it works great, but 256 0 isqrt . returns 15, clearly a rounding error. Adding the line 1 0 ud+ to isqrt fixes this :-)

Would it be OK to add the code to a lib section in the STM8EF repo under a (permissive) OS license?

[RigTig]

Thomas, Be my guest. It is the least I can do given that I got the algorithm from the net and your efforts underlie what I did anyway. Well spotted about the shortcoming. I am working on a simple GCODE interpreter for a 3D printer and I think it'll just fit onto STM8. If so, I'll be using an ESP14. I have thoroughly enjoying exploring the inner workings of stm8ef. Thank you for your public contributions too. Regards, RigTig

…

On 20 Feb 2017, at 07:19, Thomas ***@***.***> wrote: Hej RigTig, I tested the code you posted: it works great, but 256 0 isqrt . returns 15, clearly a rounding error. Adding the line 1 0 ud+ to isqrt fixes this :-) Would it be OK to add the code to a lib section in the STM8EF repo under a (permissive) OS license? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[TG9541]

Thanks :-)
I had never thought that one would actually consider moving the limits of what a tiny micro can do that far! This project sounds like a fun - I'm looking forward to more feedback!

[RigTig]

Tested the isqrt and confirmed the bug. A little bit of exploring found the culprit: d> was actually d>=. So here is a new d> that seems to be ok now, and you'd never guess that isqrt actually gives 16 as the answer for 256 0 isqrt .

: d> ( d d -- f )
d- 2dup or 0= ( d f ) rot rot
0< swap drop or not
;
Thanks Thomas, and please keep up the good work.
Regards, RigTig

[TG9541]

@RigTig

Great, thanks! I had chiefly looked in the same direction but I falsely assumed that d>= was intentional :-)

As the GitHub issues appear to work great as a discussion platform!

Here is something that might be relevant for your G-code project and also for lib/math:

trigonometry & fractional arithmetics:

• Fractional arithmetics and CORDIC source (English/Dutch) http://home.hccnet.nl/anij/fractions.zip
• CORDIC Demo (Dutch) http://forth.hcc.nl/downloads/cordic.html
• another CORDIC source https://www.complang.tuwien.ac.at/forth/programs/cordic.fs

I've got no experience with fractional arithmetics (only fixed point), I guess that I should start experimenting a bit. CORDIC used to be widely used (there is even a MCS51 µC, the XC886, with CORDIC hardware for field oriented motor control).

Another that needs solving is a "uploader" program, ideally one that does includes, and that attempts to be smart (i.e. one that tests whether a library has already been loaded). AmForth has an interesting solution, but I'm skeptical about doing constant replacement in the uploader.