Shop OBEX P1 Docs P2 Docs Learn Events
TAQOZ Reloaded v2.8 - double arithmetic (64 bit) words — Parallax Forums

TAQOZ Reloaded v2.8 - double arithmetic (64 bit) words

bob_g4bbybob_g4bby Posts: 401
edited 2022-02-22 11:23 in Forth

Very occasionally, you need a bit of 64 bit maths, so I wrote these words to give me what I needed. The assembler needs to be loaded (if it isn't already) to compile this:-

--- Double Arithmetic ver 3 for Taqoz Reloaded v2.8 by Bob Edwards Dec 2021

TAQOZ

IFDEF *64bitMaths* FORGET *64bitMaths* }
pub *64bitMaths*        PRINT" Double length maths for TAQOZ ver 3" ;

--- duplicate top 4 stack entries
: 4DUP  ( n1 n2 n3 n4 -- n1 n2 n3 n4 n1 n2 n3 n4 )
    4 0 DO 4TH LOOP
;

--- flg=true if d1=d2 else flg=false
pub D=  ( d1 d2 -- flg )
ROT = -ROT = + -2 =
;

--- cast signed long to signed double
pub L->D    ( n -- d )
    DUP 0<
    ;

--- multiply two signed doubles, result is signed double
pub D*  ( d1 d2 -- d3 )


  R SWAP >R


    2DUP UM* 2SWAP
    R> * SWAP R> * + +
    ;

--- square signed double d2 = d1 squared
pub DSQR    ( d1 -- d2 )
    2DUP D*
    ;

--- d2 = absolute value of d1, flg = true if d1 was negative
pub DABS    ( d1 -- d2 flg )
    DUP 0<
    IF
        -1 -1 D* 1
    ELSE
        0
    THEN
;

--- d2 = -d1
pub DNEGATE
    -1 -1 D*
;

--- if flg is true, d2 = -d1
pub ?DNEGATE    ( d1 flg -- d2 )
    0<>
    IF
        DNEGATE
    THEN
;

--- print a signed double
pub SD. ( d -- )
    DUP 0< IF
        SWAP
        -1 XOR 1+ DUP 0= >R
        SWAP
        -1 XOR R> IF 1+ THEN
        45 EMIT
    THEN
    D.                                      --- then print the absolute value
    ;

--- logic shift double d1 left by 1 position
pub D<shift ( d1 -- d2 )
    SWAP                --- Dhigh Dlow
    DUP 0< >R           --- is top bit of Dlow set?
    2* SWAP 2*          --- Dlow*2 Dhigh*2
    R> IF               --- if carry over to high long is reqd
        1+              --- Dhigh + 1
    THEN
;

--- add signed doubles d3 = d1 + d2 
code D+ ( d1 d2 -- d3 )
    add d,b wc
    addx c,a
    jmp #@2DROP
end

--- subtract signed doubles d3 = d1 - d2
code D- ( d1 d2 -- d3 )
    sub d,b wc
    subx c,a
    jmp #@2DROP 
end

--- left shift UD until the MS bit is 1, leftshifts is the number of shifts
pub UDNORMAL    ( UD1 -- UD2 leftshifts )
0 >R                                --- set up shift count
BEGIN
    DUP $80000000 AND 0=
WHILE
    D<shift  R> 1+ >R
REPEAT
R>
;

 --- create a 'double' type variable (64 bits )
pre double  ( -- )
    8 [C] [G] [C] bytes [C] [G]
    ;

private
double Quotient
double Remainder
double Divisor
public

--- quotient = d1 / d2 with remainder - all unsigned doubles
pri (UD/)   ( dividend divisor -- remainder quotient )
        Divisor D!
        Quotient D!
        0. Remainder D!
        64 FOR
            Quotient D@  DUP 0< >R D<shift Quotient D!
            Remainder D@ D<shift R> IF 1. D+ THEN Remainder D!
            Remainder D@ Divisor D@ D-
            DUP $80000000 AND 0=
            IF
                Remainder D!
                Quotient D@ 1. D+ Quotient D!
            ELSE
                2DROP
            THEN
        NEXT
        Remainder D@
        Quotient D@
;

--- quotient = d1 / d2 with remainder - all unsigned doubles
pub UD/ ( dividend divisor -- remainder quotient )
    DUP 0=
    IF                          --- is divisor 32 bit or less?
        DROP UM// 0 -ROT
    ELSE
        (UD/)
    THEN
;

--- quotient = d1 / d2 with remainder - all signed doubles
pub D/  ( dividend divisor -- remainder quotient )
    DABS 2* >R 2SWAP DABS R> + >R 2SWAP             --- convert d1 and d2 to +ve, remembering original signs
    UD/
    R> SWITCH                                       --- bit1 set if divisor was -ve, bit0 = true for dividend 
        1 CASE DNEGATE 2SWAP DNEGATE 2SWAP BREAK
        2 CASE DNEGATE BREAK
;

--- arithmetic shift signed double d1 right by n positions, result is d2
code D>>    ( d1 n -- d2 )
.l1
    sar b,#1 wc
    rcr c,#1
    djnz a,#l1
    jmp #@DROP
end

--- test D>> function by shifting d repeatedly and displaying the result
pub D>>TEST     ( d -- )
    CRLF
    64 1 DO
        2DUP I DUP . SPACE D>>
        SD. CRLF
    LOOP
    2DROP
;

--- arithmetic shift signed double d1 left by n positions, result is d2
code D<<    ( d1 n -- d2 )
.l1
    shl c,#1 wc
    rcl b,#1
    djnz a,#l1
    jmp #@DROP
end

--- test D>> function by shifting d repeatedly and displaying the result
pub D<<TEST     ( d -- )
    CRLF
    64 1 DO
        2DUP I DUP . SPACE D<<
        SD. CRLF
    LOOP
    2DROP
;

END


Comments

  • Thanks Bob this is great, just what I need. I have an application that applies sensor calibration using quadratic equations and 64 bit arithmetic. Cheers Pete.

  • bob_g4bbybob_g4bby Posts: 401
    edited 2021-12-18 17:24

    @Peter_F, you're quite welcome. I am guessing that it will help with your BMP280 air pressure chip driver?

    The 64 bit maths routines have only been hand tested, so if you find bugs or write some more useful functions, please add them on.

    I'm afraid division with UD/ and D/ is very slow if your divisor is greater than 2^32. It's OK otherwise. The slow division is because I used the only algorithm I could understand for division, based on shift and subtract bit by bit - I did find other 'fast' algorithms but none that were described well enough for me. Some (possibly) were only efficient for very large integers.

  • Hi Bob, you are partly right. My application is for a different pressure sensor that measures water level and produces two frequencies - one for temperature and one for pressure. There are calibration coefficients needed to convert from frequency to real units via a quadratic equation, which is where the 64 bit maths will be useful.

    I haven't looked at the BMP280 driver for a while, I got distracted with Mecrisp/Tachyon on the RP2040.

    Coming back to the pressure sensor, the P2 should be great for measuring the two frequencies accurately ( 40 kHz and 170 kHz ) and applying the calibration coefficients.
    Cheers, Pete

  • Ok @Peter_F, good luck with your water level detection, that has a lot of potential applications.

    I've been trying to get the complicated bmp280 datasheet recipe for 64 bit conversion of pressure to work, no luck yet.

    Meanwhile I've sent off for an MS5611 which is aimed at altimeter use and the conversion maths needed is much less.

  • I've upissued the 64 bit maths code - the arithmetic shift words D>> and D<< weren't working

Sign In or Register to comment.