Diff for /gforth/float.fs between versions 1.9 and 1.46

version 1.9, 1994/10/27 16:32:19 version 1.46, 2004/12/31 13:23:57
Line 1 Line 1
 \ High level floating point                            14jan94py  \ High level floating point                            14jan94py
   
 1 cells 4 = [IF]  \ Copyright (C) 1995,1997,2003,2004 Free Software Foundation, Inc.
 ' cells   Alias sfloats  
 ' cell+   Alias sfloat+  
 ' align   Alias sfalign  
 ' aligned Alias sfaligned  
 [ELSE]  
 : sfloats  2* 2* ;  
 : sfloat+  4 + ;  
 : sfaligned ( addr -- addr' )  3 + -4 and ;  
 : sfalign ( -- )  here dup sfaligned swap ?DO  bl c,  LOOP ;  
 [THEN]  
   
 1 floats 8 = [IF]  
 ' floats   Alias dfloats  
 ' float+   Alias dfloat+  
 ' falign   Alias dfalign  
 ' faligned Alias dfaligned  
 [ELSE]  
 : dfloats  2* 2* 2* ;  
 : dfloat+  8 + ;  
 : dfaligned ( addr -- addr' )  7 + -8 and ;  
 : dfalign ( -- )  here dup dfaligned swap ?DO  bl c,  LOOP ;  
 [THEN]  
   
 : f, ( f -- )  here 1 floats allot f! ;  
   
 \ !! have create produce faligned pfas  
 : fconstant  ( r -- )  
   falign here f,  Create A,  
   DOES>  @ f@ ;  
   
 : fvariable  
   falign here 0. d>f f, AConstant ;  
   
 : fdepth  ( -- n )  f0 @ fp@ - [ 1 floats ] Literal / ;  \ This file is part of Gforth.
   
 : FLit ( -- r )  r> faligned dup f@ float+ >r ;  \ Gforth is free software; you can redistribute it and/or
 : FLiteral ( r -- )  postpone FLit  falign f, ;  immediate  \ modify it under the terms of the GNU General Public License
   \ as published by the Free Software Foundation; either version 2
 &16 Value precision  \ of the License, or (at your option) any later version.
 : set-precision  to precision ;  
   \ This program is distributed in the hope that it will be useful,
   \ but WITHOUT ANY WARRANTY; without even the implied warranty of
   \ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   \ GNU General Public License for more details.
   
   \ You should have received a copy of the GNU General Public License
   \ along with this program; if not, write to the Free Software
   \ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA.
   
   \ 1 cells 4 = [IF]
   \ ' cells   Alias sfloats
   \ ' cell+   Alias sfloat+
   \ ' align   Alias sfalign
   \ ' aligned Alias sfaligned
   \ [ELSE]
   \ : sfloats  2* 2* ;
   \ : sfloat+  4 + ;
   \ : sfaligned ( addr -- addr' )  3 + -4 and ;
   \ : sfalign ( -- )  here dup sfaligned swap ?DO  bl c,  LOOP ;
   \ [THEN]
   
   \ 1 floats 8 = [IF]
   \ ' floats   Alias dfloats
   \ ' float+   Alias dfloat+
   \ ' falign   Alias dfalign
   \ ' faligned Alias dfaligned
   \ [ELSE]
   \ : dfloats  2* 2* 2* ;
   \ : dfloat+  8 + ;
   \ : dfaligned ( addr -- addr' )  7 + -8 and ;
   \ : dfalign ( -- )  here dup dfaligned swap ?DO  bl c,  LOOP ;
   \ [THEN]
   
   : sfalign ( -- ) \ float-ext s-f-align
       \G If the data-space pointer is not single-float-aligned, reserve
       \G enough space to align it.
       here dup sfaligned swap ?DO  bl c,  LOOP ;
   : dfalign ( -- ) \ float-ext d-f-align
       \G If the data-space pointer is not double-float-aligned, reserve
       \G enough space to align it.
       here dup dfaligned swap ?DO  bl c,  LOOP ;
   
   1 sfloats (Field) sfloat+ , ( sf-addr1 -- sf-addr2 ) \ float-ext s-float-plus
   \G @code{1 sfloats +}.
   
   1 dfloats (Field) dfloat+ , ( df-addr1 -- df-addr2 ) \ float-ext d-float-plus
   \G @code{1 dfloats +}.
   
   : f, ( f -- ) \ gforth
       \G Reserve data space for one floating-point number and store
       \G @i{f} in the space.
       here 1 floats allot f! ;
   
   : fconstant  ( r "name" -- ) \ float f-constant
       Create f,
   DOES> ( -- r )
       f@ ;
   
   : fdepth ( -- +n ) \ float f-depth
       \G @i{+n} is the current number of (floating-point) values on the
       \G floating-point stack.
       fp0 @ fp@ - [ 1 floats ] Literal / ;
   
   : FLiteral ( compilation r -- ; run-time -- r ) \ float f-literal
       \G Compile appropriate code such that, at run-time, @i{r} is placed
       \G on the (floating-point) stack. Interpretation semantics are undefined.
       BEGIN  here cell+ cell+ dup faligned <>  WHILE  postpone noop  REPEAT
       postpone ahead here >r f, postpone then
       r> postpone literal postpone f@ ;  immediate
   
   &15 Value precision ( -- u ) \ float-ext
   \G @i{u} is the number of significant digits currently used by
   \G @code{F.} @code{FE.} and @code{FS.} 
   : set-precision ( u -- ) \ float-ext
       \G Set the number of significant digits currently used by
       \G @code{F.} @code{FE.} and @code{FS.} to @i{u}.
       to precision ;
   
 : scratch ( r -- addr len )  : scratch ( r -- addr len )
   pad precision - precision ;    pad precision - precision ;
Line 54 Line 99
   IF  2drop  scratch 3 min type  rdrop  EXIT  THEN    IF  2drop  scratch 3 min type  rdrop  EXIT  THEN
   IF  '- emit  THEN ;    IF  '- emit  THEN ;
   
 : f.  ( r -- )  f$ dup >r 0<  : f.  ( r -- ) \ float-ext f-dot
   \G Display (the floating-point number) @i{r} without exponent,
   \G followed by a space.
     f$ dup >r 0<=
   IF    '0 emit    IF    '0 emit
   ELSE  scratch r@ min type  r@ precision - zeros  THEN    ELSE  scratch r@ min type  r@ precision - zeros  THEN
   '. emit r@ negate zeros    '. emit r@ negate zeros
Line 62 Line 110
 \ I'm afraid this does not really implement ansi semantics wrt precision.  \ I'm afraid this does not really implement ansi semantics wrt precision.
 \ Shouldn't precision indicate the number of places shown after the point?  \ Shouldn't precision indicate the number of places shown after the point?
   
 : fe. ( r -- )  f$ 1- s>d 3 fm/mod 3 * >r 1+ >r  \ Why do you think so? ANS Forth appears ambiguous on this point. -anton.
   scratch r@ min type '. emit  scratch r> /string type  
   : fe. ( r -- ) \ float-ext f-e-dot
   \G Display @i{r} using engineering notation (with exponent dividable
   \G by 3), followed by a space.
     f$ 1- s>d 3 fm/mod 3 * >r 1+ >r
     scratch r@ tuck min tuck - >r type r> zeros
     '. emit scratch r> /string type
   'E emit r> . ;    'E emit r> . ;
   
 : fs. ( r -- )  f$ 1-  : fs. ( r -- ) \ float-ext f-s-dot
   \G Display @i{r} using scientific notation (with exponent), followed
   \G by a space.
     f$ 1-
   scratch over c@ emit '. emit 1 /string type    scratch over c@ emit '. emit 1 /string type
   'E emit . ;    'E emit . ;
   
 : sfnumber ( c-addr u -- r / )  require debugs.fs
     2dup >float  
   : sfnumber ( c-addr u -- r true | false )
       2dup [CHAR] e scan ( c-addr u c-addr2 u2 )
       dup 0=
       IF
           2drop 2dup [CHAR] E scan ( c-addr u c-addr3 u3 )
       THEN
       nip
     IF      IF
         2drop state @          >float
         IF  
             postpone FLiteral  
         THEN  
     ELSE      ELSE
         defers notfound          2drop false
     THEN ;      THEN ;
   
 ' sfnumber IS notfound  :noname ( c-addr u -- )
       2dup sfnumber
 1e0 fasin 2e0 f* fconstant pi      IF
           2drop POSTPONE FLiteral
 : f2*  2e0 f* ;      ELSE
 : f2/  2e0 f/ ;          defers compiler-notfound
 : 1/f  1e0 fswap f/ ;      ENDIF ;
   IS compiler-notfound
 : falog ( f -- 10^f )  [ 10e0 fln ] FLiteral f* fexp ;  
   
 : fsinh    fexpm1 fdup fdup 1e0 f+ f/ f+ f2/ ;  :noname ( c-addr u -- r )
 : fcosh    fexp fdup 1/f f+ f2/ ;      2dup sfnumber
 : ftanh    f2* fexpm1 fdup 2e0 f+ f/ ;      IF
           2drop
       ELSE
           defers interpreter-notfound
       ENDIF ;
   IS interpreter-notfound
   
   : fvariable ( "name" -- ) \ float f-variable
       Create 0.0E0 f, ;
       \ does> ( -- f-addr )
   
   1.0e0 fasin 2.0e0 f* fconstant pi ( -- r ) \ gforth
   \G @code{Fconstant} -- @i{r} is the value pi; the ratio of a circle's area
   \G to its diameter.
   
   : f2* ( r1 -- r2 ) \ gforth
       \G Multiply @i{r1} by 2.0e0
       2.0e0 f* ;
   
   : f2/ ( r1 -- r2 ) \ gforth
       \G Multiply @i{r1} by 0.5e0
       0.5e0 f* ;
   
   : 1/f ( r1 -- r2 ) \ gforth
       \G Divide 1.0e0 by @i{r1}.
       1.0e0 fswap f/ ;
   
   get-current environment-wordlist set-current
   1.7976931348623157e308 FConstant max-float
   set-current
   
   \ We now have primitives for these, so we need not define them
   
   \ : falog ( f -- 10^f )  [ 10.0e0 fln ] FLiteral f* fexp ;
   
   \ : fsinh    fexpm1 fdup fdup 1.0e0 f+ f/ f+ f2/ ;
   \ : fcosh    fexp fdup 1/f f+ f2/ ;
   \ : ftanh    f2* fexpm1 fdup 2.0e0 f+ f/ ;
   
   \ : fatanh   fdup f0< >r fabs 1.0e0 fover f- f/  f2* flnp1 f2/
   \            r> IF  fnegate  THEN ;
   \ : facosh   fdup fdup f* 1.0e0 f- fsqrt f+ fln ;
   \ : fasinh   fdup fdup f* 1.0e0 f+ fsqrt f/ fatanh ;
   
   : f~abs ( r1 r2 r3 -- flag ) \ gforth
       \G Approximate equality with absolute error: |r1-r2|<r3.
       frot frot f- fabs fswap f< ;
   
   : f~rel ( r1 r2 r3 -- flag ) \ gforth
       \G Approximate equality with relative error: |r1-r2|<r3*|r1+r2|.
           frot frot fover fabs fover fabs f+ frot frot
           f- fabs frot frot f* f< ;
   
   : f~ ( r1 r2 r3 -- flag ) \ float-ext f-proximate
       \G ANS Forth medley for comparing r1 and r2 for equality: r3>0:
       \G @code{f~abs}; r3=0: bitwise comparison; r3<0: @code{fnegate f~rel}.
       fdup f0=
       IF \ bitwise comparison
           fp@ float+ 1 floats over float+ over str=
           fdrop fdrop fdrop
           EXIT
       THEN
       fdup f0>
       IF
           f~abs
       ELSE
           fnegate f~rel
       THEN ;
   
 : fatanh   fdup f0< >r fabs 1e0 fover f- f/  f2* flnp1 f2/  \ proposals from Krishna Myeni in <cjsp2d$47l$1@ngspool-d02.news.aol.com>
            r> IF  fnegate  THEN ;  \ not sure if they are a good idea
 : facosh   fdup fdup f* 1e0 f- fsqrt f+ fln ;  
 : fasinh   fdup fdup f* 1e0 f+ fsqrt f/ fatanh ;  
   
 : f.s  ." <" fdepth 0 .r ." > " fdepth 0 max maxdepth-.s @ min dup 0   : FTRUNC ( r1 -- r2 )
   ?DO  dup i - 1- floats fp@ + f@ f.  LOOP  drop ;       \ round towards 0
       \ !! should be implemented properly
       F>D D>F ;
   
   : FMOD ( r1 r2 -- r )
       \ remainder of r1/r2
       FOVER FOVER F/ ftrunc F* F- ;

Removed from v.1.9  
changed lines
  Added in v.1.46


FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>