\ Gforth primitives

\ Copyright (C) 1995,1996 Free Software Foundation, Inc.

\ This file is part of Gforth.

\ Gforth is free software; you can redistribute it and/or
\ modify it under the terms of the GNU General Public License
\ as published by the Free Software Foundation; either version 2
\ of the License, or (at your option) any later version.

\ 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., 675 Mass Ave, Cambridge, MA 02139, USA.


\ WARNING: This file is processed by m4. Make sure your identifiers
\ don't collide with m4's (e.g. by undefining them).
\ 
\ 
\ 
\ This file contains primitive specifications in the following format:
\ 
\ forth name	stack effect	category	[pronunciation]
\ [""glossary entry""]
\ C code
\ [:
\ Forth code]
\ 
\ prims2x is pedantic about tabs vs. blanks. The fields of the first
\ line of a primitive are separated by tabs, the stack items in a
\ stack effect by blanks.
\
\ Both pronounciation and stack items (in the stack effect) must
\ conform to the C name syntax or the C compiler will complain.
\ 
\ 
\ These specifications are automatically translated into C-code for the
\ interpreter and into some other files. I hope that your C compiler has
\ decent optimization, otherwise the automatically generated code will
\ be somewhat slow. The Forth version of the code is included for manual
\ compilers, so they will need to compile only the important words.
\ 
\ Note that stack pointer adjustment is performed according to stack
\ effect by automatically generated code and NEXT is automatically
\ appended to the C code. Also, you can use the names in the stack
\ effect in the C code. Stack access is automatic. One exception: if
\ your code does not fall through, the results are not stored into the
\ stack. Use different names on both sides of the '--', if you change a
\ value (some stores to the stack are optimized away).
\ 
\ 
\ 
\ The stack variables have the following types:
\ 
\ name matches	type
\ f.*		Bool
\ c.*		Char
\ [nw].*		Cell
\ u.*		UCell
\ d.*		DCell
\ ud.*		UDCell
\ r.*		Float
\ a_.*		Cell *
\ c_.*		Char *
\ f_.*		Float *
\ df_.*		DFloat *
\ sf_.*		SFloat *
\ xt.*		XT
\ wid.*		WID
\ f83name.*	F83Name *
\ 
\ 
\ 
\ In addition the following names can be used:
\ ip	the instruction pointer
\ sp	the data stack pointer
\ rp	the parameter stack pointer
\ lp	the locals stack pointer
\ NEXT	executes NEXT
\ cfa	
\ NEXT1	executes NEXT1
\ FLAG(x)	makes a Forth flag from a C flag
\ 
\ 
\ 
\ Percentages in comments are from Koopmans book: average/maximum use
\ (taken from four, not very representative benchmarks)
\ 
\ 
\ 
\ To do:
\ 
\ throw execute, cfa and NEXT1 out?
\ macroize *ip, ip++, *ip++ (pipelining)?

\ these m4 macros would collide with identifiers
undefine(`index')
undefine(`shift')

noop	--		gforth
;
:
 ;

lit	-- w		gforth
w = (Cell)NEXT_INST;
INC_IP(1);
:
 r> dup @ swap cell+ >r ;

execute		xt --		core
ip=IP;
IF_TOS(TOS = sp[0]);
EXEC(xt);

perform		a_addr --	gforth
""equivalent to @code{@ execute}""
/* and pfe */
ip=IP;
IF_TOS(TOS = sp[0]);
EXEC(*(Xt *)a_addr);
:
 @ execute ;

\+has-locals [IF]

branch-lp+!#	--	gforth	branch_lp_plus_store_number
/* this will probably not be used */
branch_adjust_lp:
lp += (Cell)(IP[1]);
goto branch;

\+[THEN]

branch	--		gforth
branch:
ip = (Xt *)(((Cell)IP)+(Cell)NEXT_INST);
NEXT_P0;
:
 r> dup @ + >r ;

\ condbranch(forthname,restline,code,forthcode)
\ this is non-syntactical: code must open a brace that is closed by the macro
define(condbranch,
$1	$2
$3	ip = (Xt *)(((Cell)IP)+(Cell)NEXT_INST);
        NEXT_P0;
	NEXT;
}
else
    INC_IP(1);
$4

\+has-locals [IF]

$1-lp+!#	$2_lp_plus_store_number
$3    goto branch_adjust_lp;
}
else
    INC_IP(2);

\+[THEN]
)

condbranch(?branch,f --		f83	question_branch,
if (f==0) {
    IF_TOS(TOS = sp[0]);
,)

\ we don't need an lp_plus_store version of the ?dup-stuff, because it
\ is only used in if's (yet)

\+has-xconds [IF]

?dup-?branch	f -- f	new	question_dupe_question_branch
""The run-time procedure compiled by @code{?DUP-IF}.""
if (f==0) {
  sp++;
  IF_TOS(TOS = sp[0]);
  ip = (Xt *)(((Cell)IP)+(Cell)NEXT_INST);
  NEXT_P0;
  NEXT;
}
else
  INC_IP(1);

?dup-0=-?branch	f --	new	question_dupe_zero_equals_question_branch
""The run-time procedure compiled by @code{?DUP-0=-IF}.""
/* the approach taken here of declaring the word as having the stack
effect ( f -- ) and correcting for it in the branch-taken case costs a
few cycles in that case, but is easy to convert to a CONDBRANCH
invocation */
if (f!=0) {
  sp--;
  ip = (Xt *)(((Cell)IP)+(Cell)NEXT_INST);
  NEXT_P0;
  NEXT;
}
else
  INC_IP(1);

\+[THEN]

condbranch((next),--		cmFORTH	paren_next,
if ((*rp)--) {
,:
 r> r> dup 1- >r
 IF dup @ + >r ELSE cell+ >r THEN ;)

condbranch((loop),--		gforth	paren_loop,
Cell index = *rp+1;
Cell limit = rp[1];
if (index != limit) {
    *rp = index;
,:
 r> r> 1+ r> 2dup =
 IF >r 1- >r cell+ >r
 ELSE >r >r dup @ + >r THEN ;)

condbranch((+loop),n --		gforth	paren_plus_loop,
/* !! check this thoroughly */
Cell index = *rp;
/* sign bit manipulation and test: (x^y)<0 is equivalent to (x<0) != (y<0) */
/* dependent upon two's complement arithmetic */
Cell olddiff = index-rp[1];
if ((olddiff^(olddiff+n))>=0   /* the limit is not crossed */
    || (olddiff^n)>=0          /* it is a wrap-around effect */) {
#ifdef i386
    *rp += n;
#else
    *rp = index + n;
#endif
    IF_TOS(TOS = sp[0]);
,:
 r> swap
 r> r> 2dup - >r
 2 pick r@ + r@ xor 0< 0=
 3 pick r> xor 0< 0= or
 IF    >r + >r dup @ + >r
 ELSE  >r >r drop cell+ >r THEN ;)

\+has-xconds [IF]

condbranch((-loop),u --		gforth	paren_minus_loop,
/* !! check this thoroughly */
Cell index = *rp;
UCell olddiff = index-rp[1];
if (olddiff>u) {
#ifdef i386
    *rp -= u;
#else
    *rp = index - u;
#endif
    IF_TOS(TOS = sp[0]);
,)

condbranch((s+loop),n --		gforth	paren_symmetric_plus_loop,
""The run-time procedure compiled by S+LOOP. It loops until the index
crosses the boundary between limit and limit-sign(n). I.e. a symmetric
version of (+LOOP).""
/* !! check this thoroughly */
Cell index = *rp;
Cell diff = index-rp[1];
Cell newdiff = diff+n;
if (n<0) {
    diff = -diff;
    newdiff = -newdiff;
}
if (diff>=0 || newdiff<0) {
#ifdef i386
    *rp += n;
#else
    *rp = index + n;
#endif
    IF_TOS(TOS = sp[0]);
,)

\+[THEN]

unloop		--	core
rp += 2;
:
 r> rdrop rdrop >r ;

(for)	ncount --		cmFORTH		paren_for
/* or (for) = >r -- collides with unloop! */
*--rp = 0;
*--rp = ncount;
:
 r> swap 0 >r >r >r ;

(do)	nlimit nstart --		gforth		paren_do
/* or do it in high-level? 0.09/0.23% */
*--rp = nlimit;
*--rp = nstart;
:
 r> swap rot >r >r >r ;

(?do)	nlimit nstart --	gforth	paren_question_do
*--rp = nlimit;
*--rp = nstart;
if (nstart == nlimit) {
    IF_TOS(TOS = sp[0]);
    goto branch;
    }
else {
    INC_IP(1);
}
:
  2dup =
  IF   r> swap rot >r >r
       dup @ + >r
  ELSE r> swap rot >r >r
       cell+ >r
  THEN ;				\ --> CORE-EXT

\+has-xconds [IF]

(+do)	nlimit nstart --	gforth	paren_plus_do
*--rp = nlimit;
*--rp = nstart;
if (nstart >= nlimit) {
    IF_TOS(TOS = sp[0]);
    goto branch;
    }
else {
    INC_IP(1);
}
:
 swap 2dup
 r> swap >r swap >r
 >=
 IF
     dup @ +
 ELSE
     cell+
 THEN  >r ;

(u+do)	ulimit ustart --	gforth	paren_u_plus_do
*--rp = ulimit;
*--rp = ustart;
if (ustart >= ulimit) {
    IF_TOS(TOS = sp[0]);
    goto branch;
    }
else {
    INC_IP(1);
}
:
 swap 2dup
 r> swap >r swap >r
 u>=
 IF
     dup @ +
 ELSE
     cell+
 THEN  >r ;

(-do)	nlimit nstart --	gforth	paren_minus_do
*--rp = nlimit;
*--rp = nstart;
if (nstart <= nlimit) {
    IF_TOS(TOS = sp[0]);
    goto branch;
    }
else {
    INC_IP(1);
}
:
 swap 2dup
 r> swap >r swap >r
 <=
 IF
     dup @ +
 ELSE
     cell+
 THEN  >r ;

(u-do)	ulimit ustart --	gforth	paren_u_minus_do
*--rp = ulimit;
*--rp = ustart;
if (ustart <= ulimit) {
    IF_TOS(TOS = sp[0]);
    goto branch;
    }
else {
    INC_IP(1);
}
:
 swap 2dup
 r> swap >r swap >r
 u<=
 IF
     dup @ +
 ELSE
     cell+
 THEN  >r ;

\+[THEN]

i	-- n		core
n = *rp;
:
 rp@ cell+ @ ;

i'	-- w		gforth		i_tick
""loop end value""
w = rp[1];
:
 rp@ cell+ cell+ @ ;

j	-- n		core
n = rp[2];
:
 rp@ cell+ cell+ cell+ @ ;

k	-- n		gforth
n = rp[4];
:
 rp@ [ 5 cells ] Literal + @ ;

\ digit is high-level: 0/0%

move	c_from c_to ucount --		core
memmove(c_to,c_from,ucount);
/* make an Ifdef for bsd and others? */
:
 >r 2dup u< IF r> cmove> ELSE r> cmove THEN ;

cmove	c_from c_to u --	string
while (u-- > 0)
  *c_to++ = *c_from++;
:
 bounds ?DO  dup c@ I c! 1+  LOOP  drop ;

cmove>	c_from c_to u --	string	c_move_up
while (u-- > 0)
  c_to[u] = c_from[u];
:
 dup 0= IF  drop 2drop exit  THEN
 rot over + -rot bounds swap 1-
 DO  1- dup c@ I c!  -1 +LOOP  drop ;

fill	c_addr u c --	core
memset(c_addr,c,u);
:
 -rot bounds
 ?DO  dup I c!  LOOP  drop ;

compare		c_addr1 u1 c_addr2 u2 -- n	string
""Compare the strings lexicographically. If they are equal, n is 0; if
the first string is smaller, n is -1; if the first string is larger, n
is 1. Currently this is based on the machine's character
comparison. In the future, this may change to considering the current
locale and its collation order.""
n = memcmp(c_addr1, c_addr2, u1<u2 ? u1 : u2);
if (n==0)
  n = u1-u2;
if (n<0)
  n = -1;
else if (n>0)
  n = 1;
:
 rot 2dup - >r min swap -text dup
 IF    rdrop
 ELSE  drop r@ 0>
       IF    rdrop -1
       ELSE  r> 1 and
       THEN
 THEN ;

-text		c_addr1 u c_addr2 -- n	new	dash_text
n = memcmp(c_addr1, c_addr2, u);
if (n<0)
  n = -1;
else if (n>0)
  n = 1;
:
 swap bounds
 ?DO  dup c@ I c@ = WHILE  1+  LOOP  drop 0
 ELSE  c@ I c@ - unloop  THEN  -text-flag ;
: -text-flag ( n -- -1/0/1 )
 dup 0< IF  drop -1  ELSE  0>  1 and  THEN  ;

toupper	c1 -- c2	gforth
c2 = toupper(c1);
:
 dup [char] a - [ char z char a - 1 + ] Literal u<  bl and - ;

capscomp	c_addr1 u c_addr2 -- n	new
n = memcasecmp(c_addr1, c_addr2, u); /* !! use something that works in all locales */
if (n<0)
  n = -1;
else if (n>0)
  n = 1;
:
 swap bounds
 ?DO  dup c@ I c@ <>
     IF  dup c@ toupper I c@ toupper =
     ELSE  true  THEN  WHILE  1+  LOOP  drop 0
 ELSE  c@ toupper I c@ toupper - unloop  THEN  -text-flag ;

-trailing	c_addr u1 -- c_addr u2		string	dash_trailing
u2 = u1;
while (c_addr[u2-1] == ' ')
  u2--;
:
 BEGIN  1- 2dup + c@ bl =  WHILE
        dup  0= UNTIL  ELSE  1+  THEN ;

/string		c_addr1 u1 n -- c_addr2 u2	string	slash_string
c_addr2 = c_addr1+n;
u2 = u1-n;
:
 tuck - >r + r> dup 0< IF  - 0  THEN ;

+	n1 n2 -- n		core	plus
n = n1+n2;

\ PFE-0.9.14 has it differently, but the next release will have it as follows
under+	n1 n2 n3 -- n n2	gforth	under_plus
""add @var{n3} to @var{n1} (giving @var{n})""
n = n1+n3;
:
 rot + swap ;

-	n1 n2 -- n		core	minus
n = n1-n2;
:
 negate + ;

negate	n1 -- n2		core
/* use minus as alias */
n2 = -n1;
:
 invert 1+ ;

1+	n1 -- n2		core		one_plus
n2 = n1+1;
:
 1 + ;

1-	n1 -- n2		core		one_minus
n2 = n1-1;
:
 1 - ;

max	n1 n2 -- n	core
if (n1<n2)
  n = n2;
else
  n = n1;
:
 2dup < IF swap THEN drop ;

min	n1 n2 -- n	core
if (n1<n2)
  n = n1;
else
  n = n2;
:
 2dup > IF swap THEN drop ;

abs	n1 -- n2	core
if (n1<0)
  n2 = -n1;
else
  n2 = n1;
:
 dup 0< IF negate THEN ;

*	n1 n2 -- n		core	star
n = n1*n2;
:
 um* drop ;

/	n1 n2 -- n		core	slash
n = n1/n2;
:
 /mod nip ;

mod	n1 n2 -- n		core
n = n1%n2;
:
 /mod drop ;

/mod	n1 n2 -- n3 n4		core		slash_mod
n4 = n1/n2;
n3 = n1%n2; /* !! is this correct? look into C standard! */
:
 >r s>d r> fm/mod ;

2*	n1 -- n2		core		two_star
n2 = 2*n1;
:
 dup + ;

2/	n1 -- n2		core		two_slash
/* !! is this still correct? */
n2 = n1>>1;
:
 dup MINI and IF 1 ELSE 0 THEN
 [ bits/byte cell * 1- ] literal 
 0 DO 2* swap dup 2* >r U-HIGHBIT and 
     IF 1 ELSE 0 THEN or r> swap
 LOOP nip ;

fm/mod	d1 n1 -- n2 n3		core		f_m_slash_mod
""floored division: d1 = n3*n1+n2, n1>n2>=0 or 0>=n2>n1""
#ifdef BUGGY_LONG_LONG
DCell r = fmdiv(d1,n1);
n2=r.hi;
n3=r.lo;
#else
/* assumes that the processor uses either floored or symmetric division */
n3 = d1/n1;
n2 = d1%n1;
/* note that this 1%-3>0 is optimized by the compiler */
if (1%-3>0 && (d1<0) != (n1<0) && n2!=0) {
  n3--;
  n2+=n1;
}
#endif
:
 dup >r dup 0< IF  negate >r dnegate r>  THEN
 over       0< IF  tuck + swap  THEN
 um/mod
 r> 0< IF  swap negate swap  THEN ;

sm/rem	d1 n1 -- n2 n3		core		s_m_slash_rem
""symmetric division: d1 = n3*n1+n2, sign(n2)=sign(d1) or 0""
#ifdef BUGGY_LONG_LONG
DCell r = smdiv(d1,n1);
n2=r.hi;
n3=r.lo;
#else
/* assumes that the processor uses either floored or symmetric division */
n3 = d1/n1;
n2 = d1%n1;
/* note that this 1%-3<0 is optimized by the compiler */
if (1%-3<0 && (d1<0) != (n1<0) && n2!=0) {
  n3++;
  n2-=n1;
}
#endif
:
 over >r dup >r abs -rot
 dabs rot um/mod
 r> r@ xor 0< IF       negate       THEN
 r>        0< IF  swap negate swap  THEN ;

m*	n1 n2 -- d		core	m_star
#ifdef BUGGY_LONG_LONG
d = mmul(n1,n2);
#else
d = (DCell)n1 * (DCell)n2;
#endif
:
 2dup      0< and >r
 2dup swap 0< and >r
 um* r> - r> - ;

um*	u1 u2 -- ud		core	u_m_star
/* use u* as alias */
#ifdef BUGGY_LONG_LONG
ud = ummul(u1,u2);
#else
ud = (UDCell)u1 * (UDCell)u2;
#endif
:
   >r >r 0 0 r> r> [ 8 cells ] literal 0
   DO
       over >r dup >r 0< and d2*+ drop
       r> 2* r> swap
   LOOP 2drop ;
: d2*+ ( ud n -- ud+n c )
   over MINI
   and >r >r 2dup d+ swap r> + swap r> ;

um/mod	ud u1 -- u2 u3		core	u_m_slash_mod
#ifdef BUGGY_LONG_LONG
UDCell r = umdiv(ud,u1);
u2=r.hi;
u3=r.lo;
#else
u3 = ud/u1;
u2 = ud%u1;
#endif
:
   0 swap [ 8 cells 1 + ] literal 0
   ?DO >r /modstep r> 
   LOOP drop swap 1 rshift or swap ;
: /modstep ( ud c R: u -- ud-?u c R: u )
   over I' u< 0= or IF I' - 1 ELSE 0 THEN  d2*+ ;
: d2*+ ( ud n -- ud+n c )
   over MINI
   and >r >r 2dup d+ swap r> + swap r> ;

m+	d1 n -- d2		double		m_plus
#ifdef BUGGY_LONG_LONG
d2.lo = d1.lo+n;
d2.hi = d1.hi - (n<0) + (d2.lo<d1.lo);
#else
d2 = d1+n;
#endif
:
 s>d d+ ;

d+	d1 d2 -- d		double	d_plus
#ifdef BUGGY_LONG_LONG
d.lo = d1.lo+d2.lo;
d.hi = d1.hi + d2.hi + (d.lo<d1.lo);
#else
d = d1+d2;
#endif
:
 rot + >r tuck + swap over u> r> swap - ;

d-	d1 d2 -- d		double		d_minus
#ifdef BUGGY_LONG_LONG
d.lo = d1.lo - d2.lo;
d.hi = d1.hi-d2.hi-(d1.lo<d2.lo);
#else
d = d1-d2;
#endif
:
 dnegate d+ ;

dnegate	d1 -- d2		double
/* use dminus as alias */
#ifdef BUGGY_LONG_LONG
d2 = dnegate(d1);
#else
d2 = -d1;
#endif
:
 invert swap negate tuck 0= - ;

d2*	d1 -- d2		double		d_two_star
#ifdef BUGGY_LONG_LONG
d2.lo = d1.lo<<1;
d2.hi = (d1.hi<<1) | (d1.lo>>(CELL_BITS-1));
#else
d2 = 2*d1;
#endif
:
 2dup d+ ;

d2/	d1 -- d2		double		d_two_slash
#ifdef BUGGY_LONG_LONG
d2.hi = d1.hi>>1;
d2.lo= (d1.lo>>1) | (d1.hi<<(CELL_BITS-1));
#else
d2 = d1>>1;
#endif
:
 dup 1 and >r 2/ swap 2/ [ 1 8 cells 1- lshift 1- ] Literal and
 r> IF  [ 1 8 cells 1- lshift ] Literal + THEN  swap ;

and	w1 w2 -- w		core
w = w1&w2;

or	w1 w2 -- w		core
w = w1|w2;
:
 invert swap invert and invert ;

xor	w1 w2 -- w		core
w = w1^w2;

invert	w1 -- w2		core
w2 = ~w1;
:
 MAXU xor ;

rshift	u1 n -- u2		core
  u2 = u1>>n;
:
    0 ?DO 2/ MAXI and LOOP ;

lshift	u1 n -- u2		core
  u2 = u1<<n;
:
    0 ?DO 2* LOOP ;

\ comparisons(prefix, args, prefix, arg1, arg2, wordsets...)
define(comparisons,
$1=	$2 -- f		$6	$3equals
f = FLAG($4==$5);
:
    [ char $1x char 0 = [IF]
	] IF false ELSE true THEN [
    [ELSE]
	] xor 0= [
    [THEN] ] ;

$1<>	$2 -- f		$7	$3different
f = FLAG($4!=$5);
:
    [ char $1x char 0 = [IF]
	] IF true ELSE false THEN [
    [ELSE]
	] xor 0<> [
    [THEN] ] ;

$1<	$2 -- f		$8	$3less
f = FLAG($4<$5);
:
    [ char $1x char 0 = [IF]
	] MINI and 0<> [
    [ELSE] char $1x char u = [IF]
	]   2dup xor 0<  IF nip ELSE - THEN 0<  [
	[ELSE]
	    ] MINI xor >r MINI xor r> u< [
	[THEN]
    [THEN] ] ;

$1>	$2 -- f		$9	$3greater
f = FLAG($4>$5);
:
    [ char $1x char 0 = [IF] ] negate [ [ELSE] ] swap [ [THEN] ]
    $1< ;

$1<=	$2 -- f		gforth	$3less_or_equal
f = FLAG($4<=$5);
:
    $1> 0= ;

$1>=	$2 -- f		gforth	$3greater_or_equal
f = FLAG($4>=$5);
:
    [ char $1x char 0 = [IF] ] negate [ [ELSE] ] swap [ [THEN] ]
    $1<= ;

)

comparisons(0, n, zero_, n, 0, core, core-ext, core, core-ext)
comparisons(, n1 n2, , n1, n2, core, core-ext, core, core)
comparisons(u, u1 u2, u_, u1, u2, gforth, gforth, core, core-ext)

\ dcomparisons(prefix, args, prefix, arg1, arg2, wordsets...)
define(dcomparisons,
$1=	$2 -- f		$6	$3equals
#ifdef BUGGY_LONG_LONG
f = FLAG($4.lo==$5.lo && $4.hi==$5.hi);
#else
f = FLAG($4==$5);
#endif

$1<>	$2 -- f		$7	$3different
#ifdef BUGGY_LONG_LONG
f = FLAG($4.lo!=$5.lo || $4.hi!=$5.hi);
#else
f = FLAG($4!=$5);
#endif

$1<	$2 -- f		$8	$3less
#ifdef BUGGY_LONG_LONG
f = FLAG($4.hi==$5.hi ? $4.lo<$5.lo : $4.hi<$5.hi);
#else
f = FLAG($4<$5);
#endif

$1>	$2 -- f		$9	$3greater
#ifdef BUGGY_LONG_LONG
f = FLAG($4.hi==$5.hi ? $4.lo>$5.lo : $4.hi>$5.hi);
#else
f = FLAG($4>$5);
#endif

$1<=	$2 -- f		gforth	$3less_or_equal
#ifdef BUGGY_LONG_LONG
f = FLAG($4.hi==$5.hi ? $4.lo<=$5.lo : $4.hi<=$5.hi);
#else
f = FLAG($4<=$5);
#endif

$1>=	$2 -- f		gforth	$3greater_or_equal
#ifdef BUGGY_LONG_LONG
f = FLAG($4.hi==$5.hi ? $4.lo>=$5.lo : $4.hi>=$5.hi);
#else
f = FLAG($4>=$5);
#endif

)

\+has-dcomps [IF]

dcomparisons(d, d1 d2, d_, d1, d2, double, gforth, double, gforth)
dcomparisons(d0, d, d_zero_, d, DZERO, double, gforth, double, gforth)
dcomparisons(du, ud1 ud2, d_u_, ud1, ud2, gforth, gforth, double-ext, gforth)

\+[THEN]

within	u1 u2 u3 -- f		core-ext
f = FLAG(u1-u2 < u3-u2);
:
 over - >r - r> u< ;

sp@	-- a_addr		gforth		spat
a_addr = sp+1;

sp!	a_addr --		gforth		spstore
sp = a_addr;
/* works with and without TOS caching */

rp@	-- a_addr		gforth		rpat
a_addr = rp;

rp!	a_addr --		gforth		rpstore
rp = a_addr;

\+has-floats [IF]

fp@	-- f_addr	gforth	fp_fetch
f_addr = fp;

fp!	f_addr --	gforth	fp_store
fp = f_addr;

\+[THEN]

;s	--		gforth	semis
ip = (Xt *)(*rp++);
NEXT_P0;

>r	w --		core	to_r
*--rp = w;
:
 (>r) ;
: (>r)  rp@ cell+ @ rp@ ! rp@ cell+ ! ;

r>	-- w		core	r_from
w = *rp++;
:
 rp@ cell+ @ rp@ @ rp@ cell+ ! (rdrop) rp@ ! ;
Create (rdrop) ' ;s A,

rdrop	--		gforth
rp++;
:
 r> r> drop >r ;

2>r	w1 w2 --	core-ext	two_to_r
*--rp = w1;
*--rp = w2;
:
 swap r> swap >r swap >r >r ;

2r>	-- w1 w2	core-ext	two_r_from
w2 = *rp++;
w1 = *rp++;
:
 r> r> swap r> swap >r swap ;

2r@	-- w1 w2	core-ext	two_r_fetch
w2 = rp[0];
w1 = rp[1];
:
 i' j ;

2rdrop	--		gforth	two_r_drop
rp+=2;
:
 r> r> drop r> drop >r ;

over	w1 w2 -- w1 w2 w1		core
:
 sp@ cell+ @ ;

drop	w --		core
:
 IF THEN ;

swap	w1 w2 -- w2 w1		core
:
 >r (swap) ! r> (swap) @ ;
Variable (swap)

dup	w -- w w		core
:
 sp@ @ ;

rot	w1 w2 w3 -- w2 w3 w1	core	rote
:
[ defined? (swap) [IF] ]
    (swap) ! (rot) ! >r (rot) @ (swap) @ r> ;
Variable (rot)
[ELSE] ]
    >r swap r> swap ;
[THEN]

-rot	w1 w2 w3 -- w3 w1 w2	gforth	not_rote
:
 rot rot ;

nip	w1 w2 -- w2		core-ext
:
 >r drop r> ;

tuck	w1 w2 -- w2 w1 w2	core-ext
:
 swap over ;

?dup	w -- w			core	question_dupe
if (w!=0) {
  IF_TOS(*sp-- = w;)
#ifndef USE_TOS
  *--sp = w;
#endif
}
:
 dup IF dup THEN ;

pick	u -- w			core-ext
w = sp[u+1];
:
 1+ cells sp@ + @ ;

2drop	w1 w2 --		core	two_drop
:
 drop drop ;

2dup	w1 w2 -- w1 w2 w1 w2	core	two_dupe
:
 over over ;

2over	w1 w2 w3 w4 -- w1 w2 w3 w4 w1 w2	core	two_over
:
 3 pick 3 pick ;

2swap	w1 w2 w3 w4 -- w3 w4 w1 w2	core	two_swap
:
 rot >r rot r> ;

2rot	w1 w2 w3 w4 w5 w6 -- w3 w4 w5 w6 w1 w2	double-ext	two_rote
:
 >r >r 2swap r> r> 2swap ;

2nip	w1 w2 w3 w4 -- w3 w4	gforth	two_nip
:
 2swap 2drop ;

2tuck	w1 w2 w3 w4 -- w3 w4 w1 w2 w3 w4	gforth	two_tuck
:
 2swap 2over ;

\ toggle is high-level: 0.11/0.42%

@	a_addr -- w		core	fetch
w = *a_addr;

!	w a_addr --		core	store
*a_addr = w;

+!	n a_addr --		core	plus_store
*a_addr += n;
:
 tuck @ + swap ! ;

c@	c_addr -- c		core	cfetch
c = *c_addr;
:
[ bigendian [IF] ]
    [ cell>bit 4 = [IF] ]
	dup [ 0 cell - ] Literal and @ swap 1 and
	IF  $FF and  ELSE  8>>  THEN  ;
    [ [ELSE] ]
	dup [ cell 1- ] literal and
	tuck - @ swap [ cell 1- ] literal xor
 	0 ?DO 8>> LOOP $FF and
    [ [THEN] ]
[ [ELSE] ]
    [ cell>bit 4 = [IF] ]
	dup [ 0 cell - ] Literal and @ swap 1 and
	IF  8>>  ELSE  $FF and  THEN
    [ [ELSE] ]
	dup [ cell  1- ] literal and 
	tuck - @ swap
	0 ?DO 8>> LOOP 255 and
    [ [THEN] ]
[ [THEN] ]
;
: 8>> 2/ 2/ 2/ 2/  2/ 2/ 2/ 2/ ;

c!	c c_addr --		core	cstore
*c_addr = c;
:
[ bigendian [IF] ]
    [ cell>bit 4 = [IF] ]
	tuck 1 and IF  $FF and  ELSE  8<<  THEN >r
	dup -2 and @ over 1 and cells masks + @ and
	r> or swap -2 and ! ;
	Create masks $00FF , $FF00 ,
    [ELSE] ]
	dup [ cell 1- ] literal and dup 
	[ cell 1- ] literal xor >r
	- dup @ $FF r@ 0 ?DO 8<< LOOP invert and
	rot $FF and r> 0 ?DO 8<< LOOP or swap ! ;
    [THEN]
[ELSE] ]
    [ cell>bit 4 = [IF] ]
	tuck 1 and IF  8<<  ELSE  $FF and  THEN >r
	dup -2 and @ over 1 and cells masks + @ and
	r> or swap -2 and ! ;
	Create masks $FF00 , $00FF ,
    [ELSE] ]
	dup [ cell 1- ] literal and dup >r
	- dup @ $FF r@ 0 ?DO 8<< LOOP invert and
	rot $FF and r> 0 ?DO 8<< LOOP or swap ! ;
    [THEN]
[THEN]
: 8<< 2* 2* 2* 2*  2* 2* 2* 2* ;

2!	w1 w2 a_addr --		core	two_store
a_addr[0] = w2;
a_addr[1] = w1;
:
 tuck ! cell+ ! ;

2@	a_addr -- w1 w2		core	two_fetch
w2 = a_addr[0];
w1 = a_addr[1];
:
 dup cell+ @ swap @ ;

cell+	a_addr1 -- a_addr2	core	cell_plus
a_addr2 = a_addr1+1;
:
 cell + ;

cells	n1 -- n2		core
n2 = n1 * sizeof(Cell);
:
 [ cell
 2/ dup [IF] ] 2* [ [THEN]
 2/ dup [IF] ] 2* [ [THEN]
 2/ dup [IF] ] 2* [ [THEN]
 2/ dup [IF] ] 2* [ [THEN]
 drop ] ;

char+	c_addr1 -- c_addr2	core	care_plus
c_addr2 = c_addr1 + 1;
:
 1+ ;

(chars)		n1 -- n2	gforth	paren_cares
n2 = n1 * sizeof(Char);
:
 ;

count	c_addr1 -- c_addr2 u	core
u = *c_addr1;
c_addr2 = c_addr1+1;
:
 dup 1+ swap c@ ;

(f83find)	c_addr u f83name1 -- f83name2	new	paren_f83find
for (; f83name1 != NULL; f83name1 = f83name1->next)
  if ((UCell)F83NAME_COUNT(f83name1)==u &&
      memcasecmp(c_addr, f83name1->name, u)== 0 /* or inline? */)
    break;
f83name2=f83name1;
:
    BEGIN  dup WHILE  (find-samelen)  dup  WHILE
	>r 2dup r@ cell+ char+ capscomp  0=
	IF  2drop r>  EXIT  THEN
	r> @
    REPEAT  THEN  nip nip ;
: (find-samelen) ( u f83name1 -- u f83name2/0 )
    BEGIN  2dup cell+ c@ $1F and <> WHILE  @  dup 0= UNTIL  THEN ;

\+has-hash [IF]

(hashfind)	c_addr u a_addr -- f83name2	new	paren_hashfind
F83Name *f83name1;
f83name2=NULL;
while(a_addr != NULL)
{
   f83name1=(F83Name *)(a_addr[1]);
   a_addr=(Cell *)(a_addr[0]);
   if ((UCell)F83NAME_COUNT(f83name1)==u &&
       memcasecmp(c_addr, f83name1->name, u)== 0 /* or inline? */)
     {
	f83name2=f83name1;
	break;
     }
}
:
 BEGIN  dup  WHILE
        2@ >r >r dup r@ cell+ c@ $1F and =
        IF  2dup r@ cell+ char+ capscomp 0=
	    IF  2drop r> rdrop  EXIT  THEN  THEN
	rdrop r>
 REPEAT nip nip ;

(tablefind)	c_addr u a_addr -- f83name2	new	paren_tablefind
""A case-sensitive variant of @code{(hashfind)}""
F83Name *f83name1;
f83name2=NULL;
while(a_addr != NULL)
{
   f83name1=(F83Name *)(a_addr[1]);
   a_addr=(Cell *)(a_addr[0]);
   if ((UCell)F83NAME_COUNT(f83name1)==u &&
       memcmp(c_addr, f83name1->name, u)== 0 /* or inline? */)
     {
	f83name2=f83name1;
	break;
     }
}
:
 BEGIN  dup  WHILE
        2@ >r >r dup r@ cell+ c@ $1F and =
        IF  2dup r@ cell+ char+ -text 0=
	    IF  2drop r> rdrop  EXIT  THEN  THEN
	rdrop r>
 REPEAT nip nip ;

(hashkey)	c_addr u1 -- u2		gforth	paren_hashkey
u2=0;
while(u1--)
   u2+=(Cell)toupper(*c_addr++);
:
 0 -rot bounds ?DO  I c@ toupper +  LOOP ;

(hashkey1)	c_addr u ubits -- ukey		gforth	paren_hashkey1
""ukey is the hash key for the string c_addr u fitting in ubits bits""
/* this hash function rotates the key at every step by rot bits within
   ubits bits and xors it with the character. This function does ok in
   the chi-sqare-test.  Rot should be <=7 (preferably <=5) for
   ASCII strings (larger if ubits is large), and should share no
   divisors with ubits.
*/
unsigned rot = ((char []){5,0,1,2,3,4,5,5,5,5,3,5,5,5,5,7,5,5,5,5,7,5,5,5,5,6,5,5,5,5,7,5,5})[ubits];
Char *cp = c_addr;
for (ukey=0; cp<c_addr+u; cp++)
    ukey = ((((ukey<<rot) | (ukey>>(ubits-rot))) 
	     ^ toupper(*cp))
	    & ((1<<ubits)-1));
:
 dup rot-values + c@ over 1 swap lshift 1- >r
 tuck - 2swap r> 0 2swap bounds
 ?DO  dup 4 pick lshift swap 3 pick rshift or
      I c@ toupper xor
      over and  LOOP
 nip nip nip ;
Create rot-values
  5 c, 0 c, 1 c, 2 c, 3 c,  4 c, 5 c, 5 c, 5 c, 5 c,
  3 c, 5 c, 5 c, 5 c, 5 c,  7 c, 5 c, 5 c, 5 c, 5 c,
  7 c, 5 c, 5 c, 5 c, 5 c,  6 c, 5 c, 5 c, 5 c, 5 c,
  7 c, 5 c, 5 c,

\+[THEN]

(parse-white)	c_addr1 u1 -- c_addr2 u2	gforth	paren_parse_white
/* use !isgraph instead of isspace? */
Char *endp = c_addr1+u1;
while (c_addr1<endp && isspace(*c_addr1))
  c_addr1++;
if (c_addr1<endp) {
  for (c_addr2 = c_addr1; c_addr1<endp && !isspace(*c_addr1); c_addr1++)
    ;
  u2 = c_addr1-c_addr2;
}
else {
  c_addr2 = c_addr1;
  u2 = 0;
}
:
 BEGIN  dup  WHILE  over c@ bl <=  WHILE  1 /string
 REPEAT  THEN  2dup
 BEGIN  dup  WHILE  over c@ bl >   WHILE  1 /string
 REPEAT  THEN  nip - ;

aligned		c_addr -- a_addr	core
a_addr = (Cell *)((((Cell)c_addr)+(sizeof(Cell)-1))&(-sizeof(Cell)));
:
 [ cell 1- ] Literal + [ -1 cells ] Literal and ;

faligned	c_addr -- f_addr	float	f_aligned
f_addr = (Float *)((((Cell)c_addr)+(sizeof(Float)-1))&(-sizeof(Float)));
:
 [ 1 floats 1- ] Literal + [ -1 floats ] Literal and ;

>body		xt -- a_addr	core	to_body
a_addr = PFA(xt);
:
    2 cells + ;

>code-address		xt -- c_addr		gforth	to_code_address
""c_addr is the code address of the word xt""
/* !! This behaves installation-dependently for DOES-words */
c_addr = (Address)CODE_ADDRESS(xt);
:
    @ ;

>does-code	xt -- a_addr		gforth	to_does_code
""If xt ist the execution token of a defining-word-defined word,
a_addr is the start of the Forth code after the DOES>;
Otherwise a_addr is 0.""
a_addr = (Cell *)DOES_CODE(xt);
:
    cell+ @ ;

code-address!		c_addr xt --		gforth	code_address_store
""Creates a code field with code address c_addr at xt""
MAKE_CF(xt, c_addr);
CACHE_FLUSH(xt,PFA(0));
:
    ! ;

does-code!	a_addr xt --		gforth	does_code_store
""creates a code field at xt for a defining-word-defined word; a_addr
is the start of the Forth code after DOES>""
MAKE_DOES_CF(xt, a_addr);
CACHE_FLUSH(xt,PFA(0));
:
    dodoes: over ! cell+ ! ;

does-handler!	a_addr --	gforth	does_handler_store
""creates a DOES>-handler at address a_addr. a_addr usually points
just behind a DOES>.""
MAKE_DOES_HANDLER(a_addr);
CACHE_FLUSH(a_addr,DOES_HANDLER_SIZE);
:
    drop ;

/does-handler	-- n	gforth	slash_does_handler
""the size of a does-handler (includes possible padding)""
/* !! a constant or environmental query might be better */
n = DOES_HANDLER_SIZE;
:
    2 cells ;

threading-method	-- n	gforth	threading_method
""0 if the engine is direct threaded. Note that this may change during
the lifetime of an image.""
#if defined(DOUBLY_INDIRECT)
n=2;
#else
# if defined(DIRECT_THREADED)
n=0;
# else
n=1;
# endif
#endif
:
 1 ;

\+has-os [IF]

(key)	-- n		gforth	paren_key
fflush(stdout);
/* !! noecho */
n = key();

key?	-- n		facility	key_q
fflush(stdout);
n = key_query;

stdout	-- wfileid	gforth
wfileid = (Cell)stdout;

stderr	-- wfileid	gforth
wfileid = (Cell)stderr;

form	-- urows ucols	gforth
""The number of lines and columns in the terminal. These numbers may change
with the window size.""
/* we could block SIGWINCH here to get a consistent size, but I don't
 think this is necessary or always beneficial */
urows=rows;
ucols=cols;

flush-icache	c_addr u --	gforth	flush_icache
""Make sure that the instruction cache of the processor (if there is
one) does not contain stale data at @var{c_addr} and @var{u} bytes
afterwards. @code{END-CODE} performs a @code{flush-icache}
automatically. Caveat: @code{flush-icache} might not work on your
installation; this is usually the case if direct threading is not
supported on your machine (take a look at your @file{machine.h}) and
your machine has a separate instruction cache. In such cases,
@code{flush-icache} does nothing instead of flushing the instruction
cache.""
FLUSH_ICACHE(c_addr,u);

(bye)	n --	gforth	paren_bye
return (Label *)n;

(system)	c_addr u -- wretval wior	gforth	peren_system
int old_tp=terminal_prepped;
deprep_terminal();
wretval=system(cstr(c_addr,u,1)); /* ~ expansion on first part of string? */
wior = IOR(wretval==-1 || (wretval==127 && errno != 0));
if (old_tp)
  prep_terminal();

getenv	c_addr1 u1 -- c_addr2 u2	gforth
c_addr2 = getenv(cstr(c_addr1,u1,1));
u2 = (c_addr2 == NULL ? 0 : strlen(c_addr2));

open-pipe	c_addr u ntype -- wfileid wior	gforth	open_pipe
wfileid=(Cell)popen(cstr(c_addr,u,1),fileattr[ntype]); /* ~ expansion of 1st arg? */
wior = IOR(wfileid==0); /* !! the man page says that errno is not set reliably */

close-pipe	wfileid -- wretval wior		gforth	close_pipe
wretval = pclose((FILE *)wfileid);
wior = IOR(wretval==-1);

time&date	-- nsec nmin nhour nday nmonth nyear	facility-ext	time_and_date
struct timeval time1;
struct timezone zone1;
struct tm *ltime;
gettimeofday(&time1,&zone1);
ltime=localtime((time_t *)&time1.tv_sec);
nyear =ltime->tm_year+1900;
nmonth=ltime->tm_mon+1;
nday  =ltime->tm_mday;
nhour =ltime->tm_hour;
nmin  =ltime->tm_min;
nsec  =ltime->tm_sec;

ms	n --	facility-ext
struct timeval timeout;
timeout.tv_sec=n/1000;
timeout.tv_usec=1000*(n%1000);
(void)select(0,0,0,0,&timeout);

allocate	u -- a_addr wior	memory
a_addr = (Cell *)malloc(u?u:1);
wior = IOR(a_addr==NULL);

free		a_addr -- wior		memory
free(a_addr);
wior = 0;

resize		a_addr1 u -- a_addr2 wior	memory
""Change the size of the allocated area at @i{a_addr1} to @i{u}
address units, possibly moving the contents to a different
area. @i{a_addr2} is the address of the resulting area. If
@code{a_addr1} is 0, Gforth's (but not the standard) @code{resize}
@code{allocate}s @i{u} address units.""
/* the following check is not necessary on most OSs, but it is needed
   on SunOS 4.1.2. */
if (a_addr1==NULL)
  a_addr2 = (Cell *)malloc(u);
else
  a_addr2 = (Cell *)realloc(a_addr1, u);
wior = IOR(a_addr2==NULL);	/* !! Define a return code */

strerror	n -- c_addr u	gforth
c_addr = strerror(n);
u = strlen(c_addr);

strsignal	n -- c_addr u	gforth
c_addr = strsignal(n);
u = strlen(c_addr);

call-c	w --	gforth	call_c
""Call the C function pointed to by @i{w}. The C function has to
access the stack itself. The stack pointers are exported in the global
variables @code{SP} and @code{FP}.""
/* This is a first attempt at support for calls to C. This may change in
   the future */
IF_FTOS(fp[0]=FTOS);
FP=fp;
SP=sp;
((void (*)())w)();
sp=SP;
fp=FP;
IF_TOS(TOS=sp[0]);
IF_FTOS(FTOS=fp[0]);

\+[THEN] ( has-os ) has-files [IF]

close-file	wfileid -- wior		file	close_file
wior = IOR(fclose((FILE *)wfileid)==EOF);

open-file	c_addr u ntype -- w2 wior	file	open_file
w2 = (Cell)fopen(tilde_cstr(c_addr, u, 1), fileattr[ntype]);
wior =  IOR(w2 == 0);

create-file	c_addr u ntype -- w2 wior	file	create_file
Cell	fd;
fd = open(tilde_cstr(c_addr, u, 1), O_CREAT|O_TRUNC|ufileattr[ntype], 0666);
if (fd != -1) {
  w2 = (Cell)fdopen(fd, fileattr[ntype]);
  wior = IOR(w2 == 0);
} else {
  w2 = 0;
  wior = IOR(1);
}

delete-file	c_addr u -- wior		file	delete_file
wior = IOR(unlink(tilde_cstr(c_addr, u, 1))==-1);

rename-file	c_addr1 u1 c_addr2 u2 -- wior	file-ext	rename_file
char *s1=tilde_cstr(c_addr2, u2, 1);
wior = IOR(rename(tilde_cstr(c_addr1, u1, 0), s1)==-1);

file-position	wfileid -- ud wior	file	file_position
/* !! use tell and lseek? */
ud = LONG2UD(ftell((FILE *)wfileid));
wior = IOR(UD2LONG(ud)==-1);

reposition-file	ud wfileid -- wior	file	reposition_file
wior = IOR(fseek((FILE *)wfileid, UD2LONG(ud), SEEK_SET)==-1);

file-size	wfileid -- ud wior	file	file_size
struct stat buf;
wior = IOR(fstat(fileno((FILE *)wfileid), &buf)==-1);
ud = LONG2UD(buf.st_size);

resize-file	ud wfileid -- wior	file	resize_file
wior = IOR(ftruncate(fileno((FILE *)wfileid), UD2LONG(ud))==-1);

read-file	c_addr u1 wfileid -- u2 wior	file	read_file
/* !! fread does not guarantee enough */
u2 = fread(c_addr, sizeof(Char), u1, (FILE *)wfileid);
wior = FILEIO(u2<u1 && ferror((FILE *)wfileid));
/* !! is the value of ferror errno-compatible? */
if (wior)
  clearerr((FILE *)wfileid);

read-line	c_addr u1 wfileid -- u2 flag wior	file	read_line
/*
Cell c;
flag=-1;
for(u2=0; u2<u1; u2++)
{
   *c_addr++ = (Char)(c = getc((FILE *)wfileid));
   if(c=='\n') break;
   if(c==EOF)
     {
	flag=FLAG(u2!=0);
	break;
     }
}
wior=FILEIO(ferror((FILE *)wfileid));
*/
if ((flag=FLAG(!feof((FILE *)wfileid) &&
	       fgets(c_addr,u1+1,(FILE *)wfileid) != NULL))) {
  wior=FILEIO(ferror((FILE *)wfileid)); /* !! ior? */
  if (wior)
    clearerr((FILE *)wfileid);
  u2 = strlen(c_addr);
  u2-=((u2>0) && (c_addr[u2-1]==NEWLINE));
}
else {
  wior=0;
  u2=0;
}

\+[THEN]  has-files [IF] -1 [ELSE] has-os [THEN] [IF]

write-file	c_addr u1 wfileid -- wior	file	write_file
/* !! fwrite does not guarantee enough */
{
  UCell u2 = fwrite(c_addr, sizeof(Char), u1, (FILE *)wfileid);
  wior = FILEIO(u2<u1 && ferror((FILE *)wfileid));
  if (wior)
    clearerr((FILE *)wfileid);
}

emit-file	c wfileid -- wior	gforth	emit_file
wior = FILEIO(putc(c, (FILE *)wfileid)==EOF);
if (wior)
  clearerr((FILE *)wfileid);

\+[THEN]  has-files [IF]

flush-file	wfileid -- wior		file-ext	flush_file
wior = IOR(fflush((FILE *) wfileid)==EOF);

file-status	c_addr u -- ntype wior	file-ext	file_status
char *filename=tilde_cstr(c_addr, u, 1);
if (access (filename, F_OK) != 0) {
  ntype=0;
  wior=IOR(1);
}
else if (access (filename, R_OK | W_OK) == 0) {
  ntype=2; /* r/w */
  wior=0;
}
else if (access (filename, R_OK) == 0) {
  ntype=0; /* r/o */
  wior=0;
}
else if (access (filename, W_OK) == 0) {
  ntype=4; /* w/o */
  wior=0;
}
else {
  ntype=1; /* well, we cannot access the file, but better deliver a legal
	    access mode (r/o bin), so we get a decent error later upon open. */
  wior=0;
}

\+[THEN] ( has-files ) has-floats [IF]

comparisons(f, r1 r2, f_, r1, r2, gforth, gforth, float, gforth)
comparisons(f0, r, f_zero_, r, 0., float, gforth, float, gforth)

d>f		d -- r		float	d_to_f
#ifdef BUGGY_LONG_LONG
extern double ldexp(double x, int exp);
r = ldexp((Float)d.hi,CELL_BITS) + (Float)d.lo;
#else
r = d;
#endif

f>d		r -- d		float	f_to_d
#ifdef BUGGY_LONG_LONG
d.hi = ldexp(r,-CELL_BITS) - (r<0);
d.lo = r-ldexp((Float)d.hi,CELL_BITS);
#else
d = r;
#endif

f!		r f_addr --	float	f_store
*f_addr = r;

f@		f_addr -- r	float	f_fetch
r = *f_addr;

df@		df_addr -- r	float-ext	d_f_fetch
#ifdef IEEE_FP
r = *df_addr;
#else
!! df@
#endif

df!		r df_addr --	float-ext	d_f_store
#ifdef IEEE_FP
*df_addr = r;
#else
!! df!
#endif

sf@		sf_addr -- r	float-ext	s_f_fetch
#ifdef IEEE_FP
r = *sf_addr;
#else
!! sf@
#endif

sf!		r sf_addr --	float-ext	s_f_store
#ifdef IEEE_FP
*sf_addr = r;
#else
!! sf!
#endif

f+		r1 r2 -- r3	float	f_plus
r3 = r1+r2;

f-		r1 r2 -- r3	float	f_minus
r3 = r1-r2;

f*		r1 r2 -- r3	float	f_star
r3 = r1*r2;

f/		r1 r2 -- r3	float	f_slash
r3 = r1/r2;

f**		r1 r2 -- r3	float-ext	f_star_star
""@i{r3} is @i{r1} raised to the @i{r2}th power""
r3 = pow(r1,r2);

fnegate		r1 -- r2	float
r2 = - r1;

fdrop		r --		float

fdup		r -- r r	float

fswap		r1 r2 -- r2 r1	float

fover		r1 r2 -- r1 r2 r1	float

frot		r1 r2 r3 -- r2 r3 r1	float

fnip		r1 r2 -- r2	gforth

ftuck		r1 r2 -- r2 r1 r2	gforth

float+		f_addr1 -- f_addr2	float	float_plus
f_addr2 = f_addr1+1;

floats		n1 -- n2	float
n2 = n1*sizeof(Float);

floor		r1 -- r2	float
""round towards the next smaller integral value, i.e., round toward negative infinity""
/* !! unclear wording */
r2 = floor(r1);

fround		r1 -- r2	float
""round to the nearest integral value""
/* !! unclear wording */
#ifdef HAVE_RINT
r2 = rint(r1);
#else
r2 = floor(r1+0.5);
/* !! This is not quite true to the rounding rules given in the standard */
#endif

fmax		r1 r2 -- r3	float
if (r1<r2)
  r3 = r2;
else
  r3 = r1;

fmin		r1 r2 -- r3	float
if (r1<r2)
  r3 = r1;
else
  r3 = r2;

represent		r c_addr u -- n f1 f2	float
char *sig;
int flag;
int decpt;
sig=ecvt(r, u, &decpt, &flag);
n=(r==0 ? 1 : decpt);
f1=FLAG(flag!=0);
f2=FLAG(isdigit(sig[0])!=0);
memmove(c_addr,sig,u);

>float	c_addr u -- flag	float	to_float
/* real signature: c_addr u -- r t / f */
Float r;
char *number=cstr(c_addr, u, 1);
char *endconv;
while(isspace(number[--u]) && u>0);
switch(number[u])
{
   case 'd':
   case 'D':
   case 'e':
   case 'E':  break;
   default :  u++; break;
}
number[u]='\0';
r=strtod(number,&endconv);
if((flag=FLAG(!(Cell)*endconv)))
{
   IF_FTOS(fp[0] = FTOS);
   fp += -1;
   FTOS = r;
}
else if(*endconv=='d' || *endconv=='D')
{
   *endconv='E';
   r=strtod(number,&endconv);
   if((flag=FLAG(!(Cell)*endconv)))
     {
	IF_FTOS(fp[0] = FTOS);
	fp += -1;
	FTOS = r;
     }
}

fabs		r1 -- r2	float-ext
r2 = fabs(r1);

facos		r1 -- r2	float-ext
r2 = acos(r1);

fasin		r1 -- r2	float-ext
r2 = asin(r1);

fatan		r1 -- r2	float-ext
r2 = atan(r1);

fatan2		r1 r2 -- r3	float-ext
""@i{r1/r2}=tan@i{r3}. The standard does not require, but probably
intends this to be the inverse of @code{fsincos}. In gforth it is.""
r3 = atan2(r1,r2);

fcos		r1 -- r2	float-ext
r2 = cos(r1);

fexp		r1 -- r2	float-ext
r2 = exp(r1);

fexpm1		r1 -- r2	float-ext
""@i{r2}=@i{e}**@i{r1}@minus{}1""
#ifdef HAVE_EXPM1
extern double expm1(double);
r2 = expm1(r1);
#else
r2 = exp(r1)-1.;
#endif

fln		r1 -- r2	float-ext
r2 = log(r1);

flnp1		r1 -- r2	float-ext
""@i{r2}=ln(@i{r1}+1)""
#ifdef HAVE_LOG1P
extern double log1p(double);
r2 = log1p(r1);
#else
r2 = log(r1+1.);
#endif

flog		r1 -- r2	float-ext
""the decimal logarithm""
r2 = log10(r1);

falog		r1 -- r2	float-ext
""@i{r2}=10**@i{r1}""
extern double pow10(double);
r2 = pow10(r1);

fsin		r1 -- r2	float-ext
r2 = sin(r1);

fsincos		r1 -- r2 r3	float-ext
""@i{r2}=sin(@i{r1}), @i{r3}=cos(@i{r1})""
r2 = sin(r1);
r3 = cos(r1);

fsqrt		r1 -- r2	float-ext
r2 = sqrt(r1);

ftan		r1 -- r2	float-ext
r2 = tan(r1);
:
 fsincos f/ ;

fsinh		r1 -- r2	float-ext
r2 = sinh(r1);
:
 fexpm1 fdup fdup 1. d>f f+ f/ f+ f2/ ;

fcosh		r1 -- r2	float-ext
r2 = cosh(r1);
:
 fexp fdup 1/f f+ f2/ ;

ftanh		r1 -- r2	float-ext
r2 = tanh(r1);
:
 f2* fexpm1 fdup 2. d>f f+ f/ ;

fasinh		r1 -- r2	float-ext
r2 = asinh(r1);
:
 fdup fdup f* 1. d>f f+ fsqrt f/ fatanh ;

facosh		r1 -- r2	float-ext
r2 = acosh(r1);
:
 fdup fdup f* 1. d>f f- fsqrt f+ fln ;

fatanh		r1 -- r2	float-ext
r2 = atanh(r1);
:
 fdup f0< >r fabs 1. d>f fover f- f/  f2* flnp1 f2/
 r> IF  fnegate  THEN ;

sfloats		n1 -- n2	float-ext	s_floats
n2 = n1*sizeof(SFloat);

dfloats		n1 -- n2	float-ext	d_floats
n2 = n1*sizeof(DFloat);

sfaligned	c_addr -- sf_addr	float-ext	s_f_aligned
sf_addr = (SFloat *)((((Cell)c_addr)+(sizeof(SFloat)-1))&(-sizeof(SFloat)));
:
 [ 1 sfloats 1- ] Literal + [ -1 sfloats ] Literal and ;

dfaligned	c_addr -- df_addr	float-ext	d_f_aligned
df_addr = (DFloat *)((((Cell)c_addr)+(sizeof(DFloat)-1))&(-sizeof(DFloat)));
:
 [ 1 dfloats 1- ] Literal + [ -1 dfloats ] Literal and ;

\ The following words access machine/OS/installation-dependent
\   Gforth internals
\ !! how about environmental queries DIRECT-THREADED,
\   INDIRECT-THREADED, TOS-CACHED, FTOS-CACHED, CODEFIELD-DOES */

\ local variable implementation primitives
\+[THEN] ( has-floats ) has-locals [IF]

@local#		-- w	gforth	fetch_local_number
w = *(Cell *)(lp+(Cell)NEXT_INST);
INC_IP(1);

@local0	-- w	new	fetch_local_zero
w = *(Cell *)(lp+0*sizeof(Cell));

@local1	-- w	new	fetch_local_four
w = *(Cell *)(lp+1*sizeof(Cell));

@local2	-- w	new	fetch_local_eight
w = *(Cell *)(lp+2*sizeof(Cell));

@local3	-- w	new	fetch_local_twelve
w = *(Cell *)(lp+3*sizeof(Cell));

\+has-floats [IF]

f@local#	-- r	gforth	f_fetch_local_number
r = *(Float *)(lp+(Cell)NEXT_INST);
INC_IP(1);

f@local0	-- r	new	f_fetch_local_zero
r = *(Float *)(lp+0*sizeof(Float));

f@local1	-- r	new	f_fetch_local_eight
r = *(Float *)(lp+1*sizeof(Float));

\+[THEN]

laddr#		-- c_addr	gforth	laddr_number
/* this can also be used to implement lp@ */
c_addr = (Char *)(lp+(Cell)NEXT_INST);
INC_IP(1);

lp+!#	--	gforth	lp_plus_store_number
""used with negative immediate values it allocates memory on the
local stack, a positive immediate argument drops memory from the local
stack""
lp += (Cell)NEXT_INST;
INC_IP(1);

lp-	--	new	minus_four_lp_plus_store
lp += -sizeof(Cell);

lp+	--	new	eight_lp_plus_store
lp += sizeof(Float);

lp+2	--	new	sixteen_lp_plus_store
lp += 2*sizeof(Float);

lp!	c_addr --	gforth	lp_store
lp = (Address)c_addr;

>l	w --	gforth	to_l
lp -= sizeof(Cell);
*(Cell *)lp = w;

\+has-floats [IF]

f>l	r --	gforth	f_to_l
lp -= sizeof(Float);
*(Float *)lp = r;

\+[THEN]  [THEN] \ has-locals

up!	a_addr --	gforth	up_store
UP=up=(char *)a_addr;
:
 up ! ;
Variable UP