File psl-1983/3-1/util/nbig0.red artifact 0119817a7e part of check-in 955d0a90a7


% NBIG0.RED - Vector based BIGNUM package with INUM operations
%     M. L. Griss & B Morrison,  25 June 1982.
%     Copyright (C) 1982, A. C. Norman, B. Morrison, M. Griss
%
% Revision log:
% 10 March, 1983, MLG
%   LSH in Twopower replaced by 2**n
%   Fixed a bug in SYS2BIG that did not convert negative BIGNUMS correctly
% 7 February 1983, MLG
%     Merged in NBIG1 (see its "revision history" below), plus clean-up.
%     Revision History of old NBIG1:
% 28 Dec 1982, MLG:
%	Added BigZeroP and BigOneP for NArith
%	Changed Name to NBIG1.RED from BIGFACE
% 22 Dec 1982, MLG:
%	Change way of converting from VECT to BIGN
%	Move Module dependency to .BUILD file
%       Changes for NEW-ARITH, involve name changes for MAKEFIXNUM
%       ISINUM, etc.
% 21 December, 82: MLG
%	Change PRIN1 and PRIN2 hooks to refer to RecursiveChannelprinx
%       which changed in PK:PRINTERS.RED for prinlevel stuff
%     November: Variety of Bug Fixes by A. Norman
%     Use the BIGN tag for better Interface
%
% 31 Dec 1982, MLG
%     Changed BNUM to check if arg ALREADY Big. Kludge
%     since new NARITH makes some things BIG earlier
%     since it calls the BIG funcs directly
% 20 Dec 1982, MLG
%     Changed TrimBigNUM to TrimBigNum1 in BhardDivide
%
% 14 Dec 1982, MLG
%     Changed to put LOAD and IMPORTS in BUILD file
%
% 31 August 1982, A. C . Norman
%     Adjustments to many routines: in particular corrections to BHardDivide
%     (case D6 utterly wrong), and adjustments to BExpt (for performance) and
%     all logical operators (for treatment of negative inputs);
% ---------------------------------------------------------------

% -----------------------
% A bignum will be a VECTOR of Bigits: (digits in base BigBase):
%  [BIGPOS b1 ... bn] or [BIGNEG b1 ... bn].  BigZero is thus [BIGPOS]
% All numbers are positive, with BIGNEG as 0 element to indicate negatives.

% BETA.RED - some values of BETA testing
% On DEC-20, Important Ranges are:
%  		--------------------------------           
% POSBETA       |    0          |    n         |
%  		--------------------------------           
%                  19                17 	bits
%  		--------------------------------           
% NEGBETA       |    -1         |              |
%  		--------------------------------           
%
%  		--------------------------------           
% POSINT        |    0    | 0  |               |
%  		--------------------------------           
%                 5         13       18        	bits 
%  		--------------------------------           
% NEGINT        |    -1   | -1 |               |
%  		--------------------------------           
% Thus BETA:  2^17-1       -131072 ... 131071
%      INT    2^18-1       -262144 ... 262143
%      FIX    2^35-1  -34359738368 ... 34359738367
%       [Note that one bit used for sign in 36 bit word]

fluid '(BigBetaHi!* 	% Largest BetaNum in BIG format
	BigBetaLow!* 	% Smallest BetaNum in BIG format
	BetaHi!* 	% Largest BetaNum as Inum
	BetaLow!* 	% Smallest BetaNum as Inum
	SysHi!* 	% Largest SYSINT in FixN format
	SysLow!* 	% Smallest SYSINT in FixN format
	BigSysHi!* 	% Largest SYSINT in BIG format
	BigSysLow!* 	% Smallest SYSINT in BIG format
	FloatSysHi!* 	% Largest SYSINT in Float format
	FloatSysLow!* 	% Smallest SYSINT in Float format
	BBase!* 	% BETA, base of system
	FloatBbase!*    % As a float
	BigFloatHi!* 	% Largest  Float in BIG format
	BigFloatLow!*	% Smallest Float in BIG format
	StaticBig!*	% Warray for conversion of SYS to BIG
	Bone!*          % A one
	Bzero!*		% A zero
	BBits!*         % Number of Bits in BBASE!*
	LogicalBits!*   
	Digit2Letter!*
	Carry!* 
	OutputBase!*
);

% --------------------------------------------------------------------------
% --------------------------------------------------------------------------
% Support functions:
%
% U, V, V1, V2 for arguments are Bignums.  Other arguments are usually
% fix/i-nums.

smacro procedure PutBig(b,i,val);
% Access elements of a BIGNUM
  IputV(b,i,val);

smacro procedure GetBig(b,i);
% Access elements of a BIGNUM
  IgetV(B,i);

procedure setbits x;
%
% This function sets the globals for big bignum package.
% "x" should be total # of bits per word.
Begin scalar y;
  BBits!*:=iquotient(isub1 x,2); % Total number of bits per word used.
  BBase!*:=TwoPower BBits!*;	 % "Beta", where n=A0 + A1*beta + A2*(beta^2).
  FloatBbase!* := IntFloat Bbase!*;
  LogicalBits!*:=ISub1 BBase!*;	 % Used in LAnd,Lor, etc.
  BetaHi!*:=isub1 Bbase!*;     
  BetaLow!* :=Iminus Bbase!*;
  Bone!* := Bnum 1;
  Bzero!* := Bnum 0;
  BigBetaHi!*:=BNum BetaHi!*; 	        % Highest value of Ai
  BigBetaLow!*:=BMinus BigBetaHi!*;	% Lowest value of Ai
 % here assume 2's complement

  y:=TwoPower idifference (x,2);        % eg, 36 bits, 2^35-1=2^34+2^34-1
  SysHi!*   :=y+(y-1);
  y:=-y;
  Syslow!*  :=y+y;
  BigSysHi!*:=bdifference(btwopower isub1 x,
	               Bone!*);   % Largest representable Syslisp integer.
	% Note that SYSPOS has leading 0, ie only x-1 active bits
  BigSysLow!*:=BMinus BPlus2(Bone!*, BigSysHi!*);
	% Smallest representable Syslisp integer.
end;

procedure NonBigNumError(V,L);
  StdError BldMsg(" Expect a BIGNUM in %r, given %p%n",L,V);

procedure BSize V;
% Upper Limit of [BIGxxx a1 ... An]
  If BigP V then VecLen VecInf V else 0;

procedure GtPOS N;
% Allocate [BIGPOS a1 ... an]
 Begin 
    N:=MkVect N;
    IPutV(N,0,'BIGPOS);
    Return MkBigN Vecinf N;
 End;
 
procedure GtNeg N;
% Allocate [BIGNEG a1 ... an]
 Begin 
    N:=MkVect N;
    IPutV(N,0,'BIGNEG);
    Return MkBigN VecInf N;
 End;
 
procedure TrimBigNum V3; 
% truncate trailing 0
 If Not BigP V3 then NonBigNumError(V3,'TrimBigNum)
   else TrimBigNum1(V3,BSize V3);

procedure TrimBigNum1(B,L3);
  Begin scalar v3;
     V3:=BigAsVec B;
     While IGreaterP(L3,0) and IZeroP IGetV(V3,L3) do L3:=ISub1 L3;
     If IZerop UpBv TruncateVector(V3,L3) then return GtPOS 0 
		else return B;
  end;

procedure BigAsVec B;
% In order to see BIGITS
 MkVec Inf B;

procedure VecAsBig V;
 MkBigN VecInf V;

Procedure BIG2Sys U;
% Convert a BIG to SYS, if in range
  If Blessp(U,BigSysLow!*) or Bgreaterp(U,BigSysHi!*) then
	ContinuableError(99,"BIGNUM too large to convert to SYS", U)
   else Big2SysAux U;

procedure Big2SysAux U;
% Convert a BIGN that is in range to a SYSINT
 begin scalar L,Sn,res;
  L:=BSize U;
  if IZeroP L then return 0;
  res:=IGetV(U,L);
  L:=ISub1 L;
  If BMinusP U then
   <<res:=-res;
     while L neq 0 do <<res:=ITimes2(res, Bbase!*);
	 	        res:=IDifference(res, IGetV(U,L));
		        L:=ISub1 L>>;
    >>
  else
     while L neq 0 do <<res:=ITimes2(res, Bbase!*);
	  	        res:=IPlus2(res, IGetV(U,L));
		        L:=ISub1 L>>;
  return Res;
 end;

procedure TwoPower N;	%fix/i-num 2**n
 2**n;

procedure BTwoPower N;	% gives 2**n; n is fix/i-num; result BigNum
 if not (fixp N or BigP N) then NonIntegerError(N, 'BTwoPower)
  else begin scalar quot, rem, V;
   if BigP N then n:=big2sys n;
   quot:=Quotient(N,Bbits!*);
   rem:=Remainder(N,Bbits!*);
   V:=GtPOS(IAdd1 quot);
   IFor i:=1:quot do IPutV(v,i,0);
   IPutV(V,IAdd1 quot,twopower rem);
   return TrimBigNum1(V,IAdd1 quot);
  end;

procedure BZeroP V1;
 IZerop BSize V1 and not BMinusP V1;

procedure BOneP V1;
 Not BMinusP V1 and IOneP (BSize V1) and IOneP IGetV(V1,1);

procedure BAbs V1;
 if BMinusP V1 then BMinus V1 else V1;

procedure BMax(V1,V2);
 if BGreaterP(V2,V1) then V2 else V1; 

procedure BMin(V1,V2);
 if BLessP(V2,V1) then V2 else V1;

procedure BExpt(V1,N);	
% V1 is Bignum, N is fix/i-num
 if not fixp N then NonIntegerError(N,'BEXPT)
 else if IZeroP N then Bone!*
 else if IOneP N then V1
 else if IMinusP N then BQuotient(Bone!*,BExpt(V1,IMinus N))
 else begin scalar V2;
    V2 := BExpt(V1,IQuotient(N,2));
    if IZeroP IRemainder(N,2) then return BTimes2(V2,V2)
    else return BTimes2(BTimes2(V2,V1),V2)
 end;


% ---------------------------------------
% Logical Operations
%
% All take Bignum arguments


procedure BLOr(V1,V2);
% The main body of the OR code is only obeyed when both arguments
% are positive, and so the result will be positive;
 if BMinusp V1 or BMinusp V2 then BLnot BLand(BLnot V1,BLnot V2)
 else begin scalar L1,L2,L3,V3;
     L1:=BSize V1;
     L2:=BSize V2;
     IF L2>L1 then <<L3:=L2; L2:=L1;L1:=L3;
                     V3:=V2; V2:=V1;V1:=V3>>;
     V3:=GtPOS L1;
     IFor I:=1:L2 do IPutV(V3,I,ILor(IGetV(V1,I),IGetV(V2,I)));
     IFor I:=(IAdd1 L2):L1 do IPutV(V3,i,IGetV(V1,I));
     Return V3
 end;

procedure BLXor(V1,V2);
% negative arguments are coped with using the identity
% LXor(a,b) = LNot LXor(Lnot a,b) = LNor LXor(a,Lnot b);
 begin scalar L1,L2,L3,V3,S;
     if BMinusp V1 then << V1 := BLnot V1; S := t >>;
     if BMinusp V2 then << V2 := BLnot V2; S := not S >>;
     L1:=BSize V1;
     L2:=BSize V2;
     IF L2>L1 then <<L3:=L2; L2:=L1;L1:=L3;
                     V3:=V2; V2:=V1;V1:=V3>>;
     V3:=GtPOS L1;
     IFor I:=1:L2 do IPutV(V3,I,ILXor(IGetV(V1,I),IGetV(V2,I)));
     IFor I:=(IAdd1 L2):L1 do IPutV(V3,i,IGetV(V1,I));
     V1:=TrimBigNum1(V3,L1);
     if S then V1:=BLnot V1;
     return V1
 end;

% Not Used Currently:
%
% procedure BLDiff(V1,V2);	
% ***** STILL NEEDS ADJUSTING WRT -VE ARGS *****
%  begin scalar V3,L1,L2;
%    L1:=BSize V1;
%    L2:=BSize V2;
%    V3:=GtPOS(max(L1,L2));
%    IFor i:=1:min(L1,L2) do 
% 	IPutV(V3,i,ILAnd(IGetV(V1,i),ILXor(LogicalBits!*,IGetV(V2,i))));
%    if IGreaterP(L1,L2) then IFor i:=(IAdd1 L2):L1 do IPutV(V3,i,IGetV(V1,i));
%    if IGreaterP(L2,L1) then IFor i:=(IAdd1 L1):L2 do IPutV(V3,i,0);
%    return TrimBigNum1(V3,max(L1,L2));
%  end;

procedure BLAnd(V1,V2);
% If both args are -ve the result will be too. Otherwise result will
% be positive;
 if BMinusp V1 and BMinusp V2 then BLnot BLor(BLnot V1,BLnot v2)
 else begin scalar L1,L2,L3,V3;
     L1:=BSize V1;
     L2:=BSize V2;
     L3:=Min(L1,L2);
     V3:=GtPOS L3;
     if BMinusp V1 then
       IFor I:=1:L3 do IPutV(V3,I,ILand(ILXor(Logicalbits!*,IGetV(V1,I)),
					IGetV(V2,I)))
     else if BMinusp V2 then
       IFor I:=1:L3 do IPutV(V3,I,ILand(IGetV(V1,I),
                                        ILXor(Logicalbits!*,IGetV(V2,I))))
     else IFor I:=1:L3 do IPutV(V3,I,ILand(IGetV(V1,I),IGetV(V2,I)));
     return TrimBigNum1(V3,L3);
 End;

procedure BLNot(V1);
 BMinus BSmallAdd(V1,1);

procedure BLShift(V1,V2);
% This seems a grimly inefficient way of doing things given that
% the representation of big numbers uses a base that is a power of 2.
% However it will do for now;
if BMinusP V2 then BQuotient(V1, BTwoPower BMinus V2)
  else BTimes2(V1, BTwoPower V2);



% -----------------------------------------
% Arithmetic Functions:
%
% U, V, V1, V2 are Bignum arguments.

procedure BMinus V1;	% Negates V1.
 if BZeroP V1 then V1
  else begin scalar L1,V2;
	L1:=BSize V1;
	if BMinusP V1 then V2 := GtPOS L1
	 else V2 := GtNEG L1;
	IFor I:=1:L1 do IPutV(V2,I,IGetV(V1,I));
	return V2;
  end;

% Returns V1 if V1 is strictly less than 0, NIL otherwise.
%
procedure BMinusP V1;
 if (IGetV(V1,0) eq 'BIGNEG) then V1 else NIL;

% To provide a conveninent ADD with CARRY.
procedure AddCarry A;
 begin scalar S;
   S:=IPlus2(A,Carry!*);
   if IGeq(S,BBase!*) then <<Carry!*:= 1; S:=IDifference(S,BBase!*)>>
    else Carry!*:=0;
   return S;
 end;

procedure BPlus2(V1,V2);
 begin scalar Sn1,Sn2;
     Sn1:=BMinusP V1;
     Sn2:=BMinusP V2;
     if Sn1 and Not Sn2 then return BDifference2(V2,BMinus V1,Nil);
     if Sn2 and Not Sn1 then return BDifference2(V1,BMinus V2,Nil);
     return BPlusA2(V1,V2,Sn1);
  end;

procedure BPlusA2(V1,V2,Sn1);	% Plus with signs pre-checked and
 begin scalar L1,L2,L3,V3,temp;		% identical.
     L1:=BSize V1;
     L2:=BSize V2;
     If IGreaterP(L2,L1) then <<L3:=L2; L2:=L1;L1:=L3;
				V3:=V2; V2:=V1;V1:=V3>>;
     L3:=IAdd1 L1;
     If Sn1 then V3:=GtNeg L3
      else V3:=GtPOS L3;
     Carry!*:=0;
     IFor I:=1:L2 do <<temp:=IPlus2(IGetV(V1,I),IGetV(V2,I));
			IPutV(V3,I,AddCarry temp)>>;
     temp:=IAdd1 L2;
     IFor I:=temp:L1 do IPutV(V3,I,AddCarry IGetV(V1,I));
     IPutV(V3,L3,Carry!*); % Carry Out
     Return TrimBigNum1(V3,L3);
 end;

procedure BDifference(V1,V2);
 if BZeroP V2 then V1
  else if BZeroP V1 then BMinus V2
  else begin scalar Sn1,Sn2;
     Sn1:=BMinusP V1;
     Sn2:=BMinusP V2;
     if (Sn1 and Not Sn2) or (Sn2 and Not Sn1) 
	then return BPlusA2(V1,BMinus V2,Sn1);
     return BDifference2(V1,V2,Sn1);
  end;

procedure SubCarry A;
 begin scalar S;
  S:=IDifference(A,Carry!*);
  if ILessP(S,0) then <<Carry!*:=1; S:=IPlus2(BBase!*,S)>> else Carry!*:=0;
  return S;
 end;

Procedure BDifference2(V1,V2,Sn1);  % Signs pre-checked and identical.
 begin scalar i,L1,L2,L3,V3;
  L1:=BSize V1;
  L2:=BSize V2;
  if IGreaterP(L2,L1) then <<L3:=L1;L1:=L2;L2:=L3;
			V3:=V1;V1:=V2;V2:=V3; Sn1:=not Sn1>>
   else if L1 Eq L2 then <<i:=L1;
		while (IGetV(V2,i) Eq IGetV(V1,i) and IGreaterP(i,1))
		  do i:=ISub1 i;
		if IGreaterP(IGetV(V2,i),IGetV(V1,i)) 
		   then <<L3:=L1;L1:=L2;L2:=L3;
			V3:=V1;V1:=V2;V2:=V3;Sn1:=not Sn1>> >>;
  if Sn1 then V3:=GtNEG L1
   else V3:=GtPOS L1;
  carry!*:=0;
  IFor I:=1:L2 do IPutV(V3,I,SubCarry IDifference(IGetV(V1,I),IGetV(V2,I)));
  IFor I:=(IAdd1 L2):L1 do IPutV(V3,I,SubCarry IGetV(V1,I));
  return TrimBigNum1(V3,L1);
 end;

procedure BTimes2(V1,V2);
 begin scalar L1,L2,L3,Sn1,Sn2,V3;
    L1:=BSize V1;
    L2:=BSize V2;
    if IGreaterP(L2,L1)
	 then <<V3:=V1; V1:=V2; V2:=V3;   % If V1 is larger, will be fewer
		L3:=L1; L1:=L2; L2:=L3>>; % iterations of BDigitTimes2.
    L3:=IPlus2(L1,L2);
    Sn1:=BMinusP V1;
    Sn2:=BMinusP V2;
    If (Sn1 and Sn2) or not(Sn1 or Sn2) then V3:=GtPOS L3 else V3:=GtNEG L3;
    IFor I:=1:L3 do IPutV(V3,I,0);
    IFor I:=1:L2 do BDigitTimes2(V1,IGetV(V2,I),L1,I,V3);
    return TrimBigNum1(V3,L3);
  end;

Procedure BDigitTimes2(V1,V2,L1,I,V3);
% V1 is a bignum, V2 a fixnum, L1=BSize L1, I=position of V2 in a bignum,
% and V3 is bignum receiving result.  I affects where in V3 the result of
% a calculation goes; the relationship is that positions I:I+(L1-1)
% of V3 receive the products of V2 and positions 1:L1 of V1.
% V3 is changed as a side effect here.
 begin scalar J,carry,temp1,temp2;
 if zerop V2 then return V3
  else <<
	carry:=0;
	IFor H:=1:L1 do <<
	    temp1:=ITimes2(IGetV(V1,H),V2);
	    temp2:=IPlus2(H,ISub1 I);
	    J:=IPlus2(IPlus2(temp1,IGetV(V3,temp2)),carry);
	    IPutV(V3,temp2,IRemainder(J,BBase!*));
	    carry:=IQuotient(J,BBase!*)>>;
	IPutV(V3,IPlus2(L1,I),carry)>>; % carry should be < BBase!* here 
    return V3;
 end;

Procedure BSmallTimes2(V1,C);	% V1 is a BigNum, C a fixnum.
					% Assume C positive, ignore sign(V1)
					% also assume V1 neq 0.
 if ZeroP C then return GtPOS 0		% Only used from BHardDivide, BReadAdd.
  else begin scalar J,carry,L1,L2,L3,V3;
   L1:=BSize V1;
   L2:=IPlus2(IQuotient(C,BBase!*),L1);
   L3:=IAdd1 L2;
   V3:=GtPOS L3;
   carry:=0;
   IFor H:=1:L1 do <<
	J:=IPlus2(ITimes2(IGetV(V1,H),C),carry);
	IPutV(V3,H,IRemainder(J,BBase!*));
	carry:=IQuotient(J,BBase!*)>>;
   IFor H:=(IAdd1 L1):L3 do <<
	IPutV(V3,H,IRemainder(J:=carry,BBase!*));
        carry:=IQuotient(J,BBase!*)>>;
   return TrimBigNum1(V3,L3);
 end;

procedure BQuotient(V1,V2);
 car BDivide(V1,V2);

procedure BRemainder(V1,V2);
 cdr BDivide(V1,V2);

% BDivide returns a dotted pair, (Q . R).  Q is the quotient and R is 
% the remainder.  Both are bignums.  R is of the same sign as V1.
%;

smacro procedure BSimpleQuotient(V1,L1,C,SnC);
 car BSimpleDivide(V1,L1,C,SnC);

smacro procedure BSimpleRemainder(V1,L1,C,SnC);
 cdr BSimpleDivide(V1,L1,C,SnC);

procedure BDivide(V1,V2);
 begin scalar L1,L2,Q,R,V3;
     L2:=BSize V2;
     If IZerop L2 then error(99, "Attempt to divide by 0 in BDIVIDE");
     L1:=BSize V1;
     If ILessP(L1,L2) or (L1 Eq L2 and ILessP(IGetV(V1,L1),IGetV(V2,L2)))
					% This also takes care of case
	then return (GtPOS 0 . V1);	% when V1=0.
     if IOnep L2 then return BSimpleDivide(V1,L1,IGetV(V2,1),BMinusP V2);
     return BHardDivide(V1,L1,V2,L2);
  end;


% C is a fixnum (inum?); V1 is a bignum and L1 is its length.
% SnC is T if C (which is positive) should be considered negative.
% Returns quotient . remainder; each is a bignum.
%
procedure BSimpleDivide(V1,L1,C,SnC);
 begin scalar I,P,R,RR,Sn1,V2;
  Sn1:=BMinusP V1;
  if (Sn1 and SnC) or not(Sn1 or SnC) then V2:=GtPOS L1 else V2:=GtNEG L1;
  R:=0;
  I:=L1;
  While not IZeroP I do <<P:=IPlus2(ITimes2(R,BBase!*),IGetV(V1,I));
							% Overflow.
		    IPutV(V2,I,IQuotient(P, C));
		    R:=IRemainder(P, C);
		    I:=ISub1 I>>;
  If Sn1 then RR:=GtNeg 1 else RR:=GtPOS 1;
  IPutV(RR,1,R);
  return (TrimBigNum1(V2,L1) . TrimBigNum1(RR,1));
 end;


procedure BHardDivide(U,Lu,V,Lv);
% This is an algorithm taken from Knuth.
 begin scalar U1,V1,A,D,LCV,LCV1,f,f2,J,K,Lq,carry,temp,
	      LL,M,N,N1,P,Q,QBar,SnU,SnV,U2;
     N:=Lv;
     N1:=IAdd1 N;
     M:=IDifference(Lu,Lv);
     Lq:=IAdd1 M;

     % Deal with signs of inputs;

     SnU:=BMinusP U;
     SnV:=BMinusp V;  % Note that these are not extra-boolean, i.e.
		      % for positive numbers MBinusP returns nil, for
		      % negative it returns its argument. Thus the
		      % test (SnU=SnV) does not reliably compare the signs of
		      % U and V;
     if SnU then if SnV then Q := GtPOS Lq else Q := GtNEG Lq
        else if SnV then Q := GtNEG Lq else Q := GtPOS Lq;

     U1 := GtPOS IAdd1 Lu;  % U is ALWAYS stored as if one digit longer;

     % Compute a scale factor to normalize the long division;
     D:=IQuotient(BBase!*,IAdd1 IGetV(V,Lv));
     % Now, at the same time, I remove the sign information from U and V
     % and scale them so that the leading coefficeint in V is fairly large;

     carry := 0;
     IFor i:=1:Lu do <<
	 temp := IPlus2(ITimes2(IGetV(U,I),D),carry);
	 IPutV(U1,I,IRemainder(temp,BBase!*));
	 carry := IQuotient(temp,BBase!*) >>;
     Lu := IAdd1 Lu;
     IPutV(U1,Lu,carry);

     V1:=BSmallTimes2(V,D);  % So far all variables contain safe values,
			     % i.e. numbers < BBase!*;
     IPutV(V1,0,'BIGPOS);

     if ILessp(Lv,2) then NonBigNumError(V,'BHARDDIVIDE); % To be safe;

     LCV := IGetV(V1,Lv);
     LCV1 := IGetv(V1,ISub1 Lv); % Top two digits of the scaled V accessed once
				 % here outside the main loop;

     % Now perform the main long division loop;

     IFor I:=0:M do <<
		J:=IDifference(Lu,I); 	        % J>K; working on U1[K:J] 
		K:=IDifference(J,N1);		% in this loop.
		A:=IGetV(U1,J);

		P := IPlus2(ITimes2(A,BBase!*),IGetv(U1,Isub1 J));
		   % N.B. P is up to 30 bits long. Take care! ;

		if A Eq LCV then QBar := ISub1 BBase!*
		else QBar := Iquotient(P,LCV);  % approximate next digit;

		f:=ITimes2(QBar,LCV1);
		f2:=IPlus2(ITimes2(IDifference(P,ITimes2(QBar,LCV)),BBase!*),
			   IGetV(U1,IDifference(J,2)));

		while IGreaterP(f,f2) do << % Correct most overshoots in Qbar;
			QBar:=ISub1 QBar;
			f:=IDifference(f,LCV1);;
		        f2:=IPlus2(f2,ITimes2(LCV,BBase!*)) >>;

		carry := 0;    % Ready to subtract QBar*V1 from U1;

		IFor L:=1:N do <<
		    temp := IPlus2(
				Idifference(
				   IGetV(U1,IPlus2(K,L)),
				   ITimes2(QBar,IGetV(V1,L))),
		                carry);
                    carry := IQuotient(temp,BBase!*);
		    temp := IRemainder(temp,BBase!*);
		    if IMinusp temp then <<
		       carry := ISub1 carry;
		       temp := IPlus2(temp,BBase!*) >>;
                    IPutV(U1,IPlus2(K,L),temp) >>;

		% Now propagate borrows up as far as they go;

                LL := IPlus2(K,N);
		while (not IZeroP carry) and ILessp(LL,J) do <<
		    LL := IAdd1 LL;
		    temp := IPlus2(IGetV(U1,LL),carry);
		    carry := IQuotient(temp,BBase!*);
		    temp := IRemainder(temp,BBase!*);
		    if IMinusP temp then <<
			carry := ISub1 carry;
			temp := IPlus2(temp,BBase!*) >>;
                    IPutV(U1,LL,temp) >>;

                if not IZerop carry then <<
		   % QBar was still wrong - correction step needed.
		   % This should not happen very often;
		   QBar := ISub1 QBar;

		   % Add V1 back into U1;
		   carry := 0;

		   IFor L := 1:N do <<
		       carry := IPlus2(
				   IPlus2(IGetV(U1,Iplus2(K,L)),
				          IGetV(V1,L)),
                                   carry);
                       IPutV(U1,IPlus2(K,L),IRemainder(carry,BBase!*));
		       carry := IQuotient(carry,BBase!*) >>;

                   LL := IPlus2(K,N);
		   while ILessp(LL,J) do <<
		       LL := IAdd1 LL;
		       carry := IPlus2(IGetv(U1,LL),carry);
		       IPutV(U1,LL,IRemainder(carry,BBase!*));
		       carry := IQuotient(carry,BBase!*) >> >>;

                IPutV(Q,IDifference(Lq,I),QBar)

		>>;        % End of main loop;


     U1 := TrimBigNum1(U1,IDifference(Lu,M));

     f := 0; f2 := 0; % Clean up potentially wild values;

     if not BZeroP U1 then <<
	% Unnormalize the remainder by dividing by D

        if SnU then IPutV(U1,0,'BIGNEG);
        if not IOnep D then <<
	    Lu := BSize U1;
	    carry := 0;
	    IFor L:=Lu step -1 until 1 do <<
	         P := IPlus2(ITimes2(carry,BBase!*),IGetV(U1,L));
	         IPutv(U1,L,IQuotient(P,D));
	         carry := IRemainder(P,D) >>;
     
	    P := 0;
	    if not IZeroP carry then BHardBug("remainder when unscaling",
	                            U,V,TrimBigNum1(U1,Lu),TrimBigNum1(Q,Lq));

	    U1 := TrimBigNum1(U1,Lu) >> >>;

     Q := TrimBigNum1(Q,Lq);     % In case leading digit happened to be zero;
     P := 0;  % flush out a 30 bit number;

% Here, for debugging purposes, I will try to validate the results I
% have obtained by testing if Q*V+U1=U and 0<=U1<V. I Know this slows things
% down, but I will remove it when my confidence has improved somewhat;

%    if not BZerop U1 then <<
%       if (BMinusP U and not BMinusP U1) or
%           (BMinusP U1 and not BMinusP U) then
%                  BHardBug("remainder has wrong sign",U,V,U1,Q) >>;
%    if not BAbs U1<BAbs V then BHardBug("remainder out of range",U,V,U1,Q)
%    else if not BZerop(BDifference(BPlus2(BTimes2(Q,V),U1),U)) then 
%         BHardBug("quotient or remainder incorrect",U,V,U1,Q);

     return (Q . U1)
  end;

procedure BHardBug(msg,U,V,R,Q);
% Because the inputs to BHardDivide are probably rather large, I am not
% going to rely on BldMsg to display them;
 << Prin2T "***** Internal error in BHardDivide";
    Prin2 "arg1="; Prin2T U;
    Prin2 "arg2="; Prin2T V;
    Prin2 "computed quotient="; Prin2T Q;
    Prin2 "computed remainder="; Prin2T R;
    StdError msg >>;


procedure BGreaterP(U,V);
    if BMinusP U then
       if BMinusP V then BUnsignedGreaterP(V,U)
       else nil
    else if BMinusP V then U
       else BUnsignedGreaterP(U,V);

procedure BLessp(U,V);
    if BMinusP U then
       if BMinusP V then BUnsignedGreaterP(U,V)
       else U
    else if BMinusP V then nil
       else BUnsignedGreaterP(V,U);

procedure BGeq(U,V);
    if BMinusP U then
       if BMinusP V then BUnsignedGeq(V,U)
       else nil
    else if BMinusP V then U
       else BUnsignedGeq(U,V);

procedure BLeq(U,V);
    if BMinusP U then
       if BMinusP V then BUnsignedGeq(U,V)
       else U
    else if BMinusP V then nil
       else BUnsignedGeq(V,U);

procedure BUnsignedGreaterP(U,V);
% Compare magnitudes of two bignums;
  begin
    scalar Lu,Lv,I;
    Lu := BSize U;
    Lv := BSize V;
    if not (Lu eq Lv) then <<
       if IGreaterP(Lu,Lv) then return U
       else return nil >>;
    while IGetV(U,Lv) eq IGetV(V,Lv) and IGreaterP(Lv,1) do Lv := ISub1 Lv;
    if IGreaterP(IGetV(U,Lv),IGetV(V,Lv)) then return U
    else return nil
  end;

procedure BUnsignedGeq(U,V);
% Compare magnitudes of two unsigned bignums;
  begin
    scalar Lu,Lv;
    Lu := BSize U;
    Lv := BSize V;
    if not (Lu eq Lv) then <<
       if IGreaterP(Lu,Lv) then return U
       else return nil >>;
    while IGetV(U,Lv) eq IGetV(V,Lv) and IGreaterP(Lv,1) do Lv := ISub1 Lv;
    If IGreaterP(IGetV(V,Lv),IGetV(U,Lv)) then return nil
    else return U
  end;



procedure BAdd1 V;
 BSmallAdd(V, 1);

procedure BSub1 U;
 BSmallDiff(U, 1);

% ------------------------------------------------
% Conversion to Float:

procedure FloatFromBigNum V;
 if BZeroP V then 0.0
  else if BGreaterP(V, BigFloatHi!*) or BLessp(V, BigFloatLow!*) 
	then Error(99,list("Argument, ",V," to FLOAT is too large"))
  else begin scalar L,Res,Sn,I;
% Careful, do not want to call itself recursively
    L:=BSize V;
    Sn:=BMinusP V;
    Res:=IntFloat IGetv(V,L);
    I:=ISub1 L;
    While not IZeroP I do << Res:=FloatTimes2(res,FloatBBase!*);
		             Res:=FloatPlus2(Res, IntFloat IGetV(V,I));
			     I:=ISub1 I>>;
    if Sn then Res:=minus res;
    return res;
  end;


% ------------------------------------------------
% Input and Output:
Digit2Letter!* :=		% Ascii values of digits and characters.
'[48 49 50 51 52 53 54 55 56 57 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79
80 81 82 83 84 85 86 87 88 89 90];

% OutputBase!* is assumed to be positive and less than 37.

procedure BChannelPrin2(Channel,V);
 If not BigP V then NonBigNumError(V, 'BPrin) %need?
  else begin scalar quot, rem, div, result, resultsign, myobase;
   myobase:=OutputBase!*;
   resultsign:=BMinusP V;
   div:=BSimpleDivide(V,Bsize V,OutputBase!*,nil);
   quot:=car div;
   rem:=cdr div;
   if Bzerop rem then rem:=0 else rem:=IGetV(rem,1);
   result:=rem . result;
   while Not BZeroP quot do
	<<div:=BSimpleDivide(quot,Bsize quot,OutputBase!*,nil);
	quot:=car div;
	rem:=cdr div;
	if Bzerop rem then rem:=0 else rem:=IGetV(rem,1);
	result:=rem . result>>;
   if resultsign then channelwritechar(Channel,char !-);
   if myobase neq 10 then <<ChannelWriteSysInteger(channel,myobase,10);
			ChannelWriteChar(Channel, char !#)>>;
   For each u in result do ChannelWriteChar(Channel, IGetV(digit2letter!*,u));
   OutputBase!*:=myobase;
   return;
  end;

procedure BRead(s,radix,sn);	% radix is < Bbase!*
			%s=string of digits, radix=base, sn=1 or -1
 begin scalar sz, res, ch;
  sz:=size s;
  res:=GtPOS 1;
  ch:=indx(s,0);
  if IGeq(ch,char A) and ILeq(ch,char Z)
		then ch:=IPlus2(IDifference(ch,char A),10);
  if IGeq(ch,char 0) and ILeq(ch,char 9) 
		then ch:=IDifference(ch,char 0);
  IPutV(res,1,ch);
  IFor i:=1:sz do <<ch:=indx(s,i);
		if IGeq(ch,char A) and ILeq(ch,char Z)
			then ch:=IDifference(ch,IDifference(char A,10));
		if IGeq(ch,char 0) and ILeq(ch,char 9)
			then ch:=IDifference(ch,char 0);
		res:=BReadAdd(res, radix, ch)>>;
  if iminusp sn then res:=BMinus res;
  return res;
 end;

procedure BReadAdd(V, radix, ch);
  << V:=BSmallTimes2(V, radix);
     V:=BSmallAdd(V,ch)>>;

procedure BSmallAdd(V,C);	%V big, C fix.
 if IZerop C then return V
  else if Bzerop V then return int2Big C
  else if BMinusp V then BMinus BSmallDiff(BMinus V, C)
  else if IMinusP C then BSmallDiff(V, IMinus C)
  else begin scalar V1,L1;
   Carry!*:=C;
   L1:=BSize V;
   V1:=GtPOS(IAdd1 L1);
   IFor i:=1:L1 do IPutV(V1,i,addcarry IGetV(V,i));
   if IOneP carry!* then IPutV(V1,IAdd1 L1,1) else return TrimBigNum1(V1,L1);
   return V1
  end;

procedure BNum N;	
% Creates a Bignum of one BETA digit, value N.
% N is POS or NEG
 IF BIGP N then N else BnumAux N;

procedure BNumAux N;	
% Creates a Bignum of one BIGIT value N.
% N is POS or NEG
 begin scalar B;
  if IZerop n then return GtPOS 0
   else if IMinusp N then <<b:=GtNEG 1; n:= IMinus n>> else b:=GtPos 1;
  IPutV(b,1,N);
  Return b;
 end;

procedure BSmallDiff(V,C);	%V big, C fix
 if IZerop C then V
  else if BZeroP V then int2Big IMinus C
  else if BMinusP V then BMinus BSmallAdd(BMinus V, C)
  else if IMinusP C then BSmallAdd(V, IMinus C)
  else begin scalar V1,L1;
   Carry!*:=C;
   L1:=BSize V;
   V1:=GtPOS L1;
   IFor i:=1:L1 do IPuTV(V1,i,subcarry IGetV(V,i));
   if not IZeroP carry!* then
      StdError BldMsg(" BSmallDiff V<C %p %p%n",V,C);
   return TrimBigNum1(V1,L1);
  end;

on syslisp;

syslsp procedure int2Big n;		
% Creates BigNum of value N.
% From any N, BETA,INUM,FIXNUM or BIGNUM
case tag n of
	NEGINT,POSINT:	sys2Big n;
	FIXN:		sys2Big fixval fixinf n;
	BIGN:	  	N;
	default: 	NonIntegerError(n, 'int2Big);
 End;

off syslisp;

% Convert BIGNUMs to FLOAT

procedure bigfromfloat X;
 if fixp x or bigp x then x
  else begin scalar bigpart,floatpart,power,sign,thispart;
     if minusp X then <<sign:=-1; X:=minus X>> else sign:=1;
     bigpart:=bzero!*;
     while neq(X, 0) and neq(x,0.0) do <<
	if X < bbase!* then << bigpart:=bplus2(bigpart, bnum fix x);
				X:=0 >>
	 else <<floatpart:=x;
		power:=0;
		while floatpart>=bbase!* do	% get high end of number.
			<<floatpart:=floatpart/bbase!*;
			power:=power + bbits!* >>;
		thispart:=btimes2(btwopower power, bnum fix floatpart);
		X:=X- floatfrombignum thispart;
		bigpart:=bplus2(bigpart, thispart) >> >>;
     if minusp sign then bigpart := bminus bigpart;
     return bigpart;
  end;


% Now Install Interfacing

on syslisp;

syslsp procedure SetUpGlobals;
 << Prin2t  '"SetupGlobals";
   SetBits BitsPerWord;
   Prin2T '" ... done";>>;


off syslisp;

SetupGlobals();

LoadTime <<
 	   StaticBig!*:=GtWarray 10>>;

% Assume dont need more than 10 slots to represent a BigNum
% Version of SYSint

% -- Output---

% MLG Change to interface to Recursive hooks, added for
%  Prinlevel stuff

CopyD('OldChannelPrin1,'RecursiveChannelPrin1);
CopyD('OldChannelPrin2,'RecursiveChannelPrin2);

Procedure RecursiveChannelPrin1(Channel,U,Level);
  <<if BigP U then BChannelPrin2(Channel,U)
	else OldChannelPrin1(Channel, U,Level);U>>;

Procedure RecursiveChannelPrin2(Channel,U,level);
  <<If BigP U then BChannelPrin2(Channel, U)
	else OldChannelPrin2(Channel, U,level);U>>;


procedure checkifreallybig UU;
% If BIGNUM result is in older FIXNUM or INUM range
% Convert Back.
%/ Need a faster test
 if BLessP(UU, BigSysLow!*) or BGreaterp(UU,BigSysHi!*) then UU
  else Sys2Int Big2SysAux UU;

procedure checkifreallybigpair VV;
% Used to process DIVIDE
 checkifreallybig car VV . checkifreallybig cdr VV;

procedure checkifreallybigornil UU;
% Used for EXTRA-boolean tests
 if Null UU or BLessp(UU, BigSysLow!*) or BGreaterP(UU,BigSysHi!*) then UU
  else Sys2Int Big2SysAux UU;

procedure BigPlus2(U,V);
 CheckIfReallyBig BPlus2(U,V);
  
procedure BigDifference(U,V);
 CheckIfReallyBig BDifference(U,V);

procedure BigTimes2(U,V);
 CheckIfReallyBig BTimes2(U,V);

procedure BigDivide(U,V);
 CheckIfReallyBigPair BDivide(U,V);

procedure BigQuotient(U,V);
 CheckIfReallyBig BQuotient(U,V);

procedure BigRemainder(U,V);
 CheckIfReallyBig BRemainder(U,V);

procedure BigLAnd(U,V);
 CheckIfReallyBig BLand(U,V);

procedure BigLOr(U,V);
 CheckIfReallyBig BLOr(U,V);

procedure BigLXOr(U,V);
 CheckIfReallyBig BLXor(U,V);

procedure BigLShift(U,V);
 CheckIfReallyBig BLShift(U,V);

on syslisp;

procedure Lshift(U,V);
   If BetaP U and BetaP V
	then (if V<0 then Sys2Int Wshift(U,V)
               else if V< LispVar (BBits!* ) then Sys2Int Wshift(U,V)
               else BigLshift(Sys2Big U, Sys2Big V) )
    else BigLshift(Sys2Big U, Sys2Big V) ;

off syslisp;

Copyd('LSH,'Lshift);

procedure BigGreaterP(U,V);
 CheckIfReallyBigOrNil BGreaterP(U,V);

procedure BigLessP(U,V);
 CheckIfReallyBigOrNil BLessP(U,V);

procedure BigAdd1 U;
 CheckIfReallyBig BAdd1 U;

procedure BigSub1 U;
 CheckIfReallyBig BSub1 U;

procedure BigLNot U;
 CheckIfReallyBig BLNot U;

procedure BigMinus U;
 CheckIfReallyBig BMinus U;

procedure BigMinusP U;
 CheckIfReallyBigOrNil BMinusP U;

procedure BigOneP U;
 CheckIfReallyBigOrNil BOneP U;

procedure BigZeroP U;
 CheckIfReallyBigOrNil BZeroP U;


% ---- Input ----

procedure MakeStringIntoLispInteger(S,Radix,Sn);
 CheckIfReallyBig BRead(S,Radix,Sn);

on syslisp;

procedure Int2Sys N;
% Convert a random FIXed number to WORD Integer
 case tag(N) of
	POSINT,NEGINT: 	N;
	FIXN:          	FixVal FixInf N;
	BIGN:	       	Big2SysAux N;
	default:	NonNumber1Error(N,'Int2SYS);
 End;

syslsp procedure Sys2Big N;    
% Convert a SYSint to a BIG 
% Must NOT use generic arith here
% Careful that no GC if this BIGger than INUM
Begin scalar Sn, A, B;
  If N=0 then return GtPos 0;
  A:= LispVar StaticBig!*;      % Grab the base
  If N<0 then sn:=T;
  A[1]:=N;                      % Plant number 
  N:=1;                         % now use N as counter
% Careful handling of -N in case have largest NEG, not just
% flip sign
  If Sn then <<B:=-Bbase!*;
	       While A[n]<=B do
	        <<N:=N+1; 
                  A[n]:=A[n-1]/Bbase!*; 
                  A[n-1]:=A[n-1]-a[n]*Bbase!*>>;
               B:=GtNeg N;
               For i:=1:N do Iputv(B,i,-A[i])>>
   else <<     While A[n]>=Bbase!* do
	          <<N:=N+1; 
	            A[n]:=A[n-1]/Bbase!*; 
	            A[n-1]:=A[n-1]-a[n]*Bbase!*>>;
               B:= GtPos N;
               For i:=1:N do IputV(B,i,A[i])>>;
  Return B;
End;

off syslisp;


% Coercion/Transfer Functions

copyd('oldFloatFix,'FloatFix);

procedure FloatFix U;
% Careful of sign and range
  If  FloatSysLow!* <= U and U <= FloatSysHi!* then Oldfloatfix U
   else bigfromfloat U;

on syslisp;

procedure BetaP x;
% test if NUMBER in reduced INUM range
 If Intp x then  (x  <= Lispvar(betaHi!*)) and  (x >= LispVar(betaLow!*)) 
  else NIL;

procedure BetaRangeP x;
% Test if SYSINT in reduced INUM range
 if (x  <= Lispvar(betaHi!*)) then (x>=LispVar(betaLow!*)) else NIL;

procedure Beta2P(x,y);
% Check for 2 argument arithmetic functions
 if BetaP x then BetaP y;

off syslisp;

End;
end;


REDUCE Historical
REDUCE Sourceforge Project | Historical SVN Repository | GitHub Mirror | SourceHut Mirror | NotABug Mirror | Chisel Mirror | Chisel RSS ]