Browse Source

+ software implementation of exp(<double>)

git-svn-id: trunk@5065 -
florian 19 years ago
parent
commit
02a553668f
1 changed files with 201 additions and 1 deletions
  1. 201 1
      rtl/inc/genmath.inc

+ 201 - 1
rtl/inc/genmath.inc

@@ -597,6 +597,204 @@ type
 
 
 {$ifndef FPC_SYSTEM_HAS_EXP}
+{$ifdef SUPPORT_DOUBLE}
+    {
+     This code was translated from uclib code, the original code
+     had the following copyright notice:
+
+     *
+     * ====================================================
+     * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
+     *
+     * Developed at SunPro, a Sun Microsystems, Inc. business.
+     * Permission to use, copy, modify, and distribute this
+     * software is freely granted, provided that this notice
+     * is preserved.
+     * ====================================================
+     *}
+
+    {*
+     * Returns the exponential of x.
+     *
+     * Method
+     *   1. Argument reduction:
+     *      Reduce x to an r so that |r| <= 0.5*ln2 ~ 0.34658.
+     *	Given x, find r and integer k such that
+     *
+     *               x = k*ln2 + r,  |r| <= 0.5*ln2.
+     *
+     *      Here r will be represented as r = hi-lo for better
+     *	accuracy.
+     *
+     *   2. Approximation of exp(r) by a special rational function on
+     *	the interval [0,0.34658]:
+     *	Write
+     *	    R(r**2) = r*(exp(r)+1)/(exp(r)-1) = 2 + r*r/6 - r**4/360 + ...
+     *      We use a special Reme algorithm on [0,0.34658] to generate
+     * 	a polynomial of degree 5 to approximate R. The maximum error
+     *	of this polynomial approximation is bounded by 2**-59. In
+     *	other words,
+     *	    R(z) ~ 2.0 + P1*z + P2*z**2 + P3*z**3 + P4*z**4 + P5*z**5
+     *  	(where z=r*r, and the values of P1 to P5 are listed below)
+     *	and
+     *	    |                  5          |     -59
+     *	    | 2.0+P1*z+...+P5*z   -  R(z) | <= 2
+     *	    |                             |
+     *	The computation of exp(r) thus becomes
+     *                             2*r
+     *		exp(r) = 1 + -------
+     *		              R - r
+     *                                 r*R1(r)
+     *		       = 1 + r + ----------- (for better accuracy)
+     *		                  2 - R1(r)
+     *	where
+     *			         2       4             10
+     *		R1(r) = r - (P1*r  + P2*r  + ... + P5*r   ).
+     *
+     *   3. Scale back to obtain exp(x):
+     *	From step 1, we have
+     *	   exp(x) = 2^k * exp(r)
+     *
+     * Special cases:
+     *	exp(INF) is INF, exp(NaN) is NaN;
+     *	exp(-INF) is 0, and
+     *	for finite argument, only exp(0)=1 is exact.
+     *
+     * Accuracy:
+     *	according to an error analysis, the error is always less than
+     *	1 ulp (unit in the last place).
+     *
+     * Misc. info.
+     *	For IEEE double
+     *	    if x >  7.09782712893383973096e+02 then exp(x) overflow
+     *	    if x < -7.45133219101941108420e+02 then exp(x) underflow
+     *
+     * Constants:
+     * The hexadecimal values are the intended ones for the following
+     * constants. The decimal values may be used, provided that the
+     * compiler will convert from decimal to binary accurately enough
+     * to produce the hexadecimal values shown.
+     *
+    }
+    function fpc_exp_real(x: Double):Double;compilerproc;
+      const
+        one   = 1.0,
+        halF : array[0..1] of double = (0.5,-0.5);
+        huge	= 1.0e+300;
+        twom1000 = 9.33263618503218878990e-302;     { 2**-1000=0x01700000,0}
+        o_threshold =  7.09782712893383973096e+02;  { 0x40862E42, 0xFEFA39EF }
+        u_threshold = -7.45133219101941108420e+02;  { 0xc0874910, 0xD52D3051 }
+        ln2HI  : array[0..1] of double = ( 6.93147180369123816490e-01:  { 0x3fe62e42, 0xfee00000 }
+             -6.93147180369123816490e-01); { 0xbfe62e42, 0xfee00000 }
+        ln2LO : array[0..1] of double = (1.90821492927058770002e-10;  { 0x3dea39ef, 0x35793c76 }
+             -1.90821492927058770002e-10); { 0xbdea39ef, 0x35793c76 }
+        invln2 =  1.44269504088896338700e+00; { 0x3ff71547, 0x652b82fe }
+        P1   =  1.66666666666666019037e-01; { 0x3FC55555, 0x5555553E }
+        P2   = -2.77777777770155933842e-03; { 0xBF66C16C, 0x16BEBD93 }
+        P3   =  6.61375632143793436117e-05; { 0x3F11566A, 0xAF25DE2C }
+        P4   = -1.65339022054652515390e-06; { 0xBEBBBD41, 0xC5D26BF1 }
+        P5   =  4.13813679705723846039e-08; { 0x3E663769, 0x72BEA4D0 }
+      var
+        double hi : double = 0.0;
+	double lo = 0.0;
+	c : double;
+	t : double;
+	int32_t k=0;
+	xsb : longint;
+	hx,hy,lx : dword;
+      begin
+        hx:=float64(x).high;
+        xsb := (hx shr 31) and 1;               { sign bit of x }
+        hx := hx and $7fffffff;                 { high word of |x| }
+
+        { filter out non-finite argument }
+        if hx >= $40862E42 then
+          begin                  { if |x|>=709.78... }
+            if hx >= $7ff00000
+              begin
+                lx:=float64(x).low;
+                if ((hx and $fffff) or lx)<>0 then
+                  begin
+                    result:=x+x; { NaN }
+                    exit;
+                  else
+                else
+                  begin
+                    if xsb=0 then
+                      result:=x
+                    else
+                      result:=0.0;    { exp(+-inf)=begininf,0end }
+                    exit;
+                  end;
+              end;
+            if x > o_threshold then
+              begin
+                result:=huge*huge; { overflow }
+                exit;
+              end;
+            if x < u_threshold then
+              begin
+                result:=twom1000*twom1000; { underflow }
+                exit;
+              end;
+          end;
+
+        { argument reduction }
+        if hx > $3fd62e42 then
+          begin           { if  |x| > 0.5 ln2 }
+            if hx < $3FF0A2B2 then { and |x| < 1.5 ln2 }
+              begin
+                hi := x-ln2HI[xsb];
+                lo:=ln2LO[xsb];
+                k := 1-xsb-xsb;
+              end
+            else
+              begin
+		k  := invln2*x+halF[xsb];
+		t  := k;
+                hi := x - t*ln2HI[0];    { t*ln2HI is exact here }
+		lo := t*ln2LO[0];
+              end;
+            x  := hi - lo;
+          end
+        else if hx < $3e300000 then
+          begin     { when |x|<2**-28 }
+            if huge+x>one then
+              begin
+                result:=one+x;{ trigger inexact }
+                exit;
+              end;
+          end
+	else
+          k := 0;
+
+        { x is now in primary range }
+	t:=x*x;
+	c:=x - t*(P1+t*(P2+t*(P3+t*(P4+t*P5))));
+	if k=0 then
+          begin
+            result:=one-((x*c)/(c-2.0)-x);
+            exit;
+          end
+	else
+          y := one-((lo-(x*c)/(2.0-c))-hi);
+
+        if k >= -1021
+          begin
+            hy:=float64(y).high;
+            float64(y).high:=hy+(k shl 20);        { add k to y's exponent }
+            result:=y;
+          end
+        else
+          begin
+	    hy:=float64(y).high;
+            float64(y).high:=hy+((k+1000) shl 20); { add k to y's exponent }
+	    result:=y*twom1000;
+          end;
+      end;
+
+{$else SUPPORT_DOUBLE}
+
     function fpc_exp_real(d: ValReal):ValReal;compilerproc;
     {*****************************************************************}
     { Exponential Function                                            }
@@ -668,6 +866,8 @@ type
         result := d;
       end;
     end;
+{$endif SUPPORT_DOUBLE}
+
 {$endif}
 
 
@@ -695,7 +895,7 @@ type
           else
             result:=tr;
     end;
-{$endif}
+{$endif FPC_SYSTEM_HAS_EXP}
 
 
 {$ifdef FPC_CURRENCY_IS_INT64}