| ------------------------------------------------------------------------------ |
| -- -- |
| -- GNAT RUN-TIME COMPONENTS -- |
| -- -- |
| -- ADA.NUMERICS.GENERIC_ELEMENTARY_FUNCTIONS -- |
| -- -- |
| -- B o d y -- |
| -- -- |
| -- Copyright (C) 1992-2014, Free Software Foundation, Inc. -- |
| -- -- |
| -- GNAT is free software; you can redistribute it and/or modify it under -- |
| -- terms of the GNU General Public License as published by the Free Soft- -- |
| -- ware Foundation; either version 3, or (at your option) any later ver- -- |
| -- sion. GNAT is distributed in the hope that it will be useful, but WITH- -- |
| -- OUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY -- |
| -- or FITNESS FOR A PARTICULAR PURPOSE. -- |
| -- -- |
| -- As a special exception under Section 7 of GPL version 3, you are granted -- |
| -- additional permissions described in the GCC Runtime Library Exception, -- |
| -- version 3.1, as published by the Free Software Foundation. -- |
| -- -- |
| -- You should have received a copy of the GNU General Public License and -- |
| -- a copy of the GCC Runtime Library Exception along with this program; -- |
| -- see the files COPYING3 and COPYING.RUNTIME respectively. If not, see -- |
| -- <http://www.gnu.org/licenses/>. -- |
| -- -- |
| -- GNAT was originally developed by the GNAT team at New York University. -- |
| -- Extensive contributions were provided by Ada Core Technologies Inc. -- |
| -- -- |
| ------------------------------------------------------------------------------ |
| |
| -- This body is specifically for using an Ada interface to C math.h to get |
| -- the computation engine. Many special cases are handled locally to avoid |
| -- unnecessary calls or to meet Annex G strict mode requirements. |
| |
| -- Uses functions sqrt, exp, log, pow, sin, asin, cos, acos, tan, atan, sinh, |
| -- cosh, tanh from C library via math.h |
| |
| with Ada.Numerics.Aux; |
| |
| package body Ada.Numerics.Generic_Elementary_Functions is |
| |
| use type Ada.Numerics.Aux.Double; |
| |
| Sqrt_Two : constant := 1.41421_35623_73095_04880_16887_24209_69807_85696; |
| Log_Two : constant := 0.69314_71805_59945_30941_72321_21458_17656_80755; |
| |
| Half_Log_Two : constant := Log_Two / 2; |
| |
| subtype T is Float_Type'Base; |
| subtype Double is Aux.Double; |
| |
| Two_Pi : constant T := 2.0 * Pi; |
| Half_Pi : constant T := Pi / 2.0; |
| |
| Half_Log_Epsilon : constant T := T (1 - T'Model_Mantissa) * Half_Log_Two; |
| Log_Inverse_Epsilon : constant T := T (T'Model_Mantissa - 1) * Log_Two; |
| Sqrt_Epsilon : constant T := Sqrt_Two ** (1 - T'Model_Mantissa); |
| |
| ----------------------- |
| -- Local Subprograms -- |
| ----------------------- |
| |
| function Exp_Strict (X : Float_Type'Base) return Float_Type'Base; |
| -- Cody/Waite routine, supposedly more precise than the library version. |
| -- Currently only needed for Sinh/Cosh on X86 with the largest FP type. |
| |
| function Local_Atan |
| (Y : Float_Type'Base; |
| X : Float_Type'Base := 1.0) return Float_Type'Base; |
| -- Common code for arc tangent after cycle reduction |
| |
| ---------- |
| -- "**" -- |
| ---------- |
| |
| function "**" (Left, Right : Float_Type'Base) return Float_Type'Base is |
| A_Right : Float_Type'Base; |
| Int_Part : Integer; |
| Result : Float_Type'Base; |
| R1 : Float_Type'Base; |
| Rest : Float_Type'Base; |
| |
| begin |
| if Left = 0.0 |
| and then Right = 0.0 |
| then |
| raise Argument_Error; |
| |
| elsif Left < 0.0 then |
| raise Argument_Error; |
| |
| elsif Right = 0.0 then |
| return 1.0; |
| |
| elsif Left = 0.0 then |
| if Right < 0.0 then |
| raise Constraint_Error; |
| else |
| return 0.0; |
| end if; |
| |
| elsif Left = 1.0 then |
| return 1.0; |
| |
| elsif Right = 1.0 then |
| return Left; |
| |
| else |
| begin |
| if Right = 2.0 then |
| return Left * Left; |
| |
| elsif Right = 0.5 then |
| return Sqrt (Left); |
| |
| else |
| A_Right := abs (Right); |
| |
| -- If exponent is larger than one, compute integer exponen- |
| -- tiation if possible, and evaluate fractional part with more |
| -- precision. The relative error is now proportional to the |
| -- fractional part of the exponent only. |
| |
| if A_Right > 1.0 |
| and then A_Right < Float_Type'Base (Integer'Last) |
| then |
| Int_Part := Integer (Float_Type'Base'Truncation (A_Right)); |
| Result := Left ** Int_Part; |
| Rest := A_Right - Float_Type'Base (Int_Part); |
| |
| -- Compute with two leading bits of the mantissa using |
| -- square roots. Bound to be better than logarithms, and |
| -- easily extended to greater precision. |
| |
| if Rest >= 0.5 then |
| R1 := Sqrt (Left); |
| Result := Result * R1; |
| Rest := Rest - 0.5; |
| |
| if Rest >= 0.25 then |
| Result := Result * Sqrt (R1); |
| Rest := Rest - 0.25; |
| end if; |
| |
| elsif Rest >= 0.25 then |
| Result := Result * Sqrt (Sqrt (Left)); |
| Rest := Rest - 0.25; |
| end if; |
| |
| Result := Result * |
| Float_Type'Base (Aux.Pow (Double (Left), Double (Rest))); |
| |
| if Right >= 0.0 then |
| return Result; |
| else |
| return (1.0 / Result); |
| end if; |
| else |
| return |
| Float_Type'Base (Aux.Pow (Double (Left), Double (Right))); |
| end if; |
| end if; |
| |
| exception |
| when others => |
| raise Constraint_Error; |
| end; |
| end if; |
| end "**"; |
| |
| ------------ |
| -- Arccos -- |
| ------------ |
| |
| -- Natural cycle |
| |
| function Arccos (X : Float_Type'Base) return Float_Type'Base is |
| Temp : Float_Type'Base; |
| |
| begin |
| if abs X > 1.0 then |
| raise Argument_Error; |
| |
| elsif abs X < Sqrt_Epsilon then |
| return Pi / 2.0 - X; |
| |
| elsif X = 1.0 then |
| return 0.0; |
| |
| elsif X = -1.0 then |
| return Pi; |
| end if; |
| |
| Temp := Float_Type'Base (Aux.Acos (Double (X))); |
| |
| if Temp < 0.0 then |
| Temp := Pi + Temp; |
| end if; |
| |
| return Temp; |
| end Arccos; |
| |
| -- Arbitrary cycle |
| |
| function Arccos (X, Cycle : Float_Type'Base) return Float_Type'Base is |
| Temp : Float_Type'Base; |
| |
| begin |
| if Cycle <= 0.0 then |
| raise Argument_Error; |
| |
| elsif abs X > 1.0 then |
| raise Argument_Error; |
| |
| elsif abs X < Sqrt_Epsilon then |
| return Cycle / 4.0; |
| |
| elsif X = 1.0 then |
| return 0.0; |
| |
| elsif X = -1.0 then |
| return Cycle / 2.0; |
| end if; |
| |
| Temp := Arctan (Sqrt ((1.0 - X) * (1.0 + X)) / X, 1.0, Cycle); |
| |
| if Temp < 0.0 then |
| Temp := Cycle / 2.0 + Temp; |
| end if; |
| |
| return Temp; |
| end Arccos; |
| |
| ------------- |
| -- Arccosh -- |
| ------------- |
| |
| function Arccosh (X : Float_Type'Base) return Float_Type'Base is |
| begin |
| -- Return positive branch of Log (X - Sqrt (X * X - 1.0)), or the proper |
| -- approximation for X close to 1 or >> 1. |
| |
| if X < 1.0 then |
| raise Argument_Error; |
| |
| elsif X < 1.0 + Sqrt_Epsilon then |
| return Sqrt (2.0 * (X - 1.0)); |
| |
| elsif X > 1.0 / Sqrt_Epsilon then |
| return Log (X) + Log_Two; |
| |
| else |
| return Log (X + Sqrt ((X - 1.0) * (X + 1.0))); |
| end if; |
| end Arccosh; |
| |
| ------------ |
| -- Arccot -- |
| ------------ |
| |
| -- Natural cycle |
| |
| function Arccot |
| (X : Float_Type'Base; |
| Y : Float_Type'Base := 1.0) |
| return Float_Type'Base |
| is |
| begin |
| -- Just reverse arguments |
| |
| return Arctan (Y, X); |
| end Arccot; |
| |
| -- Arbitrary cycle |
| |
| function Arccot |
| (X : Float_Type'Base; |
| Y : Float_Type'Base := 1.0; |
| Cycle : Float_Type'Base) |
| return Float_Type'Base |
| is |
| begin |
| -- Just reverse arguments |
| |
| return Arctan (Y, X, Cycle); |
| end Arccot; |
| |
| ------------- |
| -- Arccoth -- |
| ------------- |
| |
| function Arccoth (X : Float_Type'Base) return Float_Type'Base is |
| begin |
| if abs X > 2.0 then |
| return Arctanh (1.0 / X); |
| |
| elsif abs X = 1.0 then |
| raise Constraint_Error; |
| |
| elsif abs X < 1.0 then |
| raise Argument_Error; |
| |
| else |
| -- 1.0 < abs X <= 2.0. One of X + 1.0 and X - 1.0 is exact, the other |
| -- has error 0 or Epsilon. |
| |
| return 0.5 * (Log (abs (X + 1.0)) - Log (abs (X - 1.0))); |
| end if; |
| end Arccoth; |
| |
| ------------ |
| -- Arcsin -- |
| ------------ |
| |
| -- Natural cycle |
| |
| function Arcsin (X : Float_Type'Base) return Float_Type'Base is |
| begin |
| if abs X > 1.0 then |
| raise Argument_Error; |
| |
| elsif abs X < Sqrt_Epsilon then |
| return X; |
| |
| elsif X = 1.0 then |
| return Pi / 2.0; |
| |
| elsif X = -1.0 then |
| return -(Pi / 2.0); |
| end if; |
| |
| return Float_Type'Base (Aux.Asin (Double (X))); |
| end Arcsin; |
| |
| -- Arbitrary cycle |
| |
| function Arcsin (X, Cycle : Float_Type'Base) return Float_Type'Base is |
| begin |
| if Cycle <= 0.0 then |
| raise Argument_Error; |
| |
| elsif abs X > 1.0 then |
| raise Argument_Error; |
| |
| elsif X = 0.0 then |
| return X; |
| |
| elsif X = 1.0 then |
| return Cycle / 4.0; |
| |
| elsif X = -1.0 then |
| return -(Cycle / 4.0); |
| end if; |
| |
| return Arctan (X / Sqrt ((1.0 - X) * (1.0 + X)), 1.0, Cycle); |
| end Arcsin; |
| |
| ------------- |
| -- Arcsinh -- |
| ------------- |
| |
| function Arcsinh (X : Float_Type'Base) return Float_Type'Base is |
| begin |
| if abs X < Sqrt_Epsilon then |
| return X; |
| |
| elsif X > 1.0 / Sqrt_Epsilon then |
| return Log (X) + Log_Two; |
| |
| elsif X < -(1.0 / Sqrt_Epsilon) then |
| return -(Log (-X) + Log_Two); |
| |
| elsif X < 0.0 then |
| return -Log (abs X + Sqrt (X * X + 1.0)); |
| |
| else |
| return Log (X + Sqrt (X * X + 1.0)); |
| end if; |
| end Arcsinh; |
| |
| ------------ |
| -- Arctan -- |
| ------------ |
| |
| -- Natural cycle |
| |
| function Arctan |
| (Y : Float_Type'Base; |
| X : Float_Type'Base := 1.0) |
| return Float_Type'Base |
| is |
| begin |
| if X = 0.0 and then Y = 0.0 then |
| raise Argument_Error; |
| |
| elsif Y = 0.0 then |
| if X > 0.0 then |
| return 0.0; |
| else -- X < 0.0 |
| return Pi * Float_Type'Copy_Sign (1.0, Y); |
| end if; |
| |
| elsif X = 0.0 then |
| return Float_Type'Copy_Sign (Half_Pi, Y); |
| |
| else |
| return Local_Atan (Y, X); |
| end if; |
| end Arctan; |
| |
| -- Arbitrary cycle |
| |
| function Arctan |
| (Y : Float_Type'Base; |
| X : Float_Type'Base := 1.0; |
| Cycle : Float_Type'Base) |
| return Float_Type'Base |
| is |
| begin |
| if Cycle <= 0.0 then |
| raise Argument_Error; |
| |
| elsif X = 0.0 and then Y = 0.0 then |
| raise Argument_Error; |
| |
| elsif Y = 0.0 then |
| if X > 0.0 then |
| return 0.0; |
| else -- X < 0.0 |
| return Cycle / 2.0 * Float_Type'Copy_Sign (1.0, Y); |
| end if; |
| |
| elsif X = 0.0 then |
| return Float_Type'Copy_Sign (Cycle / 4.0, Y); |
| |
| else |
| return Local_Atan (Y, X) * Cycle / Two_Pi; |
| end if; |
| end Arctan; |
| |
| ------------- |
| -- Arctanh -- |
| ------------- |
| |
| function Arctanh (X : Float_Type'Base) return Float_Type'Base is |
| A, B, D, A_Plus_1, A_From_1 : Float_Type'Base; |
| |
| Mantissa : constant Integer := Float_Type'Base'Machine_Mantissa; |
| |
| begin |
| -- The naive formula: |
| |
| -- Arctanh (X) := (1/2) * Log (1 + X) / (1 - X) |
| |
| -- is not well-behaved numerically when X < 0.5 and when X is close |
| -- to one. The following is accurate but probably not optimal. |
| |
| if abs X = 1.0 then |
| raise Constraint_Error; |
| |
| elsif abs X >= 1.0 - 2.0 ** (-Mantissa) then |
| |
| if abs X >= 1.0 then |
| raise Argument_Error; |
| else |
| |
| -- The one case that overflows if put through the method below: |
| -- abs X = 1.0 - Epsilon. In this case (1/2) log (2/Epsilon) is |
| -- accurate. This simplifies to: |
| |
| return Float_Type'Copy_Sign ( |
| Half_Log_Two * Float_Type'Base (Mantissa + 1), X); |
| end if; |
| |
| -- elsif abs X <= 0.5 then |
| -- why is above line commented out ??? |
| |
| else |
| -- Use several piecewise linear approximations. A is close to X, |
| -- chosen so 1.0 + A, 1.0 - A, and X - A are exact. The two scalings |
| -- remove the low-order bits of X. |
| |
| A := Float_Type'Base'Scaling ( |
| Float_Type'Base (Long_Long_Integer |
| (Float_Type'Base'Scaling (X, Mantissa - 1))), 1 - Mantissa); |
| |
| B := X - A; -- This is exact; abs B <= 2**(-Mantissa). |
| A_Plus_1 := 1.0 + A; -- This is exact. |
| A_From_1 := 1.0 - A; -- Ditto. |
| D := A_Plus_1 * A_From_1; -- 1 - A*A. |
| |
| -- use one term of the series expansion: |
| |
| -- f (x + e) = f(x) + e * f'(x) + .. |
| |
| -- The derivative of Arctanh at A is 1/(1-A*A). Next term is |
| -- A*(B/D)**2 (if a quadratic approximation is ever needed). |
| |
| return 0.5 * (Log (A_Plus_1) - Log (A_From_1)) + B / D; |
| end if; |
| end Arctanh; |
| |
| --------- |
| -- Cos -- |
| --------- |
| |
| -- Natural cycle |
| |
| function Cos (X : Float_Type'Base) return Float_Type'Base is |
| begin |
| if abs X < Sqrt_Epsilon then |
| return 1.0; |
| end if; |
| |
| return Float_Type'Base (Aux.Cos (Double (X))); |
| end Cos; |
| |
| -- Arbitrary cycle |
| |
| function Cos (X, Cycle : Float_Type'Base) return Float_Type'Base is |
| begin |
| -- Just reuse the code for Sin. The potential small loss of speed is |
| -- negligible with proper (front-end) inlining. |
| |
| return -Sin (abs X - Cycle * 0.25, Cycle); |
| end Cos; |
| |
| ---------- |
| -- Cosh -- |
| ---------- |
| |
| function Cosh (X : Float_Type'Base) return Float_Type'Base is |
| Lnv : constant Float_Type'Base := 8#0.542714#; |
| V2minus1 : constant Float_Type'Base := 0.13830_27787_96019_02638E-4; |
| Y : constant Float_Type'Base := abs X; |
| Z : Float_Type'Base; |
| |
| begin |
| if Y < Sqrt_Epsilon then |
| return 1.0; |
| |
| elsif Y > Log_Inverse_Epsilon then |
| Z := Exp_Strict (Y - Lnv); |
| return (Z + V2minus1 * Z); |
| |
| else |
| Z := Exp_Strict (Y); |
| return 0.5 * (Z + 1.0 / Z); |
| end if; |
| |
| end Cosh; |
| |
| --------- |
| -- Cot -- |
| --------- |
| |
| -- Natural cycle |
| |
| function Cot (X : Float_Type'Base) return Float_Type'Base is |
| begin |
| if X = 0.0 then |
| raise Constraint_Error; |
| |
| elsif abs X < Sqrt_Epsilon then |
| return 1.0 / X; |
| end if; |
| |
| return 1.0 / Float_Type'Base (Aux.Tan (Double (X))); |
| end Cot; |
| |
| -- Arbitrary cycle |
| |
| function Cot (X, Cycle : Float_Type'Base) return Float_Type'Base is |
| T : Float_Type'Base; |
| |
| begin |
| if Cycle <= 0.0 then |
| raise Argument_Error; |
| end if; |
| |
| T := Float_Type'Base'Remainder (X, Cycle); |
| |
| if T = 0.0 or else abs T = 0.5 * Cycle then |
| raise Constraint_Error; |
| |
| elsif abs T < Sqrt_Epsilon then |
| return 1.0 / T; |
| |
| elsif abs T = 0.25 * Cycle then |
| return 0.0; |
| |
| else |
| T := T / Cycle * Two_Pi; |
| return Cos (T) / Sin (T); |
| end if; |
| end Cot; |
| |
| ---------- |
| -- Coth -- |
| ---------- |
| |
| function Coth (X : Float_Type'Base) return Float_Type'Base is |
| begin |
| if X = 0.0 then |
| raise Constraint_Error; |
| |
| elsif X < Half_Log_Epsilon then |
| return -1.0; |
| |
| elsif X > -Half_Log_Epsilon then |
| return 1.0; |
| |
| elsif abs X < Sqrt_Epsilon then |
| return 1.0 / X; |
| end if; |
| |
| return 1.0 / Float_Type'Base (Aux.Tanh (Double (X))); |
| end Coth; |
| |
| --------- |
| -- Exp -- |
| --------- |
| |
| function Exp (X : Float_Type'Base) return Float_Type'Base is |
| Result : Float_Type'Base; |
| |
| begin |
| if X = 0.0 then |
| return 1.0; |
| end if; |
| |
| Result := Float_Type'Base (Aux.Exp (Double (X))); |
| |
| -- Deal with case of Exp returning IEEE infinity. If Machine_Overflows |
| -- is False, then we can just leave it as an infinity (and indeed we |
| -- prefer to do so). But if Machine_Overflows is True, then we have |
| -- to raise a Constraint_Error exception as required by the RM. |
| |
| if Float_Type'Machine_Overflows and then not Result'Valid then |
| raise Constraint_Error; |
| end if; |
| |
| return Result; |
| end Exp; |
| |
| ---------------- |
| -- Exp_Strict -- |
| ---------------- |
| |
| function Exp_Strict (X : Float_Type'Base) return Float_Type'Base is |
| G : Float_Type'Base; |
| Z : Float_Type'Base; |
| |
| P0 : constant := 0.25000_00000_00000_00000; |
| P1 : constant := 0.75753_18015_94227_76666E-2; |
| P2 : constant := 0.31555_19276_56846_46356E-4; |
| |
| Q0 : constant := 0.5; |
| Q1 : constant := 0.56817_30269_85512_21787E-1; |
| Q2 : constant := 0.63121_89437_43985_02557E-3; |
| Q3 : constant := 0.75104_02839_98700_46114E-6; |
| |
| C1 : constant := 8#0.543#; |
| C2 : constant := -2.1219_44400_54690_58277E-4; |
| Le : constant := 1.4426_95040_88896_34074; |
| |
| XN : Float_Type'Base; |
| P, Q, R : Float_Type'Base; |
| |
| begin |
| if X = 0.0 then |
| return 1.0; |
| end if; |
| |
| XN := Float_Type'Base'Rounding (X * Le); |
| G := (X - XN * C1) - XN * C2; |
| Z := G * G; |
| P := G * ((P2 * Z + P1) * Z + P0); |
| Q := ((Q3 * Z + Q2) * Z + Q1) * Z + Q0; |
| R := 0.5 + P / (Q - P); |
| |
| R := Float_Type'Base'Scaling (R, Integer (XN) + 1); |
| |
| -- Deal with case of Exp returning IEEE infinity. If Machine_Overflows |
| -- is False, then we can just leave it as an infinity (and indeed we |
| -- prefer to do so). But if Machine_Overflows is True, then we have to |
| -- raise a Constraint_Error exception as required by the RM. |
| |
| if Float_Type'Machine_Overflows and then not R'Valid then |
| raise Constraint_Error; |
| else |
| return R; |
| end if; |
| |
| end Exp_Strict; |
| |
| ---------------- |
| -- Local_Atan -- |
| ---------------- |
| |
| function Local_Atan |
| (Y : Float_Type'Base; |
| X : Float_Type'Base := 1.0) return Float_Type'Base |
| is |
| Z : Float_Type'Base; |
| Raw_Atan : Float_Type'Base; |
| |
| begin |
| Z := (if abs Y > abs X then abs (X / Y) else abs (Y / X)); |
| |
| Raw_Atan := |
| (if Z < Sqrt_Epsilon then Z |
| elsif Z = 1.0 then Pi / 4.0 |
| else Float_Type'Base (Aux.Atan (Double (Z)))); |
| |
| if abs Y > abs X then |
| Raw_Atan := Half_Pi - Raw_Atan; |
| end if; |
| |
| if X > 0.0 then |
| return Float_Type'Copy_Sign (Raw_Atan, Y); |
| else |
| return Float_Type'Copy_Sign (Pi - Raw_Atan, Y); |
| end if; |
| end Local_Atan; |
| |
| --------- |
| -- Log -- |
| --------- |
| |
| -- Natural base |
| |
| function Log (X : Float_Type'Base) return Float_Type'Base is |
| begin |
| if X < 0.0 then |
| raise Argument_Error; |
| |
| elsif X = 0.0 then |
| raise Constraint_Error; |
| |
| elsif X = 1.0 then |
| return 0.0; |
| end if; |
| |
| return Float_Type'Base (Aux.Log (Double (X))); |
| end Log; |
| |
| -- Arbitrary base |
| |
| function Log (X, Base : Float_Type'Base) return Float_Type'Base is |
| begin |
| if X < 0.0 then |
| raise Argument_Error; |
| |
| elsif Base <= 0.0 or else Base = 1.0 then |
| raise Argument_Error; |
| |
| elsif X = 0.0 then |
| raise Constraint_Error; |
| |
| elsif X = 1.0 then |
| return 0.0; |
| end if; |
| |
| return Float_Type'Base (Aux.Log (Double (X)) / Aux.Log (Double (Base))); |
| end Log; |
| |
| --------- |
| -- Sin -- |
| --------- |
| |
| -- Natural cycle |
| |
| function Sin (X : Float_Type'Base) return Float_Type'Base is |
| begin |
| if abs X < Sqrt_Epsilon then |
| return X; |
| end if; |
| |
| return Float_Type'Base (Aux.Sin (Double (X))); |
| end Sin; |
| |
| -- Arbitrary cycle |
| |
| function Sin (X, Cycle : Float_Type'Base) return Float_Type'Base is |
| T : Float_Type'Base; |
| |
| begin |
| if Cycle <= 0.0 then |
| raise Argument_Error; |
| |
| -- If X is zero, return it as the result, preserving the argument sign. |
| -- Is this test really needed on any machine ??? |
| |
| elsif X = 0.0 then |
| return X; |
| end if; |
| |
| T := Float_Type'Base'Remainder (X, Cycle); |
| |
| -- The following two reductions reduce the argument to the interval |
| -- [-0.25 * Cycle, 0.25 * Cycle]. This reduction is exact and is needed |
| -- to prevent inaccuracy that may result if the sine function uses a |
| -- different (more accurate) value of Pi in its reduction than is used |
| -- in the multiplication with Two_Pi. |
| |
| if abs T > 0.25 * Cycle then |
| T := 0.5 * Float_Type'Copy_Sign (Cycle, T) - T; |
| end if; |
| |
| -- Could test for 12.0 * abs T = Cycle, and return an exact value in |
| -- those cases. It is not clear this is worth the extra test though. |
| |
| return Float_Type'Base (Aux.Sin (Double (T / Cycle * Two_Pi))); |
| end Sin; |
| |
| ---------- |
| -- Sinh -- |
| ---------- |
| |
| function Sinh (X : Float_Type'Base) return Float_Type'Base is |
| Lnv : constant Float_Type'Base := 8#0.542714#; |
| V2minus1 : constant Float_Type'Base := 0.13830_27787_96019_02638E-4; |
| Y : constant Float_Type'Base := abs X; |
| F : constant Float_Type'Base := Y * Y; |
| Z : Float_Type'Base; |
| |
| Float_Digits_1_6 : constant Boolean := Float_Type'Digits < 7; |
| |
| begin |
| if Y < Sqrt_Epsilon then |
| return X; |
| |
| elsif Y > Log_Inverse_Epsilon then |
| Z := Exp_Strict (Y - Lnv); |
| Z := Z + V2minus1 * Z; |
| |
| elsif Y < 1.0 then |
| |
| if Float_Digits_1_6 then |
| |
| -- Use expansion provided by Cody and Waite, p. 226. Note that |
| -- leading term of the polynomial in Q is exactly 1.0. |
| |
| declare |
| P0 : constant := -0.71379_3159E+1; |
| P1 : constant := -0.19033_3399E+0; |
| Q0 : constant := -0.42827_7109E+2; |
| |
| begin |
| Z := Y + Y * F * (P1 * F + P0) / (F + Q0); |
| end; |
| |
| else |
| declare |
| P0 : constant := -0.35181_28343_01771_17881E+6; |
| P1 : constant := -0.11563_52119_68517_68270E+5; |
| P2 : constant := -0.16375_79820_26307_51372E+3; |
| P3 : constant := -0.78966_12741_73570_99479E+0; |
| Q0 : constant := -0.21108_77005_81062_71242E+7; |
| Q1 : constant := 0.36162_72310_94218_36460E+5; |
| Q2 : constant := -0.27773_52311_96507_01667E+3; |
| |
| begin |
| Z := Y + Y * F * (((P3 * F + P2) * F + P1) * F + P0) |
| / (((F + Q2) * F + Q1) * F + Q0); |
| end; |
| end if; |
| |
| else |
| Z := Exp_Strict (Y); |
| Z := 0.5 * (Z - 1.0 / Z); |
| end if; |
| |
| if X > 0.0 then |
| return Z; |
| else |
| return -Z; |
| end if; |
| end Sinh; |
| |
| ---------- |
| -- Sqrt -- |
| ---------- |
| |
| function Sqrt (X : Float_Type'Base) return Float_Type'Base is |
| begin |
| if X < 0.0 then |
| raise Argument_Error; |
| |
| -- Special case Sqrt (0.0) to preserve possible minus sign per IEEE |
| |
| elsif X = 0.0 then |
| return X; |
| end if; |
| |
| return Float_Type'Base (Aux.Sqrt (Double (X))); |
| end Sqrt; |
| |
| --------- |
| -- Tan -- |
| --------- |
| |
| -- Natural cycle |
| |
| function Tan (X : Float_Type'Base) return Float_Type'Base is |
| begin |
| if abs X < Sqrt_Epsilon then |
| return X; |
| end if; |
| |
| -- Note: if X is exactly pi/2, then we should raise an exception, since |
| -- the result would overflow. But for all floating-point formats we deal |
| -- with, it is impossible for X to be exactly pi/2, and the result is |
| -- always in range. |
| |
| return Float_Type'Base (Aux.Tan (Double (X))); |
| end Tan; |
| |
| -- Arbitrary cycle |
| |
| function Tan (X, Cycle : Float_Type'Base) return Float_Type'Base is |
| T : Float_Type'Base; |
| |
| begin |
| if Cycle <= 0.0 then |
| raise Argument_Error; |
| |
| elsif X = 0.0 then |
| return X; |
| end if; |
| |
| T := Float_Type'Base'Remainder (X, Cycle); |
| |
| if abs T = 0.25 * Cycle then |
| raise Constraint_Error; |
| |
| elsif abs T = 0.5 * Cycle then |
| return 0.0; |
| |
| else |
| T := T / Cycle * Two_Pi; |
| return Sin (T) / Cos (T); |
| end if; |
| |
| end Tan; |
| |
| ---------- |
| -- Tanh -- |
| ---------- |
| |
| function Tanh (X : Float_Type'Base) return Float_Type'Base is |
| P0 : constant Float_Type'Base := -0.16134_11902_39962_28053E+4; |
| P1 : constant Float_Type'Base := -0.99225_92967_22360_83313E+2; |
| P2 : constant Float_Type'Base := -0.96437_49277_72254_69787E+0; |
| |
| Q0 : constant Float_Type'Base := 0.48402_35707_19886_88686E+4; |
| Q1 : constant Float_Type'Base := 0.22337_72071_89623_12926E+4; |
| Q2 : constant Float_Type'Base := 0.11274_47438_05349_49335E+3; |
| Q3 : constant Float_Type'Base := 0.10000_00000_00000_00000E+1; |
| |
| Half_Ln3 : constant Float_Type'Base := 0.54930_61443_34054_84570; |
| |
| P, Q, R : Float_Type'Base; |
| Y : constant Float_Type'Base := abs X; |
| G : constant Float_Type'Base := Y * Y; |
| |
| Float_Type_Digits_15_Or_More : constant Boolean := |
| Float_Type'Digits > 14; |
| |
| begin |
| if X < Half_Log_Epsilon then |
| return -1.0; |
| |
| elsif X > -Half_Log_Epsilon then |
| return 1.0; |
| |
| elsif Y < Sqrt_Epsilon then |
| return X; |
| |
| elsif Y < Half_Ln3 |
| and then Float_Type_Digits_15_Or_More |
| then |
| P := (P2 * G + P1) * G + P0; |
| Q := ((Q3 * G + Q2) * G + Q1) * G + Q0; |
| R := G * (P / Q); |
| return X + X * R; |
| |
| else |
| return Float_Type'Base (Aux.Tanh (Double (X))); |
| end if; |
| end Tanh; |
| |
| end Ada.Numerics.Generic_Elementary_Functions; |