separate ( Generic_Elementary_Functions ) function Arcsinh( X : Float_Type ) return Float_Type is -- On input, X is a floating-point value in Float_Type; -- On output, the value of Arcsinh(X) (the inverse hyperbolic sin of X) -- is returned. -- The definition of Arcsinh(Y) is log( Y + sqrt(Y*Y + 1) ) -- For symmetry, we return sign(Y)*Arcsinh(|Y|). The discussions below -- therefore assume Y >= 0. -- To obtain good accuracy, we consider several cases: -- 1) Y <= epsilon, simply return Y. -- 2) epsilon < Y <= 0.5, -- Y + sqrt(Y*Y+1) = 1 + ( Y + Y*Y/[1 + sqrt(Y*Y+1)] ) -- = 1 + ( Y + Y / [ (1/Y) + sqrt(1 + [1/Y*Y]) ] ) -- A formula best suited for the kernel function L1p. -- 3) 0.5 < Y < 10/epsilon, -- Y + sqrt(Y*Y+1) = 2( Y + 0.5/[ sqrt(Y*Y+1) + Y ] ). -- 4) 10/epsilon <= Y, then -- Y + sqrt(Y*Y+1) = 2Y for practical purposes. -- Note that (3) and (4) are suited for invoking the kernel procedure -- KP_Log(Input) which returns M, Z1, and Z2 where -- log(Input) = M * log(2) + Z1 + Z2. -- Y, Sign_X, V, M, Z1, Z2, Result : Common_Float; One : constant := 1.0; Half : constant := 0.5; Small_Threshold : constant Common_Float := Common_Float'Base'Epsilon; Large_Threshold : constant Common_Float := 10.0 / Common_Float'Base'Epsilon; Log2_Lead : constant Common_Float := 16#0.B17#; Log2_Trail : constant Common_Float := 16#0.000217F7D1CF79ABC9E3B39803F2F6AF40#; begin Y := Common_Float(X); Sign_X := Copy_Sign(One,Y); Y := abs(Y); if (Y <= Half) then if (Y < Small_Threshold) then return( Float_Type( Copy_Sign(Y,Sign_X) ) ); else V := Y + Y/( (One/Y) + KF_Sqrt(One + (One/(Y*Y))) ); Result := KF_L1p( V ); end if; else if (Y < Large_Threshold) then Y := Y + Half/( Y + KF_Sqrt(One + Y*Y) ); end if; KP_Log(Y, M, Z1, Z2); M := M + One; Result := M*Log2_Lead + (Z1 + (Z2 + M*Log2_Trail)); end if; return( Float_Type( Copy_Sign( Result, Sign_X ) ) ); end Arcsinh;