| blob | 551a324c8d882ac0309c2ff442ccf31a0e6fd026 |
1 ;;; -*- mode: lisp -*-
2 ;;; Copyright (c) 2005--2007, by A.J. Rossini <blindglobe@gmail.com>
3 ;;; See COPYRIGHT file for any additional restrictions (BSD license).
4 ;;; Since 1991, ANSI was finally finished. Edited for ANSI Common Lisp.
6 ;;; lsfloat -- Floating point specs and transcendental functions
7 ;;;
8 ;;; Copyright (c) 1991, by Luke Tierney. Permission is granted for
9 ;;; unrestricted use.
13 ;;; Common Lisp allows for four different floating point types that need
14 ;;; not be distinct. For statistical work, the type I prefer to use is
15 ;;; the one that is closest to a C double. This type is named stat-float.
16 ;;; By setting the variable *read-default-float-format* to this type, you
17 ;;; insure that data entered as floating point data is read in with this
18 ;;; type. The problem arises with data read as integers that is passed to
19 ;;; a transcendental, like sqrt. Floating point contagion rules say these
20 ;;; integers are to be converted to type single-float. Unless single-float
21 ;;; is equivalent to C double, as it is in Mac CL and KCL, this is not
22 ;;; what I want. Hence this file redefines the transcendentals to first
23 ;;; coerce their arguments to stat-float before applying the built-in
24 ;;; functions.
25 ;;;
26 ;;; No actual modifications to the transcendentals are needed if
27 ;;; single-float is the same as stat-float. The feature
28 ;;; :stat-float-is-double-float is used to indicate this.
29 ;;;
30 ;;; KCL NOTE:
31 ;;; In (A)KCL the type short-float corresponds to C float and the types
32 ;;; single-float, double-float and long-float correspond to C double.
33 ;;; But in the implementation of the transcendentals (A)KCL coerces
34 ;;; rationals to short-float, not single-float. CLtL1 is a little vague
35 ;;; on this (it talks about "single precision") but CLtL2 clarifies that
36 ;;; rationals should produce single-float results. So (A)KCL is wrong, at
37 ;;; least relative to the clarification in CLtL2. I therefore decided
38 ;;; to fix (A)KCL in files c/num_sfun.c and lsp/numlib.lsp. If these
39 ;;; fixes are applied, the feature :stat-float-is-double-float should be
40 ;;; defined.
43 ;;; ======== From the CLHS ==========
44 ;;
45 ;; Type SHORT-FLOAT, SINGLE-FLOAT, DOUBLE-FLOAT, LONG-FLOAT
46 ;;
47 ;; Supertypes:
48 ;;
49 ;; short-float: short-float, float, real, number, t
50 ;;
51 ;;
52 ;; single-float: single-float, float, real, number, t
53 ;;
54 ;; double-float: double-float, float, real, number, t
55 ;;
56 ;; long-float: long-float, float, real, number, t
57 ;;
58 ;; Description: For the four defined subtypes of type float, it is true
59 ;; that intermediate between the type short-float and the type long-float
60 ;; are the type single-float and the type double-float. The precise
61 ;; definition of these categories is implementation-defined. The
62 ;; precision (measured in ``bits'', computed as p log 2b) and the
63 ;; exponent size (also measured in ``bits,'' computed as log 2(n+1),
64 ;; where n is the maximum exponent value) is recommended to be at least
65 ;; as great as the values in the next figure. Each of the defined
66 ;; subtypes of type float might or might not have a minus zero.
67 ;;
68 ;;
69 ;; Format Minimum Precision Minimum Exponent Size
70 ;; ----------
71 ;; Short 13 bits 5 bits
72 ;; Single 24 bits 8 bits
73 ;; Double 50 bits 8 bits
74 ;; Long 50 bits 8 bits
75 ;;
76 ;; Figure 12-12. Recommended Minimum Floating-Point Precision and Exponent Size
77 ;;
78 ;; There can be fewer than four internal representations for floats. If
79 ;; there are fewer distinct representations, the following rules apply:
80 ;;
81 ;; If there is only one, it is the type single-float. In this
82 ;; representation, an object is simultaneously of types single-float,
83 ;; double-float, short-float, and long-float.
84 ;;
85 ;; Two internal representations can be arranged in either of the
86 ;; following ways:
87 ;;
88 ;; Two types are provided: single-float and short-float. An object is
89 ;; simultaneously of types single-float, double-float, and long-float.
90 ;;
91 ;; Two types are provided: single-float and double-float. An object is
92 ;; simultaneously of types single-float and short-float, or double-float
93 ;; and long-float.
94 ;;
95 ;; Three internal representations can be arranged in either of the
96 ;; following ways:
97 ;;
98 ;; Three types are provided: short-float, single-float, and
99 ;; double-float. An object can simultaneously be of type double-float and
100 ;; long-float.
101 ;;
102 ;; Three types are provided: single-float, double-float, and
103 ;; long-float. An object can simultaneously be of types single-float and
104 ;; short-float.
106 ;; This should technically be conditionalized to the Lisp
107 ;; implementation, i.e.
111 ;; etc... the above need to be verified!
113 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
114 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
115 ;;;;
116 ;;;; Constants determining default floating point type for statistical
117 ;;;; operations. This type generally corresponds to a C double.
118 ;;;;
119 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
120 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
122 ;; Why do we do this instead of defparameter? mostly to allow for doc
123 ;; defs as well as to minimize dereferencing mistakes.
132 ;(defparameter +stat-float-typing+ 'long-float)
133 ;(defparameter +stat-cfloat-typing+ '(complex long-float))
134 ;(defparameter +stat-float-template+ 0.d0)
148 ;;; FIX-BASE-DOC adds note about modification to the end of the
149 ;;; documentation string argument.
152 "~a~%Modified to coerce arguments(s) to stat-float or stat-cfloat."
155 ;;; FIX-BASE-FUN-DOC fixes documentation of SYM and installs in BASE-SYM
160 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
161 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
162 ;;;;
163 ;;;; Functions and Macros for modifying functions to coerce to standard
164 ;;;; floating point type.
165 ;;;;
166 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
167 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
169 ;;; FFIX - coerces its arguments to standard real or complex floating
170 ;;; point number if needed.
171 #-stat-float-is-double-float
177 #+stat-float-is-double-float
180 ;;; MAKEDOUBLE coerces its argument to the standard floating point
181 ;;; type specified by +stat-float-template+
185 ;; Why doesn't this optimization work?
186 ;; #+stat-float-is-double-float
187 ;; (eval-when (compile)
188 ;; (proclaim '(function makedouble (t) long-float)))
190 ;;;
191 ;;; MAKE-BASE-TRANS-FUN Macro for re-defining one argument transcendental
192 ;;; functions
193 ;;;
200 ,@doc-list
204 #+:stat-float-is-double-float
211 ;;;
212 ;;; MAKE-BASE-TRANS-FUN-2 Macro for re-defining transcendental functions
213 ;;; with an optional second argument
214 ;;;
221 ,@doc-list
225 #+:stat-float-is-double-float
231 ;;;
232 ;;; Modified base functions to coerce to standard floating point type
233 ;;;
236 ;;; BASE-FLOAT
242 #+:stat-float-is-double-float
245 ;;; BASE-EXPT
254 #+:stat-float-is-double-float
257 ;;; Others