1 ------------------------------------------------------------------------------
3 -- GNAT RUN-TIME COMPONENTS --
5 -- A D A . T E X T _ I O . F I X E D _ I O --
9 -- Copyright (C) 1992-2024, Free Software Foundation, Inc. --
11 -- GNAT is free software; you can redistribute it and/or modify it under --
12 -- terms of the GNU General Public License as published by the Free Soft- --
13 -- ware Foundation; either version 3, or (at your option) any later ver- --
14 -- sion. GNAT is distributed in the hope that it will be useful, but WITH- --
15 -- OUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY --
16 -- or FITNESS FOR A PARTICULAR PURPOSE. --
18 -- As a special exception under Section 7 of GPL version 3, you are granted --
19 -- additional permissions described in the GCC Runtime Library Exception, --
20 -- version 3.1, as published by the Free Software Foundation. --
22 -- You should have received a copy of the GNU General Public License and --
23 -- a copy of the GCC Runtime Library Exception along with this program; --
24 -- see the files COPYING3 and COPYING.RUNTIME respectively. If not, see --
25 -- <http://www.gnu.org/licenses/>. --
27 -- GNAT was originally developed by the GNAT team at New York University. --
28 -- Extensive contributions were provided by Ada Core Technologies Inc. --
30 ------------------------------------------------------------------------------
32 -- -------------------
33 -- - Fixed point I/O -
34 -- -------------------
36 -- The following text documents implementation details of the fixed point
37 -- input/output routines in the GNAT runtime. The first part describes the
38 -- general properties of fixed point types as defined by the Ada standard,
39 -- including the Information Systems Annex.
41 -- Subsequently these are reduced to implementation constraints and the impact
42 -- of these constraints on a few possible approaches to input/output is given.
43 -- Based on this analysis, a specific implementation is selected for use in
44 -- the GNAT runtime. Finally the chosen algorithms are analyzed numerically in
45 -- order to provide user-level documentation on limits for range and precision
46 -- of fixed point types as well as accuracy of input/output conversions.
48 -- -------------------------------------------
49 -- - General Properties of Fixed Point Types -
50 -- -------------------------------------------
52 -- Operations on fixed point types, other than input/output, are not important
53 -- for the purpose of this document. Only the set of values that a fixed point
54 -- type can represent and the input/output operations are significant.
59 -- The set of values of a fixed point type comprise the integral multiples of
60 -- a number called the small of the type. The small can be either a power of
61 -- two, a power of ten or (if the implementation allows) an arbitrary strictly
62 -- positive real value.
64 -- Implementations need to support ordinary fixed point types with a precision
65 -- of at least 24 bits, and (in order to comply with the Information Systems
66 -- Annex) decimal fixed point types with at least 18 digits. For the rest, no
67 -- requirements exist for the minimal small and range that must be supported.
72 -- [Wide_[Wide_]]Image attribute (see RM 3.5(27.1/2))
74 -- These attributes return a decimal real literal best approximating
75 -- the value (rounded away from zero if halfway between) with a
76 -- single leading character that is either a minus sign or a space,
77 -- one or more digits before the decimal point (with no redundant
78 -- leading zeros), a decimal point, and N digits after the decimal
79 -- point. For a subtype S, the value of N is S'Aft, the smallest
80 -- positive integer such that (10**N)*S'Delta is greater or equal to
81 -- one, see RM 3.5.10(5).
83 -- For an arbitrary small, this means large number arithmetic needs
86 -- Put (see RM A.10.9(22-26))
88 -- The requirements for Put add no extra constraints over the image
89 -- attributes, although it would be nice to be able to output more
90 -- than S'Aft digits after the decimal point for values of subtype S.
92 -- [Wide_[Wide_]]Value attribute (RM 3.5(39.1/2))
94 -- Since the input can be given in any base in the range 2..16,
95 -- accurate conversion to a fixed point number may require
96 -- arbitrary precision arithmetic if there is no limit on the
97 -- magnitude of the small of the fixed point type.
99 -- Get (see RM A.10.9(12-21))
101 -- The requirements for Get are identical to those of the Value
104 -- ------------------------------
105 -- - Implementation Constraints -
106 -- ------------------------------
108 -- The requirements listed above for the input/output operations lead to
109 -- significant complexity, if no constraints are put on supported smalls.
111 -- Implementation Strategies
112 -- -------------------------
114 -- * Floating point arithmetic
115 -- * Arbitrary-precision integer arithmetic
116 -- * Fixed-precision integer arithmetic
118 -- Although it seems convenient to convert fixed point numbers to floating
119 -- point and then print them, this leads to a number of restrictions.
120 -- The first one is precision. The widest floating-point type generally
121 -- available has 53 bits of mantissa. This means that Fine_Delta cannot
122 -- be less than 2.0**(-53).
124 -- In GNAT, Fine_Delta is 2.0**(-63), and Duration for example is a 64-bit
125 -- type. This means that a floating-point type with 64 bits of mantissa needs
126 -- to be used, which is only generally available on the x86 architecture. It
127 -- would still be possible to use multi-precision floating point to perform
128 -- calculations using longer mantissas, but this is a much harder approach.
130 -- The base conversions needed for input/output of (non-decimal) fixed point
131 -- types can be seen as pairs of integer multiplications and divisions.
133 -- Arbitrary-precision integer arithmetic would be suitable for the job at
134 -- hand, but has the drawback that it is very heavy implementation-wise.
135 -- Especially in embedded systems, where fixed point types are often used,
136 -- it may not be desirable to require large amounts of storage and time
137 -- for fixed I/O operations.
139 -- Fixed-precision integer arithmetic has the advantage of simplicity and
140 -- speed. For the most common fixed point types this would be a perfect
141 -- solution. The downside however may be a restricted set of acceptable
142 -- fixed point types.
144 -- Implementation Choices
145 -- ----------------------
147 -- The current implementation in the GNAT runtime uses fixed-precision integer
148 -- arithmetic for fixed point types whose Small is the ratio of two integers
149 -- whose magnitude is bounded relatively to the size of the mantissa, with a
150 -- two-tiered approach for 32-bit and 64-bit fixed point types. For the other
151 -- fixed point types, the implementation uses floating-point arithmetic.
153 -- The exact requirements of the algorithms are analyzed and documented along
154 -- with the implementation in their respective units.
157 with Ada
.Text_IO
.Fixed_Aux
;
158 with Ada
.Text_IO
.Float_Aux
;
159 with System
.Img_Fixed_32
; use System
.Img_Fixed_32
;
160 with System
.Img_Fixed_64
; use System
.Img_Fixed_64
;
161 with System
.Img_LFlt
; use System
.Img_LFlt
;
162 with System
.Val_Fixed_32
; use System
.Val_Fixed_32
;
163 with System
.Val_Fixed_64
; use System
.Val_Fixed_64
;
164 with System
.Val_LFlt
; use System
.Val_LFlt
;
166 package body Ada
.Text_IO
.Fixed_IO
with SPARK_Mode
=> Off
is
168 -- Note: we still use the floating-point I/O routines for types whose small
169 -- is not the ratio of two sufficiently small integers. This will result in
170 -- inaccuracies for fixed point types that require more precision than is
171 -- available in Long_Float.
173 subtype Int32
is Interfaces
.Integer_32
; use type Int32
;
174 subtype Int64
is Interfaces
.Integer_64
; use type Int64
;
177 Ada
.Text_IO
.Fixed_Aux
(Int32
, Scan_Fixed32
, Set_Image_Fixed32
);
180 Ada
.Text_IO
.Fixed_Aux
(Int64
, Scan_Fixed64
, Set_Image_Fixed64
);
182 package Aux_Long_Float
is new
183 Ada
.Text_IO
.Float_Aux
(Long_Float, Scan_Long_Float
, Set_Image_Long_Float
);
185 -- Throughout this generic body, we distinguish between the case where type
186 -- Int32 is OK and where type Int64 is OK. These boolean constants are used
187 -- to test for this, such that only code for the relevant case is included
188 -- in the instance; that's why the computation of their value must be fully
189 -- static (although it is not a static expressions in the RM sense).
191 OK_Get_32
: constant Boolean :=
192 Num
'Base'Object_Size <= 32
194 ((Num'Small_Numerator = 1 and then Num'Small_Denominator <= 2**31)
196 (Num'Small_Denominator = 1 and then Num'Small_Numerator <= 2**31)
198 (Num'Small_Numerator <= 2**27
199 and then Num'Small_Denominator <= 2**27));
200 -- These conditions are derived from the prerequisites of System.Value_F
202 OK_Put_32 : constant Boolean :=
203 Num'Base'Object_Size
<= 32
205 ((Num
'Small_Numerator = 1 and then Num
'Small_Denominator <= 2**31)
207 (Num
'Small_Denominator = 1 and then Num
'Small_Numerator <= 2**31)
209 (Num
'Small_Numerator < Num
'Small_Denominator
210 and then Num
'Small_Denominator <= 2**27)
212 (Num
'Small_Denominator < Num
'Small_Numerator
213 and then Num
'Small_Numerator <= 2**25));
214 -- These conditions are derived from the prerequisites of System.Image_F
216 OK_Get_64
: constant Boolean :=
217 Num
'Base'Object_Size <= 64
219 ((Num'Small_Numerator = 1 and then Num'Small_Denominator <= 2**63)
221 (Num'Small_Denominator = 1 and then Num'Small_Numerator <= 2**63)
223 (Num'Small_Numerator <= 2**59
224 and then Num'Small_Denominator <= 2**59));
225 -- These conditions are derived from the prerequisites of System.Value_F
227 OK_Put_64 : constant Boolean :=
228 Num'Base'Object_Size
<= 64
230 ((Num
'Small_Numerator = 1 and then Num
'Small_Denominator <= 2**63)
232 (Num
'Small_Denominator = 1 and then Num
'Small_Numerator <= 2**63)
234 (Num
'Small_Numerator < Num
'Small_Denominator
235 and then Num
'Small_Denominator <= 2**59)
237 (Num
'Small_Denominator < Num
'Small_Numerator
238 and then Num
'Small_Numerator <= 2**53));
239 -- These conditions are derived from the prerequisites of System.Image_F
241 E
: constant Natural := 63 - 32 * Boolean'Pos (OK_Put_32
);
242 -- T'Size - 1 for the selected Int{32,64}
244 F0
: constant Natural := 0;
245 F1
: constant Natural :=
246 F0
+ 18 * Boolean'Pos (2.0**E
* Num
'Small * 10.0**(-F0
) >= 1.0E+18);
247 F2
: constant Natural :=
248 F1
+ 9 * Boolean'Pos (2.0**E
* Num
'Small * 10.0**(-F1
) >= 1.0E+9);
249 F3
: constant Natural :=
250 F2
+ 5 * Boolean'Pos (2.0**E
* Num
'Small * 10.0**(-F2
) >= 1.0E+5);
251 F4
: constant Natural :=
252 F3
+ 3 * Boolean'Pos (2.0**E
* Num
'Small * 10.0**(-F3
) >= 1.0E+3);
253 F5
: constant Natural :=
254 F4
+ 2 * Boolean'Pos (2.0**E
* Num
'Small * 10.0**(-F4
) >= 1.0E+2);
255 F6
: constant Natural :=
256 F5
+ 1 * Boolean'Pos (2.0**E
* Num
'Small * 10.0**(-F5
) >= 1.0E+1);
257 -- Binary search for the number of digits - 1 before the decimal point of
258 -- the product 2.0**E * Num'Small.
260 For0
: constant Natural := 2 + F6
;
261 -- Fore value for the fixed point type whose mantissa is Int{32,64} and
262 -- whose small is Num'Small.
273 pragma Unsuppress
(Range_Check
);
277 Item
:= Num
'Fixed_Value
278 (Aux32
.Get
(File
, Width
,
279 -Num
'Small_Numerator,
280 -Num
'Small_Denominator));
282 Item
:= Num
'Fixed_Value
283 (Aux64
.Get
(File
, Width
,
284 -Num
'Small_Numerator,
285 -Num
'Small_Denominator));
287 Aux_Long_Float
.Get
(File
, Long_Float (Item
), Width
);
291 when Constraint_Error
=> raise Data_Error
;
299 Get
(Current_In
, Item
, Width
);
307 pragma Unsuppress
(Range_Check
);
311 Item
:= Num
'Fixed_Value
312 (Aux32
.Gets
(From
, Last
,
313 -Num
'Small_Numerator,
314 -Num
'Small_Denominator));
316 Item
:= Num
'Fixed_Value
317 (Aux64
.Gets
(From
, Last
,
318 -Num
'Small_Numerator,
319 -Num
'Small_Denominator));
321 Aux_Long_Float
.Gets
(From
, Long_Float (Item
), Last
);
325 when Constraint_Error
=> raise Data_Error
;
335 Fore
: Field
:= Default_Fore
;
336 Aft
: Field
:= Default_Aft
;
337 Exp
: Field
:= Default_Exp
)
341 Aux32
.Put
(File
, Int32
'Integer_Value (Item
), Fore
, Aft
, Exp
,
342 -Num
'Small_Numerator, -Num
'Small_Denominator,
345 Aux64
.Put
(File
, Int64
'Integer_Value (Item
), Fore
, Aft
, Exp
,
346 -Num
'Small_Numerator, -Num
'Small_Denominator,
349 Aux_Long_Float
.Put
(File
, Long_Float (Item
), Fore
, Aft
, Exp
);
355 Fore
: Field
:= Default_Fore
;
356 Aft
: Field
:= Default_Aft
;
357 Exp
: Field
:= Default_Exp
)
360 Put
(Current_Out
, Item
, Fore
, Aft
, Exp
);
366 Aft
: Field
:= Default_Aft
;
367 Exp
: Field
:= Default_Exp
)
371 Aux32
.Puts
(To
, Int32
'Integer_Value (Item
), Aft
, Exp
,
372 -Num
'Small_Numerator, -Num
'Small_Denominator,
375 Aux64
.Puts
(To
, Int64
'Integer_Value (Item
), Aft
, Exp
,
376 -Num
'Small_Numerator, -Num
'Small_Denominator,
379 Aux_Long_Float
.Puts
(To
, Long_Float (Item
), Aft
, Exp
);
383 end Ada
.Text_IO
.Fixed_IO
;