111
|
1 -- CXG2001.A
|
|
2 --
|
|
3 -- Grant of Unlimited Rights
|
|
4 --
|
|
5 -- Under contracts F33600-87-D-0337, F33600-84-D-0280, MDA903-79-C-0687,
|
|
6 -- F08630-91-C-0015, and DCA100-97-D-0025, the U.S. Government obtained
|
|
7 -- unlimited rights in the software and documentation contained herein.
|
|
8 -- Unlimited rights are defined in DFAR 252.227-7013(a)(19). By making
|
|
9 -- this public release, the Government intends to confer upon all
|
|
10 -- recipients unlimited rights equal to those held by the Government.
|
|
11 -- These rights include rights to use, duplicate, release or disclose the
|
|
12 -- released technical data and computer software in whole or in part, in
|
|
13 -- any manner and for any purpose whatsoever, and to have or permit others
|
|
14 -- to do so.
|
|
15 --
|
|
16 -- DISCLAIMER
|
|
17 --
|
|
18 -- ALL MATERIALS OR INFORMATION HEREIN RELEASED, MADE AVAILABLE OR
|
|
19 -- DISCLOSED ARE AS IS. THE GOVERNMENT MAKES NO EXPRESS OR IMPLIED
|
|
20 -- WARRANTY AS TO ANY MATTER WHATSOEVER, INCLUDING THE CONDITIONS OF THE
|
|
21 -- SOFTWARE, DOCUMENTATION OR OTHER INFORMATION RELEASED, MADE AVAILABLE
|
|
22 -- OR DISCLOSED, OR THE OWNERSHIP, MERCHANTABILITY, OR FITNESS FOR A
|
|
23 -- PARTICULAR PURPOSE OF SAID MATERIAL.
|
|
24 --*
|
|
25 --
|
|
26 -- OBJECTIVE:
|
|
27 -- Check that the floating point attributes Model_Mantissa,
|
|
28 -- Machine_Mantissa, Machine_Radix, and Machine_Rounds
|
|
29 -- are properly reported.
|
|
30 --
|
|
31 -- TEST DESCRIPTION:
|
|
32 -- This test uses a generic package to compute and check the
|
|
33 -- values of the Machine_ attributes listed above. The
|
|
34 -- generic package is instantiated with the standard FLOAT
|
|
35 -- type and a floating point type for the maximum number
|
|
36 -- of digits of precision.
|
|
37 --
|
|
38 -- APPLICABILITY CRITERIA:
|
|
39 -- This test applies only to implementations supporting the
|
|
40 -- Numerics Annex.
|
|
41 --
|
|
42 --
|
|
43 -- CHANGE HISTORY:
|
|
44 -- 26 JAN 96 SAIC Initial Release for 2.1
|
|
45 --
|
|
46 --!
|
|
47
|
|
48 -- References:
|
|
49 --
|
|
50 -- "Algorithms To Reveal Properties of Floating-Point Arithmetic"
|
|
51 -- Michael A. Malcolm; CACM November 1972; pgs 949-951.
|
|
52 --
|
|
53 -- Software Manual for Elementary Functions; W. J. Cody and W. Waite;
|
|
54 -- Prentice-Hall; 1980
|
|
55 -----------------------------------------------------------------------
|
|
56 --
|
|
57 -- This test relies upon the fact that
|
|
58 -- (A+2.0)-A is not necessarily 2.0. If A is large enough then adding
|
|
59 -- a small value to A does not change the value of A. Consider the case
|
|
60 -- where we have a decimal based floating point representation with 4
|
|
61 -- digits of precision. A floating point number would logically be
|
|
62 -- represented as "DDDD * 10 ** exp" where D is a value in the range 0..9.
|
|
63 -- The first loop of the test starts A at 2.0 and doubles it until
|
|
64 -- ((A+1.0)-A)-1.0 is no longer zero. For our decimal floating point
|
|
65 -- number this will be 1638 * 10**1 (the value 16384 rounded or truncated
|
|
66 -- to fit in 4 digits).
|
|
67 -- The second loop starts B at 2.0 and keeps doubling B until (A+B)-A is
|
|
68 -- no longer 0. This will keep looping until B is 8.0 because that is
|
|
69 -- the first value where rounding (assuming our machine rounds and addition
|
|
70 -- employs a guard digit) will change the upper 4 digits of the result:
|
|
71 -- 1638_
|
|
72 -- + 8
|
|
73 -- -------
|
|
74 -- 1639_
|
|
75 -- Without rounding the second loop will continue until
|
|
76 -- B is 16:
|
|
77 -- 1638_
|
|
78 -- + 16
|
|
79 -- -------
|
|
80 -- 1639_
|
|
81 --
|
|
82 -- The radix is then determined by (A+B)-A which will give 10.
|
|
83 --
|
|
84 -- The use of Tmp and ITmp in the test is to force values to be
|
|
85 -- stored into memory in the event that register precision is greater
|
|
86 -- than the stored precision of the floating point values.
|
|
87 --
|
|
88 --
|
|
89 -- The test for rounding is (ignoring the temporary variables used to
|
|
90 -- get the stored precision) is
|
|
91 -- Rounds := A + Radix/2.0 - A /= 0.0 ;
|
|
92 -- where A is the value determined in the first step that is the smallest
|
|
93 -- power of 2 such that A + 1.0 = A. This means that the true value of
|
|
94 -- A has one more digit in its value than 'Machine_Mantissa.
|
|
95 -- This check will detect the case where a value is always rounded.
|
|
96 -- There is an additional case where values are rounded to the nearest
|
|
97 -- even value. That is referred to as IEEE style rounding in the test.
|
|
98 --
|
|
99 -----------------------------------------------------------------------
|
|
100
|
|
101 with System;
|
|
102 with Report;
|
|
103 with Ada.Numerics.Generic_Elementary_Functions;
|
|
104 procedure CXG2001 is
|
|
105 Verbose : constant Boolean := False;
|
|
106
|
|
107 -- if one of the attribute computation loops exceeds Max_Iterations
|
|
108 -- it is most likely due to the compiler reordering an expression
|
|
109 -- that should not be reordered.
|
|
110 Illegal_Optimization : exception;
|
|
111 Max_Iterations : constant := 10_000;
|
|
112
|
|
113 generic
|
|
114 type Real is digits <>;
|
|
115 package Chk_Attrs is
|
|
116 procedure Do_Test;
|
|
117 end Chk_Attrs;
|
|
118
|
|
119 package body Chk_Attrs is
|
|
120 package EF is new Ada.Numerics.Generic_Elementary_Functions (Real);
|
|
121 function Log (X : Real) return Real renames EF.Log;
|
|
122
|
|
123
|
|
124 -- names used in paper
|
|
125 Radix : Integer; -- Beta
|
|
126 Mantissa_Digits : Integer; -- t
|
|
127 Rounds : Boolean; -- RND
|
|
128
|
|
129 -- made global to Determine_Attributes to help thwart optimization
|
|
130 A, B : Real := 2.0;
|
|
131 Tmp, Tmpa, Tmp1 : Real;
|
|
132 ITmp : Integer;
|
|
133 Half_Radix : Real;
|
|
134
|
|
135 -- special constants - not declared as constants so that
|
|
136 -- the "stored" precision will be used instead of a "register"
|
|
137 -- precision.
|
|
138 Zero : Real := 0.0;
|
|
139 One : Real := 1.0;
|
|
140 Two : Real := 2.0;
|
|
141
|
|
142
|
|
143 procedure Thwart_Optimization is
|
|
144 -- the purpose of this procedure is to reference the
|
|
145 -- global variables used by Determine_Attributes so
|
|
146 -- that the compiler is not likely to keep them in
|
|
147 -- a higher precision register for their entire lifetime.
|
|
148 begin
|
|
149 if Report.Ident_Bool (False) then
|
|
150 -- never executed
|
|
151 A := A + 5.0;
|
|
152 B := B + 6.0;
|
|
153 Tmp := Tmp + 1.0;
|
|
154 Tmp1 := Tmp1 + 2.0;
|
|
155 Tmpa := Tmpa + 2.0;
|
|
156 One := 12.34; Two := 56.78; Zero := 90.12;
|
|
157 end if;
|
|
158 end Thwart_Optimization;
|
|
159
|
|
160
|
|
161 -- determines values for Radix, Mantissa_Digits, and Rounds
|
|
162 -- This is mostly a straight translation of the C code.
|
|
163 -- The only significant addition is the iteration count
|
|
164 -- to prevent endless looping if things are really screwed up.
|
|
165 procedure Determine_Attributes is
|
|
166 Iterations : Integer;
|
|
167 begin
|
|
168 Rounds := True;
|
|
169
|
|
170 Iterations := 0;
|
|
171 Tmp := Real'Machine (((A + One) - A) - One);
|
|
172 while Tmp = Zero loop
|
|
173 A := Real'Machine(A + A);
|
|
174 Tmp := Real'Machine(A + One);
|
|
175 Tmp1 := Real'Machine(Tmp - A);
|
|
176 Tmp := Real'Machine(Tmp1 - One);
|
|
177
|
|
178 Iterations := Iterations + 1;
|
|
179 if Iterations > Max_Iterations then
|
|
180 raise Illegal_Optimization;
|
|
181 end if;
|
|
182 end loop;
|
|
183
|
|
184 Iterations := 0;
|
|
185 Tmp := Real'Machine(A + B);
|
|
186 ITmp := Integer (Tmp - A);
|
|
187 while ITmp = 0 loop
|
|
188 B := Real'Machine(B + B);
|
|
189 Tmp := Real'Machine(A + B);
|
|
190 ITmp := Integer (Tmp - A);
|
|
191
|
|
192 Iterations := Iterations + 1;
|
|
193 if Iterations > Max_Iterations then
|
|
194 raise Illegal_Optimization;
|
|
195 end if;
|
|
196 end loop;
|
|
197
|
|
198 Radix := ITmp;
|
|
199
|
|
200 Mantissa_Digits := 0;
|
|
201 B := 1.0;
|
|
202 Tmp := Real'Machine(((B + One) - B) - One);
|
|
203 Iterations := 0;
|
|
204 while (Tmp = Zero) loop
|
|
205 Mantissa_Digits := Mantissa_Digits + 1;
|
|
206 B := B * Real (Radix);
|
|
207 Tmp := Real'Machine(B + One);
|
|
208 Tmp1 := Real'Machine(Tmp - B);
|
|
209 Tmp := Real'Machine(Tmp1 - One);
|
|
210
|
|
211 Iterations := Iterations + 1;
|
|
212 if Iterations > Max_Iterations then
|
|
213 raise Illegal_Optimization;
|
|
214 end if;
|
|
215 end loop;
|
|
216
|
|
217 Rounds := False;
|
|
218 Half_Radix := Real (Radix) / Two;
|
|
219 Tmp := Real'Machine(A + Half_Radix);
|
|
220 Tmp1 := Real'Machine(Tmp - A);
|
|
221 if (Tmp1 /= Zero) then
|
|
222 Rounds := True;
|
|
223 end if;
|
|
224 Tmpa := Real'Machine(A + Real (Radix));
|
|
225 Tmp := Real'Machine(Tmpa + Half_Radix);
|
|
226 if not Rounds and (Tmp - TmpA /= Zero) then
|
|
227 Rounds := True;
|
|
228 if Verbose then
|
|
229 Report.Comment ("IEEE style rounding");
|
|
230 end if;
|
|
231 end if;
|
|
232
|
|
233 exception
|
|
234 when others =>
|
|
235 Thwart_Optimization;
|
|
236 raise;
|
|
237 end Determine_Attributes;
|
|
238
|
|
239
|
|
240 procedure Do_Test is
|
|
241 Show_Results : Boolean := Verbose;
|
|
242 Min_Mantissa_Digits : Integer;
|
|
243 begin
|
|
244 -- compute the actual Machine_* attribute values
|
|
245 Determine_Attributes;
|
|
246
|
|
247 if Real'Machine_Radix /= Radix then
|
|
248 Report.Failed ("'Machine_Radix incorrectly reports" &
|
|
249 Integer'Image (Real'Machine_Radix));
|
|
250 Show_Results := True;
|
|
251 end if;
|
|
252
|
|
253 if Real'Machine_Mantissa /= Mantissa_Digits then
|
|
254 Report.Failed ("'Machine_Mantissa incorrectly reports" &
|
|
255 Integer'Image (Real'Machine_Mantissa));
|
|
256 Show_Results := True;
|
|
257 end if;
|
|
258
|
|
259 if Real'Machine_Rounds /= Rounds then
|
|
260 Report.Failed ("'Machine_Rounds incorrectly reports " &
|
|
261 Boolean'Image (Real'Machine_Rounds));
|
|
262 Show_Results := True;
|
|
263 end if;
|
|
264
|
|
265 if Show_Results then
|
|
266 Report.Comment ("computed Machine_Mantissa is" &
|
|
267 Integer'Image (Mantissa_Digits));
|
|
268 Report.Comment ("computed Radix is" &
|
|
269 Integer'Image (Radix));
|
|
270 Report.Comment ("computed Rounds is " &
|
|
271 Boolean'Image (Rounds));
|
|
272 end if;
|
|
273
|
|
274 -- check the model attributes against the machine attributes
|
|
275 -- G.2.2(3)/3;6.0
|
|
276 if Real'Model_Mantissa > Real'Machine_Mantissa then
|
|
277 Report.Failed ("model mantissa > machine mantissa");
|
|
278 end if;
|
|
279
|
|
280 -- G.2.2(3)/2;6.0
|
|
281 -- 'Model_Mantissa >= ceiling(d*log(10)/log(radix))+1
|
|
282 Min_Mantissa_Digits :=
|
|
283 Integer (
|
|
284 Real'Ceiling (
|
|
285 Real(Real'Digits) * Log(10.0) / Log(Real(Real'Machine_Radix))
|
|
286 ) ) + 1;
|
|
287 if Real'Model_Mantissa < Min_Mantissa_Digits then
|
|
288 Report.Failed ("Model_Mantissa [" &
|
|
289 Integer'Image (Real'Model_Mantissa) &
|
|
290 "] < minimum mantissa digits [" &
|
|
291 Integer'Image (Min_Mantissa_Digits) &
|
|
292 "]");
|
|
293 end if;
|
|
294
|
|
295 exception
|
|
296 when Illegal_Optimization =>
|
|
297 Report.Failed ("illegal optimization of" &
|
|
298 " floating point expression");
|
|
299 end Do_Test;
|
|
300 end Chk_Attrs;
|
|
301
|
|
302 package Chk_Float is new Chk_Attrs (Float);
|
|
303
|
|
304 -- check the floating point type with the most digits
|
|
305 type A_Long_Float is digits System.Max_Digits;
|
|
306 package Chk_A_Long_Float is new Chk_Attrs (A_Long_Float);
|
|
307 begin
|
|
308 Report.Test ("CXG2001",
|
|
309 "Check the attributes Model_Mantissa," &
|
|
310 " Machine_Mantissa, Machine_Radix," &
|
|
311 " and Machine_Rounds");
|
|
312
|
|
313 Report.Comment ("checking Standard.Float");
|
|
314 Chk_Float.Do_Test;
|
|
315
|
|
316 Report.Comment ("checking a digits" &
|
|
317 Integer'Image (System.Max_Digits) &
|
|
318 " floating point type");
|
|
319 Chk_A_Long_Float.Do_Test;
|
|
320
|
|
321 Report.Result;
|
|
322 end CXG2001;
|