1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
|
(**
This file is part of the Flocq formalization of floating-point
arithmetic in Coq: http://flocq.gforge.inria.fr/
Copyright (C) 2010-2013 Sylvie Boldo
#<br />#
Copyright (C) 2010-2013 Guillaume Melquiond
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
COPYING file for more details.
*)
(** * Floating-point format with gradual underflow *)
Require Import Fcore_Raux.
Require Import Fcore_defs.
Require Import Fcore_rnd.
Require Import Fcore_generic_fmt.
Require Import Fcore_float_prop.
Require Import Fcore_FLX.
Require Import Fcore_FIX.
Require Import Fcore_ulp.
Require Import Fcore_rnd_ne.
Section RND_FLT.
Variable beta : radix.
Notation bpow e := (bpow beta e).
Variable emin prec : Z.
Context { prec_gt_0_ : Prec_gt_0 prec }.
(* floating-point format with gradual underflow *)
Definition FLT_format (x : R) :=
exists f : float beta,
x = F2R f /\ (Zabs (Fnum f) < Zpower beta prec)%Z /\ (emin <= Fexp f)%Z.
Definition FLT_exp e := Zmax (e - prec) emin.
(** Properties of the FLT format *)
Global Instance FLT_exp_valid : Valid_exp FLT_exp.
Proof.
intros k.
unfold FLT_exp.
generalize (prec_gt_0 prec).
repeat split ;
intros ; zify ; omega.
Qed.
Theorem generic_format_FLT :
forall x, FLT_format x -> generic_format beta FLT_exp x.
Proof.
clear prec_gt_0_.
intros x ((mx, ex), (H1, (H2, H3))).
simpl in H2, H3.
rewrite H1.
apply generic_format_F2R.
intros Zmx.
unfold canonic_exp, FLT_exp.
rewrite ln_beta_F2R with (1 := Zmx).
apply Zmax_lub with (2 := H3).
apply Zplus_le_reg_r with (prec - ex)%Z.
ring_simplify.
now apply ln_beta_le_Zpower.
Qed.
Theorem FLT_format_generic :
forall x, generic_format beta FLT_exp x -> FLT_format x.
Proof.
intros x.
unfold generic_format.
set (ex := canonic_exp beta FLT_exp x).
set (mx := Ztrunc (scaled_mantissa beta FLT_exp x)).
intros Hx.
rewrite Hx.
eexists ; repeat split ; simpl.
apply lt_Z2R.
rewrite Z2R_Zpower. 2: now apply Zlt_le_weak.
apply Rmult_lt_reg_r with (bpow ex).
apply bpow_gt_0.
rewrite <- bpow_plus.
change (F2R (Float beta (Zabs mx) ex) < bpow (prec + ex))%R.
rewrite F2R_Zabs.
rewrite <- Hx.
destruct (Req_dec x 0) as [Hx0|Hx0].
rewrite Hx0, Rabs_R0.
apply bpow_gt_0.
unfold canonic_exp in ex.
destruct (ln_beta beta x) as (ex', He).
simpl in ex.
specialize (He Hx0).
apply Rlt_le_trans with (1 := proj2 He).
apply bpow_le.
cut (ex' - prec <= ex)%Z. omega.
unfold ex, FLT_exp.
apply Zle_max_l.
apply Zle_max_r.
Qed.
Theorem FLT_format_bpow :
forall e, (emin <= e)%Z -> generic_format beta FLT_exp (bpow e).
Proof.
intros e He.
apply generic_format_bpow; unfold FLT_exp.
apply Z.max_case; try assumption.
unfold Prec_gt_0 in prec_gt_0_; omega.
Qed.
Theorem FLT_format_satisfies_any :
satisfies_any FLT_format.
Proof.
refine (satisfies_any_eq _ _ _ (generic_format_satisfies_any beta FLT_exp)).
intros x.
split.
apply FLT_format_generic.
apply generic_format_FLT.
Qed.
Theorem canonic_exp_FLT_FLX :
forall x,
(bpow (emin + prec - 1) <= Rabs x)%R ->
canonic_exp beta FLT_exp x = canonic_exp beta (FLX_exp prec) x.
Proof.
intros x Hx.
assert (Hx0: x <> 0%R).
intros H1; rewrite H1, Rabs_R0 in Hx.
contradict Hx; apply Rlt_not_le, bpow_gt_0.
unfold canonic_exp.
apply Zmax_left.
destruct (ln_beta beta x) as (ex, He).
unfold FLX_exp. simpl.
specialize (He Hx0).
cut (emin + prec - 1 < ex)%Z. omega.
apply (lt_bpow beta).
apply Rle_lt_trans with (1 := Hx).
apply He.
Qed.
(** Links between FLT and FLX *)
Theorem generic_format_FLT_FLX :
forall x : R,
(bpow (emin + prec - 1) <= Rabs x)%R ->
generic_format beta (FLX_exp prec) x ->
generic_format beta FLT_exp x.
Proof.
intros x Hx H.
destruct (Req_dec x 0) as [Hx0|Hx0].
rewrite Hx0.
apply generic_format_0.
unfold generic_format, scaled_mantissa.
now rewrite canonic_exp_FLT_FLX.
Qed.
Theorem generic_format_FLX_FLT :
forall x : R,
generic_format beta FLT_exp x -> generic_format beta (FLX_exp prec) x.
Proof.
clear prec_gt_0_.
intros x Hx.
unfold generic_format in Hx; rewrite Hx.
apply generic_format_F2R.
intros _.
rewrite <- Hx.
unfold canonic_exp, FLX_exp, FLT_exp.
apply Zle_max_l.
Qed.
Theorem round_FLT_FLX : forall rnd x,
(bpow (emin + prec - 1) <= Rabs x)%R ->
round beta FLT_exp rnd x = round beta (FLX_exp prec) rnd x.
intros rnd x Hx.
unfold round, scaled_mantissa.
rewrite canonic_exp_FLT_FLX ; trivial.
Qed.
(** Links between FLT and FIX (underflow) *)
Theorem canonic_exp_FLT_FIX :
forall x, x <> 0%R ->
(Rabs x < bpow (emin + prec))%R ->
canonic_exp beta FLT_exp x = canonic_exp beta (FIX_exp emin) x.
Proof.
intros x Hx0 Hx.
unfold canonic_exp.
apply Zmax_right.
unfold FIX_exp.
destruct (ln_beta beta x) as (ex, Hex).
simpl.
cut (ex - 1 < emin + prec)%Z. omega.
apply (lt_bpow beta).
apply Rle_lt_trans with (2 := Hx).
now apply Hex.
Qed.
Theorem generic_format_FIX_FLT :
forall x : R,
generic_format beta FLT_exp x ->
generic_format beta (FIX_exp emin) x.
Proof.
clear prec_gt_0_.
intros x Hx.
rewrite Hx.
apply generic_format_F2R.
intros _.
rewrite <- Hx.
apply Zle_max_r.
Qed.
Theorem generic_format_FLT_FIX :
forall x : R,
(Rabs x <= bpow (emin + prec))%R ->
generic_format beta (FIX_exp emin) x ->
generic_format beta FLT_exp x.
Proof with auto with typeclass_instances.
apply generic_inclusion_le...
intros e He.
unfold FIX_exp.
apply Zmax_lub.
omega.
apply Zle_refl.
Qed.
Theorem ulp_FLT_small: forall x, (Rabs x < bpow (emin+prec))%R ->
ulp beta FLT_exp x = bpow emin.
Proof with auto with typeclass_instances.
intros x Hx.
unfold ulp; case Req_bool_spec; intros Hx2.
(* x = 0 *)
case (negligible_exp_spec FLT_exp).
intros T; specialize (T (emin-1)%Z); contradict T.
apply Zle_not_lt; unfold FLT_exp.
apply Zle_trans with (2:=Z.le_max_r _ _); omega.
assert (V:FLT_exp emin = emin).
unfold FLT_exp; apply Z.max_r.
unfold Prec_gt_0 in prec_gt_0_; omega.
intros n H2; rewrite <-V.
apply f_equal, fexp_negligible_exp_eq...
omega.
(* x <> 0 *)
apply f_equal; unfold canonic_exp, FLT_exp.
apply Z.max_r.
assert (ln_beta beta x-1 < emin+prec)%Z;[idtac|omega].
destruct (ln_beta beta x) as (e,He); simpl.
apply lt_bpow with beta.
apply Rle_lt_trans with (2:=Hx).
now apply He.
Qed.
Theorem ulp_FLT_le :
forall x, (bpow (emin + prec - 1) <= Rabs x)%R ->
(ulp beta FLT_exp x <= Rabs x * bpow (1 - prec))%R.
Proof.
intros x Hx.
assert (Zx : (x <> 0)%R).
intros Z; contradict Hx; apply Rgt_not_le, Rlt_gt.
rewrite Z, Rabs_R0; apply bpow_gt_0.
rewrite ulp_neq_0 with (1 := Zx).
unfold canonic_exp, FLT_exp.
destruct (ln_beta beta x) as (e,He).
apply Rle_trans with (bpow (e-1)*bpow (1-prec))%R.
rewrite <- bpow_plus.
right; apply f_equal.
replace (e - 1 + (1 - prec))%Z with (e - prec)%Z by ring.
apply Z.max_l.
assert (emin+prec-1 < e)%Z; try omega.
apply lt_bpow with beta.
apply Rle_lt_trans with (1:=Hx).
now apply He.
apply Rmult_le_compat_r.
apply bpow_ge_0.
now apply He.
Qed.
Theorem ulp_FLT_gt :
forall x, (Rabs x * bpow (-prec) < ulp beta FLT_exp x)%R.
Proof.
intros x; case (Req_dec x 0); intros Hx.
rewrite Hx, ulp_FLT_small, Rabs_R0, Rmult_0_l; try apply bpow_gt_0.
rewrite Rabs_R0; apply bpow_gt_0.
rewrite ulp_neq_0; try exact Hx.
unfold canonic_exp, FLT_exp.
apply Rlt_le_trans with (bpow (ln_beta beta x)*bpow (-prec))%R.
apply Rmult_lt_compat_r.
apply bpow_gt_0.
now apply bpow_ln_beta_gt.
rewrite <- bpow_plus.
apply bpow_le.
apply Z.le_max_l.
Qed.
(** FLT is a nice format: it has a monotone exponent... *)
Global Instance FLT_exp_monotone : Monotone_exp FLT_exp.
Proof.
intros ex ey.
unfold FLT_exp.
zify ; omega.
Qed.
(** and it allows a rounding to nearest, ties to even. *)
Hypothesis NE_prop : Zeven beta = false \/ (1 < prec)%Z.
Global Instance exists_NE_FLT : Exists_NE beta FLT_exp.
Proof.
destruct NE_prop as [H|H].
now left.
right.
intros e.
unfold FLT_exp.
destruct (Zmax_spec (e - prec) emin) as [(H1,H2)|(H1,H2)] ;
rewrite H2 ; clear H2.
generalize (Zmax_spec (e + 1 - prec) emin).
generalize (Zmax_spec (e - prec + 1 - prec) emin).
omega.
generalize (Zmax_spec (e + 1 - prec) emin).
generalize (Zmax_spec (emin + 1 - prec) emin).
omega.
Qed.
End RND_FLT.
|
