IEEE_854_values
[b,p:above(1),
E_max:integer,
E_min:{i:integer|E_max>i}]: THEORY
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% This file defines the set of values that must be represented in each
% supported precision of an instance of IEEE-854
%
% Author:
% Paul S. Miner | email: [email protected]
% 1 South Wright St. / MS 130 | fax: (757) 864-4234
% NASA Langley Research Center | phone: (757) 864-6201
% Hampton, Virginia 23681 |
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
BEGIN
IMPORTING sum_hack, %for use until we have good summation library
sum_lemmas[b], % these are more specialized
% ints@mod_, %% note _ to avoid prelude version
enumerated_type_defs
Exponent: type = {i:int| E_min <= i & i <= E_max}
digits: type = [below(p)->below(b)]
fp_num: datatype
begin
finite(sign:sign_rep,Exp:Exponent,d:digits):finite?
infinite(i_sign:sign_rep): infinite?
NaN(status:NaN_type, data:NaN_data): NaN?
end fp_num
% A mapping from finite floating point numbers to real numbers
% This corresponds to the interpretation given in the standard
% section 3.1
fin : var (finite?)
value_digit(d:digits)(n:nat):nonneg_real =
if n<p then d(n)*b^(-n) else 0 endif
value(fin): real =
(-1) ^ sign(fin) * b ^ Exp(fin) * Sum(p, value_digit(d(fin)))
% Predicate testing whether an fp_num is 0
zero?(fp:fp_num):bool = if finite?(fp) then value(fp)=0 else false endif
shift_left((d: digits)): digits =
LAMBDA (i: below(p)): IF (i + 1 = p) THEN 0 ELSE d(i + 1) ENDIF
% Function normalize takes a finite and returns the finite of
% equivalent value with the smallest possible exponent.
normalize(fin:(finite?)): recursive (finite?) =
IF Exp(fin) = E_min or d(fin)(0) /= 0 then
fin
ELSE
normalize(finite(sign(fin),Exp(fin)-1,shift_left(d(fin))))
ENDIF
measure lambda (fin:(finite?)): Exp(fin) - E_min
normalize_idempotent: FACT
FORALL (fin: (finite?)): normalize(normalize(fin)) = normalize(fin)
% Predicate testing whether an fp_num is subnormal:
% Section 2 of the standard defines a subnormal (denormal in IEEE-754)
% number as a nonzero finite number whose exponent is the precision's
% minimum and whose leading significant digit is zero.
subnormal?(fp: fp_num): bool =
IF finite?(fp) THEN
not zero?(fp)
& Exp(normalize(fp)) = E_min
& d(normalize(fp))(0) = 0
ELSE FALSE
ENDIF
% Predicate testing whether an fp_num is normal:
% Section 2 of the standard defines a normal number as
% a nonzero finite number that is not subnormal
normal?((fp: fp_num)): bool =
IF finite?(fp) THEN d(normalize(fp))(0) > 0 ELSE FALSE ENDIF
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Some definitions and theorems to help illustrate the correctness
% of the definition of value.
% Maximum and Minimum representable values
max_significand:digits =
(lambda (i:below(p)): b-1)
min_significand: digits =
(LAMBDA (i: below(p)): IF i < p - 1 THEN 0 ELSE 1 ENDIF)
d_zero: digits = LAMBDA (i: below(p)): 0
m,n: var nat
Sum_value0: lemma Sum(n,value_digit(d_zero)) = 0
s: var sign_rep
E: var Exponent
zero_finite_d_zero: lemma zero?(finite(s,E,d_zero))
max_fp_pos : fp_num = finite(pos,E_max,max_significand)
min_fp_pos : fp_num = finite(pos,E_min,min_significand)
pos_zero : {fin|zero?(fin)}= finite(pos,E_min,d_zero)
max_fp_neg : fp_num = finite(neg,E_max,max_significand)
min_fp_neg : fp_num = finite(neg,E_min,min_significand)
neg_zero : {fin|zero?(fin)} = finite(neg,E_min,d_zero)
max_pos: posreal = value(max_fp_pos)
min_pos: posreal = value(min_fp_pos)
max_neg: negreal = value(max_fp_neg)
min_neg: negreal = value(min_fp_neg)
max_fp_correct: THEOREM
max_pos = b ^ (E_max +1) - b ^ (E_max + 1 - p)
min_fp_correct: THEOREM
value(min_fp_pos) = b ^ (E_min + 1 - p)
% Value of zero
value0: lemma value(fin)=0 iff d(fin)=d_zero
value_of_zero: THEOREM
value(pos_zero) = 0
% Normalization preserves the value
normal_value: lemma
value(fin) = value(normalize(fin))
% sign determines algebraic sign of value
value_positive: lemma
(sign(fin)=pos & not zero?(fin)) <=> value(fin)>0
value_negative: lemma
(sign(fin)=neg & not zero?(fin)) <=> value(fin)<0
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
finite_cover: lemma zero?(fin) or normal?(fin) or subnormal?(fin)
finite_disjoint1: lemma not (zero?(fin) & normal?(fin))
finite_disjoint2: lemma not (zero?(fin) & subnormal?(fin))
finite_disjoint3: lemma not (normal?(fin) & subnormal?(fin))
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
fp,fp1,fp2: var fp_num
% reverse the sign of a floating point number
toggle_sign(fp): fp_num
=
cases fp of
finite(s,e,d): finite(1-s,e,d),
infinite(s): infinite(1-s),
NaN(sig,data): fp
endcases
toggle_finite: lemma finite?(toggle_sign(fp)) iff finite?(fp)
toggle_infinite: lemma infinite?(toggle_sign(fp)) iff infinite?(fp)
toggle_NaN: lemma NaN?(toggle_sign(fp)) iff NaN?(fp)
value_toggle: lemma value(toggle_sign(fin)) = -value(fin)
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
toggle_d_normalize: lemma
d(normalize(toggle_sign(fin))) = d(normalize(fin))
toggle_Exp_normalize: lemma
Exp(normalize(toggle_sign(fin))) = Exp(normalize(fin))
toggle_zero?: lemma zero?(toggle_sign(fin)) iff zero?(fin)
toggle_normal?: lemma normal?(toggle_sign(fin)) iff normal?(fin)
toggle_subnormal?: lemma subnormal?(toggle_sign(fin)) iff subnormal?(fin)
% Sum lemmas
d: var digits
j : var nat
value_digit_le: lemma value_digit(d)(j) <= (b-1) * b^(-j)
Sum_d_lt_b: lemma Sum(j+1,value_digit(d)) <= b - b^(-j)
Sum_d_lt1: lemma d(0)=0 => Sum(j+1,value_digit(d)) <= 1-b^(-j)
Sum_d_ge1: lemma d(0)>0 => Sum(j+1,value_digit(d)) >= 1
value_sig_lt_b: lemma Sum(p,value_digit(d)) < b
value_sig_lt1: lemma d(0)=0 => Sum(p,value_digit(d)) < 1
value_sig_ge1: lemma d(0)>0 => Sum(p,value_digit(d)) >= 1
abs_value_fin: lemma
abs(value(fin)) = b^(Exp(fin)) * Sum(p,value_digit(d(fin)))
min_significand_min: lemma d/=d_zero =>
Sum(p,value_digit(min_significand))<= Sum(p,value_digit(d))
sign_normal: lemma sign(fin) = sign(normalize(fin))
value_normal: lemma normal?(fin) =>
b^E_min <= abs(value(fin)) & abs(value(fin)) <= max_pos
value_subnormal: lemma subnormal?(fin) =>
min_pos <= abs(value(fin)) & abs(value(fin)) < b^E_min
value_nonzero: lemma not zero?(fin) =>
min_pos <= abs(value(fin)) & abs(value(fin)) <= max_pos
%mod(floor(Sum(p, value_digit(d(normalize(fin!1)))) * b ^ x!1), b)
% = d(normalize(fin!1))(x!1)
scale_value(d,m)(n): nonneg_real = value_digit(d)(n) * b^m
scaled_Sum_int: lemma forall d,n:
forall (m:upto(n+1)): integer?(Sum(m,scale_value(d,n)))
scale_value_p(d)(n): nat = scale_value(d,p-1)(n)
i: var below(p)
scaled_Sum_i: lemma
Sum(n,value_digit(d))*b^(i) = Sum(n,scale_value(d,i))
scaled_Sum: lemma
Sum(n,value_digit(d))*b^(p-1) = Sum(n,scale_value_p(d))
floor_Sum: lemma
floor(Sum(p,value_digit(d)) * b^i) =
Sum(i+1,scale_value(d,i))
mod_Sum: lemma mod(Sum(i+1,scale_value(d,i)),b)=d(i)
%%%%%%%%%
%
% The next two declarations are here for importing purposes.
% A dummy fp_num to be used as a place holder. Any definition that
% returns holder is not yet finished.
holder:fp_num
% An undefined NaN for functions to return when the invalid exception
% should be raised. This serves as a placeholder until this specification
% accounts for exceptions.
%
invalid: (NaN?) = NaN(quiet,invalid_data)
END IEEE_854_values
¤ Dauer der Verarbeitung: 0.26 Sekunden
(vorverarbeitet)
¤
|
Haftungshinweis
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung ist noch experimentell.
|
