Class: Flt::DoubleFormat
Overview
Class to define formats such as “double double” by pairing two floating-point values which define a higher precision value by its sum. The :half parameter is a format class that defines the type of each part of the numbers. The formats defined here have a fixed precision, although these formats can actually have a variable precision. For binary formats there’s an option to gain one bit of precision by adjusting the sign of the second number. This is enabled by the :extra_prec option. For example, the “double double” format used in PowerPC is this
Flt.define :DoubleDouble, DoubleFormat, :half=>IEEE_binary64, :extra_prec=>true
Although this has a fixed 107 bits precision, the format as used in the PowerPC can have greater precision for specific values (by having greater separation between the exponents of both halves)
Constant Summary
Constants included
from Flt
APPLE, IEEE_128, IEEE_128_BE, IEEE_DEC128, IEEE_DEC32, IEEE_DEC64, IEEE_DOUBLE, IEEE_D_BE, IEEE_EXTENDED, IEEE_HALF, IEEE_H_BE, IEEE_QUAD, IEEE_Q_BE, IEEE_SINGLE, IEEE_S_BE, IEEE_X_BE, IEEE_binaryx, RPL, RPL_X
Instance Attribute Summary
Attributes inherited from FormatBase
#exponent, #sign, #significand
Class Method Summary
collapse
Instance Method Summary
collapse
Methods inherited from FormatBase
#<=>, arithmetic, arithmetic_type, bias, canonicalized, context, #convert_to, decimal_digits_necessary, decimal_digits_stored, decimal_max_exp, decimal_min_exp, endianness, epsilon, #form_class, #fp_format, from, from_bits, from_bits_text, from_bytes, from_hex, from_number, from_text, gradual_underflow?, half_epsilon, #infinite?, infinity, #initialize, join, max_value, maximum_integral_significand, min_normalized_value, min_value, minimum_normalized_integral_significand, #minus, minus_sign_value, nan, #nan?, #next_minus, #next_plus, nio_read_neutral, #nio_write_neutral, #normal?, num, num_class, pack_fields_hash, radix_log, radix_log10, radix_max_exp, radix_min_exp, radix_power, rounding_mode, sign_from_unit, sign_to_unit, #split, strict_epsilon, #subnormal?, switch_sign_value, #to, #to_a, #to_bits, #to_bits_text, #to_bytes, #to_hex, #to_num, #to_text, #ulp, unpack_fields_hash, zero, #zero?
Methods included from Flt
#*, #+, #-, #-@, #/, bcd2dpd, bitnot, convert_bytes, dbl_from_float, dbl_from_text, dbl_to_float, dpd2bcd, dpd_to_hexbcd, float_bin, float_dec, float_from_integral_sign_significand_exponent, float_from_integral_significand_exponent, float_shortest_dec, float_significant_dec, float_to_integral_sign_significand_exponent, float_to_integral_significand_exponent, hex_from_float, hex_to_float, hexbcd_to_dpd, sgl_from_float, sgl_from_text, sgl_to_float
Class Method Details
.define(params) ⇒ Object
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
|
# File 'lib/float-formats/classes.rb', line 1750
def self.define(params)
@half = params[:half]
params = @half.parameters.merge(params)
if @half.radix==2
@extra_prec = params[:extra_prec]
else
@extra_prec = false
end
params[:significand_digits] = 2*@half.significand_digits
params[:significand_digits] += 1 if @extra_prec
@significand_digits = params[:significand_digits]
@hidden_bit = params[:hidden_bit] = @half.hidden_bit
fields1 = []
fields2 = []
params[:fields].each_slice(2) do |s,v|
fields1 << s fields1 << v
fields2 << (s.to_s+"_2").to_sym
fields2 << v
end
params[:fields] = fields1 + fields2
define_fields params[:fields]
super params
end
|
.half ⇒ Object
1783
1784
1785
|
# File 'lib/float-formats/classes.rb', line 1783
def self.half
@half
end
|
.join_halfs(h1, h2) ⇒ Object
1822
1823
1824
|
# File 'lib/float-formats/classes.rb', line 1822
def self.join_halfs(h1,h2)
self.from_bytes(h1.to_bytes+h2.to_bytes)
end
|
.pack(s, m, e) ⇒ Object
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
|
# File 'lib/float-formats/classes.rb', line 1802
def self.pack(s,m,e)
if e.kind_of?(Symbol)
f1 = @half.pack(s,m,e)
f2 = @half.pack(+1,0,:zero)
else
s1,m1,e1,s2,m2,e2 = split_fp(s,m,e)
f1 = @half.pack(s1,m1,e1)
f2 = @half.pack(s2,m2,e2)
end
f1+f2
end
|
.radix ⇒ Object
1779
1780
1781
|
# File 'lib/float-formats/classes.rb', line 1779
def self.radix
@half.radix
end
|
.total_bits ⇒ Object
1832
1833
1834
|
# File 'lib/float-formats/classes.rb', line 1832
def self.total_bits
8*total_bytes
end
|
.total_bytes ⇒ Object
1829
1830
1831
|
# File 'lib/float-formats/classes.rb', line 1829
def self.total_bytes
2*@half.total_bytes
end
|
.total_nibbles ⇒ Object
1826
1827
1828
|
# File 'lib/float-formats/classes.rb', line 1826
def self.total_nibbles
2*total_bytes
end
|
.unpack(v) ⇒ Object
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
|
# File 'lib/float-formats/classes.rb', line 1787
def self.unpack(v)
sz = @half.total_bytes
v1 = @half.unpack(v[0...sz])
v2 = @half.unpack(v[sz..-1])
if v1.last == :nan
nan
elsif v1.last == :infinity
infinity(v1.sign)
else
v2 = @half.zero.split if @half.new(*v1).subnormal?
params = v1 + v2
join_fp(*params)
end
end
|
Instance Method Details
#split_halfs ⇒ Object
1814
1815
1816
1817
1818
1819
1820
|
# File 'lib/float-formats/classes.rb', line 1814
def split_halfs
b = to_bytes
sz = form_class.half.total_bytes
b1 = b[0...sz]
b2 = b[sz..-1]
[form_class.half.from_bytes(b1), form_class.half.from_bytes(b2)]
end
|