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
|
# This file is a part of Julia. License is MIT: https://julialang.org/license
# NOTES on range reduction
# [1] compute numbers near pi: http://www.cs.berkeley.edu/~wkahan/testpi/nearpi.c
# [2] range reduction: http://hal-ujm.ccsd.cnrs.fr/docs/00/08/69/04/PDF/RangeReductionIEEETC0305.pdf
# [3] precise addition, see Add22: http://ftp.nluug.nl/pub/os/BSD/FreeBSD/distfiles/crlibm/crlibm-1.0beta3.pdf
# Examples
# ΓΓ = 6411027962775774 / 2^45 # see [2] above, section 1.2
# julia> mod(ΓΓ, 2pi) # "naive" way - easily wrong
# 7.105427357601002e-15
# julia> mod2pi(ΓΓ) # using function provided here
# 2.475922546353431e-18
# Wolfram Alpha: mod(6411027962775774 / 2^45, 2pi)
# 2.475922546353430800060268586243862383453213646146648435... × 10^-18
# Test Cases. Each row contains: x and x mod 2pi (as from Wolfram Alpha)
# The values x are:
# -pi/2, pi/2, -pi, pi, 2pi, -2pi
# (or rather, the Float64 approx to those numbers.
# Thus, x mod pi will result in a small, but positive number)
# ΓΓ = 6411027962775774 / 2^47
# from [2], section 1.2:
# the Float64 greater than 8, and less than 2**63 − 1 closest to a multiple of π/4 is
# Γ = 6411027962775774 / 2^48. We take ΓΓ = 2*Γ to get cancellation with pi/2 already
# 3.14159265359, -3.14159265359
# pi/16*k +/- 0.00001 for k in [-20:20] # to cover all quadrants
# numerators of continuous fraction approximations to pi
# see http://oeis.org/A002485
# (reason: for max cancellation, we want x = k*pi + eps for small eps, so x/k ≈ pi)
testCases = [
-1.5707963267948966 4.71238898038469
1.5707963267948966 1.5707963267948966
-3.141592653589793 3.1415926535897936
3.141592653589793 3.141592653589793
6.283185307179586 6.283185307179586
-6.283185307179586 2.4492935982947064e-16
45.553093477052 1.5707963267948966
3.14159265359 3.14159265359
-3.14159265359 3.1415926535895866
-3.9269808169872418 2.356204490192345
-3.73063127613788 2.5525540310417068
-3.5342817352885176 2.748903571891069
-3.337932194439156 2.945253112740431
-3.1415826535897935 3.141602653589793
-2.9452331127404316 3.337952194439155
-2.7488835718910694 3.5343017352885173
-2.5525340310417075 3.730651276137879
-2.356184490192345 3.9270008169872415
-2.1598349493429834 4.123350357836603
-1.9634854084936209 4.319699898685966
-1.7671358676442588 4.516049439535328
-1.5707863267948967 4.71239898038469
-1.3744367859455346 4.908748521234052
-1.1780872450961726 5.105098062083414
-0.9817377042468104 5.301447602932776
-0.7853881633974483 5.4977971437821385
-0.5890386225480863 5.6941466846315
-0.3926890816987242 5.890496225480862
-0.1963395408493621 6.0868457663302244
1.0e-5 1.0e-5
0.19635954084936205 0.19635954084936205
0.3927090816987241 0.3927090816987241
0.5890586225480862 0.5890586225480862
0.7854081633974482 0.7854081633974482
0.9817577042468103 0.9817577042468103
1.1781072450961723 1.1781072450961723
1.3744567859455343 1.3744567859455343
1.5708063267948964 1.5708063267948964
1.7671558676442585 1.7671558676442585
1.9635054084936205 1.9635054084936205
2.159854949342982 2.159854949342982
2.3562044901923445 2.3562044901923445
2.5525540310417063 2.5525540310417063
2.7489035718910686 2.7489035718910686
2.9452531127404304 2.9452531127404304
3.1416026535897927 3.1416026535897927
3.3379521944391546 3.3379521944391546
3.534301735288517 3.534301735288517
3.7306512761378787 3.7306512761378787
3.927000816987241 3.927000816987241
-3.9270008169872415 2.356184490192345
-3.7306512761378796 2.552534031041707
-3.5343017352885173 2.7488835718910694
-3.3379521944391555 2.945233112740431
-3.141602653589793 3.1415826535897935
-2.9452531127404313 3.3379321944391553
-2.748903571891069 3.5342817352885176
-2.552554031041707 3.7306312761378795
-2.356204490192345 3.9269808169872418
-2.159854949342983 4.123330357836603
-1.9635054084936208 4.3196798986859655
-1.7671558676442587 4.516029439535328
-1.5708063267948966 4.71237898038469
-1.3744567859455346 4.908728521234052
-1.1781072450961725 5.105078062083414
-0.9817577042468104 5.301427602932776
-0.7854081633974483 5.497777143782138
-0.5890586225480863 5.694126684631501
-0.39270908169872415 5.890476225480862
-0.19635954084936208 6.086825766330224
-1.0e-5 6.283175307179587
0.19633954084936206 0.19633954084936206
0.39268908169872413 0.39268908169872413
0.5890386225480861 0.5890386225480861
0.7853881633974482 0.7853881633974482
0.9817377042468103 0.9817377042468103
1.1780872450961724 1.1780872450961724
1.3744367859455344 1.3744367859455344
1.5707863267948965 1.5707863267948965
1.7671358676442586 1.7671358676442586
1.9634854084936206 1.9634854084936206
2.1598349493429825 2.1598349493429825
2.3561844901923448 2.3561844901923448
2.5525340310417066 2.5525340310417066
2.748883571891069 2.748883571891069
2.9452331127404308 2.9452331127404308
3.141582653589793 3.141582653589793
3.337932194439155 3.337932194439155
3.534281735288517 3.534281735288517
3.730631276137879 3.730631276137879
3.9269808169872413 3.9269808169872413
22.0 3.1504440784612404
333.0 6.2743640266615035
355.0 3.1416227979431572
103993.0 6.283166177843807
104348.0 3.141603668607378
208341.0 3.141584539271598
312689.0 2.9006993893361787e-6
833719.0 3.1415903406703767
1.146408e6 3.1415932413697663
4.272943e6 6.283184757600089
5.419351e6 3.1415926917902683
8.0143857e7 6.283185292406739
1.65707065e8 3.1415926622445745
2.45850922e8 3.141592647471728
4.11557987e8 2.5367160519636766e-9
1.068966896e9 3.14159265254516
2.549491779e9 4.474494938161497e-10
6.167950454e9 3.141592653440059
1.4885392687e10 1.4798091093322177e-10
2.1053343141e10 3.14159265358804
1.783366216531e12 6.969482408757582e-13
3.587785776203e12 3.141592653589434
5.371151992734e12 3.1415926535901306
8.958937768937e12 6.283185307179564
1.39755218526789e14 3.1415926535898
4.28224593349304e14 3.1415926535897927
5.706674932067741e15 4.237546464512562e-16
6.134899525417045e15 3.141592653589793
]
function testModPi()
numTestCases = size(testCases,1)
modFns = [mod2pi]
xDivisors = [2pi]
errsNew, errsOld = Vector{Float64}(), Vector{Float64}()
for rowIdx in 1:numTestCases
xExact = testCases[rowIdx,1]
for colIdx in 1:1
xSoln = testCases[rowIdx,colIdx+1]
xDivisor = xDivisors[colIdx]
modFn = modFns[colIdx]
# 2. want: xNew := modFn(xExact) ≈ xSoln <--- this is the crucial bit, xNew close to xSoln
# 3. know: xOld := mod(xExact,xDivisor) might be quite a bit off from xSoln - that's expected
xNew = modFn(xExact)
xOld = mod(xExact,xDivisor)
newDiff = abs(xNew - xSoln) # should be zero, ideally (our new function)
oldDiff = abs(xOld - xSoln) # should be zero in a perfect world, but often bigger due to cancellation
oldDiff = min(oldDiff, abs(xDivisor - oldDiff)) # we are being generous here:
# if xOld happens to end up "on the wrong side of 0", eg
# if xSoln = 3.14 (correct), but xOld reports 0.01,
# we don't take the long way around the circle of 3.14 - 0.01, but the short way of 3.1415.. - (3.14 - 0.1)
push!(errsNew,abs(newDiff))
push!(errsOld,abs(oldDiff))
end
end
sort!(errsNew)
sort!(errsOld)
totalErrNew = sum(errsNew)
totalErrOld = sum(errsOld)
@test totalErrNew ≈ 0.0
end
testModPi()
# 2pi
@test mod2pi(10) ≈ mod(10,2pi)
@test mod2pi(-10) ≈ mod(-10,2pi)
@test mod2pi(355) ≈ 3.1416227979431572
@test mod2pi(Int32(355)) ≈ 3.1416227979431572
@test mod2pi(355.0) ≈ 3.1416227979431572
@test mod2pi(355.0f0) ≈ 3.1416228f0
@test mod2pi(Int64(2)^60) == mod2pi(2.0^60)
@test_throws ArgumentError mod2pi(Int64(2)^60-1)
@testset "rem_pio2_kernel" begin
# test worst case
x = 6381956970095103.0 * 2.0^797
a = setprecision(BigFloat, 4096) do
rem(big(x), big(pi)/2, RoundNearest)
end
n, yrem = Base.Math.rem_pio2_kernel(x)
y=yrem.hi+yrem.lo
@test a-y<nextfloat(y)/2
# The following has easy and hard cases in each interval. A hard case is one
# where x ≈ k*pi/2 for some integer k.
cases = [0.0, pi/6, # -π/4 <= x <= π/4
2*pi/4-0.1, 2*pi/4, # -2π/4 <= x <= 2π/4
3*pi/4-0.1, # -3π/4 <= x <= 3π/4
pi-0.1, Float64(pi), # -4π/4 <= x <= 4π/4
5*pi/4-0.1, # -5π/4 <= x <= 5π/4
6*pi/4-0.1, 6*pi/4, # -6π/4 <= x <= 6π/4
7*pi/4-0.1, # -7π/4 <= x <= 7π/4
2*pi, 2*pi-0.1, # -8π/4 <= x <= 8π/4
9*pi/4-0.1, # -9π/4 <= x <= 9π/4
2.0^10*pi/4, # -2.0^20π/2 <= x <= 2.0^20π/2
2.0^30*pi/4, # |x| >= 2.0^20π/2, idx < 0
2.0^80*pi/4] # |x| >= 2.0^20π/2, idx > 0-0.22370138542135648
# ieee754_rem_pio2_return contains the returned value from the ieee754_rem_pio2
# function in openlibm: https://github.com/JuliaLang/openlibm/blob/0598080ca09468490a13ae393ba17d8620c1b201/src/e_rem_pio2.c
ieee754_rem_pio2_return = [ 1.5707963267948966 1.5707963267948966;
1.0471975511965979 -1.0471975511965979;
0.10000000000000014 -0.10000000000000014;
6.123233995736766e-17 -6.123233995736766e-17;
-0.6853981633974481 0.6853981633974481;
0.10000000000000021 -0.10000000000000021;
1.2246467991473532e-16 -1.2246467991473532e-16;
-0.6853981633974481 0.6853981633974481;
0.09999999999999983 -0.09999999999999983;
1.8369701987210297e-16 -1.8369701987210297e-16;
-0.6853981633974484 0.6853981633974484;
2.4492935982947064e-16 -2.4492935982947064e-16;
0.0999999999999999 -0.0999999999999999;
-0.6853981633974484 0.6853981633974484;
3.135095805817224e-14 -3.135095805817224e-14;
3.287386219680602e-8 -3.287386219680602e-8;
-0.1757159771004682 0.1757159771004682 ]'
for (i, case) in enumerate(cases)
# negative argument
n, ret = Base.Math.rem_pio2_kernel(-case)
ret_sum = ret.hi+ret.lo
ulp_error = (ret_sum-ieee754_rem_pio2_return[1, i])/eps(ieee754_rem_pio2_return[1, i])
@test ulp_error <= 0.5
diff = Float64(mod(big(-case), big(pi)/2))-(ret.hi+ret.lo)
@test abs(diff) in (0.0, 1.5707963267948966, 1.5707963267948968)
# positive argument
n, ret = Base.Math.rem_pio2_kernel(case)
ret_sum = ret.hi+ret.lo
ulp_error = (ret_sum-ieee754_rem_pio2_return[2, i])/eps(ieee754_rem_pio2_return[2, i])
@test ulp_error <= 0.5
diff = Float64(mod(big(case), big(pi)/2))-(ret.hi+ret.lo)
@test abs(diff) in (0.0, 1.5707963267948966, 1.5707963267948968)
end
end
@testset "rem_pio2_kernel and mod2pi" begin
for int in (3632982096228748, 1135326194816)
bignum = int*big(pi)/2+0.00001
bigrem = rem(bignum, big(pi)/2, RoundDown)
fnum = Float64(bignum)
n, ret = Base.Math.rem_pio2_kernel(fnum)
@test mod2pi(fnum) == (ret.hi+ret.lo)
end
end
|
