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
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
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
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
|
% apnum.tex -- Arbitrary Precision Numbers
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% The documentation, Petr Olsak, 2014
% You can create the pdf version of this documentation by the command
% pdfcsplain apnum.d
% run this command four times (for the consistence of all cross references)
% If you need to generate the documentation then the packages csplain and docbytex
% have to be installed in your TeX distribution.
\input utf8off \clearmubyte % use pdfcsplain
\def\projectversion{1.4 Dec 2015}
\def\headtitle{Arbitrary Precision Numbers}
\widowpenalty=10000
\clubpenalty=10000
\emergencystretch=2em
\hbadness=2200
\showboxbreadth=1500 \showboxdepth=2
\input docby.tex
\setlinecomment{\percent} \noactive{\nb\percent} \noactive{\percent\cbrace}
\noactive{\nb fontdimen}
\noactive{\nb apSIGNa} \noactive{\nb apSIGNb}
\noactive{\nb apEa} \noactive{\nb apEb}
\noactive{\nb documentclass}
\def\tittoc{Table Of Contents}
\def\titindex{Index}
\def\titversion{version }
\def\db{\dg\nb}
\def\du#1{\api{\nb#1}}%\def\quotehook{\langleactive\let\Blue=\Red \gdef\quotehook{\langleactive}}}
\def\quotehook{\langleactive\obeyspaces}
\bgroup
\catcode`\[=1 \catcode`]=2 \catcode`\{=12 \catcode`\}=12
\gdef\obrace[{] \gdef\cbrace[}]
\egroup
\def\indexhook{%
The bold number is the number of the page where the item is documented.
Other numbers are pagenumbers of the occurrences of such item.
The items marked by $\succ$ are mentioned in user documentation.
\medskip}
\def\nn#1 {\noactive{\nb#1}}
\def\inumref[#1]{\ilink[#1]{\numref[#1]}}
\def\cnvbookmark#1{\lowercase{\lowercase{#1}}}
{\obeyspaces\global\let =\ }
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\centerline{\tt apnum.tex}
\title Arbitrary Precision Numbers
\author Petr Ol\v s\'ak
\centerline{\ulink[ftp://math.feld.cvut.cz/olsak/makra/]%
{ftp://math.feld.cvut.cz/olsak/makra/}}
\dotoc \bookmarks
\sec User's Documentation
This macro file "apnum.tex" implements addition, subtraction,
multiplication, division, power to an integer and other calculation
($\sqrt x$, $e^x$, $\ln x$, $\sin x$, $\arctan x$, \dots) with
``large numbers'' with arbitrary number of decimal digits. The numbers
are in the form:
\begtt
<sign><digits>.<digits>
\endtt
%
where optional "<sign>" is the sequence of "+" and/or "-". The nonzero number is
treated as negative if and only if there is odd number of "-" signs.
The first part or second part of decimal "<digits>" (but no both) can be empty.
The decimal point is optional if second part of "<digits>" is empty.
There can be unlimited number of digits in the operands. Only \TeX{} main
memory or your patience during calculation with very large numbers are your
limits. Note, that the "apnum.tex" implementation includes a lot of
optimization and it is above 100 times faster (on large numbers) than the
implementation of the similar task in the package "fltpoint.sty". And the
"fp.sty" doesn't implements arbitrary number of digits. The extensive
technical documentation can serve as an inspiration how to do \TeX{} macro
programming.
\subsec [eval] Evaluation of Expressions
After "\input apnum" in your document you can use the macro
\du{evaldef}"\evaldef<sequence>{<expression>}".
It gives the possibility for comfortable calculation. The "<expression>" can
include numbers (in the form described above) combined by "+", "-", "*", "/"
and "^" operators and by possible brackets "()" in an usual way. The result
is stored to the "<sequence>" as a ``literal macro''. Examples:
\begtt
\evaldef\A {2+4*(3+7)}
% ... the macro \A includes 42
\evaldef\B {\the\pageno * \A}
% ... the macro \B includes 84
\evaldef\C {123456789000123456789 * -123456789123456789123456789}
% ... \C includes -15241578765447341344197531849955953099750190521
\evaldef\D {1.23456789 + 12345678.9 - \A}
% ... the macro \D includes 12345596.13456789
\evaldef\X {1/3}
% ... the macro \X includes .3333333333333333333
\endtt
%
The limit of the number of digits of the division result can be set by
\du{apTOT}"\apTOT" and \du{apFRAC}"\apFRAC" registers.
First one declares maximum calculated digits in total
and second one declares maximum of digits after decimal point. The
result is limited by both those registers. If the "\apTOT" is negative,
then its absolute value is treated as a ``soft limit'': all digits before
decimal point are calculated even if this limit is exceeded. The digits
after decimal point are not calculated when this limit is reached.
The special value "\apTOT=0" means that the calculation is limited
only by "\apFRAC". Default values are "\apTOT=0" and "\apFRAC=20".
The operator "^" means the powering, i.e "2^8" is "256". The exponent have
to be an integer (no decimal point is allowed) and a relatively small
integer is assumed.
The scanner of the "\evaldef" macro reads (roughly speaking) the "<expression>"
in the form ``operand binary-operator operand binary-operator etc.''
without expansion.
The spaces are not significant in the "<expression>". The operands are:
\begitems
\item * numbers (in the format "<sign><digits>.<digits>") or
\item * numbers in scientific notation (see the section \inumref[sci]) or
\item * sequences "<sign>\the<token>" or "<sign>\number<token>" or
\item * any other single "<token>" optionally preceded by "<sign>" and
optionally followed by a sequence of parameters enclosed in braces, for
example "\A" or "\B{<text>}" or "-\C{<textA>}{<textB>}". This case has
two meanings:
\begitems
\item * numeric constant defined in a ``literal macro''
("\def\A{42}", "\evaldef\A{13/15}") or
\item * ``function-like'' macro which returns a value after processing.
\enditems
\enditems
The "apnum.tex" macro file provides the following ``function-like'' macros
allowed to use them as an operand in the "<expression>":
\begitems
\item * \du{ABS}"\ABS {<value>}" for absolute value,
\item * \du{SGN}"\SGN {<value>}" returns sign of the "<value>",
\item * \du{iDIV}"\iDIV {<dividend>}{<divisor>}" for integer division,
\item * \du{iMOD}"\iMOD {<dividend>}{<divisor>}" for integer remainder,
\item * \du{iROUND}"\iROUND {<value>}" for rounding the number to the integer,
\item * \du{iFRAC}"\iFRAC {<value>}" for fraction part of the "\iROUND",
\item * \du{FAC}"\FAC {<integer value>}" for factorial,
\item * \du{BINOM}"\BINOM {<integer above>}{<integer below>}" for binomial coefficient,
\item * \du{SQRT}"\SQRT {<value>}" for square root of the "<value>",
\item * \du{EXP}"\EXP {<value>}" applies exponential function to "<value>",
\item * \du{LN}"\LN {<value>}" for natural logarithm of the "<value>",
\item * \du{SIN}\du{COS}\du{TAN}"\SIN {<value>}", "\COS {<value>}", "\TAN {<value>}"
for $\sin x$, $\cos x$ and $\tan x$ functions,
\item * \du{ASIN}\du{ACOS}\du{ATAN}"\ASIN {<value>}", "\ACOS {<value>}", "\ATAN {<value>}"
for $\arcsin x$, $\arccos x$ and $\arctan x$ functions,
\item * \du{PI}\du{PIhalf}"\PI", "\PIhalf" for constants $\pi$ and $\pi/2$.
\enditems
The arguments of all these functions can be a
nested "<expressions>" with the syntax like in the "\evaldef" macro.
Example:
\begtt
\def\A{20}
\evaldef\B{ 30*\SQRT{ 100 + 1.12*\the\widowpenalty } / (4-\A) }
\endtt
%
Note that the arguments of the ``function-like'' macros are enclosed by normal
\TeX{} braces "{}" but the round brackets "()" are used for re-arranging of the
common priority of the "+", "-", "*", "/" and "^" operators.
The macros "\SQRT", "\EXP", "\LN", "\SIN", "\COS", "\TAN", "\ASIN", "\ACOS",
"\ATAN" use "\apTOT" and "\apFRAC" registers similar like during division.
The "\PI" and "\PIhalf" are ``function-like'' macros without parameters.
They returns the constant with "\apFRAC" digits after decimal point.
Users can define their own ``function-like'' macros, see the section
\inumref[pgm].
\medskip
The output of "\evaldef\foo{<expression>}" processing is stored, of course,
to the ``literal macro'' "\foo". But there are another outputs like side
effect of the processing:
\begitems
\item * The \du{OUT}"\OUT" macro includes exactly the same result as "\foo".
\item * The \du{apSIGN}"\apSIGN" register includes the value "1" or "0"
or "-1" respectively dependent on the fact that the output is positive, zero
or negative.
\item * The \du{apE}"\apE" register is equal to the decimal exponent when
scientific number format is used (see the next section~\inumref[sci]).
\enditems
For example, you can
compare long numbers using "\apSIGN" register (where direct usage of "\ifnum"
primitive may cause arithmetic overflow):
\begtt
\TEST {123456789123456789} > {123456789123456788} \iftrue OK \else KO \fi
\endtt
The "\TEST" macro is defined like:
\begtt
\def\TEST#1#2#3#4{\evaldef\tmp{#1-(#3)}\ifnum\apSIGN #2 0 }
\endtt
The "apnum.tex" macros do not provide the evaluation of the "<expression>"
at the expansion level only. There are two reasons. First, the macros can be
used in classical \TeX{} only with Knuth's plain \TeX{} macro. No e\TeX{} is
needed. And the expansion-only evaluation of any expression isn't possible
in classical \TeX. Second reason is the speed optimization
(see the section~\inumref[tests]). Anyway, users
needn't expansion-only evaluation. They can write "\evaldef\a{<expression>}"
"\edef\foo{...\a...}" instead of "\edef\foo{...<expression>...}". There is
only one case when this ``pre-processing'' trick cannot be used: while
expansion of the parameters of asynchronous "\write" commands. But you can
save the "<expression>" unexpanded into the file and you can read the file
again in the second step and do "\evaldef" during reading the file.
\subsec [sci] Scientific Notation of Numbers
The macro "\evaldef" is able
to operate with the numbers written in the notation:
\begtt
<sign><digits>.<digits>E<sign><digits>
\endtt
%
For example "1.234E9" means $1.234\cdot 10^9$, i.e.\ "1234000000" or
the text "1.234E-3" means ".001234". The decimal exponent (after the "E"
letter) have to be in the range $\pm\,2\,147\,483\,647$ because
we store this value in normal \TeX{} register.
The "\evaldef<sequence>{<expression>}" operates by ``normal way'' if there are no operands
with "E" syntax in the "<expression>". But if an operand is expressed in
scientific form then "\evaldef" provide the calculation with the mantissa
and the exponent separately. Only the mantissa of the result is found in the
"<sequence>" and "\OUT" macros. The exponent of the result is stored in the "\apE"
register. You can define the macro which shows the complete result of the
calculation, for example:
\begtt
\def\showE#1{\message{#1\ifnum\apE=0 \else*10^\the\apE\fi}}
\endtt
Suppose "\evaldef\foo{<expression>}" is processed and the complete result is
$R={}$"\foo*10^\apE". There are two possibilities how to save such complete
result $R$ to the "\foo" macro: use "\apEadd\foo" or "\apEnum\foo". Both
macros do nothing if "\apE=0". Else the \du{apEadd}"\apEadd<sequence>" macro adds
"E<exponent>" to the "<sequence>" macro and \du{apEnum}"\apEnum<sequence>" moves
the decimal point to the new right position in the "<sequence>" macro or
appends zeros. The "\apE" register is set to zero after the macro "\apEadd" or
"\apEnum" is finished. Example:
\begtt
\evaldef\foo{ 3 * 4E9 } % \foo is 12, \apE=9
\apEadd\foo % \foo is 12E+9
\evaldef\foo{ 7E9 + 5E9 } % \foo is 12, \apE=9
\apEnum\foo % \foo is 12000000000
\endtt
There are another usable macros for operations with scientific numbers.
\begitems
\item * \du{apROLL}"\apROLL <sequence>{<shift>}" \dots the "<sequence>" is assumed to
be a macro with the number. The decimal point of this number is
shifted right by "<shift>" parameter, i.e.\ the result is multiplied by
"10^<shift>". The "<sequence>" is redefined by this result.
For example the "\apEnum\A" does "\apROLL\A{\apE}".
\item * \du{apNORM}"\apNORM <sequence>{<num>}"
\dots the "<sequence>" is supposed to be a macro with "<mantissa>" and it
will be redefined. The number "<mantissa>*10^\apE" (with current value of
the "\apE" register) is assumed.
The new mantissa saved in the "<sequence>" is the ``normalized mantissa'' of
the same number. The "\apE" register is corrected so the ``normalized
mantissa''"*10^\apE" gives the same number.
The "<num>" parameter is the number of non-zero digits before the decimal
point in the outputted mantissa. If the parameter "<num>"
starts by dot following by integer (for example "{.2}"), then the
outputted mantissa has "<num>" digits after decimal point.
For example "\def\A{1.234}\apE=0" "\apNORM\A{.0}" defines "\A" as "1234"
and "\apE=-3".
\item * The \du{apROUND}"\apROUND <sequence>{<num>}"
rounds the number, which is included in
the macro "<sequence>" and redefines "<sequence>" as rounded number.
The digits after decimal point at the position greater than "<num>" are ignored
in the rounded number. The decimal point is removed, if it is the right
most character in the "\OUT". The ignored part is saved to the "\XOUT" macro
without trailing right zeros.
\enditems
Examples of "\apROUND" usage:
\begtt
\def\A{12.3456}\apROUND\A{1} % \A is "12.3", \XOUT is "456"
\def\A{12.3456}\apROUND\A{9} % \A is "12.3456", \XOUT is empty
\def\A{12.3456}\apROUND\A{0} % \A is "12", \XOUT is "3456"
\def\A{12.0000}\apROUND\A{0} % \A is "12", \XOUT is empty
\def\A{12.0001}\apROUND\A{2} % \A is "12", \XOUT is "01"
\def\A{.000010}\apROUND\A{2} % \A is "0", \XOUT is "001"
\def\A{-12.3456}\apROUND\A{2} % \A is "-12.34", \XOUT is "56"
\def\A{12.3456}\apROUND\A{-1} % \A is "10", \XOUT is "23456"
\def\A{12.3456}\apROUND\A{-4} % \A is "0", \XOUT is "00123456"
\endtt
The following example saves the result of the "\evaldef" in scientific
notation with the mantissa with maximal three digits after decimal point and one
digit before.
\begtt
\evaldef\X{...}\apNORM\X{1}\apROUND\X{3}\apEadd\X
\endtt
The macros "\apEadd", "\apEnum", "\apROLL", "\apNORM" and "\apROUND" redefine the macro
"<sequence>" given as their first argument. They are not ``function-like''
macros and they cannot be used in an "<expression>".
The macro "<sequence>" must be the number in the format
"<simple sign><digits>.<digits>" where "<simple sign>" is one minus or none
and the rest of number has the format described in the first paragraph of
this documentation. The scientific notation isn't allowed here. This format
of numbers is in accordance with the output of the "\evaldef" macro.
\subsec [pgm] Notes for macro programmers
If you plan to create a ``function-like'' macro which can be used as an
operand in the "<expression>" then observe that first token in the
macro body must be "\relax". This tells to the "<expression>" scanner that
the calculation follows. The result of this calculation must be saved into
the "\OUT" macro and into the "\apSIGN" and "\apE" registers.
Example. The "\ABS" macro for the absolute value is defined by:
\ifirst{apnum.tex} {ABS}{^^B\cbrace}{+-}
\begtt
Usage: \evaldef\A{ 2 - \ABS{3-10} }% \A includes -5.
\endtt
Note, that "\apSIGN" register is corrected by final routine of "\evaldef"
according the "\OUT" value. But setting "\apSIGN" in your macro is
recommended because user can use your macro directly outside of "\evaldef".
The "\evaldef\foo{<expression>}" is processed in two steps. The
"<expression>" scanner converts the input to the macro call of the
\du{apPLUS}"\apPLUS", \du{apMINUS}"\apMINUS", \du{apMUL}"\apMUL",
\du{apDIV}"\apDIV" or \du{apPOW}"\apPOW" macros with two
parameters. They do addition, subtraction, multiplication, division and
power to the integer. These macros are processed in the second step.
For example:
\begtt
\evaldef\A{ 2 - 3*8 } converts the input to:
\apMINUS{2}{\apMUL{3}{8}} and this is processed in the second step.
\endtt
The macros "\apPLUS", "\apMINUS", "\apMUL", "\apDIV" and "\apPOW" behave
like normal ``function-like'' macros with one important exception: they
don't accept general "<expression>" in their parameters, only single operand
(see section~\inumref[eval]) is allowed.
If your calculation is processed in the loop very intensively
than it is better to save time of such calculation and to avoid the
"<expression>" scanner processing (first step of the "\evaldef").
So, it is recommended to use directly the
Polish notation of the expression as shown in the second line in the example
above. See section~\inumref[fce] for more inspirations.
The output of the "\apPLUS", "\apMINUS", "\apMUL", "\apDIV" and "\apPOW" macros is
stored in "\OUT" macro and the registers "\apSIGN" and "\apE" are set
accordingly.
The number of digits calculated by "\apDIV" macro is limited by the
"\apTOT" and "\apFRAC" registers as described in the section~\inumref[eval].
There is another result of "\apDIV" calculation stored in the \du{XOUT}"\XOUT" macro.
It is the remainder of the division. Example:
\begtt
\apTOT=0 \apFRAC=0 \apDIV{1234567892345}{2}\ifnum\XOUT=0 even \else odd\fi
\endtt
%
You cannot apply "\ifodd" primitive on ``large numbers'' directly
because the numbers may be too big.
If you set something locally inside your ``function-like'' macro, then such
data are accessible only when your macro is called outside of "\evaldef". Each
parameter and the whole "\evaldef" is processed inside a \TeX{} group, so
your locally set data are inaccessible when your macro is used inside another
``function-like'' parameter or inside "\evaldef". The "\XOUT" output is set
locally by "\apDIV" macro, so it serves as a good example of this feature:
\begtt
{\apDIV{1}{3} ... \XOUT is .00000000000000000001 }
... \XOUT is undefined
\evaldef{1/3} ... \XOUT is undefined
\apPLUS{1}{\apDIV{1}{3}} ... \XOUT is undefined
\endtt
The macro "\apPOW{<base>}{<exponent>}" calculates the power to the integer
exponent. A slight optimization is implemented here so the usage of "\apPOW"
is faster than repeated multiplication. The decimal non-integer exponents are not
allowed. Use "\EXP" and "\LN" macros if you need to calculate non-integer
exponent:
\begtt
\def\POWER#1#2{\relax \EXP{(#2)*\LN{#1}}}
\endtt
%
Note that both parameters are excepted as an "<expression>". Thus the "#2" is
surrounded in the rounded brackets.
Examples of another common ``function-like'' macros:
\begtt
\evaldef\degcoef{PI/180}
\def\SINdeg#1{\relax \SIN{\degcoef*(#1)}}
\def\COSdeg#1{\relax \COS{\degcoef*(#1)}}
\def\SINH#1{\relax \evaldef\myE{\EXP{#1}}\evaldef\OUT{(\myE-1/\myE)/2}}
\def\ASINH#1{\relax \LN{#1+\SQRT{(#1)^2+1}}}
\def\LOG#1{\relax \apLNtenexec \apDIV{\LN{#1}}{\apLNten}}
\endtt
In another example, we implement the field "\F{<index>}" as an
``function-like'' macro. User can set values by "\set\F{<index>}={<value>}"
and then these values can be used in an "<expression>".
\begtt
\def\set#1#2#3#4{\evaldef\index{#2}\evaldef\value{#4}%
\expandafter\edef\csname \string#1[\index]\endcsname{\value}}
\def\F#1{\relax % function-like macro
\evaldef\index{#1}%
\expandafter\ifx\csname \string\F[\index]\endcsname\relax
\def\OUT{0}% undefined value
\else
\edef\OUT{\csname \string\F[\index]\endcsname}%
\fi
}
\set \F{12/2} = {28+13}
\set \F{2*4} = {144^2}
\evaldef\test { 1 + \F{6} } \message{result=\test}
\endtt
%
As an exercise, you can implement linear interpolation of known values.
The final example shows, how to implement
the macro "\usedimen{<dimen>}{<unit>}". It is ``function-like'' macro,
it can be used in the "<expression>" and it returns the
"<decimal number>" with the property "<dimen>=<decimal number><unit>".
\begtt
\def\usedimen #1#2{\relax % function-like macro
\def\OUT{0}% % default value, if the unit isn't known
\csname dimenX#2\endcsname{#1}}
\def\dimenXpt #1{\apDIV{\number#1}{65536}}
\def\dimenXcm #1{\apDIV{\number#1}{1864682.7}}
\def\dimenXmm #1{\apDIV{\number#1}{186468.27}}
%... etc.
\evaldef\a{\usedimen{\hsize}{cm}} % \a includes 15.91997501773358008845
\endtt
%
Note that user cannot write "\usedimen\hsize{cm}" without braces because
this isn't the syntactically correct operand (see section~\inumref[eval]) and the
"<expression>" scanner is unable to read it.
\subsec [tests] Experiments
The following table shows the time needed for calculation of randomly
selected examples. The comparison with the package "fltpoint.sty" is shown.
The symbol $\infty$ means that it is out of my patience.
\bigskip
\noindent\hfil\vbox{\baselineskip=13pt
\halign{&\ \hfil#\hfil\ \cr
input & \# of digits in the result & time spent by {\tt apnum.tex} &
time spent by {\tt fltpoint.sty}\cr
\noalign{\smallskip\hrule\smallskip}
200! & 375 & 0.33 sec & 173 sec \cr
1000! & 2568 & 29 sec & $\infty$ \cr
$5^{17^2}$ & 203 & 0.1 sec & 81 sec \cr
$5^{17^3}$ & 3435 & 2.1 sec & $\infty$ \cr
$1/17$ & 1000 & 0.13 sec & 113 sec \cr
$1/17$ & 100000 & 142 sec & $\infty$ \cr
}}
\sec The Implementation
\subsec Name Convention, Version, Counters
The internal control sequence names typically used in "apnum.tex" have the form
"\apNAMEsuffix", but there are exceptions. The control sequences mentioned
in the section~\inumref[eval] (user's documentation) have typically more natural names.
And the internal counter registers have names "\apnumA", "\apnumB",
"\apnumC" etc.
The code starts by the greeting. The \db apVERSION includes the version of this software.
\ifirst{apnum.tex} {apVERSION}{\empty}{+-}
We declare auxiliary counters and one Boolean variable.
\inext{newcount}{\empty}{+-}
The counters \db apSIGN , \db apE , \db apTOT and \db apFRAC are declared here:
\inext{newcount}{\empty}{+-}
Somebody sometimes sets the "@" character to the special catcode. But we
need to be sure that there is normal catcode of the "@" character.
\inext{catcode}{}{++}
\subsec Evaluation of the Expression
Suppose the following expression "\A+\B*(\C+\D)+\E" as an example.
The main task of the "\evaldef\x{\A+\B*(\C+\D)+\E}" is to prepare the macro
"\tmpb" with the content (in this example)
"\apPLUS{\apPLUS{\A}{\apMUL{\B}{\apPLUS{\C}{\D}}}}{\E}" and to execute the "\tmpb"
macro.
The expression scanner adds the "\end" at the end of the expression and
reads from left to right the couples ``operand, operator''. For our example:
"\A+", "\B*", "\C+", "\D+" and "\E\end". The "\end" operator has the
priority 0, plus, minus have priority 1, "*"~and~"/" have priority 2 and "^"
has priority 3. The brackets are ignored, but each occurrence of the opening
bracket "(" increases priority by 4 and each occurrence of closing bracket
")" decreases priority by~4. The scanner puts each couple including its
current priority to the stack and does a test at the top of the stack. The
top of the stack is executed if the priority of the top operator is less or
equal the previous operator priority. For our example the stack is only
pushed without execution until "\D+" occurs. Our example in the stack looks
like:
\begtt
\D + 1 1<=5 exec:
\C + 5 {\C+\D} + 1 1<=2 exec:
\B * 2 \B * 2 {\B*{\C+\D}} + 1 1<=1 exec:
\A + 1 \A + 1 \A + 1 {\A+{\B*{\C+\D}}} + 1
bottom 0 bottom 0 bottom 0 bottom 0
\endtt
%
Now, the priority on the top is greater, then scanner pushes next couple and
does the test on the top of the stack again.
\begtt
\E \end 0 0<=1 exec:
{\A+{\B*{\C+\D}}} + 1 {{\A+{\B*{\C+\D}}}+\E} \end 0 0<=0 exec:
bottom 0 bottom 0 RESULT
\endtt
Let $p_t$, $p_p$ are the priority on the top and the previous
priority in the stack. Let $v_t$, $v_p$ are operands on the top and in the
previous line in the stack, and the same notation is used for operators
$o_t$ and $o_p$. If $p_t\le p_p$ then: pop the stack twice, create composed
operand $v_n=v_p \, o_p \, v_t$ and push $v_n$, $o_t$, $p_t$. Else
push new couple ``operand, operator'' from the expression scanner.
In both cases try to execute the top of the stack again.
If the bottom of the stack is reached then the last operand is the result.
The \db evaldef macro is protected by "\relax". It means that it can be used
inside an "<expression>" as a ``function-like'' macro, but I don't imagine
any usual application of this. The "\apEVALa" is executed.
\inext{apEVAL}{}{++}
The macro \db apEVALa "<sequence>{<expression>}" runs the evaluation of the
expression in the group. The base priority is initialized by "\apnumA=0",
then "\apEVALb<expression>\end" scans the expression and saves the
result in the form "\apPLUS{\A}{\apMUL{\B}{\C}}" (etc.) into the "\tmpb" macro. This
macro is executed. The group is finished by "\apEND" macro, which keeps the
"\OUT", "\apSIGN" and "\apE" values unchanged. Finally the defined "<sequence>" is set
equivalent to the \db OUT macro.
\inext{apEVALa}{}{++}
The scanner is in one of the two states: reading operand or reading operator.
The first state is initialized by \db apEVALb which follows to the
"\apEVALc". The \db apEVALc reads one token and switches by its value.
If the value is a "+" or "-" sign, it is assumed to be the part of the
operand prefix. Plus sign is ignored (and "\apEVALc" is run again),
minus signs are accumulated into "\tmpa".
The auxiliary macro \db apEVALd runs the following tokens to the "\fi", but
first it closes the conditional and skips the rest of the macro "\apEVALc".
\ilabel [eval:the] {ifx\nb the}
\ilabel [eval:number] {ifx\nb number}
\ilabel [eval:num] {apTESTdigit}
\ilabel [eval:nonum] {apEVALg}
\inext{apEVALb}{def\nb apEVALd}{++}
If the next token is opening bracket, then the global priority is increased
by 4 using the macro \db apEVALe. Moreover, if the sign before bracket
generates the negative result, then the new multiplication (by $-1$)
is added using "\apEVALp" to the operand stack.
\inext{apEVALe}{^^B\cbrace}{++}
If the next token is "\the" or "\number" primitives (see lines \cite[eval:the] and
\cite[eval:number]), then one following token is
assumed as \TeX{} register and these two tokens are interpreted as an operand.
This is done by \db apEVALf. The operand is packed to the "\tmpb" macro.
\inext{apEVALf}{}{++}
If the next token is not a number (the "\apTESTdigit#1\iftrue" results like
"\iffalse" at line~\cite[eval:num]) then we save
the sign plus this token to the "\tmpb"
at line~\cite[eval:nonum]
and we do check of the following token by "\futurelet". The \db
apEVALg is processed after that. The test is performed here if the following
token is open brace (a macro with parameter). If this is true then this
parameter is appended to "\tmpb" by \db apEVALh and the test about the
existence of second parameter in braces is repeated by next "\futurelet".
The result of this loop is stored into "\tmpb" macro which includes
"<sign>" followed by "<token>" followed by all
parameters in braces. This is considered as an operand.
\inext{apEVALg}{def\nb apEVALh}{++}
If the next token after the sign is a digit or a dot (tested in "\apEVALc"
by "\apTESTdigit" at line~\cite[eval:num]), then there are two cases. The
number includes the "E" symbol as a first symbol (this is allowed in
scientific notation, mantissa is assumed to equal to one). The "\apEVALk"
is executed in such case. Else the "\apEVALn" starts the reading the
number.
The first case with "E" letter in the number is solved by macros \db apEVALk
and \db apEVALm. The number after "E" is read by "\apE=" and this number is
appended to the "\tmpb" and the expression scanner skips to "\apEVALo".
\inext{apEVALk}{def\nb apEVALm}{++}
The second case (there is normal number) is processed by the macro \db
apEVALn. This macro reads digits (token per token) and saves them to the
"\tmpb". If the next token isn't digit nor dot then the second state of the
scanner (reading an operator) is invoked by running "\apEVALo".
If the "E" is found then the exponent is read to "\apE" and it is processed by
"\apEVALm".
\inext{apEVALn}{^^B\cbrace}{++}
The reading an operator is done by the \db apEVALo macro. This is more
simple because the operator is only one token. Depending on this token the
macro \db apEVALp "<operator><priority>" pushes to the stack (by
the macro "\apEVALpush") the value from "\tmpb", the "<operator"> and the priority
increased by "\apnumA" (level of brackets).
If there is a problem (level of brackets less than zero, level of brackets not
equal to zero at the end of the expression, unknown operator) we print an
error using "\apEVALerror" macro.
The "\apNext" is set to "\apEVALb", i.e.\ scanner returns back to the state of
reading the operand. But exceptions exist: if the ")" is found then
priority is decreased and the macro "\apEVALo" is executed again.
If the end of the "<expression>" is found then the loop is ended by
"\let\apNext=\relax".
\inext{apEVALo}{\count=2 ^^B\cbrace}{++}
The \db apEVALstack macro includes the stack, three items
"{<operand>}{<operator>}{<priority>}" per level. Left part of the macro
contents is the top of the stack. The stack is initialized with empty
operand and operator and with priority zero. The dot here is only the ``total
bottom'' of the stack.
\inext{apEVALstack}{}{++}
The macro \db apEVALpush "{<operand>}{<operator>}{<priority>}" pushes its
parameters to the stack and runs "\apEVALdo<whole stack>@" to do the
desired work on the top of the stack.
\inext{apEVALpush}{^^B\cbrace}{++}
Finally, the macro
\db apEVALdo "{<vt>}{<ot>}{<pt>}{<vp>}{<op>}{<pp>}<rest of the stack>@"
performs the execution described at the beginning of this section. The new
operand "<vn>" is created as "<op>{vp}{vt}", this means
"\apPLUS{<vp>}{<vt>}" for example. The operand is not executed now, only the
result is composed by the normal \TeX{} notation. If the bottom of the stack
is reached then the result is saved to the "\tmpb" macro. This macro is
executed after group by the "\apEVALa" macro.
\inext{apEVALdo}{^^B\cbrace}{++}
The macro \db apEVALerror "<string>" prints an error message. We decide to
be better to print only "\message", no "\errmessage". The "\tmpb" is
prepared to create "\OUT" as "??" and the "\apNext" macro is set in order to skip
the rest of the scanned "<expression>".
\inext{apEVALerror}{^^B\cbrace}{++}
The auxiliary macro \db apTESTdigit "<token>\iftrue" tests, if the given token is
digit, dot or "E" letter.
\inext{apTESTdigit}{^^B\cbrace}{++}
\subsec Preparation of the Parameter
All operands of "\apPLUS", "\apMINUS", "\apMUL", "\apDIV" and "\apPOW" macros are
preprocessed by "\apPPa" macro. This macro solves (roughly speaking) the
following tasks:
\begitems
\item * It partially expands (by "\expandafter") the parameter while "<sign>" is read.
\item * The "<sign>" is removed from parameter and the appropriate "\apSIGN"
value is set.
\item * If the next token after "<sign>" is "\relax" then the rest of the
parameter is executed in the group and the results "\OUT", "\apSIGN" and
"\apE" are used.
\item * Else the number is read and saved to the parameter.
\item * If the read number has the scientific notation "<mantissa>E<exponent>"
then only "<mantissa>" is saved to the parameter and "\apE" is set as
"<exponent>". Else "\apE" is zero.
\enditems
The macro \db apPPa "<sequence><parameter>" calls
\db apPPb "<parameter>@<sequence>" and starts reading the
"<parameter>". The result will be stored to the "<sequence>".
Each token from "<sign>" is processed by three "\expandafter"s (because
there could be "\csname...\endcsname"). It means that the
parameter is partially expanded when "<sign>" is read. The "\apPPb" macro
sets the initial value of "\tmpc" and "\apSIGN" and executes the macro
\db apPPc "<parameter>@<sequence>".
\inext{apPPa}{def\nb apPPd}{++}
The "\apPPc" reads one token from "<sign>" and it is called recursively
while there are "+" or "-" signs. If the read token is "+" or "-" then
the \db apPPd closes conditionals and executes "\apPPc" again.
If "\relax" is found then the rest of parameter is executed by the
\db apPPe. The macro ends by \db apPPf "<result>@" and this macro
reverses the sign if the result is negative and removes the minus sign
from the front of the parameter.
\inext{apPPe}{def\nb apPPf}{++}
The \db apPPg "<parameter>@" macro is called when the "<sign>" was processed
and removed from the input stream. The main reason of this macro is to
remove trailing zeros from the left and to check, if there is the zero value
written for example in the form "0000.000". When this macro is started then
"\tmpc" is empty. This is a flag for removing trailing zeros. They are simply
ignored before decimal point. The "\apPPg" is called again by \db apPPh
macro which removes the rest of "\apPPg" macro and closes the conditional.
If the decimal point is found then next zeros are accumulated to the "\tmpc".
If the end of the parameter "@" is found and we are in the ``removing zeros
state'' then the whole value is assumed to be zero and this is processed by
"\apPPi @". If another digit is found (say 2) then there are two
situations: if the "\tmpc" is non-empty, then the digit is appended to the
"\tmpc" and the "\apPPi<expanded tmp>" is processed (say "\apPPi .002")
followed by the rest of the parameter. Else the digit itself is stored to
the "\tmpc" and it is returned back to the input stream (say "\apPPi"~"2")
followed by the rest of the parameter.
\inext{apPPg}{def\nb apPPh}{++}
The macro \db apPPi "<parameter without trailing zeros>@<sequence>"
switches to two cases: if the execution of the parameter was processed then
the "\OUT" doesn't include "E" notation and we can simply define
"<sequence>" as the "<parameter>" by the \db apPPj macro. This saves the
copying of the (possible) long result to the input stream again.
If the executing of the parameter was not performed, then we need to test
the existence of the "E" notation of the number by the \db apPPk macro. We
need to put the "<parameter>" to the input stream and to use \db apPPl to
test these cases. We need to remove unwanted "E" letter by the \db apPPm macro.
\inext{apPPi}{def\nb apPPm}{++}
The \db apPPn "<param>" macro does the same as "\apPPa\OUT{<param>}",
but the minus sign is returned back to the "\OUT" macro if the result is
negative.
\inext{apPPn}{}{++}
The \db apPPab "<macro>{<paramA>}{<paramB>}" is used for parameters of
all macros "\apPLUS", "\apMUL" etc. It
prepares the "<paramA>" to "\tmpa", "<paramB>" to "\tmpb", the sign and
"<decimal exponent>" of "<paramA>" to the "\apSIGNa" and "\apEa", the same
of "<paramB>" to the "\apSIGNa" and "\apEa". Finally, it executes the
"<macro>".
\inext{apPPab}{^^B\cbrace}{++}
The \db apPPs "<macro><sequence>{<param>}" prepares parameters for "\apROLL",
"\apROUND" and "\apNORM" macros. It saves the "<param>" to the "\tmpc" macro,
expands the "<sequence>" and runs the macro
\db apPPt \unskip~"<macro><expanded sequence>.@<sequence>". The macro "\apPPt"
reads first token from the "<expanded sequence>" to "#2". If "#2" is minus
sign, then "\apnumG=-1". Else
"\apnumG=1". Finally the "<macro><expanded sequence>.@<sequence>"
is executed (but without the minus sign in the input stream).
If "#2" is zero then \db apPPu "<macro><rest>.@<sequence>" is executed. If
the "<rest>" is empty, (i.e.\ the parameter is simply zero) then "<macro>"
isn't executed because there in nothing to do with zero number as a parameter of
"\apROLL", "\apROUND" or "\apNORM" macros.
\inext{apPPs}{\count=2 ^^B\cbrace}{++}
\subsec Addition and Subtraction
The significant part of the optimization in "\apPLUS", "\apMUL", "\apDIV" and
"\apPOW" macros
is the fact, that we don't treat with single decimal digits but with their
quartets. This means that we are using the numeral system with the base
10000 and we calculate four decimal digits in one elementary operation. The
base was chosen $10^4$ because the multiplication of such numbers gives
results less than $10^8$ and the maximal number in the \TeX{} register
is about $2\cdot10^9$. We'll use the word ``Digit'' (with capitalized D) in
this documentation if this means the digit in the numeral system with base
10000, i.e.\ one Digit is four digits.
Note that for addition we can use the numeral system with the base $10^8$
but we don't do it, because the auxiliary macros "\apIV*" for numeral system of the
base $10^4$ are already prepared.
Suppose the following example (the spaces between Digits are here only for
more clarity).
\begtt
123 4567 8901 9999 \apnumA=12 \apnumE=3 \apnumD=16
+ 22.423 \apnumB=0 \apnumF=2 \apnumC=12
--------------------------
sum in reversed order and without transmissions:
{4230}{10021}{8901}{4567}{123} \apnumD=-4
sum in normal order including transmissions:
123 4567 8902 0021.423
\endtt
In the first pass, we put the number with more or equal Digits before decimal
point above the second number. There are three Digits more in the example.
The "\apnumC" register saves this information (multiplied by 4). The first
pass creates the sum in reversed order without transmissions between Digits.
It simply copies the "\apnumC/4" Digits from the first number to the result in reversed
order. Then it does the sums of Digits without transmissions. The "\apnumD"
is a relative position of the decimal point to the edge of the calculated
number.
The second pass reads the result of the first pass, calculates transmissions and
saves the result in normal order.
The first Digit of the operands cannot include four digits. The number of
digits in the first Digit is saved in "\apnumE" (for first operand) and in
"\apnumF" (for second one). The rule is to have the decimal point between
Digits in all circumstances.
The \db apPLUS and \db apMINUS macros prepare parameters using "\apPPab" and execute "\apPLUSa":
\inext{apPLUS}{apMINUS}{++}
The macro \db apPLUSa does the following work:
\ilabel[plus:apE] {apEa}
\ilabel[plus:DIGa] {apDIG\nb tmpa}
\ilabel[plus:DIGb] {apDIG\nb tmpb}
\ilabel[plus:moda] {-\nb apnumE}
\ilabel[plus:modb] {-\nb apnumF}
\ilabel[plus:apnC] {apnumC=}
\ilabel[plus:xA] {apSIGNa}
\ilabel[plus:xB] {apSIGNb}
\ilabel[plus:sg] {apSIGN=}
\ilabel[plus:xAm] {PLUSxA-}
\ilabel[plus:ba] {apPLUSg}
\ilabel[plus:bb] {apnumC=-}
\ilabel[plus:G] {apnumG=0}
\ilabel[plus:next] {apNext=}
\ilabel[plus:X] {apnumX=0}
\ilabel[plus:fa] {00123}
\ilabel[plus:fb] {apPLUSy}
\begitems
\item * It gets the operands in "\tmpa" and "\tmpb" macros using the "\apPPab".
\item * If the scientific notation is used and the decimal
exponents "\apEa" and "\apEb" are not equal then the decimal point of one
operand have to be shifted (by the macro "\apPLUSxE" at line~\cite[plus:apE]).
\item * The digits before decimal point are calculated for both operands by
the "\apDIG" macro. The first result is saved to "\apnumA" and the second
result is saved to "\apnumB". The "\apDIG" macro removes decimal point (if
exists) from the parameters (lines~\cite[plus:DIGa] and~\cite[plus:DIGb]).
\item * The number of digits in the first Digit is calculated by "\apIVmod"
for both operands. This number is saved to "\apnumE" and "\apnumF". This
number is subtracted from "\apnumA" and "\apnumB", so these
registers now includes multiply of four
(lines~\cite[plus:moda] and~\cite[plus:modb]).
\item * The "\apnumC" includes the difference of Digits before the decimal
point (multiplied by four) of given operands
(line~\cite[plus:apnC]).
\item * If the first operand is negative then the minus sign is inserted to
the \db apPLUSxA macro else this macro is empty. The same for the second
operand and for the macro \db apPLUSxB is done
(lines~\cite[plus:xA] and~\cite[plus:xB]).
\item * If both operands are positive, then the sign of the result "\apSIGN"
is set to one. If both operands are negative, then the sign is set to $-1$.
But in both cases mentioned above we will do (internally) addition, so the
macros "\apPLUSxA" and "\apPLUSxB" are set to empty.
If one operand is negative and second positive then we will do
subtraction. The "\apSIGN" register is set to zero and
it will set to the right value later
(lines~\cite[plus:sg] to~\cite[plus:xAm]).
\item * The macro "\apPLUSb<first op><first dig><second op><second dig><first Dig>"
does the calculation of the first pass. The "<first op>" has to have more
or equal Digits before decimal point than "<second op>". This is reason why
this macro is called in two variants dependent on the value "\apnumC".
The macros "\apPLUSxA" and "\apPLUSxB" (with the sign of the operands) are
exchanged (by the "\apPLUSg") if the operands are exchanged
(lines~\cite[plus:ba] to~\cite[plus:bb]).
\item * The "\apnumG" is set by the macro "\apPLUSb" to the sign of the
first nonzero Digit. It is equal to zero if there are only zero Digits after
first pass. The result is zero in such case and we do nothing more
(line~\cite[plus:G]).
\item * The transmission calculation is different for addition and
subtraction. If the subtraction is processed then the sign of the result
is set (using the value "\apnumG") and the "\apPLUSm" for transmissions is
prepared. Else the "\apPLUSp" for transmissions is prepared as the "\apNext" macro
(line~\cite[plus:next])
\item * The result of the first pass is expanded in the input stream and the
"\apNext" (i.e.\ transmissions calculation) is activated at line~\cite[plus:X].
\item * if the result is in the form ".000123", then the decimal point and
the trailing zeros have to be inserted. Else the trailing zeros from the
left side of the result have to be removed by "\apPLUSy". This macro adds
the sign of the result too
(lines~\cite[plus:fa] to~\cite[plus:fb])
\enditems
\inext{apPLUSa}{^^B\cbrace}{++}
The macro \db apPLUSb "<first op><first dig><second op><second dig><first Dig>"
starts the first pass. The "<first op>" is the first operand (which have
more or equal Digits before decimal point). The "<first dig>" is the number
of digits in the first Digit in the first operand. The "<second op>" is the
second operand and the "<second dig>" is the number of digits in the first
Digit of the second operand. The "<first Dig>" is the number of Digits
before decimal point of the first operand, but without the first Digit and
multiplied by~4.
The macro"\apPLUSb" saves the second operand to "\tmpd" and appends the
$4-{}$"<second dig>" empty parameters before this operand in order to
read desired number of digits to the first Digit of this oparand.
The macro "\apPLUSb" saves the first operand to the input queue after
"\apPLUSc" macro. It inserts the appropriate number of empty parameters (in
"\tmpc") before this operand in order to read the right number of digits in
the first attempt. It appends the "\apNL" marks to the end in order to
recognize the end of the input stream. These macros expands simply to zero
but we can test the end of input stream by "\ifx".
The macro "\apPLUSb" calculates the number of digits before decimal point
(rounded up to multiply by 4) in "\apnumD" by advancing "<first DIG>" by~4.
It initializes "\apnumZ" to zero. If the first nonzero Digit will be found
then "\apnumZ" will be set to this Digit in the "\apPLUSc" macro.
\inext{apPLUSb}{^^B\cbrace}{++}
The macro \db apPLUSc is called repeatedly. It reads one Digit from input
stream and saves it to the "\apnumY". Then it calls the \db apPLUSe, which
reads (if it is allowed, i.e.\ if "\apnumC"{\tt\char`<}"=0") one digit from
second operand "\tmpd" by the "\apIVread" macro.
Then it does the addition of these digits and saves the result
into the "\OUT" macro in reverse order.
Note, that the sign "\apPLUSxA" is used when "\apnumY" is read and the sign
"\apPLUSxB" is used when advancing is performed. This means that we are
doing addition or subtraction here.
If the first nonzero Digit is reached, then the macro \db apPLUSh sets the
sign of the result to the "\apnumG" and (maybe) exchanges the "\apPLUSxA"
and "\apPLUSxB" macros (by the \db apPLUSg macro)
in order to the internal result of the subtraction will be always non-negative.
If the end of input stream is reached, then "\apNext" (used at line~\cite[plus:nn])
is reset from its original value "\apPLUSc" to the \db apPLUSd where the
"\apnumY" is simply set to zero. The reading from input stream is finished.
This occurs when there are more Digits after decimal point in the second
operand than in the first one. If the end of input stream is reached and the
"\tmpd" macro is empty (all data from second operand was read too) then the
\db apPLUSf macro removes the rest of input stream and the first pass of the
calculation is done.
\ilabel[plus:nn] {apNext^^E}
\inext{apPLUSc}{def\nb apPLUSh}{++}
Why there is a complication about reading one parameter from input stream
but second one from the macro "\tmpd"? This is more faster than to save both
parameters to the macros and using "\apIVread" for both because the
"\apIVread" must redefine its parameter. You can examine that this
parameter is very long.
The \db apPLUSm "<data>@" macro does transmissions calculation when
subtracting. The "<data>" from first pass is expanded in the input stream.
The "\apPLUSm" macro reads repeatedly one Digit from the "<data>" until the
stop mark is reached. The Digits are in the range $-9999$ to $9999$. If the
Digit is negative then we need to add $10000$ and set the transmission value
"\apnumX" to one, else "\apnumX" is zero. When the next Digit is processed then
the calculated transmission value is subtracted. The macro "\apPLUSw" writes
the result for each Digit "\apnumA" in the normal (human readable) order.
\inext{apPLUSm}{^^B\cbrace}{++}
The \db apPLUSp "<data>@" macro does transmissions calculation when
addition is processed. It is very similar to"\apPLUSm", but Digits are in
the range $0$ to $19998$. If the Digit value is greater then $9999$ then we
need to subtract $10000$ and set the transmission value "\apnumX" to one,
else "\apnumX" is zero.
\inext{apPLUSp}{^^B\cbrace}{++}
The \db apPLUSw writes the result with one Digit (saved in "\apnumA") to the
"\OUT" macro. The "\OUT" is initialized as empty. If it is empty (it means
we are after decimal point), then we need to write all four digits by
"\apIVwrite" macro (including left zeros) but we need to remove right zeros
by "\apREMzerosR". If the decimal point is reached, then it is saved to the
"\OUT". But if the previous "\OUT" is empty (it means there are no digits
after decimal point or all such digits are zero) then "\def\OUT{\empty}"
ensures that the "\OUT" is non-empty and the ignoring of right zeros are
disabled from now.
\inext{apPLUSw}{^^B\cbrace}{++}
The macro \db apPLUSy "<expanded OUT>@" removes left trailing zeros from the
"\OUT" macro and saves the possible minus sign by the \db apPLUSz macro.
\inext{apPLUSy}{def\nb apPLUSz}{++}
The macro \db apPLUSxE uses the "\apROLLa" in order to shift the decimal
point of the operand. We need to set the same decimal exponent in scientific
notation before the addition or subtraction is processed.
\inext{apPLUSxE}{^^B\cbrace}{++}
\subsec Multiplication
Suppose the following multiplication example: "1234*567=699678".
\def\begtthook{\lccode`~=`\ \lowercase{\def~{\ }}}
\begtt
Normal format: | Mirrored format:
1 2 3 4 * | 4 3 2 1 *
5 6 7 | 7 6 5
---------------- | -----------------
*7: 7 14 21 28 | *7: 28 21 14 7
*6: 6 12 18 24 | *6: 24 18 12 6
*5: 5 10 15 20 | *5: 20 15 10 5
---------------- | -----------------
6 9 9 6 7 8 | 8 7 6 9 9 6
\endtt
This example is in numeral system of base 10 only for simplification, the
macros work really with base 10000.
Because we have to do the transmissions between Digit positions
from right to left in the normal format and because it is more natural for
\TeX{} to put the data into the input stream and
read it sequentially from left to right, we use the mirrored format in our
macros.
The macro \db apMUL prepares parameters using "\apPPab" and executes "\apMULa"
\inext{apMUL}{}{++}
The macro \db apMULa does the following:
\ilabel[mul:apE] {apE}
\ilabel[mul:sgn] {apSIGN}
\ilabel[mul:sgn0] {apSIGN=0}
\ilabel[mul:diga] {apDIG\nb tmpa}
\ilabel[mul:digb] {apnumD=}
\ilabel[mul:ba] {apIVmod}
\ilabel[mul:bb] {tmpc}
\ilabel[mul:b] {*.}
\ilabel[mul:ca] {tmpb^^E}
\ilabel[mul:cb] {apMULc}
\ilabel[mul:d] {apMULd}
\ilabel[mul:g] {apMULg}
\ilabel[mul:z] {0-}
\ilabel[mul:zz] {tmpa\nb OUT}
\begitems
\item * It gets the parameters in "\tmpa" and "\tmpb" preprocessed using
the "\apPPab" macro.
\item * It evaluates the exponent of ten "\apE" which is usable when
the scientific notation of numbers is used
(line~\cite[mul:apE]).
\item * It calculates "\apSIGN" of the result
(line~\cite[mul:sgn]).
\item * If "\apSIGN=0" then the result is zero and we will do nothing more
(line~\cite[mul:sgn0]).
\item * The decimal point is removed from the parameters by
"\apDIG<param><register>". The "\apnumD" includes the number of digits
before decimal point (after the "\apDIG" is used) and the
"<register>" includes the number of digits in the rest. The "\apnumA"
or "\apnumB" includes total number of digits in the parameters "\tmpa" or
"\tmpb" respectively. The "\apnumD" is re-calculated: it saves the number
of digits after decimal point in the result
(lines~\cite[mul:diga] to~\cite[mul:digb]).
\item *
Let $A$ is the number of total digits in the "<param>" and let
$F=A \mathrel{\rm mod} 4$, but if $F=0$ then reassign it to $F=4$. Then $F$
means the number of digits in the first Digit. This calculation
is done by "\apIVmod<A><F>" macro. All another Digits will have four digits.
The "\apMULb<param>@@@@" is able to read four digits, next four digits
etc. We need to insert appropriate number of empty parameters before the "<param>".
For example "\apMULb{}{}{}<param>@@@@" reads first only one digit from "<param>",
next four digits etc. The appropriate number of empty parameters are prepared in
the "\tmpc" macro
(lines~\cite[mul:ba] to~\cite[mul:bb]).
\item * The "\apMULb" reads the "<paramA>" (all Digits) and
prepares the "\OUT" macro in the special interleaved format
(described below). The format is finished by "*." in the line~\cite[mul:b].
\item * Analogical work is done with the second parameter "<paramB>". But this
parameter is processed by "\apMULc", which reads Digits of the parameter
and inserts them to the "\tmpa" in the reversed order
(lines~\cite[mul:ca] to~\cite[mul:cb]).
\item * The main calculation is done by "\apMULd<paramB>@", which reads Digits
from "<paramB>" (in reversed order) and does multiplication of the
"<paramA>" (saved in the "\OUT") by these Digits
(line~\cite[mul:d]).
\item * The "\apMULg" macro converts the result "\OUT" to the human
readable form
(line~\cite[mul:g]).
\item * The possible minus sign and the trailing zeros of results of the
type ".00123" is prepared by "\apADDzeros\tmpa" to the "\tmpa" macro.
This macro is appended to the result in the "\OUT" macro
(lines~\cite[mul:z] to~\cite[mul:zz]).
\enditems
\inext{apMULa}{^^B\cbrace}{++}
We need to read the two data streams when the multiplication of the "<paramA>"
by one Digit from "<paramB>" is performed and the partial sum is
actualized. First: the digits of the "<paramA>" and second: the partial sum.
We can save these streams to two macros and read one piece of information
from such macros at each step, but this si not effective because the whole
stream have to be read and redefined at each step. For \TeX{} is more
natural to put one data stream to the input queue and to read pieces of
infromation thereof. Thus we interleave both
data streams into one "\OUT" in such a way that one element of data from first
stream is followed by one element from second stream and it is followed by second
element from first stream etc. Suppose that we are at the end of $i-th$ line
of the multiplication scheme where we have the partial sums $s_n, s_{n-1},
\ldots, s_0$ and the Digits of "<paramA>" are $d_k, d_{k-1}, \ldots, d_0$.
The zero index belongs to the most right position in the mirrored format.
The data will be prepared in the form:
\begtt
. {s_n} {s_(n-1)}...{s_(k+1)} * {s_k} {d_(k-1)}...{s_1} {d_1} {s_0} {d_0} *
\endtt
%
For our example (there is a simplification: numeral system of base 10 is
used and no transmissions are processed), after second line (multiplication by 6 and
calculation of partial sums) we have in "\OUT":
\begtt
. {28} * {45} {4} {32} {3} {19} {2} {6} {1} *
\endtt
%
and we need to create the following line during calculation of next line
of multiplication scheme:
\begtt
. {28} {45} * {5*4+32} {4} {5*3+19} {3} {5*2+6} {2} {5*1} {1} *
\endtt
%
This special format of data includes two parts. After the starting dot,
there is a sequence of sums which are definitely calculated. This sequence
is ended by first "*" mark. The last definitely calculated sum follows this
mark. Then the partial sums with the Digits of "<paramA>" are interleaved
and the data are finalized by second "*". If the calculation processes the
the second part of the data then the general task is to read two data
elements (partial sum and the Digit) and to write two data elements (the new
partial sum and the previous Digit). The line calculation starts by copying
of the first part of data until the first "*" and
appending the first data element after "*". Then the "*" is written and the
middle processing described above is started.
The macro \db apMULb "<paramA>@@@@" prepares the special format of the macro
"\OUT" described above where the partial sums are zero. It means:
\begtt
* . {d_k} 0 {d_(k-1)} 0 ... 0 {d_0} *
\endtt
%
where $d_i$ are Digits of "<paramA>" in reversed order.
The first ``sum'' is only dot. It will be
moved before "*" during the first line processing.
Why there is such special ``pseudo-sum''? The "\apMULe" with the parameter
delimited by the first "*" is used in the context
"\apMULe.{<sum>}*" during the third line processing
and the dot here protects from removing the braces around the first real sum.
\inext{apMULb}{^^B\cbrace}{++}
The macro \db apMULc "<paramB>@@@@" reads Digits from "<paramB>" and saves
them in reversed order into "\tmpa". Each Digit is enclosed by \TeX{} braces
"{}".
\inext{apMULc}{}{++}
The macro \db apMULd "<paramB>@" reads the Digits from "<paramB>"
(in reversed order),
uses them as a coefficient for multiplication stored in "\tmpnumA" and
processes the "\apMULe <special data format>" for each such coefficient.
This corresponds with one line in the multiplication scheme.
\inext{apMULd}{^^B\cbrace}{++}
The macro \db apMULe "<special data format>" copies the first part of data
format to the "\OUT", copies the next element after first "*", appends "*"
and does the calculation by "\apMULf". The \db apMULf is recursively
called. It reads the Digit to "#1" and the partial sum to the "#2" and
writes "{\appnumA*#1+#2}{#1}" to the "\OUT" (lines~\cite[mul:f1] to~\cite[mul:f2]).
If we are at the end of data, then
"#2" is "*" and we write the "{\apnumA*#1}{#1}" followed by ending "*" to the
"\OUT" (lines~\cite[mul:f3] to~\cite[mul:f4]).
\ilabel[mul:f1] {2^^E}
\ilabel[mul:f2] {expandafter\nb apMULf}
\ilabel[mul:f3] {ifx*}
\ilabel[mul:f4] {fi*}
\ilabel[mul:f5] {MULf0}
\inext{apMULe}{^^B\cbrace}{++}
\noindent
There are several complications in the algorithm described above.
\begitems
\item * The result isn't saved directly to the "\OUT"
macro, but partially into the macros "\apOUT:<num>", as described in the
section~\inumref[aux] where the "\apOUTx" macro is defined.
\item * The transmissions between Digit positions are calculated.
First, the transmission value "\apnumX" is set to zero in the "\apMULe".
Then this value is subtracted from the calculated value "\apnumB" and
the new transmission is calculated using the "\apIVtrans" macro
if "\apnumB"${}\ge10000$. This macro modifies "\apnumB" in order it is right
Digit in our numeral system.
\item * If the last digit has nonzero transmission, then the calculation
isn't finished, but the new pair "{<transmission>}{0}" is added to the
"\OUT". This is done by recursively call of "\apMULf" at line~\cite[mul:f5].
\item * The another situation can be occurred: the last pair has both values
zeros. Then we needn't to write this zero to the output. This is solved by
the test "\ifnum\the\apnumB#1=0" at line~\cite[mul:f4].
\enditems
The macro \db apMULg "<special data format>@" removes the first dot
(it is the "#1" parameter) and prepares the "\OUT" to writing the result in
reverse order, i.e. in human readable form. The next work is done by
"\apMULh" and "\apMULi" macros. The \db apMULh repeatedly reads the first part of the
special data format (Digits of the result are here) until the first "*"
is found. The output is stored by
"\apMULo<digits>{<data>}" macro. If the first "*" is found then the
\db apMULi macro repeatedly reads the triple
"{<Digit of result>}{<Digit of A>}{<next Digit of result>}" and saves the first
element in full (four-digits) form by the "\apIVwrite" if the third element
isn't the stop-mark "*". Else the last Digit (first Digit in the human
readable form) is saved by "\the", because we needn't the trailing zeros
here. The third element is put back to the input stream but it is ignored by
\db apMULj macro when the process is finished.
\inext{apMULg}{def\nb apMULj}{++}
The \db apMULo "<digits>{<data>}" appends "<data>" to the "\OUT" macro.
The number of digits after decimal point "\apnumD" is decreased by the
number of actually printed digits "<digits>". If the decimal point is to be
printed into "<data>" then it is performed by the \db apMULt macro.
\inext{apMULo}{def\nb apMULt}{++}
\subsec Division
Suppose the following example:
\begtt
<paramA> : <paramB> <output>
12345:678 = [12:6=2] 2 (2->1)
2*678 -1356
-1215 <0 correction! 1
12345
1*678 -678
5565 [55:6=8] 9 (9->8)
9*678 -6102
-537 <0 correction! 8
5565
8*678 -5424
1410 [14:6=2] 2
2*678 -1356
0540 [05:6=0] 0
0*678 -0
5400 [54:6=8] 9 (2x correction: 9->8, 8->7)
...
12345:678 = 182079...
\endtt
We implement the division similar like pupils do it in the school (only the
numeral system with base 10000 instead 10 is actually used, but we keep with
base 10 in our illustrations). At each step of the operation, we get first
two Digits from the dividend or remainder (called partial dividend or
remainder) and do divide it by the first nonzero Digit of the divisor (called partial
divisor). Unfortunately, the resulted Digit cannot be the definitive value
of the result. We are able to find out this after the whole divisor is
multiplied by resulted Digit and compared with the whole remainder. We
cannot do this test immediately but only after a lot of following
operations (imagine that the remainder and divisor have a huge number of Digits).
We need to subtract the remainder by the multiple of the divisor at each
step. This means that we need to calculate the transmissions from the Digit
position to the next Digit position from right to left (in the scheme
illustrated above). Thus we need to reverse the order of Digits
in the remainder and divisor. We do this reversion only once at the
preparation state of the division and we interleave the data from the
divisor and the dividend (the dividend will be replaced by the remainder,
next by next remainder etc.).
The number of Digits of the dividend can be much greater than the number of
Digits of the divisor. We need to calculate only with the first part of
dividend/remainder in such case. We need to get only one new Digit from the
rest of dividend at each calculation step. The illustration follows:
\begtt
...used dividend.. | ... rest of dividend ... | .... divisor ....
1234567890123456789 7890123456789012345678901234 : 1231231231231231231
xxxxxxxxxxxxxxxxxx 7 <- calculated remainder
xxxxxxxxxxxxxxxxx x8 <- new calculated remainder
xxxxxxxxxxxxxxxx xx9 <- new calculated remainder etc.
\endtt
We'll interleave only the ``used dividend'' part with the divisor at the
preparation state. We'll put the ``rest of dividend'' to the input stream in
the normal order. The macros do the iteration over calculation steps and
they can read only one new Digit from this input stream if they need it.
This approach needs no manipulation with the (potentially long) ``rest of the
dividend'' at each step. If the divisor has only one Digit (or comparable small Digits)
then the algorithm has only linear complexity with respect to the number of
Digits in the dividend.
The numeral system with the base 10000 brings a little problem: we are
simply able to calculate the number of digits which are multiple of four. But
user typically wishes another number of calculated decimal digits. We cannot
simply strip the trailing digits after calculation because the user needs to
read the right remainder. This is a reason why we calculate the number of
digits for the first Digit of the result. All another
calculated Digits will have four digits. We need to prepare the first
``partial dividend'' in order to the $F$ digits will be calculated first.
How to do it? Suppose the following illustration of the first two Digits in the
``partial remainder'' and ``partial divisor'':
\begtt
0000 7777 : 1111 = 7 .. one digit in the result
0007 7778 : 1111 = 70 .. two digits in the result
0077 7788 : 1111 = 700 .. three digits in the result
0777 7888 : 1111 = 7000 .. four digits in the result
7777 8888 : 1111 = ???? .. not possible in the numeral system of base 10000
\endtt
%
We need to read $F-1$ digits to the first Digit and four digits to the
second Digit of the ``partial dividend''. But this is true only if the
dividend is ``comparably greater or equal to'' divisor. The word
``comparably greater'' means that we ignore signs and the decimal point in
compared numbers and we assume the decimal points in the front of both
numbers just before the first nonzero digit. It is obvious that if the
dividend is ``comparably less'' than divisor then we need to read $F$ digits
to the first Digit.
The macro \db apDIV runs \db apDIVa macro which
uses the "\tmpa" (dividend) and "\tmpb" (divisor)
macros and does the following work:
\ilabel[div:0] {SIGNb=0}
\ilabel[div:sgn] {SIGN=}
\ilabel[div:a0] {SIGNa=0}
\ilabel[div:ape] {apE=}
\ilabel[div:diga] {apDIG}
\ilabel[div:digb] {advance\nb apnumD}
\ilabel[div:comp] {DIVcomp}
\ilabel[div:ifap] {ifapX}
\ilabel[div:ca] {apnumC=}
\ilabel[div:cb] {ifnum\nb apnumC}
\ilabel[div:h] {advance\nb apnumH}
\ilabel[div:F] {apIVmod}
\ilabel[div:rb] {apIVread\nb tmpb}
\ilabel[div:ra] {apIVreadX}
\ilabel[div:rc] {apIVread\nb tmpa}
\ilabel[div:xa] {apDIVxA}
\ilabel[div:xb] {apDIVxB}
\ilabel[div:xout] {edef\nb XOUT}
\ilabel[div:outa] {edef\nb OUT}
\ilabel[div:outb] {apADDzeros}
\ilabel[div:z] {apnumZ=}
\ilabel[div:next] {apNext=}
\ilabel[div:add] {apnumZ=\nb apnumD}
\ilabel[div:rema] {apDIVv}
\ilabel[div:remb] {apROLLa}
\begitems
\item * If the divisor "\tmpb" is equal to zero, print error and do nothing more
(line~\cite[div:0]).
\item * The "\apSIGN" of the result is calculated
(line~\cite[div:sgn]).
\item * If the dividend "\tmpa" is equal to zero, then "\OUT" and "\XOUT" are
zeros and do nothing more
(line~\cite[div:a0]).
\item * Calculate the exponent of ten "\apE" when scientific notation is used
(Line~\cite[div:ape]).
\item * The number of digits before point are counted by "\apDIG" macro
for both parameters. The difference is saved to "\apnumD" and this is the
number of digits before decimal point in the result (the exception is
mentioned later). The "\apDIG" macro removes the decimal point and
(possible) left zeros from its parameter and saves the result to the
"\apnumD" register
(lines~\cite[div:diga] to~\cite[div:digb]).
\item * The macro "\apDIVcomp<paramA><paramB>" determines if the "<paramA>"
is ``comparably greater or equal'' to "<paramB>". The result is stored in
the boolean value "apX". We can ask to this by
the "\ifapX<true>\else<false>\fi" construction
(line~\cite[div:comp]).
\item * If the dividend is ``comparably greater or equal'' to the divisor,
then the position of decimal point in the result "\apnumD" has to be shifted by one
to the right. The same is completed with "\apnumH" where the position of
decimal point of the remainder will be stored
(line~\cite[div:ifap]).
\item * The number of desired digits in the result "\apnumC" is calculated
(lines~\cite[div:ca] to~\cite[div:cb]).
\item * If the number of desired digits is zero or less than zero then do nothing more
(line~\cite[div:cb]).
\item * Finish the calculation of the position of decimal point in the
remainder "\apnumH"
(line~\cite[div:h]).
\item * Calculate the number of digits in the first Digit "\apnumF"
(line~\cite[div:F]).
\item * Read first four digits of the divisor by the macro
"\apIVread<sequence>". Note that this macro puts trailing
zeros to the right if the data stream "<param>" is shorter than four
digits. If it is empty then the macro returns zero. The returned value is saved
in "\apnumX" and the "<sequence>" is redefined by new value of
the "<param>" where the read digits are removed
(line~\cite[div:rb]).
\item * We need to read only "\apnumF" (or "\apnumF"${}-1$) digits from the
"\tmpa". This is done by the "\apIVreadX" macro at line~\cite[div:ra].
The second Digit of the ``partial dividend'' includes four digits and it
is read by "\apIVread" macro at line~\cite[div:rc].
\item * The ``partial dividend'' is saved to the "\apDIVxA" macro and
the ``partial divisor'' is stored to the "\apDIVxB" macro. Note, that
the second Digit of the ``partial dividend'' isn't expanded by simply
"\the", because when "\apnumX=11" and "\apnumA=2222" (for example), then
we need to save "22220011". These trailing zeros from left are written by
the "\apIVwrite" macro
(lines~\cite[div:xa] to~\cite[div:xb]).
\item * The "\XOUT" macro for the currently computed remainder is
initialized. The special
interleaved data format of the remainder "\XOUT" is described below
(line~\cite[div:xout]).
\item * The "\OUT" macro is initialized.
The "\OUT" is generated as literal macro. First possible "<sign>", then digits.
If the number of effective digits before decimal point "\apnumD" is
negative, the result will be in the form ".000123" and we need to add the zeros
by the "\apADDzeros" macro
(lines~\cite[div:outa] to~\cite[div:outb]).
\item * The registers for main loop are initialized. The "\apnumE"
signalizes that the remainder of the partial step is zero and we can stop
the calculation. The "\apnumZ" will include the Digit from the input
stream where the ``rest of dividend'' will be stored
(line~\cite[div:z]).
\item * The main calculation loop is processed by the "\apDIVg" macro
(line~\cite[div:next]).
\item * If the division process stops before the position of the decimal
point in the result (because there is zero remainder, for example) then we
need to add the rest of zeros by "\apADDzeros" macro. This is actual for the
results of the type "1230000"
(line~\cite[div:add]).
\item * If the remainder isn't equal to zero, we need to extract the digits
of the remainder from the special data formal to the human readable
form. This is done by the "\apDIVv" macro. The decimal point is inserted
to the remainder by the "\apROLLa" macro
(lines~\cite[div:rema] to~\cite[div:remb]).
\enditems
\inext{apDIVa}{^^B\cbrace}{++}
The macro \db apDIVcomp "<paramA><paramB>" provides the test if the
"<paramA>" is ``comparably greater or equal'' to "<paramB>". Imagine
the following examples:
\begtt
123456789 : 123456789 = 1
123456788 : 123456789 = .99999999189999992628
\endtt
%
The example shows that the last digit in the operands can be important for
the first digit in the result. This means that we need to compare whole
operands but we can stop the comparison when the first difference in the
digits is found. This is lexicographic ordering. Because we don't assume
the existence of e\TeX{} (or another extensions), we need to do this
comparison by macros. We set the "<paramA>" and "<paramB>" to the "\tmpc" and
"\tmpd" respectively. The trailing "\apNL"s are appended.
The macro \db apDIVcompA reads first 8 digits from
first parameter and the macros \db apDIVcompB reads first 8 digits from
second parameter and does the comparison. If the numbers are equal then the
loop is processed again.
\inext{apDIVcomp}{\count=3 ^^B\cbrace}{++}
The format of interleaved data with divisor and remainder is described here.
Suppose this partial step of the division process:
\begtt
R0 R1 R2 R3 ... Rn : d1 d2 d3 ... dn = ...A...
@ -A*d1 -A*d2 -A*d3 ... -A*dn [ R0 R1 : d1 = A ]
0 N0 N1 N2 ... N(n-1) Nn
\endtt
The $R_k$ are Digits of the remainder, $d_k$ are Digits of the divisor. The
$A$ is calculated Digit in this step. The calculation of the Digits of the new
remainder is hinted here. We need to do this from right to left because of
the transmissions. This implies, that the interleaved format of "\XOUT" is
in the reverse order and looks like
\begtt
dn Rn ... d3 R3 d2 R2 d1 R1 @ R0
\endtt
%
for example for "<paramA>=1234567893", "<paramB>=454502" (in the human readable form)
the "\XOUT" should be "{200}{9300}{4545}{5678}@{1234}" (in the special format).
The Digits are separated by \TeX{} braces "{}".
The resulted digit for this step is $A=12345678/1415=2716$.
The calculation of the new remainder takes $d_k$, $R_k$, $d_{k-1}$ for each
$k$ from $n$ to $0$ and creates the Digit of the new remainder $N_{k-1} = R_k -
A\cdot d_k$ (roughly speaking, actually it calculates transmissions too) and
adds the new couple $d_{k-1}$~$N_{k-1}$ to the new version of "\XOUT" macro.
The zero for $N_{-1}$ should be reached. If it is not completed then a
correction of the type $A\mathrel{:=}A-1$ have to be done and the
calculation of this step is processed again.
The result in the new "\XOUT" should be (after one step is done):
\begtt
dn Nn ... d3 N3 d2 N2 d1 N1 @ N0
\endtt
%
where $N_n$ is taken from the ``rest of the dividend'' from the input stream.
The initialization for the main loop is done by \db apDIVg macro. It reads
the Digits from "\tmpa" (dividend) and "\tmpb" macros (using "\apIVread")
and appends them to the "\XOUT" in described data format. This initialization
is finished when the "\tmpb" is empty. If the "\tmpa" is not empty in such
case, we put it to the input stream using "\expandafter\apDIVh\tmpa"
followed by four "\apNL"s (which simply expands zero digit) followed by
stop-mark. The "\apDIVh" reads one Digit from input stream.
Else we put only the stop-mark to the input stream and run the
"\apDIVi". The "\apNexti" is set to the "\apDIVi", so the macro "\apDIVh" will
be skipped forever and no new Digit is read from input stream.
\inext{apDIVg}{^^B\cbrace}{++}
The macro \db apDIVh reads one Digit from data stream (from the rest of the
dividend) and saves it to the "\apnumZ" register. If the stop-mark is
reached (this is recognized that the last digit is the "\apNL"),
then "\apNexti" is set to "\apDIVi", so the "\apDIVh" is never
processed again.
\inext{apDIVh}{^^B\cbrace}{++}
The macro \db apDIVi contains the main loop for division calculation. The
core of this loop is the macro call "\apDIVp<data>" which adds next digit
to the "\OUT" and recalculates the remainder.
The macro "\apDIVp" decreases the "\apnumC" register (the desired digits in the
output) by four, because four digits will be calculated in the next step.
The loop is processed while "\apnumC" is positive. The "\apnumZ" (new Digit
from the input stream) is initialized as zero and the "\apNexti" runs the
next step of this loop. This step starts from "\apDIVh" (reading one digit
from input stream) or directly the "\apDIVi" is repeated. If the remainder
from the previous step is calculated as zero ("\apnumE=0"), then we stop
prematurely. The \db apDIVj macro is called at the end of the loop because
we need to remove the ``rest of the dividend'' from the input stream.
\inext{apDIVi}{def\nb apDIVj}{++}
The macro \db apDIVp "<interleaved data>@" does the basic setting
before the calculation through the
expanded "\XOUT" is processed. The \db apDIVxA includes the ``partial
dividend'' and the \db apDIVxB includes the ``partial divisor''.
We need to do
"\apDIVxA" over "\apDIVxB" in order to obtain the next digit in the output.
This digit is stored in "\apnumA".
The "\apnumX" is the transmission value, the "\apnumB", "\apnumY" will be the
memory of the last two calculated Digits in the remainder. The "\apnumE"
will include the maximum of all digits of the new remainder. If it is equal to
zero, we can finish the calculation.
The new interleaved data will be stored to the "\apOUT:<num>" macros in
similar way as in the "\apMUL" macro. This increases the speed of the
calculation. The data "\apnumO", "\apnumL" and "\apOUTl" for this purpose
are initialized.
The "\apDIVq" is started and the tokens "0\apnumZ" are appended to the input
stream (i.e to the expanded "\XOUT". This zero will be ignored and the
"\apnumZ" will be used as a new $N_n$, i.e. the Digit from the ``rest of the
dividend''.
\inext{apDIVp}{^^B\cbrace}{++}
The macro \db apDIVq $\langle d_k\rangle\,\langle R_k\rangle\,\langle d_{k-1}\rangle$
calculates the Digit of the new remainder $N_{k-1}$
by the formula $N_{k-1} = - A\cdot d_k +R_k - X$ where $X$ is the
transmission from the previous Digit. If the result is negative, we need to
add minimal number of the form $X\cdot 10000$ in order the result is
non-negative. Then the $X$ is new transmission value.
The digit $N_k$ is stored in the "\apnumB"
register and then it is added to "\apOUT:<num>" in the order $d_{k-1}\, N_{k-1}$.
The "\apnumY" remembers the value of the previous "\apnumB".
The $d_{k-1}$ is put to the input stream back in order it would be read by
the next "\apDIVq" call.
If $d_{k-1}={}$"@" then we are at the end of the remainder calculation and
the "\apDIVr" is invoked.
\inext{apDIVq}{^^B\cbrace}{++}
The \db apDIVr macro does the final work after the calculation of new
remainder is done. It tests if the remainder is OK, i.e.\ the transmission
from the $R_1$ calculation is equal to $R_0$. If it is true then new Digit
"\apnumA" is added to the "\OUT" macro else the "\apnumA" is decreased (the
correction) and the calculation of the remainder is run again.
If the calculated Digit and the remainder are OK, then we do following:
\begitems
\item * The new "\XOUT" is created from "\apOUT:<num>" macros using
"\apOUTs" macro.
\item * The "\apnumA" is saved to the "\OUT". This is done with care.
If the "\apnumD" (where the decimal point is measured from the actual
point in the "\OUT") is in the interval $[0, 4)$ then the decimal point
have to be inserted between digits into the next Digit. This is done by
"\apDIVt" macro. If the remainder is zero ("\apnumE=0"), then the right
trailing zeros are removed from the Digit by the "\apDIVu" and the shift
of the "\apnumD" register is calculated from the actual digits.
All this calculation is done in "\tmpa" macro. The last step is adding
the contents of "\tmpa" to the "\OUT".
\item * The "\apnumD" is increased by the number of added digits.
\item * The new ``partial dividend'' is created from "\apnumB" and
"\apnumY".
\enditems
\inext{apDIVr}{^^B\cbrace}{++}
The \db apDIVt macro inserts the dot into digits quartet (less than four
digits are allowed too) by the "\apnumD" value. This value is assumed in the
interval $[0, 4)$. The expandable macro "\apIVdot<shift><data>" is used for
this purpose. The result from this macro has to be expanded twice.
\inext{apDIVt}{}{++}
The \db apDIVu macro removes trailing zeros from the right and removes the
dot, if it is the last token of the "\tmpa" after removing zeros. It uses
expandable macros "\apREMzerosR<data>" and "\apREMdotR<data>".
\inext{apDIVu}{}{++}
The rest of the code concerned with the division does an extraction of the
last remainder from the data and this value is saved to the \db XOUT macro in
human readable form. The \db apDIVv macro is called repeatedly on the
special format of the "\XOUT" macro and the new "\XOUT" is created.
The trailing zeros from right are ignored by the \db apDIVw.
\inext{apDIVv}{\empty}{+-}
\subsec Power to the Integer
The \db apPOW macro does the power to the integer exponent only. The \db
apPOWx is equivalent to "\apPOW" and it is used in "\evaldef" macro for the "^"
operator. If you want to redefine the meaning of the "^" operator then
redefine the "\apPOWx" sequence.
\inext{apPOWx}{}{++}
We can implement the power to the integer as repeated multiplications. This
is simple but slow. The goal of this section is to present the power to the
integer with some optimizations.
Let $a$ is the base of the powering computation and $d_1, d_2, d_3, \dots,
d_n$ are binary digits of the exponent (in reverse order). Then
$$
p = a^{1\,{d_1}+2\,{d_2}+2^2\,{d_3}+\cdots+2^{n-1}\,{d_n}} =
(a^1)^{d_1}\cdot(a^2)^{d_2}\cdot(a^{2^2})^{d_3}\cdot (a^{2^{n-1}})^{d_n}.
$$
If $d_i=0$ then $z^{d_i}$ is one and this can be omitted from the queue of
multiplications. If $d_i=1$ then we keep $z^{d_i}$ as $z$ in the queue.
We can see from this that the $p$ can be computed by the
following algorithm:
\begtt
(* "a" is initialized as the base, "e" as the exponent *)
p := 1;
while (e>0) {
if (e%2) p := p*a;
e := e/2;
if (e>0) a := a*a;
}
(* "p" includes the result *)
\endtt
The macro \db apPOWa does the following work.
\ilabel[pow:sgna] {SIGNa=}
\ilabel[pow:sgnb] {SIGNb=}
\ilabel[pow:ea] {non-integer}
\ilabel[pow:eb] {output?}
\ilabel[pow:ape] {apE=\nb apEa}
\ilabel[pow:odd] {ifodd}
\ilabel[pow:dig] {apDIG\nb tmpa}
\ilabel[pow:aa] {apIVmod}
\ilabel[pow:ab] {ifcase}
\ilabel[pow:ba] {apMULb}
\ilabel[pow:bb] {*.\nb OUT}
\ilabel[pow:p] {1*}
\ilabel[pow:rev] {apREV}
\ilabel[pow:b] {apPOWb}
\ilabel[pow:g] {apPOWg}
\ilabel[pow:outa] {-\nb OUT}
\ilabel[pow:outb] {apROLLa}
\ilabel[pow:r] {apDIVa}
\begitems
\item * After using "\apPPab" the base parameter is saved in "\tmpa" and the
exponent is saved in "\tmpb".
\item * In trivial cases, the result is set without any computing
(lines~\cite[pow:sgna] and~\cite[pow:sgnb]).
\item * If the exponent is non-integer or it is too big then the error
message is printed and the rest of the macro is skipped by the "\apPOWe"
macro
(lines~\cite[pow:ea] to~\cite[pow:eb]).
\item * The "\apE" is calculated from "\apEa" (line~\cite[pow:ape]).
\item * The sign of the result is negative only if the "\tmpb" is odd and
base is negative
(line~\cite[pow:odd]).
\item * The number of digits after decimal point for the result
is calculated and saved to "\apnumD". The total number of digits of the
base is saved to "\apnumC".
(line~\cite[pow:dig]).
\item * The first Digit of the base needn't to include all four digits, but
other Digits do it. The similar trick as in "\apMULa" is used here
(lines~\cite[pow:aa] to~\cite[pow:ab]).
\item * The base is saved in interleaved reversed format (like in
"\apMULa") into the "\OUT" macro by the "\apMULb" macro. Let it be the
$a$ value from our algorithm described above
(lines~\cite[pow:ba] and~\cite[pow:bb]).
\item * The initial value of $p=1$ from our algorithm is set in interleaved
format into "\tmpc" macro
(line~\cite[pow:p]).
\item * The main loop described above is processed by "\apPOWb" macro.
(line~\cite[pow:b]).
\item * The result in "\tmpc" is converted into human readable form by the
"\apPOWg" macro and it is stored into the "\OUT" macro
(line~\cite[pow:g]).
\item * If the result is negative or decimal point is needed to print then
use simple conversion of the "\OUT" macro (adding minus sign) or using
"\apROLLa" macro
(lines~\cite[pow:outa] and~\cite[pow:outb]).
\item * If the exponent is negative then do the $1/r$ calculation,
where $r$ is previous result
(line~\cite[pow:r]).
\enditems
\inext{apPOWa}{^^B\cbrace}{++}
The macro \db apPOWb is the body of the loop in the algorithm described above.
The code part after "\ifodd\apnumE" does "p := p*a". In order to do this, we need
to convert "\OUT" (where "a" is stored) into normal format using "\apPOWd".
The result is saved in "\tmpb". Then the multiplication is done by "\apMULd"
and the result is normalized by the "\apPOWn" macro. Because "\apMULd"
works with "\OUT" macro, we temporary set "\tmpc" to "\OUT".
The code part after "\ifnum\apnumE"{\tt\char`<}"0" does "a := a*a" using the "\apPOWt"
macro. The result is normalized by the "\apPOWn" macro.
\inext{apPOWb}{^^B\cbrace}{++}
The macro \db apPOWd "<initialized interleaved reversed format>" extracts the
Digits from its argument and saves them to the "\tmpb" macro.
\inext{apPOWd}{^^B\cbrace}{++}
The \db apPOWe macro skips the rest of the body of the "\apPOWa" macro to the
"\relax". It is used when "\errmessage" is printed.
\inext{apPOWe}{}{++}
The \db apPOWg macro provides the conversion from interleaved reversed format to
the human readable form and save the result to the "\OUT" macro.
It ignores the first two elements from the format and runs \db apPOWh.
\inext{apPOWg}{^^B\cbrace}{++}
The normalization to the initialized interleaved format of the "\OUT" is done
by the \db apPOWn "<data>@" macro. The \db apPOWna reads the first part of
the "<data>" (to the first "*", where the Digits are non-interleaved. The
\db apPOWnn reads the second part of "<data>" where the Digits of the result
are interleaved with the digits of the old coefficients. We need to set the
result as a new coefficients and prepare zeros between them for the new
calculation. The dot after the first "*" is not printed (the zero is printed
instead it) but it does not matter because this token is simply ignored
during the calculation.
\inext{apPOWn}{def\nb apPOWnn}{++}
The powering to two ("\OUT:=\OUT^2") is provided by the \db apPOWt "<data>"
macro. The macro \db apPOWu is called repeatedly for each "\apnumA="Digit from the
"<data>". One line of the multiplication scheme is processed by the \db apPOWv "<data>"
macro. We can call the "\apMULe" macro here but we don't do it because a slight
optimization is used here. You can try to multiply the number with digits "abcd" by itself
in the mirrored multiplication scheme. You'll see that first line includes
"a^2 2ab 2ac 2ad", second line is intended by two columns and includes "b^2 2bc 2bd",
next line is indented by next two columns and includes "c^2 2cd" and the last line is
intended by next two columns and includes only "d^2". Such calculation is slightly
shorter than normal multiplication and it is implemented in the "\apPOWv" macro.
\inext{apPOWt}{\empty}{+-}
\subsec apROLL, apROUND and apNORM Macros
The macros \db apROLL , \db apROUND and \db apNORM are implemented by
"\apROLLa", "\apROUNDa" and "\apNORMa" macros with common format of the
parameter text: "<expanded sequence>.@<sequence>" where "<expanded
sequence>" is the expansion of the macro "<sequence>" (given as first
parameter of "\apROLL", "\apROUND" and "\apNORM", but without optionally
minus sign. If there was the minus sign then "\apnumG=-1" else "\apnumG=1".
This preparation of the parameter "<sequence>" is done by the "\apPPs"
macro. The second parameter of the macros "\apROLL", "\apROUND" and
"\apNORM" is saved to the "\tmpc" macro.
\db apROLLa "<param>.@<sequence>" shifts the decimal point of the
"<param>" by "\tmpc" positions to the right (or to the left, if
"\tmpc" is negative) and saves the result to the "<sequence>" macro.
The "\tmpc" value is saved to the "\apnumA" register and the "\apROLLc" is
executed if we need to shift the decimal point to left. Else "\apROLLg" is
executed.
\inext{ROLLa}{ROLLc}{++}
The \db apROLLc "<param>.@<sequence>" shifts the decimal point to left
by the "-\apnumA" decimal digits.
It reads the tokens from the input stream until the dot is found
using \db apROLLd macro.
The number of such tokens is set to the "\apnumB" register and tokens
are saved to the "\tmpc" macro.
If the dot is found then \db apROLLe does the following: if the number of
read tokens is greater then the absolute value of the "<shift>", then
the number of positions from the most left digit of the number to the
desired place of the dot is set to the "\apnumA" register a the dot is
saved to this place by "\apROLLi<parameter>.@<sequence>". Else the new number looks
like ".000123" and the right number of zeros are saved to the "<sequence>" using
the "\apADDzeros" macro and the rest of the input stream (including expanded
"\tmpc" returned back) is appended to the macro "<sequence>" by the
\db apROLLf "<param>.@" macro.
\inext{apROLLc}{def\nb apROLLf}{++}
The \db apROLLg "<param>.@<sequence>" shifts the decimal point to the right by
"\apnumA" digits starting from actual position of the input stream.
It reads tokens from the input stream by the \db apROLLh and saves them to the
"\tmpd" macro where the result will be built.
When dot is found the \db apROLLi is processed. It reads next
tokens and decreases the "\apnumA" by one for each token. It ends (using
"\apROLLj\apROLLk") when "\apnumA" is equal to zero.
If the end of the input stream is reached (the "@" character) then the zero
is inserted before this character (using "\apROLLj\apROLLi0@"). This
solves the situations like "123",~"<shift>=2",~$\to$~"12300".
\ilabel[rol:g1] {0-}
\inext{apROLLg}{^^B\cbrace}{++}
The "\apROLLg" macro initializes "\apnumB=1" if the "<param>" doesn't begin
by dot. This is a flag that all digits read by "\apROLLi" have to be saved.
If the dot begins, then the number can look like ".000123" (before moving
the dot to the right) and we need to
ignore the trailing zeros. The "\apnumB" is equal to zero
in such case and this is set to "1" if here is first non-zero digit.
The \db apROLLj macro closes the conditionals and runs its parameter
separated by "\fi". It skips the rest of the "\apROLLi" macro too.
\inext{apROLLj}{}{++}
The macro \db apROLLk puts the decimal point to the "\tmpd" at current
position (using "\apROLLn") if the input stream is not fully read. Else it
ends the processing. The result is an integer without decimal digit in such
case.
\inext{apROLLk}{^^B\cbrace}{++}
The macro \db apROLLn reads the input stream until the dot is found.
Because we read now the digits after a
new position of the decimal point we need to check situations of the type
"123.000" which is needed to be written as "123" without decimal point.
This is a reason of a little complication. We save all digits to the "\tmpc"
macro and calculate the sum of such digits in "\apnumB" register. If this
sum is equal to zero then we don't append the ".\tmpc" to the "\tmpd".
The macro "\apROLLn" is finished by the \db apROLLo "@<sequence>" macro, which removes
the last token from the input stream and defines "<sequence>" as "\tmpd".
\inext{apROLLn}{def\nb apROLLo}{++}
The macro \db apROUNDa "<param>.@<sequence>" rounds the number given in the
"<param>". The number of digits after decimal point "\tmpc" is saved to
"\apnumD". If this number is negative then "\apROUNDe" is processed
else the \db apROUNDb reads the "<param>" to the decimal point and saves
this part to the "\tmpc" macro. The "\tmpd" macro (where the rest after
decimal point of the number will be stored) is initialized to empty and the
\db apROUNDc is started. This macro reads one token
from input stream repeatedly until the number of read tokens is equal to
"\apnumD" or the stop mark "@" is reached. All tokens are saved to "\tmpd".
Then the \db apROUNDd macro reads the rest of the "<param>", saves it to the
"\XOUT" macro and defines "<sequence>" (i.e.\ "#2") as the rounded number.
\inext{apROUNDa}{\count=3 ^^B\cbrace}{++}
The macro \db apROUNDe solves the ``less standard'' problem when rounding to
the negative digits after decimal point "\apnumD", i.e.\ we need to set
"-\apnumD" digits before decimal point to zero. The solution is to remove
the rest of the input stream, use "\apROLLa" to shift the decimal point left
by "-\apnumD" positions, use "\apROUNDa" to remove all digits after decimal
point and shift the decimal point back to its previous place.
\inext{apROUNDe}{^^B\cbrace}{++}
The macro \db apNORMa redefines the "<sequence>" in order to remove minus
sign because the "\apDIG" macro uses its parameter without this sign. Then
the \db apNORMb "<sequence><parameter>@" is executed where the dot in the
front of the parameter is tested. If the dot is here then the "\apDIG" macro measures
the digits after decimal point too and the \db apNORMc is executed (where
the "\apROLLa" shifts the decimal point from the right edge of the number).
Else the "\apDIG" macro doesn't measure the digits after decimal point and the
\db apNORMd is executed (where the "\apROLLa" shifts the decimal point from
the left edge of the number).
\inext{apNORMa}{\count=3 ^^B\cbrace}{++}
The macro \db apEadd "<sequence>" adds "E" in scientific format into "<sequence>" macro
and \db apEnum "<sequence>" normalizes the number in the "<sequence>". After
processing these macros the "\apE" register is set to zero.
\inext{apEadd}{\empty}{+-}
\subsec [aux] Miscelaneous Macros
The macro \db apEND closes the "\begingroup" group, but keeps the values of
"\OUT" macro and "\apSIGN", "\apE" registers.
\inext{apEND}{^^B\cbrace}{++}
The macro \db apDIG "<sequence><register or relax>" reads the content of
the macro "<sequence>" and counts the number of digits in this macro before
decimal point and saves it to "\apnumD" register. If the macro "<sequence>"
includes decimal point then it is redefined with the same content but without
decimal point. The numbers in the form ".00123" are replaced by "123"
without zeros, but "\apnumD=-2" in this example.
If the second parameter of the "\apDIG" macro is
"\relax" then the number of digits after decimal point isn't counted. Else
the number of these digits is stored to the given "<register>".
The macro "\apDIG" is developed in order to do minimal operations over a
potentially long parameters. It assumes that "<sequence>" includes a number
without "<sign>" and without left trailing zeros. This is true after
parameter preparation by the "\apPPab" macro.
The macro "\apDIG" prepares an incrementation in "\tmpc" if the second
parameter "<register>" isn't "\relax". It initializes "\apnumD" and "<register>".
It runs \db apDIGa "<data>..@<sequence>" which increments the "\apnumD" until
the dot is found. Then the "\apDIGb" is executed (if there are no digits
before dot) or the "\apDIGc" is called (if there is at least one digit
before dot). The \db apDIGb ignores zeros immediately after dot. The \db
apDIGc reads the rest of the "<data>" to the "#1" and saves it to the
"\tmpd" macro. It runs the counter over this "<data>" \db apDIGd "<data>@"
only if it is desired ("\tmpc" is non-empty). Else the "\apDIGe" is
executed. The \db apDIGe "<dot or nothing>@<sequence>" redefines
"<sequence>" if it is needed. Note, that "#1" is empty if and only if the
"<data>" include no dot (first dot was reached as the first dot from
"\apDIG", the second dot from "\apDIG" was a separator of "#1" in "\apDIGc"
and there is nothing between the second dot and the "@" mark. The
"<sequence>" isn't redefined if it doesn't include a dot. Else the sequence
is set to the "\tmpd" (the rest after dot) if there are no digits before dot.
Else the sequence is redefined using expandable macro \db apDIGf.
\inext{apDIG}{def\nb apDIGf}{++}
The macro \db apIVread "<sequence>" reads four digits from
the macro "<sequence>", sets "\apnumX" as the Digit consisting from read digits and
removes the read digits from "<sequence>".
It internally expands "<sequence>", adds the "\apNL" marks and runs
\db apIVreadA macro which sets the "\apnumX" and redefines "<sequence>".
The usage of the \db apNL as a stop-marks has the advantage: they act as
simply zero digits in the comparison but we can ask by "\ifx" if this
stop mark is reached. The "#5" parameter of "\apIVreadA" is separated by
first occurrence of "\apNL", i.e.\ the rest of the macro "<sequence>" is
here.
\inext{apNL}{def\nb apIVreadA}{++}
The macro \db apIVreadX "<num><sequence>" acts similar as
"\apIVread<sequence>", but only "<num>" digits are read. The "<num>" is
expected in the range 0 to 4. The macro prepares the appropriate number of empty
parameters in "\tmpc" and runs "\apIVreadA" with these empty parameters
inserted before the real body of the "<sequence>".
\inext{apIVreadX}{^^B\cbrace}{++}
The macro \db apIVwrite "<num>" expands the digits from "<num>" register.
The number of digits are four. If the "<num>" is less than "1000" then left
zeros are added.
\inext{apIVwrite}{}{++}
The macro \db apIVtrans calculates the transmission for the next Digit.
The value (greater or equal 10000) is assumed to be in "\apnumB". The new
value less than 10000 is stored to "\apnumB" and the transmission value is
stored in "\apnumX". The constant \db apIVbase is used instead of literal 10000
because it is quicker.
\inext{apIVbase}{^^B\cbrace}{++}
The macro \db apIVmod "<length><register>" sets "<register>" to the number
of digits to be read to the first Digit, if the number has "<length>" digits
in total. We need to read all Digits with four digits, only first Digit can
be shorter.
\inext{apIVmod}{}{++}
The macro \db apIVdot "<num><param>" adds the dot into "<param>". Let
$K={}$"<num>" and $F$ is the number of digits in the "<param>".
The macro expects that $K\in[0,4)$ and $F\in(0,4]$.
The macro inserts the dot after $K$-th digit
if $K<F$. Else no dot is inserted.
It is expandable macro, but two full expansions are needed.
After first expansion the result looks like
"\apIVdotA<dots><param>....@" where
"<dots>" are the appropriate number of dots. Then the \db apIVdotA reads
the four tokens (maybe the generated dots), ignores the dots while printing
and appends the dot after these four tokens, if the rest "#5" is non-empty.
\inext{apIVdot}{^^B\cbrace}{++}
The expandable macro \db apNUMdigits "{<param>}" expands (using the \db
apNUMdigitsA macro) to the number of
digits in the "<param>". We assume that maximal number of digits will be
four.
\inext{apNUMdigits}{\empty}{+-}
The macro \db apADDzeros "<sequence>" adds "\apnumZ" zeros to the macro
"<sequence>".
\inext{apADDzeros}{^^B\cbrace}{++}
The expandable macro \db apREMzerosR "{<param>}" removes right trailing zeros from the
"<param>". It expands to "\apREMzerosRa<param>@0@!". The
macro \db apREMzerosRa reads all text terminated by "0@" to "#1". This
termination zero can be the most right zero of the "<param>" (then "#2" is
non-empty) or "<param>" hasn't such zero digit (then "#2" is empty). If "#2"
is non-empty then the "\apREMzerosRa" is expanded again in the recursion.
Else \db apREMzerosRb removes the stop-mark "@" and the expansion is
finished.
\inext{apREMzerosR}{def\nb apREMzerosRb}{++}
The expandable macro \db apREMdotR "{<param>}" removes right trailing dot from
the "<param>" if exists. It expands to \db apREMdotRa and works similarly as
the "\apREMzerosR" macro.
\inext{apREMdotR}{def\nb apREMdotRa}{++}
The \db apREMfirst "<sequence>" macro removes the first token from the
"<sequence>" macro. It can be used for removing the ``minus'' sign from the
``number-like'' macros.
\inext{apREMfirst}{\empty}{+-}
The writing to the "\OUT" in the "\apMUL", "\apDIV" and "\apPOW" macros is optimized, which
decreases the computation time with very large numbers ten times and more.
We can do simply "\edef\OUT{\OUT<something>}" instead of
\begtt
\expandafter\edef\csname apOUT:\apOUTn\endcsname
{\csname apOUT:\apOUTn\endcsname<something>}%
\endtt
%
but "\edef\OUT{\OUT<something>}" is typically processed very often
over possibly very long macro (many thousands of tokens).
It is better to do "\edef" over more short macros "\apOUT:0", "\apOUT:1", etc.
Each such macro includes only 7 Digits pairs of the whole "\OUT".
The macro \db apOUTx is invoked each 7 digit (the "\apnumO" register is
decreased).
It uses "\apnumL" value which is the "<num>" part of the next "\apOUT:<num>"
control sequence. The "\apOUTx" defines this "<num>" as \db apOUTn and
initializes "\apOUT:<num>" as
empty and adds the "<num>" to the list \db apOUTl.
When the creating of the next "\OUT" macro is definitely finished, the "\OUT" macro is
assembled from the parts "\apOUT:0", "\apOUT:1" etc.\ by the macro
\db apOUTs \unskip~"<list of numbers><dot><comma>".
\inext{apOUTx}{def\nb apOUTs}{++}
If a ``function-like'' macro needs a local counters then it is recommended
to enclose all calculation into a group "\apINIT" ... "\apEND". The
\db apINIT opens the group and prepares a short name "\do" and the macro
"\localcounts<counters>;". The typical usage is:
\begtt
\def\MACRO#1{\relax \apINIT % function-like macro, \apINIT
\evaldef\foo{#1}% % preparing the parameter
\localounts \N \M \K ;% % local \newcount\N \newcount\M \newcount\K
... % calculation
\apEND % end of \apINIT group
}
\endtt
Note that "\localcounts" is used after preparing the parameter using "\evaldef"
in odrer to avoid name conflict of local declared ``variables'' and ``variables''
used in "#1" by user.
The "\apINIT" sets locally \db localcounts to be equivalent to \db apCOUNTS.
This macro increases the top index of allocated counters "\count10"
(used in plain \TeX) locally
and declares the counters locally. It means that if the group is closed then
the counters are deallocated and top index of counters "\count10" is returned
to its original value.
\inext{apINIT}{^^B\cbrace}{++}
The macro \db do "<sequence>=<calculation>;" allows to write
the calculation of Polish expressions more synoptic:
\begtt
\do \X=\apPLUS{2}{\the\N};% % is equivalent to:
\apPLUS{2}{\the\N}\let\X=\OUT
\endtt
The "\do" macro is locally set to be equivalent to \db apEVALxdo .
\inext{apEVALxdo}{\empty}{+-}
The \db apRETURN macro must be followed by "\fi". It skips the rest of the
block "\apINIT"..."\apEND" typically used in ``function-like'' macros.
The \db apERR "{<text>}" macro writes "<text>" as error message and returns
the processing of the block enclosed by "\apINIT"..."\apEND". User can
redefine it if the "\errmessage" isn't required.
\inext{apRETURN}{\empty}{+-}
The \db apNOPT macro removes the "pt" letters after expansion of "<dimen>"
register. This is usable when we do a classical "<dimen>" calculation, see
TBN page~80. Usage: "\expandafter\apNOPT\the<dimen>".
\inext{apNOPT}{\empty}{+-}
The \db loop macro from plain \TeX{} is redefined here in more convenient
way. It does the same as original "\loop" by D. Knuth but moreover, it
allows the construction "\if...\else...\repeat".
\inext{loop}{\empty}{+-}
\subsec [fce] Function-like Macros
The implementation of function-like macros \db ABS , \db SGN , \db iDIV ,
\db iMOD , \db iROUND , \db iFRAC are simple.
\inext{ABS}{\empty}{+-}
The \db FAC macro for {\bf factorial} doesn't use recursive call because the
\TeX{} group is opened in such case and the number of levels of \TeX{} group
is limited (to 255 in my computer). But we want to calculate more factorial
than only 255!.
\inext{FAC}{^^B\cbrace}{++}
The \db BINOM "{"$a$"}{"$b$"}" is {\bf binomial coefficient} defined by
$$
{a \choose b} = {a! \over b!\,(a-b)! } = {a\,(a-1)\cdots(a-b+1) \over b!}
\quad \hbox{for integer } b > 0, \quad {a \choose 0} = 1.
$$
We use the formula where $(a-b)!$ is missing in numerator and denominator
(second fraction) because of time optimization. Second advantage of such formula
is that $a$ need not to be integer.
That is the reason why the "\BINOM" isn't defined simply as
\begtt
\def\BINOM#1#2{\relax \evaldef{ \FAC{#1} / (\FAC{#2} * \FAC{(#1)-(#2)} }}
\endtt
The macro "\BINOM" checks if $a$ is integer. If it is true then we choose
"\C" as minimum of $b$ and $a-b$. Then we calculate factorial of "\C" in the
denominator of the formula (second fraction). And nominator includes "\C"
factors. If $a$ is non-negative integer and $a<b$ then the result is zero
because one zero occurs between the factors in the nominator. Thus we give the
result zero and we skip the rest of calculation. If $a$ is non-integer, then
"\C" must be $b$. The "\step" macro (it generates the factors in the
nominator) is prepared in two versions: for $a$ integer we use
"\advance\A by-1" which is much faster than "\apPLUS\paramA{-1}" used for $a$
non-integer.
\inext{BINOM}{^^B\cbrace}{++}
{\bf The square root} is computed in the macro \db SQRT "{"$a$"}" using Newton's
approximation method. This method solves the equation $f(x)=0$ (in this case
$x^2-a = 0$) by following way. Guess the initial value of the result $x_0$.
Create tangent to the graph of $f$ in the point $[x_0, f(x_0)]$ using the
knowledge about $f'(x_0)$ value.
The intersection of this line with the axis $x$ is the new approximation of
the result~$x_1$. Do the same with $x_1$ and find $x_2$, etc. If you apply
the general Newton method to the problem $x^2-a=0$ then you get the formula
$$
\hbox{choose } x_0 \hbox{ as an initial guess,}\quad \hbox{iterate: }
x_{n+1} = {1\over2}\left(x_n + {a\over x_n}\right)
$$
If $|x_{n+1}-x_n|$ is sufficiently small we stop the processing. In
practice, we stop the processing, if the "\OUT" representation of $x_{n+1}$
rounded to the "\apFRAC" is the same as the previous representation of
$x_n$, i.e "\ifx\Xn\OUT" in \TeX{} language. Amazingly, we need only about
four iterations for 20-digits precision and about seven iterations for 50-digits
precision, if the initial guess is good chosen.
The rest of the work in the "\SQRT" macro is about the right choose of the
initial guess (using "\apSQRTr" macro) and about shifting the decimal point
in order to set the $a$ value into the interval $[1,100)$. The decimal point
is shifted by "-\M" value. After calculation is done, the decimal point is
shifted by "\M/2" value back. If user know good initial value then he/she
can set it to \db apSQRTxo macro. The calcualtion of initial value $x_0$ is
skipped in such case.
\inext{SQRT}{^^B\cbrace}{++}
Note that if the input $a<1$, then we start the Newton's method with $b$. It
is the value $a$ with shifted decimal point, $b\in[1,100)$. On the other
hand, if $a\ge1$ then we start the Newton's method directly with $a$,
because the second derivative $(x^2)''$ is constant so the speed of Newton's
method is independent on the value of $x$. And we need to calculate the
"\apFRAC" digits after the decimal point.
The macro \db apSQRTr "<number>" excepts "<number>" in the interval
$[1,100]$ and makes a roughly estimation of square root of the "<number>" in
the "\OUT" macro. It uses only classical "<dimen>" calculation, it doesn't
use any "apnum.tex" operations. The result is based on the linear approximation
of the function $g(x)=\sqrt{x}$ with known exact points $[1,1], [4,2],
[9,3], \ldots, [100,10]$. Note, that the differences between $x_i$ values of
exact points are $3,5,7,\ldots,19$. The inverted values of these differences
are pre-calculated and inserted after "\apSQRTra" macro call.
The \db apSQRTra macro operates repeatedly for $i=1,\ldots,10$ until
"\dimen0"${}=x\string<x_i$.
Then the \db apSQRTrb is executed. We are in the situation
"\dimen0"${}= x \in[x_{i-1},x_i),\quad g(x_i) = i,\quad
g(x_{i-1}) = i-1$ and the calculation of
"\OUT"${}= g(x_{i-1}) + (x-x_{i-1})/(x_i-x_{i-1})$ is performed.
If $x\in[1,4)$ then the linear approximation is worse. So, we calculate additional
linear correction in "\dimen1" using the pre-calculated value
$\sqrt 2 - 1.33333 \mathrel{\mathop{=}\limits^.} 0.08088$ here.
\inext{apSQRTr}{\count=2 ^^B\cbrace}{++}
{\bf The exponential function} $e^x$ is implemented in the \db EXP macro using
Taylor series at zero point:
$$
e^x = 1 + {x\over 1!} + {x^2\over 2!} + {x^3\over 3!} + \cdots
$$
If $x\in(0,1)$ then this series converges relatively quickly.
The macro "\EXP" takes its argument. If it is negative, remember this fact,
remove minus sign and do "\OUT=1/\OUT" in final step. Now, the argument is
positive always. If the argument is greater than~1, do argument = argument/2
and increase $K$ register. Do this step in the loop until argument${}<1$.
After "\OUT" is calculated then we do "\OUT=\OUT"$^2$ in the loop $K$ times,
because $e^{2x} = (e^x)^2$.
Now we are ready to calculate the exponential of positive argument which is
less than 1. This is done using loop of Taylor series. "\S" is total sum,
"\Sn" is the new addition in the $n$-th step. If "\Sn" is zero (in
accordance to the "\apFRAC" register) then we stop the calculation.
\inext{EXP}{^^B\cbrace}{++}
The macro \db apTAYLOR is ready for general usage in the form:
\begtt
\def\S{...}\def\Sn{...}\N=... % setting initial values for N=0
\loop
... % auxiliary calculation
\do\Sn=\apDIV{...}{...};% % calculation of new addition \Sn
% (division must be the last activity)
\apTAYLOR \iftrue \repeat % does S = S + Sn and finishes if Sn = 0
\endtt
\inext{apTAYLOR}{\empty}{+-}
{\bf The logarithm function} $\ln x$ (inverse to $e^x$) is implemented in
\db LN macro by Taylor series in the point zero of the $\arg\tanh$ function:
$$
\ln x = 2\,\arg\tanh {x-1\over x+1} =
2\left({x-1\over x+1} + {1\over3}\left(x-1\over x+1\right)^3
+ {1\over5}\left(x-1\over x+1\right)^5 + \cdots \right).
$$
This series converges quickly when x is approximately equal to one. The idea
of the macro "\LN" includes the following steps:
\begitems
\item * Whole calculation is in the group "\apINIT"..."\apEND". Enlarge the
"\apFRAC" numeric precision by three digits in this group.
\item * Read the argument "\X" using "\evaldef".
\item * If the argument is non positive, print error and skip the next
processing.
\item * If the argument is in the interval $(0,1)$, set new argument as
1/argument and remember the ``minus'' sign for the calculated
"\OUT", else the "\OUT" remains to be positive. This uses the identity
$\ln(1/x) = -\ln x$.
\item * shift the decimal point of the argument by $M$ positions left in order to
the new argument is in the interval $[1,10)$.
\item * Let $x\in[1,10)$ be the argument calculated as mentioned before.
Calculate roughly estimated $\widetilde{\ln x}$ using "\apLNr"
macro. This macro uses linear interpolation of the function $\ln x$
in eleven points in the interval $[1,10]$.
\item * Calculate $A = x / \exp(\widetilde{\ln x})$. The result is
approximately equal to one, because $\exp(\ln x) = x$.
\item * Calculate $\ln A$ using the Taylor series above.
\item * The result of $\ln x$ is equal to $\ln A + \widetilde{\ln x}$,
because $x = A\cdot \exp(\widetilde{\ln x})$ and $\ln (ab) = \ln a + \ln b$.
\item * The real argument is in the form $x\cdot 10^M$, so "\OUT"
is equal to $\ln x + M\cdot \ln(10)$ because $\ln (ab) = \ln a + \ln b$
and $\ln(10^M) = M\ln(10)$. The $\ln(10)$ value with desired
precision is calculated by "\apLNtenexec" macro. This macro saves its
result globally when firstly calculated and use the calculated result when
the "\apLNtenexec" is called again.
\item * Round the "\OUT" to the "\apFRAC" digits.
\item * Append ``minus'' to the "\OUT" if the input argument was in the
interval $(0,1)$.
\enditems
\inext{LN}{^^B\cbrace}{++}
The macro \db apLNtaylor calculates $\ln A$ for $A\approx 1$ using Taylor
series mentioned above.
\inext{apLNtaylor}{^^B\cbrace}{++}
The macro \db apLNr finds an estimation $\widetilde{\ln x}$ for $x\in [1,10)$
using linear approximation of $\ln x$ function.
Only direct "<dimen>" and "<count>" calculation with \TeX{} registers is used,
no long numbers "apnum.tex" calculation. The $\ln x_i$ is pre-calculated for
$x_i = i$, $i\in\{1,2,3,4,5,6,7,8,9,10\}$ and the values are inserted after the
"\apLNra" macro call. The input value $x$ is set as "\dimen0".
The \db apLNra "{<valueA>}{<valueB>}" macro reads the pre-calculated values
repeatedly in the loop. The
loop ends if "\apnumC" (i.~e.~$x_i$) is greater than $x$. Then we know that
$x\in[x_{i-1},x_i)$. The linear interpolation is
$$
\widetilde{\ln x} = f(x_{i-1})+\bigl(f(x_i)-f(x_{i-1})\bigr)\,(x-x_{i-1}),
$$
where $f(x_{i-1})={}$"<valueA>", $f(x_i)={}$"<valueB>" and $x={}$"\dimen0".
The rest of the pre-calculated values is skipped by processing "\next" to "\relax".
The pre-calculated approximation of $\ln 10$ is saved in the macro \db apLNrten
because we use it at more places in the code.
\inext{apLNr}{3025}{++}
The \db apLNtenexec macro calculates the $\ln 10$ value with the precision
given by "\apFRAC". The output is prepared to the \db apLNten macro. The
"\apLNtenexec" saves globally the result to the macro "\""LNten:<apFRAC>" in
order to use it if the value is needed again. This saves time.
\inext{apLNtenexec}{^^B\cbrace}{++}
The constant $\pi$ is saved in the \db apPIvalue macro initially with 30
digits. If user needs more digits (using "\apFRAC"${}>30$) then the
"\apPIvalue" is recalculated and the \db apPIdigits is changed appropriately.
\inext{apPIvalue}{apPIdigits}{++}
The macro \db apPIexec prepares the $\pi$ constant with "\apFRAC" digits
and saves it to the \db apPI macro. And $\pi/2$ constant with "\apFRAC" digits
is saved to the \db apPIhalf macro.
The "\apPIexec" uses macros "\""apPI:<apFRAC>" and "\""apPIh:<apFRAC>" where
desired values are usually stored. If the values are not prepared here then
the macro "\apPIexecA" calculates them.
\inext{apPIexec}{^^B\cbrace}{++}
The macro \db apPIexecA creates the $\pi$ value with "\apFRAC" digits
using the "\apPIvalue", which is rounded if
"\apFRAC"${}\string<{}$"\apPIdigits". The "\apPIhalf" is calculated from
"\apPI". Finally the macros "\""apPI:<apFRAC>" and "\""apPIh:<apFRAC>" are
saved globally for saving time when we need such values again.
\inext{apPIexecA}{^^B\cbrace}{++}
If "\apFRAC"${}>{}$"\apPIdigits" then
new "\apPIvalue" with desired decimal digits is generated using \db apPIexecB
macro. The Chudnovsky formula is used:
$$
\pi = {53360\cdot \sqrt{640320} \over S}, \quad
S = \sum_{n=0}^\infty
{(6n)!\,(13591409 + 545140134\,n) \over (3n)!\,(n!)^3\,(-262537412640768000)^n }
$$
This converges very good with 14 new calculated digits per one step where
new $S_n$ is calculated. Moreover, we use the identity:
$$
F_n =
{(6n)!\over (3n)!\,(n!)^3\,(-262537412640768000)^n},\quad
F_n = F_{n-1}\cdot{8\,(6n-1)\,(6n-3)\,(6n-5) \over n^3\, (-262537412640768000)}
$$
and we use auxiliary integer constants $A_n, B_n, C_n$ with following
properties:
$$
\displaylines{A_0=B_0=C_0=1, \cr
A_n=A_{n-1}\cdot 8\,(6n-1)\,(6n-3)\,(6n-5),\quad
B_n=B_{n-1}\cdot n^3, \quad C_n=C_{n-1}\cdot (-262537412640768000), \cr
F_n = {A_n \over B_n C_n}, \cr
S_n = {A_n\,(13591409 + 545140134\,n) \over B_n C_n}
}
$$
\inext{apPIexecB}{^^B\cbrace}{++}
The macros for users \db PI and \db PIhalf are implemented as
``function-like'' macros without parameters.
\inext{PI}{PIhalf}{++}
The macros \db SIN and \db COS use the Taylor series
$$
\displaylines{
\sin x = x - {x^3\over 3!} + {x^5\over 5!} - {x^7\over 7!} + \cdots \cr
\cos x = 1 - {x^2\over 2!} + {x^4\over 4!} - {x^6\over 6!} + \cdots
}
$$
These series converge good for $|x|<1$. The main problem is to shift the
given argument $x\in{\bf R}$ to the range $[0,1)$ before the calculation of
the series is started. This task is done by \db apSINCOSa macro, the common
code for both, "\SIN" and "\COS" macros.
The macro "\apSINCOSa" does the following steps:
\begitems
\item * It advances "\apFRAC" by three and evaluates the argument.
\item * Note, that the macro "\apSINCOSx" means "\apSINx" or "\apCOSx" depending on the
given task.
\item * The macro "\signK" includes "1". It can be recalculated to "-1"
later.
\item * If the argument is zero then
the result is set and next computation is skipped. This test
is processed by "\apSINCOSo\apCOSx".
\item * If the argument is negative then remove minus and save "\sign".
This "\sign" will be applied to the result.
The "\sign" is always "+" when "\COS" is calculated. This
folows the identities $\sin(-x) = - \sin x$ and $\cos(-x) = \cos x$.
\item * The "\apFRAC" is saved and "\apTOT=0".
\item * The "\apPIexec" is processed. The "\apPI" and
"\apPIhalf" are ready after such processing.
\item * After "\X" div "\apPI" (rounded to integer) we have "\K" in "\OUT",
where $"\X"=x' + "\K"\cdot \pi$ and $x'\in[0,\pi)$.
We set "\X" := $x'$ because of the identities
$\sin x = (-1)^k \sin(x+k\pi)$, $\cos x = (-1)^k \cos(x+k\pi)$.
The sign $(-1)^k$ is saved to "\signK" macro.
\item * If the $x'$ is zero then the result is set by "\apSINCOSo\apCOSx"
and the rest of calculating is skipped.
\item * The $|"\X" - \pi/2|$ is saved to "\XmPIh" macro.
\item * If $"\X"\in (\pi/4, \pi/2)$ then $x'= "\XmPIh"$. We use identities
$\sin x = \cos(\pi/2-x)$, $\cos x = \sin(\pi/2-x)$. Set $"\X"=x'$.
The meaning of "\apSINCOSx" ("\apSINx" or "\apCOSx") is flipped in such case.
\item * If the $x'$ is zero then the result is set by "\apSINCOSo\apSINx"
and the rest of calculating is skipped.
\item * Now $"\X"\in(0,\pi/4)$, i. e. $|"\X"|<1$ and we can use Taylor
series. The "\apSINCOSx" (i.e. "\apSINx" or "\apCOSx") macro initializes
the computation of Taylor series mentioned above.
The $"\XX"="\X"^2$ is prepared. The Taylor series is processed in the loop
as usually.
\item * The the sign of the output is "\sign\signK".
\item * If the sign of the result is negative, the ``minus'' is added to the
"\OUT".
\enditems
\inext{SIN}{^^B\cbrace}{++}
The macros \db apSINx and \db apCOSx initialize the calculation
of the Taylor series.
\inext{apSINx}{apCOSx}{++}
The \db apSINCOSo "<sequence>" macro is used three times in the
"\apSINCOSa". It tests if the current result is zero. If it is true then the
"\OUT" is set as zero or it is set to "\signK"
(if processed function is equal to the "<sequence>").
\inext{apSINCOSo}{}{++}
The macro \db TAN uses the identity $\tan x = \sin x / \cos x$
and calculates the denominator first. If it is zero then "\apERR" prints
``out of range'' message else the result is calculated.
\inext{TAN}{^^B\cbrace}{++}
The macro \db ATAN calculates the inverse of tangens using series
$$
\arctan{1\over x} = {x\over 1+x^2} + {2\over3} {x\over (1+x^2)^2}
+ {2\over3}{4\over5} {x\over (1+x^2)^3}
+ {2\over3}{4\over5}{6\over7} {x\over (1+x^2)^4} + \cdots
$$
This converges relatively good for $|x|>1$. I was inspired by
the Claudio Kozick\'y's semestral work from the course ``Typography and \TeX'' at \v CVUT
in Prague.
The macro "\ATAN" takes the argument $x$ and uses identity
$\arctan(-x)=-\arctan(x)$ when $x$ is negative. If $x>1$
then the identity
$$
\arctan(x) = {\pi\over2} - \arctan{1\over x}
$$
is used and $\arctan(1/x)$ is calculated by "\apATANox" macro using the series
above. Else the argument is re-calculated $x:=1/x$ and the "\apATANox" is
used. When $x=1$ then the "\apPIhalf/2" is returned directly.
\inext{ATAN}{^^B\cbrace}{++}
The macro \db apATANox calculates $\arctan(1/x)$ using series mentioned
above.
\inext{apATANox}{^^B\cbrace}{++}
The macros \db ASIN and \db ACOS for functions
$\arcsin(x)$ and $\arccos(x)$ are implemented using following identities:
$$
\arcsin(x) = \arctan {x\over\sqrt{1-x^2}}, \qquad
\arccos(x) = {\pi\over2} - \arcsin(x)
$$
\inext{ASIN}{ACOS}{++}
\subsec Conclusion
This code is here only for backward compatibility with old versions of
"apnum.tex". Don't use these sequences if you are implementing an internal
feature because users can re-define these sequences.
\inext{PLUS}{\empty}{+-}
Here is my little joke. Of course, this macro file works in La\TeX{} without
problems because only \TeX{} primitives (from classical \TeX) and the "\newcount"
macro are used here.
But I wish to print my opinion about La\TeX. I hope that this doesn't matter
and La\TeX{} users can use my macro because a typical La\TeX{} user doesn't
read a terminal nor ".log" file.
\inext{documentclass}{\empty}{+-}
\doindex
\bye
|