summaryrefslogtreecommitdiffstats
path: root/src/erasure-code/jerasure/gf-complete/manual/gf-complete.html
blob: ed79e257661c788f00ad11b2bff0b7b592a3890e (plain)
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
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
<html>

<head>

<link rel="stylesheet" type="text/css" href="style.css">

</head>

<body>

<div id="box">

<h1>
GF-Complete: A Comprehensive Open Source Library for Galois </br>
Field Arithmetic
</h1>

<h1> Version 1.02  </h1>

<h4>James S. Plank* &nbsp&nbsp&nbsp&nbsp&nbsp&nbsp Ethan L. Miller 
Kevin M. Greenan &nbsp&nbsp&nbsp&nbsp&nbsp&nbsp Benjamin A. Arnold<br>
John A. Burnum &nbsp&nbsp&nbsp&nbsp&nbsp&nbsp Adam W. Disney  &nbsp&nbsp&nbsp&nbsp&nbsp&nbsp
Allen C. McBride

</h4> <br>



<a href="">

https://bitbucket.org/jimplank/gf-complete

 </a><br><br>
<a href=""> 
http://web.eecs.utk.edu/~plank/plank/papers/GF-Complete-Manual-1.02.pdf


 </a> <br> <br> 







</div>


<div id="pages_paragraphs_2">

This is a user's manual for GF-Complete, version 1.02. This release supersedes version 0.1 and represents the first
major release of GF-Complete. To our knowledge, this library implements every Galois Field multiplication technique
applicable to erasure coding for storage, which is why we named it GF-Complete. The primary goal of this library is
to allow storage system researchers and implementors to utilize very fast Galois Field arithmetic for Reed-Solomon
coding and the like in their storage installations. The secondary goal is to allow those who want to explore different
ways to perform Galois Field arithmetic to be able to do so effectively.


<p>
If you wish to cite GF-Complete, please cite technical report UT-CS-13-716: [PMG<sup>+</sup>13].

</p>


<h2>If You Use This Library or Document </h2>



Please send me an email to let me know how it goes. Or send me an email just to let me know you are using the
library. One of the ways in which we are evaluated both internally and externally is by the impact of our work, and if
you have found this library and/or this document useful, we would like to be able to document it. Please send mail to
<em>plank@cs.utk.edu.</em> Please send bug reports to that address as well.



<p>
The library itself is protected by the New BSD License. It is free to use and modify within the bounds of this
license. To the authors' knowledge, none of the techniques implemented in this library have been patented, and the
authors are not pursing patents. </p> <br>

 </div>
<div id="footer"> 
 
<span id="footer_bar">&nbsp&nbsp&nbsp&nbsp.*plank@cs.utk.edu (University of Tennessee), el  </span> <em>m@cs.ucsc.edu </em>(UC Santa Cruz), <em>kmgreen2@gmail.com </em> (Box). This material
is based upon work supported by the National Science Foundation under grants CNS-0917396, IIP-0934401 and CSR-1016636, plus REU supplements
CNS-1034216, CSR-1128847 and CSR-1246277. Thanks to Jens Gregor for helping us wade through compilation issues, and for Will
Houston for his initial work on this library.

</div>

<b>Finding the Code </b>
<br><br>
This code is actively maintained on bitbucket:<a href=""> https://bitbucket.org/jimplank/gf-complete. </a> There are
previous versions on my UTK site as a technical report; however, that it too hard to maintain, so the main version is
on bitbucket.<br><br>


<b>Two Related Papers </b> <br><br>

This software acccompanies a large paper that describes these implementation techniques in detail [PGM13a]. We
will refer to this as <em> "The Paper." </em> You do not have to read The Paper to use the software. However, if you want to
start exploring the various implementations, then The Paper is where you'll want to go to learn about the techniques
in detail.



<p>This library implements the techniques described in the paper "Screaming Fast Galois Field Arithmetic Using Intel
SIMD Instructions," [PGM13b]. The Paper describes all of those techniques as well.
</p><br><br>

<b>If You Would Like HelpWith the Software </b><br><br>

Please contact the first author of this manual.<br><br>

<b>Changes from Revision 1.01</b>
<br><br>
The major change is that we are using autoconf to aid with compilation, thus obviating the need for the old <b>flag_tester</b>
code. Additionally, we have added a quick timing tool, and we have modified <b>gf_methods</b> so that it may be used to
run the timing tool and the unit tester.


















<br/>
CONTENT  <span class="aligning_page_number"> 3 </span> 
<h2>Contents </h2>
<div class="index">
1 <span class="aligning_numbers">Introduction </span> <span class="aligning_page_number"> 5 </span>
  <br><br> 
2 <span class="aligning_numbers">Files in the Library </span>	<span class="aligning_page_number"> 6  </span>  <br> </div>

<div class="sub_indices">
2.1 Header files in the directory <b>"include"</b>  . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . <span class="aligning_page_number"> 6 </span>  <br>
2.2 Source files in the <b>"src"</b> directory . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . .<span class="aligning_page_number">   7  </span> <br>
2.3 Library tools files in the <b>"tools"</b> directory  . . . . . . . . . . ..  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . .   <span class="aligning_page_number"> 7   </span> <br>
2.4 The unit tester in the <b>"test"</b> directory. . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . <span class="aligning_page_number">  8  </span>  <br>
2.5 Example programs in the <b>"examples"</b> directory . . . .  . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . .<span class="aligning_page_number"> 8  </span> 

</div>
<br>
<div class="index">

3 <span class="aligning_numbers">Compilation </span><span class="aligning_page_number">  8 </span>  <br> <br>
4 <span class="aligning_numbers">Some Tools and Examples to Get You Started </span><span class="aligning_page_number">  8 </span> <br><br>  </div> 



<div class="sub_indices">
4.1 Three Simple Command Line Tools: gf_mult, gf_div and gf_add . . . . . . . . . .  . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . <span class="aligning_page_number"> 8</span>  <br>
4.2 Quick Starting Example #1: Simple multiplication and division . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . <span class="aligning_page_number">   9  </span> <br>
4.3 Quick Starting Example #2: Multiplying a region by a constant    . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . .  <span class="aligning_page_number"> 10   </span> <br>
4.4 Quick Starting Example #3: Using w = 64 . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . <span class="aligning_page_number">  11  </span>  <br>
4.5 Quick Starting Example #4: Using w = 128. . . .  . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . . <span class="aligning_page_number"> 11  </span> 
</div>
<br>


<div class="index">
5 <span class="aligning_numbers"> Important Information on Alignment when Multiplying Regions </span><span class="aligning_page_number"> 12</span> <br><br>

6 <span class="aligning_numbers"> The Defaults</span><span class="aligning_page_number"> 13 </span> <br>

</div>

<div class="sub_indices">
6.1 Changing the Defaults . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . .<span class="aligning_page_number">   14  </span> <br>


<ul style="list-style-type:none;">
<li>6.1.1 Changing the Components of a Galois Field with <b> create_gf_from_argv() </b>   . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . .  <span class="aligning_page_number"> 15   </span> <br>
</li>
<li>
6.1.2 Changing the Polynomial. . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . . . . . <span class="aligning_page_number">  16  </span>  <br>
</li>
<li>
6.1.3 Changing the Multiplication Technique. . . .  . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . .<span class="aligning_page_number"> 17  </span> 
</li>


<li>
6.1.4 Changing the Division Technique . . . . . .  . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  <span class="aligning_page_number"> 19  </span> 
</li>


<li>
6.1.5 Changing the Region Technique. . . .  . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . .  . . . ..<span class="aligning_page_number"> 19  </span> 
</li>
</ul>
6.2 Determining Supported Techniques with <b>gf_methods</b> . . . . . . . . . .  . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . <span class="aligning_page_number"> 20</span>  <br>

6.3 Testing with <b>gf_unit, gf_time,</b> and <b>time_tool.sh </b>. . . . . . . . . .  . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . .  . . <span class="aligning_page_number"> 21</span>

<ul style="list-style-type:none;">
<li>
6.3.1 <b>time_tool.sh</b> . . . . . .  . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . .  <span class="aligning_page_number"> 22 </span> 
</li>

<li>
6.3.2 An example of <b>gf_methods</b> and <b>time_tool.sh</b> . . . . . .  . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . .. . . . . . . .  . .. .  .<span class="aligning_page_number"> 23  </span> 
</li>

</ul>

6.4 Calling <b>gf_init_hard()</b> . . . . . . . . . .  . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . .. . . . . . . . .. . . . . . . .  .. . . . . . . .  . . .  <span class="aligning_page_number"> 24</span>  <br>

6.5 <b>gf_size()</b> . . . . . . . . . .  . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . .  .. . . . . . . . .. . . . . . . .  .. . . . . . . . .. . . . . . . . . . ..  .  <span class="aligning_page_number"> 26</span>  <br><br>
</div>


<div class="index">
8 <span class="aligning_numbers">  Further Information on Options and Algorithms </span><span class="aligning_page_number">   26 </span> </div> <br><br> </div>
<div class="sub_indices">
7.1 Inlining Single Multiplication and Division for Speed   . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . <span class="aligning_page_number"> 26 </span> <br>
7.2 Using different techniques for single and region multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  .  <span class="aligning_page_number"> 27 </span> <br>
7.3 General <em>w</em> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . .  . <span class="aligning_page_number"> 28  </span><br>

7.4 Arguments to <b>"SPLIT"</b> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . <span class="aligning_page_number"> 28</span>  <br>
7.5 Arguments to <b>"GROUP"</b> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  <span class="aligning_page_number">29 </span> <br>
7.6 Considerations with <b>"COMPOSITE"</b> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . .  .  <span class="aligning_page_number">30 </span> <br>
7.7 <b>"CARRY_FREE"</b> and the Primitive Polynomial  . . . . . . . . . . . . . . .  . . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . .   <span class="aligning_page_number">31 </span> <br>
7.8 More on Primitive Polynomials . .  . . . . . . . . . . . . . . . . . . .  . . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . ..   . . . . . . . . .  <span class="aligning_page_number">31 </span> <br>


<ul style="list-style-type:none;">
<li>
7.8.1 Primitive Polynomials that are not Primitive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . .  . . .  <span class="aligning_page_number"> 31</span>  <br>

</li>
<li>7.8.2 Default Polynomials for Composite Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . . . .  <span class="aligning_page_number"> 32</span>  <br>

</li>
</ul>

</div>











<br/>
CONTENT  <span class="aligning_page_number"> 4 </span> 

<div class="sub_indices">
<ul style="list-style-type:none">
<li> 7.8.3 The Program <b>gf_poly</b> for Verifying Irreducibility of Polynomials </span><span class="aligning_page_number">  33 </span> 
</li>
</ul>


7.9<span class="aligning_numbers"><b>"ALTMAP"</b> considerations and <b>extract_word()</b> </span><span class="aligning_page_number">  34 </span>  
<ul style="list-style-type:none">
<li>

7.9.1 Alternate mappings with <b>"SPLIT"</b> . . . . . . . . . .  . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .<span class="aligning_page_number"> 34</span>  <br>
</li>
<li>
7.9.2 Alternate mappings with <b>"COMPOSITE"</b> . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . . . . . <span class="aligning_page_number">   36  </span> <br>
</li>
<li>
7.9.3 The mapping of <b>"CAUCHY"</b>    . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . .. . . . . .. . . . . . . . . .  . ..  <span class="aligning_page_number"> 37   </span> <br>
</li>
</ul>
</div>


8 <span class="aligning_numbers"><b>Thread Safety </b></span><span class="aligning_page_number">  37 </span> <br><br>  </div> 

9 <span class="aligning_numbers"><b>Listing of Procedures</b> </span><span class="aligning_page_number">  37 </span> <br><br>  </div> 

10 <span class="aligning_numbers"><b>Troubleshooting</b> </span><span class="aligning_page_number">  38 </span> <br><br>  </div> 
11 <span class="aligning_numbers"><b>Timings</b> </span><span class="aligning_page_number">  41 </span> <br><br>  </div> 

<div class="sub_indices">
11.1 Multiply() . . . . . . . . . .  . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . .. . . . . . . . .. . . . . . . . .. . . . . . . . . . . . .  . . . .. . . . <span class="aligning_page_number"> 42</span>  <br>
11.2 Divide() . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . .. . . . . . . . .. . . . . . . . .. . . . . . . . .. . . . . . . . . . . .. . . . . <span class="aligning_page_number">   42  </span> <br>
11.3 Multiply Region()    . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . .. . . . . . . . .. . . . . . . . .. . . . . . . . . . . . .  . . . . .  <span class="aligning_page_number"> 43   </span> <br>
</div>






<br/>
INTRODUCTION  <span class="aligning_page_number"> 5 </span> 


<h3>1 Introduction </h3>

Galois Field arithmetic forms the backbone of erasure-coded storage systems, most famously the Reed-Solomon
erasure code. A Galois Field is defined over w-bit words and is termed <em>GF(2<sup>w</sup>).</em> As such, the elements of a Galois
Field are the integers 0, 1, . . ., 2<sup>w</sup> - 1. Galois Field arithmetic defines addition and multiplication over these closed
sets of integers in such a way that they work as you would hope they would work. Specifically, every number has a
unique multiplicative inverse. Moreover, there is a value, typically the value 2, which has the property that you can
enumerate all of the non-zero elements of the field by taking that value to successively higher powers.


<p>Addition in a Galois Field is equal to the bitwise exclusive-or operation. That's nice and convenient. Multiplication
is a little more complex, and there are many, many ways to implement it. The Paper describes them all, and the
following references providemore supporting material: [Anv09, GMS08, LHy08, LD00, LBOX12, Pla97]. The intent
of this library is to implement all of the techniques. That way, their performancemay be compared, and their tradeoffs
may be analyzed. <p>




<ol>

When used for erasure codes, there are typically five important operations:<br>
<li> <b>Adding two numbers in </b> GF(2<sup>w</sup>). That's bitwise exclusive-or. </li>
<li> <b>Multiplying two numbers in</b> GF(2<sup>w</sup>). Erasure codes are usually based on matrices in GF(2<sup>w</sup>), and constructing
these matrices requires both addition and multiplication.</li>
<li> <b>Dividing two numbers in </b>GF(2<sup>w</sup>). Sometimes you need to divide to construct matrices (for example, Cauchy
Reed-Solomon codes [BKK<sup>+</sup>95, Rab89]). More often, though, you use division to invert matrices for decoding.
Sometimes it is easier to find a number's inverse than it is to divide. In that case, you can divide by multiplying
by an inverse. </li>

<li><b>adding two regions of numbers in</b> GF(2<sup>w</sup>), which will be explained along with... </li>
<li> <b>Mutiplying a region of numbers in </b>GF(2<sup>w</sup>) by a constant in GF(2<sup>w</sup>). Erasure coding typically boils down
to performing dot products in GF(2<sup>w</sup>). For example, you may define a coding disk using the equation: </li><br>




<center>c<em><sub>0</sub></em>= d<em><sub>0</sub></em> + 2d<em><sub>1</sub></em> + 4d<em><sub>2</sub></em> + 8d<em><sub>3</sub></em>.</sup> </center><br>

That looks like three multiplications and three additions However, the way ' implemented in a disk system
looks as in Figure 1. Large regions of disks are partitioned into w-bit words in GF(2<sup>w</sup>). In the example, let us
suppose that <em>w</em> = 8, and therefore that words are bytes. Then the regions pictured are 1 KB from each disk.
The bytes on disk Di are labeled d<sub>i,0,</sub> d<sub>i,1, . . . ,</sub> d<sub>i,1023,</sub> and the equation above is replicated 1024 times. For
0 &#8804 j < 1024:
<br><br>
<center>c<em><sub>0,j</sub></em> = d<em><sub>0,j</sub></em> + 2d<em><sub>1,j</sub></em> + 4d<em><sub>2,j</sub></em> + 8d<em><sub>3,j</sub></em> . </center>
<br>


While it's possible to implement each of these 1024 equations independently, using the single multiplication
and addition operations above, it is often much more efficient to aggregate. For example, most computer architectures
support bitwise exclusive-or of 64 and 128 bit words. Thus, it makes much more sense to add regions
of numbers in 64 or 128 bit chunks rather than as words in GF(2<sup>w</sup>). Multiplying a region by a constant can
leverage similar optimizations. </ol>


<p>GF-Complete supports multiplication and division of single values for all values of <em>w</em> &#8804 32, plus <em>w</em> = 64 and <em>w</em> =
128. It also supports adding two regions of memory (for any value of <em>w</em>, since addition equals XOR), and multiplying
a region by a constant in <em>GF(2<sup>4</sup>), GF(2<sup>8</sup>), GF(2<sup>16</sup>), GF(2<sup>32</sup>), GF(2<sup>64</sup>) and GF(2<sup>128</sup>).</em> These values are chosen
because words in GF(2<sup>w</sup>) fit into machine words with these values of <em>w.</em> Other values of w don't lend themselves
to efficient multiplication of regions by constants (although see the <b>"CAUCHY"</b> option in section 6.1.5 for a way to
multiply regions for other values of <em>w</em>).</p>






<br/>

2 &nbsp &nbsp  <em>  FILES IN THE LIBRARY     </em>   <span id="index_number">6  </span> <br><br><br>



<div class="image-cell_1"> </div>  <br><br><br>

Figure 1: An example of adding two regions of numbers, and multiplying a region of numbers by a constant
in <em>GF(2<sup>w</sup>) </em>. In this example, <em>w</em> = 8, and each disk is holding a 1KB region. The same coding equation -
c<sub>0,j</sub></b> = d<sub>0,j</sub> + ad<sub>1,j</sub> + a<sup>2</sup>d<sub>2,j</sub> + a<sup>3</sup>d<sub>3,j</sub> is applied 1024 times. However, rather than executing this equation 1024
times, it is more efficient to implement this with three region-constant multiplications and three region-region additions.

<h3>2 &nbsp&nbsp&nbsp Files in the Library </h3>
This section provides an overview of the files that compose GF-Complete. They are partitioned among multiple
directories.

<h4> <b>2.1 &nbsp&nbsp&nbsp Header files in the directory  "include"</b> </h4>

The following header files are part of GF-Complete.
<ul>
<li><b>gf_complete.h:</b> This is the header file that applications should include. It defines the gf_t type, which holds
all of the data that you need to perform the various operations in GF(2<sup>w</sup>). It also defines all of the arithmetic
operations. For an application to use this library, you should include gf_complete.h and then compile with the
library src/libgf_complete.la. </li><br>

<li><b>gf_method.h:</b> If you are wanting to modify the implementation techniques from the defaults, this file provides
a "helper" function so that you can do it from the Unix command line.
</li><br>

<li><b>gf_general.h:</b> This file has helper routines for doing basic Galois Field operations with any legal value of <em>w.</em>
The problem is that <em>w </em> &#8804 32, <em>w </em> = 64 and <em> w </em> = 128 all have different data types, which is a pain. The procedures
in this file try to alleviate that pain. They are used in <b>gf_mult, gf_unit</b> and <b>gf_time.</b> I'm guessing that most
applications won't use them, as most applications use <em>w</em> &#8804 32. </li><br>

<li><b>gf_rand.h:</b> I've learned that <b>srand48()</b> and its kin are not supported in all C installations. Therefore, this file
defines some randomnumber generators to help test the programs. The randomnumber generator is the "Mother
</li>

</ul>







<br/>

2 &nbsp &nbsp  <em>  FILES IN THE LIBRARY     </em>   <span id="index_number">7  </span> <br><br><br>
<ul>

of All" random number generator [Mar94] which we've selected because it has no patent issues. <b>gf_unit</b> and
<b>gf_time</b> use these random number generators.<br><br>
<li><b>gf_int.h:</b> This is an internal header file that the various source files use. This is <em>not</em> intended for applications to
include.</li><br>
<li><b>config.xx</b> and <b>stamp-h1</b> are created by autoconf, and should be ignored by applications. </li>
</ul>

<h3>2.2 &nbsp &nbsp <b> Source files in the "src" directory" </b> </h3>
<ul>
The following C files compose <b>gf_complete.a,</b> and they are in the direcoty src. You shouldn't have to mess with these
files, but we include them in case you have to:<br><br>
<li><b> gf_.c:</b> This implements all of the procedures in both <b>gf_complete.h</b> and <b>gf_int.h.</b> </li><br>
<li><b> gf_w4.c:</b> Procedures specific to <em>w </em> = 4. </li><br>
<li> <b>gf_w8.c:</b> Procedures specific to <em>w </em> = 8</li><br>
<li> <b>gf_w16.c:</b> Procedures specific to <em>w </em> = 16</li><br>
<li> <b>gf_w32.c:</b> Procedures specific to <em>w </em> = 32</li><br>
<li><b>gf_w64.c:</b> Procedures specific to <em>w </em> = 64</li><br>
<li> <b>gf_w128.c:</b> Procedures specific to <em>w </em> = 128</li><br>
<li> <b>gf_wgen.c:</b> Procedures specific to other values of <em>w </em> between 1 and 31</li><br>
<li> <b>gf_general.c:</b> Procedures that let you manipulate general values, regardless of whether <em>w </em> &#8804 32, <em>w </em> = 64
or <em>w </em> = 128. (I.e. the procedures defined in <b>gf_ general.h</b>)</li><br>
<li> <b>gf_method.c:</b> Procedures to help you switch between the various implementation techniques. (I.e. the procedures
defined in <b>gf_method.h</b>)</li><br>
<li> <b>gf_ rand.c:</b>"The Mother of all" random number generator. (I.e. the procedures defined in <b>gf_rand.h</b>)</li><br> </ul>

<h3>2.3 &nbsp &nbsp Library tools files in the "tools" directory </h3>

<ul>
The following are tools to help you with Galois Field arithmetic, and with the library. They are explained in greater
detail elsewhere in this manual.<br><br>
<li> <b>gf_mult.c, gf_ div.c</b> and <b>gf_ add:</b> Command line tools to do multiplication, division and addition by single numbers</li><br>
<li> <b>gf_time.c:</b> A program that times the procedures for given values of <em>w </em> and implementation options</li><br>
<li> <b>time_tool.sh:</b> A shell script that helps perform rough timings of the various multiplication, division and region
operations in GF-Complete</li><br>
<li> <b>gf_methods.c:</b> A program that enumerates most of the implementation methods supported by GF-Complete</li><br>
<li> <b> gf_poly.c:</b> A program to identify irreducible polynomials in regular and composite Galois Fields</li><br>

</ul>








<br/>

3 &nbsp &nbsp  <em>  COMPILATION     </em>   <span id="index_number">8  </span> <br><br><br>


<h3>2.4 &nbsp &nbsp The unit tester in the "test" directory </h3>

The test directory contains the proram <b>gf_unit.c,</b> which performs a battery of unit tests on GF-Complete. This is
explained in more detail in section 6.3.


<h3>2.5&nbsp &nbsp Example programs in the "examples" directory </h3>

There are seven example programs to help you understand various facets of GF-Complete. They are in the files
<b>gf_example x.c </b> in the <b>examples</b> directory. They are explained in sections 4.2 through 4.5, and section 7.9.<br><br>

<h2>3 &nbsp &nbsp Compilation </h2>

<em>From revision 1.02 forward, we are using autoconf. The old "flag tester" directory is now gone, as it is no longer in
use. </em><br><br>
To compile and install, you should do the standard operations that you do with most open source Unix code:<br><br>

UNIX> ./configure <br>
... <br>
UNIX> make <br>
... <br>
UNIX> sudo make install <br><br>


<p>If you perform the <b>install,</b> then the header, source, tool, and library files will be moved to system locations. In
particular, you may then compile the library by linking with the flag <b>-lgf_complete,</b> and you may use the tools from a
global executable directory (like <b>/usr/local/bin</b>). </p>

<p>
If you don't perform the install, then the header and tool files will be in their respective directories, and the library
will be in <b>src/libgf_complete.la.</b> </p>
<p>
If your system supports the various Intel SIMD instructions, the compiler will find them, and GF-Complete will
use them by default. </p>



<h2>4 &nbsp &nbsp Some Tools and Examples to Get You Started </h2> 
<h3>4.1 Three Simple Command Line Tools: gf_mult, gf_div and gf_add </h3>


Before delving into the library, it may be helpful to explore Galois Field arithmetic with the command line tools:
<b>gf_mult, gf_div </b> and <b>gf_add.</b> These perform multiplication, division and addition on elements in <em>GF(2<sup>w</sup>).</em> If these are
not installed on your system, then you may find them in the tools directory. Their syntax is:
<ul>
<li><b>gf_mult a b</b> <em>w </em> - Multiplies a and b in <em> GF(2<sup>w</sup>)</em>. </li><br>
<li> <b>gf_div a b </b><em>w </em> - Divides a by b in GF(2<em><sup>w </sup></em>). </li><br>
<li><b>gf_add a b </b> <em>w </em> - Adds a and b in GF(2<em><sup>w </sup> </em>). </li><br>

You may use any value of <em>w </em> from 1 to 32, plus 64 and 128. By default, the values are read and printed in decimal;
however, if you append an 'h' to <em>w </em>, then <em>a, b </em> and the result will be printed in hexadecimal. For <em>w </em> = 128, the 'h' is
mandatory, and all values will be printed in hexadecimal.







<br/>

4 &nbsp &nbsp  <em>   SOME TOOLS AND EXAMPLES TO GET YOU STARTED 9     </em>   <span id="index_number">9  </span> <br><br><br>


<p>Try them out on some examples like the ones below. You of course don't need to know that, for example, 5 * 4 = 7
in <em>GF(2<sup>4 </sup>) </em>; however, once you know that, you know that 7/
5 = 4 and 7/4 = 5. You should be able to verify the <b>gf_add</b>
statements below in your head. As for the other <b>gf_mult's</b>, you can simply verify that division and multiplication work
with each other as you hope they would. </p>
<br><br>
<div id="number_spacing">

UNIX> gf_mult 5 4 4  <br>
7 <br>
UNIX> gf_div 7 5 4 <br>
4 <br>
UNIX> gf_div 7 4 4 <br>
5   <br>
UNIX> gf_mult 8000 2 16h <br>
100b  <br>
UNIX> gf_add f0f0f0f0f0f0f0f0 1313131313131313 64h <br>
e3e3e3e3e3e3e3e3 <br>
UNIX> gf_mult f0f0f0f0f0f0f0f0 1313131313131313 64h <br>
8da08da08da08da0 <br>
UNIX> gf_div 8da08da08da08da0 1313131313131313 64h <br>
f0f0f0f0f0f0f0f0  <br>
UNIX> gf_add f0f0f0f0f0f0f0f01313131313131313 1313131313131313f0f0f0f0f0f0f0f0 128h <br>
e3e3e3e3e3e3e3e3e3e3e3e3e3e3e3e3 <br>
UNIX> gf_mult f0f0f0f0f0f0f0f01313131313131313 1313131313131313f0f0f0f0f0f0f0f0 128h <br>
786278627862784982d782d782d7816e <br>
UNIX> gf_div 786278627862784982d782d782d7816e f0f0f0f0f0f0f0f01313131313131313 128h <br>
1313131313131313f0f0f0f0f0f0f0f0 <br>
UNIX> <br><br>

</div>


Don't bother trying to read the source code of these programs yet. Start with some simpler examples  like the ones
below. <br><br>

<h3>4.2 Quick Starting Example #1: Simple multiplication and division </h3>

The source files for these examples are in the examples directory.
<p>These two examples are intended for those who just want to use the library without getting too complex. The
first example is <b>gf_example 1,</b> and it takes one command line argument - w, which must be between 1 and 32. It
generates two random non-zero numbers in <em>GF(2<sup>w </sup>) </em> and multiplies them. After doing that, it divides the product by
each number. </p>
<p>
To perform multiplication and division in <em>GF(2<sup>w </sup>) </em>, you must declare an instance of the gf_t type, and then initialize
it for <em>GF(2<sup>w </sup>) </em> by calling <b>gf_init_easy().</b> This is done in <b>gf_example 1.c</b> with the following lines: </p><br><br>

gf_t gf; <br><br>r
... <br><br>
if (!gf_init_easy(&gf, w)) { <br>
fprintf(stderr, "Couldn't initialize GF structure.\n"); <br>
exit(0); <br>
}  <br>






<br/>

4 &nbsp &nbsp  <em>   SOME TOOLS AND EXAMPLES TO GET YOU STARTED      </em>   <span id="index_number">10  </span> <br><br><br>

<p>Once <b>gf</b> is initialized, you may use it for multiplication and division with the function pointers <b>multiply.w32</b> and
<b>divide.w32.</b> These work for any element of <em>GF(2<sup>w</sup>)</em> so long as w &#8804 32. </p> <br><br>

<div id="number_spacing">
<div style="padding-left:54px">
c = gf.multiply.w32(&gf, a, b);<br>
printf("%u * %u = %u\n", a, b, c);<br><br>
printf("%u / %u = %u\n", c, a, gf.divide.w32(&gf, c, a));<br>
printf("%u / %u = %u\n", c, b, gf.divide.w32(&gf, c, b));<br>


</div> </div>
<br><br>
Go ahead and test this program out. You can use <b>gf_mult</b> and <b>gf_div</b> to verify the results:<br><br>

<div id="number_spacing">
UNIX> gf_example_1 4 <br>
12 * 4 = 5  <br>
5 / 12 = 4  <br>
5 / 4 = 12  <br>
UNIX> gf_mult 12 4 4 <br>
5  <br>
UNIX> gf_example_1 16 <br>
14411 * 60911 = 44568 <br>
44568 / 14411 = 60911 <br>
44568 / 60911 = 14411  <br>
UNIX> gf_mult 14411 60911 16 <br>
44568 <br>
UNIX>  <br><br>
</div>

<b>gf_init_easy()</b> (and <b>later_gf_init_hard()</b>) do call <b>malloc()</b> to implement internal structures. To release memory, call
<b>gf_free().</b> Please see section 6.4 to see how to call <b>gf_init_hard()</b> in such a way that it doesn't call <b>malloc().</b> <br><br>



<h3>4.3 &nbsp &nbsp &nbspQuick Starting Example #2: Multiplying a region by a constant </h3>


The program <b>gf_example</b> 2 expands on <b>gf_example</b> 1. If <em>w</em> is equal to 4, 8, 16 or 32, it performs a region multiply
operation. It allocates two sixteen byte regions, <b>r1</b> and <b>r2,</b> and then multiples <b>r1</b> by a and puts the result in <b>r2</b> using
the <b>multiply_region.w32</b> function pointer: <br><br>

<div style="padding-left:52px">
gf.multiply_region.w32 (&gf, r1, r2, a, 16, 0); <br><br>
</div>

That last argument specifies whether to simply place the product into r2 or to XOR it with the contents that are already
in r2. Zero means to place the product there. When we run it, it prints the results of the <b>multiply_region.w32</b> in
hexadecimal. Again, you can verify it using <b>gf_mult</b>:<br><br>
<div id="number_spacing">
UNIX> gf_example_2 4 <br>
12 * 2 = 11 <br>
11 / 12 = 2 <br>
11 / 2 = 12 <br><br>
multiply_region by 0xc (12) <br><br>
R1 (the source): 0 2 d 9 d 6 8 a 8 d b 3 5 c 1 8 8 e b 0 6 1 5 a 2 c 4 b 3 9 3 6 <br>
R2 (the product): 0 b 3 6 3 e a 1 a 3 d 7 9 f c a a 4 d 0 e c 9 1 b f 5 d 7 6 7 e <br>

</div>










<br/>

4 &nbsp &nbsp  <em>   SOME TOOLS AND EXAMPLES TO GET YOU STARTED      </em>   <span id="index_number">11  </span> <br><br><br>

<div id="number_spacing">
<table cellpadding="6">
<tr><td>UNIX></td> <td colspan="4"> gf_example_2 16 </td> </tr>

<tr>

<td>49598</td> <td> * </td> <td> 35999</td> <td> = </td> <td>19867 </td> </tr>

<tr><td>19867 </td><td>/ </td> <td> 49598 </td> <td> =  </td> <td>35999 </td> </tr>
<tr><td>19867</td><td> /</td> <td> 35999 </td> <td> = </td> <td> 49598 </td> </tr>  </table><br>


&nbsp multiply_region by 0xc1be (49598) <br><br>


<table cellpadding="6" >
<tr>
<td>R1 (the source):</td> <td> 8c9f </td> <td> b30e </td> <td> 5bf3 </td> <td> 7cbb </td> <td>16a9 </td> <td> 105d </td> <td> 9368 </td> <td> 4bbe </td> </tr>
<td>R2 (the product):</td> <td> 4d9b</td> <td> 992d </td> <td> 02f2 </td> <td> c95c </td> <td> 228e </td> <td> ec82 </td> <td> 324e </td> <td> 35e4 </td></tr>
</table>
</div>
<div id="number_spacing">
<div style="padding-left:9px">
UNIX> gf_mult c1be 8c9f 16h<br>
4d9b <br>
UNIX> gf_mult c1be b30e 16h <br>
992d <br>
UNIX> <br><br>
</div>
</div>

<h3>4.4 &nbsp &nbsp &nbsp Quick Starting Example #3: Using <em>w </em>= 64 </h3>
The program in <b>gf_example 3.c </b> is identical to the previous program, except it uses <em> GF(2<sup>64 </sup>). </em> Now <em>a, b</em> and <em> c </em> are
<b>uint64 t'</b>s, and you have to use the function pointers that have <b>w64</b> extensions so that the larger types may be employed.
<br><br>
<div id="number_spacing">

UNIX> gf_example_31 
<table cellpadding="6">
<tr>

<td>a9af3adef0d23242 </td> <td> * </td> <td> 61fd8433b25fe7cd</td> <td> = </td> <td>bf5acdde4c41ee0c </td> </tr>

<td>bf5acdde4c41ee0c </td> <td> / </td> <td> a9af3adef0d23242 </td> <td> = </td> <td>61fd8433b25fe7cd </td> </tr>
<td>bf5acdde4c41ee0c </td> <td> / </td> <td> 61fd8433b25fe7cd  </td> <td>= </td> <td>a9af3adef0d23242 </td> </tr>
</table><br><br>

&nbsp multiply_region by a9af3adef0d23242<br><br>
<table cellpadding="6" >
<tr>
<td>R1 (the source): </td> <td> 61fd8433b25fe7cd </td> <td>272d5d4b19ca44b7 </td> <td> 3870bf7e63c3451a </td> <td> 08992149b3e2f8b7 </td> </tr>
<tr><td>R2 (the product): </td> <td> bf5acdde4c41ee0c </td> <td> ad2d786c6e4d66b7 </td> <td> 43a7d857503fd261 </td> <td> d3d29c7be46b1f7c </td> </tr>
</table>

<div style="padding-left:9px">

UNIX> gf_mult a9af3adef0d23242 61fd8433b25fe7cd 64h <br>
bf5acdde4c41ee0c<br>
UNIX><br><br>
</div>
</div>
<h3>4.5 &nbsp &nbsp &nbsp Quick Starting Example #4: Using <em>w </em>= 128 </h3>
Finally, the program in <b>gf_example_4.c</b> uses  <em>GF(2<sup>128</sup>).</em> Since there is not universal support for uint128 t, the library
represents 128-bit numbers as arrays of two uint64 t's. The function pointers for multiplication, division and region
multiplication now accept the return values as arguments:<br><br>

gf.multiply.w128(&gf, a, b, c); <br><br>

Again, we can use <b>gf_mult </b> and <b>gf_div </b>to verify the results:<br><br>
<div id="number_spacing">
<div style="padding-left:9px">
UNIX> gf_example_4 </div>
<table cellpadding="6" >
<tr>

<td>e252d9c145c0bf29b85b21a1ae2921fa </td> <td> * </td> <td> b23044e7f45daf4d70695fb7bf249432 </td> <td> = </td> </tr>
<tr><td>7883669ef3001d7fabf83784d52eb414 </td> </tr>

</table>

</div>








<br/>

4 &nbsp &nbsp  <em>   IMPORTANT INFORMATION ON ALIGNMENT WHEN MULTIPLYING REGIONS      </em>   <span id="index_number">12  </span> <br><br><br>

<div id="number_spacing">
multiply_region by e252d9c145c0bf29b85b21a1ae2921fa <br>
R1 (the source): f4f56f08fa92494c5faa57ddcd874149 b4c06a61adbbec2f4b0ffc68e43008cb <br>
R2 (the product): b1e34d34b031660676965b868b892043 382f12719ffe3978385f5d97540a13a1 <br>
UNIX> gf_mult e252d9c145c0bf29b85b21a1ae2921fa f4f56f08fa92494c5faa57ddcd874149 128h <br>
b1e34d34b031660676965b868b892043 <br>
UNIX> gf_div 382f12719ffe3978385f5d97540a13a1 b4c06a61adbbec2f4b0ffc68e43008cb 128h<br>
e252d9c145c0bf29b85b21a1ae2921fa<br>
UNIX><br><br>

</div>


<h2>5 &nbsp &nbsp &nbspImportant Information on Alignment when Multiplying Regions </h2>



In order to make multiplication of regions fast, we often employ 64 and 128 bit instructions. This has ramifications
for pointer alignment, because we want to avoid bus errors, and because on many machines, loading and manipulating
aligned quantities is much faster than unalinged quantities.<br><br>


When you perform multiply_region.wxx(<em>gf, source, dest, value, size, add </em>), there are three requirements:
<ol>
<li>
 The pointers <em>source</em> and <em>dest </em> must be aligned for <em>w</em>-bit words. For <em>w </em> = 4 and <em>w </em> = 8, there is no restriction;
however for <em>w </em> = 16, the pointers must be multiples of 2, for <em>w </em> = 32, they must be multiples of 4, and for
<em>w </em> &#1013; {64, 128}, they must be multiples of 8. </li><br>

<li> The <em>size</em> must be a multiple of &#91; <em>w /
</em> 
8 .&#93;
 With <em>w </em> = 4 and <em>w </em> = 8, <em>w/ </em>
8  = 1 and there is no restriction. The other
sizes must be multiples of <em>w </em>/
8  because you have to be multiplying whole elements of <em> GF(2<sup>w </sup>) </em>. </li><br>

<li> The <b>source</b> and <b>dest</b> pointers must be aligned identically with respect to each other for the implementation
chosen. This is subtle, and we explain it in detail in the next few paragraphs. However, if you'd rather not figure
it out, the following recommendation will <em>always </em> work in GF-Complete: </li>

</ol>



<div style="padding-left:100px">
<b>If you want to be safe, make sure that source and dest are both multiples of 16. That is not a
strict requirement, but it will always work! </b> <br><br>
</div>


If you want to relax the above recommendation, please read further.
<p>When performing <b>multiply_region.wxx() </b>, the implementation is typically optimized for a region of bytes whose
size must be a multiple of a variable <em>s </em> ,, and which must be aligned to a multiple of another variable <em>t </em>. For example,
when doing <b>multiply_region.w32() </b> in <em> GF(2<sup>16 </sup>) </em> with SSE enabled, the implementation is optimized for regions of
32 bytes, which must be aligned on a 16-byte quantity. Thus, <em>s </em> = 32 and <em>t</em> = 16. However, we don't want <b>multiply_
region.w32() </b> to be too restrictive, so instead of requiring <em>source</em> and <em> dest </em> to be aligned to 16-byte regions, we
require that (<em>source </em> mod 16) equal (<em>dest</em> mod 16). Or, in general, that (<em>source</em> mod t) equal (<em>dest</em> mod <em>t</em>). </p>


<p>
Then, <b>multiply_region.wxx()</b> proceeds in three phases. In the first phase,<b> multiply.wxx()</b> is called on successive
words until (<em>source</em> mod <em>t</em>) equals zero. The second phase then performs the optimized region multiplication on
chunks of <em> s  </em>bytes, until the remaining part of the region is less than s bytes. At that point, the third phase calls
<em>multiply.wxx() </em> on the last part of the region. </p>

A detailed example helps to illustrate. Suppose we make the following call in <em>GF(2<sup>16</sup>) </em> with SSE enabled:<br><br>
<center><b>multiply region.w32(gf, 0x10006, 0x20006, a, 274, 0)</b> </center>







<br/>

2 &nbsp &nbsp  <em>  FILES IN THE LIBRARY     </em>   <span id="index_number">13  </span> <br><br><br>



<div class="image-cell_2"> </div>  <br><br><br>

Figure 2: Example of multiplying a region of 274 bytes in GF(216) when (source mod 16) = (dest mod 16) = 6. The
alignment parameters are s = 32 and t = 16. The multiplication is in three phases, which correspond to the initial
unaligned region (10 bytes), the aligned region of s-byte chunks (256 bytes), and the final leftover region (8 bytes).


<p>First, note that <em>source</em> and <em>dest</em> are aligned on two-byte quantities, which they must be in <em>GF(2<sup>16</sup>).</em> Second, note
that size is a multiple of &#91; 16/
8 &#93 = 2. And last, note that (<em>source</em> mod 16) equals (<em>dest</em> mod 16). We illustrate the three
phases of region multiplication in Figure 2. Because (<em>source</em> mod 16) = 6, there are 10 bytes of unaligned words that
are multiplied with five calls to <b>multiply.w32()</b> in the first phase. The second phase multiplies 256 bytes (eight chunks
of <em>s</em> = 32 bytes) using the SSE instructions. That leaves 8 bytes remaining for the third phase.
</p>

<p>
When we describe the defaults and the various implementation options, we specify s and t as "alignment parameters."
</p>
<p>
One of the advanced region options is using an alternate mapping of words to memory ("ALTMAP"). These interact
in a more subtle manner with alignment. Please see Section 7.9 for details.
</p>

<h3> 6 &nbsp &nbspThe Defaults </h3>


GF-Complete implements a wide variety of techniques for multiplication, division and region multiplication. We have
set the defaults with three considerations in mind:
<ol>
<li>
<b>Speed:</b> Obviously, we want the implementations to be fast. Therefore, we choose the fastest implementations
that don’t violate the other considerations. The compilation environment is considered. For example, if SSE is
enabled, region multiplication in <em> GF(2<sup>4 </sup>) </em> employs a single multiplication table. If SSE is not enabled, then a
"double" table is employed that performs table lookup two bytes at a time. </li><br>
<li>
<b>Memory Consumption:</b> We try to keep the memory footprint of GF-Complete low. For example, the fastest
way to perform <b>multiply.w32()</b> in <em>GF(2<sup>32</sup>) </em> is to employ 1.75 MB of multiplication tables (see Section 7.4
below). We do not include this as a default, however, because we want to keep the default memory consumption
of GF-Complete low.
</li>

</ul>






<br/>

6 &nbsp &nbsp  <em>  THE DEFAULTS     </em>   <span id="index_number">14  </span> <br><br><br>

<ul>

3. &nbsp <b>Compatibility with "standard" implementations:</b> While there is no <em>de facto</em> standard of Galois Field arithmetic,
most libraries implement the same fields. For that reason, we have not selected composite fields, alternate
polynomials or memory layouts for the defaults, even though these would be faster. Again, see section 7.7 for
more information.

</ul>

<p>Table 1 shows the default methods used for each power-of-two word size, their alignment parameters <em>s</em> and <em> t,</em> their
memory consumption and their rough performance. The performance tests are on an Intel Core i7-3770 running at
3.40 GHz, and are included solely to give a flavor of performance on a standard microprocessor. Some processors
will be faster with some techniques and others will be slower, so we only put numbers in so that you can ballpark it.
For other values of <em>w</em> between 1 and 31, we use table lookup when w &#8804 8, discrete logarithms when w &#8804 16 and
"Bytwo<sub>p</sub>" for w &#8804 32. </p>
<br><br>
<center> With SSE 
<div id="data1">
<table cellpadding="6" cellspacing="0">
<tr>
<th>w </th><th class="double_border" >Memory <br> Usage </br> </th><th>multiply() <br> Implementation</th><th>Performance <br>(Mega Ops / s) </th><th>multiply region() <br> Implementation </th>
<th>s </th> <th>t </th> <th> Performance <br>(MB/s)</th>
</tr>
<tr>
<td>4 </td><td class="double_border"><1K </td><td>Table</td><td>501</td><td>Table</td>
<td>16 </td><td>16 </td> <td>11,659</td> </tr>

<tr>
<td>8 </td><td class="double_border">136K </td><td>Table</td><td>501</td><td>Split Table (8,4)</td>
<td>16 </td><td>16 </td> <td>11,824</td> </tr>

<tr>
<td>16 </td><td class="double_border">896K </td><td>Log</td><td>260</td><td>Split Table (16,4)</td>
<td>32 </td><td>16 </td> <td>7,749</td> </tr>

<tr>
<td>32 </td><td class="double_border"><1K </td><td>Carry-Free</td><td>48</td><td>Split Table (32,4)</td>
<td>64 </td><td>16 </td> <td>5,011</td> </tr>

<tr>
<td>64 </td><td class="double_border">2K </td><td>Carry-Free</td><td>84</td><td>Split Table (64,4)</td>
<td>128 </td><td>16 </td> <td>2,402</td> </tr>

<tr>
<td>128 </td><td class="double_border">64K </td><td>Carry-Free</td><td>48</td><td>Split Table (128,4)</td>
<td>16 </td><td>16 </td> <td>833</td> </tr>
</table></div>


<div id="data1">
<center>Without SE </center>
<table cellpadding="6" cellspacing="0">
<tr>
<th>w </th><th>Memory <br> Usage </br> </th><th>multiply() <br> Implementation</th><th>Performance <br>(Mega Ops / s) </th><th>multiply region() <br> Implementation </th>
<th>s </th> <th>t </th> <th> Performance <br>(MB/s)</th>
</tr>
<tr>
<td>4 </td><td>4K </td><td>Table</td><td>501</td><td>Double Table</td>
<td>16 </td><td>16 </td> <td>11,659</td> </tr>

<tr>
<td>8 </td><td>128K </td><td>Table</td><td>501</td><td>Table</td>
<td>1 </td><td>1 </td> <td>1,397</td> </tr>

<tr>
<td>16 </td><td>896K </td><td>Log</td><td>266</td><td>Split Table (16,8)</td>
<td>32 </td><td>16 </td> <td>2,135</td> </tr>

<tr>
<td>32 </td><td>4K </td><td>Bytwo<sub>p</sub></td><td>19</td><td>Split Table (32,4)</td>
<td>4 </td><td>4 </td> <td>1,149</td> </tr>

<tr>
<td>64 </td><td>16K </td><td>Bytwo<sub>p</sub></td><td>9</td><td>Split Table (64,4)</td>
<td>8 </td><td>8 </td> <td>987</td> </tr>

<tr>
<td>128 </td><td>64K </td><td>Bytwo<sub>p</sub></td><td>1.4</td><td>Split Table (128,4)</td>
<td>16 </td><td>8 </td> <td>833</td> </tr>
</table>
</div>
</center>
<br><br>
Table 1: The default implementations, memory consumption and rough performance when w is a power of two. The
variables s and t are alignment variables described in Section 5.
<p>
A few comments on Table 1 are in order. First, with SSE, the performance of <b>multiply()</b> is faster when <em> w </em> = 64
than when<em> w </em> = 32. That is because the primitive polynomial for <em> w  </em>= 32, that has historically been used in Galois
Field implementations, is sub-ideal for using carry-free multiplication (PCLMUL). You can change this polynomial
(see section 7.7) so that the performance matches <em>w </em> = 64. </p>
<p>
The region operations for <em> w  </em>= 4 and <em>w </em>= 8 without SSE have been selected to have a low memory footprint. There
are better options that consume more memory, or that only work on large memory regions (see section 6.1.5).
</p>

There are times that you may want to stray from the defaults. For example:
<ul>
<li>
You may want better performance.
</li>

</ul>










<br/>

6 &nbsp &nbsp  <em>  THE DEFAULTS     </em>   <span id="index_number">15  </span> <br><br><br>

<ul>
<li>You may want a lower memory footprint.</li>
<li>You may want to use a different Galois Field or even a ring.</li>
<li>You only care about multiplying a region by the value two.</li>

</ul>


<p>
Our command line tools allow you to deviate from the defaults, and we have two C functions <b>-gf_init_hard()</b>
and <b>create_gf_from_argv()</b> that can be called from application code to override the default methods. There are six
command-line tools that can be used to explore the many techniques implemented in GF-Complete: </p>

<ul><br>

<li> <b>gf_methods</b> is a tool that enumerates most of the possible command-line arguments that can be sent to the other
tools</li><br>
<li> <b>gf_mult</b> and <b>gf_div</b> are explained above. You may change the multiplication and division technique in these
tools if you desire</li><br>
<li> <b>gf_unit</b> performs unit tests on a set of techniques to verify correctness</li><br>
<li> <b> gf_time measures </b> the performance of a particular set of techniques</li><br>
<li> <b>time_tool.sh </b> makes some quick calls to <b>gf_time</b> so that you may gauge rough performance.</li><br>
<li> <b>gf_poly</b> tests the irreducibility of polynomials in a Galois Field</li><br>
</ul>


<p>To change the default behavior in application code, you need to call <b>gf_init_hard()</b> rather than <b>gf_init_easy().</b>
Alternatively, you can use <b>create_g_from_argv(),</b> included from <b>gf_method.h,</b> which uses an <b>argv</b>-style array of
strings to specify the options that you want. The procedure in <b>gf_method.c</b> parses the array and makes the proper
<b>gf_init_hard()</b> procedure call. This is the technique used to parse the command line in <b> gf_mult, gf_div, gf_unit </b><em>et al.</em> </p>


<h2>6.1.1 Changing the Components of a Galois Field with create <b>gf_from_argv()</b> </h2>
There are five main components to every Galois Field instance:
<ul>
<li> <em>w </em> </li>
<li> Multiplication technique </li>
<li> Division technique  </li>
<li> Region technique(s) </li>
<li> Polynomial </li>
</ul>

<p>The procedures <b>gf_init_hard()</b> and <b> create_gf_from_argv()</b> allow you to specify these parameters when you create
your Galois Field instance. We focus first on <b>create_gf_from_argv(),</b> because that is how the tools allow you to specify
the components. The prototype of <b>create_gf_from_argv()</b> is as follows: </p><br>

<div id="number_spacing">
int create_gf_from_argv(gf_t *gf, int w, int argc, char **argv, int starting);<br><br> </div>

You pass it a pointer to a gf_t, which it will initialize. You specify the word size with the parameter <em><b>w,</b></em> and then you
pass it an <b>argc/argv</b> pair as in any C or C++ program. You also specify a <b>starting</b> argument, which is where in <b>argv</b>
the specifications begin. If it successfully parses <b>argc</b> and <b>argv,</b> then it creates the <b>gf_t</b> using <b>gf_init_hard()</b> (described
below in section 6.4). It returns one past the last index of <b>argv</b> that it considered when creating the <b>gf_t.</b> If it fails, then
it returns zero, and the <b>gf_t</b> is unmodified.



<p>For example, <b>gf_mult.c</b> calls create gf_from_argv() by simply passing <b>argc</b> and <b>argv</b> from its <b>main()</b> declaration,
and setting starting to 4.</p>








<br/>

6 &nbsp &nbsp  <em>  THE DEFAULTS     </em>   <span id="index_number">16  </span> <br><br><br>

<p>
To choose defaults, <b>argv[starting]</b> should equal "-". Otherwise, you specify the component that you are changing
with "-m" for multiplication technique, "-d" for division technique, "-r" for region technique, and "-p" for the
polynomial. You may change multiple components. You end your specification with a single dash. For example, the
following call multiplies 6 and 5 in <em>GF(2<sup>4</sup>)</em> with polynomial 0x19 using the "SHIFT" technique for multiplication
(we'll explain these parameters later):
</p><br><br>

<div id="number_spacing">
UNIX> ./gf_mult 6 5 4 -p 0x19 -m SHIFT -<br>
7 <br>
UNIX> <br><br>
</div>

<p>If <b>create_gf_from_argv()</b> fails, then you can call the procedure <b>gf_error(),</b> which prints out the reason why <b>create_
gf_from_argv()</b> failed. </p>


<h2>6.1.2 Changing the Polynomial </h2>

Galois Fields are typically implemented by representing numbers as polynomials with binary coefficients, and then
using the properties of polynomials to define addition and multiplication. You do not need to understand any of that to
use this library. However, if you want to learn more about polynomial representations and how they construct fields,
please refer to The Paper.

<p>Multiplication is based on a special polynomial that we will refer to here as the "defining polynomial." This
polynomial has binary coefficients and is of degree <em> w.</em> You may change the polynomial with "-p" and then a number
in hexadecimal (the leading "0x" is optional). It is assumed that the <em>w</em>-th bit of the polynomial is set - you may include
it or omit it. For example, if you wish to set the polynomial for GF(2<sup>16</sup>) to x<sup>16</sup> + x<sup>5</sup> + x<sup>3</sup> + x<sup>2</sup> + 1, rather than its
default of x<sup>16</sup> + x<sup>12</sup> + x<sup>3</sup> + x + 1, you may say "-p 0x1002d," "-p 1002d," "-p 0x2d" or "-p 2d."
We discuss changing the polynomial for three reasons in other sections: </p>
<ul>
<li>Leveraging carry-free multiplication (section 7.7). </li>
<li>Defining composite fields (section 7.6). </li>
<li>Implementing rings (section 7.8.1). </li>

</ul>

<p>
Some words about nomenclature with respect to the polynomial. A Galois Field requires the polynomial to be
<em>irreducible </em>.. That means that it cannot be factored. For example, when the coefficients are binary, the polynomial x<sup>5</sup>+
x<sup>4</sup>+x+1 may be factored as (x<sup>4</sup>+1)(x+1). Therefore it is not irreducible and cannot be used to define a Galois Field.
It may, however, be used to define a ring. Please see section 7.8.1 for a discussion of ring support in GF-Complete. </p>
<p>
There is a subset of irreducible polynomials called primitive. These have an important property that one may enumerate
all of the elements of the field by raising 2 to successive posers. All of the default polynomials in GF-Complete 
are primitive. However, so long as a polynomial is irreducible, it defines a Galois Field. Please see section 7.7 for a
further discussion of the polynomial. </p>

<p>
One thing that we want to stress here is that changing the polynomial changes the field, so fields with different
polynomialsmay not be used interchangeably. So long as the polynomial is irreducible, it generates a Galois Field that
is isomorphic to all other Galois Fields; however the multiplication and division of elements will differ. For example,
the polynomials 0x13 (the default) and 0x19 in <em>GF(2<sup>4</sup>) </em> are both irreducible, so both generate valid Galois Fields.
However, their multiplication differs: </p><br>

<div id="number_spacing">
UNIX> gf_mult 8 2 4 -p 0x13 - <br>
3 <br>
UNIX> gf_mult 8 2 4 -p 0x19 - <br>
9 <br>
</div>









<br/>

6 &nbsp &nbsp  <em>  THE DEFAULTS     </em>   <span id="index_number">17  </span> <br><br><br>

<div id="number_spacing">
UNIX> gf_div 3 8 4 -p 0x13 -<br>
2 <br>
UNIX> gf_div 9 8 4 -p 0x19 - <br>
2 <br>
UNIX> <br>

</div>


<h3>6.1.3 &nbsp &nbsp Changing the Multiplication Technique </h3>
The following list describes the multiplication techinques that may be changed with "-m". We keep the description
here brief. Please refer to The Paper for detailed descriptions of these techniques.<br><br>


<li><b> "TABLE:" </b> Multiplication and division are implemented with tables. The tables consume quite a bit of memory
(2<sup>w</sup> &#215 2 <sup>w</sup> &#215  <sup>w</sup>/
8  bytes), so they are most useful when <em>w</em> is small. Please see <b>"SSE," "LAZY," "DOUBLE"</b> and

<b>"QUAD"</b> under region techniques below for further modifications to <b>"TABLE"</b> to perform <b>multiply_region()</b></li><br>


<li> <b>"LOG:"</b> This employs discrete (or "Zeph") logarithm <b>tables</b> to implement multiplication and division. The
memory usage is roughly (3 &#215 2<sup>w</sup> &#215 w /
8  bytes), so they are most useful when w is small, but they tolerate
larger <em>w</em> than <b>"TABLE."</b> If the polynomial is not primitive (see section 6.1.2), then you cannot use <b>"LOG"</b> as
an implementation. In that case,<b> gf_init_hard()</b> or <b>create_gf_from_argv()</b> will fail</li><br>


<li><b> "LOG_ZERO:"</b> Discrete logarithm tables which include extra room for zero entries. This more than doubles
the memory consumption to remove an <b>if</b> statement (please see [GMS08] or The Paper for more description). It
doesn’t really make a huge deal of difference in performance</li><br>

<li> <b>"LOG_ZERO_EXT:"</b> This expends even more memory to remove another <b>if</b> statement. Again, please see The
Paper for an explanation. As with <b>"LOG_ZERO,"</b> the performance difference is negligible</li><br>

<li> <b>"SHIFT:"</b> Implementation straight from the definition of Galois Field multiplication, by shifting and XOR-ing,
then reducing the product using the polynomial. This is <em>slooooooooow,</em> so we don’t recommend you use it</li><br>


<li> <b>"CARRY_FREE:"</b> This is identical to <b>"SHIFT,"</b> however it leverages the SSE instruction PCLMUL to perform
carry-freemultiplications in single instructions. As such, it is the fastest way to perform multiplication for large
values of <em>w</em> when that instruction is available. Its performance depends on the polynomial used. See The Paper
for details, and see section 7.7 below for the speedups available when <em>w </em>= 16 and <em>w</em> = 32 if you use a different
polynomial than the default one</li><br>


<li> <b>"BYTWO_p:"</b> This implements multiplication by successively multiplying the product by two and selectively
XOR-ing the multiplicand. See The Paper for more detail. It can leverage Anvin’s optimization that multiplies
64 and 128 bits of numbers in <em>GF(2<sup>w</sup>) </em> by two with just a few instructions. The SSE version requires SSE2</li><br>


<li> <b>"BYTWO_b:"</b> This implements multiplication by successively multiplying the multiplicand by two and selectively
XOR-ing it into the product. It can also leverage Anvin's optimization, and it has the feature that when
you're multiplying a region by a very small constant (like 2), it can terminate the multiplication early. As such,
if you are multiplying regions of bytes by two (as in the Linux RAID-6 Reed-Solomon code [Anv09]), this is
the fastest of the techniques, regardless of the value of <em>w.</em> The SSE version requires SSE2</li><br>


<li> <b>"SPLIT:"</b> Split multiplication tables (like the LR tables in [GMS08], or the SIMD tables for w &#8804 8 in [LHy08,
Anv09, PGM13b]). This argument must be followed by two more arguments, w<sub>a</sub> and w<sub>b</sub>, which are the index
sizes of the sub-tables. This implementation reduces the size of the table from <b>"TABLE,"</b> but requires multiple
</li><br>






<br/>

6 &nbsp &nbsp  <em>  THE DEFAULTS     </em>   <span id="index_number">18 </span> <br><br><br>
<ul>
table lookups. For example, the following multiplies 100 and 200 in <em>GF(2<sup>8</sup>) </em> using two 4K tables, as opposed 
to one 64K table when you use <b>"TABLE:"</b><br><br>
<div id="number_spacing">
UNIX> ./gf_mult 100 200 8 -m SPLIT 8 4 - <br>
79<br>
UNIX><br><br>
</div>

See section 7.4 for additional information on the arguments to <b>"SPLIT."</b> The SSE version typically requires
SSSE3.<br><br>


<li> <b>"GROUP:"</b> This implements the "left-to-right comb" technique [LBOX12]. I'm afraid we don't like that name,
so we call it <b>"GROUP,"</b> because it performs table lookup on groups of bits for shifting (left) and reducing (right).
It takes two additional arguments - g<sub>s,</sub> which is the number of bits you use while shifting (left) and g<sub>r</sub>, which
is the number of bits you use while reducing (right). Increasing these arguments can you higher computational
speed, but requires more memory. SSE version exists only for <em> w </em> = 128 and it requires SSE4. For more
description on the arguments g<sub>s</sub> and g<sub>r</sub>, see section 7.5. For a full description of <b>"GROUP"</b> algorithm, please
see The Paper.
</li><br>

<li> <b>"COMPOSITE:"</b> This allows you to perform operations on a composite Galois Field, <em> GF((2<sup>l</sup>)<sup>k</sup>)</em> as described
in [GMS08], [LBOX12] and The Paper. The field size <em>w </em> is equal to <em>lk.</em> It takes one argument, which is <em>k,</em> and
then a specification of the base field. Currently, the only value of <em>k</em> that is supported is two. However, that may
change in a future revision of the library. </li><br>


In order to specify the base field, put appropriate flags after specifying <em>k.</em> The single dash ends the base field,
and after that, you may continue making specifications for the composite field. This process can be continued
for multiple layers of <b>"COMPOSITE."</b> As an example, the following multiplies 1000000 and 2000000
in <em>GF((2<sup>16</sup>)<sup>2</sup>),</em> where the base field uses <b>BYTWO_p</b> for multiplication: <br><br>
<center>./gf_mult 1000000 2000000 32 -m COMPOSITE 2 <span style="color:red">-m BYTWO_p - -</span> </center><br>

In the above example, the red text applies to the base field, and the black text applies to the composite field.
Composite fields have two defining polynomials - one for the composite field, and one for the base field. Thus, if
you want to change polynomials, you should change both. The polynomial for the composite field must be of the
form x<sup>2</sup>+sx+1, where s is an element of <em>GF(2<sup>k</sup>).</em> To change it, you specify s (in hexadecimal)with "-p." In the
example below, we multiply 20000 and 30000 in <em>GF((2<sup>8</sup>)<sup>2</sup>) </em>, setting s to three, and using x<sup>8</sup>+x<sup>4</sup>+x<sup>3</sup>+x<sup>2</sup>+1
as the polynomial for the base field: <br><br>

<center>./gf_mult 20000 30000 16 -m COMPOSITE 2 <span style="color:red">-p 0x11d </span> - -p 0x3 - </center> <br><br>

If you use composite fields, you should consider using <b>"ALTMAP"</b> as well. The reason is that the region
operations will go much faster. Please see section 7.6.<br><br>
As with changing the polynomial, when you use a composite field, <em> GF((2<sup>l</sup>)<sup>k</sup>)</em>, you are using a different field
than the "standard" field for <em> GF((2<sup>l</sup>)<sup>k</sup>)</em>. All Galois Fields are isomorphic to each other, so they all have the
desired properties; however, the fields themselves change when you use composite fields.<br><br>
</ul>
<p>
With the exception of <b>"COMPOSITE"</b>, only one multiplication technique can be provided for a given Galois
Field instance. Composite fields may use composite fields as their base fields, in which case the specification will be
recursive. </p>








<br/>

6 &nbsp &nbsp  <em>  THE DEFAULTS     </em>   <span id="index_number">19 </span> <br><br><br>

<h3>6.1.4 &nbsp &nbsp &nbsp Changing the Division Technique </h3>

There are two techniques for division that may be set with "-d". If "-d" is not specified, then appropriate defaults
are employed. For example, when the multiplication technique is <b>"TABLE,"</b> a table is created for division as well as
multiplication. When <b>"LOG"</b> is specified, the logarithm tables are used for division. With <b>"COMPOSITE,"</b> a special
variant of Euclid's algorithm is employed that performs division using multiplication and division in the base field.
Otherwise, Euclid's algorithm is used. Please see The Paper for a description of Euclid's algorithm applied to Galois
Fields.

<p>If you use "-d", you must also specify the multiplication technique with "-m." </p>
<p>To force Euclid's algorithm instead of the defaults, you may specify it with "-d EUCLID." If instead, you would
rather convert elements of a Galois Field to a binary matrix and find an element's inverse by inverting the matrix,
then specify "-d MATRIX." In all of our tests, <b>"MATRIX"</b> is slower than <b>"EUCLID." "MATRIX" </b> is also not defined
for <em>w </em> > 32.
</p>


<h3>6.1.5  &nbsp&nbsp&nbsp Changing the Region Technique </h3>
The following are the region multiplication options ("-r"):
<ul>
<li>
<b>"SSE:"</b> Use SSE instructions. Initialization will fail if the instructions aren't supported. Table 2 details the
multiplication techniques which can leverage SSE instructions and which versions of SSE are required. </li><br>

<center>
<div id="data1">
<table cellpadding="6" cellspacing="0" style="text-align:center;font-size:19px">
<tr>
<th>Multiplication <br> Technique</th><th>multiply() </th><th>multiply_region() </th><th>SSE Version </th><th>Comments</th>

</tr>
<tr>
<td><b>"TABLE"</b></td><td >- </td><td>Yes</td><td>SSSE3</td><td>Only for <em>GF(2<sup>4</sup>). </em></td>

<tr>
<td><b>"SPLIT"</b></td><td>-</td><td>Yes</td><td>SSSE3</td><td>Only when the second argument equals 4.</td>

<tr>
<td><b>"SPLIT"</b></td><td>- </td><td>Yes</td><td>SSE4</td><td>When <em>w </em> = 64 and not using <b>"ALTMAP".</b></td>

<tr>
<td><b>"BYTWO_p"</b></td><td>- </td><td>Yes</td><td>SSE2</td><td></td>

<tr>
<td><b>"BYTWO_p"</b></td><td>- </td><td>Yes</td><td>SSE2</td><td></td>

</table></div> <br><br>
Table 2: Multiplication techniques which can leverage SSE instructions when they are available.
</center> <br><br>












<li> <b>"NOSSE:"</b> Force non-SSE version </li><br>

<li> <b> "DOUBLE:"</b> Use a table that is indexed on two words rather than one. This applies only to <em>w  </em> = 4, where
the table is indexed on bytes rather than 4-bit quantities, and to <em>w </em> = 8, where the table is indexed on shorts
rather than bytes. In each case, the table lookup performs two multiplications at a time, which makes region
multiplication faster. It doubles the size of the lookup table. </li><br>

<li> <b>"QUAD:"</b> Use a table that is indexed on four words rather than two or one. This only applies to <em>w </em> = 4, where
the table is indexed on shorts. The "Quad" table may be lazily created or created ahead of time (the default). If
the latter, then it consumes 2<sup>4</sup> &#215 2<sup>16</sup> &#215 2 = 2 MB of memory. </li><br>

<li> <b> "LAZY:"</b> Typically it's clear whether tables are constructed upon initialization or lazily when a region operation
is performed. There are two times where it is ambiguous: <b>"QUAD"</b> when <em>w </em> = 4 and <b>"DOUBLE"</b> when <em>w </em> = 8.
If you don't specify anything, these tables are created upon initialization, consuming a lot of memory. If you
specify <b>"LAZY,"</b> then the necessary row of the table is created lazily when you call <b>"multiply_region().</b>
</li>

</ul>











<br/>

6 &nbsp &nbsp  <em>  THE DEFAULTS     </em>   <span id="index_number">20 </span> <br><br><br>
<ul>

<li> <b>"ALTMAP:"</b> Use an alternate mapping, where words are split across different subregions of memory. There
are two places where this matters. The first is when implementing "<b>SPLIT</b> <em>w </em> 4" using SSE when <em>w </em> > 8. In
these cases, each byte of the word is stored in a different 128-bit vector, which allows the implementation to
better leverage 16-byte table lookups. See section 7.4 for examples, and The Paper or [PGM13b] for detailed
explanations.<br><br> </li>

The second place where it matters is when using <b>"COMPOSITE."</b> In this case, it is advantageous to split each
memory region into two chunks, and to store half of each word in a different chunk. This allows us to call
<b>region_multiply() </b> recursively on the base field, which is <em>much </em> faster than the alternative. See Section 7.6 for
examples, and The Paper for an explanation.<br><br>

It is important to note that with <b>"ALTMAP,"</b> the words are not "converted" from a standard mapping to an
alternate mapping and back again. They are assumed to always be in the alternate mapping. This typically
doesn't matter, so long as you always use the same <b>"ALTMAP"</b> calls. Please see section 7.9 for further details
on <b>"ALTMAP,"</b> especially with respect to alignment.<br><br>

<li> <b>"CAUCHY:"</b> Break memory into <em>w </em> subregions and perform only XOR's as in Cauchy Reed-Solomon coding
[BKK<sup>+</sup>95] (also described in The Paper). This works for <em>any</em> value of <em>w </em> &#8804 32, even those that are not
powers of two. If SSE2 is available, then XOR's work 128 bits at a time. For <b>"CAUCHY"</b> to work correctly,
<em>size</em> must be a multiple of <em>w </em>.</li> </ul>



<p>It is possible to combine region multiplication options. This is fully supported as long as <b>gf_methods</b> has the combination
listed. If multiple region options are required, they should be specified independently (as flags for <b>gf_init_hard()</b>
and independent options for command-line tools and <b>create_gf_from_argv()).</b> </p>


<h3>6.2  &nbsp&nbsp&nbspDetermining Supported Techniques with gf_methods </h3>


The program <b>gf_methods</b> prints a list of supported methods on standard output. It is called as follows:<br><br>
<div id="number_spacing">
<center>./gf_methods <em>w </em> -BADC -LUMDRB <br><br> </center> </div>

The first argument is <em>w </em>, which may be any legal value of <em>w </em>. The second argument has the following flags: <br><br>
<ul>

<li> <b>"B:"</b> This only prints out "basic" methods that are useful for the given value of <em>w </em>. It omits <b>"SHIFT"</b> and other
methods that are never really going to be useful.</li><br>

<li> <b> "A:"</b> In constrast, this specifies to print "all" methods. </li><br>

<li> <b>"D:"</b> This includes the <b>"EUCLID"</b> and <b>"MATRIX"</b> methods for division. By default, they are not included. </li><br>

<li> <b>"C:"</b> This includes the <b>"CAUCHY"</b> methods for region multiplication. By default, it is not included.</li> <br>
</ul>
<p>
You may specify multiple of these as the second argument. If you include both <b>"B"</b> and <b>"A,"</b> then it uses the last
one specified. </p>
<p>
The last argument determines the output format of <b>gf_methods.</b> If it is <b>"L,"</b> then it simply lists methods. If it
is <b>"U,"</b> then the output contains <b>gf_unit</b> commands for each of the methods. For the others, the output contains
<b>gf_time_tool.sh</b> commands for <b>M </b>ultiplication,<b>D</b>ivision,<b>R</b>egion multiplications with multiple buffer sizes, and the
<b>B</b>est region multiplication. </p>
<p>
<b>gf_methods</b> enumerates combinations of flags, and calls <b>create_gf_from_argv()</b> to see if the combinations are
supported. Although it enumerates a large number of combinations, it doesn't enumerate all possible parameters for
<b>"SPLIT," "GROUP"</b> or <b>"COMPOSITE."</b> </p>

<p>Some examples of calling <b>gf_methods</b> are shown below in section 6.3.2. </p>







<br/>

6 &nbsp &nbsp  <em>  THE DEFAULTS     </em>   <span id="index_number">21 </span> <br><br><br>


<h3>6.3 Testing with <b>gf_unit </b>, <b>gf_time </b>, and time_tool.sh </h3>



<b>gf_unit </b> and <b>gf_time </b> may be used to verify that a combination of arguments works correctly and efficiently on your
platform. If you plan to stray from the defaults, it is probably best to run both tools to ensure there are no issues with
your environment. <b>gf_unit </b> will run a set of unit tests based on the arguments provided to the tool, and <b>gf_time </b> will
time Galois Field methods based on the provided arguments.<br>
The usage of gf_ unit is:<br><br>
<div id="number_spacing">
<b>gf_unit </b> w tests seed method<br><br> </div>
The usage of gf_ time is:<br><br>
<div id="number_spacing">
<b>gf_time </b> w tests seed buffer-size iterations method<br><br>
</div>



The seed is an integer- negative one uses the current time. The tests are specified by a listing of characters. The
following tests are supported (All are supported by <b>gf_time.</b> Only ', 'S' and 'R' are supported by <b>gf_unit</b>):<br><br>

<ul>
<li> <b>'M':</b> Single multiplications</li><br>
<li> <b> 'D':</b> Single divisions</li><br>
<li> <b> 'I':</b> Single inverses</li><br>
<li> <b>'G': </b> Region multiplication of a buffer by a random constant</li><br>
<li> <b>'0': </b> Region multiplication of a buffer by zero (does nothing and<b>bzero()</b>)</li><br>
<li> <b>'1': </b> Region multiplication of a buffer by one (does <b>memcpy()</b> and <b>XOR</b>)</li><br>
<li> <b>'2': </b> Region multiplication of a buffer by two – sometimes this is faster than general multiplication</li><br>
<li> <b>'S':</b> All three single tests</li><br>
<li> <b>'R':</b> All four region tests</li><br>
<li> <b>'A':</b> All seven tests</li><br>
</ul>





<p>Here are some examples of calling <b>gf_unit</b> and <b>gf_time</b> to verify that <b>"-m SPLIT 32 4 -r ALTMAP -"</b> works
in <em>GF(2<sup>32</sup>),</em> and to get a feel for its performance. First, we go to the test directory and call <b>gf_unit:</b> </p><br><br>


<div id="number_spacing">
UNIX> cd test <br>
UNIX> ./gf_unit 32 A -1 -m SPLIT 32 4 -r ALTMAP - <br>
Args: 32 A -1 -m SPLIT 32 4 -r ALTMAP - / size (bytes): 684 <br>
UNIX> <br><br>
</div>

<b>gf_unit</b> reports on the arguments and how may bytes the <b>gf_t</b> consumes. If it discovers any problems or inconsistencies
with multiplication, division or region multiplication, it will report them. Here, there are no problems.
Next, we move to the <b>tools</b> directory and run performance tests on a 10K buffer, with 10,000 iterations of each test:<br><br>


UNIX> cd ../tools <br>
UNIX> ./gf_time 32 A -1 10240 10000 -m SPLIT 32 4 -r ALTMAP -<br>
Seed: 1388435794 <br>
<div id="number_spacing">
<table cellpadding="0" cellspacing="25" style="font-size:19px,font-family: 'Roboto Condensed', sans-serif;
">

<tr>

<td>Multiply:</td> <td>4.090548 s</td> <td> Mops: </td> <td> 24.414 </td> <td>5.968 Mega-ops/s </td> </tr>
<tr><td>Divide:</td> <td> 37.794962 s </td> <td>Mops: </td> <td> 24.414 </td> <td>0.646 Mega-ops/s </td> </tr>
<tr><td>Inverse:</td> <td> 33.709875 s </td> <td> Mops: </td> <td> 24.414 </td> <td> 0.724 Mega-ops/s </td> </tr>
<tr><td>Region-Random: XOR: 0 </td> <td> 0.035210 s </td> <td> MB:</td> <td> 97.656 </td> <td> 2773.527 MB/s </td></tr>
<tr><td>Region-Random: XOR: 1 </td> <td> 0.036081 s</td> <td> MB:</td> <td> 97.656 </td> <td>2706.578 MB/s </td></tr>
<tr><td>Region-By-Zero:XOR: 0 </td> <td> 0.003199 s </tD> <td> MB: </td> <td>97.656 </td> <td> 30523.884 MB/s </td> </tr>
<tr><td>Region-By-Zero: XOR: 1 </td> <td> 0.000626 s  </td> <td>MB: </td> <td> 97.656 </td> <td> 156038.095 MB/s </td></tr>

</table>
</div>










<br/>

6 &nbsp &nbsp  <em>  THE DEFAULTS     </em>   <span id="index_number">22 </span> <br><br><br>

<div id="number_spacing">
<table cellpadding="0" cellspacing="10" style="font-family: 'Roboto Condensed', sans-serif;
">

<tr>
<td>Region-By-One: XOR: 0</td> <td> 0.003810 s</td> <td> MB:</td> <td> 97.656 </td> <td> 25628.832 MB/s </td>
<tr><td>Region-By-One: XOR: 1 </td> <td> 0.008363 s </td> <td> MB:</td> <td> 97.656 </tD> <td>11677.500 MB/s </td></tr>

<tr><td>Region-By-Two: XOR: 0 </td> <td>0.032942 s  </td> <td>MB: </td> <td> 97.656 </td> <td> 2964.486 MB/s </td> </tr>
<tr><td>Region-By-Two: XOR: 1 </td> <td> 0.033488 s </td> <td> MB: </td> <td> 97.656 </td> <td> 2916.153 MB/s </td> </tr> </table>
</div>
UNIX><br><br>

<p>The first column of output displays the name of the test performed. Region tests will test with and without the XOR
flag being set (see Section 4.3 for an example). The second column displays the total time the test took to complete
measured in seconds (s). The third column displays the size of the test measured in millions of operations (Mops) for
single tests and in Megabytes (MB) for the region tests. The final column displays the speed of the tests calculated
from the second and third columns, and is where you should look to get an idea of a method's performance.</p>
<p>
If the output of <b>gf_unit</b> and <b>gf_time</b> are to your satisfaction, you can incorporate the method into application code
using create <b>gf_from_argv()</b> or <b>gf_init hard().</b></p>
<p>
The performance of "Region-By-Zero" and "Region-By-One" will not change from test to test, as all methods make
the same calls for these. "Region-By-Zero" with "XOR: 1" does nothing except set up the tests. Therefore, you may
use it as a control.</p>

<h3>6.3.1 &nbsp &nbsp &nbsp time_tool.sh </h3> 

Finally, the shell script <b>time_tool.sh</b> makes a bunch of calls to <b>gf_time</b> to give a rough estimate of performance. It is
called as follows:<br><br>
usage sh time_tool.sh M|D|R|B w method<br><br>


<p>The values for the first argument are <b>MDRB,</b> for <b>M</b>ultiplication, <b>D</b>ivision,<b>R</b>egion multiplications with multiple
buffer sizes, and the <b>B</b>est region multiplication. For the example above, let's call <b>time_tool.sh</b> to get a rough idea of
performance: </p><br><br>

<div id="number_spacing">
UNIX> sh time_tool.sh M 32 -m SPLIT 32 4 -r ALTMAP - <br>
M speed (MB/s): 6.03 W-Method: 32 -m SPLIT 32 4 -r ALTMAP - <br>
UNIX> sh time_tool.sh D 32 -m SPLIT 32 4 -r ALTMAP - <br>
D speed (MB/s): 0.65 W-Method: 32 -m SPLIT 32 4 -r ALTMAP - <br>
UNIX> sh time_tool.sh R 32 -m SPLIT 32 4 -r ALTMAP - <br>

<table cellpadding="0" cellspacing="10" style="font-family: 'Roboto Condensed', sans-serif;
">

<tr>
<td>Region Buffer-Size:</td> <td> 16K (MB/s):</td> <td>3082.91</td><td> W-Method: 32 </td> <td>-m SPLIT 32 4 </td> <td>-r ALTMAP -</td> </tr>
<tr><td>Region Buffer-Size:</td> <td>32K (MB/s): </td> <td>3529.07 </td><td> W-Method: 32 </td> <td>-m SPLIT 32 4 </td> <td>-r ALTMAP -</td> </tr>
<tr><td>Region Buffer-Size:</td> <td>64K (MB/s): </td> <td> 3749.94</td><td> W-Method: 32 </td> <td>-m SPLIT 32 4 </td> <td>-r ALTMAP -</td> </tr>
<tr><td>Region Buffer-Size:</td> <td>128K (MB/s):</td> <td>3861.27 </td> <td>W-Method: 32 </td> <td>-m SPLIT 32 4 </td> <td>-r ALTMAP -</td> </tr>
<tr><td>Region Buffer-Size:</td> <td>512K (MB/s):</td> <td>3820.82 </td><td> W-Method: 32 </td> <td>-m SPLIT 32 4 </td> <td>-r ALTMAP -</td> </tr>
<tr><td>Region Buffer-Size:</td> <td>1M (MB/s):</td> <td>3737.41 </td><td> W-Method: 32 </td> <td>-m SPLIT 32 4 </td> <td>-r ALTMAP -</td>  </tr>
<tr><td>Region Buffer-Size:</td> <td>2M (MB/s):</td> <td>3002.90 </td><td> W-Method: 32 </td> <td>-m SPLIT 32 4 </td> <td>-r ALTMAP -</td> </tr>
<tr><td>Region Buffer-Size:</td> <td>4M (MB/s): </td><td>2760.77</td><td> W-Method: 32 </td> <td>-m SPLIT 32 4 </td> <td>-r ALTMAP -</td> </tr>
<tr><td>Region Best (MB/s):</td><td> 3861.27</td><td> W-Method: 32 </td> <td>-m SPLIT 32 4 </td> <td>-r ALTMAP -</td> </tr>
</table>

UNIX> sh time_tool.sh B 32 -m SPLIT 32 4 -r ALTMAP - <br>
Region Best (MB/s): 3929.09  W-Method: 32  -m SPLIT 32 4 -r ALTMAP -</br>
UNIX><br><br>
</div>
<p>
We say that <b>time_tool.sh </b>is "rough" because it tries to limit each test to 5 ms or less. Thus, the time granularity
is fine, which means that the numbers may not be as precise as they could be were the time granularity to be course.
When in doubt, you should make your own calls to <b>gf_time</b> with a lot of iterations, so that startup costs and roundoff
error may be minimized. </p>








<br/>

6 &nbsp &nbsp  <em>  THE DEFAULTS     </em>   <span id="index_number">23 </span> <br><br><br>

<h3>6.3.2 &nbsp &nbsp &nbsp An example of gf_methods and time_tool.sh </h3><br><br>
Let's give an example of how some of these components fit together. Suppose we want to explore the basic techniques
in <em>GF(2<sup>32</sup>).</em> First, let's take a look at what <b>gf_methods</b> suggests as "basic" methods: <br><br>
<div id="number_spacing">
UNIX> gf_methods 32 -B -L <br>
w=32: - <br>
w=32: -m GROUP 4 8 - <br>
w=32: -m SPLIT 32 4 - <br>
w=32: -m SPLIT 32 4 -r ALTMAP - <br>
w=32: -m SPLIT 32 8 - <br>
w=32: -m SPLIT 8 8 - <br>
w=32: -m COMPOSITE 2 - - <br>
w=32: -m COMPOSITE 2 - -r ALTMAP - <br>
UNIX> <br><br>
</div>


<p>

You'll note, this is on my old Macbook Pro, which doesn't support (PCLMUL), so <b>"CARRY_FREE"</b> is not included
as an option. Now, let's run the unit tester on these to make sure they work, and to see their memory consumption: </p><br><br>

<div id="number_spacing">
UNIX> gf_methods 32 -B -U <br>
../test/gf_unit 32 A -1 - <br>
../test/gf_unit 32 A -1 -m GROUP 4 8 - <br>
../test/gf_unit 32 A -1 -m SPLIT 32 4 - <br>
../test/gf_unit 32 A -1 -m SPLIT 32 4 -r ALTMAP - <br>
../test/gf_unit 32 A -1 -m SPLIT 32 8 - <br>
../test/gf_unit 32 A -1 -m SPLIT 8 8 - <br>
../test/gf_unit 32 A -1 -m COMPOSITE 2 - - <br>
../test/gf_unit 32 A -1 -m COMPOSITE 2 - -r ALTMAP - <br>
UNIX> gf_methods 32 -B -U | sh <br>
Args: 32 A -1 - / size (bytes): 684 <br>
Args: 32 A -1 -m GROUP 4 8 - / size (bytes): 1296 <br>
Args: 32 A -1 -m SPLIT 32 4 - / size (bytes): 684 <br>
Args: 32 A -1 -m SPLIT 32 4 -r ALTMAP - / size (bytes): 684 <br>
Args: 32 A -1 -m SPLIT 32 8 - / size (bytes): 4268 <br>
Args: 32 A -1 -m SPLIT 8 8 - / size (bytes): 1839276 <br>
Args: 32 A -1 -m COMPOSITE 2 - - / size (bytes): 524648 <br>
Args: 32 A -1 -m COMPOSITE 2 - -r ALTMAP - / size (bytes): 524648 <br>
UNIX> <br> <br>
</div>
<p>
As anticipated, <b>"SPLIT 8 8"</b> consumes quite a bit of memory! Now, let's see how well they perform with both
single multiplications and region multiplications: </p> <br><br>
<div id="number_spacing">
UNIX> gf_methods 32 -B -M <br>
sh time_tool.sh M 32 - <br>
sh time_tool.sh M 32 -m GROUP 4 8  - <br>
sh time_tool.sh M 32 -m SPLIT 32 4 - <br>
sh time_tool.sh M 32 -m SPLIT 32 4 -r ALTMAP -<br>
sh time_tool.sh M 32 -m SPLIT 32 8 - <br>
sh time_tool.sh M 32 -m SPLIT 8 8 - <br>

</div>








<br/>

6 &nbsp &nbsp  <em>  THE DEFAULTS     </em>   <span id="index_number">24 </span> <br><br><br>

<div id="number_spacing">
sh time_tool.sh M 32 -m COMPOSITE 2 - <br>
sh time_tool.sh M 32 -m COMPOSITE 2 - -r ALTMAP <br>
UNIX> gf_methods 32 -B -M | sh
M speed (MB/s): 5.90 W-Method: 32 <br>
M speed (MB/s): 14.09 W-Method: 32 -m GROUP 4 8 <br>
M speed (MB/s): 5.60 W-Method: 32 -m SPLIT 32 4 <br>
M speed (MB/s): 5.19 W-Method: 32 -m SPLIT 32 4 -r ALTMAP <br>
M speed (MB/s): 5.98 W-Method: 32 -m SPLIT 32 8 <br>
M speed (MB/s): 22.10 W-Method: 32 -m SPLIT 8 8 <br>
M speed (MB/s): 34.98 W-Method: 32 -m COMPOSITE 2 - <br>
M speed (MB/s): 34.16 W-Method: 32 -m COMPOSITE 2 - -r ALTMAP <br>
UNIX> gf_methods 32 -B -B | sh
Region Best (MB/s): 2746.76 W-Method: 32 <br>
Region Best (MB/s): 177.06 W-Method: 32 -m GROUP 4 8 <br>
Region Best (MB/s): 2818.75 W-Method: 32 -m SPLIT 32 4 <br>
Region Best (MB/s): 3818.21 W-Method: 32 -m SPLIT 32 4 -r ALTMAP <br>
Region Best (MB/s): 728.68 W-Method: 32 -m SPLIT 32 8 <br>
Region Best (MB/s): 730.97 W-Method: 32 -m SPLIT 8 8 <br>
Region Best (MB/s): 190.20 W-Method: 32 -m COMPOSITE 2 - <br>
Region Best (MB/s): 1837.99 W-Method: 32 -m COMPOSITE 2 - -r ALTMAP <br>
UNIX>
</div>
<p>
The default is quite a bit slower than the best performing methods for both single and region multiplication. So
why are the defaults the way that they are? As detailed at the beginning of this chapter, we strive for lower memory
consumption, so we don't use <b>"SPLIT 8 8,"</b> which consumes 1.75MB.We don't implement alternate fields by default,
which is why we don't use <b>"COMPOSITE."</b> Finally, we don't implement alternate mappings of memory by default,
which is why we don't use "<b>-m SPLIT 32 4 -r ALTMAP -.</b>"</p>

<p>Of course, you may change these defaults if you please.</p>
<p>
<b>Test question:</b> Given the numbers above, it would appear that <b>"COMPOSITE"</b> yields the fastest performance of
single multiplication, while "SPLIT 32 4" yields the fastest performance of region multiplication. Should I use two
gf_t's in my application – one for single multiplication that uses <b>"COMPOSITE,"</b> and one for region multiplication
that uses <b>"SPLIT 32 4?"</b></p>
<p>
The answer to this is "no." Why? Because composite fields are different from the "standard" fields, and if you mix
these two <b>gf_t</b>'s, then you are using different fields for single multiplication and region multiplication. Please read
section 7.2 for a little more information on this.</p>

<h3>6.4 &nbsp &nbsp &nbspCalling gf_init_hard()</h3>

We recommend that you use <b>create_gf_from_argv()</b> instead of <b>gf_init_hard().</b> However, there are extra things that
you can do with <b>gf_init_hard().</b> Here's the prototype:<br><br>
<div id="number_spacing">
int gf_init_hard(gf_t *gf<br>
<div style="padding-left:100px">
int w<br>
int mult_type<br>
int region_type<br>
int divide_type<br>
uint64_t prim_poly<br>
int arg1<br>
int arg2<br>
</div>
</div>








<br/>

6 &nbsp &nbsp  <em>  THE DEFAULTS     </em>   <span id="index_number">25 </span> <br><br><br>
<div id="number_spacing">
<div style="padding-left:100px">
GFP base_gf, <br>
void *scratch_memory); </div><br><br>


The arguments mult type, region type and divide type allow for the same specifications as above, except the
types are integer constants defined in gf_complete.h: <br><br>
typedef enum {GF_MULT_DEFAULT,<br>
<div style="padding-left:124px">
GF_MULT_SHIFT<br>
GF_MULT_CARRY_FREE<br>
GF_MULT_GROUP<br>
GF_MULT_BYTWO_p<br>
GF_MULT_BYTWO_b<br>
GF_MULT_TABLE<br>
GF_MULT_LOG_TABLE<br>
GF_MULT_LOG_ZERO<br>
GF_MULT_LOG_ZERO_EXT<br>
GF_MULT_SPLIT_TABLE<br>
GF_MULT_COMPOSITE } gf_mult_type_t;<br><br>

</div>

#define GF_REGION_DEFAULT (0x0)<br>
#define GF_REGION_DOUBLE_TABLE (0x1) <br>
#define GF_REGION_QUAD_TABLE (0x2) <br>
#define GF_REGION_LAZY (0x4) <br>
#define GF_REGION_SSE (0x8) <br>
#define GF_REGION_NOSSE (0x10) <br>
#define GF_REGION_ALTMAP (0x20) <br>
#define GF_REGION_CAUCHY (0x40) <br><br>
typedef enum { GF_DIVIDE_DEFAULT<br>
<div style="padding-left:130px">GF_DIVIDE_MATRIX<br>
GF_DIVIDE_EUCLID } gf_division_type_t;<br><br>
</div>
</div>
<p>
You can mix the region types with bitwise or. The arguments to <b>GF_MULT_GROUP,GF_MULT_SPLIT_TABLE</b>
and <b>GF_MULT_COMPOSITE</b> are specified in arg1 and arg2. <b>GF_MULT_COMPOSITE</b> also takes a base field
in <b>base_gf.</b> The base field is itself a <b>gf_t,</b> which should have been created previously with <b>create_gf_fro_argv(),</b>
<b>gf_init_easy()</b> or <b>gf_init_hard().</b> Note that this <b>base_gf</b> has its own <b>base_gf</b> member and can be a composite field
itself.</p>
<p>
You can specify an alternate polynomial in <b>prim_poly.</b> For <em>w </em>&#8804 32, the leftmost one (the one in bit position <em>w</em>) is
optional. If you omit it, it will be added for you. For <em>w </em> = 64, there's no room for that one, so you have to leave it off.
For <em>w </em>= 128, your polynomial can only use the bottom-most 64 bits. Fortunately, the standard polynomial only uses
those bits. If you set <b>prim_poly</b> to zero, the library selects the "standard" polynomial.
</p>
<p>
Finally, <b>scratch_memory</b> is there in case you don't want <b>gf_init_hard()</b> to call <b>malloc()</b>. Youmay call <b>gf_scratch_size()</b>
to find out how much extra memory each technique uses, and then you may pass it a pointer for it to use in <b>scratc_memory.</b>
If you set scratch memory to NULL, then the extra memory is allocated for you with <b>malloc().</b> If you use <b>gf_init_easy()</b>
or <b>create_gf_from_argv(),</b> or you use <b>gf_init_hard()</b> and set <b>scratch_memory</b> to <b>NULL,</b> then you should call <b>gf_free()</b>
to free memory. If you use <b>gf_init_hard()</b> and use your own <b>scratch_memory</b> you can still call <b>gf_free(),</b> and it will
not do anything.</p>
<p>
Both <b>gf_init_hard()</b> and <b>gf_scratch_size()</b> return zero if the arguments don't specify a valid <b>gf_t.</b> When that happens,
you can call <b>gf_error()</b> to print why the call failed.</p>








<br/>


6 &nbsp &nbsp  <em>  FURTHER INFORMATION ON OPTIONS AND ALGORITHMS     </em>   <span id="index_number">26  </span> <br><br><br>


<p>We'll give you one example of calling <b>gf_ init_hard().</b> Suppose you want to make a <b>gf_ init_hard()</b> call to be
equivalent to "-m SPLIT 16 4 -r SSE -r ALTMAP -" and you want to allocate the scratch space yourself. Then you'd
do the following:</p><br><br>

<div id="number_spacing">
gf_t gf; <br>
void *scratch; <br>
int size; <br>
size = gf_scratch_size(16, GF_MULT_SPLIT_TABLE,<br>
GF_REGION_SSE | GF_REGION_ALTMAP,<br>
GF_DIVIDE_DEFAULT,<br>
16, 4); <br>
if (size == 0) { gf_error(); exit(1); } /* It failed. That shouldn’t happen */<br>
scratch = (void *) malloc(size); <br>
if (scratch == NULL) { perror("malloc"); exit(1); } <br>
if (!gf_init_hard(&gf, 16, GF_MULT_SPLIT_TABLE, <br>
GF_REGION_SSE | GF_REGION_ALTMAP, <br>
GF_DIVIDE_DEFAULT,<br>
0, 16, 4, NULL, scratch)) { <br>
gf_error(); <br>
exit(1); <br>
} <br>

</div>


<h3>6.5 &nbsp   &nbsp   gf_size() </h3>

You can call <b>gf_size(gf_t *gf)</b> to learn the memory consumption of the <b>gf_t.</b> It returns all memory consumed by the
<b>gf_t,</b> including the <b>gf_t</b> itself, any scratch memory required by the gf_ t, and the memory consumed by the sub-field
if the field is <b>"COMPOSITE."</b> If you provided your own memory to <b>gf_init_hard(),</b> it does not report the size of
this memory, but what the size should be, as determined by <b>gf_scratch size(). gf_ unit() </b> prints out the return value of
<b>gf_size()</b> on the given field.


<h2>7 &nbsp Further Information on Options and Algorithms </h2>
<h3>
7.1 &nbsp Inlining Single Multiplication and Division for Speed </h3>

Obviously, procedure calls are more expensive than single instructions, and the mechanics of multiplication in <b>"TABLE"</b>
and <b>"LOG"</b> are pretty simple. For that reason, we support inlining for <b>"TABLE"</b> when <em>w </em> = 4 and <em>w </em> = 8, and
for <b>"LOG"</b> when <em>w </em> = 16. We elaborate below.
<p>
When <em>w </em> = 4, you may inline multiplication and division as follows. The following procedures return pointers to
the multiplication and division tables respectively: </p> <br><br>

<div id="number_spacing">
uint8_t *gf_w4_get_mult_table(gf_t * gf);<br>
uint8_t *gf_w4_get_div_table(gf_t * gf);<br><br>
</div>
<p>The macro <b>Gf_W4_INLINE_MULTDIV </b>(<em>table, a, b</em>) then multiplies or divides <em>a </em> by <em>b</em> using the given table. This
of course only works if the multiplication technique is <b>"TABLE,"</b> which is the default for <em>w </em> = 4. If the multiplication
technique is not <b>"TABLE,"</b> then <b>gf_w4_get_mult_table()</b> will return <b>NULL.</b></p>








<br/>


6 &nbsp &nbsp  <em>  FURTHER INFORMATION ON OPTIONS AND ALGORITHMS     </em>   <span id="index_number">27  </span> <br><br><br>




<p>When <em>w </em> = 8, the procedures <b>gf_w8_et_mult_table()</b> and <b>gf_ w8_get_div_table(),</b> and the macro </p>

<b>GF_W8_INLINE_MULTDIV </b>(<em>table, a, b</em>) work identically to the <em>w </em> = 4 case.

<p>When <em>w </em> = 16, the following procedures return pointers to the logarithm table, and the two inverse logarithm tables
respectively: </p><br>

<div id="number_spacing">
uint16_t *gf_w16_get_log_table(gf_t * gf); <br>
uint16_t *gf_w16_get_mult_alog_table(gf_t * gf);<br>
uint16_t *gf_w16_get_div_alog_table(gf_t * gf);<br>

</div>
<br>
<p>
The first inverse logarithm table works for multiplication, and the second works for division. They actually point
to the same table, but to different places in the table. You may then use the macro <b>GF_W16_INLINE_MULT</b>(<em>log,
alog, a, b </em>) to multiply <em>a</em> and <em>b</em>, and the macro <b>GF_W16_INLINE_DIV </b>(<em>log, alog, a, b </em>) to divide a and b. Make
sure you use the <em>alog</em> table returned by <b>gf_w16_get_mult_alog_table()</b> for multiplication and the one returned by
<b>gf_w16_get_div_alog_table()</b> for division. Here are some timings: </p> <br><br>


UNIX> gf_time 4 M 0 10240 10240 - <br>
Seed: 0 <br>
Multiply: 0.228860 s Mops: 100.000 436.949 Mega-ops/s <br>
UNIX> gf_inline_time 4 0 10240 10240 <br>
Seed: 0 <br>
Inline mult: 0.096859 s Mops: 100.000 1032.424 Mega-ops/s <br>
UNIX> gf_time 8 M 0 10240 10240 - <br>
Seed: 0 <br>
Multiply: 0.228931 s Mops: 100.000 436.812 Mega-ops/s <br>
UNIX> gf_inline_time 8 0 10240 10240 <br>
Seed: 0 <br>
Inline mult: 0.114300 s Mops: 100.000 874.889 Mega-ops/s <br>
UNIX> gf_time 16 M 0 10240 10240 - <br>
Seed: 0 <br>
Multiply: 0.193626 s Mops: 50.000 258.229 Mega-ops/s <br>
UNIX> gf_inline_time 16 0 10240 10240 <br>
Seed: 0 <br>
Inline mult: 0.310229 s Mops: 100.000 322.342 Mega-ops/s <br>
UNIX> <br> <br>

<h3>
7.2 &nbsp &nbsp Using different techniques for single and region multiplication </h3>


You may want to "mix and match" the techniques. For example, suppose you'd like to use "-m SPLIT 8 8" for
<b>multiply()</b> in <em>GF(2<sup>32</sup>),</em> because it's fast, and you don't mind consuming all of that space for tables. However, for
<b>multiply_region(),</b> you'd like to use "-m SPLIT 32 4 -r ALTMAP," because that's the fastest way to implement
<b>multiply_region().</b> Unfortunately, There is no way to create a <b>gf_t</b> that does this combination. In this case, you should
simply create two <b>gf_t's,</b> and use one for <b>multiply()</b> and the other for <b>multiply_region().</b> All of the implementations
may be used interchangably with the following exceptions:

<ul>
<li>
<b>"COMPOSITE"</b> implements a different Galois Field. </li><br>

<li>If you change a field's polynomial, then the resulting Galois Field will be different. </li>

</ul>








<br/>


6 &nbsp &nbsp  <em>  FURTHER INFORMATION ON OPTIONS AND ALGORITHMS     </em>   <span id="index_number">28  </span> <br><br><br>

<ul>
<li>

If you are using <b>"ALTMAP"</b> to multiply regions, then the contents of the resulting regions of memory will
depend on the multiplication technique, the size of the region and its alignment. Please see section 7.9 for a
detailed explanation of this. </li>

<li>If you are using <b>"CAUCHY"</b> to multiply regions, then like <b>"ALTMAP,"</b> the contents of the result regions of
memory the multiplication technique and the size of the region. You don't have to worry about alignment. </li>

<h3>7.3 &nbsp &nbsp General <em>w </em>  </h3>
The library supports Galois Field arithmetic with 2 < <em>w </em> &#8804 32. Values of <em>w </em> which are not whole number powers of
2 are handled by the functions in <b>gf_wgen.c</b> . For these values of <em>w </em>, the available multiplication types are <b>"SHIFT,"
"BYT<em>w </em>O p," "BYT<em>w </em>O b," "GROUP," "TABLE"</b> and <b>"LOG." "LOG" </b> is only valid for <em>w </em> < 28 and <b>"TABLE"</b>

is only valid for <em>w </em> < 15. The defaults for these values of <em>w </em> are <b>"TABLE"</b> for <em>w </em> < 8, <b>"LOG"</b> for <em>w </em> < 16, and
<b>"BYT<em>w </em>O p"</b> for <em>w </em> < 32.<br><br>

<h3>7.4 Arguments to "SPLIT" </h3>

The "SPLIT" technique is based on the distributive property of multiplication and addition: <br><br>
<center>
a * (b + c) = (a * b) + (a * c). </center>
<br>
This property allo<em>w </em>s us to, for example, split an eight bit <em>w </em>ord into t<em>w </em>o four-bit components and calculate the product
by performing t<em>w </em>o table lookups in 16-element tables on each of the compoents, and adding the result. There is much
more information on <b>"SPLIT"</b> in The Paper. Here <em>w </em>e describe the version of <b>"SPLIT"</b> implemented in GF-Complete.

<p>
<b>"SPLIT"</b> takes t<em>w </em>o arguments, <em>w </em>hich are the number of bits in each component of a, <em>w </em>hich <em>w </em>e call <em>w </em><sub>a</sub>, and the
number of bits in each component of b, <em>w </em>hich <em>w </em>e call <em>w </em><sub>b.</sub> If the t<em>w </em>o differ, it does not matter <em>w </em>hich is bigger - the
library recognizes this and performs the correct implementation. The legal values of <em>w </em><sub>a</sub> and <em>w </em><sub>b</sub> fall into five categories:
</p><br>


<ol>
<li>
 <em>w </em><sub>a</sub> is equal to <em>w </em> and <em>w </em><sub>b</sub> is equal to four. In this case, b is broken up into <em>w </em>/4
four-bit <em>w </em>ords <em>w </em>hich are used
in 16-element lookup tables. The tables are created on demand in <b>multiply_region()</b> and the SSSE3 instruction

<b>mm_shuffle_epi8()</b> is leveraged to perform 16 lookups in parallel. Thus, these are very fast implementations.
<em>w </em>hen <em>w </em> &#8805 16, you should combine this <em>w </em>ith <b>"ALTMAP"</b> to get the best performance (see The Paper
or [PGM13b] for explanation). If you do this please see section 7.9 for information about <b>"ALTMAP"</b> and
alignment.<br><br>


If you don't use <b>"ALTMAP,"</b> the implementations for <em>w </em> &#8712 {16, 32, 64} convert the standard representation into
<b>"ALTMAP,"</b> perform the multiplication <em>w </em>ith <b>"ALTMAP"</b> and then convert back to the standard representation.
The performance difference using <b>"ALTMAP"</b> can be significant: <br><br><br>

<div id="number_spacing">
<center>
<div id="table_page28">
<table cellpadding="6" cellspacing="0" style="text-align:center;font-size:19px">
<tr>
<td> gf_time 16 G 0 1048576 100 -m SPLIT 16 4 -</td> <td>Speed = 8,389 MB/s </td> 
</tr>
<tr>
<td>gf_time 16 G 0 1048576 100 -m SPLIT 16 4 -r ALTMAP - </td> <td>Speed = 8,389 MB/s </td> 
</tr>

<tr>
<td>gf_time 32 G 0 1048576 100 -m SPLIT 32 4 -</td> <td> Speed = 5,304 MB/s</td> 
</tr>
<tr>
<td>gf_time 32 G 0 1048576 100 -m SPLIT 32 4 -r ALTMAP -</td> <td> Speed = 7,146 MB/s</td> 
</tr>


<tr>
<td>gf_time 64 G 0 1048576 100 -m SPLIT 64 4 - </td> <td>Speed = 2,595 MB/s </td> 
</tr>

<tr>
<td>gf_time 64 G 0 1048576 100 -m SPLIT 64 4 -r ALTMAP - </td> <td>Speed = 3,436 MB/s </td> 
</tr>
</div>



</table>
</div>









<br/>


6 &nbsp &nbsp  <em>  FURTHER INFORMATION ON OPTIONS AND ALGORITHMS     </em>   <span id="index_number">29  </span> <br><br><br>

<ol style="list-style-type:none">


<li>2. &nbsp w<sub>a</sub> is equal to <em>w </em> and w<sub>b</sub> is equal to eight. Now, b is broken into bytes, each of these is used in its own 256-element
lookup table. This is typically the best w<sub>a</sub>y to perform <b>multiply_region()</b> without SSE.</li> 
Because this is a region optimization, when you specify these options, you get a default <b>multiply()</b>  see
Table 1 for a listing of the defaults. See section 7.2 for using a different <b>multiply()</b> than the defaults.<br><br>


<li>
3. &nbsp w<sub>a</sub> is equal to <em>w </em> and <em>w </em><sub>b</sub> is equal to 16. This is only valid for <em>w </em> = 32 and <em>w </em> = 64. No<em>w </em>, b is broken into shorts,
each of these is used in its own 64K-element lookup table. This is typically slower than when <em>w </em><sub>b</suB> equals 8, and
requires more amortization (larger buffer sizes) to be effective. </li><br>


<li>4. &nbsp <em>w </em><sub>a</sub> and <em>w </em><sub>b</sub> are both equal to eight. Now  both <em>a</em> and <em>b</em> are broken into bytes, 
and the products of the various bytes
are looked up in multiple 256 &#215 256 tables. In <em>GF(2<sup>16</sup>),</em> there are three of these tables. In <em>GF(232),</em> there are
seven, and in <em>GF(2<sup>64</sup>)</em> there are fifteen. Thus, this implementation can be a space hog. How ever, for <em>w </em> = 32,
this is the fastest way to perform <b>multiply()</b> on some machines.
when this option is employed, <b>multiply_region()</b> is implemented in an identical fashion to when <em>w </em><sub>a</sub> = <em>w </em>
and <em>w </em><sub>b</sub> = 8. </li><br>

<li>5.&nbsp w<sub>a</sub> = 32 and w<sub>b</sub> = 2. (<em>w</em> = 32 only). I was playing with a different way to use <b>mm_shuffle_epi8().</b> It works,
but it's slower than when w<sub>b</sub> = 4.
</li>

</ul>



<h2>7.5 &nbsp&nbsp Arguments to "GROUP" </h3>

The <b>"GROUP"</b> multiplication option takes t<em>w </em>o arguments, g<sub>s</sub> and g<sub>r</sub>. It implements multiplication in the same manner
as <b>"SHIFT,"</b> except it uses a table of size 2<sup>gs</sup> to perform g<sup>s</sup> shifts at a time, and a table of size 2<sup>gr</sup> to perform g<sup>r</sup>
reductions at at time. The program <b>gf_methods</b> only prints the options 4 4 and 4 8 as arguments for <b>"GROUP."</b>
However, other values of g<sub>s</sub> and g<sub>r</sub> are legal and sometimes desirable: <br><br>

<ol>
<li>
 For <em>w </em> &#8804 32 and <em>w </em> = 64, any values of g<sub>s</sub> and g<sub>r</sub> may be used, so long as they are less than or equal to <em>w </em> and so
long as the tables fit into memory. There are four exceptions to this, listed belo<em>w </em>. </li><br>
<li> For <em>w </em> = 4, <b>"GROUP"</b> is not supported. </li><br>
<li> For <em>w </em> = 8, <b>"GROUP"</b> is not supported. </li><br>
<li> For <em>w </em> = 16, <b>"GROUP"</b> is only supported for gs = gr = 4. </li><br>
<li> For <em>w </em> = 128 <b>"GROUP"</b> only supports <em>g<sub>s</sub></em> = 4 and <em> g<sub>r</b> </em> &#8712 {4, 8, 16}.</li><br>
</ol>
<p>
The way that gs and gr impact performance is as follows. The <b>"SHIFT"</b> implementation works by performing a
carry-free multiplication in <em>w </em> steps, and then performing reduction in <em>w </em> steps. In "GROUP," the carry-free multiplication
is reduced to  <em>w /</em>g<sub>s</sub>steps, and the reduction is reduced to <em>w /</em>g<sub>r</sub>

. Both require tables. The table for the carry-free
multiplication must be created at the beginning of each <b>multiply()</b> or <b>multiply_region(),</b> while the table for reduction
is created when the <b>gf_t</b> is initialized. For that reason, it makes sense for g<sub>r</sub> to be bigger than g<sub>s.</sub></p>

<p>
To give a flavor for the impact of these arguments, Figure 3 show </em>s the performance of varying g<sub>s</sub> and g<sub>r</sub> for
single multiplication and region multiplication respectively, in <em> GF(2<sup>32</sup>)</em> and <em>GF(2<sup>64</sup>).</em> As the graphs demonstrate,
<b>multiply()</b> performs better <em>w </em>ith smaller values of gs, <em>w </em>hile multiply region() amortizes the creation of the shifting
table, and can tolerate larger values of g<sub>s.</sub> <em>w </em>hen g<sub>s</sub> equals g<sub>r,</sub> there are some optimizations that we hand-encode.
These can be seen clearly in the <b>multiply_region()</b> graphs.
</p>








<br/>
7 &nbsp &nbsp  <em>   FURTHER INFORMATION ON OPTIONS AND ALGORITHMS     </em>   <span id="index_number">30 </span> 


<div id="box_1"> 
 
<div class="image-cell_3"> </div>

<div class="image-cell_4"> </div>
</div>
Figure 3: The performance of <b>multiply()</b> and <b>multiply_region()</b> using <b>"GROUP,"</b> and varying the arguments <br> g<sub>s</sub>
and g<sub>r.</sub> All graphs are heat maps with black equaling zero. The region size is 100KB.

<h3>7.6 &nbspConsiderations with "COMPOSITE" </h3>


As mentioned above, using <b>"ALTMAP"</b> with <b>"COMPOSITE"</b> allows <b>multiply_region()</b> to recursively call <b>multiply_
region(),</b> rather than simply calling <b>multiply()</b> on every word in the region. The difference can be pronounced:<br><br>

<div id="table_page28"><center>

<table cellpadding="6" cellspacing="0" style="text-align:center;font-size:19px"><tr>
<td>
gf_time 32 G 0 10240 10240 -m COMPOSITE 2 - -
Speed = 322 MB/s </td> </tr>
<tr>
<td>gf_time 32 G 0 10240 10240 -m COMPOSITE 2 - -r ALTMAP -
Speed = 3,368 MB/s </td> </tr>

<tr>
<td>
gf_time 32 G 0 10240 10240 -m COMPOSITE 2 -m SPLIT 16 4 -r ALTMAP - -r ALTMAP -
Speed = 3,925 MB/s </td> </tr>
</center>
</table>
</div>


<br><br>
<p>
There is support for performing <b>multiply()</b> inline for the <b>"TABLE"</b> implementations for w &#8712 {4, 8} and for the
"LOG" implementation for <em>w</em> = 16 (see section 7.1). These are leveraged by <b>multiply()</b> in <b>"COMPOSITE,"</b> and
by <b>multiply_region()</b> if you are not using <b>"ALTMAP."</b> To demonstrate this, in the table below, you can see that the
performance of <b>multiply()</b> with <b>"SPLIT 8 4"</b> is 88 percent as fast than the default in <em>w</em> = 8 (which is <b>"TABLE"</b>).
When you use each as a base field for <b>"COMPOSITE"</b> with <em>w</em> = 16, the one with <b>"SPLIT 8 4"</b> is now just 37 percent
as fast. The difference is the inlining of multiplication in the base field when <b>"TABLE"</b> is employed:</p><br><br>

<div id="table_page28" border="0"><center>

    <table cellpadding="6" cellspacing="0" style="text-align:center;font-size:19px">

      <tr><td>gf_time 8 M 0 1048576 100 - Speed = 501 Mega-ops/s</td> </tr>
      <tr><td>gf_time 8 M 0 1048576 100 -m SPLIT 8 4 - Speed = 439 Mega-ops/s </td> </tr>
      <tr><td>gf_time 8 M 0 1048576 100 -m COMPOSITE 2 - - Speed = 207 Mega-ops/s </td> </tr>
      <tr><td>gf_time 8 M 0 1048576 100 -m COMPOSITE 2 -m SPLIT 8 4 - - Speed = 77 Mega-ops/s </td> </tr>

    </table> 
    </center>
<br><br>
</div>

You can keep making recursive definitions of composites field if you want. For example, this one's not too slow for
region operations (641 MB/s):








<br/>
<br/>


6 &nbsp &nbsp  <em>  FURTHER INFORMATION ON OPTIONS AND ALGORITHMS     </em>   <span id="index_number">31  </span> <br><br><br>

<div id="number_spacing">
<center>
gf_time 128 G 0 1048576 100 -m COMPOSITE 2 <span style="color:red">-m COMPOSITE 2 </span> <span style="color:blue">-m COMPOSITE 2 </span> <br>
<span style="color:rgb(250, 149, 167)">-m SPLIT 16 4 -r ALTMAP -</span> <span style="color:blue">-r ALTMAP -</span> <span style="color:red"> -r ALTMAP -</span> -r ALTMAP -
</center>
</div><br>

<p>Please see section 7.8.1 for a discussion of polynomials in composite fields.</p>

<h2>7.7 &nbsp &nbsp &nbsp "CARRY_FREE" and the Primitive Polynomial </h2>


If your machine supports the PCLMUL instruction, then we leverage that in <b>"CARRY_FREE."</b> This implementation
first performs a carry free multiplication of two <em>w</em>-bit numbers, which yields a 2<em>w</em>-bit number. It does this with
one PCLMUL instruction. To reduce the 2<em>w</em>-bit number back to a <em>w</em>-bit number requires some manipulation of the
polynomial. As it turns out, if the polynomial has a lot of contiguous zeroes following its leftmost one, the number of
reduction steps may be minimized. For example, with <em>w </em> = 32, we employ the polynomial 0x100400007, because that
is what other libraries employ. This only has 9 contiguous zeros following the one, which means that the reduction
takes four steps. If we instead use 0x1000000c5, which has 24 contiguous zeros, the reduction takes just two steps.
You can see the difference in performance:
<br><br>
<center>
<div id="table_page28">

<table cellpadding="6" cellspacing="0" style="text-align:center;font-size:19px">
<tr>

<td>gf_time 32 M 0 1048576 100 -m CARRY_FREE - </td> <td> Speed = 48 Mega-ops/s</td> </tr>

<tr><td>gf_time 32 M 0 1048576 100 -m CARRY_FREE -p 0xc5 -</td> <td> Speed = 81 Mega-ops/s </td> </tr>

</table></center>
</div>
<br><br>

<p>
This is relevant for <em>w </em> = 16 and <em>w </em> = 32, where the "standard" polynomials are sub-optimal with respect to
<b>"CARRY_FREE."</b> For <em>w </em> = 16, the polynomial 0x1002d has the desired property. It’s less important, of course,
with <em>w </em> = 16, because <b>"LOG"</b> is so much faster than <b>CARRY_FREE.</b> </p>

<h2>7.8 &nbsp  More on Primitive Polynomials </h3>

<h3>7.8.1 &nbsp Primitive Polynomials that are not Primitive </h4>

The library is willing to work with most polynomials, even if they are not primitive or irreducible. For example, the
polynomial x<sup>4</sup> + x<sup>3</sup> +x<sup>2</sup> +x+1 is irreducible, and therefore generates a valid Galois Field for <em>GF(2<sup>4</sup>).</em> However, it
is not primitive, because 2<sup>5</sup> = 1. For that reason, if you use this polynomial, you cannot use the <b>"LOG"</b> method. The
other methods will work fine: <br><br>

<div id="number_spacing">

UNIX> gf_mult 2 2 4 -p 0xf -  <br>
4 <br>
UNIX> gf_mult 4 2 4 -p 0xf - <br>
8 <br>
UNIX> gf_mult 8 2 4 -p 0xf - <br>
15 <br>
UNIX> gf_mult 15  2 4 -p 0xf - <br>
1 <br>
UNIX> gf_div 1 15 4 -p 0xf - <br>
2 <br>
UNIX> gf_div 1 15 4 -p 0xf -m LOG - <br>
usage: gf_div a b w [method] - does division of a and b in GF(2&#710;w) <br>
Bad Method Specification: Cannot use Log tables because the polynomial is not primitive. <br>
UNIX>  <br>
</div>
<p>
If a polynomial is reducible, then it does not define a Galois Field, but instead a ring. GF-Complete attempts to
work here where it can; however certain parts of the library will not work:
</p>






<br/>


6 &nbsp &nbsp  <em>  FURTHER INFORMATION ON OPTIONS AND ALGORITHMS     </em>   <span id="index_number">32  </span> <br><br><br>
<ol>
<li>
Division is a best effort service. The problemis that often quotients are not unique. If <b>divide()</b> returns a non-zero
number, then that number will be a valid quotient, but it may be one of many. If the multiplication technique is
<b>"TABLE,"</b> then if a quotient exists, one is returned. Otherwise, zero is returned. Here are some examples - the
polynomial x<sup>4</sup> + 1 is reducible, and therefore produces a ring. Below, we see that with this polynomal, 1*6 = 6
and 14*6 = 6. Therefore, 6/6 has two valid quotients: 1 and 14. GF-Complete returns 14 as the quotient:</li><br>

<div id="number_spacing">
UNIX> gf_mult 1 6 4 -p 0x1 -<br>
6 <br>
UNIX> gf_mult 14 6 4 -p 0x1 - <br>
6 <br>
UNIX> gf_div 6 6 4 -p 0x1 - <br>
14 <br>
UNIX> <br><br>
</div>


<li>When <b>"EUCLID"</b> is employed for division, it uses the extended Euclidean algorithm for GCD to find a number's
inverse, and then it multiplies by the inverse. The problem is that not all numbers in a ring have inverses. For
example, in the above ring, there is no number <em>a</em> such that 6a = 1. Thus, 6 has no inverse. This means that even
though 6/6 has quotients in this ring, <b>"EUCLID"</b> will fail on it because it is unable to find the inverse of 6. It will
return 0:
</li><br>
<div id="number_spacing">
UNIX> gf_div 6 6 4 -p 0x1 -m TABLE -d EUCLID -<br>
0<br>
UNIX><br>
</div><br>

<li> Inverses only work if a number has an inverse. Inverses may not be unique. </li><br>

<li> <b>"LOG"</b> will not work. In cases where the default would be <b>"LOG,"</b> <b>"SHIFT"</b> is used instead. </li>
</ol>

<p>
Due to problems with division, <b>gf_unit</b> may fail on a reducible polynomial. If you are determined to use such a
polynomial, don't let this error discourage you.
</p>

<h3>7.8.2 Default Polynomials for Composite Fields </h3>

GF-Complete will successfully select a default polynomial in the following composite fields:
<ul>
<li> <em>w </em> = 8 and the default polynomial (0x13) is employed for <em>GF(2<sup>4</sup>)</em></li><br>
<li> w = 16 and the default polynomial (0x11d) is employed for <em>GF(2<sup>8</sup>)</em></li><br>
<li> <em>w </em> = 32 and the default polynomial (0x1100b) is employed for <em>GF(2<sup>16</sup>) </em></li><br>
<li> <em>w </em> = 32 and 0x1002d is employed for <em>GF(2<sup>16</sup>) </em></li><br>
<li> <em>w </em> = 32 and the base field for <em>GF(w<em>16</em>) </em> is a composite field that uses a default polynomial</li><br>
<li> <em>w </em> = 64 and the default polynomial (0x100400007) is employed for <em>GF(2<sup>32</sup>)</em></li><br>
<li> <em>w </em> = 64 and 0x1000000c5 is employed for <em>GF(2<sup>32</sup>) </em></li><br>
<li> <em>w </em> = 64 and the base field for <em>GF(w<sup>32</sup>) </em> is a composite field that uses a default polynomial</li><br>
<li> <em>w </em> = 128 and the default polynomial (0x1b) is employed for <em>GF(2<sup>64</sup>) </em></li><br>
<li> <em>w </em> = 128 and the base field for <em> GF(w<sup>64 </sup>) </em> is a composite field that uses a default polynomial</li><br>
</ul>








<br/>


6 &nbsp &nbsp  <em>  FURTHER INFORMATION ON OPTIONS AND ALGORITHMS     </em>   <span id="index_number">33  </span> <br><br><br>


<h3>7.8.3 The Program gf_poly for Verifying Irreducibility of Polynomials </h3>

The program <b>gf_poly</b> uses the Ben-Or algorithm[GP97] to determine whether a polynomial with coefficients in <em> GF(2<sup>w </sup>) </em>
is reducible. Its syntax is:<br><br>
<div id="number_spacing">
gf_poly w method power:coef power:coef ... 
</div>

<br>
<p>You can use it to test for irreducible polynomials with binary coefficients by specifying w = 1. For example, from
the discussion above, we know that x<sup>4</sup> +x+1 and x<sup>4</sup> +x<sup>3</sup> +x<sup>2</sup> +x+1 are both irreducible, but x<sup>4</sup> +1 is reducible.
<b>gf_poly</b> confirms:<p><br>

<div id="number_spacing">
UNIX> gf_poly 1 - 4:1 1:1 0:1 <br>
Poly: x&#710;4 + x + 1 <br>
Irreducible. <br>
UNIX> gf_poly 1 - 4:1 3:1 2:1 1:1 0:1 <rb>
Poly: x&#710;4 + x&#710;3 + x&#710;2 + x + 1 <br>
Irreducible. <br>
UNIX> gf_poly 1 - 4:1 0:1 r<br>
Poly: x&#710;4 + 1 <br>
Reducible. <br>
UNIX> <br>

</div>


<p>
For composite fields <em>GF((2<sup>l</sup>)<sup>2</sup>),</em> we are looking for a value s such that x<sup>2</sup> + sx + 1 is irreducible. That value
depends on the base field. For example, for the default field <em>GF(2<sup>32</sup>),</em> a value of <em>s</em> = 2 makes the polynomial
irreducible. However, if the polynomial 0xc5 is used (so that PCLMUL is fast - see section 7.7), then <em>s</em> = 2 yields a
reducible polynomial, but <em>s</em> = 3 yields an irreducible one. You can use <b>gf_poly</b> to help verify these things, and to help
define s if you need to stray from the defaults:</p> <br>

<div id="number_spacing">
UNIX> gf_poly 32 - 2:1 1:2 0:1<br>
Poly: x&#710;2 + (0x2)x + 1 <br>
Irreducible. <br>
UNIX> gf_poly 32 -p 0xc5 - 2:1 1:2 0:1 <br>
Poly: x&#710;2 + (0x2)x + 1 <br>
Reducible. <br>
UNIX> gf_poly 32 -p 0xc5 - 2:1 1:3 0:1 <br>
Poly: x&#710;2 + (0x3)x + 1 <br>
Irreducible. <br>
UNIX> <br>
</div>

<p>
<b>gf_unit</b> does random sampling to test for problems. In particular, it chooses a random a and a random b, multiplies
them, and then tests the result by dividing it by a and b. When w is large, this sampling does not come close to
providing complete coverage to check for problems. In particular, if the polynomial is reducible, there is a good
chance that <b>gf_unit</b> won't discover any problems. For example, the following <b>gf_unit</b> call does not flag any problems,
even though the polynomial is reducible.</p>
<br>
<div id="number_spacing">
UNIX> gf_unit 64 A 0 -m COMPOSITE 2 -p 0xc5 - -p 2 -<br>
UNIX>
</div>

<p>
How can we demonstrate that this particular field has a problem? Well, when the polynomial is 0xc5, we can factor
x<sup>2</sup> + 2x + 1 as (x + 0x7f6f95f9)(x + 0x7f6f95fb). Thus, in the composite field, when we multiply 0x17f6f95f9 by
0x17f6f95fb, we get zero. That's the problem:
</p>








<br/>


6 &nbsp &nbsp  <em>  FURTHER INFORMATION ON OPTIONS AND ALGORITHMS     </em>   <span id="index_number">34  </span> <br><br><br>

<div id="number_spacing">

UNIX> gf_mult 7f6f95f9 7f6f95fb 32h -p 0xc5 - <br>
1 <br>
UNIX> gf_mult 17f6f95f9 17f6f95fb 64h -m COMPOSITE 2 -p 0xc5 - -p 2 - <br>
0 <br>
UNIX> <br>

</div>

<h2>7.9 "ALTMAP" considerations and extract_word() </h2>

There are two times when you may employ alternate memory mappings:
<ol>
<li> When using <b>"SPLIT"</b> and w<sub>b</sub> = 4. </li>
<li> When using <b>"COMPOSITE."</b> </li>
</ol>

Additionally, by default, the <b>"CAUCHY"</b> region option also employs an alternate memory mapping.

<p>When you use alternate memory mappings, the exact mapping of words in <em> GF(2<sup>w </sup>) </em> to memory depends on the
situation, the size of the region, and the alignment of the pointers. To help you figure things out, we have included the
procedures <b>extract_word.wxx()</b> as part of the <b>gf_t</b> struct. This procedure takes four parameters: </p>
<ul>
<li>A pointer to the <b>gf_t.</b> </li>
<li> The beginning of the memory region. </li>
<li>The number of bytes in the memory region. </li>
<li>The desired word number: <em>n.</em> </li>
</ul>

<p>
It then returns the <em>n</em>-th word in memory. When the standard mapping is employed, this simply returns the <em>n</em>-
th contiguous word in memory. With alternate mappings, each word may be split over several memory regions, so
<b>extract_word()</b> grabs the relevant parts of each memory region to extract the word. Below, we go over each of the
above situations in detail. Please refer to Figure 2 in Section 5 for reference. </p>


<h3>7.9.1 Alternate mappings with "SPLIT" </h3>

The alternate mapping with <b>"SPLIT"</b> is employed so that we can best leverage <b>mm_shuffle_epi8().</b> Please read [PGM13b]
for details as to why. Consider an example when <em>w</em> = 16. In the main region of memory (the middle region in Figure
2), multiplication proceeds in units of 32 bytes, which are each broken into two 16-byte regions. The first region
holds the high bytes of each word in <em>GF(2<sup>16</sup>),</em> and the second region holds the low bytes.
Let's look at a very detailed example, from <b>gf_example_5.c.</b> This program makes the following call, where <b>gf</b> has

been initialized for <em>w</em> = 16, using <b>"SPLIT"</b> and <b>"ALTMAP:"</b><br><br>
<div id="number_spacing">
gf.multiply_region.w32(&gf, a, b, 0x1234, 30*2, 0);
</div><br>


<p>In other words, it is multiplying a region a of 60 bytes (30 words) by the constant 0x1234 in <em> GF(2<sup>16</sup>),</em> and placing
the result into <em>b.</em> The pointers <em>a</em> and <em>b</em> have been set up so that they are not multiples of 16. The first line of output
prints <em>a</em> and <em>b:</em></p><br>

a: 0x10010008c b: 0x10010015c <br><br>

As described in Section 5, the regions of memory are split into three parts:








<br/>


6 &nbsp &nbsp  <em>  FURTHER INFORMATION ON OPTIONS AND ALGORITHMS     </em>   <span id="index_number">35  </span> <br><br><br>


<ol>
<li> 4 bytes starting at 0x1001008c / 0x10010015c. </li>
<li> 32 bytes starting at 0x10010090 / 0x100100160. </li>
<li> 24 bytes starting at 0x100100b0 / 0x100100180. </li>

</ol>


<p>In the first and third parts, the bytes are laid out according to the standard mapping. However, the second part is
split into two 16-byte regions- one that holds the high bytes of each word and one that holds the low bytes. To help
illustrate, the remainder of the output prints the 30 words of <em>a</em> and <em>b</em> as they appear in memory, and then the 30 return
values of <b>extract_word.w32():</b> </p><br>

<div id="number_spacing">
<table cellspacing="6" style="text-align:right">

<tr>
<td></td> <td> 1</td> <td> 2 </td> <td> 3 </td> <td> 4</td> <td> 5 </td> <td> 6 </td> <td> 7</td> <td> 8 </td> <td> 9</td> </tr>
<tr>
<td>a:</td><td> 640b</td> <td> 07e5</td> <td> 2fba </td> <td> ce5d </td> <td> f1f9</td> <td> 3ab8</td> <td> c518 </td> <td> 1d97</td> <td> 45a7</td>
 <td> 0160</td> </tr>
 
<tr><td>b:</td> <td>1ba3</td><td> 644e</td> <td> 84f8</td> <td> be3c</td> <td> 4318</td> <td> 4905</td> <td> b2fb </td> <td> 46eb </td> <td> ef01 </td>
 <td>a503</td> 
</tr>
</table> 
 <br><br>
<table cellspacing="6" style="text-align:right">

<tr>
<td> 10</td> <td> 11 </td> <td> 12</td> <td> 13</td> <td> 14 </td> <td> 15 </td> <td> 16</td> <td> 17</td> <td>18</td> <td> 19 </td></tr>
<tr>
<td>a:</td><td> 3759</td> <td> b107</td> <td> 9660 </td> <td> 3fde </td> <td> b3ea</td> <td> 8a53</td> <td> 75ff </td> <td> 46dc</td> <td> c504</td>
 <td> 72c2</td> </tr>
 
<tr><td>b:</td> <td>da27</td><td> e166</td> <td> a0d2</td> <td> b3a2</td> <td> 1699</td> <td> 3a3e</td> <td> 47fb </td> <td> 39af </td> <td> 1314 </td>
 <td>8e76</td> 
</tr>
</table> 

<table cellspacing="6" style="text-align:right">
<br><br>
<tr>
<td> 20</td> <td> 21 </td> <td> 22</td> <td> 23</td> <td> 24 </td> <td> 25 </td> <td> 26</td> <td> 27</td> <td>28</td> <td> 29 </td></tr>
<tr>
<td>a:</td><td> b469</td> <td> 1b97</td> <td> e91d </td> <td> 1dbc </td> <td> 131e</td> <td> 47e0</td> <td> c11a </td> <td> 7f07</td> <td> 76e0</td>
 <td> fe86</td> </tr>
 
<tr><td>b:</td> <td>937c</td><td> a5db</td> <td> 01b7</td> <td> 7f5f</td> <td> 8974</td> <td> 05e1</td> <td> cff3 </td> <td> a09c </td> <td> de3c </td>
 <td>4ac0</td> 
</tr>
</table> 
<br><br>
<table cellspacing="6">


<tr><td>Word</td><td> 0:</td> <td>0x640b </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x1ba3 Word 15:</td> <td>0x4575 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0xef47</td></tr>     
<tr><td>Word</td> <td> 1:</td> <td>0x07e5 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x644e Word 16:</td> <td>0x60dc </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x03af</td></tr>
<tr><td>Word</td> <td> 2:</td> <td>0xba59 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0xf827 Word 17:</td> <td>0x0146 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0xa539 </td> </tr>
<tr><td>Word</td> <td>3:</td> <td>0x2f37 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x84da Word 18:</td> <td>0xc504 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x1314 </td> </tr>
<tr><td>Word</td> <td>4:</td> <td>0x5d07 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x3c66 Word 19:</td> <td>0x72c2 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x8e76 </td> </tr>
<tr><td>Word</td> <td>5:</td> <td>0xceb1 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0xbee1 Word 20:</td> <td>0xb469 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x937c </td> </tr>
<tr><td>Word</td> <td>6:</td> <td>0xf960 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x18d2 Word 21:</td> <td>0x1b97 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0xa5db </td> </tr>
<tr><td>Word</td> <td>7:</td> <td>0xf196 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x43a0 Word 22:</td> <td>0xe91d </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x01b7 </td> </tr>
<tr><td>Word</td> <td>8:</td> <td>0xb8de </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x05a2 Word 23:</td> <td>0x1dbc </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x7f5f </td> </tr>
<tr><td>Word</td> <td>9:</td> <td>0x3a3f </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x49b3 Word 24:</td> <td>0x131e </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x8974 </td> </tr>
<tr><td>Word</td> <td>10:</td> <td>0x18ea </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0xfb99 Word 25:</td> <td>0x47e0 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x05e1 </td> </tr>
<tr><td>Word</td> <td>11:</td> <td>0xc5b3 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0xb216 Word 26:</td> <td>0xc11a </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0xcff3  </td> </tr>
<tr><td>Word</td> <td>12:</td> <td>0x9753 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0xeb3e Word 27:</td> <td>0x7f07 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0xa09c  </td> </tr>
<tr><td>Word</td> <td>13:</td> <td>0x1d8a </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x463a Word 28:</td> <td>0x76e0 </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0xde3c  </td> </tr>
<tr><td>Word</td> <td>14:</td> <td>0xa7ff </td><td>*</td> <td>0x1234</td> <td>=</td> <td>0x01fb Word 29:</td> <td>0xfe86 <td>*</td> <td>0x1234</td> <td>=</td> <td>0x4ac0 </td> </tr>

</table>
</div>
<br>
In the first region are words 0 and 1, which are identical to how they appear in memory: 0x640b and 0x07e5. In
the second region are words 2 through 17. These words are split among the two sixteen-byte regions. For example,
word 2, which <b>extract_word()</b> reports is 0xba59, is constructed from the low byte in word 2 (0xba) and the low byte
in word 10 (0x59). Since 0xba59 * 0x1234 = 0xf827, we see that the low byte in word 2 of <em> b </em> is 0xf8, and the low byte
in word 10 is 0x27.
<p>When we reach word 22, we are in the third region of memory, and words are once again identical to how they
appear in memory.</p>

<p>While this is confusing, we stress that that so long as you call <b>multiply_region()</b> with pointers of the same alignment
and regions of the same size, your results with <b>ALTMAP</b> will be consistent. If you call it with pointers of </p>






<br/>


7 &nbsp &nbsp  <em>  FURTHER INFORMATION ON OPTIONS AND ALGORITHMS     </em>   <span id="index_number">36  </span> <br><br><br>

different alignments, or with different region sizes, then the results will not be consistent. To reiterate, if you don't use
<b>ALTMAP,</b> you don't have to worry about any of this - words will always be laid out contiguously in memory.
<p>
When <em>w</em> = 32, the middle region is a multiple of 64, and each word in the middle region is broken into bytes, each
of which is in a different 16-byte region. When <em>w</em> = 64, the middle region is a multiple of 128, and each word is
stored in eight 16-byte regions. And finally, when<em>w</em> = 128, the middle region is a multiple of 128, and each word is
stored in 16 16-byte regions.</p><br>

<h3>7.9.2 &nbsp Alternate mappings with "COMPOSITE" </h3>

With <b>"COMPOSITE,"</b> the alternate mapping divides the middle region in half. The lower half of each word is stored
in the first half of the middle region, and the higher half is stored in the second half. To illustrate, gf_example_6
performs the same example as gf_example_5, except it is using <b>"COMPOSITE"</b> in GF((2<sup>16</sup>)<sup>2</sup>), and it is multiplying
a region of 120 bytes rather than 60. As before, the pointers are not aligned on 16-bit quantities, so the region is broken
into three regions of 4 bytes, 96 bytes, and 20 bytes. In the first and third region, each consecutive four byte word is a
word in <em>GF(2<sup>32</sup>).</em> For example, word 0 is 0x562c640b, and word 25 is 0x46bc47e0. In the middle region, the low two
bytes of each word come from the first half, and the high two bytes come from the second half. For example, word 1
as reported by <b>extract_word()</b> is composed of the lower two bytes of word 1 of memory (0x07e5), and the lower two
bytes of word 13 (0x3fde). The product of 0x3fde07e5 and 0x12345678 is 0x211c880d, which is stored in the lower
two bytes of words 1 and 13 of <em>b.</em><br><br>

a: 0x10010011c b: 0x1001001ec

<br><br>

<div id="number_spacing">
<table cellspacing="6" style="text-align:right">

<tr>
<td></td> <td> 1</td> <td> 2 </td> <td> 3 </td> <td> 4</td> <td> 5 </td> <td> 6 </td> <td> 7</td> <td> 8 </td> <td> 9</td> </tr>
<tr>
<td>a:</td><td> 562c640b</td> <td> 959407e5</td> <td> 56592fba </td> <td> cbadce5d </td> <td> 1d1cf1f9</td> <td> 35d73ab8</td> <td> 6493c518 </td> <td> b37c1d97</td> 
<td> 8e4545a7</td>
 <td> c0d80160</td> </tr>
 
<tr><td>b:</td> <td>f589f36c</td><td> f146880d</td> <td> 74f7b349</td> <td> 7ea7c5c6</td> <td> 34827c1a</td> <td> 93cc3746</td> <td> bfd9288b </td>
 <td> 763941d1 </td> 
<td> bcd33a5d </td>
 <td>da695e64</td> 
</tr>
</table> 


<br><br>
<table cellspacing="6" style="text-align:right">

<tr>
<td> 10</td> <td> 11 </td> <td> 12</td> <td> 13</td> <td> 14 </td> <td> 15 </td> <td> 16</td> <td> 17</td> <td>18</td> <td> 19 </td></tr>
<tr>
<td>a:</td><td> 965b3759</td> <td> cb3eb107</td> <td> 1b129660 </td> <td> 95a33fde </td> <td> 95a7b3ea</td> <td> d16c8a53</td> <td> 153375ff </td> 
<td> f74646dc</td> <td> 35aac504</td>
 <td> 98f972c2</td> </tr>
 
<tr><td>b:</td> <td>fd70f125</td><td> 3274fa8f</td> <td> d9dd34ee</td> <td> c01a211c</td> <td> d4402403</td> <td> 8b55c08b</td> <td> da45f0ad </td> 
<td> 90992e18 </td> <td> b65e0902 </td>
 <td>d91069b5</td> 
</tr>
</table> 


<table cellspacing="6" style="text-align:right">
<br><br>
<tr>
<td> 20</td> <td> 21 </td> <td> 22</td> <td> 23</td> <td> 24 </td> <td> 25 </td> <td> 26</td> <td> 27</td> <td>28</td> <td> 29 </td></tr>
<tr>
<td>a:</td><td> 5509b469</td> <td> 7f8a1b97</td> <td> 3472e91d </td> <td> 9ee71dbc </td> <td> de4e131e</td> <td> 46bc47e0</td> <td> 5bc9c11a </td>
 <td> 931d7f07</td> <td> c85cfe86</td>
 <td> fe86</td> </tr>
 
<tr><td>b:</td> <td>fc92b8f5</td><td> edd59668</td> <td> b4bc0d90</td> <td> a679e4ce</td> <td> 1a98f7d0</td> <td> 6038765f</td> <td> b2ff333f </td> <td> e7937e49 </td> 
<td> fa5a5867 </td>
 <td>79c00ea2</td> 
</tr>
</table> 
<br><br>


<table cellspacing="6" style="text-align:right">


<tr><td>Word</td><td> 0:</td> <td>0x562c640b </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0xf589f36c Word 15:</td> <td>0xb46945a7 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0xb8f53a5d</td></tr>     
<tr><td>Word</td> <td> 1:</td> <td>0x3fde07e5 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0x211c880d Word 16:</td> <td>0x55098e45 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0xfc92bcd3</td></tr>
<tr><td>Word</td> <td> 2:</td> <td>0x95a39594 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0xc01af146 Word 17:</td> <td>0x1b970160 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0x96685e64 </td> </tr>
<tr><td>Word</td> <td>3:</td> <td>0xb3ea2fba </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0x2403b349 Word 18:</td> <td>0x7f8ac0d8 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0xedd5da69 </td> </tr>
<tr><td>Word</td> <td>4:</td> <td>0x95a75659 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0xd44074f7 Word 19:</td> <td>0xe91d3759 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0x0d90f125 </td> </tr>
<tr><td>Word</td> <td>5:</td> <td>0x8a53ce5d </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0xc08bc5c6 Word 20:</td> <td>0x3472965b </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0xb4bcfd70 </td> </tr>
<tr><td>Word</td> <td>6:</td> <td>0xd16ccbad </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0x8b557ea7 Word 21:</td> <td>0x1dbcb107 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0xe4cefa8f </td> </tr>
<tr><td>Word</td> <td>7:</td> <td>0x75fff1f9 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0xf0ad7c1a Word 22:</td> <td>0x9ee7cb3e </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0xa6793274 </td> </tr>
<tr><td>Word</td> <td>8:</td> <td>0x15331d1c </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0xda453482 Word 23:</td> <td>0x131e9660 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0xf7d034ee </td> </tr>
<tr><td>Word</td> <td>9:</td> <td>0x46dc3ab8 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0x2e183746 Word 24:</td> <td>0xde4e1b12 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0x1a98d9dd </td> </tr>
<tr><td>Word</td> <td>10:</td> <td>0xf74635d7 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0x909993cc Word 25:</td> <td>0x46bc47e0 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0x6038765f </td> </tr>
<tr><td>Word</td> <td>11:</td> <td>0xc504c518 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0x0902288b Word 26:</td> <td>0x5bc9c11a </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0xb2ff333f  </td> </tr>
<tr><td>Word</td> <td>12:</td> <td>0x35aa6493 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0xb65ebfd9 Word 27:</td> <td>0x931d7f07 </td><td>*</td> <td>0x12345678</td> <td>=</td> <td>0xe7937e49  </td> </tr>

</table>
</div>








<br/>


8 &nbsp &nbsp  <em>  THREAD SAFETY     </em>   <span id="index_number">37  </span> <br><br><br>
<div id="number_spacing">
<table cellpadding="6" cellspacing="0">
<tr>
<td>Word 13:</td> <td> 0x72c21d97</td> <td> *</td> <td> 0x12345678</td> <td> =</td> <td> 0x69b541d1</td> <td> Word 28:</tD>

<td> 0xd40676e0 </td> <td> * </td> <td> 0x12345678 </td> <td> = </td> <td> 0xfa5a5867 </td> </tr>

<tr><td>Word 14:</td> <td> 0x98f9b37c</td> <td> * </td> <td> 0x12345678 </td> <td> = </td> <td> 0xd9107639</td> <td> Word 29:</td>
<td> 0xc85cfe86</td> <td>*</td> <td> 0x12345678</td> <td> =</td> <td> 0x79c00ea2</td></tr>

</table>
</div><br>


<p>
As with <b>"SPLIT,"</b> using <b>multiply_region()</b> with <b>"COMPOSITE"</b> and <b>"ALTMAP"</b> will be consistent only if the
alignment of pointers and region sizes are identical. </p>


<h3>7.9.3 The mapping of "CAUCHY" </h3>

With <b>"CAUCHY,"</b> the region is partitioned into <em>w</em> subregions, and each word in the region is broken into <em>w</em> bits,
each of which is stored in a different subregion. To illustrate, <b>gf_example_7</b> multiplies a region of three bytes by 5
in <em>GF(2<sup>3</sup>)</em> using <b>"CAUCHY:"</b><br><br>

<div id="number_spacing">

UNIX> gf_example_7 <br>
a: 0x100100190 b: 0x1001001a0 <br><br>
a: 0x0b 0xe5 0xba <br>
b: 0xee 0xba 0x0b <br><br>
a bits: 00001011 11100101 10111010 <br>
b bits: 11101110 10111010 00001011<br><br>
Word 0: 3 * 5 = 4 <br>
Word 1: 5 * 5 = 7 <br>
Word 2: 2 * 5 = 1 <br>
Word 3: 5 * 5 = 7 <br>  
Word 4: 4 * 5 = 2 <br>
Word 5: 6 * 5 = 3 <br>
Word 6: 2 * 5 = 1 <br>
Word 7: 6 * 5 = 3 <br>
UNIX><br><br> </div>
<p>

The program prints the three bytes of a and b in hexadecimal and in binary. To see how words are broken up,
consider word 0, which is the lowest bit of each of the three bytes of a (and b). These are the bits 1, 1 and 0 in a, and
0, 0, and 1 in b. Accordingly, the word is 3 in a, and 3*5 = 4 in b. Similarly, word 7 is the high bit in each byte: 0, 1, 1
(6) in a, and 1, 1, 0 (3) in b.</p>
<p>With <b>"CAUCHY," multiply_region()</b>may be implemented exclusively with XOR operations. Please see [BKK<sup>+</sup>95]
for more information on the motivation behind <b>"CAUCHY."</b> </p>

<h2>8 &nbsp Thread Safety </h2>

Once you initialize a <b>gf_t,</b> you may use it wontonly in multiple threads for all operations except for the ones below.
With the implementations listed below, the scratch space in the <b>gf_t</b> is used for temporary tables, and therefore you
cannot call <b>region_multiply,</b> and in some cases <b>multiply</b> from multiple threads because they will overwrite each
others' tables. In these cases, if you want to call the procedures from multiple threads, you should allocate a separate
gf_t for each thread:
<ul>
<li>
 All "GROUP" implementations are not thread safe for either <b>region_multiply()</b> or <b> multiply().</b> Other than
<b>"GROUP," multiply() </b> is always thread-safe.

</li>
</ul>









<br/>


9 &nbsp &nbsp  <em>  LISTING OF PROCEDURES     </em>   <span id="index_number">38  </span> <br><br><br>
<ul>
<li>

For <em>w </em> = 4, <b>region_multiply.w32()</b> is unsafe in in "-m TABLE -r QUAD -r LAZY." </li><br>
<li> For <em>w </em> = 8, <b> region_multiply.w32()</b> is unsafe in in "-m TABLE -r DOUBLE -r LAZY."</li><br>
<li> For <em>w </em> = 16, <b>region_multiply.w32() </b> is unsafe in in "-m TABLE."</li><br>
<li> For <em>w </em> &#8712 {32, 64, 128}, all <b>"SPLIT"</b> implementations are unsafe for <b>region_multiply().</b> This means that if the
default uses <b>"SPLIT"</b> (see Table 1 for when that occurs), then <b>region_multiply()</b> is not thread safe.</li><br>
<li> The <b>"COMPOSITE"</b> operations are only safe if the implementations of the underlying fields are safe.</li>
</ul>

<h2>9 &nbspListing of Procedures </h2>

The following is an alphabetical listing of the procedures, data types and global variables for users to employ in
GF-complete.<br>

<ul>
<li> <b>GF_W16_INLINE_DIV()</b> in <b>gf_complete.h:</b> This is a macro for inline division when <em>w </em> = 16. See section 7.1.</li><br>
<li> <b>GF_W16_INLINE_MULT()</b> in <b>gf_complete.h:</b> This is a macro for inline multiplication when <em>w </em> = 16. See
section 7.1.</li><br>
<li> <b>GF_W4_INLINE_MULTDIV()</b> in <b>gf_complete.h:</b> This is a macro for inline multiplication/division when <em>w </em> =
4. See section 7.1.</li><br>

<li> <b>GF_W8_INLINE_MULTDIV()</b> in <b>gf_complete.h:</b> This is a macro for inline multiplication/division when <em>w </em> =
8. See section 7.1.</li><br>
<li> <b>MOA_Fill_Random_Region()</b> in <b>gf_rand.h:</b> Fills a region with random numbers.</li><br>
<li> <b>MOA_Random_128()</b> in <b>gf_rand.h:</b> Creates a random 128-bit number.</li><br>
<li> <b>MOA_Random_32()</b> in <b>gf_rand.h:</b> Creates a random 32-bit number. </li><br>
<li> <b>MOA_Random_64()</b> in <b>gf_rand.h:</b> Creates a random 64-bit number. </li><br>
<li> <b>MOA_Random_W()</b> in <b>gf_rand.h:</b> Creates a random w-bit number, where <em>w </em> &#8804 32. </li><br>
<li> <b>MOA_Seed()</b> in <b>gf_rand.h:</b> Sets the seed for the random number generator. </li><br>
<li> <b>gf_errno</b> in <b>gf_complete.h:</b> This is to help figure out why an initialization call failed. See section 6.1.</li><br>
<li> <b>gf_create_gf_from_argv()</b> in <b>gf_method.h:</b> Creates a gf_t using C style argc/argv. See section 6.1.1. </li><br>
<li> <b>gf_division_type_t</b> in <b>gf_complete.h:</b> the different ways to specify division when using <b>gf_init_hard().</b> See 
section 6.4. </li><br>
<li> <b>gf_error()</b> in <b>gf_complete.h:</b> This prints out why an initialization call failed. See section 6.1. </li><br>

<li> <b>gf_extract</b> in <b>gf_complete.h:</b> This is the data type of <b>extract_word()</b> in a gf_t. See section 7.9 for an example
of how to use extract word().</li>
</ul>





<br/>


9 &nbsp &nbsp  <em>  LISTING OF PROCEDURES     </em>   <span id="index_number">39  </span> <br><br><br>
<ul>
<li>
<b>gf_free()</b> in <b>gf_complete.h:</b> If <b>gf_init easy(), gf_init hard()</b> or <b>create_gf_from_argv()</b> allocated memory, this
frees it. See section 6.4. </li>

<li> <b>gf_func_a_b</b> in <b>gf_complete.h:</b> This is the data type of <b>multiply()</b> and <b>divide()</b> in a gf_t. See section 4.2 for
examples of how to use <b>multiply()</b> and <b>divide()</b></li><br>

<li> <b>gf_func_a_b</b> in <b>gf_complete.h:</b> This is the data type of <b>multiply()</b> and <b>divide()</b> in a <b>gf_t.</b> See section 4.2 for
examples of how to use <b>multiply()</b> and <b>divide()</b></li><br>

<li> <b>gf_func_a</b> in <b>gf_complete.h:</b> This is the data type of <b>inverse()</b> in a <b>gf_t</b></li><br>

<li> <b>gf_general_add()</b> in <b>gf_general.h:</b> This adds two <b>gf_general_t's </b></li><br>

<li> <b>gf_general_divide()</b> in <b>gf_general.h:</b> This divides two <b>gf_general t's </b></li><br>

<li> <b>gf_general_do_region_check() </b> in <b>gf_general.h:</b> This checks a region multiply of <b>gf_general_t's </b></li><br>

<li> <b>gf_general_do_region_multiply() </b> in <b>gf_general.h:</b> This does a region multiply of <b>gf_general_t's </b></li><br>

<li> <b>gf_general_do_single_timing_test()</b> in <b>gf_general.h:</b> Used in <b>gf_time.c </b></li><br>

<li> <b>gf_general_inverse() </b> in <b>gf_general.h:</b> This takes the inverse of a <b>gf_general_t </b></li><br>

<li> <b>gf_general_is_one() </b> in <b>gf_general.h:</b> This tests whether a <b>gf_general_t </b> is one</li><br>

<li> <b>gf_general_is_two() </b> in <b>gf_general.h:</b> This tests whether a <b>gf_general_t  </b>is two</li><br>

<li> <b>gf_general_is_zero() </b> in <b>gf_general.h:</b> This tests whether a <b>gf_general_t </b> is zero</li><br>

<li> <b>gf_general_multiply() </b> in <b>gf_general.h:</b> This multiplies two <b>gf_general_t's.</b> See the implementation of gf_mult.c

for an example</li><br>
<li> <b>gf_general_s_to_val() </b> in <b>gf_general.h:</b> This converts a string to a <b>gf_general t.</b> See the implementation of
gf_mult.c for an example</li><br>
<li> <b>gf_general_set_one() </b> in <b>gf_general.h:</b> This sets a <b>gf_general_t</b> to one</li><br>
<li> <b>gf_general_set_random()</b> in <b>gf_general.h:</b> This sets a <b>gf_general_t </b> to a random number</li><br>
<li> <b>gf_general_set_two() in </b><b>gf_general.h:</b> This sets a <b>gf_general_t </b> to two</li><br>
<li> <b>gf_general_set_up_single_timing_test() </b> in <b>gf_general.h:</b> Used in <b>gf_time.c</b></li><br>
<li> <b>gf_general_set_zero() in </b><b>gf_general.h:</b> This sets a <b>gf_general_t_to_zero</b></li><br>
<li> <b>gf_general_t_in .</b><b>gf_general.h:</b> This is a general data type for all values of w. See the implementation of gf_mult.c
for examples of using these</li><br>
<li> <b>gf_general_val_to_s()</b> in<b>gf_general.h:</b> This converts a <b>gf_general_t </b> to a string. See the implementation of
<b>gf_mult.c</b> for an example</li><br>

<li> <b>gf_init_easy()</b> in <b>gf_complete.h:</b> This is how you initialize a default <b>gf_t.</b> See 4.2 through 4.5 for examples of
calling <b>gf_init_easy()</b></li><br>
</ul>







<br/>


9 &nbsp &nbsp  <em>  LISTING OF PROCEDURES     </em>   <span id="index_number">40  </span> <br><br><br>

<ul>

<li><b>gf_init hard()</b> in <b>gf_complete.h: </b> This allows you to initialize a <b>gf_t</b> without using the defaults. See 6.4. We
recommend calling create <b>gf_from argv()</b> when you can, instead of <b>gf_ init_hard()</b></li><br>

<li> <b>gf_ mult_type_t </b> in <b>gf_complete.h: </b> the different ways to specify multiplication when using <b>gf_init hard()</b>. See
section 6.4</li><br>

<li> <b>gf_region_type_t</b> in <b>gf_complete.h: </b> the different ways to specify region multiplication when using <b>gf_init_hard()</b>.
See section 6.4</li><br>

<li> <b>gf_region_in</b> <b>gf_complete.h: </b> This is the data type of <b>multiply_region()</b> in a <b>gf_t.</b> See section 4.3 for an example
of how to use <b>multiply_region()</b></li><br>

<li> <b>gf_scratch_size()</b> in <b>gf_complete.h: </b> This is how you calculate how much memory a <b>gf_t</b> needs. See section 6.4.</li><br>

<li> <b>gf_size()</b> in <b>gf_complete.h: </b> Returns the memory consumption of a <b>gf_t.</b> See section 6.5.</li><br>

<li> <b>gf_ val_128_t</b> in <b>gf_complete.h: </b> This is how you store a value where <em>w </em> &#8804 128. It is a pointer to two 64-bit
unsigned integers. See section 4.4</li><br>


<li> <b>gf_val_32_t</b> in <b>gf_ complete.h: </b> This is how you store a value where <em>w </em> &#8804 32. It is equivalent to a 32-bit unsigned
integer. See section 4.2</li><br>

<li> <b>gf_ val_64_t</b> in <b>gf_complete.h: </b> This is how you store a value where <em>w </em> &#8804 64. It is equivalent to a 64-bit unsigned
integer. See section 4.5</li><br>

<li> <b>gf_w16_get_div_alog_table()</b> in <b>gf_ complete.h: </b> This returns a pointer to an inverse logarithm table that can be
used for inlining division when <em>w </em> = 16. See section 7.1</li><br>


<li> <b>gf_w16_get_log_table()</b> in <b>gf_complete.h: </b> This returns a pointer to a logarithm table that can be used for inlining
when <em>w </em> = 16. See section 7.1</li><br>


<li> <b>gf_w16_get_mult_alog_table()</b> in <b>gf_complete.h: </b> This returns a pointer to an inverse logarithm table that can be
used for inlining multiplication when <em>w </em> = 16. See section 7.1</li><br>


<li> <b>gf_ w4 get div table()</b> in <b>gf_complete.h: </b> This returns a pointer to a division table that can be used for inlining
when <em>w </em> = 4. See section 7.1</li><br>


<li> <b>gf_w4_get_mult_table()</b> in <b>gf_complete.h: </b> This returns a pointer to a multiplication table that can be used for
inlining when <em>w </em> = 4. See section 7.1</li><br>

<li> <b>gf_w8_get_div_table()</b> in <b>gf_complete.h: </b> This returns a pointer to a division table that can be used for inlining
when <em>w </em> = 8. See section 7.1</li><br>

<li> <b>gf_w8_get_mult_table()</b> in <b>gf_complete.h: </b> This returns a pointer to a multiplication table that can be used for
inlining when <em>w </em> = 8. See section 7.1</li><br>

</ul>









<br/>
10 &nbsp &nbsp  <em>TROUBLESHOOTING </em>   <span id="index_number">41  </span> <br><br><br>

<ul>
<li><b> SSE support.</b> Leveraging SSE instructions requires processor support as well as compiler support. For example,
the Mac OS 10.8.4 (and possibly earlier versions) default compile environment fails to properly compile
PCLMUL instructions. This issue can be fixed by installing an alternative compiler; see Section 3 for details</li><br>

<li> <b>Initialization segfaults.</b> You have to already have allocated your <b>gf_t</b> before you pass a pointer to it in
<b>bgf_init_easy()</b>, <b>create_gf_ from_argv()</b>, or <b>bgf_ini_hard()</b></li><br>


<li> <b>GF-Complete is slower than it should be.</b> Perhaps your machine has SSE, but you haven't specified the SSE
compilation flags. See section 3 for how to compile using the proper flags</li><br>


<li> <b>Bad alignment.</b> If you get alignment errors, see Section 5</li><br>

<li> <b>Mutually exclusive region types.</b> Some combinations of region types are invalid. All valid and implemented
combinations are printed by <b>bgf_methods.c </b></li><br>

<li><b>Incompatible division types.</b> Some choices of multiplication type constrain choice of divide type. For example,
<b>"COMPOSITE"</b> methods only allow the default division type, which divides by finding inverses (i.e.,
neither <b>"EUCLID"</b> nor <b>"MATRIX"</b> are allowed). For each multiplication method printed by <b>gf_methods.c,</b> the
corresponding valid division types are also printed</li><br>


<li><b> Arbitrary "GROUP" arguments.</b> The legal arguments to <b>"GROUP"</b> are specified in section 7.5</li><br>

<li> <b> Arbitrary "SPLIt" arguments.</b> The legal arguments to <b>"SPLIt"</b> are specified in section 7.4</li><br>

<li> <b>Threading problems.</b> For threading questions, see Section 8</li><br>

<li> <b>No default polynomial.</b> If you change the polynomial in a base field using <b>"COMPOSITE,"</b> then unless it is
a special case for which GF-Complete finds a default polynomial, you'll need to specify the polynomial of the
composite field too. See 7.8.2 for the fields where GF-Complete will support default polynomials</li><br>
<li> Encoding/decoding with different fields. Certain fields are not compatible. Please see section 7.2 for an
explanation</li><br>


<li> <b>"ALTMAP" is confusing.</b> We agree. Please see section 7.9 for more explanation.</li><br>

<li> <b>I used "ALTMAP" and it doesn't appear to be functioning correctly.</b> With 7.9, the size of the region and
its alignment both matter in terms of how <b>"ALTMAP"</b> performs <b>multiply_region()</b>. Please see section 7.9 for
detailed explanation</li><br>

<li><b>Where are the erasure codes?.</b> This library only implements Galois Field arithmetic, which is an underlying
component for erasure coding. Jerasure will eventually be ported to this library, so that you can have fast erasure
coding</li><br>
</ul>
<h2>11 &nbsp &nbsp Timings </h2>

We don't want to get too detailed with timing, because it is quite machine specific. However, here are the timings on
an Intel Core i7-3770 CPU running at 3.40 GHz, with 4 &#215 256 KB L2 caches and an 8MB L3 cache. All timings are
obtained with <b>gf_time</b> or <b>gf_inline_time,</b> in user mode with the machine dedicated solely to running these jobs.









<br/>
10 &nbsp &nbsp  <em>TROUBLESHOOTING </em>   <span id="index_number">41  </span> <br><br><br>

<div class="image-cell_5"> </div>
<center>Figure 4: Speed of doing single multiplications for w &#8712 {4, 8, 16}. </center>
<h2>11.1 &nbsp Multiply() </h2>

The performance of <b>multiply()</b> is displayed in Figures 4 for w &#8712 {4, 8, 16} and 5 for w &#8712 {32, 64, 128}. These
numbers were obtained by calling <b>gf_time</b> with the size and iterations both set to 10240. We plot the speed in megaops
per second.

<p>As would be anticipated, the inlined operations (see section 7.1) outperform the others. Additionally, in all
cases with the exception of <em>w</em> = 32, the defaults are the fastest performing implementations. With w = 32,
"CARRY_FREE" is the fastest with an alternate polynomial (see section 7.7). Because we require the defaults to
use a "standard" polynomial, we cannot use this implementation as the default. </p>

<h2>11.2 &nbsp Divide() </h2>

For the  <b>"TABLE"</b> and <b>"LOG"</b> implementations, the performance of division is the same as multiplication. This means
that for w &#8712 {4, 8, 16}, it is very fast indeed. For the other implementations, division is implemented with Euclid's
method, and is several factors slower than multiplication.
In Figure 6, we plot the speed of a few implementations of the larger word sizes. Compared to the <b>"TABLE"</b> and
<b>"LOG"</b> implemenations for the smaller word sizes, where the speeds are in the hundreds of mega-ops per second,
these are very slow. Of note is the <b>"COMPOSITE"</b> implementation for <em>w</em> = 32, which is much faster than the others








<br/>
10 &nbsp &nbsp  <em>TROUBLESHOOTING </em>   <span id="index_number">43  </span> <br><br><br>

<div class="image-cell_6"> </div>

<center>Figure 5: Speed of doing single multiplications for w &#8712 {32, 64, 128}. </center><br>

because it uses a special application of Euclid's method, which relies on division in <em>GF(2<sup>16</sup>),</em> which is very fast.<br><br>

<h3>11.3 &nbsp Multiply_Region() </h2>

Tables 3 through 8 show the performance of the various region operations. It should be noted that for <em>GF(2<sup>16 </sup>) </em>
through <em>GF(2<sup>128</sup>),</em> the default is not the fastest implementation of <b>multiply_region().</b> The reasons for this are outlined
in section 6
<p>
For these tables, we performed 1GB worth of <b>multiply_region()</b> calls for all regions of size 2i bytes for 10 &#8804 i &#8804
30. In the table, we plot the fastest speed obtained.</p>
<p>We note that the performance of <b>"CAUCHY"</b> can be improved with techniques from [LSXP13] and [PSR12].</p>









<br/>
<em>REFERENCES </em>   <span id="index_number">44  </span> <br><br><br>

<div class="image-cell_7"> </div>

<center>Figure 6: Speed of doing single divisions for w &#8712 {32, 64, 128}. </center><br>

<center>
<div id="data2">
<table cellpadding="2" cellspacing="0" style="text-align:center;font-size:19px">

<tr><th>Method</td> <th>Speed (MB/s)</td> </tr>

<tr><td>-m TABLE (Default) -</td> <td>11879.909</td> </tr>
<tr><td>-m TABLE -r CAUCHY -</td> <td>9079.712</td> </tr>
<tr><td>-m BYTWO_b -</td> <td>5242.400</td> </tr>
<tr><td>-m BYTWO_p -</td> <td>4078.431</td> </tr>
<tr><td>-m BYTWO_b -r NOSSE -</td> <td>3799.699</td> </tr>
<tr><td>-m TABLE -r QUAD -</td> <td>3014.315</td> </tr>

<tr><td>-m TABLE -r DOUBLE -</td> <td>2253.627</td> </tr>
<tr><td>-m TABLE -r NOSSE -</td> <td>2021.237</td> </tr>
<tr><td>-m TABLE -r NOSSE -</td> <td>1061.497</td> </tr>
<tr><td>-m LOG -</td> <td>503.310</td> </tr>


<tr><td>m SHIFT -</td> <td>157.749</td> </tr>
<tr><td>-m CARRY_FREE -</td> <td>86.202</td> </tr>
</div>
</table> <br><br>
</div> </center>
<center>Table 3: Speed of various calls to <b>multiply_region()</b> for <em>w</em> = 4. </center>

<h3>References </h3>

[Anv09] H. P. Anvin. The mathematics of RAID-6.<a href=""> http://kernel.org/pub/linux/kernel/people/hpa/
raid6.pdf,</a> 2009.<br><br>

[BKK<sup>+</sup>95] J. Blomer, M. Kalfane, M. Karpinski, R. Karp, M. Luby, and D. Zuckerman. An XOR-based erasureresilient
coding scheme. Technical Report TR-95-048, International Computer Science Institute, August
1995. <br><br>

[GMS08] K. Greenan, E. Miller, and T. J. Schwartz. Optimizing Galois Field arithmetic for diverse processor
architectures and applications. In MASCOTS 2008: <em>16th IEEE Symposium on Modeling, Analysis and
Simulation of Computer and Telecommunication Systems,</em> Baltimore, MD, September 2008.<br><br>


[GP97] S. Gao and D. Panario. Tests and constructions of irreducible polynomials over finite fields. In <em> Foundations
of Computational Mathematics,</em> pages 346–361. Springer Verlag, 1997.
















<br/>
<em>REFERENCES </em>   <span id="index_number">45  </span> <br><br><br>


<center>
<div id="data2">
<table cellpadding="2" cellspacing="0" style="text-align:center;font-size:19px">

<tr><th>Method</td> <th>Speed (MB/s)</td> </tr>
<tr><td>-m SPLIT 8 4 (Default)</td> <td>13279.146</td> </tr>
<tr><td>-m COMPOSITE 2 - -r ALTMAP -</td> <td>5516.588</td> </tr>
<tr><td>-m TABLE -r CAUCHY -</td> <td>4968.721</td> </tr>
<tr><td>-m BYTWO_b -</td> <td>2656.463</td> </tr>
<tr><td>-m TABLE -r DOUBLE -</td> <td>2561.225</td> </tr>
<tr><td>-m TABLE -</td> <td>1408.577</td> </tr>

<tr><td>-m BYTWO_b -r NOSSE -</td> <td>1382.409</td> </tr>
<tr><td>-m BYTWO_p -</td> <td>1376.661</td> </tr>
<tr><td>-m LOG_ZERO_EXT -</td> <td>1175.739</td> </tr>
<tr><td>-m LOG_ZERO -</td> <td>1174.694</td> </tr>


<tr><td>-m LOG -</td> <td>997.838</td> </tr>
<tr><td>-m SPLIT 8 4 -r NOSSE -</td> <td>885.897</td> </tr>


<tr><td>-m BYTWO_p -r NOSSE -</td> <td>589.520</td> </tr>
<tr><td>-m COMPOSITE 2 - -</td> <td>327.039</td> </tr>


<tr><td>-m SHIFT -</td> <td>106.115</td> </tr>

<tr><td>-m CARRY_FREE -</td> <td>104.299</td> </tr>


</div>
</table> <br><br>
</div> </center>
<center>Table 4: Speed of various calls to multiply region() for <em>w</em> = 4. </center><br><br>

[LBOX12] J. Luo, K. D. Bowers, A. Oprea, and L. Xu. Efficient software implementations of large finite fields
<em>GF(2<sup>n</sup>) </em> for secure storage applications.<em> ACM Transactions on Storage, 8(2),</em> February 2012.<br><br>

[LD00] J. Lopez and R. Dahab. High-speed software multiplication in f<sub>2<sup>m</sup></sub>. In <em>Annual International Conference
on Cryptology in India,</em> 2000.<br><br>

[LHy08] H. Li and Q. Huan-yan. Parallelized network coding with SIMD instruction sets. In <em>International Symposium
on Computer Science and Computational Technology,</em> pages 364-369. IEEE, December 2008.<br><br>

[LSXP13] J. Luo, M. Shrestha, L. Xu, and J. S. Plank. Efficient encoding schedules for XOR-based erasure codes.
<em>IEEE Transactions on Computing,</em>May 2013.<br><br>

[Mar94] G. Marsaglia. The mother of all random generators.<a href=""> ftp://ftp.taygeta.com/pub/c/mother.
c,</a> October 1994.<br>

[PGM13a] J. S. Plank, K. M. Greenan, and E. L. Miller. A complete treatment of software implementations of
finite field arithmetic for erasure coding applications. Technical Report UT-CS-13-717, University of
Tennessee, September 2013.<br><br>

[PGM13b] J. S. Plank, K. M. Greenan, and E. L. Miller. Screaming fast Galois Field arithmetic using Intel SIMD
instructions. In FAST-2013: <em>11th Usenix Conference on File and Storage Technologies,</em> San Jose, February
2013.<br><br>

[Pla97] J. S. Plank. A tutorial on Reed-Solomon coding for fault-tolerance in RAID-like systems.<em> Software -
Practice & Experience,</em> 27(9):995-1012, September 1997.












<br/>
<em>REFERENCES </em>   <span id="index_number">46  </span> <br><br><br>


<center>
<div id="data2">
<table cellpadding="2" cellspacing="0" style="text-align:center;font-size:19px">

<tr><th>Method</td> <th>Speed (MB/s)</td> </tr>
<tr><td>-m SPLIT 16 4 -r ALTMAP -</td> <td>10460.834</td> </tr>
<tr><td>-m SPLIT 16 4 -r SSE (Default) - </td> <td>8473.793</td> </tr>
<tr><td>-m COMPOSITE 2 - -r ALTMAP -</td> <td>5215.073</td> </tr>
<tr><td>-m LOG -r CAUCHY -</td> <td>2428.824</td> </tr>
<tr><td>-m TABLE -</td> <td>2319.129</td> </tr>
<tr><td>-m SPLIT 16 8 -</td> <td>2164.111</td> </tr>

<tr><td>-m SPLIT 8 8 -</td> <td>2163.993</td> </tr>
<tr><td>-m SPLIT 16 4 -r NOSSE -</td> <td>1148.810</td> </tr>
<tr><td>-m LOG -</td> <td>1019.896</td> </tr>
<tr><td>-m LOG_ZERO -</td> <td>1016.814</td> </tr>
<tr><td>-m BYTWO_b -</td> <td>738.879</td> </tr>
<tr><td>-m COMPOSITE 2 - -</td> <td>596.819</td> </tr>
<tr><td>-m BYTWO_p -</td> <td>560.972</td> </tr>
<tr><td>-m GROUP 4 4 -</td> <td>450.815</td> </tr>
<tr><td>-m BYTWO_b -r NOSSE -</td> <td>332.967</td> </tr>
<tr><td>-m BYTWO_p -r NOSSE -</td> <td>249.849</td> </tr>
<tr><td>-m CARRY_FREE -</td> <td>111.582</td> </tr>
<tr><td>-m SHIFT -</td> <td>95.813</td> </tr>


</div>
</table> <br><br>
</div> </center>
<center>Table 5: Speed of various calls to multiply region()  for <em>w</em> = 4. </center><br><br>

[PMG<sup>+</sup>13] J. S. Plank, E. L. Miller, K. M. Greenan, B. A. Arnold, J. A. Burnum, A. W. Disney, and A. C. McBride.
GF-Complete: A comprehensive open source library for Galois Field arithmetic. version 1.0. Technical
Report UT-CS-13-716, University of Tennessee, September 2013.<br><br>

[PSR12] J. S. Plank, C. D. Schuman, and B. D. Robison. Heuristics for optimizing matrix-based erasure codes for
fault-tolerant storage systems. In DSN-2012:<em> The International Conference on Dependable Systems and
Networks,</em> Boston, MA, June 2012. IEEE.<br><br>

[Rab89] M. O. Rabin. Efficient dispersal of information for security, load balancing, and fault tolerance. <em>Journal
of the Association for Computing Machinery,</em> 36(2):335-348, April 1989.









<br/>
<em>REFERENCES </em>   <span id="index_number">47  </span> <br><br><br>
<center>
<div id="data2">
<table cellpadding="2" cellspacing="0" style="text-align:center;font-size:19px">
<tr><th>Method</td> <th>Speed (MB/s)</td> </tr>
<tr>
 
<td>

-m SPLIT 32 4 -r SSE -r ALTMAP - <br>
-m SPLIT 32 4 (Default)  <br>
-m COMPOSITE 2 -m SPLIT 16 4 -r ALTMAP - -r ALTMAP - <br>
-m COMPOSITE 2 - -r ALTMAP -  <br>
-m SPLIT 8 8 - <br> 
-m SPLIT 32 8 - <br> 
-m SPLIT 32 16 - <br> 
-m SPLIT 8 8 -r CAUCHY <br> 
-m SPLIT 32 4 -r NOSSE <br> 
-m CARRY_FREE -p 0xc5 <br> 
-m COMPOSITE 2 - <br> 
-m BYTWO_b - <br> 
-m BYTWO_p - <br> 
-m GROUP 4 8 - <br> 
-m GROUP 4 4 - <br> 
-m CARRY_FREE - <br> 
-m BYTWO_b -r NOSSE - <br> 
-m BYTWO_p -r NOSSE - <br>
-m SHIFT - <br> 

</td>

<td>
7185.440 <br>
5063.966 <br>
 4176.440 <br>
3360.860 <br>
1345.678 <br>
1340.656 <br>
1262.676 <br>
1143.263  <br>
 480.859 <br>
393.185 <br>
332.964 <br>
309.971 <br>
258.623 <br>
242.076 <br>
227.399 <br>
226.785 <br>
143.403 <br>
111.956 <br>
52.295 <br>
</td>


</tr>

</div>
</table> <br><br>
</div> </center>
<center>Table 6: Speed of various calls to multiply region() <em>w</em> = 4. </center><br><br>

<center>
<div id="data2">
<table cellpadding="2" cellspacing="0" style="text-align:center;font-size:19px">
<tr><th>Method</td> <th>Speed (MB/s)</td> </tr>
<tr>
 
<td>
-m SPLIT 64 4 -r ALTMAP - <br>
-m SPLIT 64 4 -r SSE (Default) - <br>
-m COMPOSITE 2 -m SPLIT 32 4 -r ALTMAP - -r ALTMAP - <br>
-m COMPOSITE 2 - -r ALTMAP -  <br>
-m SPLIT 64 16 - <br>
-m SPLIT 64 8 -  <br>
-m CARRY_FREE -  <br>
-m SPLIT 64 4 -r NOSSE - <br>
-m GROUP 4 4 -  <br>
-m GROUP 4 8 -  <br>
-m BYTWO_b -  <br>
-m BYTWO_p -  <br>
-m SPLIT 8 8 - <br>
-m BYTWO_p -r NOSSE - <br>
-m COMPOSITE 2 - - <br>
-m BYTWO_b -r NOSSE - <br>
-m SHIFT - <br>

</td>

<td>3522.798 <br>
 2647.862 <br>
2461.572 <br>
1860.921 <br>
1066.490 <br>
998.461 <br>
975.290 <br>
545.479 <br>
230.137 <br>
153.947 <br>
144.052 <br>
124.538 <br>
98.892 <br>
77.912 <br>
77.522 <br>
36.391 <br>
25.282 <br>
</td>


</tr>

</div>
</table> <br><br>
</div> </center>
<center>Table 7: Speed of various calls to multiply region() for  <em>w</em> = 4. </center><br><br>













<br/>
<em>REFERENCES </em>   <span id="index_number">48  </span> <br><br><br>

<center>
<div id="data2">
<table cellpadding="2" cellspacing="0" style="text-align:center;font-size:19px">
<tr><th>Method</td> <th>Speed (MB/s)</td> </tr>
<tr>
 
<td>

-m SPLIT 128 4 -r ALTMAP - <br>
-m COMPOSITE 2 -m SPLIT 64 4 -r ALTMAP - -r ALTMAP - <br> 
-m COMPOSITE 2 - -r ALTMAP - <br> 
-m SPLIT 128 8 (Default) - <br>
-m CARRY_FREE -<br> 
-m SPLIT 128 4 -<br> 
-m COMPOSITE 2 - <br>
-m GROUP 4 8 -<br> 
-m GROUP 4 4 -<br> 
-m BYTWO_p -<br> 
-m BYTWO_b -<br> 
-m SHIFT -<br> 
</td>

<td>
1727.683 <br>
1385.693 <br>
1041.456 <br>
872.619 <br>
814.030 <br>
500.133  <br>
289.207 <br>
133.583 <br>
116.187 <br>
25.162 <br>
25.157 <br>
14.183 <br>
</td>


</tr>

</div>
</table> <br><br>
</div> </center>
<center>Table 8: Speed of various calls to multiply region() for <em>w</em> = 4. </center><br><br>