summaryrefslogtreecommitdiff
path: root/tune/tuneup.c
blob: fd2bf833a5ad797c6950796028eac737ba429e52 (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
/* Create tuned thresholds for various algorithms.

Copyright 1999-2003, 2005, 2006, 2008-2017 Free Software Foundation, Inc.

This file is part of the GNU MP Library.

The GNU MP Library is free software; you can redistribute it and/or modify
it under the terms of either:

  * the GNU Lesser General Public License as published by the Free
    Software Foundation; either version 3 of the License, or (at your
    option) any later version.

or

  * the GNU General Public License as published by the Free Software
    Foundation; either version 2 of the License, or (at your option) any
    later version.

or both in parallel, as here.

The GNU MP Library is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
for more details.

You should have received copies of the GNU General Public License and the
GNU Lesser General Public License along with the GNU MP Library.  If not,
see https://www.gnu.org/licenses/.  */


/* Usage: tuneup [-t] [-t] [-p precision]

   -t turns on some diagnostic traces, a second -t turns on more traces.

   Notes:

   The code here isn't a vision of loveliness, mainly because it's subject
   to ongoing changes according to new things wanting to be tuned, and
   practical requirements of systems tested.

   Sometimes running the program twice produces slightly different results.
   This is probably because there's so little separating algorithms near
   their crossover, and on that basis it should make little or no difference
   to the final speed of the relevant routines, but nothing has been done to
   check that carefully.

   Algorithm:

   The thresholds are determined as follows.  A crossover may not be a
   single size but rather a range where it oscillates between method A or
   method B faster.  If the threshold is set making B used where A is faster
   (or vice versa) that's bad.  Badness is the percentage time lost and
   total badness is the sum of this over all sizes measured.  The threshold
   is set to minimize total badness.

   Suppose, as sizes increase, method B becomes faster than method A.  The
   effect of the rule is that, as you look at increasing sizes, isolated
   points where B is faster are ignored, but when it's consistently faster,
   or faster on balance, then the threshold is set there.  The same result
   is obtained thinking in the other direction of A becoming faster at
   smaller sizes.

   In practice the thresholds tend to be chosen to bring on the next
   algorithm fairly quickly.

   This rule is attractive because it's got a basis in reason and is fairly
   easy to implement, but no work has been done to actually compare it in
   absolute terms to other possibilities.

   Implementation:

   In a normal library build the thresholds are constants.  To tune them
   selected objects are recompiled with the thresholds as global variables
   instead.  #define TUNE_PROGRAM_BUILD does this, with help from code at
   the end of gmp-impl.h, and rules in tune/Makefile.am.

   MUL_TOOM22_THRESHOLD for example uses a recompiled mpn_mul_n.  The
   threshold is set to "size+1" to avoid karatsuba, or to "size" to use one
   level, but recurse into the basecase.

   MUL_TOOM33_THRESHOLD makes use of the tuned MUL_TOOM22_THRESHOLD value.
   Other routines in turn will make use of both of those.  Naturally the
   dependants must be tuned first.

   In a couple of cases, like DIVEXACT_1_THRESHOLD, there's no recompiling,
   just a threshold based on comparing two routines (mpn_divrem_1 and
   mpn_divexact_1), and no further use of the value determined.

   Flags like USE_PREINV_MOD_1 or JACOBI_BASE_METHOD are even simpler, being
   just comparisons between certain routines on representative data.

   Shortcuts are applied when native (assembler) versions of routines exist.
   For instance a native mpn_sqr_basecase is assumed to be always faster
   than mpn_mul_basecase, with no measuring.

   No attempt is made to tune within assembler routines, for instance
   DIVREM_1_NORM_THRESHOLD.  An assembler mpn_divrem_1 is expected to be
   written and tuned all by hand.  Assembler routines that might have hard
   limits are recompiled though, to make them accept a bigger range of sizes
   than normal, eg. mpn_sqr_basecase to compare against mpn_toom2_sqr.

   Limitations:

   The FFTs aren't subject to the same badness rule as the other thresholds,
   so each k is probably being brought on a touch early.  This isn't likely
   to make a difference, and the simpler probing means fewer tests.

*/

#define TUNE_PROGRAM_BUILD  1   /* for gmp-impl.h */

#include "config.h"

#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#if HAVE_UNISTD_H
#include <unistd.h>
#endif

#include "gmp-impl.h"
#include "longlong.h"

#include "tests.h"
#include "speed.h"

#if !HAVE_DECL_OPTARG
extern char *optarg;
extern int optind, opterr;
#endif


#define DEFAULT_MAX_SIZE   1000  /* limbs */

#if WANT_FFT
mp_size_t  option_fft_max_size = 50000;  /* limbs */
#else
mp_size_t  option_fft_max_size = 0;
#endif
int        option_trace = 0;
int        option_fft_trace = 0;
struct speed_params  s;

struct dat_t {
  mp_size_t  size;
  double     d;
} *dat = NULL;
int  ndat = 0;
int  allocdat = 0;

/* This is not defined if mpn_sqr_basecase doesn't declare a limit.  In that
   case use zero here, which for params.max_size means no limit.  */
#ifndef TUNE_SQR_TOOM2_MAX
#define TUNE_SQR_TOOM2_MAX  0
#endif

mp_size_t  mul_toom22_threshold         = MP_SIZE_T_MAX;
mp_size_t  mul_toom33_threshold         = MUL_TOOM33_THRESHOLD_LIMIT;
mp_size_t  mul_toom44_threshold         = MUL_TOOM44_THRESHOLD_LIMIT;
mp_size_t  mul_toom6h_threshold         = MUL_TOOM6H_THRESHOLD_LIMIT;
mp_size_t  mul_toom8h_threshold         = MUL_TOOM8H_THRESHOLD_LIMIT;
mp_size_t  mul_toom32_to_toom43_threshold = MP_SIZE_T_MAX;
mp_size_t  mul_toom32_to_toom53_threshold = MP_SIZE_T_MAX;
mp_size_t  mul_toom42_to_toom53_threshold = MP_SIZE_T_MAX;
mp_size_t  mul_toom42_to_toom63_threshold = MP_SIZE_T_MAX;
mp_size_t  mul_toom43_to_toom54_threshold = MP_SIZE_T_MAX;
mp_size_t  mul_fft_threshold            = MP_SIZE_T_MAX;
mp_size_t  mul_fft_modf_threshold       = MP_SIZE_T_MAX;
mp_size_t  sqr_basecase_threshold       = MP_SIZE_T_MAX;
mp_size_t  sqr_toom2_threshold
  = (TUNE_SQR_TOOM2_MAX == 0 ? MP_SIZE_T_MAX : TUNE_SQR_TOOM2_MAX);
mp_size_t  sqr_toom3_threshold          = SQR_TOOM3_THRESHOLD_LIMIT;
mp_size_t  sqr_toom4_threshold          = SQR_TOOM4_THRESHOLD_LIMIT;
mp_size_t  sqr_toom6_threshold          = SQR_TOOM6_THRESHOLD_LIMIT;
mp_size_t  sqr_toom8_threshold          = SQR_TOOM8_THRESHOLD_LIMIT;
mp_size_t  sqr_fft_threshold            = MP_SIZE_T_MAX;
mp_size_t  sqr_fft_modf_threshold       = MP_SIZE_T_MAX;
mp_size_t  mullo_basecase_threshold     = MP_SIZE_T_MAX;
mp_size_t  mullo_dc_threshold           = MP_SIZE_T_MAX;
mp_size_t  mullo_mul_n_threshold        = MP_SIZE_T_MAX;
mp_size_t  sqrlo_basecase_threshold     = MP_SIZE_T_MAX;
mp_size_t  sqrlo_dc_threshold           = MP_SIZE_T_MAX;
mp_size_t  sqrlo_sqr_threshold          = MP_SIZE_T_MAX;
mp_size_t  mulmid_toom42_threshold      = MP_SIZE_T_MAX;
mp_size_t  mulmod_bnm1_threshold        = MP_SIZE_T_MAX;
mp_size_t  sqrmod_bnm1_threshold        = MP_SIZE_T_MAX;
mp_size_t  div_qr_2_pi2_threshold       = MP_SIZE_T_MAX;
mp_size_t  dc_div_qr_threshold          = MP_SIZE_T_MAX;
mp_size_t  dc_divappr_q_threshold       = MP_SIZE_T_MAX;
mp_size_t  mu_div_qr_threshold          = MP_SIZE_T_MAX;
mp_size_t  mu_divappr_q_threshold       = MP_SIZE_T_MAX;
mp_size_t  mupi_div_qr_threshold        = MP_SIZE_T_MAX;
mp_size_t  mu_div_q_threshold           = MP_SIZE_T_MAX;
mp_size_t  dc_bdiv_qr_threshold         = MP_SIZE_T_MAX;
mp_size_t  dc_bdiv_q_threshold          = MP_SIZE_T_MAX;
mp_size_t  mu_bdiv_qr_threshold         = MP_SIZE_T_MAX;
mp_size_t  mu_bdiv_q_threshold          = MP_SIZE_T_MAX;
mp_size_t  inv_mulmod_bnm1_threshold    = MP_SIZE_T_MAX;
mp_size_t  inv_newton_threshold         = MP_SIZE_T_MAX;
mp_size_t  inv_appr_threshold           = MP_SIZE_T_MAX;
mp_size_t  binv_newton_threshold        = MP_SIZE_T_MAX;
mp_size_t  redc_1_to_redc_2_threshold   = MP_SIZE_T_MAX;
mp_size_t  redc_1_to_redc_n_threshold   = MP_SIZE_T_MAX;
mp_size_t  redc_2_to_redc_n_threshold   = MP_SIZE_T_MAX;
mp_size_t  matrix22_strassen_threshold  = MP_SIZE_T_MAX;
mp_size_t  hgcd_threshold               = MP_SIZE_T_MAX;
mp_size_t  hgcd_appr_threshold          = MP_SIZE_T_MAX;
mp_size_t  hgcd_reduce_threshold        = MP_SIZE_T_MAX;
mp_size_t  gcd_dc_threshold             = MP_SIZE_T_MAX;
mp_size_t  gcdext_dc_threshold          = MP_SIZE_T_MAX;
int	   div_qr_1n_pi1_method		= 0;
mp_size_t  div_qr_1_norm_threshold      = MP_SIZE_T_MAX;
mp_size_t  div_qr_1_unnorm_threshold    = MP_SIZE_T_MAX;
mp_size_t  divrem_1_norm_threshold      = MP_SIZE_T_MAX;
mp_size_t  divrem_1_unnorm_threshold    = MP_SIZE_T_MAX;
mp_size_t  mod_1_norm_threshold         = MP_SIZE_T_MAX;
mp_size_t  mod_1_unnorm_threshold       = MP_SIZE_T_MAX;
int	   mod_1_1p_method		= 0;
mp_size_t  mod_1n_to_mod_1_1_threshold  = MP_SIZE_T_MAX;
mp_size_t  mod_1u_to_mod_1_1_threshold  = MP_SIZE_T_MAX;
mp_size_t  mod_1_1_to_mod_1_2_threshold = MP_SIZE_T_MAX;
mp_size_t  mod_1_2_to_mod_1_4_threshold = MP_SIZE_T_MAX;
mp_size_t  preinv_mod_1_to_mod_1_threshold = MP_SIZE_T_MAX;
mp_size_t  divrem_2_threshold           = MP_SIZE_T_MAX;
mp_size_t  get_str_dc_threshold         = MP_SIZE_T_MAX;
mp_size_t  get_str_precompute_threshold = MP_SIZE_T_MAX;
mp_size_t  set_str_dc_threshold         = MP_SIZE_T_MAX;
mp_size_t  set_str_precompute_threshold = MP_SIZE_T_MAX;
mp_size_t  fac_odd_threshold            = 0;
mp_size_t  fac_dsc_threshold            = FAC_DSC_THRESHOLD_LIMIT;

mp_size_t  fft_modf_sqr_threshold = MP_SIZE_T_MAX;
mp_size_t  fft_modf_mul_threshold = MP_SIZE_T_MAX;

struct param_t {
  const char        *name;
  speed_function_t  function;
  speed_function_t  function2;
  double            step_factor;    /* how much to step relatively */
  int               step;           /* how much to step absolutely */
  double            function_fudge; /* multiplier for "function" speeds */
  int               stop_since_change;
  double            stop_factor;
  mp_size_t         min_size;
  int               min_is_always;
  mp_size_t         max_size;
  mp_size_t         check_size;
  mp_size_t         size_extra;

#define DATA_HIGH_LT_R  1
#define DATA_HIGH_GE_R  2
  int               data_high;

  int               noprint;
};


/* These are normally undefined when false, which suits "#if" fine.
   But give them zero values so they can be used in plain C "if"s.  */
#ifndef UDIV_PREINV_ALWAYS
#define UDIV_PREINV_ALWAYS 0
#endif
#ifndef HAVE_NATIVE_mpn_divexact_1
#define HAVE_NATIVE_mpn_divexact_1 0
#endif
#ifndef HAVE_NATIVE_mpn_div_qr_1n_pi1
#define HAVE_NATIVE_mpn_div_qr_1n_pi1 0
#endif
#ifndef HAVE_NATIVE_mpn_divrem_1
#define HAVE_NATIVE_mpn_divrem_1 0
#endif
#ifndef HAVE_NATIVE_mpn_divrem_2
#define HAVE_NATIVE_mpn_divrem_2 0
#endif
#ifndef HAVE_NATIVE_mpn_mod_1
#define HAVE_NATIVE_mpn_mod_1 0
#endif
#ifndef HAVE_NATIVE_mpn_mod_1_1p
#define HAVE_NATIVE_mpn_mod_1_1p 0
#endif
#ifndef HAVE_NATIVE_mpn_modexact_1_odd
#define HAVE_NATIVE_mpn_modexact_1_odd 0
#endif
#ifndef HAVE_NATIVE_mpn_preinv_divrem_1
#define HAVE_NATIVE_mpn_preinv_divrem_1 0
#endif
#ifndef HAVE_NATIVE_mpn_preinv_mod_1
#define HAVE_NATIVE_mpn_preinv_mod_1 0
#endif
#ifndef HAVE_NATIVE_mpn_sqr_basecase
#define HAVE_NATIVE_mpn_sqr_basecase 0
#endif


#define MAX3(a,b,c)  MAX (MAX (a, b), c)

mp_limb_t
randlimb_norm (void)
{
  mp_limb_t  n;
  mpn_random (&n, 1);
  n |= GMP_NUMB_HIGHBIT;
  return n;
}

#define GMP_NUMB_HALFMASK  ((CNST_LIMB(1) << (GMP_NUMB_BITS/2)) - 1)

mp_limb_t
randlimb_half (void)
{
  mp_limb_t  n;
  mpn_random (&n, 1);
  n &= GMP_NUMB_HALFMASK;
  n += (n==0);
  return n;
}


/* Add an entry to the end of the dat[] array, reallocing to make it bigger
   if necessary.  */
void
add_dat (mp_size_t size, double d)
{
#define ALLOCDAT_STEP  500

  ASSERT_ALWAYS (ndat <= allocdat);

  if (ndat == allocdat)
    {
      dat = (struct dat_t *) __gmp_allocate_or_reallocate
        (dat, allocdat * sizeof(dat[0]),
         (allocdat+ALLOCDAT_STEP) * sizeof(dat[0]));
      allocdat += ALLOCDAT_STEP;
    }

  dat[ndat].size = size;
  dat[ndat].d = d;
  ndat++;
}


/* Return the threshold size based on the data accumulated. */
mp_size_t
analyze_dat (int final)
{
  double  x, min_x;
  int     j, min_j;

  /* If the threshold is set at dat[0].size, any positive values are bad. */
  x = 0.0;
  for (j = 0; j < ndat; j++)
    if (dat[j].d > 0.0)
      x += dat[j].d;

  if (option_trace >= 2 && final)
    {
      printf ("\n");
      printf ("x is the sum of the badness from setting thresh at given size\n");
      printf ("  (minimum x is sought)\n");
      printf ("size=%ld  first x=%.4f\n", (long) dat[j].size, x);
    }

  min_x = x;
  min_j = 0;


  /* When stepping to the next dat[j].size, positive values are no longer
     bad (so subtracted), negative values become bad (so add the absolute
     value, meaning subtract). */
  for (j = 0; j < ndat; x -= dat[j].d, j++)
    {
      if (option_trace >= 2 && final)
        printf ("size=%ld  x=%.4f\n", (long) dat[j].size, x);

      if (x < min_x)
        {
          min_x = x;
          min_j = j;
        }
    }

  return min_j;
}


/* Measuring for recompiled mpn/generic/div_qr_1.c,
 * mpn/generic/divrem_1.c, mpn/generic/mod_1.c and mpz/fac_ui.c */

mp_limb_t mpn_div_qr_1_tune (mp_ptr, mp_limb_t *, mp_srcptr, mp_size_t, mp_limb_t);

#if defined (__cplusplus)
extern "C" {
#endif

mp_limb_t mpn_divrem_1_tune (mp_ptr, mp_size_t, mp_srcptr, mp_size_t, mp_limb_t);
mp_limb_t mpn_mod_1_tune (mp_srcptr, mp_size_t, mp_limb_t);
void mpz_fac_ui_tune (mpz_ptr, unsigned long);

#if defined (__cplusplus)
}
#endif

double
speed_mpn_mod_1_tune (struct speed_params *s)
{
  SPEED_ROUTINE_MPN_MOD_1 (mpn_mod_1_tune);
}
double
speed_mpn_divrem_1_tune (struct speed_params *s)
{
  SPEED_ROUTINE_MPN_DIVREM_1 (mpn_divrem_1_tune);
}
double
speed_mpz_fac_ui_tune (struct speed_params *s)
{
  SPEED_ROUTINE_MPZ_FAC_UI (mpz_fac_ui_tune);
}
double
speed_mpn_div_qr_1_tune (struct speed_params *s)
{
  SPEED_ROUTINE_MPN_DIV_QR_1 (mpn_div_qr_1_tune);
}

double
tuneup_measure (speed_function_t fun,
                const struct param_t *param,
                struct speed_params *s)
{
  static struct param_t  dummy;
  double   t;
  TMP_DECL;

  if (! param)
    param = &dummy;

  s->size += param->size_extra;

  TMP_MARK;
  SPEED_TMP_ALLOC_LIMBS (s->xp, s->size, 0);
  SPEED_TMP_ALLOC_LIMBS (s->yp, s->size, 0);

  mpn_random (s->xp, s->size);
  mpn_random (s->yp, s->size);

  switch (param->data_high) {
  case DATA_HIGH_LT_R:
    s->xp[s->size-1] %= s->r;
    s->yp[s->size-1] %= s->r;
    break;
  case DATA_HIGH_GE_R:
    s->xp[s->size-1] |= s->r;
    s->yp[s->size-1] |= s->r;
    break;
  }

  t = speed_measure (fun, s);

  s->size -= param->size_extra;

  TMP_FREE;
  return t;
}


#define PRINT_WIDTH  31

void
print_define_start (const char *name)
{
  printf ("#define %-*s  ", PRINT_WIDTH, name);
  if (option_trace)
    printf ("...\n");
}

void
print_define_end_remark (const char *name, mp_size_t value, const char *remark)
{
  if (option_trace)
    printf ("#define %-*s  ", PRINT_WIDTH, name);

  if (value == MP_SIZE_T_MAX)
    printf ("MP_SIZE_T_MAX");
  else
    printf ("%5ld", (long) value);

  if (remark != NULL)
    printf ("  /* %s */", remark);
  printf ("\n");
  fflush (stdout);
}

void
print_define_end (const char *name, mp_size_t value)
{
  const char  *remark;
  if (value == MP_SIZE_T_MAX)
    remark = "never";
  else if (value == 0)
    remark = "always";
  else
    remark = NULL;
  print_define_end_remark (name, value, remark);
}

void
print_define (const char *name, mp_size_t value)
{
  print_define_start (name);
  print_define_end (name, value);
}

void
print_define_remark (const char *name, mp_size_t value, const char *remark)
{
  print_define_start (name);
  print_define_end_remark (name, value, remark);
}


void
one (mp_size_t *threshold, struct param_t *param)
{
  int  since_positive, since_thresh_change;
  int  thresh_idx, new_thresh_idx;

#define DEFAULT(x,n)  do { if (! (x))  (x) = (n); } while (0)

  DEFAULT (param->function_fudge, 1.0);
  DEFAULT (param->function2, param->function);
  DEFAULT (param->step_factor, 0.01);  /* small steps by default */
  DEFAULT (param->step, 1);            /* small steps by default */
  DEFAULT (param->stop_since_change, 80);
  DEFAULT (param->stop_factor, 1.2);
  DEFAULT (param->min_size, 10);
  DEFAULT (param->max_size, DEFAULT_MAX_SIZE);

  if (param->check_size != 0)
    {
      double   t1, t2;
      s.size = param->check_size;

      *threshold = s.size+1;
      t1 = tuneup_measure (param->function, param, &s);

      *threshold = s.size;
      t2 = tuneup_measure (param->function2, param, &s);
      if (t1 == -1.0 || t2 == -1.0)
        {
          printf ("Oops, can't run both functions at size %ld\n",
                  (long) s.size);
          abort ();
        }
      t1 *= param->function_fudge;

      /* ask that t2 is at least 4% below t1 */
      if (t1 < t2*1.04)
        {
          if (option_trace)
            printf ("function2 never enough faster: t1=%.9f t2=%.9f\n", t1, t2);
          *threshold = MP_SIZE_T_MAX;
          if (! param->noprint)
            print_define (param->name, *threshold);
          return;
        }

      if (option_trace >= 2)
        printf ("function2 enough faster at size=%ld: t1=%.9f t2=%.9f\n",
                (long) s.size, t1, t2);
    }

  if (! param->noprint || option_trace)
    print_define_start (param->name);

  ndat = 0;
  since_positive = 0;
  since_thresh_change = 0;
  thresh_idx = 0;

  if (option_trace >= 2)
    {
      printf ("             algorithm-A  algorithm-B   ratio  possible\n");
      printf ("              (seconds)    (seconds)    diff    thresh\n");
    }

  for (s.size = param->min_size;
       s.size < param->max_size;
       s.size += MAX ((mp_size_t) floor (s.size * param->step_factor), param->step))
    {
      double   ti, tiplus1, d;

      /*
        FIXME: check minimum size requirements are met, possibly by just
        checking for the -1 returns from the speed functions.
      */

      /* using method A at this size */
      *threshold = s.size+1;
      ti = tuneup_measure (param->function, param, &s);
      if (ti == -1.0)
        abort ();
      ti *= param->function_fudge;

      /* using method B at this size */
      *threshold = s.size;
      tiplus1 = tuneup_measure (param->function2, param, &s);
      if (tiplus1 == -1.0)
        abort ();

      /* Calculate the fraction by which the one or the other routine is
         slower.  */
      if (tiplus1 >= ti)
        d = (tiplus1 - ti) / tiplus1;  /* negative */
      else
        d = (tiplus1 - ti) / ti;       /* positive */

      add_dat (s.size, d);

      new_thresh_idx = analyze_dat (0);

      if (option_trace >= 2)
        printf ("size=%ld  %.9f  %.9f  % .4f %c  %ld\n",
                (long) s.size, ti, tiplus1, d,
                ti > tiplus1 ? '#' : ' ',
                (long) dat[new_thresh_idx].size);

      /* Stop if the last time method i was faster was more than a
         certain number of measurements ago.  */
#define STOP_SINCE_POSITIVE  200
      if (d >= 0)
        since_positive = 0;
      else
        if (++since_positive > STOP_SINCE_POSITIVE)
          {
            if (option_trace >= 1)
              printf ("stopped due to since_positive (%d)\n",
                      STOP_SINCE_POSITIVE);
            break;
          }

      /* Stop if method A has become slower by a certain factor. */
      if (ti >= tiplus1 * param->stop_factor)
        {
          if (option_trace >= 1)
            printf ("stopped due to ti >= tiplus1 * factor (%.1f)\n",
                    param->stop_factor);
          break;
        }

      /* Stop if the threshold implied hasn't changed in a certain
         number of measurements.  (It's this condition that usually
         stops the loop.) */
      if (thresh_idx != new_thresh_idx)
        since_thresh_change = 0, thresh_idx = new_thresh_idx;
      else
        if (++since_thresh_change > param->stop_since_change)
          {
            if (option_trace >= 1)
              printf ("stopped due to since_thresh_change (%d)\n",
                      param->stop_since_change);
            break;
          }

      /* Stop if the threshold implied is more than a certain number of
         measurements ago.  */
#define STOP_SINCE_AFTER   500
      if (ndat - thresh_idx > STOP_SINCE_AFTER)
        {
          if (option_trace >= 1)
            printf ("stopped due to ndat - thresh_idx > amount (%d)\n",
                    STOP_SINCE_AFTER);
          break;
        }

      /* Stop when the size limit is reached before the end of the
         crossover, but only show this as an error for >= the default max
         size.  FIXME: Maybe should make it a param choice whether this is
         an error.  */
      if (s.size >= param->max_size && param->max_size >= DEFAULT_MAX_SIZE)
        {
          fprintf (stderr, "%s\n", param->name);
          fprintf (stderr, "sizes %ld to %ld total %d measurements\n",
                   (long) dat[0].size, (long) dat[ndat-1].size, ndat);
          fprintf (stderr, "    max size reached before end of crossover\n");
          break;
        }
    }

  if (option_trace >= 1)
    printf ("sizes %ld to %ld total %d measurements\n",
            (long) dat[0].size, (long) dat[ndat-1].size, ndat);

  *threshold = dat[analyze_dat (1)].size;

  if (param->min_is_always)
    {
      if (*threshold == param->min_size)
        *threshold = 0;
    }

  if (! param->noprint || option_trace)
    print_define_end (param->name, *threshold);
}


/* Special probing for the fft thresholds.  The size restrictions on the
   FFTs mean the graph of time vs size has a step effect.  See this for
   example using

       ./speed -s 4096-16384 -t 128 -P foo mpn_mul_fft.8 mpn_mul_fft.9
       gnuplot foo.gnuplot

   The current approach is to compare routines at the midpoint of relevant
   steps.  Arguably a more sophisticated system of threshold data is wanted
   if this step effect remains. */

struct fft_param_t {
  const char        *table_name;
  const char        *threshold_name;
  const char        *modf_threshold_name;
  mp_size_t         *p_threshold;
  mp_size_t         *p_modf_threshold;
  mp_size_t         first_size;
  mp_size_t         max_size;
  speed_function_t  function;
  speed_function_t  mul_modf_function;
  speed_function_t  mul_function;
  mp_size_t         sqr;
};


/* mpn_mul_fft requires pl a multiple of 2^k limbs, but with
   N=pl*BIT_PER_MP_LIMB it internally also pads out so N/2^k is a multiple
   of 2^(k-1) bits. */

mp_size_t
fft_step_size (int k)
{
  mp_size_t  step;

  step = MAX ((mp_size_t) 1 << (k-1), GMP_LIMB_BITS) / GMP_LIMB_BITS;
  step *= (mp_size_t) 1 << k;

  if (step <= 0)
    {
      printf ("Can't handle k=%d\n", k);
      abort ();
    }

  return step;
}

mp_size_t
fft_next_size (mp_size_t pl, int k)
{
  mp_size_t  m = fft_step_size (k);

/*    printf ("[k=%d %ld] %ld ->", k, m, pl); */

  if (pl == 0 || (pl & (m-1)) != 0)
    pl = (pl | (m-1)) + 1;

/*    printf (" %ld\n", pl); */
  return pl;
}

#define NMAX_DEFAULT 1000000
#define MAX_REPS 25
#define MIN_REPS 5

static inline size_t
mpn_mul_fft_lcm (size_t a, unsigned int k)
{
  unsigned int l = k;

  while (a % 2 == 0 && k > 0)
    {
      a >>= 1;
      k--;
    }
  return a << l;
}

mp_size_t
fftfill (mp_size_t pl, int k, int sqr)
{
  mp_size_t maxLK;
  mp_bitcnt_t N, Nprime, nprime, M;

  N = pl * GMP_NUMB_BITS;
  M = N >> k;

  maxLK = mpn_mul_fft_lcm ((unsigned long) GMP_NUMB_BITS, k);

  Nprime = (1 + (2 * M + k + 2) / maxLK) * maxLK;
  nprime = Nprime / GMP_NUMB_BITS;
  if (nprime >= (sqr ? SQR_FFT_MODF_THRESHOLD : MUL_FFT_MODF_THRESHOLD))
    {
      size_t K2;
      for (;;)
	{
	  K2 = 1L << mpn_fft_best_k (nprime, sqr);
	  if ((nprime & (K2 - 1)) == 0)
	    break;
	  nprime = (nprime + K2 - 1) & -K2;
	  Nprime = nprime * GMP_LIMB_BITS;
	}
    }
  ASSERT_ALWAYS (nprime < pl);

  return Nprime;
}

static int
compare_double (const void *ap, const void *bp)
{
  double a = * (const double *) ap;
  double b = * (const double *) bp;

  if (a < b)
    return -1;
  else if (a > b)
    return 1;
  else
    return 0;
}

double
median (double *times, int n)
{
  qsort (times, n, sizeof (double), compare_double);
  return times[n/2];
}

#define FFT_CACHE_SIZE 25
typedef struct fft_cache
{
  mp_size_t n;
  double time;
} fft_cache_t;

fft_cache_t fft_cache[FFT_CACHE_SIZE];

double
cached_measure (mp_ptr rp, mp_srcptr ap, mp_srcptr bp, mp_size_t n, int k,
		int n_measurements)
{
  int i;
  double t, ttab[MAX_REPS];

  if (fft_cache[k].n == n)
    return fft_cache[k].time;

  for (i = 0; i < n_measurements; i++)
    {
      speed_starttime ();
      mpn_mul_fft (rp, n, ap, n, bp, n, k);
      ttab[i] = speed_endtime ();
    }

  t = median (ttab, n_measurements);
  fft_cache[k].n = n;
  fft_cache[k].time = t;
  return t;
}

#define INSERT_FFTTAB(idx, nval, kval)					\
  do {									\
    fft_tab[idx].n = nval;						\
    fft_tab[idx].k = kval;						\
    fft_tab[idx+1].n = (1 << 27) - 1;	/* sentinel, 27b wide field */	\
    fft_tab[idx+1].k = (1 <<  5) - 1;					\
  } while (0)

int
fftmes (mp_size_t nmin, mp_size_t nmax, int initial_k, struct fft_param_t *p, int idx, int print)
{
  mp_size_t n, n1, prev_n1;
  int k, best_k, last_best_k, kmax;
  int eff, prev_eff;
  double t0, t1;
  int n_measurements;
  mp_limb_t *ap, *bp, *rp;
  mp_size_t alloc;
  struct fft_table_nk *fft_tab;

  fft_tab = mpn_fft_table3[p->sqr];

  for (k = 0; k < FFT_CACHE_SIZE; k++)
    fft_cache[k].n = 0;

  if (nmin < (p->sqr ? SQR_FFT_MODF_THRESHOLD : MUL_FFT_MODF_THRESHOLD))
    {
      nmin = (p->sqr ? SQR_FFT_MODF_THRESHOLD : MUL_FFT_MODF_THRESHOLD);
    }

  if (print)
    printf ("#define %s%*s", p->table_name, 38, "");

  if (idx == 0)
    {
      INSERT_FFTTAB (0, nmin, initial_k);

      if (print)
	{
	  printf ("\\\n  { ");
	  printf ("{%7u,%2u}", fft_tab[0].n, fft_tab[0].k);
	}

      idx = 1;
    }

  ap = (mp_ptr) malloc (sizeof (mp_limb_t));
  if (p->sqr)
    bp = ap;
  else
    bp = (mp_ptr) malloc (sizeof (mp_limb_t));
  rp = (mp_ptr) malloc (sizeof (mp_limb_t));
  alloc = 1;

  /* Round n to comply to initial k value */
  n = (nmin + ((1ul << initial_k) - 1)) & (MP_SIZE_T_MAX << initial_k);

  n_measurements = (18 - initial_k) | 1;
  n_measurements = MAX (n_measurements, MIN_REPS);
  n_measurements = MIN (n_measurements, MAX_REPS);

  last_best_k = initial_k;
  best_k = initial_k;

  while (n < nmax)
    {
      int start_k, end_k;

      /* Assume the current best k is best until we hit its next FFT step.  */
      t0 = 99999;

      prev_n1 = n + 1;

      start_k = MAX (4, best_k - 4);
      end_k = MIN (24, best_k + 4);
      for (k = start_k; k <= end_k; k++)
	{
          n1 = mpn_fft_next_size (prev_n1, k);

	  eff = 200 * (n1 * GMP_NUMB_BITS >> k) / fftfill (n1, k, p->sqr);

	  if (eff < 70)		/* avoid measuring too slow fft:s */
	    continue;

	  if (n1 > alloc)
	    {
	      alloc = n1;
	      if (p->sqr)
		{
		  ap = (mp_ptr) realloc (ap, sizeof (mp_limb_t));
		  rp = (mp_ptr) realloc (rp, sizeof (mp_limb_t));
		  ap = bp = (mp_ptr) realloc (ap, alloc * sizeof (mp_limb_t));
		  mpn_random (ap, alloc);
		  rp = (mp_ptr) realloc (rp, alloc * sizeof (mp_limb_t));
		}
	      else
		{
		  ap = (mp_ptr) realloc (ap, sizeof (mp_limb_t));
		  bp = (mp_ptr) realloc (bp, sizeof (mp_limb_t));
		  rp = (mp_ptr) realloc (rp, sizeof (mp_limb_t));
		  ap = (mp_ptr) realloc (ap, alloc * sizeof (mp_limb_t));
		  mpn_random (ap, alloc);
		  bp = (mp_ptr) realloc (bp, alloc * sizeof (mp_limb_t));
		  mpn_random (bp, alloc);
		  rp = (mp_ptr) realloc (rp, alloc * sizeof (mp_limb_t));
		}
	    }

	  t1 = cached_measure (rp, ap, bp, n1, k, n_measurements);

	  if (t1 * n_measurements > 0.3)
	    n_measurements -= 2;
	  n_measurements = MAX (n_measurements, MIN_REPS);

	  if (t1 < t0)
	    {
	      best_k = k;
	      t0 = t1;
	    }
	}

      n1 = mpn_fft_next_size (prev_n1, best_k);

      if (last_best_k != best_k)
	{
	  ASSERT_ALWAYS ((prev_n1 & ((1ul << last_best_k) - 1)) == 1);

	  if (idx >= FFT_TABLE3_SIZE)
	    {
	      printf ("FFT table exhausted, increase FFT_TABLE3_SIZE in gmp-impl.h\n");
	      abort ();
	    }
	  INSERT_FFTTAB (idx, prev_n1 >> last_best_k, best_k);

	  if (print)
	    {
	      printf (", ");
	      if (idx % 4 == 0)
		printf ("\\\n    ");
	      printf ("{%7u,%2u}", fft_tab[idx].n, fft_tab[idx].k);
	    }

	  if (option_trace >= 2)
	    {
	      printf ("{%lu,%u}\n", prev_n1, best_k);
	      fflush (stdout);
	    }

	  last_best_k = best_k;
	  idx++;
	}

      for (;;)
	{
	  prev_n1 = n1;
	  prev_eff = fftfill (prev_n1, best_k, p->sqr);
	  n1 = mpn_fft_next_size (prev_n1 + 1, best_k);
	  eff = fftfill (n1, best_k, p->sqr);

	  if (eff != prev_eff)
	    break;
	}

      n = prev_n1;
    }

  kmax = sizeof (mp_size_t) * 4;	/* GMP_MP_SIZE_T_BITS / 2 */
  kmax = MIN (kmax, 25-1);
  for (k = last_best_k + 1; k <= kmax; k++)
    {
      if (idx >= FFT_TABLE3_SIZE)
	{
	  printf ("FFT table exhausted, increase FFT_TABLE3_SIZE in gmp-impl.h\n");
	  abort ();
	}
      INSERT_FFTTAB (idx, ((1ul << (2*k-2)) + 1) >> (k-1), k);

      if (print)
	{
	  printf (", ");
	  if (idx % 4 == 0)
	    printf ("\\\n    ");
	  printf ("{%7u,%2u}", fft_tab[idx].n, fft_tab[idx].k);
	}

      idx++;
    }

  if (print)
    printf (" }\n");

  free (ap);
  if (! p->sqr)
    free (bp);
  free (rp);

  return idx;
}

void
fft (struct fft_param_t *p)
{
  mp_size_t  size;
  int        k, idx, initial_k;

  /*** Generate MUL_FFT_MODF_THRESHOLD / SQR_FFT_MODF_THRESHOLD ***/

#if 1
  {
    /* Use plain one() mechanism, for some reasonable initial values of k.  The
       advantage is that we don't depend on mpn_fft_table3, which can therefore
       leave it completely uninitialized.  */

    static struct param_t param;
    mp_size_t thres, best_thres;
    int best_k;
    char buf[20];

    best_thres = MP_SIZE_T_MAX;
    best_k = -1;

    for (k = 5; k <= 7; k++)
      {
	param.name = p->modf_threshold_name;
	param.min_size = 100;
	param.max_size = 2000;
	param.function  = p->mul_function;
	param.step_factor = 0.0;
	param.step = 4;
	param.function2 = p->mul_modf_function;
	param.noprint = 1;
	s.r = k;
	one (&thres, &param);
	if (thres < best_thres)
	  {
	    best_thres = thres;
	    best_k = k;
	  }
      }

    *(p->p_modf_threshold) = best_thres;
    sprintf (buf, "k = %d", best_k);
    print_define_remark (p->modf_threshold_name, best_thres, buf);
    initial_k = best_k;
  }
#else
  size = p->first_size;
  for (;;)
    {
      double  tk, tm;

      size = mpn_fft_next_size (size+1, mpn_fft_best_k (size+1, p->sqr));
      k = mpn_fft_best_k (size, p->sqr);

      if (size >= p->max_size)
        break;

      s.size = size + fft_step_size (k) / 2;
      s.r = k;
      tk = tuneup_measure (p->mul_modf_function, NULL, &s);
      if (tk == -1.0)
        abort ();

      tm = tuneup_measure (p->mul_function, NULL, &s);
      if (tm == -1.0)
        abort ();

      if (option_trace >= 2)
        printf ("at %ld   size=%ld  k=%d  %.9f   size=%ld modf %.9f\n",
                (long) size,
                (long) size + fft_step_size (k) / 2, k, tk,
                (long) s.size, tm);

      if (tk < tm)
        {
	  *p->p_modf_threshold = s.size;
	  print_define (p->modf_threshold_name, *p->p_modf_threshold);
	  break;
        }
    }
  initial_k = ?;
#endif

  /*** Generate MUL_FFT_TABLE3 / SQR_FFT_TABLE3 ***/

  idx = fftmes (*p->p_modf_threshold, p->max_size, initial_k, p, 0, 1);
  printf ("#define %s_SIZE %d\n", p->table_name, idx);

  /*** Generate MUL_FFT_THRESHOLD / SQR_FFT_THRESHOLD ***/

  size = 2 * *p->p_modf_threshold;	/* OK? */
  for (;;)
    {
      double  tk, tm;
      mp_size_t mulmod_size, mul_size;;

      if (size >= p->max_size)
        break;

      mulmod_size = mpn_mulmod_bnm1_next_size (2 * (size + 1)) / 2;
      mul_size = (size + mulmod_size) / 2;	/* middle of step */

      s.size = mulmod_size;
      tk = tuneup_measure (p->function, NULL, &s);
      if (tk == -1.0)
        abort ();

      s.size = mul_size;
      tm = tuneup_measure (p->mul_function, NULL, &s);
      if (tm == -1.0)
        abort ();

      if (option_trace >= 2)
        printf ("at %ld   size=%ld  %.9f   size=%ld mul %.9f\n",
                (long) size,
                (long) mulmod_size, tk,
                (long) mul_size, tm);

      size = mulmod_size;

      if (tk < tm)
        {
	  *p->p_threshold = s.size;
	  print_define (p->threshold_name, *p->p_threshold);
	  break;
        }
    }
}

/* Compare mpn_mul_1 to whatever fast exact single-limb division we have.  This
   is currently mpn_divexact_1, but will become mpn_bdiv_1_qr_pi2 or somesuch.
   This is used in get_str and set_str.  */
void
relspeed_div_1_vs_mul_1 (void)
{
  const size_t max_opsize = 100;
  mp_size_t n;
  long j;
  mp_limb_t rp[max_opsize];
  mp_limb_t ap[max_opsize];
  double multime, divtime;

  mpn_random (ap, max_opsize);

  multime = 0;
  for (n = max_opsize; n > 1; n--)
    {
      mpn_mul_1 (rp, ap, n, MP_BASES_BIG_BASE_10);
      speed_starttime ();
      for (j = speed_precision; j != 0 ; j--)
	mpn_mul_1 (rp, ap, n, MP_BASES_BIG_BASE_10);
      multime += speed_endtime () / n;
    }

  divtime = 0;
  for (n = max_opsize; n > 1; n--)
    {
      /* Make input divisible for good measure.  */
      ap[n - 1] = mpn_mul_1 (ap, ap, n - 1, MP_BASES_BIG_BASE_10);

#if HAVE_NATIVE_mpn_pi1_bdiv_q_1 || ! HAVE_NATIVE_mpn_divexact_1
	  mpn_pi1_bdiv_q_1 (rp, ap, n, MP_BASES_BIG_BASE_10,
			    MP_BASES_BIG_BASE_BINVERTED_10,
			    MP_BASES_BIG_BASE_CTZ_10);
#else
	  mpn_divexact_1 (rp, ap, n, MP_BASES_BIG_BASE_10);
#endif
      speed_starttime ();
      for (j = speed_precision; j != 0 ; j--)
	{
#if HAVE_NATIVE_mpn_pi1_bdiv_q_1 || ! HAVE_NATIVE_mpn_divexact_1
	  mpn_pi1_bdiv_q_1 (rp, ap, n, MP_BASES_BIG_BASE_10,
			    MP_BASES_BIG_BASE_BINVERTED_10,
			    MP_BASES_BIG_BASE_CTZ_10);
#else
	  mpn_divexact_1 (rp, ap, n, MP_BASES_BIG_BASE_10);
#endif
	}
      divtime += speed_endtime () / n;
    }

  print_define ("DIV_1_VS_MUL_1_PERCENT", (int) (100 * divtime/multime));
}


/* Start karatsuba from 4, since the Cray t90 ieee code is much faster at 2,
   giving wrong results.  */
void
tune_mul_n (void)
{
  static struct param_t  param;
  mp_size_t next_toom_start;
  int something_changed;

  param.function = speed_mpn_mul_n;

  param.name = "MUL_TOOM22_THRESHOLD";
  param.min_size = MAX (4, MPN_TOOM22_MUL_MINSIZE);
  param.max_size = MUL_TOOM22_THRESHOLD_LIMIT-1;
  one (&mul_toom22_threshold, &param);

  param.noprint = 1;

  /* Threshold sequence loop.  Disable functions that would be used in a very
     narrow range, re-measuring things when that happens.  */
  something_changed = 1;
  while (something_changed)
    {
      something_changed = 0;

	next_toom_start = mul_toom22_threshold;

	if (mul_toom33_threshold != 0)
	  {
	    param.name = "MUL_TOOM33_THRESHOLD";
	    param.min_size = MAX (next_toom_start, MPN_TOOM33_MUL_MINSIZE);
	    param.max_size = MUL_TOOM33_THRESHOLD_LIMIT-1;
	    one (&mul_toom33_threshold, &param);

	    if (next_toom_start * 1.05 >= mul_toom33_threshold)
	      {
		mul_toom33_threshold = 0;
		something_changed = 1;
	      }
	  }

	next_toom_start = MAX (next_toom_start, mul_toom33_threshold);

	if (mul_toom44_threshold != 0)
	  {
	    param.name = "MUL_TOOM44_THRESHOLD";
	    param.min_size = MAX (next_toom_start, MPN_TOOM44_MUL_MINSIZE);
	    param.max_size = MUL_TOOM44_THRESHOLD_LIMIT-1;
	    one (&mul_toom44_threshold, &param);

	    if (next_toom_start * 1.05 >= mul_toom44_threshold)
	      {
		mul_toom44_threshold = 0;
		something_changed = 1;
	      }
	  }

	next_toom_start = MAX (next_toom_start, mul_toom44_threshold);

	if (mul_toom6h_threshold != 0)
	  {
	    param.name = "MUL_TOOM6H_THRESHOLD";
	    param.min_size = MAX (next_toom_start, MPN_TOOM6H_MUL_MINSIZE);
	    param.max_size = MUL_TOOM6H_THRESHOLD_LIMIT-1;
	    one (&mul_toom6h_threshold, &param);

	    if (next_toom_start * 1.05 >= mul_toom6h_threshold)
	      {
		mul_toom6h_threshold = 0;
		something_changed = 1;
	      }
	  }

	next_toom_start = MAX (next_toom_start, mul_toom6h_threshold);

	if (mul_toom8h_threshold != 0)
	  {
	    param.name = "MUL_TOOM8H_THRESHOLD";
	    param.min_size = MAX (next_toom_start, MPN_TOOM8H_MUL_MINSIZE);
	    param.max_size = MUL_TOOM8H_THRESHOLD_LIMIT-1;
	    one (&mul_toom8h_threshold, &param);

	    if (next_toom_start * 1.05 >= mul_toom8h_threshold)
	      {
		mul_toom8h_threshold = 0;
		something_changed = 1;
	      }
	  }
    }

    print_define ("MUL_TOOM33_THRESHOLD", MUL_TOOM33_THRESHOLD);
    print_define ("MUL_TOOM44_THRESHOLD", MUL_TOOM44_THRESHOLD);
    print_define ("MUL_TOOM6H_THRESHOLD", MUL_TOOM6H_THRESHOLD);
    print_define ("MUL_TOOM8H_THRESHOLD", MUL_TOOM8H_THRESHOLD);

  /* disabled until tuned */
  MUL_FFT_THRESHOLD = MP_SIZE_T_MAX;
}

void
tune_mul (void)
{
  static struct param_t  param;
  mp_size_t thres;

  param.noprint = 1;

  param.function = speed_mpn_toom32_for_toom43_mul;
  param.function2 = speed_mpn_toom43_for_toom32_mul;
  param.name = "MUL_TOOM32_TO_TOOM43_THRESHOLD";
  param.min_size = MPN_TOOM43_MUL_MINSIZE * 24 / 17;
  one (&thres, &param);
  mul_toom32_to_toom43_threshold = thres * 17 / 24;
  print_define ("MUL_TOOM32_TO_TOOM43_THRESHOLD", mul_toom32_to_toom43_threshold);

  param.function = speed_mpn_toom32_for_toom53_mul;
  param.function2 = speed_mpn_toom53_for_toom32_mul;
  param.name = "MUL_TOOM32_TO_TOOM53_THRESHOLD";
  param.min_size = MPN_TOOM53_MUL_MINSIZE * 30 / 19;
  one (&thres, &param);
  mul_toom32_to_toom53_threshold = thres * 19 / 30;
  print_define ("MUL_TOOM32_TO_TOOM53_THRESHOLD", mul_toom32_to_toom53_threshold);

  param.function = speed_mpn_toom42_for_toom53_mul;
  param.function2 = speed_mpn_toom53_for_toom42_mul;
  param.name = "MUL_TOOM42_TO_TOOM53_THRESHOLD";
  param.min_size = MPN_TOOM53_MUL_MINSIZE * 20 / 11;
  one (&thres, &param);
  mul_toom42_to_toom53_threshold = thres * 11 / 20;
  print_define ("MUL_TOOM42_TO_TOOM53_THRESHOLD", mul_toom42_to_toom53_threshold);

  param.function = speed_mpn_toom42_mul;
  param.function2 = speed_mpn_toom63_mul;
  param.name = "MUL_TOOM42_TO_TOOM63_THRESHOLD";
  param.min_size = MPN_TOOM63_MUL_MINSIZE * 2;
  one (&thres, &param);
  mul_toom42_to_toom63_threshold = thres / 2;
  print_define ("MUL_TOOM42_TO_TOOM63_THRESHOLD", mul_toom42_to_toom63_threshold);

  /* Use ratio 5/6 when measuring, the middle of the range 2/3 to 1. */
  param.function = speed_mpn_toom43_for_toom54_mul;
  param.function2 = speed_mpn_toom54_for_toom43_mul;
  param.name = "MUL_TOOM43_TO_TOOM54_THRESHOLD";
  param.min_size = MPN_TOOM54_MUL_MINSIZE * 6 / 5;
  one (&thres, &param);
  mul_toom43_to_toom54_threshold = thres * 5 / 6;
  print_define ("MUL_TOOM43_TO_TOOM54_THRESHOLD", mul_toom43_to_toom54_threshold);
}


void
tune_mullo (void)
{
  static struct param_t  param;

  param.function = speed_mpn_mullo_n;

  param.name = "MULLO_BASECASE_THRESHOLD";
  param.min_size = 1;
  param.min_is_always = 1;
  param.max_size = MULLO_BASECASE_THRESHOLD_LIMIT-1;
  param.stop_factor = 1.5;
  param.noprint = 1;
  one (&mullo_basecase_threshold, &param);

  param.name = "MULLO_DC_THRESHOLD";
  param.min_size = 8;
  param.min_is_always = 0;
  param.max_size = 1000;
  one (&mullo_dc_threshold, &param);

  if (mullo_basecase_threshold >= mullo_dc_threshold)
    {
      print_define ("MULLO_BASECASE_THRESHOLD", mullo_dc_threshold);
      print_define_remark ("MULLO_DC_THRESHOLD", 0, "never mpn_mullo_basecase");
    }
  else
    {
      print_define ("MULLO_BASECASE_THRESHOLD", mullo_basecase_threshold);
      print_define ("MULLO_DC_THRESHOLD", mullo_dc_threshold);
    }

  if (WANT_FFT && mul_fft_threshold < MP_SIZE_T_MAX / 2)
    {
      param.name = "MULLO_MUL_N_THRESHOLD";
      param.min_size = mullo_dc_threshold;
      param.max_size = 2 * mul_fft_threshold;
      param.noprint = 0;
      param.step_factor = 0.03;
      one (&mullo_mul_n_threshold, &param);
    }
  else
    print_define_remark ("MULLO_MUL_N_THRESHOLD", MP_SIZE_T_MAX,
			 "without FFT use mullo forever");
}

void
tune_sqrlo (void)
{
  static struct param_t  param;

  param.function = speed_mpn_sqrlo;

  param.name = "SQRLO_BASECASE_THRESHOLD";
  param.min_size = 1;
  param.min_is_always = 1;
  param.max_size = SQRLO_BASECASE_THRESHOLD_LIMIT-1;
  param.stop_factor = 1.5;
  param.noprint = 1;
  one (&sqrlo_basecase_threshold, &param);

  param.name = "SQRLO_DC_THRESHOLD";
  param.min_size = 8;
  param.min_is_always = 0;
  param.max_size = SQRLO_DC_THRESHOLD_LIMIT-1;
  one (&sqrlo_dc_threshold, &param);

  if (sqrlo_basecase_threshold >= sqrlo_dc_threshold)
    {
      print_define ("SQRLO_BASECASE_THRESHOLD", sqrlo_dc_threshold);
      print_define_remark ("SQRLO_DC_THRESHOLD", 0, "never mpn_sqrlo_basecase");
    }
  else
    {
      print_define ("SQRLO_BASECASE_THRESHOLD", sqrlo_basecase_threshold);
      print_define ("SQRLO_DC_THRESHOLD", sqrlo_dc_threshold);
    }

  if (WANT_FFT && sqr_fft_threshold < MP_SIZE_T_MAX / 2)
    {
      param.name = "SQRLO_SQR_THRESHOLD";
      param.min_size = sqrlo_dc_threshold;
      param.max_size = 2 * sqr_fft_threshold;
      param.noprint = 0;
      param.step_factor = 0.03;
      one (&sqrlo_sqr_threshold, &param);
    }
  else
    print_define_remark ("SQRLO_SQR_THRESHOLD", MP_SIZE_T_MAX,
			 "without FFT use sqrlo forever");
}

void
tune_mulmid (void)
{
  static struct param_t  param;

  param.name = "MULMID_TOOM42_THRESHOLD";
  param.function = speed_mpn_mulmid_n;
  param.min_size = 4;
  param.max_size = 100;
  one (&mulmid_toom42_threshold, &param);
}

void
tune_mulmod_bnm1 (void)
{
  static struct param_t  param;

  param.name = "MULMOD_BNM1_THRESHOLD";
  param.function = speed_mpn_mulmod_bnm1;
  param.min_size = 4;
  param.max_size = 100;
  one (&mulmod_bnm1_threshold, &param);
}

void
tune_sqrmod_bnm1 (void)
{
  static struct param_t  param;

  param.name = "SQRMOD_BNM1_THRESHOLD";
  param.function = speed_mpn_sqrmod_bnm1;
  param.min_size = 4;
  param.max_size = 100;
  one (&sqrmod_bnm1_threshold, &param);
}


/* Start the basecase from 3, since 1 is a special case, and if mul_basecase
   is faster only at size==2 then we don't want to bother with extra code
   just for that.  Start karatsuba from 4 same as MUL above.  */

void
tune_sqr (void)
{
  /* disabled until tuned */
  SQR_FFT_THRESHOLD = MP_SIZE_T_MAX;

  if (HAVE_NATIVE_mpn_sqr_basecase)
    {
      print_define_remark ("SQR_BASECASE_THRESHOLD", 0, "always (native)");
      sqr_basecase_threshold = 0;
    }
  else
    {
      static struct param_t  param;
      param.name = "SQR_BASECASE_THRESHOLD";
      param.function = speed_mpn_sqr;
      param.min_size = 3;
      param.min_is_always = 1;
      param.max_size = TUNE_SQR_TOOM2_MAX;
      param.noprint = 1;
      one (&sqr_basecase_threshold, &param);
    }

  {
    static struct param_t  param;
    param.name = "SQR_TOOM2_THRESHOLD";
    param.function = speed_mpn_sqr;
    param.min_size = MAX (4, MPN_TOOM2_SQR_MINSIZE);
    param.max_size = TUNE_SQR_TOOM2_MAX;
    param.noprint = 1;
    one (&sqr_toom2_threshold, &param);

    if (! HAVE_NATIVE_mpn_sqr_basecase
        && sqr_toom2_threshold < sqr_basecase_threshold)
      {
        /* Karatsuba becomes faster than mul_basecase before
           sqr_basecase does.  Arrange for the expression
           "BELOW_THRESHOLD (un, SQR_TOOM2_THRESHOLD))" which
           selects mpn_sqr_basecase in mpn_sqr to be false, by setting
           SQR_TOOM2_THRESHOLD to zero, making
           SQR_BASECASE_THRESHOLD the toom2 threshold.  */

        sqr_basecase_threshold = SQR_TOOM2_THRESHOLD;
        SQR_TOOM2_THRESHOLD = 0;

        print_define_remark ("SQR_BASECASE_THRESHOLD", sqr_basecase_threshold,
                             "toom2");
        print_define_remark ("SQR_TOOM2_THRESHOLD",SQR_TOOM2_THRESHOLD,
                             "never sqr_basecase");
      }
    else
      {
        if (! HAVE_NATIVE_mpn_sqr_basecase)
          print_define ("SQR_BASECASE_THRESHOLD", sqr_basecase_threshold);
        print_define ("SQR_TOOM2_THRESHOLD", SQR_TOOM2_THRESHOLD);
      }
  }

  {
    static struct param_t  param;
    mp_size_t next_toom_start;
    int something_changed;

    param.function = speed_mpn_sqr;
    param.noprint = 1;

  /* Threshold sequence loop.  Disable functions that would be used in a very
     narrow range, re-measuring things when that happens.  */
    something_changed = 1;
    while (something_changed)
      {
	something_changed = 0;

	next_toom_start = MAX (sqr_toom2_threshold, sqr_basecase_threshold);

	sqr_toom3_threshold = SQR_TOOM3_THRESHOLD_LIMIT;
	param.name = "SQR_TOOM3_THRESHOLD";
	param.min_size = MAX (next_toom_start, MPN_TOOM3_SQR_MINSIZE);
	param.max_size = SQR_TOOM3_THRESHOLD_LIMIT-1;
	one (&sqr_toom3_threshold, &param);

	next_toom_start = MAX (next_toom_start, sqr_toom3_threshold);

	if (sqr_toom4_threshold != 0)
	  {
	    param.name = "SQR_TOOM4_THRESHOLD";
	    sqr_toom4_threshold = SQR_TOOM4_THRESHOLD_LIMIT;
	    param.min_size = MAX (next_toom_start, MPN_TOOM4_SQR_MINSIZE);
	    param.max_size = SQR_TOOM4_THRESHOLD_LIMIT-1;
	    one (&sqr_toom4_threshold, &param);

	    if (next_toom_start * 1.05 >= sqr_toom4_threshold)
	      {
		sqr_toom4_threshold = 0;
		something_changed = 1;
	      }
	  }

	next_toom_start = MAX (next_toom_start, sqr_toom4_threshold);

	if (sqr_toom6_threshold != 0)
	  {
	    param.name = "SQR_TOOM6_THRESHOLD";
	    sqr_toom6_threshold = SQR_TOOM6_THRESHOLD_LIMIT;
	    param.min_size = MAX (next_toom_start, MPN_TOOM6_SQR_MINSIZE);
	    param.max_size = SQR_TOOM6_THRESHOLD_LIMIT-1;
	    one (&sqr_toom6_threshold, &param);

	    if (next_toom_start * 1.05 >= sqr_toom6_threshold)
	      {
		sqr_toom6_threshold = 0;
		something_changed = 1;
	      }
	  }

	next_toom_start = MAX (next_toom_start, sqr_toom6_threshold);

	if (sqr_toom8_threshold != 0)
	  {
	    param.name = "SQR_TOOM8_THRESHOLD";
	    sqr_toom8_threshold = SQR_TOOM8_THRESHOLD_LIMIT;
	    param.min_size = MAX (next_toom_start, MPN_TOOM8_SQR_MINSIZE);
	    param.max_size = SQR_TOOM8_THRESHOLD_LIMIT-1;
	    one (&sqr_toom8_threshold, &param);

	    if (next_toom_start * 1.05 >= sqr_toom8_threshold)
	      {
		sqr_toom8_threshold = 0;
		something_changed = 1;
	      }
	  }
      }

    print_define ("SQR_TOOM3_THRESHOLD", SQR_TOOM3_THRESHOLD);
    print_define ("SQR_TOOM4_THRESHOLD", SQR_TOOM4_THRESHOLD);
    print_define ("SQR_TOOM6_THRESHOLD", SQR_TOOM6_THRESHOLD);
    print_define ("SQR_TOOM8_THRESHOLD", SQR_TOOM8_THRESHOLD);
  }
}


void
tune_dc_div (void)
{
  s.r = 0;		/* clear to make speed function do 2n/n */
  {
    static struct param_t  param;
    param.name = "DC_DIV_QR_THRESHOLD";
    param.function = speed_mpn_sbpi1_div_qr;
    param.function2 = speed_mpn_dcpi1_div_qr;
    param.min_size = 6;
    one (&dc_div_qr_threshold, &param);
  }
  {
    static struct param_t  param;
    param.name = "DC_DIVAPPR_Q_THRESHOLD";
    param.function = speed_mpn_sbpi1_divappr_q;
    param.function2 = speed_mpn_dcpi1_divappr_q;
    param.min_size = 6;
    one (&dc_divappr_q_threshold, &param);
  }
}

static double
speed_mpn_sbordcpi1_div_qr (struct speed_params *s)
{
  if (s->size < DC_DIV_QR_THRESHOLD)
    return speed_mpn_sbpi1_div_qr (s);
  else
    return speed_mpn_dcpi1_div_qr (s);
}

void
tune_mu_div (void)
{
  s.r = 0;		/* clear to make speed function do 2n/n */
  {
    static struct param_t  param;
    param.name = "MU_DIV_QR_THRESHOLD";
    param.function = speed_mpn_dcpi1_div_qr;
    param.function2 = speed_mpn_mu_div_qr;
    param.min_size = mul_toom22_threshold;
    param.max_size = 5000;
    param.step_factor = 0.02;
    one (&mu_div_qr_threshold, &param);
  }
  {
    static struct param_t  param;
    param.name = "MU_DIVAPPR_Q_THRESHOLD";
    param.function = speed_mpn_dcpi1_divappr_q;
    param.function2 = speed_mpn_mu_divappr_q;
    param.min_size = mul_toom22_threshold;
    param.max_size = 5000;
    param.step_factor = 0.02;
    one (&mu_divappr_q_threshold, &param);
  }
  {
    static struct param_t  param;
    param.name = "MUPI_DIV_QR_THRESHOLD";
    param.function = speed_mpn_sbordcpi1_div_qr;
    param.function2 = speed_mpn_mupi_div_qr;
    param.min_size = 6;
    param.min_is_always = 1;
    param.max_size = 1000;
    param.step_factor = 0.02;
    one (&mupi_div_qr_threshold, &param);
  }
}

void
tune_dc_bdiv (void)
{
  s.r = 0;		/* clear to make speed function do 2n/n*/
  {
    static struct param_t  param;
    param.name = "DC_BDIV_QR_THRESHOLD";
    param.function = speed_mpn_sbpi1_bdiv_qr;
    param.function2 = speed_mpn_dcpi1_bdiv_qr;
    param.min_size = 4;
    one (&dc_bdiv_qr_threshold, &param);
  }
  {
    static struct param_t  param;
    param.name = "DC_BDIV_Q_THRESHOLD";
    param.function = speed_mpn_sbpi1_bdiv_q;
    param.function2 = speed_mpn_dcpi1_bdiv_q;
    param.min_size = 4;
    one (&dc_bdiv_q_threshold, &param);
  }
}

void
tune_mu_bdiv (void)
{
  s.r = 0;		/* clear to make speed function do 2n/n*/
  {
    static struct param_t  param;
    param.name = "MU_BDIV_QR_THRESHOLD";
    param.function = speed_mpn_dcpi1_bdiv_qr;
    param.function2 = speed_mpn_mu_bdiv_qr;
    param.min_size = dc_bdiv_qr_threshold;
    param.max_size = 5000;
    param.step_factor = 0.02;
    one (&mu_bdiv_qr_threshold, &param);
  }
  {
    static struct param_t  param;
    param.name = "MU_BDIV_Q_THRESHOLD";
    param.function = speed_mpn_dcpi1_bdiv_q;
    param.function2 = speed_mpn_mu_bdiv_q;
    param.min_size = dc_bdiv_q_threshold;
    param.max_size = 5000;
    param.step_factor = 0.02;
    one (&mu_bdiv_q_threshold, &param);
  }
}

void
tune_invertappr (void)
{
  static struct param_t  param;

  param.function = speed_mpn_ni_invertappr;
  param.name = "INV_MULMOD_BNM1_THRESHOLD";
  param.min_size = 5;
  one (&inv_mulmod_bnm1_threshold, &param);

  param.function = speed_mpn_invertappr;
  param.name = "INV_NEWTON_THRESHOLD";
  param.min_size = 5;
  one (&inv_newton_threshold, &param);
}

void
tune_invert (void)
{
  static struct param_t  param;

  param.function = speed_mpn_invert;
  param.name = "INV_APPR_THRESHOLD";
  param.min_size = 5;
  one (&inv_appr_threshold, &param);
}

void
tune_binvert (void)
{
  static struct param_t  param;

  param.function = speed_mpn_binvert;
  param.name = "BINV_NEWTON_THRESHOLD";
  param.min_size = 8;		/* pointless with smaller operands */
  one (&binv_newton_threshold, &param);
}

void
tune_redc (void)
{
#define TUNE_REDC_2_MAX 100
#if HAVE_NATIVE_mpn_addmul_2 || HAVE_NATIVE_mpn_redc_2
#define WANT_REDC_2 1
#endif

#if WANT_REDC_2
  {
    static struct param_t  param;
    param.name = "REDC_1_TO_REDC_2_THRESHOLD";
    param.function = speed_mpn_redc_1;
    param.function2 = speed_mpn_redc_2;
    param.min_size = 1;
    param.min_is_always = 1;
    param.max_size = TUNE_REDC_2_MAX;
    param.noprint = 1;
    param.stop_factor = 1.5;
    one (&redc_1_to_redc_2_threshold, &param);
  }
  {
    static struct param_t  param;
    param.name = "REDC_2_TO_REDC_N_THRESHOLD";
    param.function = speed_mpn_redc_2;
    param.function2 = speed_mpn_redc_n;
    param.min_size = 16;
    param.noprint = 1;
    one (&redc_2_to_redc_n_threshold, &param);
  }
  if (redc_1_to_redc_2_threshold >= redc_2_to_redc_n_threshold)
    {
      redc_2_to_redc_n_threshold = 0;	/* disable redc_2 */

      /* Never use redc2, measure redc_1 -> redc_n cutoff, store result as
	 REDC_1_TO_REDC_2_THRESHOLD.  */
      {
	static struct param_t  param;
	param.name = "REDC_1_TO_REDC_2_THRESHOLD";
	param.function = speed_mpn_redc_1;
	param.function2 = speed_mpn_redc_n;
	param.min_size = 16;
	param.noprint = 1;
	one (&redc_1_to_redc_2_threshold, &param);
      }
    }
  print_define ("REDC_1_TO_REDC_2_THRESHOLD", REDC_1_TO_REDC_2_THRESHOLD);
  print_define ("REDC_2_TO_REDC_N_THRESHOLD", REDC_2_TO_REDC_N_THRESHOLD);
#else
  {
    static struct param_t  param;
    param.name = "REDC_1_TO_REDC_N_THRESHOLD";
    param.function = speed_mpn_redc_1;
    param.function2 = speed_mpn_redc_n;
    param.min_size = 16;
    one (&redc_1_to_redc_n_threshold, &param);
  }
#endif
}

void
tune_matrix22_mul (void)
{
  static struct param_t  param;
  param.name = "MATRIX22_STRASSEN_THRESHOLD";
  param.function = speed_mpn_matrix22_mul;
  param.min_size = 2;
  one (&matrix22_strassen_threshold, &param);
}

void
tune_hgcd (void)
{
  static struct param_t  param;
  param.name = "HGCD_THRESHOLD";
  param.function = speed_mpn_hgcd;
  /* We seem to get strange results for small sizes */
  param.min_size = 30;
  one (&hgcd_threshold, &param);
}

void
tune_hgcd_appr (void)
{
  static struct param_t  param;
  param.name = "HGCD_APPR_THRESHOLD";
  param.function = speed_mpn_hgcd_appr;
  /* We seem to get strange results for small sizes */
  param.min_size = 50;
  param.stop_since_change = 150;
  one (&hgcd_appr_threshold, &param);
}

void
tune_hgcd_reduce (void)
{
  static struct param_t  param;
  param.name = "HGCD_REDUCE_THRESHOLD";
  param.function = speed_mpn_hgcd_reduce;
  param.min_size = 30;
  param.max_size = 7000;
  param.step_factor = 0.04;
  one (&hgcd_reduce_threshold, &param);
}

void
tune_gcd_dc (void)
{
  static struct param_t  param;
  param.name = "GCD_DC_THRESHOLD";
  param.function = speed_mpn_gcd;
  param.min_size = hgcd_threshold;
  param.max_size = 3000;
  param.step_factor = 0.02;
  one (&gcd_dc_threshold, &param);
}

void
tune_gcdext_dc (void)
{
  static struct param_t  param;
  param.name = "GCDEXT_DC_THRESHOLD";
  param.function = speed_mpn_gcdext;
  param.min_size = hgcd_threshold;
  param.max_size = 3000;
  param.step_factor = 0.02;
  one (&gcdext_dc_threshold, &param);
}

/* In tune_powm_sec we compute the table used by the win_size function.  The
   cutoff points are in exponent bits, disregarding other operand sizes.  It is
   not possible to use the one framework since it currently uses a granularity
   of full limbs.
*/

/* This win_size replaces the variant in the powm code, allowing us to
   control k in the k-ary algorithms.  */
int winsize;
int
win_size (mp_bitcnt_t eb)
{
  return winsize;
}

void
tune_powm_sec (void)
{
  mp_size_t n;
  int k, i;
  mp_size_t itch;
  mp_bitcnt_t nbits, nbits_next, possible_nbits_cutoff;
  const int n_max = 3000 / GMP_NUMB_BITS;
  const int n_measurements = 5;
  mp_ptr rp, bp, ep, mp, tp;
  double ttab[n_measurements], tk, tkp1;
  TMP_DECL;
  TMP_MARK;

  possible_nbits_cutoff = 0;

  k = 1;

  winsize = 10;			/* the itch function needs this */
  itch = mpn_sec_powm_itch (n_max, n_max * GMP_NUMB_BITS, n_max);

  rp = TMP_ALLOC_LIMBS (n_max);
  bp = TMP_ALLOC_LIMBS (n_max);
  ep = TMP_ALLOC_LIMBS (n_max);
  mp = TMP_ALLOC_LIMBS (n_max);
  tp = TMP_ALLOC_LIMBS (itch);

  mpn_random (bp, n_max);
  mpn_random (mp, n_max);
  mp[0] |= 1;

/* How about taking the M operand size into account?

   An operation R=powm(B,E,N) will take time O(log(E)*M(log(N))) (assuming
   B = O(M)).

   Using k-ary and no sliding window, the precomputation will need time
   O(2^(k-1)*M(log(N))) and the main computation will need O(log(E)*S(N)) +
   O(log(E)/k*M(N)), for the squarings, multiplications, respectively.

   An operation R=powm_sec(B,E,N) will take time like powm.

   Using k-ary, the precomputation will need time O(2^k*M(log(N))) and the
   main computation will need O(log(E)*S(N)) + O(log(E)/k*M(N)) +
   O(log(E)/k*2^k*log(N)), for the squarings, multiplications, and full
   table reads, respectively.  */

  printf ("#define POWM_SEC_TABLE  ");

  /* For nbits == 1, we should always use k == 1, so no need to tune
     that. Starting with nbits == 2 also ensure that nbits always is
     larger than the windowsize k+1. */
  for (nbits = 2; nbits <= n_max * GMP_NUMB_BITS; )
    {
      n = (nbits - 1) / GMP_NUMB_BITS + 1;

      /* Generate E such that sliding-window for k and k+1 works equally
	 well/poorly (but sliding is not used in powm_sec, of course). */
      for (i = 0; i < n; i++)
	ep[i] = ~CNST_LIMB(0);

      winsize = k;
      for (i = 0; i < n_measurements; i++)
	{
	  speed_starttime ();
	  mpn_sec_powm (rp, bp, n, ep, nbits, mp, n, tp);
	  ttab[i] = speed_endtime ();
	}
      tk = median (ttab, n_measurements);

      winsize = k + 1;
      speed_starttime ();
      for (i = 0; i < n_measurements; i++)
	{
	  speed_starttime ();
	  mpn_sec_powm (rp, bp, n, ep, nbits, mp, n, tp);
	  ttab[i] = speed_endtime ();
	}
      tkp1 = median (ttab, n_measurements);
/*
      printf ("testing: %ld, %d", nbits, k, ep[n-1]);
      printf ("   %10.5f  %10.5f\n", tk, tkp1);
*/
      if (tkp1 < tk)
	{
	  if (possible_nbits_cutoff)
	    {
	      /* Two consecutive sizes indicate k increase, obey.  */

	      /* Must always have x[k] >= k */
	      ASSERT_ALWAYS (possible_nbits_cutoff >= k);

	      if (k > 1)
		printf (",");
	      printf ("%ld", (long) possible_nbits_cutoff);
	      k++;
	      possible_nbits_cutoff = 0;
	    }
	  else
	    {
	      /* One measurement indicate k increase, save nbits for further
		 consideration.  */
	      /* The new larger k gets used for sizes > the cutoff
		 value, hence the cutoff should be one less than the
		 smallest size where it gives a speedup. */
	      possible_nbits_cutoff = nbits - 1;
	    }
	}
      else
	possible_nbits_cutoff = 0;

      nbits_next = nbits * 65 / 64;
      nbits = nbits_next + (nbits_next == nbits);
    }
  printf ("\n");
  TMP_FREE;
}


/* size_extra==1 reflects the fact that with high<divisor one division is
   always skipped.  Forcing high<divisor while testing ensures consistency
   while stepping through sizes, ie. that size-1 divides will be done each
   time.

   min_size==2 and min_is_always are used so that if plain division is only
   better at size==1 then don't bother including that code just for that
   case, instead go with preinv always and get a size saving.  */

#define DIV_1_PARAMS                    \
  param.check_size = 256;               \
  param.min_size = 2;                   \
  param.min_is_always = 1;              \
  param.data_high = DATA_HIGH_LT_R;     \
  param.size_extra = 1;                 \
  param.stop_factor = 2.0;


double (*tuned_speed_mpn_divrem_1) (struct speed_params *);

void
tune_divrem_1 (void)
{
  /* plain version by default */
  tuned_speed_mpn_divrem_1 = speed_mpn_divrem_1;

  /* No support for tuning native assembler code, do that by hand and put
     the results in the .asm file, there's no need for such thresholds to
     appear in gmp-mparam.h.  */
  if (HAVE_NATIVE_mpn_divrem_1)
    return;

  if (GMP_NAIL_BITS != 0)
    {
      print_define_remark ("DIVREM_1_NORM_THRESHOLD", MP_SIZE_T_MAX,
                           "no preinv with nails");
      print_define_remark ("DIVREM_1_UNNORM_THRESHOLD", MP_SIZE_T_MAX,
                           "no preinv with nails");
      return;
    }

  if (UDIV_PREINV_ALWAYS)
    {
      print_define_remark ("DIVREM_1_NORM_THRESHOLD", 0L, "preinv always");
      print_define ("DIVREM_1_UNNORM_THRESHOLD", 0L);
      return;
    }

  tuned_speed_mpn_divrem_1 = speed_mpn_divrem_1_tune;

  /* Tune for the integer part of mpn_divrem_1.  This will very possibly be
     a bit out for the fractional part, but that's too bad, the integer part
     is more important. */
  {
    static struct param_t  param;
    param.name = "DIVREM_1_NORM_THRESHOLD";
    DIV_1_PARAMS;
    s.r = randlimb_norm ();
    param.function = speed_mpn_divrem_1_tune;
    one (&divrem_1_norm_threshold, &param);
  }
  {
    static struct param_t  param;
    param.name = "DIVREM_1_UNNORM_THRESHOLD";
    DIV_1_PARAMS;
    s.r = randlimb_half ();
    param.function = speed_mpn_divrem_1_tune;
    one (&divrem_1_unnorm_threshold, &param);
  }
}

void
tune_div_qr_1 (void)
{
  static struct param_t  param;
  double            t1, t2;

  if (!HAVE_NATIVE_mpn_div_qr_1n_pi1)
    {
      static struct param_t  param;
      double   t1, t2;

      s.size = 10;
      s.r = randlimb_norm ();

      t1 = tuneup_measure (speed_mpn_div_qr_1n_pi1_1, &param, &s);
      t2 = tuneup_measure (speed_mpn_div_qr_1n_pi1_2, &param, &s);

      if (t1 == -1.0 || t2 == -1.0)
	{
	  printf ("Oops, can't measure all mpn_div_qr_1n_pi1 methods at %ld\n",
		  (long) s.size);
	  abort ();
	}
      div_qr_1n_pi1_method = (t1 < t2) ? 1 : 2;
      print_define ("DIV_QR_1N_PI1_METHOD", div_qr_1n_pi1_method);
    }

  {
    static struct param_t  param;
    param.name = "DIV_QR_1_NORM_THRESHOLD";
    DIV_1_PARAMS;
    param.min_size = 1;
    param.min_is_always = 0;
    s.r = randlimb_norm ();
    param.function = speed_mpn_div_qr_1_tune;
    one (&div_qr_1_norm_threshold, &param);
  }
  {
    static struct param_t  param;
    param.name = "DIV_QR_1_UNNORM_THRESHOLD";
    DIV_1_PARAMS;
    param.min_size = 1;
    param.min_is_always = 0;
    s.r = randlimb_half();
    param.function = speed_mpn_div_qr_1_tune;
    one (&div_qr_1_unnorm_threshold, &param);
  }
}


void
tune_mod_1 (void)
{
  /* No support for tuning native assembler code, do that by hand and put
     the results in the .asm file, there's no need for such thresholds to
     appear in gmp-mparam.h.  */
  if (HAVE_NATIVE_mpn_mod_1)
    return;

  if (GMP_NAIL_BITS != 0)
    {
      print_define_remark ("MOD_1_NORM_THRESHOLD", MP_SIZE_T_MAX,
                           "no preinv with nails");
      print_define_remark ("MOD_1_UNNORM_THRESHOLD", MP_SIZE_T_MAX,
                           "no preinv with nails");
      return;
    }

  if (!HAVE_NATIVE_mpn_mod_1_1p)
    {
      static struct param_t  param;
      double   t1, t2;

      s.size = 10;
      s.r = randlimb_half ();

      t1 = tuneup_measure (speed_mpn_mod_1_1_1, &param, &s);
      t2 = tuneup_measure (speed_mpn_mod_1_1_2, &param, &s);

      if (t1 == -1.0 || t2 == -1.0)
	{
	  printf ("Oops, can't measure all mpn_mod_1_1 methods at %ld\n",
		  (long) s.size);
	  abort ();
	}
      mod_1_1p_method = (t1 < t2) ? 1 : 2;
      print_define ("MOD_1_1P_METHOD", mod_1_1p_method);
    }

  if (UDIV_PREINV_ALWAYS)
    {
      print_define ("MOD_1_NORM_THRESHOLD", 0L);
      print_define ("MOD_1_UNNORM_THRESHOLD", 0L);
    }
  else
    {
      {
	static struct param_t  param;
	param.name = "MOD_1_NORM_THRESHOLD";
	DIV_1_PARAMS;
	s.r = randlimb_norm ();
	param.function = speed_mpn_mod_1_tune;
	one (&mod_1_norm_threshold, &param);
      }
      {
	static struct param_t  param;
	param.name = "MOD_1_UNNORM_THRESHOLD";
	DIV_1_PARAMS;
	s.r = randlimb_half ();
	param.function = speed_mpn_mod_1_tune;
	one (&mod_1_unnorm_threshold, &param);
      }
    }
  {
    static struct param_t  param;

    param.check_size = 256;

    s.r = randlimb_norm ();
    param.function = speed_mpn_mod_1_tune;

    param.name = "MOD_1N_TO_MOD_1_1_THRESHOLD";
    param.min_size = 2;
    one (&mod_1n_to_mod_1_1_threshold, &param);
  }

  {
    static struct param_t  param;

    param.check_size = 256;
    s.r = randlimb_half ();
    param.noprint = 1;

    param.function = speed_mpn_mod_1_1;
    param.function2 = speed_mpn_mod_1_2;
    param.min_is_always = 1;
    param.name = "MOD_1_1_TO_MOD_1_2_THRESHOLD";
    param.min_size = 2;
    one (&mod_1_1_to_mod_1_2_threshold, &param);

    param.function = speed_mpn_mod_1_2;
    param.function2 = speed_mpn_mod_1_4;
    param.min_is_always = 1;
    param.name = "MOD_1_2_TO_MOD_1_4_THRESHOLD";
    param.min_size = 1;
    one (&mod_1_2_to_mod_1_4_threshold, &param);

    if (mod_1_1_to_mod_1_2_threshold >= mod_1_2_to_mod_1_4_threshold)
      {
	/* Never use mod_1_2, measure mod_1_1 -> mod_1_4 */
	mod_1_2_to_mod_1_4_threshold = 0;

	param.function = speed_mpn_mod_1_1;
	param.function2 = speed_mpn_mod_1_4;
	param.min_is_always = 1;
	param.name = "MOD_1_1_TO_MOD_1_4_THRESHOLD fake";
	param.min_size = 2;
	one (&mod_1_1_to_mod_1_2_threshold, &param);
      }

    param.function = speed_mpn_mod_1_tune;
    param.function2 = NULL;
    param.name = "MOD_1U_TO_MOD_1_1_THRESHOLD";
    param.min_size = 2;
    param.min_is_always = 0;
    one (&mod_1u_to_mod_1_1_threshold, &param);

    if (mod_1u_to_mod_1_1_threshold >= mod_1_1_to_mod_1_2_threshold)
      mod_1_1_to_mod_1_2_threshold = 0;
    if (mod_1u_to_mod_1_1_threshold >= mod_1_2_to_mod_1_4_threshold)
      mod_1_2_to_mod_1_4_threshold = 0;

    print_define_remark ("MOD_1U_TO_MOD_1_1_THRESHOLD", mod_1u_to_mod_1_1_threshold, NULL);
    print_define_remark ("MOD_1_1_TO_MOD_1_2_THRESHOLD", mod_1_1_to_mod_1_2_threshold,
			 mod_1_1_to_mod_1_2_threshold == 0 ? "never mpn_mod_1_1p" : NULL);
    print_define_remark ("MOD_1_2_TO_MOD_1_4_THRESHOLD", mod_1_2_to_mod_1_4_threshold,
			 mod_1_2_to_mod_1_4_threshold == 0 ? "never mpn_mod_1s_2p" : NULL);
  }

  {
    static struct param_t  param;

    param.check_size = 256;

    param.name = "PREINV_MOD_1_TO_MOD_1_THRESHOLD";
    s.r = randlimb_norm ();
    param.function = speed_mpn_preinv_mod_1;
    param.function2 = speed_mpn_mod_1_tune;
    param.min_size = 1;
    one (&preinv_mod_1_to_mod_1_threshold, &param);
  }
}


/* A non-zero DIVREM_1_UNNORM_THRESHOLD (or DIVREM_1_NORM_THRESHOLD) would
   imply that udiv_qrnnd_preinv is worth using, but it seems most
   straightforward to compare mpn_preinv_divrem_1 and mpn_divrem_1_div
   directly.  */

void
tune_preinv_divrem_1 (void)
{
  static struct param_t  param;
  speed_function_t  divrem_1;
  const char        *divrem_1_name;
  double            t1, t2;

  if (GMP_NAIL_BITS != 0)
    {
      print_define_remark ("USE_PREINV_DIVREM_1", 0, "no preinv with nails");
      return;
    }

  /* Any native version of mpn_preinv_divrem_1 is assumed to exist because
     it's faster than mpn_divrem_1.  */
  if (HAVE_NATIVE_mpn_preinv_divrem_1)
    {
      print_define_remark ("USE_PREINV_DIVREM_1", 1, "native");
      return;
    }

  /* If udiv_qrnnd_preinv is the only division method then of course
     mpn_preinv_divrem_1 should be used.  */
  if (UDIV_PREINV_ALWAYS)
    {
      print_define_remark ("USE_PREINV_DIVREM_1", 1, "preinv always");
      return;
    }

  /* If we've got an assembler version of mpn_divrem_1, then compare against
     that, not the mpn_divrem_1_div generic C.  */
  if (HAVE_NATIVE_mpn_divrem_1)
    {
      divrem_1 = speed_mpn_divrem_1;
      divrem_1_name = "mpn_divrem_1";
    }
  else
    {
      divrem_1 = speed_mpn_divrem_1_div;
      divrem_1_name = "mpn_divrem_1_div";
    }

  param.data_high = DATA_HIGH_LT_R; /* allow skip one division */
  s.size = 200;                     /* generous but not too big */
  /* Divisor, nonzero.  Unnormalized so as to exercise the shift!=0 case,
     since in general that's probably most common, though in fact for a
     64-bit limb mp_bases[10].big_base is normalized.  */
  s.r = urandom() & (GMP_NUMB_MASK >> 4);
  if (s.r == 0) s.r = 123;

  t1 = tuneup_measure (speed_mpn_preinv_divrem_1, &param, &s);
  t2 = tuneup_measure (divrem_1, &param, &s);
  if (t1 == -1.0 || t2 == -1.0)
    {
      printf ("Oops, can't measure mpn_preinv_divrem_1 and %s at %ld\n",
              divrem_1_name, (long) s.size);
      abort ();
    }
  if (option_trace >= 1)
    printf ("size=%ld, mpn_preinv_divrem_1 %.9f, %s %.9f\n",
            (long) s.size, t1, divrem_1_name, t2);

  print_define_remark ("USE_PREINV_DIVREM_1", (mp_size_t) (t1 < t2), NULL);
}



void
tune_divrem_2 (void)
{
  static struct param_t  param;

  /* No support for tuning native assembler code, do that by hand and put
     the results in the .asm file, and there's no need for such thresholds
     to appear in gmp-mparam.h.  */
  if (HAVE_NATIVE_mpn_divrem_2)
    return;

  if (GMP_NAIL_BITS != 0)
    {
      print_define_remark ("DIVREM_2_THRESHOLD", MP_SIZE_T_MAX,
                           "no preinv with nails");
      return;
    }

  if (UDIV_PREINV_ALWAYS)
    {
      print_define_remark ("DIVREM_2_THRESHOLD", 0L, "preinv always");
      return;
    }

  /* Tune for the integer part of mpn_divrem_2.  This will very possibly be
     a bit out for the fractional part, but that's too bad, the integer part
     is more important.

     min_size must be >=2 since nsize>=2 is required, but is set to 4 to save
     code space if plain division is better only at size==2 or size==3. */
  param.name = "DIVREM_2_THRESHOLD";
  param.check_size = 256;
  param.min_size = 4;
  param.min_is_always = 1;
  param.size_extra = 2;      /* does qsize==nsize-2 divisions */
  param.stop_factor = 2.0;

  s.r = randlimb_norm ();
  param.function = speed_mpn_divrem_2;
  one (&divrem_2_threshold, &param);
}

void
tune_div_qr_2 (void)
{
  static struct param_t  param;
  param.name = "DIV_QR_2_PI2_THRESHOLD";
  param.function = speed_mpn_div_qr_2n;
  param.check_size = 500;
  param.min_size = 4;
  one (&div_qr_2_pi2_threshold, &param);
}

/* mpn_divexact_1 is vaguely expected to be used on smallish divisors, so
   tune for that.  Its speed can differ on odd or even divisor, so take an
   average threshold for the two.

   mpn_divrem_1 can vary with high<divisor or not, whereas mpn_divexact_1
   might not vary that way, but don't test this since high<divisor isn't
   expected to occur often with small divisors.  */

void
tune_divexact_1 (void)
{
  static struct param_t  param;
  mp_size_t  thresh[2], average;
  int        low, i;

  /* Any native mpn_divexact_1 is assumed to incorporate all the speed of a
     full mpn_divrem_1.  */
  if (HAVE_NATIVE_mpn_divexact_1)
    {
      print_define_remark ("DIVEXACT_1_THRESHOLD", 0, "always (native)");
      return;
    }

  ASSERT_ALWAYS (tuned_speed_mpn_divrem_1 != NULL);

  param.name = "DIVEXACT_1_THRESHOLD";
  param.data_high = DATA_HIGH_GE_R;
  param.check_size = 256;
  param.min_size = 2;
  param.stop_factor = 1.5;
  param.function  = tuned_speed_mpn_divrem_1;
  param.function2 = speed_mpn_divexact_1;
  param.noprint = 1;

  print_define_start (param.name);

  for (low = 0; low <= 1; low++)
    {
      s.r = randlimb_half();
      if (low == 0)
        s.r |= 1;
      else
        s.r &= ~CNST_LIMB(7);

      one (&thresh[low], &param);
      if (option_trace)
        printf ("low=%d thresh %ld\n", low, (long) thresh[low]);

      if (thresh[low] == MP_SIZE_T_MAX)
        {
          average = MP_SIZE_T_MAX;
          goto divexact_1_done;
        }
    }

  if (option_trace)
    {
      printf ("average of:");
      for (i = 0; i < numberof(thresh); i++)
        printf (" %ld", (long) thresh[i]);
      printf ("\n");
    }

  average = 0;
  for (i = 0; i < numberof(thresh); i++)
    average += thresh[i];
  average /= numberof(thresh);

  /* If divexact turns out to be better as early as 3 limbs, then use it
     always, so as to reduce code size and conditional jumps.  */
  if (average <= 3)
    average = 0;

 divexact_1_done:
  print_define_end (param.name, average);
}


/* The generic mpn_modexact_1_odd skips a divide step if high<divisor, the
   same as mpn_mod_1, but this might not be true of an assembler
   implementation.  The threshold used is an average based on data where a
   divide can be skipped and where it can't.

   If modexact turns out to be better as early as 3 limbs, then use it
   always, so as to reduce code size and conditional jumps.  */

void
tune_modexact_1_odd (void)
{
  static struct param_t  param;
  mp_size_t  thresh_lt, thresh_ge, average;

#if 0
  /* Any native mpn_modexact_1_odd is assumed to incorporate all the speed
     of a full mpn_mod_1.  */
  if (HAVE_NATIVE_mpn_modexact_1_odd)
    {
      print_define_remark ("BMOD_1_TO_MOD_1_THRESHOLD", MP_SIZE_T_MAX, "always bmod_1");
      return;
    }
#endif

  param.name = "BMOD_1_TO_MOD_1_THRESHOLD";
  param.check_size = 256;
  param.min_size = 2;
  param.stop_factor = 1.5;
  param.function  = speed_mpn_modexact_1c_odd;
  param.function2 = speed_mpn_mod_1_tune;
  param.noprint = 1;
  s.r = randlimb_half () | 1;

  print_define_start (param.name);

  param.data_high = DATA_HIGH_LT_R;
  one (&thresh_lt, &param);
  if (option_trace)
    printf ("lt thresh %ld\n", (long) thresh_lt);

  average = thresh_lt;
  if (thresh_lt != MP_SIZE_T_MAX)
    {
      param.data_high = DATA_HIGH_GE_R;
      one (&thresh_ge, &param);
      if (option_trace)
        printf ("ge thresh %ld\n", (long) thresh_ge);

      if (thresh_ge != MP_SIZE_T_MAX)
        {
          average = (thresh_ge + thresh_lt) / 2;
          if (thresh_ge <= 3)
            average = 0;
        }
    }

  print_define_end (param.name, average);
}


void
tune_jacobi_base (void)
{
  static struct param_t  param;
  double   t1, t2, t3, t4;
  int      method;

  s.size = GMP_LIMB_BITS * 3 / 4;

  t1 = tuneup_measure (speed_mpn_jacobi_base_1, &param, &s);
  if (option_trace >= 1)
    printf ("size=%ld, mpn_jacobi_base_1 %.9f\n", (long) s.size, t1);

  t2 = tuneup_measure (speed_mpn_jacobi_base_2, &param, &s);
  if (option_trace >= 1)
    printf ("size=%ld, mpn_jacobi_base_2 %.9f\n", (long) s.size, t2);

  t3 = tuneup_measure (speed_mpn_jacobi_base_3, &param, &s);
  if (option_trace >= 1)
    printf ("size=%ld, mpn_jacobi_base_3 %.9f\n", (long) s.size, t3);

  t4 = tuneup_measure (speed_mpn_jacobi_base_4, &param, &s);
  if (option_trace >= 1)
    printf ("size=%ld, mpn_jacobi_base_4 %.9f\n", (long) s.size, t4);

  if (t1 == -1.0 || t2 == -1.0 || t3 == -1.0 || t4 == -1.0)
    {
      printf ("Oops, can't measure all mpn_jacobi_base methods at %ld\n",
              (long) s.size);
      abort ();
    }

  if (t1 < t2 && t1 < t3 && t1 < t4)
    method = 1;
  else if (t2 < t3 && t2 < t4)
    method = 2;
  else if (t3 < t4)
    method = 3;
  else
    method = 4;

  print_define ("JACOBI_BASE_METHOD", method);
}


void
tune_get_str (void)
{
  /* Tune for decimal, it being most common.  Some rough testing suggests
     other bases are different, but not by very much.  */
  s.r = 10;
  {
    static struct param_t  param;
    GET_STR_PRECOMPUTE_THRESHOLD = 0;
    param.name = "GET_STR_DC_THRESHOLD";
    param.function = speed_mpn_get_str;
    param.min_size = 4;
    param.max_size = GET_STR_THRESHOLD_LIMIT;
    one (&get_str_dc_threshold, &param);
  }
  {
    static struct param_t  param;
    param.name = "GET_STR_PRECOMPUTE_THRESHOLD";
    param.function = speed_mpn_get_str;
    param.min_size = GET_STR_DC_THRESHOLD;
    param.max_size = GET_STR_THRESHOLD_LIMIT;
    one (&get_str_precompute_threshold, &param);
  }
}


double
speed_mpn_pre_set_str (struct speed_params *s)
{
  unsigned char *str;
  mp_ptr     wp;
  mp_size_t  wn;
  unsigned   i;
  int        base;
  double     t;
  mp_ptr powtab_mem, tp;
  powers_t powtab[GMP_LIMB_BITS];
  mp_size_t un;
  int chars_per_limb;
  TMP_DECL;

  SPEED_RESTRICT_COND (s->size >= 1);

  base = s->r == 0 ? 10 : s->r;
  SPEED_RESTRICT_COND (base >= 2 && base <= 256);

  TMP_MARK;

  str = (unsigned char *) TMP_ALLOC (s->size);
  for (i = 0; i < s->size; i++)
    str[i] = s->xp[i] % base;

  LIMBS_PER_DIGIT_IN_BASE (wn, s->size, base);
  SPEED_TMP_ALLOC_LIMBS (wp, wn, s->align_wp);

  /* use this during development to check wn is big enough */
  /*
  ASSERT_ALWAYS (mpn_set_str (wp, str, s->size, base) <= wn);
  */

  speed_operand_src (s, (mp_ptr) str, s->size/GMP_LIMB_BYTES);
  speed_operand_dst (s, wp, wn);
  speed_cache_fill (s);

  chars_per_limb = mp_bases[base].chars_per_limb;
  un = s->size / chars_per_limb + 1;
  powtab_mem = TMP_BALLOC_LIMBS (mpn_str_powtab_alloc (un));
  size_t n_pows = mpn_compute_powtab (powtab, powtab_mem, un, base);
  powers_t *pt = powtab + n_pows;
  tp = TMP_BALLOC_LIMBS (mpn_dc_set_str_itch (un));

  speed_starttime ();
  i = s->reps;
  do
    {
      mpn_pre_set_str (wp, str, s->size, pt, tp);
    }
  while (--i != 0);
  t = speed_endtime ();

  TMP_FREE;
  return t;
}

void
tune_set_str (void)
{
  s.r = 10;  /* decimal */
  {
    static struct param_t  param;
    SET_STR_PRECOMPUTE_THRESHOLD = 0;
    param.step_factor = 0.01;
    param.name = "SET_STR_DC_THRESHOLD";
    param.function = speed_mpn_pre_set_str;
    param.min_size = 100;
    param.max_size = 50000;
    one (&set_str_dc_threshold, &param);
  }
  {
    static struct param_t  param;
    param.step_factor = 0.02;
    param.name = "SET_STR_PRECOMPUTE_THRESHOLD";
    param.function = speed_mpn_set_str;
    param.min_size = SET_STR_DC_THRESHOLD;
    param.max_size = 100000;
    one (&set_str_precompute_threshold, &param);
  }
}


void
tune_fft_mul (void)
{
  static struct fft_param_t  param;

  if (option_fft_max_size == 0)
    return;

  param.table_name          = "MUL_FFT_TABLE3";
  param.threshold_name      = "MUL_FFT_THRESHOLD";
  param.p_threshold         = &mul_fft_threshold;
  param.modf_threshold_name = "MUL_FFT_MODF_THRESHOLD";
  param.p_modf_threshold    = &mul_fft_modf_threshold;
  param.first_size          = MUL_TOOM33_THRESHOLD / 2;
  param.max_size            = option_fft_max_size;
  param.function            = speed_mpn_fft_mul;
  param.mul_modf_function   = speed_mpn_mul_fft;
  param.mul_function        = speed_mpn_mul_n;
  param.sqr = 0;
  fft (&param);
}


void
tune_fft_sqr (void)
{
  static struct fft_param_t  param;

  if (option_fft_max_size == 0)
    return;

  param.table_name          = "SQR_FFT_TABLE3";
  param.threshold_name      = "SQR_FFT_THRESHOLD";
  param.p_threshold         = &sqr_fft_threshold;
  param.modf_threshold_name = "SQR_FFT_MODF_THRESHOLD";
  param.p_modf_threshold    = &sqr_fft_modf_threshold;
  param.first_size          = SQR_TOOM3_THRESHOLD / 2;
  param.max_size            = option_fft_max_size;
  param.function            = speed_mpn_fft_sqr;
  param.mul_modf_function   = speed_mpn_mul_fft_sqr;
  param.mul_function        = speed_mpn_sqr;
  param.sqr = 1;
  fft (&param);
}

void
tune_fac_ui (void)
{
  static struct param_t  param;

  param.function = speed_mpz_fac_ui_tune;

  param.name = "FAC_DSC_THRESHOLD";
  param.min_size = 70;
  param.max_size = FAC_DSC_THRESHOLD_LIMIT;
  one (&fac_dsc_threshold, &param);

  param.name = "FAC_ODD_THRESHOLD";
  param.min_size = 22;
  param.stop_factor = 1.7;
  param.min_is_always = 1;
  one (&fac_odd_threshold, &param);
}

void
all (void)
{
  time_t  start_time, end_time;
  TMP_DECL;

  TMP_MARK;
  SPEED_TMP_ALLOC_LIMBS (s.xp_block, SPEED_BLOCK_SIZE, 0);
  SPEED_TMP_ALLOC_LIMBS (s.yp_block, SPEED_BLOCK_SIZE, 0);

  mpn_random (s.xp_block, SPEED_BLOCK_SIZE);
  mpn_random (s.yp_block, SPEED_BLOCK_SIZE);

  fprintf (stderr, "Parameters for %s\n", GMP_MPARAM_H_SUGGEST);

  speed_time_init ();
  fprintf (stderr, "Using: %s\n", speed_time_string);

  fprintf (stderr, "speed_precision %d", speed_precision);
  if (speed_unittime == 1.0)
    fprintf (stderr, ", speed_unittime 1 cycle");
  else
    fprintf (stderr, ", speed_unittime %.2e secs", speed_unittime);
  if (speed_cycletime == 1.0 || speed_cycletime == 0.0)
    fprintf (stderr, ", CPU freq unknown\n");
  else
    fprintf (stderr, ", CPU freq %.2f MHz\n", 1e-6/speed_cycletime);

  fprintf (stderr, "DEFAULT_MAX_SIZE %d, fft_max_size %ld\n",
           DEFAULT_MAX_SIZE, (long) option_fft_max_size);
  fprintf (stderr, "\n");

  time (&start_time);
  {
    struct tm  *tp;
    tp = localtime (&start_time);
    printf ("/* Generated by tuneup.c, %d-%02d-%02d, ",
            tp->tm_year+1900, tp->tm_mon+1, tp->tm_mday);

#ifdef __GNUC__
    /* gcc sub-minor version doesn't seem to come through as a define */
    printf ("gcc %d.%d */\n", __GNUC__, __GNUC_MINOR__);
#define PRINTED_COMPILER
#endif
#if defined (__SUNPRO_C)
    printf ("Sun C %d.%d */\n", __SUNPRO_C / 0x100, __SUNPRO_C % 0x100);
#define PRINTED_COMPILER
#endif
#if ! defined (__GNUC__) && defined (__sgi) && defined (_COMPILER_VERSION)
    /* gcc defines __sgi and _COMPILER_VERSION on irix 6, avoid that */
    printf ("MIPSpro C %d.%d.%d */\n",
	    _COMPILER_VERSION / 100,
	    _COMPILER_VERSION / 10 % 10,
	    _COMPILER_VERSION % 10);
#define PRINTED_COMPILER
#endif
#if defined (__DECC) && defined (__DECC_VER)
    printf ("DEC C %d */\n", __DECC_VER);
#define PRINTED_COMPILER
#endif
#if ! defined (PRINTED_COMPILER)
    printf ("system compiler */\n");
#endif
  }
  printf ("\n");

  tune_divrem_1 ();
  tune_mod_1 ();
  tune_preinv_divrem_1 ();
  tune_div_qr_1 ();
#if 0
  tune_divrem_2 ();
#endif
  tune_div_qr_2 ();
  tune_divexact_1 ();
  tune_modexact_1_odd ();
  printf("\n");

  relspeed_div_1_vs_mul_1 ();
  printf("\n");

  tune_mul_n ();
  printf("\n");

  tune_mul ();
  printf("\n");

  tune_sqr ();
  printf("\n");

  tune_mulmid ();
  printf("\n");

  tune_mulmod_bnm1 ();
  tune_sqrmod_bnm1 ();
  printf("\n");

  tune_fft_mul ();
  printf("\n");

  tune_fft_sqr ();
  printf ("\n");

  tune_mullo ();
  tune_sqrlo ();
  printf("\n");

  tune_dc_div ();
  tune_dc_bdiv ();

  printf("\n");
  tune_invertappr ();
  tune_invert ();
  printf("\n");

  tune_binvert ();
  tune_redc ();
  printf("\n");

  tune_mu_div ();
  tune_mu_bdiv ();
  printf("\n");

  tune_powm_sec ();
  printf("\n");

  tune_get_str ();
  tune_set_str ();
  printf("\n");

  tune_fac_ui ();
  printf("\n");

  tune_matrix22_mul ();
  tune_hgcd ();
  tune_hgcd_appr ();
  tune_hgcd_reduce();
  tune_gcd_dc ();
  tune_gcdext_dc ();
  tune_jacobi_base ();
  printf("\n");

  time (&end_time);
  printf ("/* Tuneup completed successfully, took %ld seconds */\n",
          (long) (end_time - start_time));

  TMP_FREE;
}


int
main (int argc, char *argv[])
{
  int  opt;

  /* Unbuffered so if output is redirected to a file it isn't lost if the
     program is killed part way through.  */
  setbuf (stdout, NULL);
  setbuf (stderr, NULL);

  while ((opt = getopt(argc, argv, "f:o:p:t")) != EOF)
    {
      switch (opt) {
      case 'f':
        if (optarg[0] == 't')
          option_fft_trace = 2;
        else
          option_fft_max_size = atol (optarg);
        break;
      case 'o':
        speed_option_set (optarg);
        break;
      case 'p':
        speed_precision = atoi (optarg);
        break;
      case 't':
        option_trace++;
        break;
      case '?':
        exit(1);
      }
    }

  all ();
  exit (0);
}