File: isl_ast_codegen.c

package info (click to toggle)
isl 0.20-2
  • links: PTS
  • area: main
  • in suites: sid
  • size: 12,708 kB
  • sloc: ansic: 125,702; python: 5,770; sh: 4,291; cpp: 1,057; makefile: 497; lisp: 54; perl: 9
file content (5762 lines) | stat: -rw-r--r-- 185,607 bytes parent folder | download | duplicates (4)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595
4596
4597
4598
4599
4600
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
4616
4617
4618
4619
4620
4621
4622
4623
4624
4625
4626
4627
4628
4629
4630
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640
4641
4642
4643
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
4665
4666
4667
4668
4669
4670
4671
4672
4673
4674
4675
4676
4677
4678
4679
4680
4681
4682
4683
4684
4685
4686
4687
4688
4689
4690
4691
4692
4693
4694
4695
4696
4697
4698
4699
4700
4701
4702
4703
4704
4705
4706
4707
4708
4709
4710
4711
4712
4713
4714
4715
4716
4717
4718
4719
4720
4721
4722
4723
4724
4725
4726
4727
4728
4729
4730
4731
4732
4733
4734
4735
4736
4737
4738
4739
4740
4741
4742
4743
4744
4745
4746
4747
4748
4749
4750
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765
4766
4767
4768
4769
4770
4771
4772
4773
4774
4775
4776
4777
4778
4779
4780
4781
4782
4783
4784
4785
4786
4787
4788
4789
4790
4791
4792
4793
4794
4795
4796
4797
4798
4799
4800
4801
4802
4803
4804
4805
4806
4807
4808
4809
4810
4811
4812
4813
4814
4815
4816
4817
4818
4819
4820
4821
4822
4823
4824
4825
4826
4827
4828
4829
4830
4831
4832
4833
4834
4835
4836
4837
4838
4839
4840
4841
4842
4843
4844
4845
4846
4847
4848
4849
4850
4851
4852
4853
4854
4855
4856
4857
4858
4859
4860
4861
4862
4863
4864
4865
4866
4867
4868
4869
4870
4871
4872
4873
4874
4875
4876
4877
4878
4879
4880
4881
4882
4883
4884
4885
4886
4887
4888
4889
4890
4891
4892
4893
4894
4895
4896
4897
4898
4899
4900
4901
4902
4903
4904
4905
4906
4907
4908
4909
4910
4911
4912
4913
4914
4915
4916
4917
4918
4919
4920
4921
4922
4923
4924
4925
4926
4927
4928
4929
4930
4931
4932
4933
4934
4935
4936
4937
4938
4939
4940
4941
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
4952
4953
4954
4955
4956
4957
4958
4959
4960
4961
4962
4963
4964
4965
4966
4967
4968
4969
4970
4971
4972
4973
4974
4975
4976
4977
4978
4979
4980
4981
4982
4983
4984
4985
4986
4987
4988
4989
4990
4991
4992
4993
4994
4995
4996
4997
4998
4999
5000
5001
5002
5003
5004
5005
5006
5007
5008
5009
5010
5011
5012
5013
5014
5015
5016
5017
5018
5019
5020
5021
5022
5023
5024
5025
5026
5027
5028
5029
5030
5031
5032
5033
5034
5035
5036
5037
5038
5039
5040
5041
5042
5043
5044
5045
5046
5047
5048
5049
5050
5051
5052
5053
5054
5055
5056
5057
5058
5059
5060
5061
5062
5063
5064
5065
5066
5067
5068
5069
5070
5071
5072
5073
5074
5075
5076
5077
5078
5079
5080
5081
5082
5083
5084
5085
5086
5087
5088
5089
5090
5091
5092
5093
5094
5095
5096
5097
5098
5099
5100
5101
5102
5103
5104
5105
5106
5107
5108
5109
5110
5111
5112
5113
5114
5115
5116
5117
5118
5119
5120
5121
5122
5123
5124
5125
5126
5127
5128
5129
5130
5131
5132
5133
5134
5135
5136
5137
5138
5139
5140
5141
5142
5143
5144
5145
5146
5147
5148
5149
5150
5151
5152
5153
5154
5155
5156
5157
5158
5159
5160
5161
5162
5163
5164
5165
5166
5167
5168
5169
5170
5171
5172
5173
5174
5175
5176
5177
5178
5179
5180
5181
5182
5183
5184
5185
5186
5187
5188
5189
5190
5191
5192
5193
5194
5195
5196
5197
5198
5199
5200
5201
5202
5203
5204
5205
5206
5207
5208
5209
5210
5211
5212
5213
5214
5215
5216
5217
5218
5219
5220
5221
5222
5223
5224
5225
5226
5227
5228
5229
5230
5231
5232
5233
5234
5235
5236
5237
5238
5239
5240
5241
5242
5243
5244
5245
5246
5247
5248
5249
5250
5251
5252
5253
5254
5255
5256
5257
5258
5259
5260
5261
5262
5263
5264
5265
5266
5267
5268
5269
5270
5271
5272
5273
5274
5275
5276
5277
5278
5279
5280
5281
5282
5283
5284
5285
5286
5287
5288
5289
5290
5291
5292
5293
5294
5295
5296
5297
5298
5299
5300
5301
5302
5303
5304
5305
5306
5307
5308
5309
5310
5311
5312
5313
5314
5315
5316
5317
5318
5319
5320
5321
5322
5323
5324
5325
5326
5327
5328
5329
5330
5331
5332
5333
5334
5335
5336
5337
5338
5339
5340
5341
5342
5343
5344
5345
5346
5347
5348
5349
5350
5351
5352
5353
5354
5355
5356
5357
5358
5359
5360
5361
5362
5363
5364
5365
5366
5367
5368
5369
5370
5371
5372
5373
5374
5375
5376
5377
5378
5379
5380
5381
5382
5383
5384
5385
5386
5387
5388
5389
5390
5391
5392
5393
5394
5395
5396
5397
5398
5399
5400
5401
5402
5403
5404
5405
5406
5407
5408
5409
5410
5411
5412
5413
5414
5415
5416
5417
5418
5419
5420
5421
5422
5423
5424
5425
5426
5427
5428
5429
5430
5431
5432
5433
5434
5435
5436
5437
5438
5439
5440
5441
5442
5443
5444
5445
5446
5447
5448
5449
5450
5451
5452
5453
5454
5455
5456
5457
5458
5459
5460
5461
5462
5463
5464
5465
5466
5467
5468
5469
5470
5471
5472
5473
5474
5475
5476
5477
5478
5479
5480
5481
5482
5483
5484
5485
5486
5487
5488
5489
5490
5491
5492
5493
5494
5495
5496
5497
5498
5499
5500
5501
5502
5503
5504
5505
5506
5507
5508
5509
5510
5511
5512
5513
5514
5515
5516
5517
5518
5519
5520
5521
5522
5523
5524
5525
5526
5527
5528
5529
5530
5531
5532
5533
5534
5535
5536
5537
5538
5539
5540
5541
5542
5543
5544
5545
5546
5547
5548
5549
5550
5551
5552
5553
5554
5555
5556
5557
5558
5559
5560
5561
5562
5563
5564
5565
5566
5567
5568
5569
5570
5571
5572
5573
5574
5575
5576
5577
5578
5579
5580
5581
5582
5583
5584
5585
5586
5587
5588
5589
5590
5591
5592
5593
5594
5595
5596
5597
5598
5599
5600
5601
5602
5603
5604
5605
5606
5607
5608
5609
5610
5611
5612
5613
5614
5615
5616
5617
5618
5619
5620
5621
5622
5623
5624
5625
5626
5627
5628
5629
5630
5631
5632
5633
5634
5635
5636
5637
5638
5639
5640
5641
5642
5643
5644
5645
5646
5647
5648
5649
5650
5651
5652
5653
5654
5655
5656
5657
5658
5659
5660
5661
5662
5663
5664
5665
5666
5667
5668
5669
5670
5671
5672
5673
5674
5675
5676
5677
5678
5679
5680
5681
5682
5683
5684
5685
5686
5687
5688
5689
5690
5691
5692
5693
5694
5695
5696
5697
5698
5699
5700
5701
5702
5703
5704
5705
5706
5707
5708
5709
5710
5711
5712
5713
5714
5715
5716
5717
5718
5719
5720
5721
5722
5723
5724
5725
5726
5727
5728
5729
5730
5731
5732
5733
5734
5735
5736
5737
5738
5739
5740
5741
5742
5743
5744
5745
5746
5747
5748
5749
5750
5751
5752
5753
5754
5755
5756
5757
5758
5759
5760
5761
5762
/*
 * Copyright 2012-2014 Ecole Normale Superieure
 * Copyright 2014      INRIA Rocquencourt
 *
 * Use of this software is governed by the MIT license
 *
 * Written by Sven Verdoolaege,
 * Ecole Normale Superieure, 45 rue d’Ulm, 75230 Paris, France
 * and Inria Paris - Rocquencourt, Domaine de Voluceau - Rocquencourt,
 * B.P. 105 - 78153 Le Chesnay, France
 */

#include <limits.h>
#include <isl/id.h>
#include <isl/val.h>
#include <isl/space.h>
#include <isl/aff.h>
#include <isl/constraint.h>
#include <isl/set.h>
#include <isl/ilp.h>
#include <isl/union_set.h>
#include <isl/union_map.h>
#include <isl/schedule_node.h>
#include <isl_sort.h>
#include <isl_tarjan.h>
#include <isl_ast_private.h>
#include <isl_ast_build_expr.h>
#include <isl_ast_build_private.h>
#include <isl_ast_graft_private.h>

/* Data used in generate_domain.
 *
 * "build" is the input build.
 * "list" collects the results.
 */
struct isl_generate_domain_data {
	isl_ast_build *build;

	isl_ast_graft_list *list;
};

static __isl_give isl_ast_graft_list *generate_next_level(
	__isl_take isl_union_map *executed,
	__isl_take isl_ast_build *build);
static __isl_give isl_ast_graft_list *generate_code(
	__isl_take isl_union_map *executed, __isl_take isl_ast_build *build,
	int internal);

/* Generate an AST for a single domain based on
 * the (non single valued) inverse schedule "executed".
 *
 * We extend the schedule with the iteration domain
 * and continue generating through a call to generate_code.
 *
 * In particular, if executed has the form
 *
 *	S -> D
 *
 * then we continue generating code on
 *
 *	[S -> D] -> D
 *
 * The extended inverse schedule is clearly single valued
 * ensuring that the nested generate_code will not reach this function,
 * but will instead create calls to all elements of D that need
 * to be executed from the current schedule domain.
 */
static isl_stat generate_non_single_valued(__isl_take isl_map *executed,
	struct isl_generate_domain_data *data)
{
	isl_map *identity;
	isl_ast_build *build;
	isl_ast_graft_list *list;

	build = isl_ast_build_copy(data->build);

	identity = isl_set_identity(isl_map_range(isl_map_copy(executed)));
	executed = isl_map_domain_product(executed, identity);
	build = isl_ast_build_set_single_valued(build, 1);

	list = generate_code(isl_union_map_from_map(executed), build, 1);

	data->list = isl_ast_graft_list_concat(data->list, list);

	return isl_stat_ok;
}

/* Call the at_each_domain callback, if requested by the user,
 * after recording the current inverse schedule in the build.
 */
static __isl_give isl_ast_graft *at_each_domain(__isl_take isl_ast_graft *graft,
	__isl_keep isl_map *executed, __isl_keep isl_ast_build *build)
{
	if (!graft || !build)
		return isl_ast_graft_free(graft);
	if (!build->at_each_domain)
		return graft;

	build = isl_ast_build_copy(build);
	build = isl_ast_build_set_executed(build,
			isl_union_map_from_map(isl_map_copy(executed)));
	if (!build)
		return isl_ast_graft_free(graft);

	graft->node = build->at_each_domain(graft->node,
					build, build->at_each_domain_user);
	isl_ast_build_free(build);

	if (!graft->node)
		graft = isl_ast_graft_free(graft);

	return graft;
}

/* Generate a call expression for the single executed
 * domain element "map" and put a guard around it based its (simplified)
 * domain.  "executed" is the original inverse schedule from which "map"
 * has been derived.  In particular, "map" is either identical to "executed"
 * or it is the result of gisting "executed" with respect to the build domain.
 * "executed" is only used if there is an at_each_domain callback.
 *
 * At this stage, any pending constraints in the build can no longer
 * be simplified with respect to any enforced constraints since
 * the call node does not have any enforced constraints.
 * Since all pending constraints not covered by any enforced constraints
 * will be added as a guard to the graft in create_node_scaled,
 * even in the eliminated case, the pending constraints
 * can be considered to have been generated by outer constructs.
 *
 * If the user has set an at_each_domain callback, it is called
 * on the constructed call expression node.
 */
static isl_stat add_domain(__isl_take isl_map *executed,
	__isl_take isl_map *map, struct isl_generate_domain_data *data)
{
	isl_ast_build *build;
	isl_ast_graft *graft;
	isl_ast_graft_list *list;
	isl_set *guard, *pending;

	build = isl_ast_build_copy(data->build);
	pending = isl_ast_build_get_pending(build);
	build = isl_ast_build_replace_pending_by_guard(build, pending);

	guard = isl_map_domain(isl_map_copy(map));
	guard = isl_set_compute_divs(guard);
	guard = isl_set_coalesce(guard);
	guard = isl_set_gist(guard, isl_ast_build_get_generated(build));
	guard = isl_ast_build_specialize(build, guard);

	graft = isl_ast_graft_alloc_domain(map, build);
	graft = at_each_domain(graft, executed, build);
	isl_ast_build_free(build);
	isl_map_free(executed);
	graft = isl_ast_graft_add_guard(graft, guard, data->build);

	list = isl_ast_graft_list_from_ast_graft(graft);
	data->list = isl_ast_graft_list_concat(data->list, list);

	return isl_stat_ok;
}

/* Generate an AST for a single domain based on
 * the inverse schedule "executed" and add it to data->list.
 *
 * If there is more than one domain element associated to the current
 * schedule "time", then we need to continue the generation process
 * in generate_non_single_valued.
 * Note that the inverse schedule being single-valued may depend
 * on constraints that are only available in the original context
 * domain specified by the user.  We therefore first introduce
 * some of the constraints of data->build->domain.  In particular,
 * we intersect with a single-disjunct approximation of this set.
 * We perform this approximation to avoid further splitting up
 * the executed relation, possibly introducing a disjunctive guard
 * on the statement.
 *
 * On the other hand, we only perform the test after having taken the gist
 * of the domain as the resulting map is the one from which the call
 * expression is constructed.  Using this map to construct the call
 * expression usually yields simpler results in cases where the original
 * map is not obviously single-valued.
 * If the original map is obviously single-valued, then the gist
 * operation is skipped.
 *
 * Because we perform the single-valuedness test on the gisted map,
 * we may in rare cases fail to recognize that the inverse schedule
 * is single-valued.  This becomes problematic if this happens
 * from the recursive call through generate_non_single_valued
 * as we would then end up in an infinite recursion.
 * We therefore check if we are inside a call to generate_non_single_valued
 * and revert to the ungisted map if the gisted map turns out not to be
 * single-valued.
 *
 * Otherwise, call add_domain to generate a call expression (with guard) and
 * to call the at_each_domain callback, if any.
 */
static isl_stat generate_domain(__isl_take isl_map *executed, void *user)
{
	struct isl_generate_domain_data *data = user;
	isl_set *domain;
	isl_map *map = NULL;
	int empty, sv;

	domain = isl_ast_build_get_domain(data->build);
	domain = isl_set_from_basic_set(isl_set_simple_hull(domain));
	executed = isl_map_intersect_domain(executed, domain);
	empty = isl_map_is_empty(executed);
	if (empty < 0)
		goto error;
	if (empty) {
		isl_map_free(executed);
		return isl_stat_ok;
	}

	sv = isl_map_plain_is_single_valued(executed);
	if (sv < 0)
		goto error;
	if (sv)
		return add_domain(executed, isl_map_copy(executed), data);

	executed = isl_map_coalesce(executed);
	map = isl_map_copy(executed);
	map = isl_ast_build_compute_gist_map_domain(data->build, map);
	sv = isl_map_is_single_valued(map);
	if (sv < 0)
		goto error;
	if (!sv) {
		isl_map_free(map);
		if (data->build->single_valued)
			map = isl_map_copy(executed);
		else
			return generate_non_single_valued(executed, data);
	}

	return add_domain(executed, map, data);
error:
	isl_map_free(map);
	isl_map_free(executed);
	return isl_stat_error;
}

/* Call build->create_leaf to a create "leaf" node in the AST,
 * encapsulate the result in an isl_ast_graft and return the result
 * as a 1-element list.
 *
 * Note that the node returned by the user may be an entire tree.
 *
 * Since the node itself cannot enforce any constraints, we turn
 * all pending constraints into guards and add them to the resulting
 * graft to ensure that they will be generated.
 *
 * Before we pass control to the user, we first clear some information
 * from the build that is (presumbably) only meaningful
 * for the current code generation.
 * This includes the create_leaf callback itself, so we make a copy
 * of the build first.
 */
static __isl_give isl_ast_graft_list *call_create_leaf(
	__isl_take isl_union_map *executed, __isl_take isl_ast_build *build)
{
	isl_set *guard;
	isl_ast_node *node;
	isl_ast_graft *graft;
	isl_ast_build *user_build;

	guard = isl_ast_build_get_pending(build);
	user_build = isl_ast_build_copy(build);
	user_build = isl_ast_build_replace_pending_by_guard(user_build,
							isl_set_copy(guard));
	user_build = isl_ast_build_set_executed(user_build, executed);
	user_build = isl_ast_build_clear_local_info(user_build);
	if (!user_build)
		node = NULL;
	else
		node = build->create_leaf(user_build, build->create_leaf_user);
	graft = isl_ast_graft_alloc(node, build);
	graft = isl_ast_graft_add_guard(graft, guard, build);
	isl_ast_build_free(build);
	return isl_ast_graft_list_from_ast_graft(graft);
}

static __isl_give isl_ast_graft_list *build_ast_from_child(
	__isl_take isl_ast_build *build, __isl_take isl_schedule_node *node,
	__isl_take isl_union_map *executed);

/* Generate an AST after having handled the complete schedule
 * of this call to the code generator or the complete band
 * if we are generating an AST from a schedule tree.
 *
 * If we are inside a band node, then move on to the child of the band.
 *
 * If the user has specified a create_leaf callback, control
 * is passed to the user in call_create_leaf.
 *
 * Otherwise, we generate one or more calls for each individual
 * domain in generate_domain.
 */
static __isl_give isl_ast_graft_list *generate_inner_level(
	__isl_take isl_union_map *executed, __isl_take isl_ast_build *build)
{
	isl_ctx *ctx;
	struct isl_generate_domain_data data = { build };

	if (!build || !executed)
		goto error;

	if (isl_ast_build_has_schedule_node(build)) {
		isl_schedule_node *node;
		node = isl_ast_build_get_schedule_node(build);
		build = isl_ast_build_reset_schedule_node(build);
		return build_ast_from_child(build, node, executed);
	}

	if (build->create_leaf)
		return call_create_leaf(executed, build);

	ctx = isl_union_map_get_ctx(executed);
	data.list = isl_ast_graft_list_alloc(ctx, 0);
	if (isl_union_map_foreach_map(executed, &generate_domain, &data) < 0)
		data.list = isl_ast_graft_list_free(data.list);

	if (0)
error:		data.list = NULL;
	isl_ast_build_free(build);
	isl_union_map_free(executed);
	return data.list;
}

/* Call the before_each_for callback, if requested by the user.
 */
static __isl_give isl_ast_node *before_each_for(__isl_take isl_ast_node *node,
	__isl_keep isl_ast_build *build)
{
	isl_id *id;

	if (!node || !build)
		return isl_ast_node_free(node);
	if (!build->before_each_for)
		return node;
	id = build->before_each_for(build, build->before_each_for_user);
	node = isl_ast_node_set_annotation(node, id);
	return node;
}

/* Call the after_each_for callback, if requested by the user.
 */
static __isl_give isl_ast_graft *after_each_for(__isl_take isl_ast_graft *graft,
	__isl_keep isl_ast_build *build)
{
	if (!graft || !build)
		return isl_ast_graft_free(graft);
	if (!build->after_each_for)
		return graft;
	graft->node = build->after_each_for(graft->node, build,
						build->after_each_for_user);
	if (!graft->node)
		return isl_ast_graft_free(graft);
	return graft;
}

/* Plug in all the know values of the current and outer dimensions
 * in the domain of "executed".  In principle, we only need to plug
 * in the known value of the current dimension since the values of
 * outer dimensions have been plugged in already.
 * However, it turns out to be easier to just plug in all known values.
 */
static __isl_give isl_union_map *plug_in_values(
	__isl_take isl_union_map *executed, __isl_keep isl_ast_build *build)
{
	return isl_ast_build_substitute_values_union_map_domain(build,
								    executed);
}

/* Check if the constraint "c" is a lower bound on dimension "pos",
 * an upper bound, or independent of dimension "pos".
 */
static int constraint_type(isl_constraint *c, int pos)
{
	if (isl_constraint_is_lower_bound(c, isl_dim_set, pos))
		return 1;
	if (isl_constraint_is_upper_bound(c, isl_dim_set, pos))
		return 2;
	return 0;
}

/* Compare the types of the constraints "a" and "b",
 * resulting in constraints that are independent of "depth"
 * to be sorted before the lower bounds on "depth", which in
 * turn are sorted before the upper bounds on "depth".
 */
static int cmp_constraint(__isl_keep isl_constraint *a,
	__isl_keep isl_constraint *b, void *user)
{
	int *depth = user;
	int t1 = constraint_type(a, *depth);
	int t2 = constraint_type(b, *depth);

	return t1 - t2;
}

/* Extract a lower bound on dimension "pos" from constraint "c".
 *
 * If the constraint is of the form
 *
 *	a x + f(...) >= 0
 *
 * then we essentially return
 *
 *	l = ceil(-f(...)/a)
 *
 * However, if the current dimension is strided, then we need to make
 * sure that the lower bound we construct is of the form
 *
 *	f + s a
 *
 * with f the offset and s the stride.
 * We therefore compute
 *
 *	f + s * ceil((l - f)/s)
 */
static __isl_give isl_aff *lower_bound(__isl_keep isl_constraint *c,
	int pos, __isl_keep isl_ast_build *build)
{
	isl_aff *aff;

	aff = isl_constraint_get_bound(c, isl_dim_set, pos);
	aff = isl_aff_ceil(aff);

	if (isl_ast_build_has_stride(build, pos)) {
		isl_aff *offset;
		isl_val *stride;

		offset = isl_ast_build_get_offset(build, pos);
		stride = isl_ast_build_get_stride(build, pos);

		aff = isl_aff_sub(aff, isl_aff_copy(offset));
		aff = isl_aff_scale_down_val(aff, isl_val_copy(stride));
		aff = isl_aff_ceil(aff);
		aff = isl_aff_scale_val(aff, stride);
		aff = isl_aff_add(aff, offset);
	}

	aff = isl_ast_build_compute_gist_aff(build, aff);

	return aff;
}

/* Return the exact lower bound (or upper bound if "upper" is set)
 * of "domain" as a piecewise affine expression.
 *
 * If we are computing a lower bound (of a strided dimension), then
 * we need to make sure it is of the form
 *
 *	f + s a
 *
 * where f is the offset and s is the stride.
 * We therefore need to include the stride constraint before computing
 * the minimum.
 */
static __isl_give isl_pw_aff *exact_bound(__isl_keep isl_set *domain,
	__isl_keep isl_ast_build *build, int upper)
{
	isl_set *stride;
	isl_map *it_map;
	isl_pw_aff *pa;
	isl_pw_multi_aff *pma;

	domain = isl_set_copy(domain);
	if (!upper) {
		stride = isl_ast_build_get_stride_constraint(build);
		domain = isl_set_intersect(domain, stride);
	}
	it_map = isl_ast_build_map_to_iterator(build, domain);
	if (upper)
		pma = isl_map_lexmax_pw_multi_aff(it_map);
	else
		pma = isl_map_lexmin_pw_multi_aff(it_map);
	pa = isl_pw_multi_aff_get_pw_aff(pma, 0);
	isl_pw_multi_aff_free(pma);
	pa = isl_ast_build_compute_gist_pw_aff(build, pa);
	pa = isl_pw_aff_coalesce(pa);

	return pa;
}

/* Callback for sorting the isl_pw_aff_list passed to reduce_list and
 * remove_redundant_lower_bounds.
 */
static int reduce_list_cmp(__isl_keep isl_pw_aff *a, __isl_keep isl_pw_aff *b,
	void *user)
{
	return isl_pw_aff_plain_cmp(a, b);
}

/* Given a list of lower bounds "list", remove those that are redundant
 * with respect to the other bounds in "list" and the domain of "build".
 *
 * We first sort the bounds in the same way as they would be sorted
 * by set_for_node_expressions so that we can try and remove the last
 * bounds first.
 *
 * For a lower bound to be effective, there needs to be at least
 * one domain element for which it is larger than all other lower bounds.
 * For each lower bound we therefore intersect the domain with
 * the conditions that it is larger than all other bounds and
 * check whether the result is empty.  If so, the bound can be removed.
 */
static __isl_give isl_pw_aff_list *remove_redundant_lower_bounds(
	__isl_take isl_pw_aff_list *list, __isl_keep isl_ast_build *build)
{
	int i, j, n;
	isl_set *domain;

	list = isl_pw_aff_list_sort(list, &reduce_list_cmp, NULL);
	if (!list)
		return NULL;

	n = isl_pw_aff_list_n_pw_aff(list);
	if (n <= 1)
		return list;

	domain = isl_ast_build_get_domain(build);

	for (i = n - 1; i >= 0; --i) {
		isl_pw_aff *pa_i;
		isl_set *domain_i;
		int empty;

		domain_i = isl_set_copy(domain);
		pa_i = isl_pw_aff_list_get_pw_aff(list, i);

		for (j = 0; j < n; ++j) {
			isl_pw_aff *pa_j;
			isl_set *better;

			if (j == i)
				continue;

			pa_j = isl_pw_aff_list_get_pw_aff(list, j);
			better = isl_pw_aff_gt_set(isl_pw_aff_copy(pa_i), pa_j);
			domain_i = isl_set_intersect(domain_i, better);
		}

		empty = isl_set_is_empty(domain_i);

		isl_set_free(domain_i);
		isl_pw_aff_free(pa_i);

		if (empty < 0)
			goto error;
		if (!empty)
			continue;
		list = isl_pw_aff_list_drop(list, i, 1);
		n--;
	}

	isl_set_free(domain);

	return list;
error:
	isl_set_free(domain);
	return isl_pw_aff_list_free(list);
}

/* Extract a lower bound on dimension "pos" from each constraint
 * in "constraints" and return the list of lower bounds.
 * If "constraints" has zero elements, then we extract a lower bound
 * from "domain" instead.
 *
 * If the current dimension is strided, then the lower bound
 * is adjusted by lower_bound to match the stride information.
 * This modification may make one or more lower bounds redundant
 * with respect to the other lower bounds.  We therefore check
 * for this condition and remove the redundant lower bounds.
 */
static __isl_give isl_pw_aff_list *lower_bounds(
	__isl_keep isl_constraint_list *constraints, int pos,
	__isl_keep isl_set *domain, __isl_keep isl_ast_build *build)
{
	isl_ctx *ctx;
	isl_pw_aff_list *list;
	int i, n;

	if (!build)
		return NULL;

	n = isl_constraint_list_n_constraint(constraints);
	if (n == 0) {
		isl_pw_aff *pa;
		pa = exact_bound(domain, build, 0);
		return isl_pw_aff_list_from_pw_aff(pa);
	}

	ctx = isl_ast_build_get_ctx(build);
	list = isl_pw_aff_list_alloc(ctx,n);

	for (i = 0; i < n; ++i) {
		isl_aff *aff;
		isl_constraint *c;

		c = isl_constraint_list_get_constraint(constraints, i);
		aff = lower_bound(c, pos, build);
		isl_constraint_free(c);
		list = isl_pw_aff_list_add(list, isl_pw_aff_from_aff(aff));
	}

	if (isl_ast_build_has_stride(build, pos))
		list = remove_redundant_lower_bounds(list, build);

	return list;
}

/* Extract an upper bound on dimension "pos" from each constraint
 * in "constraints" and return the list of upper bounds.
 * If "constraints" has zero elements, then we extract an upper bound
 * from "domain" instead.
 */
static __isl_give isl_pw_aff_list *upper_bounds(
	__isl_keep isl_constraint_list *constraints, int pos,
	__isl_keep isl_set *domain, __isl_keep isl_ast_build *build)
{
	isl_ctx *ctx;
	isl_pw_aff_list *list;
	int i, n;

	n = isl_constraint_list_n_constraint(constraints);
	if (n == 0) {
		isl_pw_aff *pa;
		pa = exact_bound(domain, build, 1);
		return isl_pw_aff_list_from_pw_aff(pa);
	}

	ctx = isl_ast_build_get_ctx(build);
	list = isl_pw_aff_list_alloc(ctx,n);

	for (i = 0; i < n; ++i) {
		isl_aff *aff;
		isl_constraint *c;

		c = isl_constraint_list_get_constraint(constraints, i);
		aff = isl_constraint_get_bound(c, isl_dim_set, pos);
		isl_constraint_free(c);
		aff = isl_aff_floor(aff);
		list = isl_pw_aff_list_add(list, isl_pw_aff_from_aff(aff));
	}

	return list;
}

/* Return an isl_ast_expr that performs the reduction of type "type"
 * on AST expressions corresponding to the elements in "list".
 *
 * The list is assumed to contain at least one element.
 * If the list contains exactly one element, then the returned isl_ast_expr
 * simply computes that affine expression.
 * If the list contains more than one element, then we sort it
 * using a fairly abitrary but hopefully reasonably stable order.
 */
static __isl_give isl_ast_expr *reduce_list(enum isl_ast_op_type type,
	__isl_keep isl_pw_aff_list *list, __isl_keep isl_ast_build *build)
{
	int i, n;
	isl_ctx *ctx;
	isl_ast_expr *expr;

	if (!list)
		return NULL;

	n = isl_pw_aff_list_n_pw_aff(list);

	if (n == 1)
		return isl_ast_build_expr_from_pw_aff_internal(build,
				isl_pw_aff_list_get_pw_aff(list, 0));

	ctx = isl_pw_aff_list_get_ctx(list);
	expr = isl_ast_expr_alloc_op(ctx, type, n);
	if (!expr)
		return NULL;

	list = isl_pw_aff_list_copy(list);
	list = isl_pw_aff_list_sort(list, &reduce_list_cmp, NULL);
	if (!list)
		return isl_ast_expr_free(expr);

	for (i = 0; i < n; ++i) {
		isl_ast_expr *expr_i;

		expr_i = isl_ast_build_expr_from_pw_aff_internal(build,
				isl_pw_aff_list_get_pw_aff(list, i));
		if (!expr_i)
			goto error;
		expr->u.op.args[i] = expr_i;
	}

	isl_pw_aff_list_free(list);
	return expr;
error:
	isl_pw_aff_list_free(list);
	isl_ast_expr_free(expr);
	return NULL;
}

/* Add guards implied by the "generated constraints",
 * but not (necessarily) enforced by the generated AST to "guard".
 * In particular, if there is any stride constraints,
 * then add the guard implied by those constraints.
 * If we have generated a degenerate loop, then add the guard
 * implied by "bounds" on the outer dimensions, i.e., the guard
 * that ensures that the single value actually exists.
 * Since there may also be guards implied by a combination
 * of these constraints, we first combine them before
 * deriving the implied constraints.
 */
static __isl_give isl_set *add_implied_guards(__isl_take isl_set *guard,
	int degenerate, __isl_keep isl_basic_set *bounds,
	__isl_keep isl_ast_build *build)
{
	int depth, has_stride;
	isl_space *space;
	isl_set *dom, *set;

	depth = isl_ast_build_get_depth(build);
	has_stride = isl_ast_build_has_stride(build, depth);
	if (!has_stride && !degenerate)
		return guard;

	space = isl_basic_set_get_space(bounds);
	dom = isl_set_universe(space);

	if (degenerate) {
		bounds = isl_basic_set_copy(bounds);
		bounds = isl_basic_set_drop_constraints_not_involving_dims(
					bounds, isl_dim_set, depth, 1);
		set = isl_set_from_basic_set(bounds);
		dom = isl_set_intersect(dom, set);
	}

	if (has_stride) {
		set = isl_ast_build_get_stride_constraint(build);
		dom = isl_set_intersect(dom, set);
	}

	dom = isl_set_eliminate(dom, isl_dim_set, depth, 1);
	dom = isl_ast_build_compute_gist(build, dom);
	guard = isl_set_intersect(guard, dom);

	return guard;
}

/* Update "graft" based on "sub_build" for the degenerate case.
 *
 * "build" is the build in which graft->node was created
 * "sub_build" contains information about the current level itself,
 * including the single value attained.
 *
 * We set the initialization part of the for loop to the single
 * value attained by the current dimension.
 * The increment and condition are not strictly needed as the are known
 * to be "1" and "iterator <= value" respectively.
 */
static __isl_give isl_ast_graft *refine_degenerate(
	__isl_take isl_ast_graft *graft, __isl_keep isl_ast_build *build,
	__isl_keep isl_ast_build *sub_build)
{
	isl_pw_aff *value;

	if (!graft || !sub_build)
		return isl_ast_graft_free(graft);

	value = isl_pw_aff_copy(sub_build->value);

	graft->node->u.f.init = isl_ast_build_expr_from_pw_aff_internal(build,
						value);
	if (!graft->node->u.f.init)
		return isl_ast_graft_free(graft);

	return graft;
}

/* Return the intersection of constraints in "list" as a set.
 */
static __isl_give isl_set *intersect_constraints(
	__isl_keep isl_constraint_list *list)
{
	int i, n;
	isl_basic_set *bset;

	n = isl_constraint_list_n_constraint(list);
	if (n < 1)
		isl_die(isl_constraint_list_get_ctx(list), isl_error_internal,
			"expecting at least one constraint", return NULL);

	bset = isl_basic_set_from_constraint(
				isl_constraint_list_get_constraint(list, 0));
	for (i = 1; i < n; ++i) {
		isl_basic_set *bset_i;

		bset_i = isl_basic_set_from_constraint(
				isl_constraint_list_get_constraint(list, i));
		bset = isl_basic_set_intersect(bset, bset_i);
	}

	return isl_set_from_basic_set(bset);
}

/* Compute the constraints on the outer dimensions enforced by
 * graft->node and add those constraints to graft->enforced,
 * in case the upper bound is expressed as a set "upper".
 *
 * In particular, if l(...) is a lower bound in "lower", and
 *
 *	-a i + f(...) >= 0		or	a i <= f(...)
 *
 * is an upper bound ocnstraint on the current dimension i,
 * then the for loop enforces the constraint
 *
 *	-a l(...) + f(...) >= 0		or	a l(...) <= f(...)
 *
 * We therefore simply take each lower bound in turn, plug it into
 * the upper bounds and compute the intersection over all lower bounds.
 *
 * If a lower bound is a rational expression, then
 * isl_basic_set_preimage_multi_aff will force this rational
 * expression to have only integer values.  However, the loop
 * itself does not enforce this integrality constraint.  We therefore
 * use the ceil of the lower bounds instead of the lower bounds themselves.
 * Other constraints will make sure that the for loop is only executed
 * when each of the lower bounds attains an integral value.
 * In particular, potentially rational values only occur in
 * lower_bound if the offset is a (seemingly) rational expression,
 * but then outer conditions will make sure that this rational expression
 * only attains integer values.
 */
static __isl_give isl_ast_graft *set_enforced_from_set(
	__isl_take isl_ast_graft *graft,
	__isl_keep isl_pw_aff_list *lower, int pos, __isl_keep isl_set *upper)
{
	isl_space *space;
	isl_basic_set *enforced;
	isl_pw_multi_aff *pma;
	int i, n;

	if (!graft || !lower)
		return isl_ast_graft_free(graft);

	space = isl_set_get_space(upper);
	enforced = isl_basic_set_universe(isl_space_copy(space));

	space = isl_space_map_from_set(space);
	pma = isl_pw_multi_aff_identity(space);

	n = isl_pw_aff_list_n_pw_aff(lower);
	for (i = 0; i < n; ++i) {
		isl_pw_aff *pa;
		isl_set *enforced_i;
		isl_basic_set *hull;
		isl_pw_multi_aff *pma_i;

		pa = isl_pw_aff_list_get_pw_aff(lower, i);
		pa = isl_pw_aff_ceil(pa);
		pma_i = isl_pw_multi_aff_copy(pma);
		pma_i = isl_pw_multi_aff_set_pw_aff(pma_i, pos, pa);
		enforced_i = isl_set_copy(upper);
		enforced_i = isl_set_preimage_pw_multi_aff(enforced_i, pma_i);
		hull = isl_set_simple_hull(enforced_i);
		enforced = isl_basic_set_intersect(enforced, hull);
	}

	isl_pw_multi_aff_free(pma);

	graft = isl_ast_graft_enforce(graft, enforced);

	return graft;
}

/* Compute the constraints on the outer dimensions enforced by
 * graft->node and add those constraints to graft->enforced,
 * in case the upper bound is expressed as
 * a list of affine expressions "upper".
 *
 * The enforced condition is that each lower bound expression is less
 * than or equal to each upper bound expression.
 */
static __isl_give isl_ast_graft *set_enforced_from_list(
	__isl_take isl_ast_graft *graft,
	__isl_keep isl_pw_aff_list *lower, __isl_keep isl_pw_aff_list *upper)
{
	isl_set *cond;
	isl_basic_set *enforced;

	lower = isl_pw_aff_list_copy(lower);
	upper = isl_pw_aff_list_copy(upper);
	cond = isl_pw_aff_list_le_set(lower, upper);
	enforced = isl_set_simple_hull(cond);
	graft = isl_ast_graft_enforce(graft, enforced);

	return graft;
}

/* Does "aff" have a negative constant term?
 */
static isl_stat aff_constant_is_negative(__isl_take isl_set *set,
	__isl_take isl_aff *aff, void *user)
{
	int *neg = user;
	isl_val *v;

	v = isl_aff_get_constant_val(aff);
	*neg = isl_val_is_neg(v);
	isl_val_free(v);
	isl_set_free(set);
	isl_aff_free(aff);

	return *neg ? isl_stat_ok : isl_stat_error;
}

/* Does "pa" have a negative constant term over its entire domain?
 */
static isl_stat pw_aff_constant_is_negative(__isl_take isl_pw_aff *pa,
	void *user)
{
	isl_stat r;
	int *neg = user;

	r = isl_pw_aff_foreach_piece(pa, &aff_constant_is_negative, user);
	isl_pw_aff_free(pa);

	return (*neg && r >= 0) ? isl_stat_ok : isl_stat_error;
}

/* Does each element in "list" have a negative constant term?
 *
 * The callback terminates the iteration as soon an element has been
 * found that does not have a negative constant term.
 */
static int list_constant_is_negative(__isl_keep isl_pw_aff_list *list)
{
	int neg = 1;

	if (isl_pw_aff_list_foreach(list,
				&pw_aff_constant_is_negative, &neg) < 0 && neg)
		return -1;

	return neg;
}

/* Add 1 to each of the elements in "list", where each of these elements
 * is defined over the internal schedule space of "build".
 */
static __isl_give isl_pw_aff_list *list_add_one(
	__isl_take isl_pw_aff_list *list, __isl_keep isl_ast_build *build)
{
	int i, n;
	isl_space *space;
	isl_aff *aff;
	isl_pw_aff *one;

	space = isl_ast_build_get_space(build, 1);
	aff = isl_aff_zero_on_domain(isl_local_space_from_space(space));
	aff = isl_aff_add_constant_si(aff, 1);
	one = isl_pw_aff_from_aff(aff);

	n = isl_pw_aff_list_n_pw_aff(list);
	for (i = 0; i < n; ++i) {
		isl_pw_aff *pa;
		pa = isl_pw_aff_list_get_pw_aff(list, i);
		pa = isl_pw_aff_add(pa, isl_pw_aff_copy(one));
		list = isl_pw_aff_list_set_pw_aff(list, i, pa);
	}

	isl_pw_aff_free(one);

	return list;
}

/* Set the condition part of the for node graft->node in case
 * the upper bound is represented as a list of piecewise affine expressions.
 *
 * In particular, set the condition to
 *
 *	iterator <= min(list of upper bounds)
 *
 * If each of the upper bounds has a negative constant term, then
 * set the condition to
 *
 *	iterator < min(list of (upper bound + 1)s)
 *
 */
static __isl_give isl_ast_graft *set_for_cond_from_list(
	__isl_take isl_ast_graft *graft, __isl_keep isl_pw_aff_list *list,
	__isl_keep isl_ast_build *build)
{
	int neg;
	isl_ast_expr *bound, *iterator, *cond;
	enum isl_ast_op_type type = isl_ast_op_le;

	if (!graft || !list)
		return isl_ast_graft_free(graft);

	neg = list_constant_is_negative(list);
	if (neg < 0)
		return isl_ast_graft_free(graft);
	list = isl_pw_aff_list_copy(list);
	if (neg) {
		list = list_add_one(list, build);
		type = isl_ast_op_lt;
	}

	bound = reduce_list(isl_ast_op_min, list, build);
	iterator = isl_ast_expr_copy(graft->node->u.f.iterator);
	cond = isl_ast_expr_alloc_binary(type, iterator, bound);
	graft->node->u.f.cond = cond;

	isl_pw_aff_list_free(list);
	if (!graft->node->u.f.cond)
		return isl_ast_graft_free(graft);
	return graft;
}

/* Set the condition part of the for node graft->node in case
 * the upper bound is represented as a set.
 */
static __isl_give isl_ast_graft *set_for_cond_from_set(
	__isl_take isl_ast_graft *graft, __isl_keep isl_set *set,
	__isl_keep isl_ast_build *build)
{
	isl_ast_expr *cond;

	if (!graft)
		return NULL;

	cond = isl_ast_build_expr_from_set_internal(build, isl_set_copy(set));
	graft->node->u.f.cond = cond;
	if (!graft->node->u.f.cond)
		return isl_ast_graft_free(graft);
	return graft;
}

/* Construct an isl_ast_expr for the increment (i.e., stride) of
 * the current dimension.
 */
static __isl_give isl_ast_expr *for_inc(__isl_keep isl_ast_build *build)
{
	int depth;
	isl_val *v;
	isl_ctx *ctx;

	if (!build)
		return NULL;
	ctx = isl_ast_build_get_ctx(build);
	depth = isl_ast_build_get_depth(build);

	if (!isl_ast_build_has_stride(build, depth))
		return isl_ast_expr_alloc_int_si(ctx, 1);

	v = isl_ast_build_get_stride(build, depth);
	return isl_ast_expr_from_val(v);
}

/* Should we express the loop condition as
 *
 *	iterator <= min(list of upper bounds)
 *
 * or as a conjunction of constraints?
 *
 * The first is constructed from a list of upper bounds.
 * The second is constructed from a set.
 *
 * If there are no upper bounds in "constraints", then this could mean
 * that "domain" simply doesn't have an upper bound or that we didn't
 * pick any upper bound.  In the first case, we want to generate the
 * loop condition as a(n empty) conjunction of constraints
 * In the second case, we will compute
 * a single upper bound from "domain" and so we use the list form.
 *
 * If there are upper bounds in "constraints",
 * then we use the list form iff the atomic_upper_bound option is set.
 */
static int use_upper_bound_list(isl_ctx *ctx, int n_upper,
	__isl_keep isl_set *domain, int depth)
{
	if (n_upper > 0)
		return isl_options_get_ast_build_atomic_upper_bound(ctx);
	else
		return isl_set_dim_has_upper_bound(domain, isl_dim_set, depth);
}

/* Fill in the expressions of the for node in graft->node.
 *
 * In particular,
 * - set the initialization part of the loop to the maximum of the lower bounds
 * - extract the increment from the stride of the current dimension
 * - construct the for condition either based on a list of upper bounds
 *	or on a set of upper bound constraints.
 */
static __isl_give isl_ast_graft *set_for_node_expressions(
	__isl_take isl_ast_graft *graft, __isl_keep isl_pw_aff_list *lower,
	int use_list, __isl_keep isl_pw_aff_list *upper_list,
	__isl_keep isl_set *upper_set, __isl_keep isl_ast_build *build)
{
	isl_ast_node *node;

	if (!graft)
		return NULL;

	build = isl_ast_build_copy(build);

	node = graft->node;
	node->u.f.init = reduce_list(isl_ast_op_max, lower, build);
	node->u.f.inc = for_inc(build);

	if (!node->u.f.init || !node->u.f.inc)
		graft = isl_ast_graft_free(graft);

	if (use_list)
		graft = set_for_cond_from_list(graft, upper_list, build);
	else
		graft = set_for_cond_from_set(graft, upper_set, build);

	isl_ast_build_free(build);

	return graft;
}

/* Update "graft" based on "bounds" and "domain" for the generic,
 * non-degenerate, case.
 *
 * "c_lower" and "c_upper" contain the lower and upper bounds
 * that the loop node should express.
 * "domain" is the subset of the intersection of the constraints
 * for which some code is executed.
 *
 * There may be zero lower bounds or zero upper bounds in "constraints"
 * in case the list of constraints was created
 * based on the atomic option or based on separation with explicit bounds.
 * In that case, we use "domain" to derive lower and/or upper bounds.
 *
 * We first compute a list of one or more lower bounds.
 *
 * Then we decide if we want to express the condition as
 *
 *	iterator <= min(list of upper bounds)
 *
 * or as a conjunction of constraints.
 *
 * The set of enforced constraints is then computed either based on
 * a list of upper bounds or on a set of upper bound constraints.
 * We do not compute any enforced constraints if we were forced
 * to compute a lower or upper bound using exact_bound.  The domains
 * of the resulting expressions may imply some bounds on outer dimensions
 * that we do not want to appear in the enforced constraints since
 * they are not actually enforced by the corresponding code.
 *
 * Finally, we fill in the expressions of the for node.
 */
static __isl_give isl_ast_graft *refine_generic_bounds(
	__isl_take isl_ast_graft *graft,
	__isl_take isl_constraint_list *c_lower,
	__isl_take isl_constraint_list *c_upper,
	__isl_keep isl_set *domain, __isl_keep isl_ast_build *build)
{
	int depth;
	isl_ctx *ctx;
	isl_pw_aff_list *lower;
	int use_list;
	isl_set *upper_set = NULL;
	isl_pw_aff_list *upper_list = NULL;
	int n_lower, n_upper;

	if (!graft || !c_lower || !c_upper || !build)
		goto error;

	depth = isl_ast_build_get_depth(build);
	ctx = isl_ast_graft_get_ctx(graft);

	n_lower = isl_constraint_list_n_constraint(c_lower);
	n_upper = isl_constraint_list_n_constraint(c_upper);

	use_list = use_upper_bound_list(ctx, n_upper, domain, depth);

	lower = lower_bounds(c_lower, depth, domain, build);

	if (use_list)
		upper_list = upper_bounds(c_upper, depth, domain, build);
	else if (n_upper > 0)
		upper_set = intersect_constraints(c_upper);
	else
		upper_set = isl_set_universe(isl_set_get_space(domain));

	if (n_lower == 0 || n_upper == 0)
		;
	else if (use_list)
		graft = set_enforced_from_list(graft, lower, upper_list);
	else
		graft = set_enforced_from_set(graft, lower, depth, upper_set);

	graft = set_for_node_expressions(graft, lower, use_list, upper_list,
					upper_set, build);

	isl_pw_aff_list_free(lower);
	isl_pw_aff_list_free(upper_list);
	isl_set_free(upper_set);
	isl_constraint_list_free(c_lower);
	isl_constraint_list_free(c_upper);

	return graft;
error:
	isl_constraint_list_free(c_lower);
	isl_constraint_list_free(c_upper);
	return isl_ast_graft_free(graft);
}

/* Internal data structure used inside count_constraints to keep
 * track of the number of constraints that are independent of dimension "pos",
 * the lower bounds in "pos" and the upper bounds in "pos".
 */
struct isl_ast_count_constraints_data {
	int pos;

	int n_indep;
	int n_lower;
	int n_upper;
};

/* Increment data->n_indep, data->lower or data->upper depending
 * on whether "c" is independenct of dimensions data->pos,
 * a lower bound or an upper bound.
 */
static isl_stat count_constraints(__isl_take isl_constraint *c, void *user)
{
	struct isl_ast_count_constraints_data *data = user;

	if (isl_constraint_is_lower_bound(c, isl_dim_set, data->pos))
		data->n_lower++;
	else if (isl_constraint_is_upper_bound(c, isl_dim_set, data->pos))
		data->n_upper++;
	else
		data->n_indep++;

	isl_constraint_free(c);

	return isl_stat_ok;
}

/* Update "graft" based on "bounds" and "domain" for the generic,
 * non-degenerate, case.
 *
 * "list" respresent the list of bounds that need to be encoded by
 * the for loop.  Only the constraints that involve the iterator
 * are relevant here.  The other constraints are taken care of by
 * the caller and are included in the generated constraints of "build".
 * "domain" is the subset of the intersection of the constraints
 * for which some code is executed.
 * "build" is the build in which graft->node was created.
 *
 * We separate lower bounds, upper bounds and constraints that
 * are independent of the loop iterator.
 *
 * The actual for loop bounds are generated in refine_generic_bounds.
 */
static __isl_give isl_ast_graft *refine_generic_split(
	__isl_take isl_ast_graft *graft, __isl_take isl_constraint_list *list,
	__isl_keep isl_set *domain, __isl_keep isl_ast_build *build)
{
	struct isl_ast_count_constraints_data data;
	isl_constraint_list *lower;
	isl_constraint_list *upper;

	if (!list)
		return isl_ast_graft_free(graft);

	data.pos = isl_ast_build_get_depth(build);

	list = isl_constraint_list_sort(list, &cmp_constraint, &data.pos);
	if (!list)
		return isl_ast_graft_free(graft);

	data.n_indep = data.n_lower = data.n_upper = 0;
	if (isl_constraint_list_foreach(list, &count_constraints, &data) < 0) {
		isl_constraint_list_free(list);
		return isl_ast_graft_free(graft);
	}

	lower = isl_constraint_list_drop(list, 0, data.n_indep);
	upper = isl_constraint_list_copy(lower);
	lower = isl_constraint_list_drop(lower, data.n_lower, data.n_upper);
	upper = isl_constraint_list_drop(upper, 0, data.n_lower);

	return refine_generic_bounds(graft, lower, upper, domain, build);
}

/* Update "graft" based on "bounds" and "domain" for the generic,
 * non-degenerate, case.
 *
 * "bounds" respresent the bounds that need to be encoded by
 * the for loop (or a guard around the for loop).
 * "domain" is the subset of "bounds" for which some code is executed.
 * "build" is the build in which graft->node was created.
 *
 * We break up "bounds" into a list of constraints and continue with
 * refine_generic_split.
 */
static __isl_give isl_ast_graft *refine_generic(
	__isl_take isl_ast_graft *graft,
	__isl_keep isl_basic_set *bounds, __isl_keep isl_set *domain,
	__isl_keep isl_ast_build *build)
{
	isl_constraint_list *list;

	if (!build || !graft)
		return isl_ast_graft_free(graft);

	list = isl_basic_set_get_constraint_list(bounds);

	graft = refine_generic_split(graft, list, domain, build);

	return graft;
}

/* Create a for node for the current level.
 *
 * Mark the for node degenerate if "degenerate" is set.
 */
static __isl_give isl_ast_node *create_for(__isl_keep isl_ast_build *build,
	int degenerate)
{
	int depth;
	isl_id *id;
	isl_ast_node *node;

	if (!build)
		return NULL;

	depth = isl_ast_build_get_depth(build);
	id = isl_ast_build_get_iterator_id(build, depth);
	node = isl_ast_node_alloc_for(id);
	if (degenerate)
		node = isl_ast_node_for_mark_degenerate(node);

	return node;
}

/* If the ast_build_exploit_nested_bounds option is set, then return
 * the constraints enforced by all elements in "list".
 * Otherwise, return the universe.
 */
static __isl_give isl_basic_set *extract_shared_enforced(
	__isl_keep isl_ast_graft_list *list, __isl_keep isl_ast_build *build)
{
	isl_ctx *ctx;
	isl_space *space;

	if (!list)
		return NULL;

	ctx = isl_ast_graft_list_get_ctx(list);
	if (isl_options_get_ast_build_exploit_nested_bounds(ctx))
		return isl_ast_graft_list_extract_shared_enforced(list, build);

	space = isl_ast_build_get_space(build, 1);
	return isl_basic_set_universe(space);
}

/* Return the pending constraints of "build" that are not already taken
 * care of (by a combination of "enforced" and the generated constraints
 * of "build").
 */
static __isl_give isl_set *extract_pending(__isl_keep isl_ast_build *build,
	__isl_keep isl_basic_set *enforced)
{
	isl_set *guard, *context;

	guard = isl_ast_build_get_pending(build);
	context = isl_set_from_basic_set(isl_basic_set_copy(enforced));
	context = isl_set_intersect(context,
					isl_ast_build_get_generated(build));
	return isl_set_gist(guard, context);
}

/* Create an AST node for the current dimension based on
 * the schedule domain "bounds" and return the node encapsulated
 * in an isl_ast_graft.
 *
 * "executed" is the current inverse schedule, taking into account
 * the bounds in "bounds"
 * "domain" is the domain of "executed", with inner dimensions projected out.
 * It may be a strict subset of "bounds" in case "bounds" was created
 * based on the atomic option or based on separation with explicit bounds.
 *
 * "domain" may satisfy additional equalities that result
 * from intersecting "executed" with "bounds" in add_node.
 * It may also satisfy some global constraints that were dropped out because
 * we performed separation with explicit bounds.
 * The very first step is then to copy these constraints to "bounds".
 *
 * Since we may be calling before_each_for and after_each_for
 * callbacks, we record the current inverse schedule in the build.
 *
 * We consider three builds,
 * "build" is the one in which the current level is created,
 * "body_build" is the build in which the next level is created,
 * "sub_build" is essentially the same as "body_build", except that
 * the depth has not been increased yet.
 *
 * "build" already contains information (in strides and offsets)
 * about the strides at the current level, but this information is not
 * reflected in the build->domain.
 * We first add this information and the "bounds" to the sub_build->domain.
 * isl_ast_build_set_loop_bounds adds the stride information and
 * checks whether the current dimension attains
 * only a single value and whether this single value can be represented using
 * a single affine expression.
 * In the first case, the current level is considered "degenerate".
 * In the second, sub-case, the current level is considered "eliminated".
 * Eliminated levels don't need to be reflected in the AST since we can
 * simply plug in the affine expression.  For degenerate, but non-eliminated,
 * levels, we do introduce a for node, but mark is as degenerate so that
 * it can be printed as an assignment of the single value to the loop
 * "iterator".
 *
 * If the current level is eliminated, we explicitly plug in the value
 * for the current level found by isl_ast_build_set_loop_bounds in the
 * inverse schedule.  This ensures that if we are working on a slice
 * of the domain based on information available in the inverse schedule
 * and the build domain, that then this information is also reflected
 * in the inverse schedule.  This operation also eliminates the current
 * dimension from the inverse schedule making sure no inner dimensions depend
 * on the current dimension.  Otherwise, we create a for node, marking
 * it degenerate if appropriate.  The initial for node is still incomplete
 * and will be completed in either refine_degenerate or refine_generic.
 *
 * We then generate a sequence of grafts for the next level,
 * create a surrounding graft for the current level and insert
 * the for node we created (if the current level is not eliminated).
 * Before creating a graft for the current level, we first extract
 * hoistable constraints from the child guards and combine them
 * with the pending constraints in the build.  These constraints
 * are used to simplify the child guards and then added to the guard
 * of the current graft to ensure that they will be generated.
 * If the hoisted guard is a disjunction, then we use it directly
 * to gist the guards on the children before intersect it with the
 * pending constraints.  We do so because this disjunction is typically
 * identical to the guards on the children such that these guards
 * can be effectively removed completely.  After the intersection,
 * the gist operation would have a harder time figuring this out.
 *
 * Finally, we set the bounds of the for loop in either
 * refine_degenerate or refine_generic.
 * We do so in a context where the pending constraints of the build
 * have been replaced by the guard of the current graft.
 */
static __isl_give isl_ast_graft *create_node_scaled(
	__isl_take isl_union_map *executed,
	__isl_take isl_basic_set *bounds, __isl_take isl_set *domain,
	__isl_take isl_ast_build *build)
{
	int depth;
	int degenerate;
	isl_bool eliminated;
	isl_basic_set *hull;
	isl_basic_set *enforced;
	isl_set *guard, *hoisted;
	isl_ast_node *node = NULL;
	isl_ast_graft *graft;
	isl_ast_graft_list *children;
	isl_ast_build *sub_build;
	isl_ast_build *body_build;

	domain = isl_ast_build_eliminate_divs(build, domain);
	domain = isl_set_detect_equalities(domain);
	hull = isl_set_unshifted_simple_hull(isl_set_copy(domain));
	bounds = isl_basic_set_intersect(bounds, hull);
	build = isl_ast_build_set_executed(build, isl_union_map_copy(executed));

	depth = isl_ast_build_get_depth(build);
	sub_build = isl_ast_build_copy(build);
	bounds = isl_basic_set_remove_redundancies(bounds);
	bounds = isl_ast_build_specialize_basic_set(sub_build, bounds);
	sub_build = isl_ast_build_set_loop_bounds(sub_build,
						isl_basic_set_copy(bounds));
	degenerate = isl_ast_build_has_value(sub_build);
	eliminated = isl_ast_build_has_affine_value(sub_build, depth);
	if (degenerate < 0 || eliminated < 0)
		executed = isl_union_map_free(executed);
	if (!degenerate)
		bounds = isl_ast_build_compute_gist_basic_set(build, bounds);
	sub_build = isl_ast_build_set_pending_generated(sub_build,
						isl_basic_set_copy(bounds));
	if (eliminated)
		executed = plug_in_values(executed, sub_build);
	else
		node = create_for(build, degenerate);

	body_build = isl_ast_build_copy(sub_build);
	body_build = isl_ast_build_increase_depth(body_build);
	if (!eliminated)
		node = before_each_for(node, body_build);
	children = generate_next_level(executed,
				    isl_ast_build_copy(body_build));

	enforced = extract_shared_enforced(children, build);
	guard = extract_pending(sub_build, enforced);
	hoisted = isl_ast_graft_list_extract_hoistable_guard(children, build);
	if (isl_set_n_basic_set(hoisted) > 1)
		children = isl_ast_graft_list_gist_guards(children,
						    isl_set_copy(hoisted));
	guard = isl_set_intersect(guard, hoisted);
	if (!eliminated)
		guard = add_implied_guards(guard, degenerate, bounds, build);

	graft = isl_ast_graft_alloc_from_children(children,
			    isl_set_copy(guard), enforced, build, sub_build);

	if (!eliminated) {
		isl_ast_build *for_build;

		graft = isl_ast_graft_insert_for(graft, node);
		for_build = isl_ast_build_copy(build);
		for_build = isl_ast_build_replace_pending_by_guard(for_build,
							isl_set_copy(guard));
		if (degenerate)
			graft = refine_degenerate(graft, for_build, sub_build);
		else
			graft = refine_generic(graft, bounds,
					domain, for_build);
		isl_ast_build_free(for_build);
	}
	isl_set_free(guard);
	if (!eliminated)
		graft = after_each_for(graft, body_build);

	isl_ast_build_free(body_build);
	isl_ast_build_free(sub_build);
	isl_ast_build_free(build);
	isl_basic_set_free(bounds);
	isl_set_free(domain);

	return graft;
}

/* Internal data structure for checking if all constraints involving
 * the input dimension "depth" are such that the other coefficients
 * are multiples of "m", reducing "m" if they are not.
 * If "m" is reduced all the way down to "1", then the check has failed
 * and we break out of the iteration.
 */
struct isl_check_scaled_data {
	int depth;
	isl_val *m;
};

/* If constraint "c" involves the input dimension data->depth,
 * then make sure that all the other coefficients are multiples of data->m,
 * reducing data->m if needed.
 * Break out of the iteration if data->m has become equal to "1".
 */
static isl_stat constraint_check_scaled(__isl_take isl_constraint *c,
	void *user)
{
	struct isl_check_scaled_data *data = user;
	int i, j, n;
	enum isl_dim_type t[] = { isl_dim_param, isl_dim_in, isl_dim_out,
				    isl_dim_div };

	if (!isl_constraint_involves_dims(c, isl_dim_in, data->depth, 1)) {
		isl_constraint_free(c);
		return isl_stat_ok;
	}

	for (i = 0; i < 4; ++i) {
		n = isl_constraint_dim(c, t[i]);
		for (j = 0; j < n; ++j) {
			isl_val *d;

			if (t[i] == isl_dim_in && j == data->depth)
				continue;
			if (!isl_constraint_involves_dims(c, t[i], j, 1))
				continue;
			d = isl_constraint_get_coefficient_val(c, t[i], j);
			data->m = isl_val_gcd(data->m, d);
			if (isl_val_is_one(data->m))
				break;
		}
		if (j < n)
			break;
	}

	isl_constraint_free(c);

	return i < 4 ? isl_stat_error : isl_stat_ok;
}

/* For each constraint of "bmap" that involves the input dimension data->depth,
 * make sure that all the other coefficients are multiples of data->m,
 * reducing data->m if needed.
 * Break out of the iteration if data->m has become equal to "1".
 */
static isl_stat basic_map_check_scaled(__isl_take isl_basic_map *bmap,
	void *user)
{
	isl_stat r;

	r = isl_basic_map_foreach_constraint(bmap,
						&constraint_check_scaled, user);
	isl_basic_map_free(bmap);

	return r;
}

/* For each constraint of "map" that involves the input dimension data->depth,
 * make sure that all the other coefficients are multiples of data->m,
 * reducing data->m if needed.
 * Break out of the iteration if data->m has become equal to "1".
 */
static isl_stat map_check_scaled(__isl_take isl_map *map, void *user)
{
	isl_stat r;

	r = isl_map_foreach_basic_map(map, &basic_map_check_scaled, user);
	isl_map_free(map);

	return r;
}

/* Create an AST node for the current dimension based on
 * the schedule domain "bounds" and return the node encapsulated
 * in an isl_ast_graft.
 *
 * "executed" is the current inverse schedule, taking into account
 * the bounds in "bounds"
 * "domain" is the domain of "executed", with inner dimensions projected out.
 *
 *
 * Before moving on to the actual AST node construction in create_node_scaled,
 * we first check if the current dimension is strided and if we can scale
 * down this stride.  Note that we only do this if the ast_build_scale_strides
 * option is set.
 *
 * In particular, let the current dimension take on values
 *
 *	f + s a
 *
 * with a an integer.  We check if we can find an integer m that (obviously)
 * divides both f and s.
 *
 * If so, we check if the current dimension only appears in constraints
 * where the coefficients of the other variables are multiples of m.
 * We perform this extra check to avoid the risk of introducing
 * divisions by scaling down the current dimension.
 *
 * If so, we scale the current dimension down by a factor of m.
 * That is, we plug in
 *
 *	i = m i'							(1)
 *
 * Note that in principle we could always scale down strided loops
 * by plugging in
 *
 *	i = f + s i'
 *
 * but this may result in i' taking on larger values than the original i,
 * due to the shift by "f".
 * By constrast, the scaling in (1) can only reduce the (absolute) value "i".
 */
static __isl_give isl_ast_graft *create_node(__isl_take isl_union_map *executed,
	__isl_take isl_basic_set *bounds, __isl_take isl_set *domain,
	__isl_take isl_ast_build *build)
{
	struct isl_check_scaled_data data;
	isl_ctx *ctx;
	isl_aff *offset;
	isl_val *d;

	ctx = isl_ast_build_get_ctx(build);
	if (!isl_options_get_ast_build_scale_strides(ctx))
		return create_node_scaled(executed, bounds, domain, build);

	data.depth = isl_ast_build_get_depth(build);
	if (!isl_ast_build_has_stride(build, data.depth))
		return create_node_scaled(executed, bounds, domain, build);

	offset = isl_ast_build_get_offset(build, data.depth);
	data.m = isl_ast_build_get_stride(build, data.depth);
	if (!data.m)
		offset = isl_aff_free(offset);
	offset = isl_aff_scale_down_val(offset, isl_val_copy(data.m));
	d = isl_aff_get_denominator_val(offset);
	if (!d)
		executed = isl_union_map_free(executed);

	if (executed && isl_val_is_divisible_by(data.m, d))
		data.m = isl_val_div(data.m, d);
	else {
		data.m = isl_val_set_si(data.m, 1);
		isl_val_free(d);
	}

	if (!isl_val_is_one(data.m)) {
		if (isl_union_map_foreach_map(executed, &map_check_scaled,
						&data) < 0 &&
		    !isl_val_is_one(data.m))
			executed = isl_union_map_free(executed);
	}

	if (!isl_val_is_one(data.m)) {
		isl_space *space;
		isl_multi_aff *ma;
		isl_aff *aff;
		isl_map *map;
		isl_union_map *umap;

		space = isl_ast_build_get_space(build, 1);
		space = isl_space_map_from_set(space);
		ma = isl_multi_aff_identity(space);
		aff = isl_multi_aff_get_aff(ma, data.depth);
		aff = isl_aff_scale_val(aff, isl_val_copy(data.m));
		ma = isl_multi_aff_set_aff(ma, data.depth, aff);

		bounds = isl_basic_set_preimage_multi_aff(bounds,
						isl_multi_aff_copy(ma));
		domain = isl_set_preimage_multi_aff(domain,
						isl_multi_aff_copy(ma));
		map = isl_map_reverse(isl_map_from_multi_aff(ma));
		umap = isl_union_map_from_map(map);
		executed = isl_union_map_apply_domain(executed,
						isl_union_map_copy(umap));
		build = isl_ast_build_scale_down(build, isl_val_copy(data.m),
						umap);
	}
	isl_aff_free(offset);
	isl_val_free(data.m);

	return create_node_scaled(executed, bounds, domain, build);
}

/* Add the basic set to the list that "user" points to.
 */
static isl_stat collect_basic_set(__isl_take isl_basic_set *bset, void *user)
{
	isl_basic_set_list **list = user;

	*list = isl_basic_set_list_add(*list, bset);

	return isl_stat_ok;
}

/* Extract the basic sets of "set" and collect them in an isl_basic_set_list.
 */
static __isl_give isl_basic_set_list *isl_basic_set_list_from_set(
	__isl_take isl_set *set)
{
	int n;
	isl_ctx *ctx;
	isl_basic_set_list *list;

	if (!set)
		return NULL;

	ctx = isl_set_get_ctx(set);

	n = isl_set_n_basic_set(set);
	list = isl_basic_set_list_alloc(ctx, n);
	if (isl_set_foreach_basic_set(set, &collect_basic_set, &list) < 0)
		list = isl_basic_set_list_free(list);

	isl_set_free(set);
	return list;
}

/* Generate code for the schedule domain "bounds"
 * and add the result to "list".
 *
 * We mainly detect strides here and check if the bounds do not
 * conflict with the current build domain
 * and then pass over control to create_node.
 *
 * "bounds" reflects the bounds on the current dimension and possibly
 * some extra conditions on outer dimensions.
 * It does not, however, include any divs involving the current dimension,
 * so it does not capture any stride constraints.
 * We therefore need to compute that part of the schedule domain that
 * intersects with "bounds" and derive the strides from the result.
 */
static __isl_give isl_ast_graft_list *add_node(
	__isl_take isl_ast_graft_list *list, __isl_take isl_union_map *executed,
	__isl_take isl_basic_set *bounds, __isl_take isl_ast_build *build)
{
	isl_ast_graft *graft;
	isl_set *domain = NULL;
	isl_union_set *uset;
	int empty, disjoint;

	uset = isl_union_set_from_basic_set(isl_basic_set_copy(bounds));
	executed = isl_union_map_intersect_domain(executed, uset);
	empty = isl_union_map_is_empty(executed);
	if (empty < 0)
		goto error;
	if (empty)
		goto done;

	uset = isl_union_map_domain(isl_union_map_copy(executed));
	domain = isl_set_from_union_set(uset);
	domain = isl_ast_build_specialize(build, domain);

	domain = isl_set_compute_divs(domain);
	domain = isl_ast_build_eliminate_inner(build, domain);
	disjoint = isl_set_is_disjoint(domain, build->domain);
	if (disjoint < 0)
		goto error;
	if (disjoint)
		goto done;

	build = isl_ast_build_detect_strides(build, isl_set_copy(domain));

	graft = create_node(executed, bounds, domain,
				isl_ast_build_copy(build));
	list = isl_ast_graft_list_add(list, graft);
	isl_ast_build_free(build);
	return list;
error:
	list = isl_ast_graft_list_free(list);
done:
	isl_set_free(domain);
	isl_basic_set_free(bounds);
	isl_union_map_free(executed);
	isl_ast_build_free(build);
	return list;
}

/* Does any element of i follow or coincide with any element of j
 * at the current depth for equal values of the outer dimensions?
 */
static isl_bool domain_follows_at_depth(__isl_keep isl_basic_set *i,
	__isl_keep isl_basic_set *j, void *user)
{
	int depth = *(int *) user;
	isl_basic_map *test;
	isl_bool empty;
	int l;

	test = isl_basic_map_from_domain_and_range(isl_basic_set_copy(i),
						    isl_basic_set_copy(j));
	for (l = 0; l < depth; ++l)
		test = isl_basic_map_equate(test, isl_dim_in, l,
						isl_dim_out, l);
	test = isl_basic_map_order_ge(test, isl_dim_in, depth,
					isl_dim_out, depth);
	empty = isl_basic_map_is_empty(test);
	isl_basic_map_free(test);

	return empty < 0 ? isl_bool_error : !empty;
}

/* Split up each element of "list" into a part that is related to "bset"
 * according to "gt" and a part that is not.
 * Return a list that consist of "bset" and all the pieces.
 */
static __isl_give isl_basic_set_list *add_split_on(
	__isl_take isl_basic_set_list *list, __isl_take isl_basic_set *bset,
	__isl_keep isl_basic_map *gt)
{
	int i, n;
	isl_basic_set_list *res;

	if (!list)
		bset = isl_basic_set_free(bset);

	gt = isl_basic_map_copy(gt);
	gt = isl_basic_map_intersect_domain(gt, isl_basic_set_copy(bset));
	n = isl_basic_set_list_n_basic_set(list);
	res = isl_basic_set_list_from_basic_set(bset);
	for (i = 0; res && i < n; ++i) {
		isl_basic_set *bset;
		isl_set *set1, *set2;
		isl_basic_map *bmap;
		int empty;

		bset = isl_basic_set_list_get_basic_set(list, i);
		bmap = isl_basic_map_copy(gt);
		bmap = isl_basic_map_intersect_range(bmap, bset);
		bset = isl_basic_map_range(bmap);
		empty = isl_basic_set_is_empty(bset);
		if (empty < 0)
			res = isl_basic_set_list_free(res);
		if (empty)  {
			isl_basic_set_free(bset);
			bset = isl_basic_set_list_get_basic_set(list, i);
			res = isl_basic_set_list_add(res, bset);
			continue;
		}

		res = isl_basic_set_list_add(res, isl_basic_set_copy(bset));
		set1 = isl_set_from_basic_set(bset);
		bset = isl_basic_set_list_get_basic_set(list, i);
		set2 = isl_set_from_basic_set(bset);
		set1 = isl_set_subtract(set2, set1);
		set1 = isl_set_make_disjoint(set1);

		res = isl_basic_set_list_concat(res,
					    isl_basic_set_list_from_set(set1));
	}
	isl_basic_map_free(gt);
	isl_basic_set_list_free(list);
	return res;
}

static __isl_give isl_ast_graft_list *generate_sorted_domains(
	__isl_keep isl_basic_set_list *domain_list,
	__isl_keep isl_union_map *executed,
	__isl_keep isl_ast_build *build);

/* Internal data structure for add_nodes.
 *
 * "executed" and "build" are extra arguments to be passed to add_node.
 * "list" collects the results.
 */
struct isl_add_nodes_data {
	isl_union_map *executed;
	isl_ast_build *build;

	isl_ast_graft_list *list;
};

/* Generate code for the schedule domains in "scc"
 * and add the results to "list".
 *
 * The domains in "scc" form a strongly connected component in the ordering.
 * If the number of domains in "scc" is larger than 1, then this means
 * that we cannot determine a valid ordering for the domains in the component.
 * This should be fairly rare because the individual domains
 * have been made disjoint first.
 * The problem is that the domains may be integrally disjoint but not
 * rationally disjoint.  For example, we may have domains
 *
 *	{ [i,i] : 0 <= i <= 1 }		and	{ [i,1-i] : 0 <= i <= 1 }
 *
 * These two domains have an empty intersection, but their rational
 * relaxations do intersect.  It is impossible to order these domains
 * in the second dimension because the first should be ordered before
 * the second for outer dimension equal to 0, while it should be ordered
 * after for outer dimension equal to 1.
 *
 * This may happen in particular in case of unrolling since the domain
 * of each slice is replaced by its simple hull.
 *
 * For each basic set i in "scc" and for each of the following basic sets j,
 * we split off that part of the basic set i that shares the outer dimensions
 * with j and lies before j in the current dimension.
 * We collect all the pieces in a new list that replaces "scc".
 *
 * While the elements in "scc" should be disjoint, we double-check
 * this property to avoid running into an infinite recursion in case
 * they intersect due to some internal error.
 */
static isl_stat add_nodes(__isl_take isl_basic_set_list *scc, void *user)
{
	struct isl_add_nodes_data *data = user;
	int i, n, depth;
	isl_basic_set *bset, *first;
	isl_basic_set_list *list;
	isl_space *space;
	isl_basic_map *gt;

	n = isl_basic_set_list_n_basic_set(scc);
	bset = isl_basic_set_list_get_basic_set(scc, 0);
	if (n == 1) {
		isl_basic_set_list_free(scc);
		data->list = add_node(data->list,
				isl_union_map_copy(data->executed), bset,
				isl_ast_build_copy(data->build));
		return data->list ? isl_stat_ok : isl_stat_error;
	}

	depth = isl_ast_build_get_depth(data->build);
	space = isl_basic_set_get_space(bset);
	space = isl_space_map_from_set(space);
	gt = isl_basic_map_universe(space);
	for (i = 0; i < depth; ++i)
		gt = isl_basic_map_equate(gt, isl_dim_in, i, isl_dim_out, i);
	gt = isl_basic_map_order_gt(gt, isl_dim_in, depth, isl_dim_out, depth);

	first = isl_basic_set_copy(bset);
	list = isl_basic_set_list_from_basic_set(bset);
	for (i = 1; i < n; ++i) {
		int disjoint;

		bset = isl_basic_set_list_get_basic_set(scc, i);

		disjoint = isl_basic_set_is_disjoint(bset, first);
		if (disjoint < 0)
			list = isl_basic_set_list_free(list);
		else if (!disjoint)
			isl_die(isl_basic_set_list_get_ctx(scc),
				isl_error_internal,
				"basic sets in scc are assumed to be disjoint",
				list = isl_basic_set_list_free(list));

		list = add_split_on(list, bset, gt);
	}
	isl_basic_set_free(first);
	isl_basic_map_free(gt);
	isl_basic_set_list_free(scc);
	scc = list;
	data->list = isl_ast_graft_list_concat(data->list,
		    generate_sorted_domains(scc, data->executed, data->build));
	isl_basic_set_list_free(scc);

	return data->list ? isl_stat_ok : isl_stat_error;
}

/* Sort the domains in "domain_list" according to the execution order
 * at the current depth (for equal values of the outer dimensions),
 * generate code for each of them, collecting the results in a list.
 * If no code is generated (because the intersection of the inverse schedule
 * with the domains turns out to be empty), then an empty list is returned.
 *
 * The caller is responsible for ensuring that the basic sets in "domain_list"
 * are pair-wise disjoint.  It can, however, in principle happen that
 * two basic sets should be ordered one way for one value of the outer
 * dimensions and the other way for some other value of the outer dimensions.
 * We therefore play safe and look for strongly connected components.
 * The function add_nodes takes care of handling non-trivial components.
 */
static __isl_give isl_ast_graft_list *generate_sorted_domains(
	__isl_keep isl_basic_set_list *domain_list,
	__isl_keep isl_union_map *executed, __isl_keep isl_ast_build *build)
{
	isl_ctx *ctx;
	struct isl_add_nodes_data data;
	int depth;
	int n;

	if (!domain_list)
		return NULL;

	ctx = isl_basic_set_list_get_ctx(domain_list);
	n = isl_basic_set_list_n_basic_set(domain_list);
	data.list = isl_ast_graft_list_alloc(ctx, n);
	if (n == 0)
		return data.list;
	if (n == 1)
		return add_node(data.list, isl_union_map_copy(executed),
			isl_basic_set_list_get_basic_set(domain_list, 0),
			isl_ast_build_copy(build));

	depth = isl_ast_build_get_depth(build);
	data.executed = executed;
	data.build = build;
	if (isl_basic_set_list_foreach_scc(domain_list,
					&domain_follows_at_depth, &depth,
					&add_nodes, &data) < 0)
		data.list = isl_ast_graft_list_free(data.list);

	return data.list;
}

/* Do i and j share any values for the outer dimensions?
 */
static isl_bool shared_outer(__isl_keep isl_basic_set *i,
	__isl_keep isl_basic_set *j, void *user)
{
	int depth = *(int *) user;
	isl_basic_map *test;
	isl_bool empty;
	int l;

	test = isl_basic_map_from_domain_and_range(isl_basic_set_copy(i),
						    isl_basic_set_copy(j));
	for (l = 0; l < depth; ++l)
		test = isl_basic_map_equate(test, isl_dim_in, l,
						isl_dim_out, l);
	empty = isl_basic_map_is_empty(test);
	isl_basic_map_free(test);

	return empty < 0 ? isl_bool_error : !empty;
}

/* Internal data structure for generate_sorted_domains_wrap.
 *
 * "n" is the total number of basic sets
 * "executed" and "build" are extra arguments to be passed
 *	to generate_sorted_domains.
 *
 * "single" is set to 1 by generate_sorted_domains_wrap if there
 * is only a single component.
 * "list" collects the results.
 */
struct isl_ast_generate_parallel_domains_data {
	int n;
	isl_union_map *executed;
	isl_ast_build *build;

	int single;
	isl_ast_graft_list *list;
};

/* Call generate_sorted_domains on "scc", fuse the result into a list
 * with either zero or one graft and collect the these single element
 * lists into data->list.
 *
 * If there is only one component, i.e., if the number of basic sets
 * in the current component is equal to the total number of basic sets,
 * then data->single is set to 1 and the result of generate_sorted_domains
 * is not fused.
 */
static isl_stat generate_sorted_domains_wrap(__isl_take isl_basic_set_list *scc,
	void *user)
{
	struct isl_ast_generate_parallel_domains_data *data = user;
	isl_ast_graft_list *list;

	list = generate_sorted_domains(scc, data->executed, data->build);
	data->single = isl_basic_set_list_n_basic_set(scc) == data->n;
	if (!data->single)
		list = isl_ast_graft_list_fuse(list, data->build);
	if (!data->list)
		data->list = list;
	else
		data->list = isl_ast_graft_list_concat(data->list, list);

	isl_basic_set_list_free(scc);
	if (!data->list)
		return isl_stat_error;

	return isl_stat_ok;
}

/* Look for any (weakly connected) components in the "domain_list"
 * of domains that share some values of the outer dimensions.
 * That is, domains in different components do not share any values
 * of the outer dimensions.  This means that these components
 * can be freely reordered.
 * Within each of the components, we sort the domains according
 * to the execution order at the current depth.
 *
 * If there is more than one component, then generate_sorted_domains_wrap
 * fuses the result of each call to generate_sorted_domains
 * into a list with either zero or one graft and collects these (at most)
 * single element lists into a bigger list. This means that the elements of the
 * final list can be freely reordered.  In particular, we sort them
 * according to an arbitrary but fixed ordering to ease merging of
 * graft lists from different components.
 */
static __isl_give isl_ast_graft_list *generate_parallel_domains(
	__isl_keep isl_basic_set_list *domain_list,
	__isl_keep isl_union_map *executed, __isl_keep isl_ast_build *build)
{
	int depth;
	struct isl_ast_generate_parallel_domains_data data;

	if (!domain_list)
		return NULL;

	data.n = isl_basic_set_list_n_basic_set(domain_list);
	if (data.n <= 1)
		return generate_sorted_domains(domain_list, executed, build);

	depth = isl_ast_build_get_depth(build);
	data.list = NULL;
	data.executed = executed;
	data.build = build;
	data.single = 0;
	if (isl_basic_set_list_foreach_scc(domain_list, &shared_outer, &depth,
					    &generate_sorted_domains_wrap,
					    &data) < 0)
		data.list = isl_ast_graft_list_free(data.list);

	if (!data.single)
		data.list = isl_ast_graft_list_sort_guard(data.list);

	return data.list;
}

/* Internal data for separate_domain.
 *
 * "explicit" is set if we only want to use explicit bounds.
 *
 * "domain" collects the separated domains.
 */
struct isl_separate_domain_data {
	isl_ast_build *build;
	int explicit;
	isl_set *domain;
};

/* Extract implicit bounds on the current dimension for the executed "map".
 *
 * The domain of "map" may involve inner dimensions, so we
 * need to eliminate them.
 */
static __isl_give isl_set *implicit_bounds(__isl_take isl_map *map,
	__isl_keep isl_ast_build *build)
{
	isl_set *domain;

	domain = isl_map_domain(map);
	domain = isl_ast_build_eliminate(build, domain);

	return domain;
}

/* Extract explicit bounds on the current dimension for the executed "map".
 *
 * Rather than eliminating the inner dimensions as in implicit_bounds,
 * we simply drop any constraints involving those inner dimensions.
 * The idea is that most bounds that are implied by constraints on the
 * inner dimensions will be enforced by for loops and not by explicit guards.
 * There is then no need to separate along those bounds.
 */
static __isl_give isl_set *explicit_bounds(__isl_take isl_map *map,
	__isl_keep isl_ast_build *build)
{
	isl_set *domain;
	int depth, dim;

	dim = isl_map_dim(map, isl_dim_out);
	map = isl_map_drop_constraints_involving_dims(map, isl_dim_out, 0, dim);

	domain = isl_map_domain(map);
	depth = isl_ast_build_get_depth(build);
	dim = isl_set_dim(domain, isl_dim_set);
	domain = isl_set_detect_equalities(domain);
	domain = isl_set_drop_constraints_involving_dims(domain,
				isl_dim_set, depth + 1, dim - (depth + 1));
	domain = isl_set_remove_divs_involving_dims(domain,
				isl_dim_set, depth, 1);
	domain = isl_set_remove_unknown_divs(domain);

	return domain;
}

/* Split data->domain into pieces that intersect with the range of "map"
 * and pieces that do not intersect with the range of "map"
 * and then add that part of the range of "map" that does not intersect
 * with data->domain.
 */
static isl_stat separate_domain(__isl_take isl_map *map, void *user)
{
	struct isl_separate_domain_data *data = user;
	isl_set *domain;
	isl_set *d1, *d2;

	if (data->explicit)
		domain = explicit_bounds(map, data->build);
	else
		domain = implicit_bounds(map, data->build);

	domain = isl_set_coalesce(domain);
	domain = isl_set_make_disjoint(domain);
	d1 = isl_set_subtract(isl_set_copy(domain), isl_set_copy(data->domain));
	d2 = isl_set_subtract(isl_set_copy(data->domain), isl_set_copy(domain));
	data->domain = isl_set_intersect(data->domain, domain);
	data->domain = isl_set_union(data->domain, d1);
	data->domain = isl_set_union(data->domain, d2);

	return isl_stat_ok;
}

/* Separate the schedule domains of "executed".
 *
 * That is, break up the domain of "executed" into basic sets,
 * such that for each basic set S, every element in S is associated with
 * the same domain spaces.
 *
 * "space" is the (single) domain space of "executed".
 */
static __isl_give isl_set *separate_schedule_domains(
	__isl_take isl_space *space, __isl_take isl_union_map *executed,
	__isl_keep isl_ast_build *build)
{
	struct isl_separate_domain_data data = { build };
	isl_ctx *ctx;

	ctx = isl_ast_build_get_ctx(build);
	data.explicit = isl_options_get_ast_build_separation_bounds(ctx) ==
				    ISL_AST_BUILD_SEPARATION_BOUNDS_EXPLICIT;
	data.domain = isl_set_empty(space);
	if (isl_union_map_foreach_map(executed, &separate_domain, &data) < 0)
		data.domain = isl_set_free(data.domain);

	isl_union_map_free(executed);
	return data.domain;
}

/* Temporary data used during the search for a lower bound for unrolling.
 *
 * "build" is the build in which the unrolling will be performed
 * "domain" is the original set for which to find a lower bound
 * "depth" is the dimension for which to find a lower boudn
 * "expansion" is the expansion that needs to be applied to "domain"
 * in the unrolling that will be performed
 *
 * "lower" is the best lower bound found so far.  It is NULL if we have not
 * found any yet.
 * "n" is the corresponding size.  If lower is NULL, then the value of n
 * is undefined.
 * "n_div" is the maximal number of integer divisions in the first
 * unrolled iteration (after expansion).  It is set to -1 if it hasn't
 * been computed yet.
 */
struct isl_find_unroll_data {
	isl_ast_build *build;
	isl_set *domain;
	int depth;
	isl_basic_map *expansion;

	isl_aff *lower;
	int *n;
	int n_div;
};

/* Return the constraint
 *
 *	i_"depth" = aff + offset
 */
static __isl_give isl_constraint *at_offset(int depth, __isl_keep isl_aff *aff,
	int offset)
{
	aff = isl_aff_copy(aff);
	aff = isl_aff_add_coefficient_si(aff, isl_dim_in, depth, -1);
	aff = isl_aff_add_constant_si(aff, offset);
	return isl_equality_from_aff(aff);
}

/* Update *user to the number of integer divsions in the first element
 * of "ma", if it is larger than the current value.
 */
static isl_stat update_n_div(__isl_take isl_set *set,
	__isl_take isl_multi_aff *ma, void *user)
{
	isl_aff *aff;
	int *n = user;
	int n_div;

	aff = isl_multi_aff_get_aff(ma, 0);
	n_div = isl_aff_dim(aff, isl_dim_div);
	isl_aff_free(aff);
	isl_multi_aff_free(ma);
	isl_set_free(set);

	if (n_div > *n)
		*n = n_div;

	return aff ? isl_stat_ok : isl_stat_error;
}

/* Get the number of integer divisions in the expression for the iterator
 * value at the first slice in the unrolling based on lower bound "lower",
 * taking into account the expansion that needs to be performed on this slice.
 */
static int get_expanded_n_div(struct isl_find_unroll_data *data,
	__isl_keep isl_aff *lower)
{
	isl_constraint *c;
	isl_set *set;
	isl_map *it_map, *expansion;
	isl_pw_multi_aff *pma;
	int n;

	c = at_offset(data->depth, lower, 0);
	set = isl_set_copy(data->domain);
	set = isl_set_add_constraint(set, c);
	expansion = isl_map_from_basic_map(isl_basic_map_copy(data->expansion));
	set = isl_set_apply(set, expansion);
	it_map = isl_ast_build_map_to_iterator(data->build, set);
	pma = isl_pw_multi_aff_from_map(it_map);
	n = 0;
	if (isl_pw_multi_aff_foreach_piece(pma, &update_n_div, &n) < 0)
		n = -1;
	isl_pw_multi_aff_free(pma);

	return n;
}

/* Is the lower bound "lower" with corresponding iteration count "n"
 * better than the one stored in "data"?
 * If there is no upper bound on the iteration count ("n" is infinity) or
 * if the count is too large, then we cannot use this lower bound.
 * Otherwise, if there was no previous lower bound or
 * if the iteration count of the new lower bound is smaller than
 * the iteration count of the previous lower bound, then we consider
 * the new lower bound to be better.
 * If the iteration count is the same, then compare the number
 * of integer divisions that would be needed to express
 * the iterator value at the first slice in the unrolling
 * according to the lower bound.  If we end up computing this
 * number, then store the lowest value in data->n_div.
 */
static int is_better_lower_bound(struct isl_find_unroll_data *data,
	__isl_keep isl_aff *lower, __isl_keep isl_val *n)
{
	int cmp;
	int n_div;

	if (!n)
		return -1;
	if (isl_val_is_infty(n))
		return 0;
	if (isl_val_cmp_si(n, INT_MAX) > 0)
		return 0;
	if (!data->lower)
		return 1;
	cmp = isl_val_cmp_si(n, *data->n);
	if (cmp < 0)
		return 1;
	if (cmp > 0)
		return 0;
	if (data->n_div < 0)
		data->n_div = get_expanded_n_div(data, data->lower);
	if (data->n_div < 0)
		return -1;
	if (data->n_div == 0)
		return 0;
	n_div = get_expanded_n_div(data, lower);
	if (n_div < 0)
		return -1;
	if (n_div >= data->n_div)
		return 0;
	data->n_div = n_div;

	return 1;
}

/* Check if we can use "c" as a lower bound and if it is better than
 * any previously found lower bound.
 *
 * If "c" does not involve the dimension at the current depth,
 * then we cannot use it.
 * Otherwise, let "c" be of the form
 *
 *	i >= f(j)/a
 *
 * We compute the maximal value of
 *
 *	-ceil(f(j)/a)) + i + 1
 *
 * over the domain.  If there is such a value "n", then we know
 *
 *	-ceil(f(j)/a)) + i + 1 <= n
 *
 * or
 *
 *	i < ceil(f(j)/a)) + n
 *
 * meaning that we can use ceil(f(j)/a)) as a lower bound for unrolling.
 * We just need to check if we have found any lower bound before and
 * if the new lower bound is better (smaller n or fewer integer divisions)
 * than the previously found lower bounds.
 */
static isl_stat update_unrolling_lower_bound(struct isl_find_unroll_data *data,
	__isl_keep isl_constraint *c)
{
	isl_aff *aff, *lower;
	isl_val *max;
	int better;

	if (!isl_constraint_is_lower_bound(c, isl_dim_set, data->depth))
		return isl_stat_ok;

	lower = isl_constraint_get_bound(c, isl_dim_set, data->depth);
	lower = isl_aff_ceil(lower);
	aff = isl_aff_copy(lower);
	aff = isl_aff_neg(aff);
	aff = isl_aff_add_coefficient_si(aff, isl_dim_in, data->depth, 1);
	aff = isl_aff_add_constant_si(aff, 1);
	max = isl_set_max_val(data->domain, aff);
	isl_aff_free(aff);

	better = is_better_lower_bound(data, lower, max);
	if (better < 0 || !better) {
		isl_val_free(max);
		isl_aff_free(lower);
		return better < 0 ? isl_stat_error : isl_stat_ok;
	}

	isl_aff_free(data->lower);
	data->lower = lower;
	*data->n = isl_val_get_num_si(max);
	isl_val_free(max);

	return isl_stat_ok;
}

/* Check if we can use "c" as a lower bound and if it is better than
 * any previously found lower bound.
 */
static isl_stat constraint_find_unroll(__isl_take isl_constraint *c, void *user)
{
	struct isl_find_unroll_data *data;
	isl_stat r;

	data = (struct isl_find_unroll_data *) user;
	r = update_unrolling_lower_bound(data, c);
	isl_constraint_free(c);

	return r;
}

/* Look for a lower bound l(i) on the dimension at "depth"
 * and a size n such that "domain" is a subset of
 *
 *	{ [i] : l(i) <= i_d < l(i) + n }
 *
 * where d is "depth" and l(i) depends only on earlier dimensions.
 * Furthermore, try and find a lower bound such that n is as small as possible.
 * In particular, "n" needs to be finite.
 * "build" is the build in which the unrolling will be performed.
 * "expansion" is the expansion that needs to be applied to "domain"
 * in the unrolling that will be performed.
 *
 * Inner dimensions have been eliminated from "domain" by the caller.
 *
 * We first construct a collection of lower bounds on the input set
 * by computing its simple hull.  We then iterate through them,
 * discarding those that we cannot use (either because they do not
 * involve the dimension at "depth" or because they have no corresponding
 * upper bound, meaning that "n" would be unbounded) and pick out the
 * best from the remaining ones.
 *
 * If we cannot find a suitable lower bound, then we consider that
 * to be an error.
 */
static __isl_give isl_aff *find_unroll_lower_bound(
	__isl_keep isl_ast_build *build, __isl_keep isl_set *domain,
	int depth, __isl_keep isl_basic_map *expansion, int *n)
{
	struct isl_find_unroll_data data =
			{ build, domain, depth, expansion, NULL, n, -1 };
	isl_basic_set *hull;

	hull = isl_set_simple_hull(isl_set_copy(domain));

	if (isl_basic_set_foreach_constraint(hull,
					    &constraint_find_unroll, &data) < 0)
		goto error;

	isl_basic_set_free(hull);

	if (!data.lower)
		isl_die(isl_set_get_ctx(domain), isl_error_invalid,
			"cannot find lower bound for unrolling", return NULL);

	return data.lower;
error:
	isl_basic_set_free(hull);
	return isl_aff_free(data.lower);
}

/* Call "fn" on each iteration of the current dimension of "domain".
 * If "init" is not NULL, then it is called with the number of
 * iterations before any call to "fn".
 * Return -1 on failure.
 *
 * Since we are going to be iterating over the individual values,
 * we first check if there are any strides on the current dimension.
 * If there is, we rewrite the current dimension i as
 *
 *		i = stride i' + offset
 *
 * and then iterate over individual values of i' instead.
 *
 * We then look for a lower bound on i' and a size such that the domain
 * is a subset of
 *
 *	{ [j,i'] : l(j) <= i' < l(j) + n }
 *
 * and then take slices of the domain at values of i'
 * between l(j) and l(j) + n - 1.
 *
 * We compute the unshifted simple hull of each slice to ensure that
 * we have a single basic set per offset.  The slicing constraint
 * may get simplified away before the unshifted simple hull is taken
 * and may therefore in some rare cases disappear from the result.
 * We therefore explicitly add the constraint back after computing
 * the unshifted simple hull to ensure that the basic sets
 * remain disjoint.  The constraints that are dropped by taking the hull
 * will be taken into account at the next level, as in the case of the
 * atomic option.
 *
 * Finally, we map i' back to i and call "fn".
 */
static int foreach_iteration(__isl_take isl_set *domain,
	__isl_keep isl_ast_build *build, int (*init)(int n, void *user),
	int (*fn)(__isl_take isl_basic_set *bset, void *user), void *user)
{
	int i, n;
	int empty;
	int depth;
	isl_multi_aff *expansion;
	isl_basic_map *bmap;
	isl_aff *lower = NULL;
	isl_ast_build *stride_build;

	depth = isl_ast_build_get_depth(build);

	domain = isl_ast_build_eliminate_inner(build, domain);
	domain = isl_set_intersect(domain, isl_ast_build_get_domain(build));
	stride_build = isl_ast_build_copy(build);
	stride_build = isl_ast_build_detect_strides(stride_build,
							isl_set_copy(domain));
	expansion = isl_ast_build_get_stride_expansion(stride_build);

	domain = isl_set_preimage_multi_aff(domain,
					    isl_multi_aff_copy(expansion));
	domain = isl_ast_build_eliminate_divs(stride_build, domain);
	isl_ast_build_free(stride_build);

	bmap = isl_basic_map_from_multi_aff(expansion);

	empty = isl_set_is_empty(domain);
	if (empty < 0) {
		n = -1;
	} else if (empty) {
		n = 0;
	} else {
		lower = find_unroll_lower_bound(build, domain, depth, bmap, &n);
		if (!lower)
			n = -1;
	}
	if (n >= 0 && init && init(n, user) < 0)
		n = -1;
	for (i = 0; i < n; ++i) {
		isl_set *set;
		isl_basic_set *bset;
		isl_constraint *slice;

		slice = at_offset(depth, lower, i);
		set = isl_set_copy(domain);
		set = isl_set_add_constraint(set, isl_constraint_copy(slice));
		bset = isl_set_unshifted_simple_hull(set);
		bset = isl_basic_set_add_constraint(bset, slice);
		bset = isl_basic_set_apply(bset, isl_basic_map_copy(bmap));

		if (fn(bset, user) < 0)
			break;
	}

	isl_aff_free(lower);
	isl_set_free(domain);
	isl_basic_map_free(bmap);

	return n < 0 || i < n ? -1 : 0;
}

/* Data structure for storing the results and the intermediate objects
 * of compute_domains.
 *
 * "list" is the main result of the function and contains a list
 * of disjoint basic sets for which code should be generated.
 *
 * "executed" and "build" are inputs to compute_domains.
 * "schedule_domain" is the domain of "executed".
 *
 * "option" contains the domains at the current depth that should by
 * atomic, separated or unrolled.  These domains are as specified by
 * the user, except that inner dimensions have been eliminated and
 * that they have been made pair-wise disjoint.
 *
 * "sep_class" contains the user-specified split into separation classes
 * specialized to the current depth.
 * "done" contains the union of the separation domains that have already
 * been handled.
 */
struct isl_codegen_domains {
	isl_basic_set_list *list;

	isl_union_map *executed;
	isl_ast_build *build;
	isl_set *schedule_domain;

	isl_set *option[4];

	isl_map *sep_class;
	isl_set *done;
};

/* Internal data structure for do_unroll.
 *
 * "domains" stores the results of compute_domains.
 * "class_domain" is the original class domain passed to do_unroll.
 * "unroll_domain" collects the unrolled iterations.
 */
struct isl_ast_unroll_data {
	struct isl_codegen_domains *domains;
	isl_set *class_domain;
	isl_set *unroll_domain;
};

/* Given an iteration of an unrolled domain represented by "bset",
 * add it to data->domains->list.
 * Since we may have dropped some constraints, we intersect with
 * the class domain again to ensure that each element in the list
 * is disjoint from the other class domains.
 */
static int do_unroll_iteration(__isl_take isl_basic_set *bset, void *user)
{
	struct isl_ast_unroll_data *data = user;
	isl_set *set;
	isl_basic_set_list *list;

	set = isl_set_from_basic_set(bset);
	data->unroll_domain = isl_set_union(data->unroll_domain,
					    isl_set_copy(set));
	set = isl_set_intersect(set, isl_set_copy(data->class_domain));
	set = isl_set_make_disjoint(set);
	list = isl_basic_set_list_from_set(set);
	data->domains->list = isl_basic_set_list_concat(data->domains->list,
							list);

	return 0;
}

/* Extend domains->list with a list of basic sets, one for each value
 * of the current dimension in "domain" and remove the corresponding
 * sets from the class domain.  Return the updated class domain.
 * The divs that involve the current dimension have not been projected out
 * from this domain.
 *
 * We call foreach_iteration to iterate over the individual values and
 * in do_unroll_iteration we collect the individual basic sets in
 * domains->list and their union in data->unroll_domain, which is then
 * used to update the class domain.
 */
static __isl_give isl_set *do_unroll(struct isl_codegen_domains *domains,
	__isl_take isl_set *domain, __isl_take isl_set *class_domain)
{
	struct isl_ast_unroll_data data;

	if (!domain)
		return isl_set_free(class_domain);
	if (!class_domain)
		return isl_set_free(domain);

	data.domains = domains;
	data.class_domain = class_domain;
	data.unroll_domain = isl_set_empty(isl_set_get_space(domain));

	if (foreach_iteration(domain, domains->build, NULL,
				&do_unroll_iteration, &data) < 0)
		data.unroll_domain = isl_set_free(data.unroll_domain);

	class_domain = isl_set_subtract(class_domain, data.unroll_domain);

	return class_domain;
}

/* Add domains to domains->list for each individual value of the current
 * dimension, for that part of the schedule domain that lies in the
 * intersection of the option domain and the class domain.
 * Remove the corresponding sets from the class domain and
 * return the updated class domain.
 *
 * We first break up the unroll option domain into individual pieces
 * and then handle each of them separately.  The unroll option domain
 * has been made disjoint in compute_domains_init_options,
 *
 * Note that we actively want to combine different pieces of the
 * schedule domain that have the same value at the current dimension.
 * We therefore need to break up the unroll option domain before
 * intersecting with class and schedule domain, hoping that the
 * unroll option domain specified by the user is relatively simple.
 */
static __isl_give isl_set *compute_unroll_domains(
	struct isl_codegen_domains *domains, __isl_take isl_set *class_domain)
{
	isl_set *unroll_domain;
	isl_basic_set_list *unroll_list;
	int i, n;
	int empty;

	empty = isl_set_is_empty(domains->option[isl_ast_loop_unroll]);
	if (empty < 0)
		return isl_set_free(class_domain);
	if (empty)
		return class_domain;

	unroll_domain = isl_set_copy(domains->option[isl_ast_loop_unroll]);
	unroll_list = isl_basic_set_list_from_set(unroll_domain);

	n = isl_basic_set_list_n_basic_set(unroll_list);
	for (i = 0; i < n; ++i) {
		isl_basic_set *bset;

		bset = isl_basic_set_list_get_basic_set(unroll_list, i);
		unroll_domain = isl_set_from_basic_set(bset);
		unroll_domain = isl_set_intersect(unroll_domain,
						    isl_set_copy(class_domain));
		unroll_domain = isl_set_intersect(unroll_domain,
					isl_set_copy(domains->schedule_domain));

		empty = isl_set_is_empty(unroll_domain);
		if (empty >= 0 && empty) {
			isl_set_free(unroll_domain);
			continue;
		}

		class_domain = do_unroll(domains, unroll_domain, class_domain);
	}

	isl_basic_set_list_free(unroll_list);

	return class_domain;
}

/* Try and construct a single basic set that includes the intersection of
 * the schedule domain, the atomic option domain and the class domain.
 * Add the resulting basic set(s) to domains->list and remove them
 * from class_domain.  Return the updated class domain.
 *
 * We construct a single domain rather than trying to combine
 * the schedule domains of individual domains because we are working
 * within a single component so that non-overlapping schedule domains
 * should already have been separated.
 * We do however need to make sure that this single domains is a subset
 * of the class domain so that it would not intersect with any other
 * class domains.  This means that we may end up splitting up the atomic
 * domain in case separation classes are being used.
 *
 * "domain" is the intersection of the schedule domain and the class domain,
 * with inner dimensions projected out.
 */
static __isl_give isl_set *compute_atomic_domain(
	struct isl_codegen_domains *domains, __isl_take isl_set *class_domain)
{
	isl_basic_set *bset;
	isl_basic_set_list *list;
	isl_set *domain, *atomic_domain;
	int empty;

	domain = isl_set_copy(domains->option[isl_ast_loop_atomic]);
	domain = isl_set_intersect(domain, isl_set_copy(class_domain));
	domain = isl_set_intersect(domain,
				isl_set_copy(domains->schedule_domain));
	empty = isl_set_is_empty(domain);
	if (empty < 0)
		class_domain = isl_set_free(class_domain);
	if (empty) {
		isl_set_free(domain);
		return class_domain;
	}

	domain = isl_ast_build_eliminate(domains->build, domain);
	domain = isl_set_coalesce(domain);
	bset = isl_set_unshifted_simple_hull(domain);
	domain = isl_set_from_basic_set(bset);
	atomic_domain = isl_set_copy(domain);
	domain = isl_set_intersect(domain, isl_set_copy(class_domain));
	class_domain = isl_set_subtract(class_domain, atomic_domain);
	domain = isl_set_make_disjoint(domain);
	list = isl_basic_set_list_from_set(domain);
	domains->list = isl_basic_set_list_concat(domains->list, list);

	return class_domain;
}

/* Split up the schedule domain into uniform basic sets,
 * in the sense that each element in a basic set is associated to
 * elements of the same domains, and add the result to domains->list.
 * Do this for that part of the schedule domain that lies in the
 * intersection of "class_domain" and the separate option domain.
 *
 * "class_domain" may or may not include the constraints
 * of the schedule domain, but this does not make a difference
 * since we are going to intersect it with the domain of the inverse schedule.
 * If it includes schedule domain constraints, then they may involve
 * inner dimensions, but we will eliminate them in separation_domain.
 */
static int compute_separate_domain(struct isl_codegen_domains *domains,
	__isl_keep isl_set *class_domain)
{
	isl_space *space;
	isl_set *domain;
	isl_union_map *executed;
	isl_basic_set_list *list;
	int empty;

	domain = isl_set_copy(domains->option[isl_ast_loop_separate]);
	domain = isl_set_intersect(domain, isl_set_copy(class_domain));
	executed = isl_union_map_copy(domains->executed);
	executed = isl_union_map_intersect_domain(executed,
				    isl_union_set_from_set(domain));
	empty = isl_union_map_is_empty(executed);
	if (empty < 0 || empty) {
		isl_union_map_free(executed);
		return empty < 0 ? -1 : 0;
	}

	space = isl_set_get_space(class_domain);
	domain = separate_schedule_domains(space, executed, domains->build);

	list = isl_basic_set_list_from_set(domain);
	domains->list = isl_basic_set_list_concat(domains->list, list);

	return 0;
}

/* Split up the domain at the current depth into disjoint
 * basic sets for which code should be generated separately
 * for the given separation class domain.
 *
 * If any separation classes have been defined, then "class_domain"
 * is the domain of the current class and does not refer to inner dimensions.
 * Otherwise, "class_domain" is the universe domain.
 *
 * We first make sure that the class domain is disjoint from
 * previously considered class domains.
 *
 * The separate domains can be computed directly from the "class_domain".
 *
 * The unroll, atomic and remainder domains need the constraints
 * from the schedule domain.
 *
 * For unrolling, the actual schedule domain is needed (with divs that
 * may refer to the current dimension) so that stride detection can be
 * performed.
 *
 * For atomic and remainder domains, inner dimensions and divs involving
 * the current dimensions should be eliminated.
 * In case we are working within a separation class, we need to intersect
 * the result with the current "class_domain" to ensure that the domains
 * are disjoint from those generated from other class domains.
 *
 * The domain that has been made atomic may be larger than specified
 * by the user since it needs to be representable as a single basic set.
 * This possibly larger domain is removed from class_domain by
 * compute_atomic_domain.  It is computed first so that the extended domain
 * would not overlap with any domains computed before.
 * Similary, the unrolled domains may have some constraints removed and
 * may therefore also be larger than specified by the user.
 *
 * If anything is left after handling separate, unroll and atomic,
 * we split it up into basic sets and append the basic sets to domains->list.
 */
static isl_stat compute_partial_domains(struct isl_codegen_domains *domains,
	__isl_take isl_set *class_domain)
{
	isl_basic_set_list *list;
	isl_set *domain;

	class_domain = isl_set_subtract(class_domain,
					isl_set_copy(domains->done));
	domains->done = isl_set_union(domains->done,
					isl_set_copy(class_domain));

	class_domain = compute_atomic_domain(domains, class_domain);
	class_domain = compute_unroll_domains(domains, class_domain);

	domain = isl_set_copy(class_domain);

	if (compute_separate_domain(domains, domain) < 0)
		goto error;
	domain = isl_set_subtract(domain,
			isl_set_copy(domains->option[isl_ast_loop_separate]));

	domain = isl_set_intersect(domain,
				isl_set_copy(domains->schedule_domain));

	domain = isl_ast_build_eliminate(domains->build, domain);
	domain = isl_set_intersect(domain, isl_set_copy(class_domain));

	domain = isl_set_coalesce(domain);
	domain = isl_set_make_disjoint(domain);

	list = isl_basic_set_list_from_set(domain);
	domains->list = isl_basic_set_list_concat(domains->list, list);

	isl_set_free(class_domain);

	return isl_stat_ok;
error:
	isl_set_free(domain);
	isl_set_free(class_domain);
	return isl_stat_error;
}

/* Split up the domain at the current depth into disjoint
 * basic sets for which code should be generated separately
 * for the separation class identified by "pnt".
 *
 * We extract the corresponding class domain from domains->sep_class,
 * eliminate inner dimensions and pass control to compute_partial_domains.
 */
static isl_stat compute_class_domains(__isl_take isl_point *pnt, void *user)
{
	struct isl_codegen_domains *domains = user;
	isl_set *class_set;
	isl_set *domain;
	int disjoint;

	class_set = isl_set_from_point(pnt);
	domain = isl_map_domain(isl_map_intersect_range(
				isl_map_copy(domains->sep_class), class_set));
	domain = isl_ast_build_compute_gist(domains->build, domain);
	domain = isl_ast_build_eliminate(domains->build, domain);

	disjoint = isl_set_plain_is_disjoint(domain, domains->schedule_domain);
	if (disjoint < 0)
		return isl_stat_error;
	if (disjoint) {
		isl_set_free(domain);
		return isl_stat_ok;
	}

	return compute_partial_domains(domains, domain);
}

/* Extract the domains at the current depth that should be atomic,
 * separated or unrolled and store them in option.
 *
 * The domains specified by the user might overlap, so we make
 * them disjoint by subtracting earlier domains from later domains.
 */
static void compute_domains_init_options(isl_set *option[4],
	__isl_keep isl_ast_build *build)
{
	enum isl_ast_loop_type type, type2;
	isl_set *unroll;

	for (type = isl_ast_loop_atomic;
	    type <= isl_ast_loop_separate; ++type) {
		option[type] = isl_ast_build_get_option_domain(build, type);
		for (type2 = isl_ast_loop_atomic; type2 < type; ++type2)
			option[type] = isl_set_subtract(option[type],
						isl_set_copy(option[type2]));
	}

	unroll = option[isl_ast_loop_unroll];
	unroll = isl_set_coalesce(unroll);
	unroll = isl_set_make_disjoint(unroll);
	option[isl_ast_loop_unroll] = unroll;
}

/* Split up the domain at the current depth into disjoint
 * basic sets for which code should be generated separately,
 * based on the user-specified options.
 * Return the list of disjoint basic sets.
 *
 * There are three kinds of domains that we need to keep track of.
 * - the "schedule domain" is the domain of "executed"
 * - the "class domain" is the domain corresponding to the currrent
 *	separation class
 * - the "option domain" is the domain corresponding to one of the options
 *	atomic, unroll or separate
 *
 * We first consider the individial values of the separation classes
 * and split up the domain for each of them separately.
 * Finally, we consider the remainder.  If no separation classes were
 * specified, then we call compute_partial_domains with the universe
 * "class_domain".  Otherwise, we take the "schedule_domain" as "class_domain",
 * with inner dimensions removed.  We do this because we want to
 * avoid computing the complement of the class domains (i.e., the difference
 * between the universe and domains->done).
 */
static __isl_give isl_basic_set_list *compute_domains(
	__isl_keep isl_union_map *executed, __isl_keep isl_ast_build *build)
{
	struct isl_codegen_domains domains;
	isl_ctx *ctx;
	isl_set *domain;
	isl_union_set *schedule_domain;
	isl_set *classes;
	isl_space *space;
	int n_param;
	enum isl_ast_loop_type type;
	int empty;

	if (!executed)
		return NULL;

	ctx = isl_union_map_get_ctx(executed);
	domains.list = isl_basic_set_list_alloc(ctx, 0);

	schedule_domain = isl_union_map_domain(isl_union_map_copy(executed));
	domain = isl_set_from_union_set(schedule_domain);

	compute_domains_init_options(domains.option, build);

	domains.sep_class = isl_ast_build_get_separation_class(build);
	classes = isl_map_range(isl_map_copy(domains.sep_class));
	n_param = isl_set_dim(classes, isl_dim_param);
	classes = isl_set_project_out(classes, isl_dim_param, 0, n_param);

	space = isl_set_get_space(domain);
	domains.build = build;
	domains.schedule_domain = isl_set_copy(domain);
	domains.executed = executed;
	domains.done = isl_set_empty(space);

	if (isl_set_foreach_point(classes, &compute_class_domains, &domains) < 0)
		domains.list = isl_basic_set_list_free(domains.list);
	isl_set_free(classes);

	empty = isl_set_is_empty(domains.done);
	if (empty < 0) {
		domains.list = isl_basic_set_list_free(domains.list);
		domain = isl_set_free(domain);
	} else if (empty) {
		isl_set_free(domain);
		domain = isl_set_universe(isl_set_get_space(domains.done));
	} else {
		domain = isl_ast_build_eliminate(build, domain);
	}
	if (compute_partial_domains(&domains, domain) < 0)
		domains.list = isl_basic_set_list_free(domains.list);

	isl_set_free(domains.schedule_domain);
	isl_set_free(domains.done);
	isl_map_free(domains.sep_class);
	for (type = isl_ast_loop_atomic; type <= isl_ast_loop_separate; ++type)
		isl_set_free(domains.option[type]);

	return domains.list;
}

/* Generate code for a single component, after shifting (if any)
 * has been applied, in case the schedule was specified as a union map.
 *
 * We first split up the domain at the current depth into disjoint
 * basic sets based on the user-specified options.
 * Then we generated code for each of them and concatenate the results.
 */
static __isl_give isl_ast_graft_list *generate_shifted_component_flat(
	__isl_take isl_union_map *executed, __isl_take isl_ast_build *build)
{
	isl_basic_set_list *domain_list;
	isl_ast_graft_list *list = NULL;

	domain_list = compute_domains(executed, build);
	list = generate_parallel_domains(domain_list, executed, build);

	isl_basic_set_list_free(domain_list);
	isl_union_map_free(executed);
	isl_ast_build_free(build);

	return list;
}

/* Generate code for a single component, after shifting (if any)
 * has been applied, in case the schedule was specified as a schedule tree
 * and the separate option was specified.
 *
 * We perform separation on the domain of "executed" and then generate
 * an AST for each of the resulting disjoint basic sets.
 */
static __isl_give isl_ast_graft_list *generate_shifted_component_tree_separate(
	__isl_take isl_union_map *executed, __isl_take isl_ast_build *build)
{
	isl_space *space;
	isl_set *domain;
	isl_basic_set_list *domain_list;
	isl_ast_graft_list *list;

	space = isl_ast_build_get_space(build, 1);
	domain = separate_schedule_domains(space,
					isl_union_map_copy(executed), build);
	domain_list = isl_basic_set_list_from_set(domain);

	list = generate_parallel_domains(domain_list, executed, build);

	isl_basic_set_list_free(domain_list);
	isl_union_map_free(executed);
	isl_ast_build_free(build);

	return list;
}

/* Internal data structure for generate_shifted_component_tree_unroll.
 *
 * "executed" and "build" are inputs to generate_shifted_component_tree_unroll.
 * "list" collects the constructs grafts.
 */
struct isl_ast_unroll_tree_data {
	isl_union_map *executed;
	isl_ast_build *build;
	isl_ast_graft_list *list;
};

/* Initialize data->list to a list of "n" elements.
 */
static int init_unroll_tree(int n, void *user)
{
	struct isl_ast_unroll_tree_data *data = user;
	isl_ctx *ctx;

	ctx = isl_ast_build_get_ctx(data->build);
	data->list = isl_ast_graft_list_alloc(ctx, n);

	return 0;
}

/* Given an iteration of an unrolled domain represented by "bset",
 * generate the corresponding AST and add the result to data->list.
 */
static int do_unroll_tree_iteration(__isl_take isl_basic_set *bset, void *user)
{
	struct isl_ast_unroll_tree_data *data = user;

	data->list = add_node(data->list, isl_union_map_copy(data->executed),
				bset, isl_ast_build_copy(data->build));

	return 0;
}

/* Generate code for a single component, after shifting (if any)
 * has been applied, in case the schedule was specified as a schedule tree
 * and the unroll option was specified.
 *
 * We call foreach_iteration to iterate over the individual values and
 * construct and collect the corresponding grafts in do_unroll_tree_iteration.
 */
static __isl_give isl_ast_graft_list *generate_shifted_component_tree_unroll(
	__isl_take isl_union_map *executed, __isl_take isl_set *domain,
	__isl_take isl_ast_build *build)
{
	struct isl_ast_unroll_tree_data data = { executed, build, NULL };

	if (foreach_iteration(domain, build, &init_unroll_tree,
				&do_unroll_tree_iteration, &data) < 0)
		data.list = isl_ast_graft_list_free(data.list);

	isl_union_map_free(executed);
	isl_ast_build_free(build);

	return data.list;
}

/* Does "domain" involve a disjunction that is purely based on
 * constraints involving only outer dimension?
 *
 * In particular, is there a disjunction such that the constraints
 * involving the current and later dimensions are the same over
 * all the disjuncts?
 */
static isl_bool has_pure_outer_disjunction(__isl_keep isl_set *domain,
	__isl_keep isl_ast_build *build)
{
	isl_basic_set *hull;
	isl_set *shared, *inner;
	isl_bool equal;
	int depth, dim;

	if (isl_set_n_basic_set(domain) <= 1)
		return isl_bool_false;

	inner = isl_set_copy(domain);
	depth = isl_ast_build_get_depth(build);
	dim = isl_set_dim(inner, isl_dim_set);
	inner = isl_set_drop_constraints_not_involving_dims(inner,
					    isl_dim_set, depth, dim - depth);
	hull = isl_set_plain_unshifted_simple_hull(isl_set_copy(inner));
	shared = isl_set_from_basic_set(hull);
	equal = isl_set_plain_is_equal(inner, shared);
	isl_set_free(inner);
	isl_set_free(shared);

	return equal;
}

/* Generate code for a single component, after shifting (if any)
 * has been applied, in case the schedule was specified as a schedule tree.
 * In particular, handle the base case where there is either no isolated
 * set or we are within the isolated set (in which case "isolated" is set)
 * or the iterations that precede or follow the isolated set.
 *
 * The schedule domain is broken up or combined into basic sets
 * according to the AST generation option specified in the current
 * schedule node, which may be either atomic, separate, unroll or
 * unspecified.  If the option is unspecified, then we currently simply
 * split the schedule domain into disjoint basic sets.
 *
 * In case the separate option is specified, the AST generation is
 * handled by generate_shifted_component_tree_separate.
 * In the other cases, we need the global schedule domain.
 * In the unroll case, the AST generation is then handled by
 * generate_shifted_component_tree_unroll which needs the actual
 * schedule domain (with divs that may refer to the current dimension)
 * so that stride detection can be performed.
 * In the atomic or unspecified case, inner dimensions and divs involving
 * the current dimensions should be eliminated.
 * The result is then either combined into a single basic set or
 * split up into disjoint basic sets.
 * Finally an AST is generated for each basic set and the results are
 * concatenated.
 *
 * If the schedule domain involves a disjunction that is purely based on
 * constraints involving only outer dimension, then it is treated as
 * if atomic was specified.  This ensures that only a single loop
 * is generated instead of a sequence of identical loops with
 * different guards.
 */
static __isl_give isl_ast_graft_list *generate_shifted_component_tree_base(
	__isl_take isl_union_map *executed, __isl_take isl_ast_build *build,
	int isolated)
{
	isl_bool outer_disjunction;
	isl_union_set *schedule_domain;
	isl_set *domain;
	isl_basic_set_list *domain_list;
	isl_ast_graft_list *list;
	enum isl_ast_loop_type type;

	type = isl_ast_build_get_loop_type(build, isolated);
	if (type < 0)
		goto error;

	if (type == isl_ast_loop_separate)
		return generate_shifted_component_tree_separate(executed,
								build);

	schedule_domain = isl_union_map_domain(isl_union_map_copy(executed));
	domain = isl_set_from_union_set(schedule_domain);

	if (type == isl_ast_loop_unroll)
		return generate_shifted_component_tree_unroll(executed, domain,
								build);

	domain = isl_ast_build_eliminate(build, domain);
	domain = isl_set_coalesce(domain);

	outer_disjunction = has_pure_outer_disjunction(domain, build);
	if (outer_disjunction < 0)
		domain = isl_set_free(domain);

	if (outer_disjunction || type == isl_ast_loop_atomic) {
		isl_basic_set *hull;
		hull = isl_set_unshifted_simple_hull(domain);
		domain_list = isl_basic_set_list_from_basic_set(hull);
	} else {
		domain = isl_set_make_disjoint(domain);
		domain_list = isl_basic_set_list_from_set(domain);
	}

	list = generate_parallel_domains(domain_list, executed, build);

	isl_basic_set_list_free(domain_list);
	isl_union_map_free(executed);
	isl_ast_build_free(build);

	return list;
error:
	isl_union_map_free(executed);
	isl_ast_build_free(build);
	return NULL;
}

/* Extract out the disjunction imposed by "domain" on the outer
 * schedule dimensions.
 *
 * In particular, remove all inner dimensions from "domain" (including
 * the current dimension) and then remove the constraints that are shared
 * by all disjuncts in the result.
 */
static __isl_give isl_set *extract_disjunction(__isl_take isl_set *domain,
	__isl_keep isl_ast_build *build)
{
	isl_set *hull;
	int depth, dim;

	domain = isl_ast_build_specialize(build, domain);
	depth = isl_ast_build_get_depth(build);
	dim = isl_set_dim(domain, isl_dim_set);
	domain = isl_set_eliminate(domain, isl_dim_set, depth, dim - depth);
	domain = isl_set_remove_unknown_divs(domain);
	hull = isl_set_copy(domain);
	hull = isl_set_from_basic_set(isl_set_unshifted_simple_hull(hull));
	domain = isl_set_gist(domain, hull);

	return domain;
}

/* Add "guard" to the grafts in "list".
 * "build" is the outer AST build, while "sub_build" includes "guard"
 * in its generated domain.
 *
 * First combine the grafts into a single graft and then add the guard.
 * If the list is empty, or if some error occurred, then simply return
 * the list.
 */
static __isl_give isl_ast_graft_list *list_add_guard(
	__isl_take isl_ast_graft_list *list, __isl_keep isl_set *guard,
	__isl_keep isl_ast_build *build, __isl_keep isl_ast_build *sub_build)
{
	isl_ast_graft *graft;

	list = isl_ast_graft_list_fuse(list, sub_build);

	if (isl_ast_graft_list_n_ast_graft(list) != 1)
		return list;

	graft = isl_ast_graft_list_get_ast_graft(list, 0);
	graft = isl_ast_graft_add_guard(graft, isl_set_copy(guard), build);
	list = isl_ast_graft_list_set_ast_graft(list, 0, graft);

	return list;
}

/* Generate code for a single component, after shifting (if any)
 * has been applied, in case the schedule was specified as a schedule tree.
 * In particular, do so for the specified subset of the schedule domain.
 *
 * If we are outside of the isolated part, then "domain" may include
 * a disjunction.  Explicitly generate this disjunction at this point
 * instead of relying on the disjunction getting hoisted back up
 * to this level.
 */
static __isl_give isl_ast_graft_list *generate_shifted_component_tree_part(
	__isl_keep isl_union_map *executed, __isl_take isl_set *domain,
	__isl_keep isl_ast_build *build, int isolated)
{
	isl_union_set *uset;
	isl_ast_graft_list *list;
	isl_ast_build *sub_build;
	int empty;

	uset = isl_union_set_from_set(isl_set_copy(domain));
	executed = isl_union_map_copy(executed);
	executed = isl_union_map_intersect_domain(executed, uset);
	empty = isl_union_map_is_empty(executed);
	if (empty < 0)
		goto error;
	if (empty) {
		isl_ctx *ctx;
		isl_union_map_free(executed);
		isl_set_free(domain);
		ctx = isl_ast_build_get_ctx(build);
		return isl_ast_graft_list_alloc(ctx, 0);
	}

	sub_build = isl_ast_build_copy(build);
	if (!isolated) {
		domain = extract_disjunction(domain, build);
		sub_build = isl_ast_build_restrict_generated(sub_build,
							isl_set_copy(domain));
	}
	list = generate_shifted_component_tree_base(executed,
				isl_ast_build_copy(sub_build), isolated);
	if (!isolated)
		list = list_add_guard(list, domain, build, sub_build);
	isl_ast_build_free(sub_build);
	isl_set_free(domain);
	return list;
error:
	isl_union_map_free(executed);
	isl_set_free(domain);
	return NULL;
}

/* Generate code for a single component, after shifting (if any)
 * has been applied, in case the schedule was specified as a schedule tree.
 * In particular, do so for the specified sequence of subsets
 * of the schedule domain, "before", "isolated", "after" and "other",
 * where only the "isolated" part is considered to be isolated.
 */
static __isl_give isl_ast_graft_list *generate_shifted_component_parts(
	__isl_take isl_union_map *executed, __isl_take isl_set *before,
	__isl_take isl_set *isolated, __isl_take isl_set *after,
	__isl_take isl_set *other, __isl_take isl_ast_build *build)
{
	isl_ast_graft_list *list, *res;

	res = generate_shifted_component_tree_part(executed, before, build, 0);
	list = generate_shifted_component_tree_part(executed, isolated,
						    build, 1);
	res = isl_ast_graft_list_concat(res, list);
	list = generate_shifted_component_tree_part(executed, after, build, 0);
	res = isl_ast_graft_list_concat(res, list);
	list = generate_shifted_component_tree_part(executed, other, build, 0);
	res = isl_ast_graft_list_concat(res, list);

	isl_union_map_free(executed);
	isl_ast_build_free(build);

	return res;
}

/* Does "set" intersect "first", but not "second"?
 */
static isl_bool only_intersects_first(__isl_keep isl_set *set,
	__isl_keep isl_set *first, __isl_keep isl_set *second)
{
	isl_bool disjoint;

	disjoint = isl_set_is_disjoint(set, first);
	if (disjoint < 0)
		return isl_bool_error;
	if (disjoint)
		return isl_bool_false;

	return isl_set_is_disjoint(set, second);
}

/* Generate code for a single component, after shifting (if any)
 * has been applied, in case the schedule was specified as a schedule tree.
 * In particular, do so in case of isolation where there is
 * only an "isolated" part and an "after" part.
 * "dead1" and "dead2" are freed by this function in order to simplify
 * the caller.
 *
 * The "before" and "other" parts are set to empty sets.
 */
static __isl_give isl_ast_graft_list *generate_shifted_component_only_after(
	__isl_take isl_union_map *executed, __isl_take isl_set *isolated,
	__isl_take isl_set *after, __isl_take isl_ast_build *build,
	__isl_take isl_set *dead1, __isl_take isl_set *dead2)
{
	isl_set *empty;

	empty = isl_set_empty(isl_set_get_space(after));
	isl_set_free(dead1);
	isl_set_free(dead2);
	return generate_shifted_component_parts(executed, isl_set_copy(empty),
						isolated, after, empty, build);
}

/* Generate code for a single component, after shifting (if any)
 * has been applied, in case the schedule was specified as a schedule tree.
 *
 * We first check if the user has specified an isolated schedule domain
 * and that we are not already outside of this isolated schedule domain.
 * If so, we break up the schedule domain into iterations that
 * precede the isolated domain, the isolated domain itself,
 * the iterations that follow the isolated domain and
 * the remaining iterations (those that are incomparable
 * to the isolated domain).
 * We generate an AST for each piece and concatenate the results.
 *
 * If the isolated domain is not convex, then it is replaced
 * by a convex superset to ensure that the sets of preceding and
 * following iterations are properly defined and, in particular,
 * that there are no intermediate iterations that do not belong
 * to the isolated domain.
 *
 * In the special case where at least one element of the schedule
 * domain that does not belong to the isolated domain needs
 * to be scheduled after this isolated domain, but none of those
 * elements need to be scheduled before, break up the schedule domain
 * in only two parts, the isolated domain, and a part that will be
 * scheduled after the isolated domain.
 *
 * If no isolated set has been specified, then we generate an
 * AST for the entire inverse schedule.
 */
static __isl_give isl_ast_graft_list *generate_shifted_component_tree(
	__isl_take isl_union_map *executed, __isl_take isl_ast_build *build)
{
	int i, depth;
	int empty, has_isolate;
	isl_space *space;
	isl_union_set *schedule_domain;
	isl_set *domain;
	isl_basic_set *hull;
	isl_set *isolated, *before, *after, *test;
	isl_map *gt, *lt;
	isl_bool pure;

	build = isl_ast_build_extract_isolated(build);
	has_isolate = isl_ast_build_has_isolated(build);
	if (has_isolate < 0)
		executed = isl_union_map_free(executed);
	else if (!has_isolate)
		return generate_shifted_component_tree_base(executed, build, 0);

	schedule_domain = isl_union_map_domain(isl_union_map_copy(executed));
	domain = isl_set_from_union_set(schedule_domain);

	isolated = isl_ast_build_get_isolated(build);
	isolated = isl_set_intersect(isolated, isl_set_copy(domain));
	test = isl_ast_build_specialize(build, isl_set_copy(isolated));
	empty = isl_set_is_empty(test);
	isl_set_free(test);
	if (empty < 0)
		goto error;
	if (empty) {
		isl_set_free(isolated);
		isl_set_free(domain);
		return generate_shifted_component_tree_base(executed, build, 0);
	}
	isolated = isl_ast_build_eliminate(build, isolated);
	hull = isl_set_unshifted_simple_hull(isolated);
	isolated = isl_set_from_basic_set(hull);

	depth = isl_ast_build_get_depth(build);
	space = isl_space_map_from_set(isl_set_get_space(isolated));
	gt = isl_map_universe(space);
	for (i = 0; i < depth; ++i)
		gt = isl_map_equate(gt, isl_dim_in, i, isl_dim_out, i);
	gt = isl_map_order_gt(gt, isl_dim_in, depth, isl_dim_out, depth);
	lt = isl_map_reverse(isl_map_copy(gt));
	before = isl_set_apply(isl_set_copy(isolated), gt);
	after = isl_set_apply(isl_set_copy(isolated), lt);

	domain = isl_set_subtract(domain, isl_set_copy(isolated));
	pure = only_intersects_first(domain, after, before);
	if (pure < 0)
		executed = isl_union_map_free(executed);
	else if (pure)
		return generate_shifted_component_only_after(executed, isolated,
						domain, build, before, after);
	domain = isl_set_subtract(domain, isl_set_copy(before));
	domain = isl_set_subtract(domain, isl_set_copy(after));
	after = isl_set_subtract(after, isl_set_copy(isolated));
	after = isl_set_subtract(after, isl_set_copy(before));
	before = isl_set_subtract(before, isl_set_copy(isolated));

	return generate_shifted_component_parts(executed, before, isolated,
						after, domain, build);
error:
	isl_set_free(domain);
	isl_set_free(isolated);
	isl_union_map_free(executed);
	isl_ast_build_free(build);
	return NULL;
}

/* Generate code for a single component, after shifting (if any)
 * has been applied.
 *
 * Call generate_shifted_component_tree or generate_shifted_component_flat
 * depending on whether the schedule was specified as a schedule tree.
 */
static __isl_give isl_ast_graft_list *generate_shifted_component(
	__isl_take isl_union_map *executed, __isl_take isl_ast_build *build)
{
	if (isl_ast_build_has_schedule_node(build))
		return generate_shifted_component_tree(executed, build);
	else
		return generate_shifted_component_flat(executed, build);
}

struct isl_set_map_pair {
	isl_set *set;
	isl_map *map;
};

/* Given an array "domain" of isl_set_map_pairs and an array "order"
 * of indices into the "domain" array,
 * return the union of the "map" fields of the elements
 * indexed by the first "n" elements of "order".
 */
static __isl_give isl_union_map *construct_component_executed(
	struct isl_set_map_pair *domain, int *order, int n)
{
	int i;
	isl_map *map;
	isl_union_map *executed;

	map = isl_map_copy(domain[order[0]].map);
	executed = isl_union_map_from_map(map);
	for (i = 1; i < n; ++i) {
		map = isl_map_copy(domain[order[i]].map);
		executed = isl_union_map_add_map(executed, map);
	}

	return executed;
}

/* Generate code for a single component, after shifting (if any)
 * has been applied.
 *
 * The component inverse schedule is specified as the "map" fields
 * of the elements of "domain" indexed by the first "n" elements of "order".
 */
static __isl_give isl_ast_graft_list *generate_shifted_component_from_list(
	struct isl_set_map_pair *domain, int *order, int n,
	__isl_take isl_ast_build *build)
{
	isl_union_map *executed;

	executed = construct_component_executed(domain, order, n);
	return generate_shifted_component(executed, build);
}

/* Does set dimension "pos" of "set" have an obviously fixed value?
 */
static int dim_is_fixed(__isl_keep isl_set *set, int pos)
{
	int fixed;
	isl_val *v;

	v = isl_set_plain_get_val_if_fixed(set, isl_dim_set, pos);
	if (!v)
		return -1;
	fixed = !isl_val_is_nan(v);
	isl_val_free(v);

	return fixed;
}

/* Given an array "domain" of isl_set_map_pairs and an array "order"
 * of indices into the "domain" array,
 * do all (except for at most one) of the "set" field of the elements
 * indexed by the first "n" elements of "order" have a fixed value
 * at position "depth"?
 */
static int at_most_one_non_fixed(struct isl_set_map_pair *domain,
	int *order, int n, int depth)
{
	int i;
	int non_fixed = -1;

	for (i = 0; i < n; ++i) {
		int f;

		f = dim_is_fixed(domain[order[i]].set, depth);
		if (f < 0)
			return -1;
		if (f)
			continue;
		if (non_fixed >= 0)
			return 0;
		non_fixed = i;
	}

	return 1;
}

/* Given an array "domain" of isl_set_map_pairs and an array "order"
 * of indices into the "domain" array,
 * eliminate the inner dimensions from the "set" field of the elements
 * indexed by the first "n" elements of "order", provided the current
 * dimension does not have a fixed value.
 *
 * Return the index of the first element in "order" with a corresponding
 * "set" field that does not have an (obviously) fixed value.
 */
static int eliminate_non_fixed(struct isl_set_map_pair *domain,
	int *order, int n, int depth, __isl_keep isl_ast_build *build)
{
	int i;
	int base = -1;

	for (i = n - 1; i >= 0; --i) {
		int f;
		f = dim_is_fixed(domain[order[i]].set, depth);
		if (f < 0)
			return -1;
		if (f)
			continue;
		domain[order[i]].set = isl_ast_build_eliminate_inner(build,
							domain[order[i]].set);
		base = i;
	}

	return base;
}

/* Given an array "domain" of isl_set_map_pairs and an array "order"
 * of indices into the "domain" array,
 * find the element of "domain" (amongst those indexed by the first "n"
 * elements of "order") with the "set" field that has the smallest
 * value for the current iterator.
 *
 * Note that the domain with the smallest value may depend on the parameters
 * and/or outer loop dimension.  Since the result of this function is only
 * used as heuristic, we only make a reasonable attempt at finding the best
 * domain, one that should work in case a single domain provides the smallest
 * value for the current dimension over all values of the parameters
 * and outer dimensions.
 *
 * In particular, we compute the smallest value of the first domain
 * and replace it by that of any later domain if that later domain
 * has a smallest value that is smaller for at least some value
 * of the parameters and outer dimensions.
 */
static int first_offset(struct isl_set_map_pair *domain, int *order, int n,
	__isl_keep isl_ast_build *build)
{
	int i;
	isl_map *min_first;
	int first = 0;

	min_first = isl_ast_build_map_to_iterator(build,
					isl_set_copy(domain[order[0]].set));
	min_first = isl_map_lexmin(min_first);

	for (i = 1; i < n; ++i) {
		isl_map *min, *test;
		int empty;

		min = isl_ast_build_map_to_iterator(build,
					isl_set_copy(domain[order[i]].set));
		min = isl_map_lexmin(min);
		test = isl_map_copy(min);
		test = isl_map_apply_domain(isl_map_copy(min_first), test);
		test = isl_map_order_lt(test, isl_dim_in, 0, isl_dim_out, 0);
		empty = isl_map_is_empty(test);
		isl_map_free(test);
		if (empty >= 0 && !empty) {
			isl_map_free(min_first);
			first = i;
			min_first = min;
		} else
			isl_map_free(min);

		if (empty < 0)
			break;
	}

	isl_map_free(min_first);

	return i < n ? -1 : first;
}

/* Construct a shifted inverse schedule based on the original inverse schedule,
 * the stride and the offset.
 *
 * The original inverse schedule is specified as the "map" fields
 * of the elements of "domain" indexed by the first "n" elements of "order".
 *
 * "stride" and "offset" are such that the difference
 * between the values of the current dimension of domain "i"
 * and the values of the current dimension for some reference domain are
 * equal to
 *
 *	stride * integer + offset[i]
 *
 * Moreover, 0 <= offset[i] < stride.
 *
 * For each domain, we create a map
 *
 *	{ [..., j, ...] -> [..., j - offset[i], offset[i], ....] }
 *
 * where j refers to the current dimension and the other dimensions are
 * unchanged, and apply this map to the original schedule domain.
 *
 * For example, for the original schedule
 *
 *	{ A[i] -> [2i]: 0 <= i < 10; B[i] -> [2i+1] : 0 <= i < 10 }
 *
 * and assuming the offset is 0 for the A domain and 1 for the B domain,
 * we apply the mapping
 *
 *	{ [j] -> [j, 0] }
 *
 * to the schedule of the "A" domain and the mapping
 *
 *	{ [j - 1] -> [j, 1] }
 *
 * to the schedule of the "B" domain.
 *
 *
 * Note that after the transformation, the differences between pairs
 * of values of the current dimension over all domains are multiples
 * of stride and that we have therefore exposed the stride.
 *
 *
 * To see that the mapping preserves the lexicographic order,
 * first note that each of the individual maps above preserves the order.
 * If the value of the current iterator is j1 in one domain and j2 in another,
 * then if j1 = j2, we know that the same map is applied to both domains
 * and the order is preserved.
 * Otherwise, let us assume, without loss of generality, that j1 < j2.
 * If c1 >= c2 (with c1 and c2 the corresponding offsets), then
 *
 *	j1 - c1 < j2 - c2
 *
 * and the order is preserved.
 * If c1 < c2, then we know
 *
 *	0 <= c2 - c1 < s
 *
 * We also have
 *
 *	j2 - j1 = n * s + r
 *
 * with n >= 0 and 0 <= r < s.
 * In other words, r = c2 - c1.
 * If n > 0, then
 *
 *	j1 - c1 < j2 - c2
 *
 * If n = 0, then
 *
 *	j1 - c1 = j2 - c2
 *
 * and so
 *
 *	(j1 - c1, c1) << (j2 - c2, c2)
 *
 * with "<<" the lexicographic order, proving that the order is preserved
 * in all cases.
 */
static __isl_give isl_union_map *construct_shifted_executed(
	struct isl_set_map_pair *domain, int *order, int n,
	__isl_keep isl_val *stride, __isl_keep isl_multi_val *offset,
	__isl_take isl_ast_build *build)
{
	int i;
	isl_union_map *executed;
	isl_space *space;
	isl_map *map;
	int depth;
	isl_constraint *c;

	depth = isl_ast_build_get_depth(build);
	space = isl_ast_build_get_space(build, 1);
	executed = isl_union_map_empty(isl_space_copy(space));
	space = isl_space_map_from_set(space);
	map = isl_map_identity(isl_space_copy(space));
	map = isl_map_eliminate(map, isl_dim_out, depth, 1);
	map = isl_map_insert_dims(map, isl_dim_out, depth + 1, 1);
	space = isl_space_insert_dims(space, isl_dim_out, depth + 1, 1);

	c = isl_constraint_alloc_equality(isl_local_space_from_space(space));
	c = isl_constraint_set_coefficient_si(c, isl_dim_in, depth, 1);
	c = isl_constraint_set_coefficient_si(c, isl_dim_out, depth, -1);

	for (i = 0; i < n; ++i) {
		isl_map *map_i;
		isl_val *v;

		v = isl_multi_val_get_val(offset, i);
		if (!v)
			break;
		map_i = isl_map_copy(map);
		map_i = isl_map_fix_val(map_i, isl_dim_out, depth + 1,
					isl_val_copy(v));
		v = isl_val_neg(v);
		c = isl_constraint_set_constant_val(c, v);
		map_i = isl_map_add_constraint(map_i, isl_constraint_copy(c));

		map_i = isl_map_apply_domain(isl_map_copy(domain[order[i]].map),
						map_i);
		executed = isl_union_map_add_map(executed, map_i);
	}

	isl_constraint_free(c);
	isl_map_free(map);

	if (i < n)
		executed = isl_union_map_free(executed);

	return executed;
}

/* Generate code for a single component, after exposing the stride,
 * given that the schedule domain is "shifted strided".
 *
 * The component inverse schedule is specified as the "map" fields
 * of the elements of "domain" indexed by the first "n" elements of "order".
 *
 * The schedule domain being "shifted strided" means that the differences
 * between the values of the current dimension of domain "i"
 * and the values of the current dimension for some reference domain are
 * equal to
 *
 *	stride * integer + offset[i]
 *
 * We first look for the domain with the "smallest" value for the current
 * dimension and adjust the offsets such that the offset of the "smallest"
 * domain is equal to zero.  The other offsets are reduced modulo stride.
 *
 * Based on this information, we construct a new inverse schedule in
 * construct_shifted_executed that exposes the stride.
 * Since this involves the introduction of a new schedule dimension,
 * the build needs to be changed accordingly.
 * After computing the AST, the newly introduced dimension needs
 * to be removed again from the list of grafts.  We do this by plugging
 * in a mapping that represents the new schedule domain in terms of the
 * old schedule domain.
 */
static __isl_give isl_ast_graft_list *generate_shift_component(
	struct isl_set_map_pair *domain, int *order, int n,
	__isl_keep isl_val *stride, __isl_keep isl_multi_val *offset,
	__isl_take isl_ast_build *build)
{
	isl_ast_graft_list *list;
	int first;
	int depth;
	isl_val *val;
	isl_multi_val *mv;
	isl_space *space;
	isl_multi_aff *ma, *zero;
	isl_union_map *executed;

	depth = isl_ast_build_get_depth(build);

	first = first_offset(domain, order, n, build);
	if (first < 0)
		goto error;

	mv = isl_multi_val_copy(offset);
	val = isl_multi_val_get_val(offset, first);
	val = isl_val_neg(val);
	mv = isl_multi_val_add_val(mv, val);
	mv = isl_multi_val_mod_val(mv, isl_val_copy(stride));

	executed = construct_shifted_executed(domain, order, n, stride, mv,
						build);
	space = isl_ast_build_get_space(build, 1);
	space = isl_space_map_from_set(space);
	ma = isl_multi_aff_identity(isl_space_copy(space));
	space = isl_space_from_domain(isl_space_domain(space));
	space = isl_space_add_dims(space, isl_dim_out, 1);
	zero = isl_multi_aff_zero(space);
	ma = isl_multi_aff_range_splice(ma, depth + 1, zero);
	build = isl_ast_build_insert_dim(build, depth + 1);
	list = generate_shifted_component(executed, build);

	list = isl_ast_graft_list_preimage_multi_aff(list, ma);

	isl_multi_val_free(mv);

	return list;
error:
	isl_ast_build_free(build);
	return NULL;
}

/* Does any node in the schedule tree rooted at the current schedule node
 * of "build" depend on outer schedule nodes?
 */
static int has_anchored_subtree(__isl_keep isl_ast_build *build)
{
	isl_schedule_node *node;
	int dependent = 0;

	node = isl_ast_build_get_schedule_node(build);
	dependent = isl_schedule_node_is_subtree_anchored(node);
	isl_schedule_node_free(node);

	return dependent;
}

/* Generate code for a single component.
 *
 * The component inverse schedule is specified as the "map" fields
 * of the elements of "domain" indexed by the first "n" elements of "order".
 *
 * This function may modify the "set" fields of "domain".
 *
 * Before proceeding with the actual code generation for the component,
 * we first check if there are any "shifted" strides, meaning that
 * the schedule domains of the individual domains are all strided,
 * but that they have different offsets, resulting in the union
 * of schedule domains not being strided anymore.
 *
 * The simplest example is the schedule
 *
 *	{ A[i] -> [2i]: 0 <= i < 10; B[i] -> [2i+1] : 0 <= i < 10 }
 *
 * Both schedule domains are strided, but their union is not.
 * This function detects such cases and then rewrites the schedule to
 *
 *	{ A[i] -> [2i, 0]: 0 <= i < 10; B[i] -> [2i, 1] : 0 <= i < 10 }
 *
 * In the new schedule, the schedule domains have the same offset (modulo
 * the stride), ensuring that the union of schedule domains is also strided.
 *
 *
 * If there is only a single domain in the component, then there is
 * nothing to do.   Similarly, if the current schedule dimension has
 * a fixed value for almost all domains then there is nothing to be done.
 * In particular, we need at least two domains where the current schedule
 * dimension does not have a fixed value.
 * Finally, in case of a schedule map input,
 * if any of the options refer to the current schedule dimension,
 * then we bail out as well.  It would be possible to reformulate the options
 * in terms of the new schedule domain, but that would introduce constraints
 * that separate the domains in the options and that is something we would
 * like to avoid.
 * In the case of a schedule tree input, we bail out if any of
 * the descendants of the current schedule node refer to outer
 * schedule nodes in any way.
 *
 *
 * To see if there is any shifted stride, we look at the differences
 * between the values of the current dimension in pairs of domains
 * for equal values of outer dimensions.  These differences should be
 * of the form
 *
 *	m x + r
 *
 * with "m" the stride and "r" a constant.  Note that we cannot perform
 * this analysis on individual domains as the lower bound in each domain
 * may depend on parameters or outer dimensions and so the current dimension
 * itself may not have a fixed remainder on division by the stride.
 *
 * In particular, we compare the first domain that does not have an
 * obviously fixed value for the current dimension to itself and all
 * other domains and collect the offsets and the gcd of the strides.
 * If the gcd becomes one, then we failed to find shifted strides.
 * If the gcd is zero, then the differences were all fixed, meaning
 * that some domains had non-obviously fixed values for the current dimension.
 * If all the offsets are the same (for those domains that do not have
 * an obviously fixed value for the current dimension), then we do not
 * apply the transformation.
 * If none of the domains were skipped, then there is nothing to do.
 * If some of them were skipped, then if we apply separation, the schedule
 * domain should get split in pieces with a (non-shifted) stride.
 *
 * Otherwise, we apply a shift to expose the stride in
 * generate_shift_component.
 */
static __isl_give isl_ast_graft_list *generate_component(
	struct isl_set_map_pair *domain, int *order, int n,
	__isl_take isl_ast_build *build)
{
	int i, d;
	int depth;
	isl_ctx *ctx;
	isl_map *map;
	isl_set *deltas;
	isl_val *gcd = NULL;
	isl_multi_val *mv;
	int fixed, skip;
	int base;
	isl_ast_graft_list *list;
	int res = 0;

	depth = isl_ast_build_get_depth(build);

	skip = n == 1;
	if (skip >= 0 && !skip)
		skip = at_most_one_non_fixed(domain, order, n, depth);
	if (skip >= 0 && !skip) {
		if (isl_ast_build_has_schedule_node(build))
			skip = has_anchored_subtree(build);
		else
			skip = isl_ast_build_options_involve_depth(build);
	}
	if (skip < 0)
		goto error;
	if (skip)
		return generate_shifted_component_from_list(domain,
							    order, n, build);

	base = eliminate_non_fixed(domain, order, n, depth, build);
	if (base < 0)
		goto error;

	ctx = isl_ast_build_get_ctx(build);

	mv = isl_multi_val_zero(isl_space_set_alloc(ctx, 0, n));

	fixed = 1;
	for (i = 0; i < n; ++i) {
		isl_val *r, *m;

		map = isl_map_from_domain_and_range(
					isl_set_copy(domain[order[base]].set),
					isl_set_copy(domain[order[i]].set));
		for (d = 0; d < depth; ++d)
			map = isl_map_equate(map, isl_dim_in, d,
						    isl_dim_out, d);
		deltas = isl_map_deltas(map);
		res = isl_set_dim_residue_class_val(deltas, depth, &m, &r);
		isl_set_free(deltas);
		if (res < 0)
			break;

		if (i == 0)
			gcd = m;
		else
			gcd = isl_val_gcd(gcd, m);
		if (isl_val_is_one(gcd)) {
			isl_val_free(r);
			break;
		}
		mv = isl_multi_val_set_val(mv, i, r);

		res = dim_is_fixed(domain[order[i]].set, depth);
		if (res < 0)
			break;
		if (res)
			continue;

		if (fixed && i > base) {
			isl_val *a, *b;
			a = isl_multi_val_get_val(mv, i);
			b = isl_multi_val_get_val(mv, base);
			if (isl_val_ne(a, b))
				fixed = 0;
			isl_val_free(a);
			isl_val_free(b);
		}
	}

	if (res < 0 || !gcd) {
		isl_ast_build_free(build);
		list = NULL;
	} else if (i < n || fixed || isl_val_is_zero(gcd)) {
		list = generate_shifted_component_from_list(domain,
							    order, n, build);
	} else {
		list = generate_shift_component(domain, order, n, gcd, mv,
						build);
	}

	isl_val_free(gcd);
	isl_multi_val_free(mv);

	return list;
error:
	isl_ast_build_free(build);
	return NULL;
}

/* Store both "map" itself and its domain in the
 * structure pointed to by *next and advance to the next array element.
 */
static isl_stat extract_domain(__isl_take isl_map *map, void *user)
{
	struct isl_set_map_pair **next = user;

	(*next)->map = isl_map_copy(map);
	(*next)->set = isl_map_domain(map);
	(*next)++;

	return isl_stat_ok;
}

static int after_in_tree(__isl_keep isl_union_map *umap,
	__isl_keep isl_schedule_node *node);

/* Is any domain element of "umap" scheduled after any of
 * the corresponding image elements by the tree rooted at
 * the child of "node"?
 */
static int after_in_child(__isl_keep isl_union_map *umap,
	__isl_keep isl_schedule_node *node)
{
	isl_schedule_node *child;
	int after;

	child = isl_schedule_node_get_child(node, 0);
	after = after_in_tree(umap, child);
	isl_schedule_node_free(child);

	return after;
}

/* Is any domain element of "umap" scheduled after any of
 * the corresponding image elements by the tree rooted at
 * the band node "node"?
 *
 * We first check if any domain element is scheduled after any
 * of the corresponding image elements by the band node itself.
 * If not, we restrict "map" to those pairs of element that
 * are scheduled together by the band node and continue with
 * the child of the band node.
 * If there are no such pairs then the map passed to after_in_child
 * will be empty causing it to return 0.
 */
static int after_in_band(__isl_keep isl_union_map *umap,
	__isl_keep isl_schedule_node *node)
{
	isl_multi_union_pw_aff *mupa;
	isl_union_map *partial, *test, *gt, *universe, *umap1, *umap2;
	isl_union_set *domain, *range;
	isl_space *space;
	int empty;
	int after;

	if (isl_schedule_node_band_n_member(node) == 0)
		return after_in_child(umap, node);

	mupa = isl_schedule_node_band_get_partial_schedule(node);
	space = isl_multi_union_pw_aff_get_space(mupa);
	partial = isl_union_map_from_multi_union_pw_aff(mupa);
	test = isl_union_map_copy(umap);
	test = isl_union_map_apply_domain(test, isl_union_map_copy(partial));
	test = isl_union_map_apply_range(test, isl_union_map_copy(partial));
	gt = isl_union_map_from_map(isl_map_lex_gt(space));
	test = isl_union_map_intersect(test, gt);
	empty = isl_union_map_is_empty(test);
	isl_union_map_free(test);

	if (empty < 0 || !empty) {
		isl_union_map_free(partial);
		return empty < 0 ? -1 : 1;
	}

	universe = isl_union_map_universe(isl_union_map_copy(umap));
	domain = isl_union_map_domain(isl_union_map_copy(universe));
	range = isl_union_map_range(universe);
	umap1 = isl_union_map_copy(partial);
	umap1 = isl_union_map_intersect_domain(umap1, domain);
	umap2 = isl_union_map_intersect_domain(partial, range);
	test = isl_union_map_apply_range(umap1, isl_union_map_reverse(umap2));
	test = isl_union_map_intersect(test, isl_union_map_copy(umap));
	after = after_in_child(test, node);
	isl_union_map_free(test);
	return after;
}

/* Is any domain element of "umap" scheduled after any of
 * the corresponding image elements by the tree rooted at
 * the context node "node"?
 *
 * The context constraints apply to the schedule domain,
 * so we cannot apply them directly to "umap", which contains
 * pairs of statement instances.  Instead, we add them
 * to the range of the prefix schedule for both domain and
 * range of "umap".
 */
static int after_in_context(__isl_keep isl_union_map *umap,
	__isl_keep isl_schedule_node *node)
{
	isl_union_map *prefix, *universe, *umap1, *umap2;
	isl_union_set *domain, *range;
	isl_set *context;
	int after;

	umap = isl_union_map_copy(umap);
	context = isl_schedule_node_context_get_context(node);
	prefix = isl_schedule_node_get_prefix_schedule_union_map(node);
	universe = isl_union_map_universe(isl_union_map_copy(umap));
	domain = isl_union_map_domain(isl_union_map_copy(universe));
	range = isl_union_map_range(universe);
	umap1 = isl_union_map_copy(prefix);
	umap1 = isl_union_map_intersect_domain(umap1, domain);
	umap2 = isl_union_map_intersect_domain(prefix, range);
	umap1 = isl_union_map_intersect_range(umap1,
					    isl_union_set_from_set(context));
	umap1 = isl_union_map_apply_range(umap1, isl_union_map_reverse(umap2));
	umap = isl_union_map_intersect(umap, umap1);

	after = after_in_child(umap, node);

	isl_union_map_free(umap);

	return after;
}

/* Is any domain element of "umap" scheduled after any of
 * the corresponding image elements by the tree rooted at
 * the expansion node "node"?
 *
 * We apply the expansion to domain and range of "umap" and
 * continue with its child.
 */
static int after_in_expansion(__isl_keep isl_union_map *umap,
	__isl_keep isl_schedule_node *node)
{
	isl_union_map *expansion;
	int after;

	expansion = isl_schedule_node_expansion_get_expansion(node);
	umap = isl_union_map_copy(umap);
	umap = isl_union_map_apply_domain(umap, isl_union_map_copy(expansion));
	umap = isl_union_map_apply_range(umap, expansion);

	after = after_in_child(umap, node);

	isl_union_map_free(umap);

	return after;
}

/* Is any domain element of "umap" scheduled after any of
 * the corresponding image elements by the tree rooted at
 * the extension node "node"?
 *
 * Since the extension node may add statement instances before or
 * after the pairs of statement instances in "umap", we return 1
 * to ensure that these pairs are not broken up.
 */
static int after_in_extension(__isl_keep isl_union_map *umap,
	__isl_keep isl_schedule_node *node)
{
	return 1;
}

/* Is any domain element of "umap" scheduled after any of
 * the corresponding image elements by the tree rooted at
 * the filter node "node"?
 *
 * We intersect domain and range of "umap" with the filter and
 * continue with its child.
 */
static int after_in_filter(__isl_keep isl_union_map *umap,
	__isl_keep isl_schedule_node *node)
{
	isl_union_set *filter;
	int after;

	umap = isl_union_map_copy(umap);
	filter = isl_schedule_node_filter_get_filter(node);
	umap = isl_union_map_intersect_domain(umap, isl_union_set_copy(filter));
	umap = isl_union_map_intersect_range(umap, filter);

	after = after_in_child(umap, node);

	isl_union_map_free(umap);

	return after;
}

/* Is any domain element of "umap" scheduled after any of
 * the corresponding image elements by the tree rooted at
 * the set node "node"?
 *
 * This is only the case if this condition holds in any
 * of the (filter) children of the set node.
 * In particular, if the domain and the range of "umap"
 * are contained in different children, then the condition
 * does not hold.
 */
static int after_in_set(__isl_keep isl_union_map *umap,
	__isl_keep isl_schedule_node *node)
{
	int i, n;

	n = isl_schedule_node_n_children(node);
	for (i = 0; i < n; ++i) {
		isl_schedule_node *child;
		int after;

		child = isl_schedule_node_get_child(node, i);
		after = after_in_tree(umap, child);
		isl_schedule_node_free(child);

		if (after < 0 || after)
			return after;
	}

	return 0;
}

/* Return the filter of child "i" of "node".
 */
static __isl_give isl_union_set *child_filter(
	__isl_keep isl_schedule_node *node, int i)
{
	isl_schedule_node *child;
	isl_union_set *filter;

	child = isl_schedule_node_get_child(node, i);
	filter = isl_schedule_node_filter_get_filter(child);
	isl_schedule_node_free(child);

	return filter;
}

/* Is any domain element of "umap" scheduled after any of
 * the corresponding image elements by the tree rooted at
 * the sequence node "node"?
 *
 * This happens in particular if any domain element is
 * contained in a later child than one containing a range element or
 * if the condition holds within a given child in the sequence.
 * The later part of the condition is checked by after_in_set.
 */
static int after_in_sequence(__isl_keep isl_union_map *umap,
	__isl_keep isl_schedule_node *node)
{
	int i, j, n;
	isl_union_map *umap_i;
	int empty, after = 0;

	n = isl_schedule_node_n_children(node);
	for (i = 1; i < n; ++i) {
		isl_union_set *filter_i;

		umap_i = isl_union_map_copy(umap);
		filter_i = child_filter(node, i);
		umap_i = isl_union_map_intersect_domain(umap_i, filter_i);
		empty = isl_union_map_is_empty(umap_i);
		if (empty < 0)
			goto error;
		if (empty) {
			isl_union_map_free(umap_i);
			continue;
		}

		for (j = 0; j < i; ++j) {
			isl_union_set *filter_j;
			isl_union_map *umap_ij;

			umap_ij = isl_union_map_copy(umap_i);
			filter_j = child_filter(node, j);
			umap_ij = isl_union_map_intersect_range(umap_ij,
								filter_j);
			empty = isl_union_map_is_empty(umap_ij);
			isl_union_map_free(umap_ij);

			if (empty < 0)
				goto error;
			if (!empty)
				after = 1;
			if (after)
				break;
		}

		isl_union_map_free(umap_i);
		if (after)
			break;
	}

	if (after < 0 || after)
		return after;

	return after_in_set(umap, node);
error:
	isl_union_map_free(umap_i);
	return -1;
}

/* Is any domain element of "umap" scheduled after any of
 * the corresponding image elements by the tree rooted at "node"?
 *
 * If "umap" is empty, then clearly there is no such element.
 * Otherwise, consider the different types of nodes separately.
 */
static int after_in_tree(__isl_keep isl_union_map *umap,
	__isl_keep isl_schedule_node *node)
{
	int empty;
	enum isl_schedule_node_type type;

	empty = isl_union_map_is_empty(umap);
	if (empty < 0)
		return -1;
	if (empty)
		return 0;
	if (!node)
		return -1;

	type = isl_schedule_node_get_type(node);
	switch (type) {
	case isl_schedule_node_error:
		return -1;
	case isl_schedule_node_leaf:
		return 0;
	case isl_schedule_node_band:
		return after_in_band(umap, node);
	case isl_schedule_node_domain:
		isl_die(isl_schedule_node_get_ctx(node), isl_error_internal,
			"unexpected internal domain node", return -1);
	case isl_schedule_node_context:
		return after_in_context(umap, node);
	case isl_schedule_node_expansion:
		return after_in_expansion(umap, node);
	case isl_schedule_node_extension:
		return after_in_extension(umap, node);
	case isl_schedule_node_filter:
		return after_in_filter(umap, node);
	case isl_schedule_node_guard:
	case isl_schedule_node_mark:
		return after_in_child(umap, node);
	case isl_schedule_node_set:
		return after_in_set(umap, node);
	case isl_schedule_node_sequence:
		return after_in_sequence(umap, node);
	}

	return 1;
}

/* Is any domain element of "map1" scheduled after any domain
 * element of "map2" by the subtree underneath the current band node,
 * while at the same time being scheduled together by the current
 * band node, i.e., by "map1" and "map2?
 *
 * If the child of the current band node is a leaf, then
 * no element can be scheduled after any other element.
 *
 * Otherwise, we construct a relation between domain elements
 * of "map1" and domain elements of "map2" that are scheduled
 * together and then check if the subtree underneath the current
 * band node determines their relative order.
 */
static int after_in_subtree(__isl_keep isl_ast_build *build,
	__isl_keep isl_map *map1, __isl_keep isl_map *map2)
{
	isl_schedule_node *node;
	isl_map *map;
	isl_union_map *umap;
	int after;

	node = isl_ast_build_get_schedule_node(build);
	if (!node)
		return -1;
	node = isl_schedule_node_child(node, 0);
	if (isl_schedule_node_get_type(node) == isl_schedule_node_leaf) {
		isl_schedule_node_free(node);
		return 0;
	}
	map = isl_map_copy(map2);
	map = isl_map_apply_domain(map, isl_map_copy(map1));
	umap = isl_union_map_from_map(map);
	after = after_in_tree(umap, node);
	isl_union_map_free(umap);
	isl_schedule_node_free(node);
	return after;
}

/* Internal data for any_scheduled_after.
 *
 * "build" is the build in which the AST is constructed.
 * "depth" is the number of loops that have already been generated
 * "group_coscheduled" is a local copy of options->ast_build_group_coscheduled
 * "domain" is an array of set-map pairs corresponding to the different
 * iteration domains.  The set is the schedule domain, i.e., the domain
 * of the inverse schedule, while the map is the inverse schedule itself.
 */
struct isl_any_scheduled_after_data {
	isl_ast_build *build;
	int depth;
	int group_coscheduled;
	struct isl_set_map_pair *domain;
};

/* Is any element of domain "i" scheduled after any element of domain "j"
 * (for a common iteration of the first data->depth loops)?
 *
 * data->domain[i].set contains the domain of the inverse schedule
 * for domain "i", i.e., elements in the schedule domain.
 *
 * If we are inside a band of a schedule tree and there is a pair
 * of elements in the two domains that is schedule together by
 * the current band, then we check if any element of "i" may be schedule
 * after element of "j" by the descendants of the band node.
 *
 * If data->group_coscheduled is set, then we also return 1 if there
 * is any pair of elements in the two domains that are scheduled together.
 */
static isl_bool any_scheduled_after(int i, int j, void *user)
{
	struct isl_any_scheduled_after_data *data = user;
	int dim = isl_set_dim(data->domain[i].set, isl_dim_set);
	int pos;

	for (pos = data->depth; pos < dim; ++pos) {
		int follows;

		follows = isl_set_follows_at(data->domain[i].set,
						data->domain[j].set, pos);

		if (follows < -1)
			return isl_bool_error;
		if (follows > 0)
			return isl_bool_true;
		if (follows < 0)
			return isl_bool_false;
	}

	if (isl_ast_build_has_schedule_node(data->build)) {
		int after;

		after = after_in_subtree(data->build, data->domain[i].map,
					    data->domain[j].map);
		if (after < 0 || after)
			return after;
	}

	return data->group_coscheduled;
}

/* Look for independent components at the current depth and generate code
 * for each component separately.  The resulting lists of grafts are
 * merged in an attempt to combine grafts with identical guards.
 *
 * Code for two domains can be generated separately if all the elements
 * of one domain are scheduled before (or together with) all the elements
 * of the other domain.  We therefore consider the graph with as nodes
 * the domains and an edge between two nodes if any element of the first
 * node is scheduled after any element of the second node.
 * If the ast_build_group_coscheduled is set, then we also add an edge if
 * there is any pair of elements in the two domains that are scheduled
 * together.
 * Code is then generated (by generate_component)
 * for each of the strongly connected components in this graph
 * in their topological order.
 *
 * Since the test is performed on the domain of the inverse schedules of
 * the different domains, we precompute these domains and store
 * them in data.domain.
 */
static __isl_give isl_ast_graft_list *generate_components(
	__isl_take isl_union_map *executed, __isl_take isl_ast_build *build)
{
	int i;
	isl_ctx *ctx = isl_ast_build_get_ctx(build);
	int n = isl_union_map_n_map(executed);
	struct isl_any_scheduled_after_data data;
	struct isl_set_map_pair *next;
	struct isl_tarjan_graph *g = NULL;
	isl_ast_graft_list *list = NULL;
	int n_domain = 0;

	data.domain = isl_calloc_array(ctx, struct isl_set_map_pair, n);
	if (!data.domain)
		goto error;
	n_domain = n;

	next = data.domain;
	if (isl_union_map_foreach_map(executed, &extract_domain, &next) < 0)
		goto error;

	if (!build)
		goto error;
	data.build = build;
	data.depth = isl_ast_build_get_depth(build);
	data.group_coscheduled = isl_options_get_ast_build_group_coscheduled(ctx);
	g = isl_tarjan_graph_init(ctx, n, &any_scheduled_after, &data);
	if (!g)
		goto error;

	list = isl_ast_graft_list_alloc(ctx, 0);

	i = 0;
	while (list && n) {
		isl_ast_graft_list *list_c;
		int first = i;

		if (g->order[i] == -1)
			isl_die(ctx, isl_error_internal, "cannot happen",
				goto error);
		++i; --n;
		while (g->order[i] != -1) {
			++i; --n;
		}

		list_c = generate_component(data.domain,
					    g->order + first, i - first,
					    isl_ast_build_copy(build));
		list = isl_ast_graft_list_merge(list, list_c, build);

		++i;
	}

	if (0)
error:		list = isl_ast_graft_list_free(list);
	isl_tarjan_graph_free(g);
	for (i = 0; i < n_domain; ++i) {
		isl_map_free(data.domain[i].map);
		isl_set_free(data.domain[i].set);
	}
	free(data.domain);
	isl_union_map_free(executed);
	isl_ast_build_free(build);

	return list;
}

/* Generate code for the next level (and all inner levels).
 *
 * If "executed" is empty, i.e., no code needs to be generated,
 * then we return an empty list.
 *
 * If we have already generated code for all loop levels, then we pass
 * control to generate_inner_level.
 *
 * If "executed" lives in a single space, i.e., if code needs to be
 * generated for a single domain, then there can only be a single
 * component and we go directly to generate_shifted_component.
 * Otherwise, we call generate_components to detect the components
 * and to call generate_component on each of them separately.
 */
static __isl_give isl_ast_graft_list *generate_next_level(
	__isl_take isl_union_map *executed, __isl_take isl_ast_build *build)
{
	int depth;

	if (!build || !executed)
		goto error;

	if (isl_union_map_is_empty(executed)) {
		isl_ctx *ctx = isl_ast_build_get_ctx(build);
		isl_union_map_free(executed);
		isl_ast_build_free(build);
		return isl_ast_graft_list_alloc(ctx, 0);
	}

	depth = isl_ast_build_get_depth(build);
	if (depth >= isl_ast_build_dim(build, isl_dim_set))
		return generate_inner_level(executed, build);

	if (isl_union_map_n_map(executed) == 1)
		return generate_shifted_component(executed, build);

	return generate_components(executed, build);
error:
	isl_union_map_free(executed);
	isl_ast_build_free(build);
	return NULL;
}

/* Internal data structure used by isl_ast_build_node_from_schedule_map.
 * internal, executed and build are the inputs to generate_code.
 * list collects the output.
 */
struct isl_generate_code_data {
	int internal;
	isl_union_map *executed;
	isl_ast_build *build;

	isl_ast_graft_list *list;
};

/* Given an inverse schedule in terms of the external build schedule, i.e.,
 *
 *	[E -> S] -> D
 *
 * with E the external build schedule and S the additional schedule "space",
 * reformulate the inverse schedule in terms of the internal schedule domain,
 * i.e., return
 *
 *	[I -> S] -> D
 *
 * We first obtain a mapping
 *
 *	I -> E
 *
 * take the inverse and the product with S -> S, resulting in
 *
 *	[I -> S] -> [E -> S]
 *
 * Applying the map to the input produces the desired result.
 */
static __isl_give isl_union_map *internal_executed(
	__isl_take isl_union_map *executed, __isl_keep isl_space *space,
	__isl_keep isl_ast_build *build)
{
	isl_map *id, *proj;

	proj = isl_ast_build_get_schedule_map(build);
	proj = isl_map_reverse(proj);
	space = isl_space_map_from_set(isl_space_copy(space));
	id = isl_map_identity(space);
	proj = isl_map_product(proj, id);
	executed = isl_union_map_apply_domain(executed,
						isl_union_map_from_map(proj));
	return executed;
}

/* Generate an AST that visits the elements in the range of data->executed
 * in the relative order specified by the corresponding domain element(s)
 * for those domain elements that belong to "set".
 * Add the result to data->list.
 *
 * The caller ensures that "set" is a universe domain.
 * "space" is the space of the additional part of the schedule.
 * It is equal to the space of "set" if build->domain is parametric.
 * Otherwise, it is equal to the range of the wrapped space of "set".
 *
 * If the build space is not parametric and
 * if isl_ast_build_node_from_schedule_map
 * was called from an outside user (data->internal not set), then
 * the (inverse) schedule refers to the external build domain and needs to
 * be transformed to refer to the internal build domain.
 *
 * If the build space is parametric, then we add some of the parameter
 * constraints to the executed relation.  Adding these constraints
 * allows for an earlier detection of conflicts in some cases.
 * However, we do not want to divide the executed relation into
 * more disjuncts than necessary.  We therefore approximate
 * the constraints on the parameters by a single disjunct set.
 *
 * The build is extended to include the additional part of the schedule.
 * If the original build space was not parametric, then the options
 * in data->build refer only to the additional part of the schedule
 * and they need to be adjusted to refer to the complete AST build
 * domain.
 *
 * After having adjusted inverse schedule and build, we start generating
 * code with the outer loop of the current code generation
 * in generate_next_level.
 *
 * If the original build space was not parametric, we undo the embedding
 * on the resulting isl_ast_node_list so that it can be used within
 * the outer AST build.
 */
static isl_stat generate_code_in_space(struct isl_generate_code_data *data,
	__isl_take isl_set *set, __isl_take isl_space *space)
{
	isl_union_map *executed;
	isl_ast_build *build;
	isl_ast_graft_list *list;
	int embed;

	executed = isl_union_map_copy(data->executed);
	executed = isl_union_map_intersect_domain(executed,
						 isl_union_set_from_set(set));

	embed = !isl_set_is_params(data->build->domain);
	if (embed && !data->internal)
		executed = internal_executed(executed, space, data->build);
	if (!embed) {
		isl_set *domain;
		domain = isl_ast_build_get_domain(data->build);
		domain = isl_set_from_basic_set(isl_set_simple_hull(domain));
		executed = isl_union_map_intersect_params(executed, domain);
	}

	build = isl_ast_build_copy(data->build);
	build = isl_ast_build_product(build, space);

	list = generate_next_level(executed, build);

	list = isl_ast_graft_list_unembed(list, embed);

	data->list = isl_ast_graft_list_concat(data->list, list);

	return isl_stat_ok;
}

/* Generate an AST that visits the elements in the range of data->executed
 * in the relative order specified by the corresponding domain element(s)
 * for those domain elements that belong to "set".
 * Add the result to data->list.
 *
 * The caller ensures that "set" is a universe domain.
 *
 * If the build space S is not parametric, then the space of "set"
 * need to be a wrapped relation with S as domain.  That is, it needs
 * to be of the form
 *
 *	[S -> T]
 *
 * Check this property and pass control to generate_code_in_space
 * passing along T.
 * If the build space is not parametric, then T is the space of "set".
 */
static isl_stat generate_code_set(__isl_take isl_set *set, void *user)
{
	struct isl_generate_code_data *data = user;
	isl_space *space, *build_space;
	int is_domain;

	space = isl_set_get_space(set);

	if (isl_set_is_params(data->build->domain))
		return generate_code_in_space(data, set, space);

	build_space = isl_ast_build_get_space(data->build, data->internal);
	space = isl_space_unwrap(space);
	is_domain = isl_space_is_domain(build_space, space);
	isl_space_free(build_space);
	space = isl_space_range(space);

	if (is_domain < 0)
		goto error;
	if (!is_domain)
		isl_die(isl_set_get_ctx(set), isl_error_invalid,
			"invalid nested schedule space", goto error);

	return generate_code_in_space(data, set, space);
error:
	isl_set_free(set);
	isl_space_free(space);
	return isl_stat_error;
}

/* Generate an AST that visits the elements in the range of "executed"
 * in the relative order specified by the corresponding domain element(s).
 *
 * "build" is an isl_ast_build that has either been constructed by
 * isl_ast_build_from_context or passed to a callback set by
 * isl_ast_build_set_create_leaf.
 * In the first case, the space of the isl_ast_build is typically
 * a parametric space, although this is currently not enforced.
 * In the second case, the space is never a parametric space.
 * If the space S is not parametric, then the domain space(s) of "executed"
 * need to be wrapped relations with S as domain.
 *
 * If the domain of "executed" consists of several spaces, then an AST
 * is generated for each of them (in arbitrary order) and the results
 * are concatenated.
 *
 * If "internal" is set, then the domain "S" above refers to the internal
 * schedule domain representation.  Otherwise, it refers to the external
 * representation, as returned by isl_ast_build_get_schedule_space.
 *
 * We essentially run over all the spaces in the domain of "executed"
 * and call generate_code_set on each of them.
 */
static __isl_give isl_ast_graft_list *generate_code(
	__isl_take isl_union_map *executed, __isl_take isl_ast_build *build,
	int internal)
{
	isl_ctx *ctx;
	struct isl_generate_code_data data = { 0 };
	isl_space *space;
	isl_union_set *schedule_domain;
	isl_union_map *universe;

	if (!build)
		goto error;
	space = isl_ast_build_get_space(build, 1);
	space = isl_space_align_params(space,
				    isl_union_map_get_space(executed));
	space = isl_space_align_params(space,
				    isl_union_map_get_space(build->options));
	build = isl_ast_build_align_params(build, isl_space_copy(space));
	executed = isl_union_map_align_params(executed, space);
	if (!executed || !build)
		goto error;

	ctx = isl_ast_build_get_ctx(build);

	data.internal = internal;
	data.executed = executed;
	data.build = build;
	data.list = isl_ast_graft_list_alloc(ctx, 0);

	universe = isl_union_map_universe(isl_union_map_copy(executed));
	schedule_domain = isl_union_map_domain(universe);
	if (isl_union_set_foreach_set(schedule_domain, &generate_code_set,
					&data) < 0)
		data.list = isl_ast_graft_list_free(data.list);

	isl_union_set_free(schedule_domain);
	isl_union_map_free(executed);

	isl_ast_build_free(build);
	return data.list;
error:
	isl_union_map_free(executed);
	isl_ast_build_free(build);
	return NULL;
}

/* Generate an AST that visits the elements in the domain of "schedule"
 * in the relative order specified by the corresponding image element(s).
 *
 * "build" is an isl_ast_build that has either been constructed by
 * isl_ast_build_from_context or passed to a callback set by
 * isl_ast_build_set_create_leaf.
 * In the first case, the space of the isl_ast_build is typically
 * a parametric space, although this is currently not enforced.
 * In the second case, the space is never a parametric space.
 * If the space S is not parametric, then the range space(s) of "schedule"
 * need to be wrapped relations with S as domain.
 *
 * If the range of "schedule" consists of several spaces, then an AST
 * is generated for each of them (in arbitrary order) and the results
 * are concatenated.
 *
 * We first initialize the local copies of the relevant options.
 * We do this here rather than when the isl_ast_build is created
 * because the options may have changed between the construction
 * of the isl_ast_build and the call to isl_generate_code.
 *
 * The main computation is performed on an inverse schedule (with
 * the schedule domain in the domain and the elements to be executed
 * in the range) called "executed".
 */
__isl_give isl_ast_node *isl_ast_build_node_from_schedule_map(
	__isl_keep isl_ast_build *build, __isl_take isl_union_map *schedule)
{
	isl_ast_graft_list *list;
	isl_ast_node *node;
	isl_union_map *executed;

	build = isl_ast_build_copy(build);
	build = isl_ast_build_set_single_valued(build, 0);
	schedule = isl_union_map_coalesce(schedule);
	schedule = isl_union_map_remove_redundancies(schedule);
	executed = isl_union_map_reverse(schedule);
	list = generate_code(executed, isl_ast_build_copy(build), 0);
	node = isl_ast_node_from_graft_list(list, build);
	isl_ast_build_free(build);

	return node;
}

/* The old name for isl_ast_build_node_from_schedule_map.
 * It is being kept for backward compatibility, but
 * it will be removed in the future.
 */
__isl_give isl_ast_node *isl_ast_build_ast_from_schedule(
	__isl_keep isl_ast_build *build, __isl_take isl_union_map *schedule)
{
	return isl_ast_build_node_from_schedule_map(build, schedule);
}

/* Generate an AST that visits the elements in the domain of "executed"
 * in the relative order specified by the band node "node" and its descendants.
 *
 * The relation "executed" maps the outer generated loop iterators
 * to the domain elements executed by those iterations.
 *
 * If the band is empty, we continue with its descendants.
 * Otherwise, we extend the build and the inverse schedule with
 * the additional space/partial schedule and continue generating
 * an AST in generate_next_level.
 * As soon as we have extended the inverse schedule with the additional
 * partial schedule, we look for equalities that may exists between
 * the old and the new part.
 */
static __isl_give isl_ast_graft_list *build_ast_from_band(
	__isl_take isl_ast_build *build, __isl_take isl_schedule_node *node,
	__isl_take isl_union_map *executed)
{
	isl_space *space;
	isl_multi_union_pw_aff *extra;
	isl_union_map *extra_umap;
	isl_ast_graft_list *list;
	unsigned n1, n2;

	if (!build || !node || !executed)
		goto error;

	if (isl_schedule_node_band_n_member(node) == 0)
		return build_ast_from_child(build, node, executed);

	extra = isl_schedule_node_band_get_partial_schedule(node);
	extra = isl_multi_union_pw_aff_align_params(extra,
				isl_ast_build_get_space(build, 1));
	space = isl_multi_union_pw_aff_get_space(extra);

	extra_umap = isl_union_map_from_multi_union_pw_aff(extra);
	extra_umap = isl_union_map_reverse(extra_umap);

	executed = isl_union_map_domain_product(executed, extra_umap);
	executed = isl_union_map_detect_equalities(executed);

	n1 = isl_ast_build_dim(build, isl_dim_param);
	build = isl_ast_build_product(build, space);
	n2 = isl_ast_build_dim(build, isl_dim_param);
	if (n2 > n1)
		isl_die(isl_ast_build_get_ctx(build), isl_error_invalid,
			"band node is not allowed to introduce new parameters",
			build = isl_ast_build_free(build));
	build = isl_ast_build_set_schedule_node(build, node);

	list = generate_next_level(executed, build);

	list = isl_ast_graft_list_unembed(list, 1);

	return list;
error:
	isl_schedule_node_free(node);
	isl_union_map_free(executed);
	isl_ast_build_free(build);
	return NULL;
}

/* Hoist a list of grafts (in practice containing a single graft)
 * from "sub_build" (which includes extra context information)
 * to "build".
 *
 * In particular, project out all additional parameters introduced
 * by the context node from the enforced constraints and the guard
 * of the single graft.
 */
static __isl_give isl_ast_graft_list *hoist_out_of_context(
	__isl_take isl_ast_graft_list *list, __isl_keep isl_ast_build *build,
	__isl_keep isl_ast_build *sub_build)
{
	isl_ast_graft *graft;
	isl_basic_set *enforced;
	isl_set *guard;
	unsigned n_param, extra_param;

	if (!build || !sub_build)
		return isl_ast_graft_list_free(list);

	n_param = isl_ast_build_dim(build, isl_dim_param);
	extra_param = isl_ast_build_dim(sub_build, isl_dim_param);

	if (extra_param == n_param)
		return list;

	extra_param -= n_param;
	enforced = isl_ast_graft_list_extract_shared_enforced(list, sub_build);
	enforced = isl_basic_set_project_out(enforced, isl_dim_param,
							n_param, extra_param);
	enforced = isl_basic_set_remove_unknown_divs(enforced);
	guard = isl_ast_graft_list_extract_hoistable_guard(list, sub_build);
	guard = isl_set_remove_divs_involving_dims(guard, isl_dim_param,
							n_param, extra_param);
	guard = isl_set_project_out(guard, isl_dim_param, n_param, extra_param);
	guard = isl_set_compute_divs(guard);
	graft = isl_ast_graft_alloc_from_children(list, guard, enforced,
							build, sub_build);
	list = isl_ast_graft_list_from_ast_graft(graft);

	return list;
}

/* Generate an AST that visits the elements in the domain of "executed"
 * in the relative order specified by the context node "node"
 * and its descendants.
 *
 * The relation "executed" maps the outer generated loop iterators
 * to the domain elements executed by those iterations.
 *
 * The context node may introduce additional parameters as well as
 * constraints on the outer schedule dimensions or original parameters.
 *
 * We add the extra parameters to a new build and the context
 * constraints to both the build and (as a single disjunct)
 * to the domain of "executed".  Since the context constraints
 * are specified in terms of the input schedule, we first need
 * to map them to the internal schedule domain.
 *
 * After constructing the AST from the descendants of "node",
 * we combine the list of grafts into a single graft within
 * the new build, in order to be able to exploit the additional
 * context constraints during this combination.
 *
 * Additionally, if the current node is the outermost node in
 * the schedule tree (apart from the root domain node), we generate
 * all pending guards, again to be able to exploit the additional
 * context constraints.  We currently do not do this for internal
 * context nodes since we may still want to hoist conditions
 * to outer AST nodes.
 *
 * If the context node introduced any new parameters, then they
 * are removed from the set of enforced constraints and guard
 * in hoist_out_of_context.
 */
static __isl_give isl_ast_graft_list *build_ast_from_context(
	__isl_take isl_ast_build *build, __isl_take isl_schedule_node *node,
	__isl_take isl_union_map *executed)
{
	isl_set *context;
	isl_space *space;
	isl_multi_aff *internal2input;
	isl_ast_build *sub_build;
	isl_ast_graft_list *list;
	int n, depth;

	depth = isl_schedule_node_get_tree_depth(node);
	space = isl_ast_build_get_space(build, 1);
	context = isl_schedule_node_context_get_context(node);
	context = isl_set_align_params(context, space);
	sub_build = isl_ast_build_copy(build);
	space = isl_set_get_space(context);
	sub_build = isl_ast_build_align_params(sub_build, space);
	internal2input = isl_ast_build_get_internal2input(sub_build);
	context = isl_set_preimage_multi_aff(context, internal2input);
	sub_build = isl_ast_build_restrict_generated(sub_build,
					isl_set_copy(context));
	context = isl_set_from_basic_set(isl_set_simple_hull(context));
	executed = isl_union_map_intersect_domain(executed,
					isl_union_set_from_set(context));

	list = build_ast_from_child(isl_ast_build_copy(sub_build),
						node, executed);
	n = isl_ast_graft_list_n_ast_graft(list);
	if (n < 0)
		list = isl_ast_graft_list_free(list);

	list = isl_ast_graft_list_fuse(list, sub_build);
	if (depth == 1)
		list = isl_ast_graft_list_insert_pending_guard_nodes(list,
								sub_build);
	if (n >= 1)
		list = hoist_out_of_context(list, build, sub_build);

	isl_ast_build_free(build);
	isl_ast_build_free(sub_build);

	return list;
}

/* Generate an AST that visits the elements in the domain of "executed"
 * in the relative order specified by the expansion node "node" and
 * its descendants.
 *
 * The relation "executed" maps the outer generated loop iterators
 * to the domain elements executed by those iterations.
 *
 * We expand the domain elements by the expansion and
 * continue with the descendants of the node.
 */
static __isl_give isl_ast_graft_list *build_ast_from_expansion(
	__isl_take isl_ast_build *build, __isl_take isl_schedule_node *node,
	__isl_take isl_union_map *executed)
{
	isl_union_map *expansion;
	unsigned n1, n2;

	expansion = isl_schedule_node_expansion_get_expansion(node);
	expansion = isl_union_map_align_params(expansion,
				isl_union_map_get_space(executed));

	n1 = isl_union_map_dim(executed, isl_dim_param);
	executed = isl_union_map_apply_range(executed, expansion);
	n2 = isl_union_map_dim(executed, isl_dim_param);
	if (n2 > n1)
		isl_die(isl_ast_build_get_ctx(build), isl_error_invalid,
			"expansion node is not allowed to introduce "
			"new parameters", goto error);

	return build_ast_from_child(build, node, executed);
error:
	isl_ast_build_free(build);
	isl_schedule_node_free(node);
	isl_union_map_free(executed);
	return NULL;
}

/* Generate an AST that visits the elements in the domain of "executed"
 * in the relative order specified by the extension node "node" and
 * its descendants.
 *
 * The relation "executed" maps the outer generated loop iterators
 * to the domain elements executed by those iterations.
 *
 * Extend the inverse schedule with the extension applied to current
 * set of generated constraints.  Since the extension if formulated
 * in terms of the input schedule, it first needs to be transformed
 * to refer to the internal schedule.
 */
static __isl_give isl_ast_graft_list *build_ast_from_extension(
	__isl_take isl_ast_build *build, __isl_take isl_schedule_node *node,
	__isl_take isl_union_map *executed)
{
	isl_union_set *schedule_domain;
	isl_union_map *extension;
	isl_set *set;

	set = isl_ast_build_get_generated(build);
	set = isl_set_from_basic_set(isl_set_simple_hull(set));
	schedule_domain = isl_union_set_from_set(set);

	extension = isl_schedule_node_extension_get_extension(node);

	extension = isl_union_map_preimage_domain_multi_aff(extension,
			isl_multi_aff_copy(build->internal2input));
	extension = isl_union_map_intersect_domain(extension, schedule_domain);
	extension = isl_ast_build_substitute_values_union_map_domain(build,
								    extension);
	executed = isl_union_map_union(executed, extension);

	return build_ast_from_child(build, node, executed);
}

/* Generate an AST that visits the elements in the domain of "executed"
 * in the relative order specified by the filter node "node" and
 * its descendants.
 *
 * The relation "executed" maps the outer generated loop iterators
 * to the domain elements executed by those iterations.
 *
 * We simply intersect the iteration domain (i.e., the range of "executed")
 * with the filter and continue with the descendants of the node,
 * unless the resulting inverse schedule is empty, in which
 * case we return an empty list.
 *
 * If the result of the intersection is equal to the original "executed"
 * relation, then keep the original representation since the intersection
 * may have unnecessarily broken up the relation into a greater number
 * of disjuncts.
 */
static __isl_give isl_ast_graft_list *build_ast_from_filter(
	__isl_take isl_ast_build *build, __isl_take isl_schedule_node *node,
	__isl_take isl_union_map *executed)
{
	isl_ctx *ctx;
	isl_union_set *filter;
	isl_union_map *orig;
	isl_ast_graft_list *list;
	int empty;
	isl_bool unchanged;
	unsigned n1, n2;

	orig = isl_union_map_copy(executed);
	if (!build || !node || !executed)
		goto error;

	filter = isl_schedule_node_filter_get_filter(node);
	filter = isl_union_set_align_params(filter,
				isl_union_map_get_space(executed));
	n1 = isl_union_map_dim(executed, isl_dim_param);
	executed = isl_union_map_intersect_range(executed, filter);
	n2 = isl_union_map_dim(executed, isl_dim_param);
	if (n2 > n1)
		isl_die(isl_ast_build_get_ctx(build), isl_error_invalid,
			"filter node is not allowed to introduce "
			"new parameters", goto error);

	unchanged = isl_union_map_is_subset(orig, executed);
	empty = isl_union_map_is_empty(executed);
	if (unchanged < 0 || empty < 0)
		goto error;
	if (unchanged) {
		isl_union_map_free(executed);
		return build_ast_from_child(build, node, orig);
	}
	isl_union_map_free(orig);
	if (!empty)
		return build_ast_from_child(build, node, executed);

	ctx = isl_ast_build_get_ctx(build);
	list = isl_ast_graft_list_alloc(ctx, 0);
	isl_ast_build_free(build);
	isl_schedule_node_free(node);
	isl_union_map_free(executed);
	return list;
error:
	isl_ast_build_free(build);
	isl_schedule_node_free(node);
	isl_union_map_free(executed);
	isl_union_map_free(orig);
	return NULL;
}

/* Generate an AST that visits the elements in the domain of "executed"
 * in the relative order specified by the guard node "node" and
 * its descendants.
 *
 * The relation "executed" maps the outer generated loop iterators
 * to the domain elements executed by those iterations.
 *
 * Ensure that the associated guard is enforced by the outer AST
 * constructs by adding it to the guard of the graft.
 * Since we know that we will enforce the guard, we can also include it
 * in the generated constraints used to construct an AST for
 * the descendant nodes.
 */
static __isl_give isl_ast_graft_list *build_ast_from_guard(
	__isl_take isl_ast_build *build, __isl_take isl_schedule_node *node,
	__isl_take isl_union_map *executed)
{
	isl_space *space;
	isl_set *guard, *hoisted;
	isl_basic_set *enforced;
	isl_ast_build *sub_build;
	isl_ast_graft *graft;
	isl_ast_graft_list *list;
	unsigned n1, n2;

	space = isl_ast_build_get_space(build, 1);
	guard = isl_schedule_node_guard_get_guard(node);
	n1 = isl_space_dim(space, isl_dim_param);
	guard = isl_set_align_params(guard, space);
	n2 = isl_set_dim(guard, isl_dim_param);
	if (n2 > n1)
		isl_die(isl_ast_build_get_ctx(build), isl_error_invalid,
			"guard node is not allowed to introduce "
			"new parameters", guard = isl_set_free(guard));
	guard = isl_set_preimage_multi_aff(guard,
			isl_multi_aff_copy(build->internal2input));
	guard = isl_ast_build_specialize(build, guard);
	guard = isl_set_gist(guard, isl_set_copy(build->generated));

	sub_build = isl_ast_build_copy(build);
	sub_build = isl_ast_build_restrict_generated(sub_build,
							isl_set_copy(guard));

	list = build_ast_from_child(isl_ast_build_copy(sub_build),
							node, executed);

	hoisted = isl_ast_graft_list_extract_hoistable_guard(list, sub_build);
	if (isl_set_n_basic_set(hoisted) > 1)
		list = isl_ast_graft_list_gist_guards(list,
						    isl_set_copy(hoisted));
	guard = isl_set_intersect(guard, hoisted);
	enforced = extract_shared_enforced(list, build);
	graft = isl_ast_graft_alloc_from_children(list, guard, enforced,
						    build, sub_build);

	isl_ast_build_free(sub_build);
	isl_ast_build_free(build);
	return isl_ast_graft_list_from_ast_graft(graft);
}

/* Call the before_each_mark callback, if requested by the user.
 *
 * Return 0 on success and -1 on error.
 *
 * The caller is responsible for recording the current inverse schedule
 * in "build".
 */
static isl_stat before_each_mark(__isl_keep isl_id *mark,
	__isl_keep isl_ast_build *build)
{
	if (!build)
		return isl_stat_error;
	if (!build->before_each_mark)
		return isl_stat_ok;
	return build->before_each_mark(mark, build,
					build->before_each_mark_user);
}

/* Call the after_each_mark callback, if requested by the user.
 *
 * The caller is responsible for recording the current inverse schedule
 * in "build".
 */
static __isl_give isl_ast_graft *after_each_mark(
	__isl_take isl_ast_graft *graft, __isl_keep isl_ast_build *build)
{
	if (!graft || !build)
		return isl_ast_graft_free(graft);
	if (!build->after_each_mark)
		return graft;
	graft->node = build->after_each_mark(graft->node, build,
						build->after_each_mark_user);
	if (!graft->node)
		return isl_ast_graft_free(graft);
	return graft;
}


/* Generate an AST that visits the elements in the domain of "executed"
 * in the relative order specified by the mark node "node" and
 * its descendants.
 *
 * The relation "executed" maps the outer generated loop iterators
 * to the domain elements executed by those iterations.

 * Since we may be calling before_each_mark and after_each_mark
 * callbacks, we record the current inverse schedule in the build.
 *
 * We generate an AST for the child of the mark node, combine
 * the graft list into a single graft and then insert the mark
 * in the AST of that single graft.
 */
static __isl_give isl_ast_graft_list *build_ast_from_mark(
	__isl_take isl_ast_build *build, __isl_take isl_schedule_node *node,
	__isl_take isl_union_map *executed)
{
	isl_id *mark;
	isl_ast_graft *graft;
	isl_ast_graft_list *list;
	int n;

	build = isl_ast_build_set_executed(build, isl_union_map_copy(executed));

	mark = isl_schedule_node_mark_get_id(node);
	if (before_each_mark(mark, build) < 0)
		node = isl_schedule_node_free(node);

	list = build_ast_from_child(isl_ast_build_copy(build), node, executed);
	list = isl_ast_graft_list_fuse(list, build);
	n = isl_ast_graft_list_n_ast_graft(list);
	if (n < 0)
		list = isl_ast_graft_list_free(list);
	if (n == 0) {
		isl_id_free(mark);
	} else {
		graft = isl_ast_graft_list_get_ast_graft(list, 0);
		graft = isl_ast_graft_insert_mark(graft, mark);
		graft = after_each_mark(graft, build);
		list = isl_ast_graft_list_set_ast_graft(list, 0, graft);
	}
	isl_ast_build_free(build);

	return list;
}

static __isl_give isl_ast_graft_list *build_ast_from_schedule_node(
	__isl_take isl_ast_build *build, __isl_take isl_schedule_node *node,
	__isl_take isl_union_map *executed);

/* Generate an AST that visits the elements in the domain of "executed"
 * in the relative order specified by the sequence (or set) node "node" and
 * its descendants.
 *
 * The relation "executed" maps the outer generated loop iterators
 * to the domain elements executed by those iterations.
 *
 * We simply generate an AST for each of the children and concatenate
 * the results.
 */
static __isl_give isl_ast_graft_list *build_ast_from_sequence(
	__isl_take isl_ast_build *build, __isl_take isl_schedule_node *node,
	__isl_take isl_union_map *executed)
{
	int i, n;
	isl_ctx *ctx;
	isl_ast_graft_list *list;

	ctx = isl_ast_build_get_ctx(build);
	list = isl_ast_graft_list_alloc(ctx, 0);

	n = isl_schedule_node_n_children(node);
	for (i = 0; i < n; ++i) {
		isl_schedule_node *child;
		isl_ast_graft_list *list_i;

		child = isl_schedule_node_get_child(node, i);
		list_i = build_ast_from_schedule_node(isl_ast_build_copy(build),
					child, isl_union_map_copy(executed));
		list = isl_ast_graft_list_concat(list, list_i);
	}
	isl_ast_build_free(build);
	isl_schedule_node_free(node);
	isl_union_map_free(executed);

	return list;
}

/* Generate an AST that visits the elements in the domain of "executed"
 * in the relative order specified by the node "node" and its descendants.
 *
 * The relation "executed" maps the outer generated loop iterators
 * to the domain elements executed by those iterations.
 *
 * If the node is a leaf, then we pass control to generate_inner_level.
 * Note that the current build does not refer to any band node, so
 * that generate_inner_level will not try to visit the child of
 * the leaf node.
 *
 * The other node types are handled in separate functions.
 * Set nodes are currently treated in the same way as sequence nodes.
 * The children of a set node may be executed in any order,
 * including the order of the children.
 */
static __isl_give isl_ast_graft_list *build_ast_from_schedule_node(
	__isl_take isl_ast_build *build, __isl_take isl_schedule_node *node,
	__isl_take isl_union_map *executed)
{
	enum isl_schedule_node_type type;

	type = isl_schedule_node_get_type(node);

	switch (type) {
	case isl_schedule_node_error:
		goto error;
	case isl_schedule_node_leaf:
		isl_schedule_node_free(node);
		return generate_inner_level(executed, build);
	case isl_schedule_node_band:
		return build_ast_from_band(build, node, executed);
	case isl_schedule_node_context:
		return build_ast_from_context(build, node, executed);
	case isl_schedule_node_domain:
		isl_die(isl_schedule_node_get_ctx(node), isl_error_unsupported,
			"unexpected internal domain node", goto error);
	case isl_schedule_node_expansion:
		return build_ast_from_expansion(build, node, executed);
	case isl_schedule_node_extension:
		return build_ast_from_extension(build, node, executed);
	case isl_schedule_node_filter:
		return build_ast_from_filter(build, node, executed);
	case isl_schedule_node_guard:
		return build_ast_from_guard(build, node, executed);
	case isl_schedule_node_mark:
		return build_ast_from_mark(build, node, executed);
	case isl_schedule_node_sequence:
	case isl_schedule_node_set:
		return build_ast_from_sequence(build, node, executed);
	}

	isl_die(isl_ast_build_get_ctx(build), isl_error_internal,
		"unhandled type", goto error);
error:
	isl_union_map_free(executed);
	isl_schedule_node_free(node);
	isl_ast_build_free(build);

	return NULL;
}

/* Generate an AST that visits the elements in the domain of "executed"
 * in the relative order specified by the (single) child of "node" and
 * its descendants.
 *
 * The relation "executed" maps the outer generated loop iterators
 * to the domain elements executed by those iterations.
 *
 * This function is never called on a leaf, set or sequence node,
 * so the node always has exactly one child.
 */
static __isl_give isl_ast_graft_list *build_ast_from_child(
	__isl_take isl_ast_build *build, __isl_take isl_schedule_node *node,
	__isl_take isl_union_map *executed)
{
	node = isl_schedule_node_child(node, 0);
	return build_ast_from_schedule_node(build, node, executed);
}

/* Generate an AST that visits the elements in the domain of the domain
 * node "node" in the relative order specified by its descendants.
 *
 * An initial inverse schedule is created that maps a zero-dimensional
 * schedule space to the node domain.
 * The input "build" is assumed to have a parametric domain and
 * is replaced by the same zero-dimensional schedule space.
 *
 * We also add some of the parameter constraints in the build domain
 * to the executed relation.  Adding these constraints
 * allows for an earlier detection of conflicts in some cases.
 * However, we do not want to divide the executed relation into
 * more disjuncts than necessary.  We therefore approximate
 * the constraints on the parameters by a single disjunct set.
 */
static __isl_give isl_ast_node *build_ast_from_domain(
	__isl_take isl_ast_build *build, __isl_take isl_schedule_node *node)
{
	isl_ctx *ctx;
	isl_union_set *domain, *schedule_domain;
	isl_union_map *executed;
	isl_space *space;
	isl_set *set;
	isl_ast_graft_list *list;
	isl_ast_node *ast;
	int is_params;

	if (!build)
		goto error;

	ctx = isl_ast_build_get_ctx(build);
	space = isl_ast_build_get_space(build, 1);
	is_params = isl_space_is_params(space);
	isl_space_free(space);
	if (is_params < 0)
		goto error;
	if (!is_params)
		isl_die(ctx, isl_error_unsupported,
			"expecting parametric initial context", goto error);

	domain = isl_schedule_node_domain_get_domain(node);
	domain = isl_union_set_coalesce(domain);

	space = isl_union_set_get_space(domain);
	space = isl_space_set_from_params(space);
	build = isl_ast_build_product(build, space);

	set = isl_ast_build_get_domain(build);
	set = isl_set_from_basic_set(isl_set_simple_hull(set));
	schedule_domain = isl_union_set_from_set(set);

	executed = isl_union_map_from_domain_and_range(schedule_domain, domain);
	list = build_ast_from_child(isl_ast_build_copy(build), node, executed);
	ast = isl_ast_node_from_graft_list(list, build);
	isl_ast_build_free(build);

	return ast;
error:
	isl_schedule_node_free(node);
	isl_ast_build_free(build);
	return NULL;
}

/* Generate an AST that visits the elements in the domain of "schedule"
 * in the relative order specified by the schedule tree.
 *
 * "build" is an isl_ast_build that has been created using
 * isl_ast_build_alloc or isl_ast_build_from_context based
 * on a parametric set.
 *
 * The construction starts at the root node of the schedule,
 * which is assumed to be a domain node.
 */
__isl_give isl_ast_node *isl_ast_build_node_from_schedule(
	__isl_keep isl_ast_build *build, __isl_take isl_schedule *schedule)
{
	isl_ctx *ctx;
	isl_schedule_node *node;

	if (!build || !schedule)
		goto error;

	ctx = isl_ast_build_get_ctx(build);

	node = isl_schedule_get_root(schedule);
	if (!node)
		goto error;
	isl_schedule_free(schedule);

	build = isl_ast_build_copy(build);
	build = isl_ast_build_set_single_valued(build, 0);
	if (isl_schedule_node_get_type(node) != isl_schedule_node_domain)
		isl_die(ctx, isl_error_unsupported,
			"expecting root domain node",
			build = isl_ast_build_free(build));
	return build_ast_from_domain(build, node);
error:
	isl_schedule_free(schedule);
	return NULL;
}