summarylogtreecommitdiffstats
path: root/66219.patch
blob: c15de90ed4700909aa7034bb8a20804b7bfb2a78 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
From 55473c81535db8890d35e29cff852b737954ce80 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Wed, 6 Oct 2021 11:52:58 -0700
Subject: [PATCH 01/30] Refactor cub namespace

---
 aten/src/ATen/cuda/cub.cuh                    | 102 +++++++++---------
 aten/src/ATen/cuda/cub_definitions.cuh        |  17 +++
 caffe2/core/context_gpu.cu                    |   1 +
 caffe2/operators/accuracy_op.cu               |   1 +
 caffe2/operators/affine_channel_op.cu         |   1 +
 caffe2/operators/arg_ops.cu                   |   2 +-
 caffe2/operators/batch_moments_op.cu          |   1 +
 caffe2/operators/batch_sparse_to_dense_op.cu  |   1 +
 caffe2/operators/boolean_mask_ops.cu          |   2 +-
 caffe2/operators/cross_entropy_op.cu          |   1 +
 caffe2/operators/distance_op.cu               |   1 +
 caffe2/operators/elementwise_div_op.cu        |   2 +-
 caffe2/operators/elementwise_linear_op.cu     |   1 +
 caffe2/operators/elementwise_mul_op.cu        |   2 +-
 caffe2/operators/elementwise_ops.cu           |   1 +
 caffe2/operators/find_op.cu                   |   1 +
 caffe2/operators/generate_proposals_op.cu     |   2 +-
 caffe2/operators/normalize_ops.cu             |   1 +
 caffe2/operators/one_hot_ops.cu               |   1 +
 caffe2/operators/pack_segments.cu             |   2 +-
 caffe2/operators/prelu_op.cu                  |   1 +
 caffe2/operators/reduce_front_back_max_ops.cu |   1 +
 .../reduce_front_back_sum_mean_ops.cu         |   1 +
 caffe2/operators/reduction_ops.cu             |   2 +-
 caffe2/operators/rmac_regions_op.cu           |   7 ++
 caffe2/operators/segment_reduction_op_gpu.cuh |   1 +
 caffe2/operators/sequence_ops.cu              |   2 +-
 caffe2/operators/softmax_ops.cu               |   1 +
 .../operators/spatial_batch_norm_op_impl.cuh  |   2 +-
 caffe2/sgd/adagrad_fused_op_gpu.cu            |   1 +
 caffe2/sgd/adagrad_op_gpu.cu                  |   1 +
 caffe2/sgd/adam_op_gpu.cu                     |   1 +
 caffe2/utils/math/reduce.cu                   |   2 +-
 caffe2/utils/math/reduce.cuh                  |   2 +-
 caffe2/utils/math_gpu.cu                      |   2 +-
 cmake/Dependencies.cmake                      |   5 +
 36 files changed, 115 insertions(+), 60 deletions(-)
 create mode 100644 aten/src/ATen/cuda/cub_definitions.cuh

diff --git a/aten/src/ATen/cuda/cub.cuh b/aten/src/ATen/cuda/cub.cuh
index 5d8ae777ebef..39938efc48be 100644
--- a/aten/src/ATen/cuda/cub.cuh
+++ b/aten/src/ATen/cuda/cub.cuh
@@ -5,16 +5,24 @@
 #include <iterator>
 #include <limits>
 
-// include cub in a safe manner, see:
-// https://github.com/pytorch/pytorch/pull/55292
+#include <ATen/cuda/cub_definitions.cuh>
+
+#if CUB_SUPPORTS_WRAPPED_NAMESPACE()
+
+#include <cub/cub.cuh>
+
+#else
+
 #undef CUB_NS_POSTFIX //undef to avoid redefinition warnings
 #undef CUB_NS_PREFIX
-#define CUB_NS_PREFIX namespace at { namespace cuda { namespace detail {
-#define CUB_NS_POSTFIX }}}
+#define CUB_NS_PREFIX namespace at_cuda_detail {
+#define CUB_NS_POSTFIX }
 #include <cub/cub.cuh>
 #undef CUB_NS_POSTFIX
 #undef CUB_NS_PREFIX
 
+#endif
+
 #include <ATen/cuda/Exceptions.h>
 #include <c10/cuda/CUDACachingAllocator.h>
 #include <c10/cuda/CUDAStream.h>
@@ -33,16 +41,41 @@
 #define NO_ROCM(x)
 #else
 #define NO_ROCM(x) x
+#endif
 
-namespace at { namespace native {
+#if !defined(USE_ROCM) && !CUB_SUPPORTS_NV_BFLOAT16()
 
-namespace cub = at::cuda::detail::cub;
+namespace at_cuda_detail {
+// backport https://github.com/NVIDIA/cub/pull/306 for c10::BFloat16
 
-}}
+template <>
+struct cub::FpLimits<c10::BFloat16>
+{
+    static __host__ __device__ __forceinline__ c10::BFloat16 Max() {
+        unsigned short max_word = 0x7F7F;
+        return reinterpret_cast<c10::BFloat16&>(max_word);
+    }
+
+    static __host__ __device__ __forceinline__ c10::BFloat16 Lowest() {
+        unsigned short lowest_word = 0xFF7F;
+        return reinterpret_cast<c10::BFloat16&>(lowest_word);
+    }
+};
+
+template <> struct cub::NumericTraits<c10::BFloat16>: cub::BaseTraits<cub::FLOATING_POINT, true, false, unsigned short, c10::BFloat16> {};
+}
 #endif
 
+namespace at { namespace native {
+namespace cub = at_cuda_detail::cub;
+}}
+namespace caffe2 {
+namespace cub = at_cuda_detail::cub;
+}
+
 namespace at {
 namespace cuda {
+namespace cub {
 
 namespace detail {
 
@@ -55,44 +88,17 @@ struct cuda_type<c10::Half> {
   using type = __half;
 };
 
-#if defined(CUDA_VERSION) && CUDA_VERSION >= 11050
-// cub sort support for __nv_bfloat16 is added to cub 1.13 in
-// https://github.com/NVIDIA/cub/pull/306 and according to
-// https://github.com/NVIDIA/cub#releases, 1.13 is included in
-// CUDA Toolkit 11.5
+#if CUB_SUPPORTS_NV_BFLOAT16()
 
-// waiting for https://github.com/NVIDIA/cub/pull/306 to land on CUDA
 template<>
 struct cuda_type<c10::BFloat16> {
   using type = __nv_bfloat16;
 };
 
-#elif !defined(__HIP_PLATFORM_HCC__)
-
-// backport https://github.com/NVIDIA/cub/pull/306 for c10::BFloat16
-
-template <>
-struct cub::FpLimits<c10::BFloat16>
-{
-    static __host__ __device__ __forceinline__ c10::BFloat16 Max() {
-        unsigned short max_word = 0x7F7F;
-        return reinterpret_cast<c10::BFloat16&>(max_word);
-    }
-
-    static __host__ __device__ __forceinline__ c10::BFloat16 Lowest() {
-        unsigned short lowest_word = 0xFF7F;
-        return reinterpret_cast<c10::BFloat16&>(lowest_word);
-    }
-};
-
-template <> struct cub::NumericTraits<c10::BFloat16>: cub::BaseTraits<cub::FLOATING_POINT, true, false, unsigned short, c10::BFloat16> {};
-
 #endif
 
 }  // namespace detail
 
-namespace cub {
-
 inline int get_num_bits(uint64_t max_key) {
   int num_bits = 1;
   while (max_key > 1) {
@@ -115,11 +121,11 @@ static inline void sort_keys(
   key_t_ *keys_out_ = reinterpret_cast<key_t_*>(keys_out);
 
   if (descending) {
-    CUB_WRAPPER(NO_ROCM(detail)::cub::DeviceRadixSort::SortKeysDescending,
+    CUB_WRAPPER(NO_ROCM(at_cuda_detail)::cub::DeviceRadixSort::SortKeysDescending,
       keys_in_, keys_out_, n,
       begin_bit, end_bit, c10::cuda::getCurrentCUDAStream());
   } else {
-    CUB_WRAPPER(NO_ROCM(detail)::cub::DeviceRadixSort::SortKeys,
+    CUB_WRAPPER(NO_ROCM(at_cuda_detail)::cub::DeviceRadixSort::SortKeys,
       keys_in_, keys_out_, n,
       begin_bit, end_bit, c10::cuda::getCurrentCUDAStream());
   }
@@ -147,11 +153,11 @@ static inline void sort_pairs(
   key_t_ *keys_out_ = reinterpret_cast<key_t_*>(keys_out);
 
   if (descending) {
-    CUB_WRAPPER(NO_ROCM(detail)::cub::DeviceRadixSort::SortPairsDescending,
+    CUB_WRAPPER(NO_ROCM(at_cuda_detail)::cub::DeviceRadixSort::SortPairsDescending,
       keys_in_, keys_out_, values_in, values_out, n,
       begin_bit, end_bit, c10::cuda::getCurrentCUDAStream());
   } else {
-    CUB_WRAPPER(NO_ROCM(detail)::cub::DeviceRadixSort::SortPairs,
+    CUB_WRAPPER(NO_ROCM(at_cuda_detail)::cub::DeviceRadixSort::SortPairs,
       keys_in_, keys_out_, values_in, values_out, n,
       begin_bit, end_bit, c10::cuda::getCurrentCUDAStream());
   }
@@ -183,12 +189,12 @@ static inline void segmented_sort_pairs(
   key_t_ *keys_out_ = reinterpret_cast<key_t_*>(keys_out);
 
   if (descending) {
-    CUB_WRAPPER(NO_ROCM(detail)::cub::DeviceSegmentedRadixSort::SortPairsDescending,
+    CUB_WRAPPER(NO_ROCM(at_cuda_detail)::cub::DeviceSegmentedRadixSort::SortPairsDescending,
       keys_in_, keys_out_, values_in, values_out,
       num_elements, num_segments, begin_offsets, end_offsets,
       begin_bit, end_bit, c10::cuda::getCurrentCUDAStream());
   } else {
-    CUB_WRAPPER(NO_ROCM(detail)::cub::DeviceSegmentedRadixSort::SortPairs,
+    CUB_WRAPPER(NO_ROCM(at_cuda_detail)::cub::DeviceSegmentedRadixSort::SortPairs,
       keys_in_, keys_out_, values_in, values_out,
       num_elements, num_segments, begin_offsets, end_offsets,
       begin_bit, end_bit, c10::cuda::getCurrentCUDAStream());
@@ -240,7 +246,7 @@ inline void inclusive_scan(InputIteratorT input, OutputIteratorT output, ScanOpT
   // so split at int_max/2
   constexpr int max_cub_size = std::numeric_limits<int>::max() / 2 + 1; // 2**30
   int size_cub = std::min<int64_t>(num_items, max_cub_size);
-  CUB_WRAPPER(NO_ROCM(detail)::cub::DeviceScan::InclusiveScan,
+  CUB_WRAPPER(NO_ROCM(at_cuda_detail)::cub::DeviceScan::InclusiveScan,
       input,
       output,
       scan_op,
@@ -260,7 +266,7 @@ inline void inclusive_scan(InputIteratorT input, OutputIteratorT output, ScanOpT
         first_elem_ptr,
         scan_op);
     C10_CUDA_KERNEL_LAUNCH_CHECK();
-    using ArgIndexInputIterator = NO_ROCM(detail)::cub::ArgIndexInputIterator<InputIteratorT>;
+    using ArgIndexInputIterator = NO_ROCM(at_cuda_detail)::cub::ArgIndexInputIterator<InputIteratorT>;
     using tuple = typename ArgIndexInputIterator::value_type;
     auto input_iter_transform = [=] __device__ (const tuple &x)->input_t  {
       if (x.key == 0) {
@@ -269,9 +275,9 @@ inline void inclusive_scan(InputIteratorT input, OutputIteratorT output, ScanOpT
         return x.value;
       }
     };
-    auto input_ = NO_ROCM(detail)::cub::TransformInputIterator<input_t, decltype(input_iter_transform), ArgIndexInputIterator>(
+    auto input_ = NO_ROCM(at_cuda_detail)::cub::TransformInputIterator<input_t, decltype(input_iter_transform), ArgIndexInputIterator>(
       ArgIndexInputIterator(input + i), input_iter_transform);
-    CUB_WRAPPER(NO_ROCM(detail)::cub::DeviceScan::InclusiveScan,
+    CUB_WRAPPER(NO_ROCM(at_cuda_detail)::cub::DeviceScan::InclusiveScan,
         input_,
         output + i,
         scan_op,
@@ -287,7 +293,7 @@ inline void exclusive_scan(InputIteratorT input, OutputIteratorT output, ScanOpT
   // so split at int_max/2
   constexpr int max_cub_size = std::numeric_limits<int>::max() / 2 + 1; // 2**30
   int size_cub = std::min<int64_t>(num_items, max_cub_size);
-  CUB_WRAPPER(NO_ROCM(detail)::cub::DeviceScan::ExclusiveScan,
+  CUB_WRAPPER(NO_ROCM(at_cuda_detail)::cub::DeviceScan::ExclusiveScan,
       input,
       output,
       scan_op,
@@ -309,7 +315,7 @@ inline void exclusive_scan(InputIteratorT input, OutputIteratorT output, ScanOpT
     C10_CUDA_KERNEL_LAUNCH_CHECK();
     auto input_ = impl::chained_iterator<InitValueT, InputIteratorT>{
       input + i, first_elem_ptr};
-    CUB_WRAPPER(NO_ROCM(detail)::cub::DeviceScan::InclusiveScan,
+    CUB_WRAPPER(NO_ROCM(at_cuda_detail)::cub::DeviceScan::InclusiveScan,
         input_,
         output + i,
         scan_op,
@@ -322,7 +328,7 @@ template<typename InputIteratorT , typename OutputIteratorT , typename NumSelect
 inline void unique(InputIteratorT input, OutputIteratorT output, NumSelectedIteratorT num_selected_out, int64_t num_items) {
   TORCH_CHECK(num_items <= std::numeric_limits<int>::max(),
     "cub unique does not support more than INT_MAX elements");
-  CUB_WRAPPER(NO_ROCM(detail)::cub::DeviceSelect::Unique,
+  CUB_WRAPPER(NO_ROCM(at_cuda_detail)::cub::DeviceSelect::Unique,
     input, output, num_selected_out, num_items, at::cuda::getCurrentCUDAStream());
 }
 
diff --git a/aten/src/ATen/cuda/cub_definitions.cuh b/aten/src/ATen/cuda/cub_definitions.cuh
new file mode 100644
index 000000000000..a7694fda4168
--- /dev/null
+++ b/aten/src/ATen/cuda/cub_definitions.cuh
@@ -0,0 +1,17 @@
+#include <cub/version.cuh>
+
+// cub sort support for __nv_bfloat16 is added to cub 1.13 in:
+// https://github.com/NVIDIA/cub/pull/306
+#if CUB_VERSION >= 101300
+#define CUB_SUPPORTS_NV_BFLOAT16() true
+#elif
+#define CUB_SUPPORTS_NV_BFLOAT16() false
+#endif
+
+// cub sort support for CUB_WRAPPED_NAMESPACE is added to cub 1.14 in:
+// https://github.com/NVIDIA/cub/pull/326
+#if CUB_VERSION >= 101400
+#define CUB_SUPPORTS_WRAPPED_NAMESPACE() true
+#elif
+#define CUB_SUPPORTS_WRAPPED_NAMESPACE() false
+#endif
\ No newline at end of file
diff --git a/caffe2/core/context_gpu.cu b/caffe2/core/context_gpu.cu
index c2b89945ada9..475ed61ab4f7 100644
--- a/caffe2/core/context_gpu.cu
+++ b/caffe2/core/context_gpu.cu
@@ -4,6 +4,7 @@
 #include <string>
 #include <unordered_map>
 
+#include <ATen/cuda/cub.cuh>
 #include <ATen/Context.h>
 #include <c10/cuda/CUDAFunctions.h>
 #include <c10/cuda/CUDACachingAllocator.h>
diff --git a/caffe2/operators/accuracy_op.cu b/caffe2/operators/accuracy_op.cu
index f06663d71a90..7ad2b09c238a 100644
--- a/caffe2/operators/accuracy_op.cu
+++ b/caffe2/operators/accuracy_op.cu
@@ -3,6 +3,7 @@
 #include "caffe2/utils/GpuAtomics.cuh"
 #include "caffe2/utils/math.h"
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 namespace caffe2 {
diff --git a/caffe2/operators/affine_channel_op.cu b/caffe2/operators/affine_channel_op.cu
index adf4ac55c0fc..f3d9e22c7e8c 100644
--- a/caffe2/operators/affine_channel_op.cu
+++ b/caffe2/operators/affine_channel_op.cu
@@ -1,5 +1,6 @@
 #include "caffe2/operators/affine_channel_op.h"
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/arg_ops.cu b/caffe2/operators/arg_ops.cu
index 7e90d25b836b..fbefe0774376 100644
--- a/caffe2/operators/arg_ops.cu
+++ b/caffe2/operators/arg_ops.cu
@@ -2,8 +2,8 @@
 
 #include <limits>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
-#include <cub/cub.cuh>
 
 #include "caffe2/core/common_gpu.h"
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/batch_moments_op.cu b/caffe2/operators/batch_moments_op.cu
index 4b693b5c04e2..65c43200e5bd 100644
--- a/caffe2/operators/batch_moments_op.cu
+++ b/caffe2/operators/batch_moments_op.cu
@@ -1,5 +1,6 @@
 #include "caffe2/operators/batch_moments_op.h"
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/batch_sparse_to_dense_op.cu b/caffe2/operators/batch_sparse_to_dense_op.cu
index aea2035a5d3d..2cb09deb8668 100644
--- a/caffe2/operators/batch_sparse_to_dense_op.cu
+++ b/caffe2/operators/batch_sparse_to_dense_op.cu
@@ -1,5 +1,6 @@
 #include "caffe2/operators/batch_sparse_to_dense_op.h"
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/device/device_scan.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/boolean_mask_ops.cu b/caffe2/operators/boolean_mask_ops.cu
index 214b7c13ba3c..c87688f51d64 100644
--- a/caffe2/operators/boolean_mask_ops.cu
+++ b/caffe2/operators/boolean_mask_ops.cu
@@ -3,7 +3,7 @@
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/boolean_mask_ops.h"
 
-#include <cub/cub.cuh>
+#include <ATen/cuda/cub.cuh>
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/cross_entropy_op.cu b/caffe2/operators/cross_entropy_op.cu
index 380e80399fc3..95f3ffddbf1f 100644
--- a/caffe2/operators/cross_entropy_op.cu
+++ b/caffe2/operators/cross_entropy_op.cu
@@ -1,4 +1,5 @@
 #include <assert.h>
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/distance_op.cu b/caffe2/operators/distance_op.cu
index 3a8bb337d541..d94691d5a9d9 100644
--- a/caffe2/operators/distance_op.cu
+++ b/caffe2/operators/distance_op.cu
@@ -4,6 +4,7 @@
 #include "caffe2/operators/distance_op.h"
 #include "caffe2/utils/conversions.h"
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 namespace caffe2 {
diff --git a/caffe2/operators/elementwise_div_op.cu b/caffe2/operators/elementwise_div_op.cu
index 42b103a0f110..ca9682326324 100644
--- a/caffe2/operators/elementwise_div_op.cu
+++ b/caffe2/operators/elementwise_div_op.cu
@@ -3,8 +3,8 @@
 #include <algorithm>
 #include <functional>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
-#include <cub/cub.cuh>
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/elementwise_ops_utils.h"
diff --git a/caffe2/operators/elementwise_linear_op.cu b/caffe2/operators/elementwise_linear_op.cu
index cc49115bffc5..c1c45263f34c 100644
--- a/caffe2/operators/elementwise_linear_op.cu
+++ b/caffe2/operators/elementwise_linear_op.cu
@@ -5,6 +5,7 @@
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/operator_fallback_gpu.h"
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 namespace caffe2 {
diff --git a/caffe2/operators/elementwise_mul_op.cu b/caffe2/operators/elementwise_mul_op.cu
index bdbf760cf95b..88c3da00edc3 100644
--- a/caffe2/operators/elementwise_mul_op.cu
+++ b/caffe2/operators/elementwise_mul_op.cu
@@ -3,8 +3,8 @@
 #include <algorithm>
 #include <functional>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
-#include <cub/cub.cuh>
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/elementwise_ops_utils.h"
diff --git a/caffe2/operators/elementwise_ops.cu b/caffe2/operators/elementwise_ops.cu
index c9ced33cf806..1ac0426d2ca7 100644
--- a/caffe2/operators/elementwise_ops.cu
+++ b/caffe2/operators/elementwise_ops.cu
@@ -1,5 +1,6 @@
 #include "caffe2/operators/elementwise_ops.h"
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_load.cuh>
 #include <cub/block/block_reduce.cuh>
 #include <cub/device/device_reduce.cuh>
diff --git a/caffe2/operators/find_op.cu b/caffe2/operators/find_op.cu
index f8ff2bab1637..666df335ce42 100644
--- a/caffe2/operators/find_op.cu
+++ b/caffe2/operators/find_op.cu
@@ -1,3 +1,4 @@
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/find_op.h"
diff --git a/caffe2/operators/generate_proposals_op.cu b/caffe2/operators/generate_proposals_op.cu
index cab0ad3d0b88..84906a8e8182 100644
--- a/caffe2/operators/generate_proposals_op.cu
+++ b/caffe2/operators/generate_proposals_op.cu
@@ -1,4 +1,4 @@
-#include <cub/cub.cuh>
+#include <ATen/cuda/cub.cuh>
 #include "caffe2/core/context.h"
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/generate_proposals_op.h"
diff --git a/caffe2/operators/normalize_ops.cu b/caffe2/operators/normalize_ops.cu
index 26df05308d88..468175df985f 100644
--- a/caffe2/operators/normalize_ops.cu
+++ b/caffe2/operators/normalize_ops.cu
@@ -1,5 +1,6 @@
 #include <algorithm>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/one_hot_ops.cu b/caffe2/operators/one_hot_ops.cu
index e521b3dd09df..86f82f78bb82 100644
--- a/caffe2/operators/one_hot_ops.cu
+++ b/caffe2/operators/one_hot_ops.cu
@@ -1,3 +1,4 @@
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/pack_segments.cu b/caffe2/operators/pack_segments.cu
index 7475100fd368..b9ed413d1e7b 100644
--- a/caffe2/operators/pack_segments.cu
+++ b/caffe2/operators/pack_segments.cu
@@ -1,4 +1,4 @@
-#include <cub/cub.cuh>
+#include <ATen/cuda/cub.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/pack_segments.h"
 
diff --git a/caffe2/operators/prelu_op.cu b/caffe2/operators/prelu_op.cu
index 745a393f075f..d29882086754 100644
--- a/caffe2/operators/prelu_op.cu
+++ b/caffe2/operators/prelu_op.cu
@@ -1,6 +1,7 @@
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/prelu_op.h"
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 namespace caffe2 {
diff --git a/caffe2/operators/reduce_front_back_max_ops.cu b/caffe2/operators/reduce_front_back_max_ops.cu
index ae91f8a6da72..ba62b2eff671 100644
--- a/caffe2/operators/reduce_front_back_max_ops.cu
+++ b/caffe2/operators/reduce_front_back_max_ops.cu
@@ -1,3 +1,4 @@
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/reduce_front_back_max_ops.h"
diff --git a/caffe2/operators/reduce_front_back_sum_mean_ops.cu b/caffe2/operators/reduce_front_back_sum_mean_ops.cu
index 476596f08425..586c20fe8d8e 100644
--- a/caffe2/operators/reduce_front_back_sum_mean_ops.cu
+++ b/caffe2/operators/reduce_front_back_sum_mean_ops.cu
@@ -1,3 +1,4 @@
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/reduce_front_back_sum_mean_ops.h"
diff --git a/caffe2/operators/reduction_ops.cu b/caffe2/operators/reduction_ops.cu
index ba55a66de588..0d94fab22a7f 100644
--- a/caffe2/operators/reduction_ops.cu
+++ b/caffe2/operators/reduction_ops.cu
@@ -2,7 +2,7 @@
 #include "caffe2/operators/reduction_ops.h"
 #include "caffe2/utils/conversions.h"
 
-#include <cub/cub.cuh>
+#include <ATen/cuda/cub.cuh>
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/rmac_regions_op.cu b/caffe2/operators/rmac_regions_op.cu
index 76c4d012d71a..39cc5fbc988d 100644
--- a/caffe2/operators/rmac_regions_op.cu
+++ b/caffe2/operators/rmac_regions_op.cu
@@ -1,3 +1,4 @@
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
@@ -10,6 +11,9 @@
 #if defined(USE_ROCM)
 namespace rocprim {
 #else
+#if CUB_SUPPORTS_WRAPPED_NAMESPACE()
+namespace at_cuda_detail {
+#endif
 namespace cub {
 #endif
 
@@ -22,6 +26,9 @@ inline __host__ __device__ bool operator<(
 }
 
 } // namespace cub
+#if CUB_SUPPORTS_WRAPPED_NAMESPACE()
+}  // namespace at_cuda_detail
+#endif
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/segment_reduction_op_gpu.cuh b/caffe2/operators/segment_reduction_op_gpu.cuh
index 8d51196ee138..447617c6e9de 100644
--- a/caffe2/operators/segment_reduction_op_gpu.cuh
+++ b/caffe2/operators/segment_reduction_op_gpu.cuh
@@ -1,3 +1,4 @@
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include <cub/device/device_reduce.cuh>
 #include <cub/device/device_scan.cuh>
diff --git a/caffe2/operators/sequence_ops.cu b/caffe2/operators/sequence_ops.cu
index cc34effd3f22..e66d491f85e6 100644
--- a/caffe2/operators/sequence_ops.cu
+++ b/caffe2/operators/sequence_ops.cu
@@ -1,6 +1,6 @@
 #include <algorithm>
 
-#include <cub/cub.cuh>
+#include <ATen/cuda/cub.cuh>
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/sequence_ops.h"
diff --git a/caffe2/operators/softmax_ops.cu b/caffe2/operators/softmax_ops.cu
index 51c0cbc2bf6a..c01fcf3e0a48 100644
--- a/caffe2/operators/softmax_ops.cu
+++ b/caffe2/operators/softmax_ops.cu
@@ -1,4 +1,5 @@
 #include <cfloat>
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/spatial_batch_norm_op_impl.cuh b/caffe2/operators/spatial_batch_norm_op_impl.cuh
index edc076c7d718..f9b9fb58adc8 100644
--- a/caffe2/operators/spatial_batch_norm_op_impl.cuh
+++ b/caffe2/operators/spatial_batch_norm_op_impl.cuh
@@ -5,8 +5,8 @@
 
 #include <limits>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
-#include <cub/cub.cuh>
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/utils/math.h"
diff --git a/caffe2/sgd/adagrad_fused_op_gpu.cu b/caffe2/sgd/adagrad_fused_op_gpu.cu
index 2c2ad2cf76ae..396da5195125 100644
--- a/caffe2/sgd/adagrad_fused_op_gpu.cu
+++ b/caffe2/sgd/adagrad_fused_op_gpu.cu
@@ -2,6 +2,7 @@
 #include <c10/core/GeneratorImpl.h>
 #include <algorithm>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/device/device_radix_sort.cuh>
 #include "caffe2/sgd/adagrad_fused_op_gpu.cuh"
 #include "caffe2/utils/math.h"
diff --git a/caffe2/sgd/adagrad_op_gpu.cu b/caffe2/sgd/adagrad_op_gpu.cu
index 8abb3376ca87..a6fa842ddc80 100644
--- a/caffe2/sgd/adagrad_op_gpu.cu
+++ b/caffe2/sgd/adagrad_op_gpu.cu
@@ -1,5 +1,6 @@
 #include <algorithm>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/common_gpu.h"
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/sgd/adam_op_gpu.cu b/caffe2/sgd/adam_op_gpu.cu
index 42ab975faacb..4b59836b6a68 100644
--- a/caffe2/sgd/adam_op_gpu.cu
+++ b/caffe2/sgd/adam_op_gpu.cu
@@ -1,3 +1,4 @@
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/common_gpu.h"
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/utils/math/reduce.cu b/caffe2/utils/math/reduce.cu
index fc3e476b288b..20919334da50 100644
--- a/caffe2/utils/math/reduce.cu
+++ b/caffe2/utils/math/reduce.cu
@@ -6,8 +6,8 @@
 #include <numeric>
 #include <vector>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
-#include <cub/cub.cuh>
 
 #include <thrust/execution_policy.h>
 #include <thrust/reduce.h>
diff --git a/caffe2/utils/math/reduce.cuh b/caffe2/utils/math/reduce.cuh
index 0c43ad45a379..39ad553eec76 100644
--- a/caffe2/utils/math/reduce.cuh
+++ b/caffe2/utils/math/reduce.cuh
@@ -1,8 +1,8 @@
 #ifndef CAFFE2_UTILS_MATH_REDUCE_CUH_
 #define CAFFE2_UTILS_MATH_REDUCE_CUH_
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
-#include <cub/cub.cuh>
 
 #include "caffe2/core/common_gpu.h"
 
diff --git a/caffe2/utils/math_gpu.cu b/caffe2/utils/math_gpu.cu
index a37d4b744d73..b0a44fed34fb 100644
--- a/caffe2/utils/math_gpu.cu
+++ b/caffe2/utils/math_gpu.cu
@@ -7,8 +7,8 @@
 #include <numeric>
 #include <vector>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
-#include <cub/cub.cuh>
 
 #include <thrust/host_vector.h>
 #include <thrust/device_vector.h>
diff --git a/cmake/Dependencies.cmake b/cmake/Dependencies.cmake
index ca560288a41a..f127d8f2e5b1 100644
--- a/cmake/Dependencies.cmake
+++ b/cmake/Dependencies.cmake
@@ -1622,6 +1622,11 @@ if(NOT INTERN_BUILD_MOBILE)
     list(APPEND CUDA_NVCC_FLAGS "-Xcompiler" "-fPIC")
   endif()
 
+  # include cub in a safe manner, see:
+  # https://github.com/pytorch/pytorch/pull/55292
+  # https://github.com/NVIDIA/cub/releases/tag/1.14.0
+  list(APPEND CUDA_NVCC_FLAGS "-DCUB_WRAPPED_NAMESPACE=at_cuda_detail")
+
   if(CUDA_HAS_FP16 OR NOT ${CUDA_VERSION} LESS 7.5)
     message(STATUS "Found CUDA with FP16 support, compiling with torch.cuda.HalfTensor")
     list(APPEND CUDA_NVCC_FLAGS "-DCUDA_HAS_FP16=1" "-D__CUDA_NO_HALF_OPERATORS__" "-D__CUDA_NO_HALF_CONVERSIONS__"

From f0a6afd7f0a5c6210289869282c77a680d7dfa2e Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Wed, 6 Oct 2021 12:16:48 -0700
Subject: [PATCH 02/30] fix

---
 .github/workflows/lint.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.github/workflows/lint.yml b/.github/workflows/lint.yml
index 4c469878e63f..af472e00140a 100644
--- a/.github/workflows/lint.yml
+++ b/.github/workflows/lint.yml
@@ -97,7 +97,7 @@ jobs:
       - name: Ensure no direct cub include
         if: always()
         run: |
-          (! git --no-pager grep -I -no $'#include <cub/' --  ./aten  ':(exclude)aten/src/ATen/cuda/cub.cuh' || (echo "The above files have direct cub include; please include ATen/cuda/cub.cuh instead and wrap your cub calls in at::native namespace if necessary"; false))
+          (! git --no-pager grep -I -no $'#include <cub/' --  ./aten  ':(exclude)aten/src/ATen/cuda/cub.cuh' ':(exclude)aten/src/ATen/cuda/cub_definitions.cuh' || (echo "The above files have direct cub include; please include ATen/cuda/cub.cuh instead and wrap your cub calls in at::native namespace if necessary"; false))
       - name: Ensure no raw cuda api calls
         if: always()
         run: |

From 8edc6e961f1911f0fe04cb57af13b251f66c9153 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Wed, 6 Oct 2021 12:26:19 -0700
Subject: [PATCH 03/30] save

---
 .github/workflows/lint.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.github/workflows/lint.yml b/.github/workflows/lint.yml
index af472e00140a..0b2abe7a7483 100644
--- a/.github/workflows/lint.yml
+++ b/.github/workflows/lint.yml
@@ -97,7 +97,7 @@ jobs:
       - name: Ensure no direct cub include
         if: always()
         run: |
-          (! git --no-pager grep -I -no $'#include <cub/' --  ./aten  ':(exclude)aten/src/ATen/cuda/cub.cuh' ':(exclude)aten/src/ATen/cuda/cub_definitions.cuh' || (echo "The above files have direct cub include; please include ATen/cuda/cub.cuh instead and wrap your cub calls in at::native namespace if necessary"; false))
+          (! git --no-pager grep -I -no $'#include <cub/' --  ./aten  ':(exclude)aten/src/ATen/cuda/cub*.cuh' || (echo "The above files have direct cub include; please include ATen/cuda/cub.cuh instead and wrap your cub calls in at::native namespace if necessary"; false))
       - name: Ensure no raw cuda api calls
         if: always()
         run: |

From 0d50954a3747ed0f2c7cbfcd58f17a2a81d5929c Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Wed, 6 Oct 2021 12:32:16 -0700
Subject: [PATCH 04/30] fix

---
 aten/src/ATen/cuda/cub_definitions.cuh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/aten/src/ATen/cuda/cub_definitions.cuh b/aten/src/ATen/cuda/cub_definitions.cuh
index a7694fda4168..802f67f22a0f 100644
--- a/aten/src/ATen/cuda/cub_definitions.cuh
+++ b/aten/src/ATen/cuda/cub_definitions.cuh
@@ -14,4 +14,4 @@
 #define CUB_SUPPORTS_WRAPPED_NAMESPACE() true
 #elif
 #define CUB_SUPPORTS_WRAPPED_NAMESPACE() false
-#endif
\ No newline at end of file
+#endif

From 2ac5cf508d694995eee4710d9a6499bb3135d324 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Wed, 6 Oct 2021 14:04:40 -0700
Subject: [PATCH 05/30] fix

---
 aten/src/ATen/cuda/cub_definitions.cuh | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/aten/src/ATen/cuda/cub_definitions.cuh b/aten/src/ATen/cuda/cub_definitions.cuh
index 802f67f22a0f..1f5628fa9d3a 100644
--- a/aten/src/ATen/cuda/cub_definitions.cuh
+++ b/aten/src/ATen/cuda/cub_definitions.cuh
@@ -4,7 +4,7 @@
 // https://github.com/NVIDIA/cub/pull/306
 #if CUB_VERSION >= 101300
 #define CUB_SUPPORTS_NV_BFLOAT16() true
-#elif
+#else
 #define CUB_SUPPORTS_NV_BFLOAT16() false
 #endif
 
@@ -12,6 +12,6 @@
 // https://github.com/NVIDIA/cub/pull/326
 #if CUB_VERSION >= 101400
 #define CUB_SUPPORTS_WRAPPED_NAMESPACE() true
-#elif
+#else
 #define CUB_SUPPORTS_WRAPPED_NAMESPACE() false
 #endif

From 815f5a5981919f61aaab2d65597826ddf0495ac9 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Wed, 6 Oct 2021 14:06:32 -0700
Subject: [PATCH 06/30] fix rocm

---
 aten/src/ATen/cuda/cub_definitions.cuh | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/aten/src/ATen/cuda/cub_definitions.cuh b/aten/src/ATen/cuda/cub_definitions.cuh
index 1f5628fa9d3a..7da3ab716e41 100644
--- a/aten/src/ATen/cuda/cub_definitions.cuh
+++ b/aten/src/ATen/cuda/cub_definitions.cuh
@@ -1,4 +1,10 @@
+#pragma once
+
+#if !defined(USE_ROCM)
 #include <cub/version.cuh>
+#else
+#define CUB_VERSION 0
+#endif
 
 // cub sort support for __nv_bfloat16 is added to cub 1.13 in:
 // https://github.com/NVIDIA/cub/pull/306

From 02e9ca26bdc658edd1d960dc5d31e277b9b9afc1 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Mon, 11 Oct 2021 10:06:27 -0700
Subject: [PATCH 07/30] revert caffe2 change

---
 caffe2/core/context_gpu.cu                         | 1 -
 caffe2/operators/accuracy_op.cu                    | 1 -
 caffe2/operators/affine_channel_op.cu              | 1 -
 caffe2/operators/arg_ops.cu                        | 2 +-
 caffe2/operators/batch_moments_op.cu               | 1 -
 caffe2/operators/batch_sparse_to_dense_op.cu       | 1 -
 caffe2/operators/boolean_mask_ops.cu               | 2 +-
 caffe2/operators/cross_entropy_op.cu               | 1 -
 caffe2/operators/distance_op.cu                    | 1 -
 caffe2/operators/elementwise_div_op.cu             | 2 +-
 caffe2/operators/elementwise_linear_op.cu          | 1 -
 caffe2/operators/elementwise_mul_op.cu             | 2 +-
 caffe2/operators/elementwise_ops.cu                | 1 -
 caffe2/operators/find_op.cu                        | 1 -
 caffe2/operators/generate_proposals_op.cu          | 2 +-
 caffe2/operators/normalize_ops.cu                  | 1 -
 caffe2/operators/one_hot_ops.cu                    | 1 -
 caffe2/operators/pack_segments.cu                  | 2 +-
 caffe2/operators/prelu_op.cu                       | 1 -
 caffe2/operators/reduce_front_back_max_ops.cu      | 1 -
 caffe2/operators/reduce_front_back_sum_mean_ops.cu | 1 -
 caffe2/operators/reduction_ops.cu                  | 2 +-
 caffe2/operators/rmac_regions_op.cu                | 7 -------
 caffe2/operators/segment_reduction_op_gpu.cuh      | 1 -
 caffe2/operators/sequence_ops.cu                   | 2 +-
 caffe2/operators/softmax_ops.cu                    | 1 -
 caffe2/operators/spatial_batch_norm_op_impl.cuh    | 2 +-
 caffe2/sgd/adagrad_fused_op_gpu.cu                 | 1 -
 caffe2/sgd/adagrad_op_gpu.cu                       | 1 -
 caffe2/sgd/adam_op_gpu.cu                          | 1 -
 caffe2/utils/math/reduce.cu                        | 2 +-
 caffe2/utils/math/reduce.cuh                       | 2 +-
 caffe2/utils/math_gpu.cu                           | 2 +-
 33 files changed, 12 insertions(+), 39 deletions(-)

diff --git a/caffe2/core/context_gpu.cu b/caffe2/core/context_gpu.cu
index 475ed61ab4f7..c2b89945ada9 100644
--- a/caffe2/core/context_gpu.cu
+++ b/caffe2/core/context_gpu.cu
@@ -4,7 +4,6 @@
 #include <string>
 #include <unordered_map>
 
-#include <ATen/cuda/cub.cuh>
 #include <ATen/Context.h>
 #include <c10/cuda/CUDAFunctions.h>
 #include <c10/cuda/CUDACachingAllocator.h>
diff --git a/caffe2/operators/accuracy_op.cu b/caffe2/operators/accuracy_op.cu
index 7ad2b09c238a..f06663d71a90 100644
--- a/caffe2/operators/accuracy_op.cu
+++ b/caffe2/operators/accuracy_op.cu
@@ -3,7 +3,6 @@
 #include "caffe2/utils/GpuAtomics.cuh"
 #include "caffe2/utils/math.h"
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 namespace caffe2 {
diff --git a/caffe2/operators/affine_channel_op.cu b/caffe2/operators/affine_channel_op.cu
index f3d9e22c7e8c..adf4ac55c0fc 100644
--- a/caffe2/operators/affine_channel_op.cu
+++ b/caffe2/operators/affine_channel_op.cu
@@ -1,6 +1,5 @@
 #include "caffe2/operators/affine_channel_op.h"
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/arg_ops.cu b/caffe2/operators/arg_ops.cu
index fbefe0774376..7e90d25b836b 100644
--- a/caffe2/operators/arg_ops.cu
+++ b/caffe2/operators/arg_ops.cu
@@ -2,8 +2,8 @@
 
 #include <limits>
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
+#include <cub/cub.cuh>
 
 #include "caffe2/core/common_gpu.h"
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/batch_moments_op.cu b/caffe2/operators/batch_moments_op.cu
index 65c43200e5bd..4b693b5c04e2 100644
--- a/caffe2/operators/batch_moments_op.cu
+++ b/caffe2/operators/batch_moments_op.cu
@@ -1,6 +1,5 @@
 #include "caffe2/operators/batch_moments_op.h"
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/batch_sparse_to_dense_op.cu b/caffe2/operators/batch_sparse_to_dense_op.cu
index 2cb09deb8668..aea2035a5d3d 100644
--- a/caffe2/operators/batch_sparse_to_dense_op.cu
+++ b/caffe2/operators/batch_sparse_to_dense_op.cu
@@ -1,6 +1,5 @@
 #include "caffe2/operators/batch_sparse_to_dense_op.h"
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/device/device_scan.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/boolean_mask_ops.cu b/caffe2/operators/boolean_mask_ops.cu
index c87688f51d64..214b7c13ba3c 100644
--- a/caffe2/operators/boolean_mask_ops.cu
+++ b/caffe2/operators/boolean_mask_ops.cu
@@ -3,7 +3,7 @@
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/boolean_mask_ops.h"
 
-#include <ATen/cuda/cub.cuh>
+#include <cub/cub.cuh>
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/cross_entropy_op.cu b/caffe2/operators/cross_entropy_op.cu
index 95f3ffddbf1f..380e80399fc3 100644
--- a/caffe2/operators/cross_entropy_op.cu
+++ b/caffe2/operators/cross_entropy_op.cu
@@ -1,5 +1,4 @@
 #include <assert.h>
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/distance_op.cu b/caffe2/operators/distance_op.cu
index d94691d5a9d9..3a8bb337d541 100644
--- a/caffe2/operators/distance_op.cu
+++ b/caffe2/operators/distance_op.cu
@@ -4,7 +4,6 @@
 #include "caffe2/operators/distance_op.h"
 #include "caffe2/utils/conversions.h"
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 namespace caffe2 {
diff --git a/caffe2/operators/elementwise_div_op.cu b/caffe2/operators/elementwise_div_op.cu
index ca9682326324..42b103a0f110 100644
--- a/caffe2/operators/elementwise_div_op.cu
+++ b/caffe2/operators/elementwise_div_op.cu
@@ -3,8 +3,8 @@
 #include <algorithm>
 #include <functional>
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
+#include <cub/cub.cuh>
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/elementwise_ops_utils.h"
diff --git a/caffe2/operators/elementwise_linear_op.cu b/caffe2/operators/elementwise_linear_op.cu
index c1c45263f34c..cc49115bffc5 100644
--- a/caffe2/operators/elementwise_linear_op.cu
+++ b/caffe2/operators/elementwise_linear_op.cu
@@ -5,7 +5,6 @@
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/operator_fallback_gpu.h"
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 namespace caffe2 {
diff --git a/caffe2/operators/elementwise_mul_op.cu b/caffe2/operators/elementwise_mul_op.cu
index 88c3da00edc3..bdbf760cf95b 100644
--- a/caffe2/operators/elementwise_mul_op.cu
+++ b/caffe2/operators/elementwise_mul_op.cu
@@ -3,8 +3,8 @@
 #include <algorithm>
 #include <functional>
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
+#include <cub/cub.cuh>
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/elementwise_ops_utils.h"
diff --git a/caffe2/operators/elementwise_ops.cu b/caffe2/operators/elementwise_ops.cu
index 1ac0426d2ca7..c9ced33cf806 100644
--- a/caffe2/operators/elementwise_ops.cu
+++ b/caffe2/operators/elementwise_ops.cu
@@ -1,6 +1,5 @@
 #include "caffe2/operators/elementwise_ops.h"
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_load.cuh>
 #include <cub/block/block_reduce.cuh>
 #include <cub/device/device_reduce.cuh>
diff --git a/caffe2/operators/find_op.cu b/caffe2/operators/find_op.cu
index 666df335ce42..f8ff2bab1637 100644
--- a/caffe2/operators/find_op.cu
+++ b/caffe2/operators/find_op.cu
@@ -1,4 +1,3 @@
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/find_op.h"
diff --git a/caffe2/operators/generate_proposals_op.cu b/caffe2/operators/generate_proposals_op.cu
index 84906a8e8182..cab0ad3d0b88 100644
--- a/caffe2/operators/generate_proposals_op.cu
+++ b/caffe2/operators/generate_proposals_op.cu
@@ -1,4 +1,4 @@
-#include <ATen/cuda/cub.cuh>
+#include <cub/cub.cuh>
 #include "caffe2/core/context.h"
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/generate_proposals_op.h"
diff --git a/caffe2/operators/normalize_ops.cu b/caffe2/operators/normalize_ops.cu
index 468175df985f..26df05308d88 100644
--- a/caffe2/operators/normalize_ops.cu
+++ b/caffe2/operators/normalize_ops.cu
@@ -1,6 +1,5 @@
 #include <algorithm>
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/one_hot_ops.cu b/caffe2/operators/one_hot_ops.cu
index 86f82f78bb82..e521b3dd09df 100644
--- a/caffe2/operators/one_hot_ops.cu
+++ b/caffe2/operators/one_hot_ops.cu
@@ -1,4 +1,3 @@
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/pack_segments.cu b/caffe2/operators/pack_segments.cu
index b9ed413d1e7b..7475100fd368 100644
--- a/caffe2/operators/pack_segments.cu
+++ b/caffe2/operators/pack_segments.cu
@@ -1,4 +1,4 @@
-#include <ATen/cuda/cub.cuh>
+#include <cub/cub.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/pack_segments.h"
 
diff --git a/caffe2/operators/prelu_op.cu b/caffe2/operators/prelu_op.cu
index d29882086754..745a393f075f 100644
--- a/caffe2/operators/prelu_op.cu
+++ b/caffe2/operators/prelu_op.cu
@@ -1,7 +1,6 @@
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/prelu_op.h"
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 namespace caffe2 {
diff --git a/caffe2/operators/reduce_front_back_max_ops.cu b/caffe2/operators/reduce_front_back_max_ops.cu
index ba62b2eff671..ae91f8a6da72 100644
--- a/caffe2/operators/reduce_front_back_max_ops.cu
+++ b/caffe2/operators/reduce_front_back_max_ops.cu
@@ -1,4 +1,3 @@
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/reduce_front_back_max_ops.h"
diff --git a/caffe2/operators/reduce_front_back_sum_mean_ops.cu b/caffe2/operators/reduce_front_back_sum_mean_ops.cu
index 586c20fe8d8e..476596f08425 100644
--- a/caffe2/operators/reduce_front_back_sum_mean_ops.cu
+++ b/caffe2/operators/reduce_front_back_sum_mean_ops.cu
@@ -1,4 +1,3 @@
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/reduce_front_back_sum_mean_ops.h"
diff --git a/caffe2/operators/reduction_ops.cu b/caffe2/operators/reduction_ops.cu
index 0d94fab22a7f..ba55a66de588 100644
--- a/caffe2/operators/reduction_ops.cu
+++ b/caffe2/operators/reduction_ops.cu
@@ -2,7 +2,7 @@
 #include "caffe2/operators/reduction_ops.h"
 #include "caffe2/utils/conversions.h"
 
-#include <ATen/cuda/cub.cuh>
+#include <cub/cub.cuh>
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/rmac_regions_op.cu b/caffe2/operators/rmac_regions_op.cu
index 39cc5fbc988d..76c4d012d71a 100644
--- a/caffe2/operators/rmac_regions_op.cu
+++ b/caffe2/operators/rmac_regions_op.cu
@@ -1,4 +1,3 @@
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
@@ -11,9 +10,6 @@
 #if defined(USE_ROCM)
 namespace rocprim {
 #else
-#if CUB_SUPPORTS_WRAPPED_NAMESPACE()
-namespace at_cuda_detail {
-#endif
 namespace cub {
 #endif
 
@@ -26,9 +22,6 @@ inline __host__ __device__ bool operator<(
 }
 
 } // namespace cub
-#if CUB_SUPPORTS_WRAPPED_NAMESPACE()
-}  // namespace at_cuda_detail
-#endif
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/segment_reduction_op_gpu.cuh b/caffe2/operators/segment_reduction_op_gpu.cuh
index 447617c6e9de..8d51196ee138 100644
--- a/caffe2/operators/segment_reduction_op_gpu.cuh
+++ b/caffe2/operators/segment_reduction_op_gpu.cuh
@@ -1,4 +1,3 @@
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include <cub/device/device_reduce.cuh>
 #include <cub/device/device_scan.cuh>
diff --git a/caffe2/operators/sequence_ops.cu b/caffe2/operators/sequence_ops.cu
index e66d491f85e6..cc34effd3f22 100644
--- a/caffe2/operators/sequence_ops.cu
+++ b/caffe2/operators/sequence_ops.cu
@@ -1,6 +1,6 @@
 #include <algorithm>
 
-#include <ATen/cuda/cub.cuh>
+#include <cub/cub.cuh>
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/sequence_ops.h"
diff --git a/caffe2/operators/softmax_ops.cu b/caffe2/operators/softmax_ops.cu
index c01fcf3e0a48..51c0cbc2bf6a 100644
--- a/caffe2/operators/softmax_ops.cu
+++ b/caffe2/operators/softmax_ops.cu
@@ -1,5 +1,4 @@
 #include <cfloat>
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/spatial_batch_norm_op_impl.cuh b/caffe2/operators/spatial_batch_norm_op_impl.cuh
index f9b9fb58adc8..edc076c7d718 100644
--- a/caffe2/operators/spatial_batch_norm_op_impl.cuh
+++ b/caffe2/operators/spatial_batch_norm_op_impl.cuh
@@ -5,8 +5,8 @@
 
 #include <limits>
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
+#include <cub/cub.cuh>
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/utils/math.h"
diff --git a/caffe2/sgd/adagrad_fused_op_gpu.cu b/caffe2/sgd/adagrad_fused_op_gpu.cu
index 396da5195125..2c2ad2cf76ae 100644
--- a/caffe2/sgd/adagrad_fused_op_gpu.cu
+++ b/caffe2/sgd/adagrad_fused_op_gpu.cu
@@ -2,7 +2,6 @@
 #include <c10/core/GeneratorImpl.h>
 #include <algorithm>
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/device/device_radix_sort.cuh>
 #include "caffe2/sgd/adagrad_fused_op_gpu.cuh"
 #include "caffe2/utils/math.h"
diff --git a/caffe2/sgd/adagrad_op_gpu.cu b/caffe2/sgd/adagrad_op_gpu.cu
index a6fa842ddc80..8abb3376ca87 100644
--- a/caffe2/sgd/adagrad_op_gpu.cu
+++ b/caffe2/sgd/adagrad_op_gpu.cu
@@ -1,6 +1,5 @@
 #include <algorithm>
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/common_gpu.h"
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/sgd/adam_op_gpu.cu b/caffe2/sgd/adam_op_gpu.cu
index 4b59836b6a68..42ab975faacb 100644
--- a/caffe2/sgd/adam_op_gpu.cu
+++ b/caffe2/sgd/adam_op_gpu.cu
@@ -1,4 +1,3 @@
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/common_gpu.h"
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/utils/math/reduce.cu b/caffe2/utils/math/reduce.cu
index 20919334da50..fc3e476b288b 100644
--- a/caffe2/utils/math/reduce.cu
+++ b/caffe2/utils/math/reduce.cu
@@ -6,8 +6,8 @@
 #include <numeric>
 #include <vector>
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
+#include <cub/cub.cuh>
 
 #include <thrust/execution_policy.h>
 #include <thrust/reduce.h>
diff --git a/caffe2/utils/math/reduce.cuh b/caffe2/utils/math/reduce.cuh
index 39ad553eec76..0c43ad45a379 100644
--- a/caffe2/utils/math/reduce.cuh
+++ b/caffe2/utils/math/reduce.cuh
@@ -1,8 +1,8 @@
 #ifndef CAFFE2_UTILS_MATH_REDUCE_CUH_
 #define CAFFE2_UTILS_MATH_REDUCE_CUH_
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
+#include <cub/cub.cuh>
 
 #include "caffe2/core/common_gpu.h"
 
diff --git a/caffe2/utils/math_gpu.cu b/caffe2/utils/math_gpu.cu
index b0a44fed34fb..a37d4b744d73 100644
--- a/caffe2/utils/math_gpu.cu
+++ b/caffe2/utils/math_gpu.cu
@@ -7,8 +7,8 @@
 #include <numeric>
 #include <vector>
 
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
+#include <cub/cub.cuh>
 
 #include <thrust/host_vector.h>
 #include <thrust/device_vector.h>

From a9ca6d97ff78a4a7a6a6dca90b21ad24666bf6fe Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Mon, 11 Oct 2021 10:27:36 -0700
Subject: [PATCH 08/30] save

---
 aten/src/ATen/cuda/cub.cuh | 1 +
 1 file changed, 1 insertion(+)

diff --git a/aten/src/ATen/cuda/cub.cuh b/aten/src/ATen/cuda/cub.cuh
index 39938efc48be..23e90a81a16c 100644
--- a/aten/src/ATen/cuda/cub.cuh
+++ b/aten/src/ATen/cuda/cub.cuh
@@ -9,6 +9,7 @@
 
 #if CUB_SUPPORTS_WRAPPED_NAMESPACE()
 
+#define CUB_WRAPPED_NAMESPACE at_cuda_detail
 #include <cub/cub.cuh>
 
 #else

From 3695222a89a011cf58dfdb907515ffc72231f3af Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Mon, 11 Oct 2021 10:51:25 -0700
Subject: [PATCH 09/30] fix

---
 aten/src/ATen/cuda/cub.cuh | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/aten/src/ATen/cuda/cub.cuh b/aten/src/ATen/cuda/cub.cuh
index 23e90a81a16c..9014d4b7d2ec 100644
--- a/aten/src/ATen/cuda/cub.cuh
+++ b/aten/src/ATen/cuda/cub.cuh
@@ -67,12 +67,11 @@ template <> struct cub::NumericTraits<c10::BFloat16>: cub::BaseTraits<cub::FLOAT
 }
 #endif
 
+#if !defined(USE_ROCM)
 namespace at { namespace native {
 namespace cub = at_cuda_detail::cub;
 }}
-namespace caffe2 {
-namespace cub = at_cuda_detail::cub;
-}
+#endif
 
 namespace at {
 namespace cuda {

From 4d94df2cf61598bb6078364a0f628cb0321cdff5 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Tue, 12 Oct 2021 07:54:47 -0700
Subject: [PATCH 10/30] Revert "revert caffe2 change"

This reverts commit 02e9ca26bdc658edd1d960dc5d31e277b9b9afc1.
---
 caffe2/core/context_gpu.cu                         | 1 +
 caffe2/operators/accuracy_op.cu                    | 1 +
 caffe2/operators/affine_channel_op.cu              | 1 +
 caffe2/operators/arg_ops.cu                        | 2 +-
 caffe2/operators/batch_moments_op.cu               | 1 +
 caffe2/operators/batch_sparse_to_dense_op.cu       | 1 +
 caffe2/operators/boolean_mask_ops.cu               | 2 +-
 caffe2/operators/cross_entropy_op.cu               | 1 +
 caffe2/operators/distance_op.cu                    | 1 +
 caffe2/operators/elementwise_div_op.cu             | 2 +-
 caffe2/operators/elementwise_linear_op.cu          | 1 +
 caffe2/operators/elementwise_mul_op.cu             | 2 +-
 caffe2/operators/elementwise_ops.cu                | 1 +
 caffe2/operators/find_op.cu                        | 1 +
 caffe2/operators/generate_proposals_op.cu          | 2 +-
 caffe2/operators/normalize_ops.cu                  | 1 +
 caffe2/operators/one_hot_ops.cu                    | 1 +
 caffe2/operators/pack_segments.cu                  | 2 +-
 caffe2/operators/prelu_op.cu                       | 1 +
 caffe2/operators/reduce_front_back_max_ops.cu      | 1 +
 caffe2/operators/reduce_front_back_sum_mean_ops.cu | 1 +
 caffe2/operators/reduction_ops.cu                  | 2 +-
 caffe2/operators/rmac_regions_op.cu                | 7 +++++++
 caffe2/operators/segment_reduction_op_gpu.cuh      | 1 +
 caffe2/operators/sequence_ops.cu                   | 2 +-
 caffe2/operators/softmax_ops.cu                    | 1 +
 caffe2/operators/spatial_batch_norm_op_impl.cuh    | 2 +-
 caffe2/sgd/adagrad_fused_op_gpu.cu                 | 1 +
 caffe2/sgd/adagrad_op_gpu.cu                       | 1 +
 caffe2/sgd/adam_op_gpu.cu                          | 1 +
 caffe2/utils/math/reduce.cu                        | 2 +-
 caffe2/utils/math/reduce.cuh                       | 2 +-
 caffe2/utils/math_gpu.cu                           | 2 +-
 33 files changed, 39 insertions(+), 12 deletions(-)

diff --git a/caffe2/core/context_gpu.cu b/caffe2/core/context_gpu.cu
index c2b89945ada9..475ed61ab4f7 100644
--- a/caffe2/core/context_gpu.cu
+++ b/caffe2/core/context_gpu.cu
@@ -4,6 +4,7 @@
 #include <string>
 #include <unordered_map>
 
+#include <ATen/cuda/cub.cuh>
 #include <ATen/Context.h>
 #include <c10/cuda/CUDAFunctions.h>
 #include <c10/cuda/CUDACachingAllocator.h>
diff --git a/caffe2/operators/accuracy_op.cu b/caffe2/operators/accuracy_op.cu
index f06663d71a90..7ad2b09c238a 100644
--- a/caffe2/operators/accuracy_op.cu
+++ b/caffe2/operators/accuracy_op.cu
@@ -3,6 +3,7 @@
 #include "caffe2/utils/GpuAtomics.cuh"
 #include "caffe2/utils/math.h"
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 namespace caffe2 {
diff --git a/caffe2/operators/affine_channel_op.cu b/caffe2/operators/affine_channel_op.cu
index adf4ac55c0fc..f3d9e22c7e8c 100644
--- a/caffe2/operators/affine_channel_op.cu
+++ b/caffe2/operators/affine_channel_op.cu
@@ -1,5 +1,6 @@
 #include "caffe2/operators/affine_channel_op.h"
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/arg_ops.cu b/caffe2/operators/arg_ops.cu
index 7e90d25b836b..fbefe0774376 100644
--- a/caffe2/operators/arg_ops.cu
+++ b/caffe2/operators/arg_ops.cu
@@ -2,8 +2,8 @@
 
 #include <limits>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
-#include <cub/cub.cuh>
 
 #include "caffe2/core/common_gpu.h"
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/batch_moments_op.cu b/caffe2/operators/batch_moments_op.cu
index 4b693b5c04e2..65c43200e5bd 100644
--- a/caffe2/operators/batch_moments_op.cu
+++ b/caffe2/operators/batch_moments_op.cu
@@ -1,5 +1,6 @@
 #include "caffe2/operators/batch_moments_op.h"
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/batch_sparse_to_dense_op.cu b/caffe2/operators/batch_sparse_to_dense_op.cu
index aea2035a5d3d..2cb09deb8668 100644
--- a/caffe2/operators/batch_sparse_to_dense_op.cu
+++ b/caffe2/operators/batch_sparse_to_dense_op.cu
@@ -1,5 +1,6 @@
 #include "caffe2/operators/batch_sparse_to_dense_op.h"
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/device/device_scan.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/boolean_mask_ops.cu b/caffe2/operators/boolean_mask_ops.cu
index 214b7c13ba3c..c87688f51d64 100644
--- a/caffe2/operators/boolean_mask_ops.cu
+++ b/caffe2/operators/boolean_mask_ops.cu
@@ -3,7 +3,7 @@
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/boolean_mask_ops.h"
 
-#include <cub/cub.cuh>
+#include <ATen/cuda/cub.cuh>
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/cross_entropy_op.cu b/caffe2/operators/cross_entropy_op.cu
index 380e80399fc3..95f3ffddbf1f 100644
--- a/caffe2/operators/cross_entropy_op.cu
+++ b/caffe2/operators/cross_entropy_op.cu
@@ -1,4 +1,5 @@
 #include <assert.h>
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/distance_op.cu b/caffe2/operators/distance_op.cu
index 3a8bb337d541..d94691d5a9d9 100644
--- a/caffe2/operators/distance_op.cu
+++ b/caffe2/operators/distance_op.cu
@@ -4,6 +4,7 @@
 #include "caffe2/operators/distance_op.h"
 #include "caffe2/utils/conversions.h"
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 namespace caffe2 {
diff --git a/caffe2/operators/elementwise_div_op.cu b/caffe2/operators/elementwise_div_op.cu
index 42b103a0f110..ca9682326324 100644
--- a/caffe2/operators/elementwise_div_op.cu
+++ b/caffe2/operators/elementwise_div_op.cu
@@ -3,8 +3,8 @@
 #include <algorithm>
 #include <functional>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
-#include <cub/cub.cuh>
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/elementwise_ops_utils.h"
diff --git a/caffe2/operators/elementwise_linear_op.cu b/caffe2/operators/elementwise_linear_op.cu
index cc49115bffc5..c1c45263f34c 100644
--- a/caffe2/operators/elementwise_linear_op.cu
+++ b/caffe2/operators/elementwise_linear_op.cu
@@ -5,6 +5,7 @@
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/operator_fallback_gpu.h"
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 namespace caffe2 {
diff --git a/caffe2/operators/elementwise_mul_op.cu b/caffe2/operators/elementwise_mul_op.cu
index bdbf760cf95b..88c3da00edc3 100644
--- a/caffe2/operators/elementwise_mul_op.cu
+++ b/caffe2/operators/elementwise_mul_op.cu
@@ -3,8 +3,8 @@
 #include <algorithm>
 #include <functional>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
-#include <cub/cub.cuh>
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/elementwise_ops_utils.h"
diff --git a/caffe2/operators/elementwise_ops.cu b/caffe2/operators/elementwise_ops.cu
index c9ced33cf806..1ac0426d2ca7 100644
--- a/caffe2/operators/elementwise_ops.cu
+++ b/caffe2/operators/elementwise_ops.cu
@@ -1,5 +1,6 @@
 #include "caffe2/operators/elementwise_ops.h"
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_load.cuh>
 #include <cub/block/block_reduce.cuh>
 #include <cub/device/device_reduce.cuh>
diff --git a/caffe2/operators/find_op.cu b/caffe2/operators/find_op.cu
index f8ff2bab1637..666df335ce42 100644
--- a/caffe2/operators/find_op.cu
+++ b/caffe2/operators/find_op.cu
@@ -1,3 +1,4 @@
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/find_op.h"
diff --git a/caffe2/operators/generate_proposals_op.cu b/caffe2/operators/generate_proposals_op.cu
index cab0ad3d0b88..84906a8e8182 100644
--- a/caffe2/operators/generate_proposals_op.cu
+++ b/caffe2/operators/generate_proposals_op.cu
@@ -1,4 +1,4 @@
-#include <cub/cub.cuh>
+#include <ATen/cuda/cub.cuh>
 #include "caffe2/core/context.h"
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/generate_proposals_op.h"
diff --git a/caffe2/operators/normalize_ops.cu b/caffe2/operators/normalize_ops.cu
index 26df05308d88..468175df985f 100644
--- a/caffe2/operators/normalize_ops.cu
+++ b/caffe2/operators/normalize_ops.cu
@@ -1,5 +1,6 @@
 #include <algorithm>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/one_hot_ops.cu b/caffe2/operators/one_hot_ops.cu
index e521b3dd09df..86f82f78bb82 100644
--- a/caffe2/operators/one_hot_ops.cu
+++ b/caffe2/operators/one_hot_ops.cu
@@ -1,3 +1,4 @@
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/pack_segments.cu b/caffe2/operators/pack_segments.cu
index 7475100fd368..b9ed413d1e7b 100644
--- a/caffe2/operators/pack_segments.cu
+++ b/caffe2/operators/pack_segments.cu
@@ -1,4 +1,4 @@
-#include <cub/cub.cuh>
+#include <ATen/cuda/cub.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/pack_segments.h"
 
diff --git a/caffe2/operators/prelu_op.cu b/caffe2/operators/prelu_op.cu
index 745a393f075f..d29882086754 100644
--- a/caffe2/operators/prelu_op.cu
+++ b/caffe2/operators/prelu_op.cu
@@ -1,6 +1,7 @@
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/prelu_op.h"
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 namespace caffe2 {
diff --git a/caffe2/operators/reduce_front_back_max_ops.cu b/caffe2/operators/reduce_front_back_max_ops.cu
index ae91f8a6da72..ba62b2eff671 100644
--- a/caffe2/operators/reduce_front_back_max_ops.cu
+++ b/caffe2/operators/reduce_front_back_max_ops.cu
@@ -1,3 +1,4 @@
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/reduce_front_back_max_ops.h"
diff --git a/caffe2/operators/reduce_front_back_sum_mean_ops.cu b/caffe2/operators/reduce_front_back_sum_mean_ops.cu
index 476596f08425..586c20fe8d8e 100644
--- a/caffe2/operators/reduce_front_back_sum_mean_ops.cu
+++ b/caffe2/operators/reduce_front_back_sum_mean_ops.cu
@@ -1,3 +1,4 @@
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/reduce_front_back_sum_mean_ops.h"
diff --git a/caffe2/operators/reduction_ops.cu b/caffe2/operators/reduction_ops.cu
index ba55a66de588..0d94fab22a7f 100644
--- a/caffe2/operators/reduction_ops.cu
+++ b/caffe2/operators/reduction_ops.cu
@@ -2,7 +2,7 @@
 #include "caffe2/operators/reduction_ops.h"
 #include "caffe2/utils/conversions.h"
 
-#include <cub/cub.cuh>
+#include <ATen/cuda/cub.cuh>
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/rmac_regions_op.cu b/caffe2/operators/rmac_regions_op.cu
index 76c4d012d71a..39cc5fbc988d 100644
--- a/caffe2/operators/rmac_regions_op.cu
+++ b/caffe2/operators/rmac_regions_op.cu
@@ -1,3 +1,4 @@
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
@@ -10,6 +11,9 @@
 #if defined(USE_ROCM)
 namespace rocprim {
 #else
+#if CUB_SUPPORTS_WRAPPED_NAMESPACE()
+namespace at_cuda_detail {
+#endif
 namespace cub {
 #endif
 
@@ -22,6 +26,9 @@ inline __host__ __device__ bool operator<(
 }
 
 } // namespace cub
+#if CUB_SUPPORTS_WRAPPED_NAMESPACE()
+}  // namespace at_cuda_detail
+#endif
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/segment_reduction_op_gpu.cuh b/caffe2/operators/segment_reduction_op_gpu.cuh
index 8d51196ee138..447617c6e9de 100644
--- a/caffe2/operators/segment_reduction_op_gpu.cuh
+++ b/caffe2/operators/segment_reduction_op_gpu.cuh
@@ -1,3 +1,4 @@
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include <cub/device/device_reduce.cuh>
 #include <cub/device/device_scan.cuh>
diff --git a/caffe2/operators/sequence_ops.cu b/caffe2/operators/sequence_ops.cu
index cc34effd3f22..e66d491f85e6 100644
--- a/caffe2/operators/sequence_ops.cu
+++ b/caffe2/operators/sequence_ops.cu
@@ -1,6 +1,6 @@
 #include <algorithm>
 
-#include <cub/cub.cuh>
+#include <ATen/cuda/cub.cuh>
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/sequence_ops.h"
diff --git a/caffe2/operators/softmax_ops.cu b/caffe2/operators/softmax_ops.cu
index 51c0cbc2bf6a..c01fcf3e0a48 100644
--- a/caffe2/operators/softmax_ops.cu
+++ b/caffe2/operators/softmax_ops.cu
@@ -1,4 +1,5 @@
 #include <cfloat>
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/spatial_batch_norm_op_impl.cuh b/caffe2/operators/spatial_batch_norm_op_impl.cuh
index edc076c7d718..f9b9fb58adc8 100644
--- a/caffe2/operators/spatial_batch_norm_op_impl.cuh
+++ b/caffe2/operators/spatial_batch_norm_op_impl.cuh
@@ -5,8 +5,8 @@
 
 #include <limits>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
-#include <cub/cub.cuh>
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/utils/math.h"
diff --git a/caffe2/sgd/adagrad_fused_op_gpu.cu b/caffe2/sgd/adagrad_fused_op_gpu.cu
index 2c2ad2cf76ae..396da5195125 100644
--- a/caffe2/sgd/adagrad_fused_op_gpu.cu
+++ b/caffe2/sgd/adagrad_fused_op_gpu.cu
@@ -2,6 +2,7 @@
 #include <c10/core/GeneratorImpl.h>
 #include <algorithm>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/device/device_radix_sort.cuh>
 #include "caffe2/sgd/adagrad_fused_op_gpu.cuh"
 #include "caffe2/utils/math.h"
diff --git a/caffe2/sgd/adagrad_op_gpu.cu b/caffe2/sgd/adagrad_op_gpu.cu
index 8abb3376ca87..a6fa842ddc80 100644
--- a/caffe2/sgd/adagrad_op_gpu.cu
+++ b/caffe2/sgd/adagrad_op_gpu.cu
@@ -1,5 +1,6 @@
 #include <algorithm>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/common_gpu.h"
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/sgd/adam_op_gpu.cu b/caffe2/sgd/adam_op_gpu.cu
index 42ab975faacb..4b59836b6a68 100644
--- a/caffe2/sgd/adam_op_gpu.cu
+++ b/caffe2/sgd/adam_op_gpu.cu
@@ -1,3 +1,4 @@
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/common_gpu.h"
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/utils/math/reduce.cu b/caffe2/utils/math/reduce.cu
index fc3e476b288b..20919334da50 100644
--- a/caffe2/utils/math/reduce.cu
+++ b/caffe2/utils/math/reduce.cu
@@ -6,8 +6,8 @@
 #include <numeric>
 #include <vector>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
-#include <cub/cub.cuh>
 
 #include <thrust/execution_policy.h>
 #include <thrust/reduce.h>
diff --git a/caffe2/utils/math/reduce.cuh b/caffe2/utils/math/reduce.cuh
index 0c43ad45a379..39ad553eec76 100644
--- a/caffe2/utils/math/reduce.cuh
+++ b/caffe2/utils/math/reduce.cuh
@@ -1,8 +1,8 @@
 #ifndef CAFFE2_UTILS_MATH_REDUCE_CUH_
 #define CAFFE2_UTILS_MATH_REDUCE_CUH_
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
-#include <cub/cub.cuh>
 
 #include "caffe2/core/common_gpu.h"
 
diff --git a/caffe2/utils/math_gpu.cu b/caffe2/utils/math_gpu.cu
index a37d4b744d73..b0a44fed34fb 100644
--- a/caffe2/utils/math_gpu.cu
+++ b/caffe2/utils/math_gpu.cu
@@ -7,8 +7,8 @@
 #include <numeric>
 #include <vector>
 
+#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
-#include <cub/cub.cuh>
 
 #include <thrust/host_vector.h>
 #include <thrust/device_vector.h>

From 359666307907780e946284445e609131652a4739 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Tue, 12 Oct 2021 08:04:01 -0700
Subject: [PATCH 11/30] save

---
 aten/src/ATen/cuda/cub.cuh             | 4 +++-
 aten/src/ATen/cuda/cub_definitions.cuh | 5 ++++-
 cmake/Dependencies.cmake               | 4 ++++
 3 files changed, 11 insertions(+), 2 deletions(-)

diff --git a/aten/src/ATen/cuda/cub.cuh b/aten/src/ATen/cuda/cub.cuh
index 9014d4b7d2ec..4f39bde4b3c8 100644
--- a/aten/src/ATen/cuda/cub.cuh
+++ b/aten/src/ATen/cuda/cub.cuh
@@ -9,7 +9,6 @@
 
 #if CUB_SUPPORTS_WRAPPED_NAMESPACE()
 
-#define CUB_WRAPPED_NAMESPACE at_cuda_detail
 #include <cub/cub.cuh>
 
 #else
@@ -71,6 +70,9 @@ template <> struct cub::NumericTraits<c10::BFloat16>: cub::BaseTraits<cub::FLOAT
 namespace at { namespace native {
 namespace cub = at_cuda_detail::cub;
 }}
+namespace caffew {
+namespace cub = at_cuda_detail::cub;
+}
 #endif
 
 namespace at {
diff --git a/aten/src/ATen/cuda/cub_definitions.cuh b/aten/src/ATen/cuda/cub_definitions.cuh
index 7da3ab716e41..b4c82c9ae975 100644
--- a/aten/src/ATen/cuda/cub_definitions.cuh
+++ b/aten/src/ATen/cuda/cub_definitions.cuh
@@ -1,6 +1,7 @@
 #pragma once
 
 #if !defined(USE_ROCM)
+#include <cuda.h>
 #include <cub/version.cuh>
 #else
 #define CUB_VERSION 0
@@ -16,7 +17,9 @@
 
 // cub sort support for CUB_WRAPPED_NAMESPACE is added to cub 1.14 in:
 // https://github.com/NVIDIA/cub/pull/326
-#if CUB_VERSION >= 101400
+// CUB_WRAPPED_NAMESPACE is defined globally in cmake/Dependencies.cmake
+// starting from CUDA 11.4
+#if defined(CUDA_VERSION) && CUDA_VERSION >= 11040
 #define CUB_SUPPORTS_WRAPPED_NAMESPACE() true
 #else
 #define CUB_SUPPORTS_WRAPPED_NAMESPACE() false
diff --git a/cmake/Dependencies.cmake b/cmake/Dependencies.cmake
index 564fcebc0b08..bfee8f6dc78a 100644
--- a/cmake/Dependencies.cmake
+++ b/cmake/Dependencies.cmake
@@ -1618,6 +1618,10 @@ if(NOT INTERN_BUILD_MOBILE)
     set(CMAKE_CXX_STANDARD 14)
   endif()
 
+  if(NOT ${CUDA_VERSION} LESS 11.4)
+    string(APPEND CMAKE_CUDA_FLAGS " -DCUB_WRAPPED_NAMESPACE=at_cuda_detail")
+  endif()
+
   if(CUDA_HAS_FP16 OR NOT ${CUDA_VERSION} LESS 7.5)
     message(STATUS "Found CUDA with FP16 support, compiling with torch.cuda.HalfTensor")
     string(APPEND CMAKE_CUDA_FLAGS " -DCUDA_HAS_FP16=1"

From d4b3679d45cadbf9123a869cfefa001832bce04a Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Tue, 12 Oct 2021 08:09:00 -0700
Subject: [PATCH 12/30] save

---
 aten/src/ATen/cuda/cub.cuh             | 2 +-
 aten/src/ATen/cuda/cub_definitions.cuh | 4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/aten/src/ATen/cuda/cub.cuh b/aten/src/ATen/cuda/cub.cuh
index 4f39bde4b3c8..2583b7e640ec 100644
--- a/aten/src/ATen/cuda/cub.cuh
+++ b/aten/src/ATen/cuda/cub.cuh
@@ -7,7 +7,7 @@
 
 #include <ATen/cuda/cub_definitions.cuh>
 
-#if CUB_SUPPORTS_WRAPPED_NAMESPACE()
+#if USE_GLOBAL_CUB_WRAPPED_NAMESPACE()
 
 #include <cub/cub.cuh>
 
diff --git a/aten/src/ATen/cuda/cub_definitions.cuh b/aten/src/ATen/cuda/cub_definitions.cuh
index b4c82c9ae975..160d6f157396 100644
--- a/aten/src/ATen/cuda/cub_definitions.cuh
+++ b/aten/src/ATen/cuda/cub_definitions.cuh
@@ -20,7 +20,7 @@
 // CUB_WRAPPED_NAMESPACE is defined globally in cmake/Dependencies.cmake
 // starting from CUDA 11.4
 #if defined(CUDA_VERSION) && CUDA_VERSION >= 11040
-#define CUB_SUPPORTS_WRAPPED_NAMESPACE() true
+#define USE_GLOBAL_CUB_WRAPPED_NAMESPACE() true
 #else
-#define CUB_SUPPORTS_WRAPPED_NAMESPACE() false
+#define USE_GLOBAL_CUB_WRAPPED_NAMESPACE() false
 #endif

From 2b710dab756fceadf393d1c38cec91ef000e8fa8 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Tue, 12 Oct 2021 08:23:39 -0700
Subject: [PATCH 13/30] fix

---
 aten/src/ATen/cuda/cub.cuh | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/aten/src/ATen/cuda/cub.cuh b/aten/src/ATen/cuda/cub.cuh
index 2583b7e640ec..042009f3d8b4 100644
--- a/aten/src/ATen/cuda/cub.cuh
+++ b/aten/src/ATen/cuda/cub.cuh
@@ -68,10 +68,10 @@ template <> struct cub::NumericTraits<c10::BFloat16>: cub::BaseTraits<cub::FLOAT
 
 #if !defined(USE_ROCM)
 namespace at { namespace native {
-namespace cub = at_cuda_detail::cub;
+namespace cub = ::at_cuda_detail::cub;
 }}
-namespace caffew {
-namespace cub = at_cuda_detail::cub;
+namespace caffe2 {
+namespace cub = ::at_cuda_detail::cub;
 }
 #endif
 

From 34c57ca996c3249fe23e7037f98d1a9638187371 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Tue, 12 Oct 2021 08:31:35 -0700
Subject: [PATCH 14/30] fix

---
 caffe2/operators/rmac_regions_op.cu | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/caffe2/operators/rmac_regions_op.cu b/caffe2/operators/rmac_regions_op.cu
index 39cc5fbc988d..2f6c230574b0 100644
--- a/caffe2/operators/rmac_regions_op.cu
+++ b/caffe2/operators/rmac_regions_op.cu
@@ -11,7 +11,7 @@
 #if defined(USE_ROCM)
 namespace rocprim {
 #else
-#if CUB_SUPPORTS_WRAPPED_NAMESPACE()
+#if USE_GLOBAL_CUB_WRAPPED_NAMESPACE()
 namespace at_cuda_detail {
 #endif
 namespace cub {
@@ -26,7 +26,7 @@ inline __host__ __device__ bool operator<(
 }
 
 } // namespace cub
-#if CUB_SUPPORTS_WRAPPED_NAMESPACE()
+#if USE_GLOBAL_CUB_WRAPPED_NAMESPACE()
 }  // namespace at_cuda_detail
 #endif
 

From 3f6bce369e20fc75160c8151064e5cf3c4d871ce Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Tue, 12 Oct 2021 08:56:27 -0700
Subject: [PATCH 16/30] save

---
 aten/src/ATen/cuda/cub_definitions.cuh | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/aten/src/ATen/cuda/cub_definitions.cuh b/aten/src/ATen/cuda/cub_definitions.cuh
index 160d6f157396..7746885d6428 100644
--- a/aten/src/ATen/cuda/cub_definitions.cuh
+++ b/aten/src/ATen/cuda/cub_definitions.cuh
@@ -15,11 +15,11 @@
 #define CUB_SUPPORTS_NV_BFLOAT16() false
 #endif
 
-// cub sort support for CUB_WRAPPED_NAMESPACE is added to cub 1.14 in:
+// cub sort support for CUB_WRAPPED_NAMESPACE is added to cub 1.13.1 in:
 // https://github.com/NVIDIA/cub/pull/326
 // CUB_WRAPPED_NAMESPACE is defined globally in cmake/Dependencies.cmake
 // starting from CUDA 11.4
-#if defined(CUDA_VERSION) && CUDA_VERSION >= 11040
+#if CUB_VERSION >= 101301
 #define USE_GLOBAL_CUB_WRAPPED_NAMESPACE() true
 #else
 #define USE_GLOBAL_CUB_WRAPPED_NAMESPACE() false

From a2346ec6c5804085e49e81007ed70fd1aae0f333 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Tue, 12 Oct 2021 09:04:04 -0700
Subject: [PATCH 17/30] save

---
 aten/src/ATen/cuda/cub.cuh | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/aten/src/ATen/cuda/cub.cuh b/aten/src/ATen/cuda/cub.cuh
index 042009f3d8b4..17062586b0e3 100644
--- a/aten/src/ATen/cuda/cub.cuh
+++ b/aten/src/ATen/cuda/cub.cuh
@@ -17,9 +17,8 @@
 #undef CUB_NS_PREFIX
 #define CUB_NS_PREFIX namespace at_cuda_detail {
 #define CUB_NS_POSTFIX }
+#define CUB_NS_QUALIFIER ::at_cuda_detail::cub
 #include <cub/cub.cuh>
-#undef CUB_NS_POSTFIX
-#undef CUB_NS_PREFIX
 
 #endif
 

From add451ed3a579a794609bb26e82b31cec8527d03 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Tue, 12 Oct 2021 09:06:53 -0700
Subject: [PATCH 18/30] save

---
 aten/src/ATen/cuda/cub_definitions.cuh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/aten/src/ATen/cuda/cub_definitions.cuh b/aten/src/ATen/cuda/cub_definitions.cuh
index 7746885d6428..fb43441148fc 100644
--- a/aten/src/ATen/cuda/cub_definitions.cuh
+++ b/aten/src/ATen/cuda/cub_definitions.cuh
@@ -19,7 +19,7 @@
 // https://github.com/NVIDIA/cub/pull/326
 // CUB_WRAPPED_NAMESPACE is defined globally in cmake/Dependencies.cmake
 // starting from CUDA 11.4
-#if CUB_VERSION >= 101301
+#if CUB_VERSION >= 101400
 #define USE_GLOBAL_CUB_WRAPPED_NAMESPACE() true
 #else
 #define USE_GLOBAL_CUB_WRAPPED_NAMESPACE() false

From d1113be1236633978a51581ad9270069ddea5c30 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Tue, 12 Oct 2021 09:07:12 -0700
Subject: [PATCH 19/30] save

---
 aten/src/ATen/cuda/cub_definitions.cuh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/aten/src/ATen/cuda/cub_definitions.cuh b/aten/src/ATen/cuda/cub_definitions.cuh
index fb43441148fc..4b339b92c6dc 100644
--- a/aten/src/ATen/cuda/cub_definitions.cuh
+++ b/aten/src/ATen/cuda/cub_definitions.cuh
@@ -18,7 +18,7 @@
 // cub sort support for CUB_WRAPPED_NAMESPACE is added to cub 1.13.1 in:
 // https://github.com/NVIDIA/cub/pull/326
 // CUB_WRAPPED_NAMESPACE is defined globally in cmake/Dependencies.cmake
-// starting from CUDA 11.4
+// starting from CUDA 11.5
 #if CUB_VERSION >= 101400
 #define USE_GLOBAL_CUB_WRAPPED_NAMESPACE() true
 #else

From 59949c2c4945f7df89888bbbfb44060ea19e31ff Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Tue, 12 Oct 2021 11:45:18 -0700
Subject: [PATCH 20/30] save

---
 aten/src/ATen/cuda/cub.cuh                         | 8 ++++----
 aten/src/ATen/cuda/cub_definitions.cuh             | 8 ++++----
 caffe2/core/context_gpu.cu                         | 2 +-
 caffe2/operators/accuracy_op.cu                    | 2 +-
 caffe2/operators/affine_channel_op.cu              | 2 +-
 caffe2/operators/arg_ops.cu                        | 2 +-
 caffe2/operators/batch_moments_op.cu               | 2 +-
 caffe2/operators/batch_sparse_to_dense_op.cu       | 2 +-
 caffe2/operators/boolean_mask_ops.cu               | 4 ++--
 caffe2/operators/cross_entropy_op.cu               | 2 +-
 caffe2/operators/distance_op.cu                    | 2 +-
 caffe2/operators/elementwise_div_op.cu             | 2 +-
 caffe2/operators/elementwise_linear_op.cu          | 2 +-
 caffe2/operators/elementwise_mul_op.cu             | 2 +-
 caffe2/operators/elementwise_ops.cu                | 2 +-
 caffe2/operators/find_op.cu                        | 2 +-
 caffe2/operators/generate_proposals_op.cu          | 3 ++-
 caffe2/operators/normalize_ops.cu                  | 2 +-
 caffe2/operators/one_hot_ops.cu                    | 2 +-
 caffe2/operators/pack_segments.cu                  | 3 ++-
 caffe2/operators/prelu_op.cu                       | 2 +-
 caffe2/operators/reduce_front_back_max_ops.cu      | 2 +-
 caffe2/operators/reduce_front_back_sum_mean_ops.cu | 2 +-
 caffe2/operators/reduction_ops.cu                  | 2 +-
 caffe2/operators/rmac_regions_op.cu                | 2 +-
 caffe2/operators/segment_reduction_op_gpu.cuh      | 2 +-
 caffe2/operators/sequence_ops.cu                   | 3 ++-
 caffe2/operators/softmax_ops.cu                    | 2 +-
 caffe2/operators/spatial_batch_norm_op_impl.cuh    | 2 +-
 caffe2/sgd/adagrad_fused_op_gpu.cu                 | 2 +-
 caffe2/sgd/adagrad_op_gpu.cu                       | 2 +-
 caffe2/sgd/adam_op_gpu.cu                          | 2 +-
 caffe2/utils/cub_namespace.cuh                     | 7 +++++++
 caffe2/utils/math/reduce.cu                        | 3 +--
 caffe2/utils/math/reduce.cuh                       | 2 +-
 caffe2/utils/math_gpu.cu                           | 3 ++-
 cmake/Dependencies.cmake                           | 2 +-
 37 files changed, 54 insertions(+), 44 deletions(-)
 create mode 100644 caffe2/utils/cub_namespace.cuh

diff --git a/aten/src/ATen/cuda/cub.cuh b/aten/src/ATen/cuda/cub.cuh
index 17062586b0e3..de14455adb98 100644
--- a/aten/src/ATen/cuda/cub.cuh
+++ b/aten/src/ATen/cuda/cub.cuh
@@ -15,11 +15,14 @@
 
 #undef CUB_NS_POSTFIX //undef to avoid redefinition warnings
 #undef CUB_NS_PREFIX
+#undef CUB_NS_QUALIFIER
 #define CUB_NS_PREFIX namespace at_cuda_detail {
 #define CUB_NS_POSTFIX }
 #define CUB_NS_QUALIFIER ::at_cuda_detail::cub
 #include <cub/cub.cuh>
-
+#undef CUB_NS_POSTFIX
+#undef CUB_NS_PREFIX
+#undef CUB_NS_QUALIFIER
 #endif
 
 #include <ATen/cuda/Exceptions.h>
@@ -69,9 +72,6 @@ template <> struct cub::NumericTraits<c10::BFloat16>: cub::BaseTraits<cub::FLOAT
 namespace at { namespace native {
 namespace cub = ::at_cuda_detail::cub;
 }}
-namespace caffe2 {
-namespace cub = ::at_cuda_detail::cub;
-}
 #endif
 
 namespace at {
diff --git a/aten/src/ATen/cuda/cub_definitions.cuh b/aten/src/ATen/cuda/cub_definitions.cuh
index 4b339b92c6dc..07cffe9d34e4 100644
--- a/aten/src/ATen/cuda/cub_definitions.cuh
+++ b/aten/src/ATen/cuda/cub_definitions.cuh
@@ -15,12 +15,12 @@
 #define CUB_SUPPORTS_NV_BFLOAT16() false
 #endif
 
-// cub sort support for CUB_WRAPPED_NAMESPACE is added to cub 1.13.1 in:
+// cub sort support for CUB_WRAPPED_NAMESPACE is added to cub 1.14 in:
 // https://github.com/NVIDIA/cub/pull/326
 // CUB_WRAPPED_NAMESPACE is defined globally in cmake/Dependencies.cmake
-// starting from CUDA 11.5
-#if CUB_VERSION >= 101400
+// starting from CUDA 11.6
+#if defined(CUB_WRAPPED_NAMESPACE) || defined(THRUST_CUB_WRAPPED_NAMESPACE)
 #define USE_GLOBAL_CUB_WRAPPED_NAMESPACE() true
 #else
 #define USE_GLOBAL_CUB_WRAPPED_NAMESPACE() false
-#endif
+#endif
\ No newline at end of file
diff --git a/caffe2/core/context_gpu.cu b/caffe2/core/context_gpu.cu
index 475ed61ab4f7..9ba9f74d5376 100644
--- a/caffe2/core/context_gpu.cu
+++ b/caffe2/core/context_gpu.cu
@@ -4,7 +4,7 @@
 #include <string>
 #include <unordered_map>
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <ATen/Context.h>
 #include <c10/cuda/CUDAFunctions.h>
 #include <c10/cuda/CUDACachingAllocator.h>
diff --git a/caffe2/operators/accuracy_op.cu b/caffe2/operators/accuracy_op.cu
index 7ad2b09c238a..29df54e752d3 100644
--- a/caffe2/operators/accuracy_op.cu
+++ b/caffe2/operators/accuracy_op.cu
@@ -3,7 +3,7 @@
 #include "caffe2/utils/GpuAtomics.cuh"
 #include "caffe2/utils/math.h"
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 namespace caffe2 {
diff --git a/caffe2/operators/affine_channel_op.cu b/caffe2/operators/affine_channel_op.cu
index f3d9e22c7e8c..efae0a3fc695 100644
--- a/caffe2/operators/affine_channel_op.cu
+++ b/caffe2/operators/affine_channel_op.cu
@@ -1,6 +1,6 @@
 #include "caffe2/operators/affine_channel_op.h"
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/arg_ops.cu b/caffe2/operators/arg_ops.cu
index fbefe0774376..56deaa636356 100644
--- a/caffe2/operators/arg_ops.cu
+++ b/caffe2/operators/arg_ops.cu
@@ -2,7 +2,7 @@
 
 #include <limits>
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/common_gpu.h"
diff --git a/caffe2/operators/batch_moments_op.cu b/caffe2/operators/batch_moments_op.cu
index 65c43200e5bd..81359f6440b0 100644
--- a/caffe2/operators/batch_moments_op.cu
+++ b/caffe2/operators/batch_moments_op.cu
@@ -1,6 +1,6 @@
 #include "caffe2/operators/batch_moments_op.h"
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/batch_sparse_to_dense_op.cu b/caffe2/operators/batch_sparse_to_dense_op.cu
index 2cb09deb8668..3e7ad8af9a5b 100644
--- a/caffe2/operators/batch_sparse_to_dense_op.cu
+++ b/caffe2/operators/batch_sparse_to_dense_op.cu
@@ -1,6 +1,6 @@
 #include "caffe2/operators/batch_sparse_to_dense_op.h"
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/device/device_scan.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/boolean_mask_ops.cu b/caffe2/operators/boolean_mask_ops.cu
index c87688f51d64..501dd3b191c8 100644
--- a/caffe2/operators/boolean_mask_ops.cu
+++ b/caffe2/operators/boolean_mask_ops.cu
@@ -2,8 +2,8 @@
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/boolean_mask_ops.h"
-
-#include <ATen/cuda/cub.cuh>
+#include <cub/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/cross_entropy_op.cu b/caffe2/operators/cross_entropy_op.cu
index 95f3ffddbf1f..15cb8a4f574a 100644
--- a/caffe2/operators/cross_entropy_op.cu
+++ b/caffe2/operators/cross_entropy_op.cu
@@ -1,5 +1,5 @@
 #include <assert.h>
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/distance_op.cu b/caffe2/operators/distance_op.cu
index d94691d5a9d9..a360166854ff 100644
--- a/caffe2/operators/distance_op.cu
+++ b/caffe2/operators/distance_op.cu
@@ -4,7 +4,7 @@
 #include "caffe2/operators/distance_op.h"
 #include "caffe2/utils/conversions.h"
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 namespace caffe2 {
diff --git a/caffe2/operators/elementwise_div_op.cu b/caffe2/operators/elementwise_div_op.cu
index ca9682326324..33118a8f5e16 100644
--- a/caffe2/operators/elementwise_div_op.cu
+++ b/caffe2/operators/elementwise_div_op.cu
@@ -3,7 +3,7 @@
 #include <algorithm>
 #include <functional>
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/elementwise_linear_op.cu b/caffe2/operators/elementwise_linear_op.cu
index c1c45263f34c..8f749644b295 100644
--- a/caffe2/operators/elementwise_linear_op.cu
+++ b/caffe2/operators/elementwise_linear_op.cu
@@ -5,7 +5,7 @@
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/operator_fallback_gpu.h"
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 namespace caffe2 {
diff --git a/caffe2/operators/elementwise_mul_op.cu b/caffe2/operators/elementwise_mul_op.cu
index 88c3da00edc3..1991b8b513af 100644
--- a/caffe2/operators/elementwise_mul_op.cu
+++ b/caffe2/operators/elementwise_mul_op.cu
@@ -3,7 +3,7 @@
 #include <algorithm>
 #include <functional>
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/elementwise_ops.cu b/caffe2/operators/elementwise_ops.cu
index 1ac0426d2ca7..932bd5dafda0 100644
--- a/caffe2/operators/elementwise_ops.cu
+++ b/caffe2/operators/elementwise_ops.cu
@@ -1,6 +1,6 @@
 #include "caffe2/operators/elementwise_ops.h"
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_load.cuh>
 #include <cub/block/block_reduce.cuh>
 #include <cub/device/device_reduce.cuh>
diff --git a/caffe2/operators/find_op.cu b/caffe2/operators/find_op.cu
index 666df335ce42..20d42560b506 100644
--- a/caffe2/operators/find_op.cu
+++ b/caffe2/operators/find_op.cu
@@ -1,4 +1,4 @@
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/find_op.h"
diff --git a/caffe2/operators/generate_proposals_op.cu b/caffe2/operators/generate_proposals_op.cu
index 84906a8e8182..b63726651939 100644
--- a/caffe2/operators/generate_proposals_op.cu
+++ b/caffe2/operators/generate_proposals_op.cu
@@ -1,10 +1,11 @@
-#include <ATen/cuda/cub.cuh>
+#include <cub/cub.cuh>
 #include "caffe2/core/context.h"
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/generate_proposals_op.h"
 #include "caffe2/operators/generate_proposals_op_util_boxes.h" // BBOX_XFORM_CLIP_DEFAULT
 #include "caffe2/operators/generate_proposals_op_util_nms.h"
 #include "caffe2/operators/generate_proposals_op_util_nms_gpu.h"
+#include "caffe2/utils/cub_namespace.cuh"
 
 #if defined(USE_ROCM)
 #include <cfloat>
diff --git a/caffe2/operators/normalize_ops.cu b/caffe2/operators/normalize_ops.cu
index 468175df985f..952c4a772fa5 100644
--- a/caffe2/operators/normalize_ops.cu
+++ b/caffe2/operators/normalize_ops.cu
@@ -1,6 +1,6 @@
 #include <algorithm>
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/one_hot_ops.cu b/caffe2/operators/one_hot_ops.cu
index 86f82f78bb82..4b1e054b0806 100644
--- a/caffe2/operators/one_hot_ops.cu
+++ b/caffe2/operators/one_hot_ops.cu
@@ -1,4 +1,4 @@
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/pack_segments.cu b/caffe2/operators/pack_segments.cu
index b9ed413d1e7b..372638abdd24 100644
--- a/caffe2/operators/pack_segments.cu
+++ b/caffe2/operators/pack_segments.cu
@@ -1,6 +1,7 @@
-#include <ATen/cuda/cub.cuh>
+#include <cub/cub.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/pack_segments.h"
+#include "caffe2/utils/cub_namespace.cuh"
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/prelu_op.cu b/caffe2/operators/prelu_op.cu
index d29882086754..6303b70b4a89 100644
--- a/caffe2/operators/prelu_op.cu
+++ b/caffe2/operators/prelu_op.cu
@@ -1,7 +1,7 @@
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/prelu_op.h"
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 namespace caffe2 {
diff --git a/caffe2/operators/reduce_front_back_max_ops.cu b/caffe2/operators/reduce_front_back_max_ops.cu
index ba62b2eff671..d6bb862e4dbb 100644
--- a/caffe2/operators/reduce_front_back_max_ops.cu
+++ b/caffe2/operators/reduce_front_back_max_ops.cu
@@ -1,4 +1,4 @@
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/reduce_front_back_max_ops.h"
diff --git a/caffe2/operators/reduce_front_back_sum_mean_ops.cu b/caffe2/operators/reduce_front_back_sum_mean_ops.cu
index 586c20fe8d8e..2b5cb7110edf 100644
--- a/caffe2/operators/reduce_front_back_sum_mean_ops.cu
+++ b/caffe2/operators/reduce_front_back_sum_mean_ops.cu
@@ -1,4 +1,4 @@
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/reduce_front_back_sum_mean_ops.h"
diff --git a/caffe2/operators/reduction_ops.cu b/caffe2/operators/reduction_ops.cu
index 0d94fab22a7f..9649b85d015c 100644
--- a/caffe2/operators/reduction_ops.cu
+++ b/caffe2/operators/reduction_ops.cu
@@ -2,7 +2,7 @@
 #include "caffe2/operators/reduction_ops.h"
 #include "caffe2/utils/conversions.h"
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/rmac_regions_op.cu b/caffe2/operators/rmac_regions_op.cu
index 2f6c230574b0..6b79e0c4c8b4 100644
--- a/caffe2/operators/rmac_regions_op.cu
+++ b/caffe2/operators/rmac_regions_op.cu
@@ -1,5 +1,5 @@
-#include <ATen/cuda/cub.cuh>
 #include <cub/block/block_reduce.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/rmac_regions_op.h"
diff --git a/caffe2/operators/segment_reduction_op_gpu.cuh b/caffe2/operators/segment_reduction_op_gpu.cuh
index 447617c6e9de..bb3f3be13c72 100644
--- a/caffe2/operators/segment_reduction_op_gpu.cuh
+++ b/caffe2/operators/segment_reduction_op_gpu.cuh
@@ -1,4 +1,4 @@
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 #include <cub/device/device_reduce.cuh>
 #include <cub/device/device_scan.cuh>
diff --git a/caffe2/operators/sequence_ops.cu b/caffe2/operators/sequence_ops.cu
index e66d491f85e6..2ceb5236ef72 100644
--- a/caffe2/operators/sequence_ops.cu
+++ b/caffe2/operators/sequence_ops.cu
@@ -1,6 +1,7 @@
 #include <algorithm>
 
-#include <ATen/cuda/cub.cuh>
+#include <cub/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/sequence_ops.h"
diff --git a/caffe2/operators/softmax_ops.cu b/caffe2/operators/softmax_ops.cu
index c01fcf3e0a48..b0afac3332a6 100644
--- a/caffe2/operators/softmax_ops.cu
+++ b/caffe2/operators/softmax_ops.cu
@@ -1,5 +1,5 @@
 #include <cfloat>
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/operators/spatial_batch_norm_op_impl.cuh b/caffe2/operators/spatial_batch_norm_op_impl.cuh
index f9b9fb58adc8..6fdb4c63f8ef 100644
--- a/caffe2/operators/spatial_batch_norm_op_impl.cuh
+++ b/caffe2/operators/spatial_batch_norm_op_impl.cuh
@@ -5,7 +5,7 @@
 
 #include <limits>
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/sgd/adagrad_fused_op_gpu.cu b/caffe2/sgd/adagrad_fused_op_gpu.cu
index 396da5195125..63d0712e3970 100644
--- a/caffe2/sgd/adagrad_fused_op_gpu.cu
+++ b/caffe2/sgd/adagrad_fused_op_gpu.cu
@@ -2,7 +2,7 @@
 #include <c10/core/GeneratorImpl.h>
 #include <algorithm>
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/device/device_radix_sort.cuh>
 #include "caffe2/sgd/adagrad_fused_op_gpu.cuh"
 #include "caffe2/utils/math.h"
diff --git a/caffe2/sgd/adagrad_op_gpu.cu b/caffe2/sgd/adagrad_op_gpu.cu
index a6fa842ddc80..0b7f499345be 100644
--- a/caffe2/sgd/adagrad_op_gpu.cu
+++ b/caffe2/sgd/adagrad_op_gpu.cu
@@ -1,6 +1,6 @@
 #include <algorithm>
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/common_gpu.h"
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/sgd/adam_op_gpu.cu b/caffe2/sgd/adam_op_gpu.cu
index 4b59836b6a68..a93812fabbe8 100644
--- a/caffe2/sgd/adam_op_gpu.cu
+++ b/caffe2/sgd/adam_op_gpu.cu
@@ -1,4 +1,4 @@
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/common_gpu.h"
 #include "caffe2/core/context_gpu.h"
diff --git a/caffe2/utils/cub_namespace.cuh b/caffe2/utils/cub_namespace.cuh
new file mode 100644
index 000000000000..c7a5db0dc013
--- /dev/null
+++ b/caffe2/utils/cub_namespace.cuh
@@ -0,0 +1,7 @@
+#include <ATen/cuda/cub_definitions.cuh>
+
+#if USE_GLOBAL_CUB_WRAPPED_NAMESPACE()
+namespace caffe2 {
+namespace cub = ::CUB_WRAPPED_NAMESPACE::cub;
+}
+#endif
diff --git a/caffe2/utils/math/reduce.cu b/caffe2/utils/math/reduce.cu
index 20919334da50..69a6469d8ed1 100644
--- a/caffe2/utils/math/reduce.cu
+++ b/caffe2/utils/math/reduce.cu
@@ -5,8 +5,7 @@
 #include <limits>
 #include <numeric>
 #include <vector>
-
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 #include <thrust/execution_policy.h>
diff --git a/caffe2/utils/math/reduce.cuh b/caffe2/utils/math/reduce.cuh
index 39ad553eec76..18bdca11b9de 100644
--- a/caffe2/utils/math/reduce.cuh
+++ b/caffe2/utils/math/reduce.cuh
@@ -1,7 +1,7 @@
 #ifndef CAFFE2_UTILS_MATH_REDUCE_CUH_
 #define CAFFE2_UTILS_MATH_REDUCE_CUH_
 
-#include <ATen/cuda/cub.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/common_gpu.h"
diff --git a/caffe2/utils/math_gpu.cu b/caffe2/utils/math_gpu.cu
index b0a44fed34fb..54b0a9391c26 100644
--- a/caffe2/utils/math_gpu.cu
+++ b/caffe2/utils/math_gpu.cu
@@ -7,8 +7,9 @@
 #include <numeric>
 #include <vector>
 
-#include <ATen/cuda/cub.cuh>
+#include <cub/cub.cuh>
 #include <cub/block/block_reduce.cuh>
+#include "caffe2/utils/cub_namespace.cuh"
 
 #include <thrust/host_vector.h>
 #include <thrust/device_vector.h>
diff --git a/cmake/Dependencies.cmake b/cmake/Dependencies.cmake
index bfee8f6dc78a..a6a3946bac8e 100644
--- a/cmake/Dependencies.cmake
+++ b/cmake/Dependencies.cmake
@@ -1618,7 +1618,7 @@ if(NOT INTERN_BUILD_MOBILE)
     set(CMAKE_CXX_STANDARD 14)
   endif()
 
-  if(NOT ${CUDA_VERSION} LESS 11.4)
+  if(NOT ${CUDA_VERSION} LESS 11.6)
     string(APPEND CMAKE_CUDA_FLAGS " -DCUB_WRAPPED_NAMESPACE=at_cuda_detail")
   endif()
 

From c22df90d62acfcb572b4b7bf55646fe338843a69 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Tue, 12 Oct 2021 11:47:52 -0700
Subject: [PATCH 21/30] save

---
 aten/src/ATen/cuda/cub.cuh             | 1 +
 aten/src/ATen/cuda/cub_definitions.cuh | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)

diff --git a/aten/src/ATen/cuda/cub.cuh b/aten/src/ATen/cuda/cub.cuh
index de14455adb98..fecf9e077a6f 100644
--- a/aten/src/ATen/cuda/cub.cuh
+++ b/aten/src/ATen/cuda/cub.cuh
@@ -23,6 +23,7 @@
 #undef CUB_NS_POSTFIX
 #undef CUB_NS_PREFIX
 #undef CUB_NS_QUALIFIER
+
 #endif
 
 #include <ATen/cuda/Exceptions.h>
diff --git a/aten/src/ATen/cuda/cub_definitions.cuh b/aten/src/ATen/cuda/cub_definitions.cuh
index 07cffe9d34e4..d639d28e66b6 100644
--- a/aten/src/ATen/cuda/cub_definitions.cuh
+++ b/aten/src/ATen/cuda/cub_definitions.cuh
@@ -23,4 +23,4 @@
 #define USE_GLOBAL_CUB_WRAPPED_NAMESPACE() true
 #else
 #define USE_GLOBAL_CUB_WRAPPED_NAMESPACE() false
-#endif
\ No newline at end of file
+#endif

From 64adfaaaca8267cd6cf03ae7a524eee239530425 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Tue, 12 Oct 2021 12:29:41 -0700
Subject: [PATCH 22/30] save

---
 caffe2/core/context_gpu.cu                         | 2 +-
 caffe2/operators/cross_entropy_op.cu               | 2 +-
 caffe2/operators/find_op.cu                        | 2 +-
 caffe2/operators/normalize_ops.cu                  | 2 +-
 caffe2/operators/one_hot_ops.cu                    | 2 +-
 caffe2/operators/reduce_front_back_max_ops.cu      | 2 +-
 caffe2/operators/reduce_front_back_sum_mean_ops.cu | 2 +-
 caffe2/operators/softmax_ops.cu                    | 2 +-
 caffe2/sgd/adagrad_op_gpu.cu                       | 2 +-
 caffe2/sgd/adam_op_gpu.cu                          | 2 +-
 cmake/Dependencies.cmake                           | 2 +-
 11 files changed, 11 insertions(+), 11 deletions(-)

diff --git a/caffe2/core/context_gpu.cu b/caffe2/core/context_gpu.cu
index 9ba9f74d5376..6d537400913e 100644
--- a/caffe2/core/context_gpu.cu
+++ b/caffe2/core/context_gpu.cu
@@ -4,7 +4,6 @@
 #include <string>
 #include <unordered_map>
 
-#include "caffe2/utils/cub_namespace.cuh"
 #include <ATen/Context.h>
 #include <c10/cuda/CUDAFunctions.h>
 #include <c10/cuda/CUDACachingAllocator.h>
@@ -22,6 +21,7 @@
 #include "caffe2/core/logging.h"
 #include "caffe2/core/tensor.h"
 #include "caffe2/utils/string_utils.h"
+#include "caffe2/utils/cub_namespace.cuh"
 
 C10_DEFINE_string(
     caffe2_cuda_memory_pool,
diff --git a/caffe2/operators/cross_entropy_op.cu b/caffe2/operators/cross_entropy_op.cu
index 15cb8a4f574a..c23f05f8e5c2 100644
--- a/caffe2/operators/cross_entropy_op.cu
+++ b/caffe2/operators/cross_entropy_op.cu
@@ -1,10 +1,10 @@
 #include <assert.h>
-#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/cross_entropy_op.h"
 #include "caffe2/operators/operator_fallback_gpu.h"
+#include "caffe2/utils/cub_namespace.cuh"
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/find_op.cu b/caffe2/operators/find_op.cu
index 20d42560b506..0418a71fbcda 100644
--- a/caffe2/operators/find_op.cu
+++ b/caffe2/operators/find_op.cu
@@ -1,7 +1,7 @@
-#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/find_op.h"
+#include "caffe2/utils/cub_namespace.cuh"
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/normalize_ops.cu b/caffe2/operators/normalize_ops.cu
index 952c4a772fa5..e4d1f34b754c 100644
--- a/caffe2/operators/normalize_ops.cu
+++ b/caffe2/operators/normalize_ops.cu
@@ -1,11 +1,11 @@
 #include <algorithm>
 
-#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/normalize_l1_op.h"
 #include "caffe2/operators/normalize_op.h"
+#include "caffe2/utils/cub_namespace.cuh"
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/one_hot_ops.cu b/caffe2/operators/one_hot_ops.cu
index 4b1e054b0806..87e8196765ef 100644
--- a/caffe2/operators/one_hot_ops.cu
+++ b/caffe2/operators/one_hot_ops.cu
@@ -1,8 +1,8 @@
-#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/one_hot_ops.h"
+#include "caffe2/utils/cub_namespace.cuh"
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/reduce_front_back_max_ops.cu b/caffe2/operators/reduce_front_back_max_ops.cu
index d6bb862e4dbb..2ea25de46009 100644
--- a/caffe2/operators/reduce_front_back_max_ops.cu
+++ b/caffe2/operators/reduce_front_back_max_ops.cu
@@ -1,7 +1,7 @@
-#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/reduce_front_back_max_ops.h"
+#include "caffe2/utils/cub_namespace.cuh"
 
 #if defined(USE_ROCM)
 #include <cfloat>
diff --git a/caffe2/operators/reduce_front_back_sum_mean_ops.cu b/caffe2/operators/reduce_front_back_sum_mean_ops.cu
index 2b5cb7110edf..a7ad6dd50084 100644
--- a/caffe2/operators/reduce_front_back_sum_mean_ops.cu
+++ b/caffe2/operators/reduce_front_back_sum_mean_ops.cu
@@ -1,7 +1,7 @@
-#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/reduce_front_back_sum_mean_ops.h"
+#include "caffe2/utils/cub_namespace.cuh"
 
 namespace caffe2 {
 
diff --git a/caffe2/operators/softmax_ops.cu b/caffe2/operators/softmax_ops.cu
index b0afac3332a6..ebf0700c9ef0 100644
--- a/caffe2/operators/softmax_ops.cu
+++ b/caffe2/operators/softmax_ops.cu
@@ -1,11 +1,11 @@
 #include <cfloat>
-#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/operators/softmax_op.h"
 #include "caffe2/operators/softmax_with_loss_op.h"
 #include "caffe2/operators/spatial_softmax_with_loss_op.h"
+#include "caffe2/utils/cub_namespace.cuh"
 
 namespace caffe2 {
 
diff --git a/caffe2/sgd/adagrad_op_gpu.cu b/caffe2/sgd/adagrad_op_gpu.cu
index 0b7f499345be..b80d29700c3f 100644
--- a/caffe2/sgd/adagrad_op_gpu.cu
+++ b/caffe2/sgd/adagrad_op_gpu.cu
@@ -1,10 +1,10 @@
 #include <algorithm>
 
-#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/common_gpu.h"
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/sgd/adagrad_op.h"
+#include "caffe2/utils/cub_namespace.cuh"
 
 namespace caffe2 {
 
diff --git a/caffe2/sgd/adam_op_gpu.cu b/caffe2/sgd/adam_op_gpu.cu
index a93812fabbe8..6f9c3234204d 100644
--- a/caffe2/sgd/adam_op_gpu.cu
+++ b/caffe2/sgd/adam_op_gpu.cu
@@ -1,8 +1,8 @@
-#include "caffe2/utils/cub_namespace.cuh"
 #include <cub/block/block_reduce.cuh>
 #include "caffe2/core/common_gpu.h"
 #include "caffe2/core/context_gpu.h"
 #include "caffe2/sgd/adam_op.h"
+#include "caffe2/utils/cub_namespace.cuh"
 
 namespace caffe2 {
 
diff --git a/cmake/Dependencies.cmake b/cmake/Dependencies.cmake
index a6a3946bac8e..5074bab1e999 100644
--- a/cmake/Dependencies.cmake
+++ b/cmake/Dependencies.cmake
@@ -1618,7 +1618,7 @@ if(NOT INTERN_BUILD_MOBILE)
     set(CMAKE_CXX_STANDARD 14)
   endif()
 
-  if(NOT ${CUDA_VERSION} LESS 11.6)
+  if(NOT ${CUDA_VERSION} LESS 11.5)
     string(APPEND CMAKE_CUDA_FLAGS " -DCUB_WRAPPED_NAMESPACE=at_cuda_detail")
   endif()
 

From 21abede594471e4f225def12796c1d7c73a96b20 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Tue, 12 Oct 2021 12:46:32 -0700
Subject: [PATCH 23/30] save

---
 aten/src/ATen/cuda/cub_definitions.cuh | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/aten/src/ATen/cuda/cub_definitions.cuh b/aten/src/ATen/cuda/cub_definitions.cuh
index d639d28e66b6..9828abdfc99a 100644
--- a/aten/src/ATen/cuda/cub_definitions.cuh
+++ b/aten/src/ATen/cuda/cub_definitions.cuh
@@ -15,10 +15,10 @@
 #define CUB_SUPPORTS_NV_BFLOAT16() false
 #endif
 
-// cub sort support for CUB_WRAPPED_NAMESPACE is added to cub 1.14 in:
+// cub sort support for CUB_WRAPPED_NAMESPACE is added to cub 1.13.1 in:
 // https://github.com/NVIDIA/cub/pull/326
 // CUB_WRAPPED_NAMESPACE is defined globally in cmake/Dependencies.cmake
-// starting from CUDA 11.6
+// starting from CUDA 11.5
 #if defined(CUB_WRAPPED_NAMESPACE) || defined(THRUST_CUB_WRAPPED_NAMESPACE)
 #define USE_GLOBAL_CUB_WRAPPED_NAMESPACE() true
 #else

From c73b101377d5adb8f96eb31988ceef7c2e760839 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Tue, 12 Oct 2021 12:59:12 -0700
Subject: [PATCH 24/30] comment

---
 aten/src/ATen/cuda/cub.cuh | 2 ++
 cmake/Dependencies.cmake   | 2 ++
 2 files changed, 4 insertions(+)

diff --git a/aten/src/ATen/cuda/cub.cuh b/aten/src/ATen/cuda/cub.cuh
index fecf9e077a6f..0532470d74ec 100644
--- a/aten/src/ATen/cuda/cub.cuh
+++ b/aten/src/ATen/cuda/cub.cuh
@@ -13,6 +13,8 @@
 
 #else
 
+// include cub in a safe manner, see:
+// https://github.com/pytorch/pytorch/pull/55292
 #undef CUB_NS_POSTFIX //undef to avoid redefinition warnings
 #undef CUB_NS_PREFIX
 #undef CUB_NS_QUALIFIER
diff --git a/cmake/Dependencies.cmake b/cmake/Dependencies.cmake
index 5074bab1e999..bde42bf1719f 100644
--- a/cmake/Dependencies.cmake
+++ b/cmake/Dependencies.cmake
@@ -1618,6 +1618,8 @@ if(NOT INTERN_BUILD_MOBILE)
     set(CMAKE_CXX_STANDARD 14)
   endif()
 
+  # use cub in a safe manner, see:
+  # https://github.com/pytorch/pytorch/pull/55292
   if(NOT ${CUDA_VERSION} LESS 11.5)
     string(APPEND CMAKE_CUDA_FLAGS " -DCUB_WRAPPED_NAMESPACE=at_cuda_detail")
   endif()

From 1cb29c9816e795767236854644a1ba2c28da6e0d Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Fri, 22 Oct 2021 10:42:32 -0700
Subject: [PATCH 25/30] fix

---
 caffe2/utils/cub_namespace.cuh | 19 ++++++++++++++++++-
 1 file changed, 18 insertions(+), 1 deletion(-)

diff --git a/caffe2/utils/cub_namespace.cuh b/caffe2/utils/cub_namespace.cuh
index c7a5db0dc013..752f273128ee 100644
--- a/caffe2/utils/cub_namespace.cuh
+++ b/caffe2/utils/cub_namespace.cuh
@@ -1,4 +1,21 @@
-#include <ATen/cuda/cub_definitions.cuh>
+#pragma once
+
+#if !defined(USE_ROCM)
+#include <cuda.h>
+#include <cub/version.cuh>
+#else
+#define CUB_VERSION 0
+#endif
+
+// cub sort support for CUB_WRAPPED_NAMESPACE is added to cub 1.13.1 in:
+// https://github.com/NVIDIA/cub/pull/326
+// CUB_WRAPPED_NAMESPACE is defined globally in cmake/Dependencies.cmake
+// starting from CUDA 11.5
+#if defined(CUB_WRAPPED_NAMESPACE) || defined(THRUST_CUB_WRAPPED_NAMESPACE)
+#define USE_GLOBAL_CUB_WRAPPED_NAMESPACE() true
+#else
+#define USE_GLOBAL_CUB_WRAPPED_NAMESPACE() false
+#endif
 
 #if USE_GLOBAL_CUB_WRAPPED_NAMESPACE()
 namespace caffe2 {

From 3ca182d3c6d1dcc45396c3f2f3169cc84496ca1f Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Sun, 24 Oct 2021 14:27:09 -0700
Subject: [PATCH 26/30] fix some cuda 10 build

---
 aten/src/ATen/cuda/cub_definitions.cuh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/aten/src/ATen/cuda/cub_definitions.cuh b/aten/src/ATen/cuda/cub_definitions.cuh
index 9828abdfc99a..5243e551f4a7 100644
--- a/aten/src/ATen/cuda/cub_definitions.cuh
+++ b/aten/src/ATen/cuda/cub_definitions.cuh
@@ -1,6 +1,6 @@
 #pragma once
 
-#if !defined(USE_ROCM)
+#if defined(CUDA_VERSION) && CUDA_VERSION >= 11000
 #include <cuda.h>
 #include <cub/version.cuh>
 #else

From 3a7c2bc6da6f2be060be879d0b2277b833ab7cc9 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Sun, 24 Oct 2021 14:28:53 -0700
Subject: [PATCH 27/30] fix caffe2

---
 caffe2/utils/cub_namespace.cuh | 7 -------
 1 file changed, 7 deletions(-)

diff --git a/caffe2/utils/cub_namespace.cuh b/caffe2/utils/cub_namespace.cuh
index 752f273128ee..188a9936f9c6 100644
--- a/caffe2/utils/cub_namespace.cuh
+++ b/caffe2/utils/cub_namespace.cuh
@@ -1,12 +1,5 @@
 #pragma once
 
-#if !defined(USE_ROCM)
-#include <cuda.h>
-#include <cub/version.cuh>
-#else
-#define CUB_VERSION 0
-#endif
-
 // cub sort support for CUB_WRAPPED_NAMESPACE is added to cub 1.13.1 in:
 // https://github.com/NVIDIA/cub/pull/326
 // CUB_WRAPPED_NAMESPACE is defined globally in cmake/Dependencies.cmake

From 960606e0dc1da7c16b6cb0ae11b04bace95816ac Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Sun, 24 Oct 2021 14:50:13 -0700
Subject: [PATCH 28/30] fix

---
 aten/src/ATen/cuda/cub_definitions.cuh | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/aten/src/ATen/cuda/cub_definitions.cuh b/aten/src/ATen/cuda/cub_definitions.cuh
index 5243e551f4a7..b921af480655 100644
--- a/aten/src/ATen/cuda/cub_definitions.cuh
+++ b/aten/src/ATen/cuda/cub_definitions.cuh
@@ -1,7 +1,10 @@
 #pragma once
 
-#if defined(CUDA_VERSION) && CUDA_VERSION >= 11000
+#if !define(USE_ROCM)
 #include <cuda.h>
+#endif
+
+#if defined(CUDA_VERSION) && CUDA_VERSION >= 11000
 #include <cub/version.cuh>
 #else
 #define CUB_VERSION 0

From 48f128f55396622bfb952b474888813334fae303 Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Sun, 24 Oct 2021 14:51:12 -0700
Subject: [PATCH 29/30] save

---
 aten/src/ATen/cuda/cub_definitions.cuh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/aten/src/ATen/cuda/cub_definitions.cuh b/aten/src/ATen/cuda/cub_definitions.cuh
index b921af480655..8c99f4951c6c 100644
--- a/aten/src/ATen/cuda/cub_definitions.cuh
+++ b/aten/src/ATen/cuda/cub_definitions.cuh
@@ -1,7 +1,7 @@
 #pragma once
 
 #if !define(USE_ROCM)
-#include <cuda.h>
+#include <cuda.h>  // for CUDA_VERSION
 #endif
 
 #if defined(CUDA_VERSION) && CUDA_VERSION >= 11000

From b2023a036afe9ddff4f710953ad6ba55f8396ecf Mon Sep 17 00:00:00 2001
From: Xiang Gao <qasdfgtyuiop@gmail.com>
Date: Sun, 24 Oct 2021 14:51:59 -0700
Subject: [PATCH 30/30] fixed

---
 aten/src/ATen/cuda/cub_definitions.cuh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/aten/src/ATen/cuda/cub_definitions.cuh b/aten/src/ATen/cuda/cub_definitions.cuh
index 8c99f4951c6c..61119fc17458 100644
--- a/aten/src/ATen/cuda/cub_definitions.cuh
+++ b/aten/src/ATen/cuda/cub_definitions.cuh
@@ -1,6 +1,6 @@
 #pragma once
 
-#if !define(USE_ROCM)
+#if !defined(USE_ROCM)
 #include <cuda.h>  // for CUDA_VERSION
 #endif