aboutsummaryrefslogtreecommitdiffstats
path: root/lib/Transforms/Instrumentation/MemorySanitizer.cpp
blob: c2aa1e2f772ec7462ac643110bc8065d798905c4 (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
//===-- MemorySanitizer.cpp - detector of uninitialized reads -------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
/// \file
/// This file is a part of MemorySanitizer, a detector of uninitialized
/// reads.
///
/// The algorithm of the tool is similar to Memcheck
/// (http://goo.gl/QKbem). We associate a few shadow bits with every
/// byte of the application memory, poison the shadow of the malloc-ed
/// or alloca-ed memory, load the shadow bits on every memory read,
/// propagate the shadow bits through some of the arithmetic
/// instruction (including MOV), store the shadow bits on every memory
/// write, report a bug on some other instructions (e.g. JMP) if the
/// associated shadow is poisoned.
///
/// But there are differences too. The first and the major one:
/// compiler instrumentation instead of binary instrumentation. This
/// gives us much better register allocation, possible compiler
/// optimizations and a fast start-up. But this brings the major issue
/// as well: msan needs to see all program events, including system
/// calls and reads/writes in system libraries, so we either need to
/// compile *everything* with msan or use a binary translation
/// component (e.g. DynamoRIO) to instrument pre-built libraries.
/// Another difference from Memcheck is that we use 8 shadow bits per
/// byte of application memory and use a direct shadow mapping. This
/// greatly simplifies the instrumentation code and avoids races on
/// shadow updates (Memcheck is single-threaded so races are not a
/// concern there. Memcheck uses 2 shadow bits per byte with a slow
/// path storage that uses 8 bits per byte).
///
/// The default value of shadow is 0, which means "clean" (not poisoned).
///
/// Every module initializer should call __msan_init to ensure that the
/// shadow memory is ready. On error, __msan_warning is called. Since
/// parameters and return values may be passed via registers, we have a
/// specialized thread-local shadow for return values
/// (__msan_retval_tls) and parameters (__msan_param_tls).
///
///                           Origin tracking.
///
/// MemorySanitizer can track origins (allocation points) of all uninitialized
/// values. This behavior is controlled with a flag (msan-track-origins) and is
/// disabled by default.
///
/// Origins are 4-byte values created and interpreted by the runtime library.
/// They are stored in a second shadow mapping, one 4-byte value for 4 bytes
/// of application memory. Propagation of origins is basically a bunch of
/// "select" instructions that pick the origin of a dirty argument, if an
/// instruction has one.
///
/// Every 4 aligned, consecutive bytes of application memory have one origin
/// value associated with them. If these bytes contain uninitialized data
/// coming from 2 different allocations, the last store wins. Because of this,
/// MemorySanitizer reports can show unrelated origins, but this is unlikely in
/// practice.
///
/// Origins are meaningless for fully initialized values, so MemorySanitizer
/// avoids storing origin to memory when a fully initialized value is stored.
/// This way it avoids needless overwritting origin of the 4-byte region on
/// a short (i.e. 1 byte) clean store, and it is also good for performance.
///
///                            Atomic handling.
///
/// Ideally, every atomic store of application value should update the
/// corresponding shadow location in an atomic way. Unfortunately, atomic store
/// of two disjoint locations can not be done without severe slowdown.
///
/// Therefore, we implement an approximation that may err on the safe side.
/// In this implementation, every atomically accessed location in the program
/// may only change from (partially) uninitialized to fully initialized, but
/// not the other way around. We load the shadow _after_ the application load,
/// and we store the shadow _before_ the app store. Also, we always store clean
/// shadow (if the application store is atomic). This way, if the store-load
/// pair constitutes a happens-before arc, shadow store and load are correctly
/// ordered such that the load will get either the value that was stored, or
/// some later value (which is always clean).
///
/// This does not work very well with Compare-And-Swap (CAS) and
/// Read-Modify-Write (RMW) operations. To follow the above logic, CAS and RMW
/// must store the new shadow before the app operation, and load the shadow
/// after the app operation. Computers don't work this way. Current
/// implementation ignores the load aspect of CAS/RMW, always returning a clean
/// value. It implements the store part as a simple atomic store by storing a
/// clean shadow.

//===----------------------------------------------------------------------===//

#include "llvm/Transforms/Instrumentation.h"
#include "llvm/ADT/DepthFirstIterator.h"
#include "llvm/ADT/SmallString.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/ADT/Triple.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/InlineAsm.h"
#include "llvm/IR/InstVisitor.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/MDBuilder.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/Type.h"
#include "llvm/IR/ValueMap.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Transforms/Utils/ModuleUtils.h"

using namespace llvm;

#define DEBUG_TYPE "msan"

static const unsigned kOriginSize = 4;
static const unsigned kMinOriginAlignment = 4;
static const unsigned kShadowTLSAlignment = 8;

// These constants must be kept in sync with the ones in msan.h.
static const unsigned kParamTLSSize = 800;
static const unsigned kRetvalTLSSize = 800;

// Accesses sizes are powers of two: 1, 2, 4, 8.
static const size_t kNumberOfAccessSizes = 4;

/// \brief Track origins of uninitialized values.
///
/// Adds a section to MemorySanitizer report that points to the allocation
/// (stack or heap) the uninitialized bits came from originally.
static cl::opt<int> ClTrackOrigins("msan-track-origins",
       cl::desc("Track origins (allocation sites) of poisoned memory"),
       cl::Hidden, cl::init(0));
static cl::opt<bool> ClKeepGoing("msan-keep-going",
       cl::desc("keep going after reporting a UMR"),
       cl::Hidden, cl::init(false));
static cl::opt<bool> ClPoisonStack("msan-poison-stack",
       cl::desc("poison uninitialized stack variables"),
       cl::Hidden, cl::init(true));
static cl::opt<bool> ClPoisonStackWithCall("msan-poison-stack-with-call",
       cl::desc("poison uninitialized stack variables with a call"),
       cl::Hidden, cl::init(false));
static cl::opt<int> ClPoisonStackPattern("msan-poison-stack-pattern",
       cl::desc("poison uninitialized stack variables with the given patter"),
       cl::Hidden, cl::init(0xff));
static cl::opt<bool> ClPoisonUndef("msan-poison-undef",
       cl::desc("poison undef temps"),
       cl::Hidden, cl::init(true));

static cl::opt<bool> ClHandleICmp("msan-handle-icmp",
       cl::desc("propagate shadow through ICmpEQ and ICmpNE"),
       cl::Hidden, cl::init(true));

static cl::opt<bool> ClHandleICmpExact("msan-handle-icmp-exact",
       cl::desc("exact handling of relational integer ICmp"),
       cl::Hidden, cl::init(false));

// This flag controls whether we check the shadow of the address
// operand of load or store. Such bugs are very rare, since load from
// a garbage address typically results in SEGV, but still happen
// (e.g. only lower bits of address are garbage, or the access happens
// early at program startup where malloc-ed memory is more likely to
// be zeroed. As of 2012-08-28 this flag adds 20% slowdown.
static cl::opt<bool> ClCheckAccessAddress("msan-check-access-address",
       cl::desc("report accesses through a pointer which has poisoned shadow"),
       cl::Hidden, cl::init(true));

static cl::opt<bool> ClDumpStrictInstructions("msan-dump-strict-instructions",
       cl::desc("print out instructions with default strict semantics"),
       cl::Hidden, cl::init(false));

static cl::opt<int> ClInstrumentationWithCallThreshold(
    "msan-instrumentation-with-call-threshold",
    cl::desc(
        "If the function being instrumented requires more than "
        "this number of checks and origin stores, use callbacks instead of "
        "inline checks (-1 means never use callbacks)."),
    cl::Hidden, cl::init(3500));

// This is an experiment to enable handling of cases where shadow is a non-zero
// compile-time constant. For some unexplainable reason they were silently
// ignored in the instrumentation.
static cl::opt<bool> ClCheckConstantShadow("msan-check-constant-shadow",
       cl::desc("Insert checks for constant shadow values"),
       cl::Hidden, cl::init(false));

namespace {

// Memory map parameters used in application-to-shadow address calculation.
// Offset = (Addr & ~AndMask) ^ XorMask
// Shadow = ShadowBase + Offset
// Origin = OriginBase + Offset
struct MemoryMapParams {
  uint64_t AndMask;
  uint64_t XorMask;
  uint64_t ShadowBase;
  uint64_t OriginBase;
};

struct PlatformMemoryMapParams {
  const MemoryMapParams *bits32;
  const MemoryMapParams *bits64;
};

// i386 Linux
static const MemoryMapParams Linux_I386_MemoryMapParams = {
  0x000080000000,  // AndMask
  0,               // XorMask (not used)
  0,               // ShadowBase (not used)
  0x000040000000,  // OriginBase
};

// x86_64 Linux
static const MemoryMapParams Linux_X86_64_MemoryMapParams = {
  0x400000000000,  // AndMask
  0,               // XorMask (not used)
  0,               // ShadowBase (not used)
  0x200000000000,  // OriginBase
};

// mips64 Linux
static const MemoryMapParams Linux_MIPS64_MemoryMapParams = {
  0x004000000000,  // AndMask
  0,               // XorMask (not used)
  0,               // ShadowBase (not used)
  0x002000000000,  // OriginBase
};

// i386 FreeBSD
static const MemoryMapParams FreeBSD_I386_MemoryMapParams = {
  0x000180000000,  // AndMask
  0x000040000000,  // XorMask
  0x000020000000,  // ShadowBase
  0x000700000000,  // OriginBase
};

// x86_64 FreeBSD
static const MemoryMapParams FreeBSD_X86_64_MemoryMapParams = {
  0xc00000000000,  // AndMask
  0x200000000000,  // XorMask
  0x100000000000,  // ShadowBase
  0x380000000000,  // OriginBase
};

static const PlatformMemoryMapParams Linux_X86_MemoryMapParams = {
  &Linux_I386_MemoryMapParams,
  &Linux_X86_64_MemoryMapParams,
};

static const PlatformMemoryMapParams Linux_MIPS_MemoryMapParams = {
  NULL,
  &Linux_MIPS64_MemoryMapParams,
};

static const PlatformMemoryMapParams FreeBSD_X86_MemoryMapParams = {
  &FreeBSD_I386_MemoryMapParams,
  &FreeBSD_X86_64_MemoryMapParams,
};

/// \brief An instrumentation pass implementing detection of uninitialized
/// reads.
///
/// MemorySanitizer: instrument the code in module to find
/// uninitialized reads.
class MemorySanitizer : public FunctionPass {
 public:
  MemorySanitizer(int TrackOrigins = 0)
      : FunctionPass(ID),
        TrackOrigins(std::max(TrackOrigins, (int)ClTrackOrigins)),
        WarningFn(nullptr) {}
  const char *getPassName() const override { return "MemorySanitizer"; }
  bool runOnFunction(Function &F) override;
  bool doInitialization(Module &M) override;
  static char ID;  // Pass identification, replacement for typeid.

 private:
  void initializeCallbacks(Module &M);

  /// \brief Track origins (allocation points) of uninitialized values.
  int TrackOrigins;

  LLVMContext *C;
  Type *IntptrTy;
  Type *OriginTy;
  /// \brief Thread-local shadow storage for function parameters.
  GlobalVariable *ParamTLS;
  /// \brief Thread-local origin storage for function parameters.
  GlobalVariable *ParamOriginTLS;
  /// \brief Thread-local shadow storage for function return value.
  GlobalVariable *RetvalTLS;
  /// \brief Thread-local origin storage for function return value.
  GlobalVariable *RetvalOriginTLS;
  /// \brief Thread-local shadow storage for in-register va_arg function
  /// parameters (x86_64-specific).
  GlobalVariable *VAArgTLS;
  /// \brief Thread-local shadow storage for va_arg overflow area
  /// (x86_64-specific).
  GlobalVariable *VAArgOverflowSizeTLS;
  /// \brief Thread-local space used to pass origin value to the UMR reporting
  /// function.
  GlobalVariable *OriginTLS;

  /// \brief The run-time callback to print a warning.
  Value *WarningFn;
  // These arrays are indexed by log2(AccessSize).
  Value *MaybeWarningFn[kNumberOfAccessSizes];
  Value *MaybeStoreOriginFn[kNumberOfAccessSizes];

  /// \brief Run-time helper that generates a new origin value for a stack
  /// allocation.
  Value *MsanSetAllocaOrigin4Fn;
  /// \brief Run-time helper that poisons stack on function entry.
  Value *MsanPoisonStackFn;
  /// \brief Run-time helper that records a store (or any event) of an
  /// uninitialized value and returns an updated origin id encoding this info.
  Value *MsanChainOriginFn;
  /// \brief MSan runtime replacements for memmove, memcpy and memset.
  Value *MemmoveFn, *MemcpyFn, *MemsetFn;

  /// \brief Memory map parameters used in application-to-shadow calculation.
  const MemoryMapParams *MapParams;

  MDNode *ColdCallWeights;
  /// \brief Branch weights for origin store.
  MDNode *OriginStoreWeights;
  /// \brief An empty volatile inline asm that prevents callback merge.
  InlineAsm *EmptyAsm;

  friend struct MemorySanitizerVisitor;
  friend struct VarArgAMD64Helper;
  friend struct VarArgMIPS64Helper;
};
}  // namespace

char MemorySanitizer::ID = 0;
INITIALIZE_PASS(MemorySanitizer, "msan",
                "MemorySanitizer: detects uninitialized reads.",
                false, false)

FunctionPass *llvm::createMemorySanitizerPass(int TrackOrigins) {
  return new MemorySanitizer(TrackOrigins);
}

/// \brief Create a non-const global initialized with the given string.
///
/// Creates a writable global for Str so that we can pass it to the
/// run-time lib. Runtime uses first 4 bytes of the string to store the
/// frame ID, so the string needs to be mutable.
static GlobalVariable *createPrivateNonConstGlobalForString(Module &M,
                                                            StringRef Str) {
  Constant *StrConst = ConstantDataArray::getString(M.getContext(), Str);
  return new GlobalVariable(M, StrConst->getType(), /*isConstant=*/false,
                            GlobalValue::PrivateLinkage, StrConst, "");
}


/// \brief Insert extern declaration of runtime-provided functions and globals.
void MemorySanitizer::initializeCallbacks(Module &M) {
  // Only do this once.
  if (WarningFn)
    return;

  IRBuilder<> IRB(*C);
  // Create the callback.
  // FIXME: this function should have "Cold" calling conv,
  // which is not yet implemented.
  StringRef WarningFnName = ClKeepGoing ? "__msan_warning"
                                        : "__msan_warning_noreturn";
  WarningFn = M.getOrInsertFunction(WarningFnName, IRB.getVoidTy(), nullptr);

  for (size_t AccessSizeIndex = 0; AccessSizeIndex < kNumberOfAccessSizes;
       AccessSizeIndex++) {
    unsigned AccessSize = 1 << AccessSizeIndex;
    std::string FunctionName = "__msan_maybe_warning_" + itostr(AccessSize);
    MaybeWarningFn[AccessSizeIndex] = M.getOrInsertFunction(
        FunctionName, IRB.getVoidTy(), IRB.getIntNTy(AccessSize * 8),
        IRB.getInt32Ty(), nullptr);

    FunctionName = "__msan_maybe_store_origin_" + itostr(AccessSize);
    MaybeStoreOriginFn[AccessSizeIndex] = M.getOrInsertFunction(
        FunctionName, IRB.getVoidTy(), IRB.getIntNTy(AccessSize * 8),
        IRB.getInt8PtrTy(), IRB.getInt32Ty(), nullptr);
  }

  MsanSetAllocaOrigin4Fn = M.getOrInsertFunction(
    "__msan_set_alloca_origin4", IRB.getVoidTy(), IRB.getInt8PtrTy(), IntptrTy,
    IRB.getInt8PtrTy(), IntptrTy, nullptr);
  MsanPoisonStackFn =
      M.getOrInsertFunction("__msan_poison_stack", IRB.getVoidTy(),
                            IRB.getInt8PtrTy(), IntptrTy, nullptr);
  MsanChainOriginFn = M.getOrInsertFunction(
    "__msan_chain_origin", IRB.getInt32Ty(), IRB.getInt32Ty(), nullptr);
  MemmoveFn = M.getOrInsertFunction(
    "__msan_memmove", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
    IRB.getInt8PtrTy(), IntptrTy, nullptr);
  MemcpyFn = M.getOrInsertFunction(
    "__msan_memcpy", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
    IntptrTy, nullptr);
  MemsetFn = M.getOrInsertFunction(
    "__msan_memset", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(), IRB.getInt32Ty(),
    IntptrTy, nullptr);

  // Create globals.
  RetvalTLS = new GlobalVariable(
    M, ArrayType::get(IRB.getInt64Ty(), kRetvalTLSSize / 8), false,
    GlobalVariable::ExternalLinkage, nullptr, "__msan_retval_tls", nullptr,
    GlobalVariable::InitialExecTLSModel);
  RetvalOriginTLS = new GlobalVariable(
    M, OriginTy, false, GlobalVariable::ExternalLinkage, nullptr,
    "__msan_retval_origin_tls", nullptr, GlobalVariable::InitialExecTLSModel);

  ParamTLS = new GlobalVariable(
    M, ArrayType::get(IRB.getInt64Ty(), kParamTLSSize / 8), false,
    GlobalVariable::ExternalLinkage, nullptr, "__msan_param_tls", nullptr,
    GlobalVariable::InitialExecTLSModel);
  ParamOriginTLS = new GlobalVariable(
    M, ArrayType::get(OriginTy, kParamTLSSize / 4), false,
    GlobalVariable::ExternalLinkage, nullptr, "__msan_param_origin_tls",
    nullptr, GlobalVariable::InitialExecTLSModel);

  VAArgTLS = new GlobalVariable(
    M, ArrayType::get(IRB.getInt64Ty(), kParamTLSSize / 8), false,
    GlobalVariable::ExternalLinkage, nullptr, "__msan_va_arg_tls", nullptr,
    GlobalVariable::InitialExecTLSModel);
  VAArgOverflowSizeTLS = new GlobalVariable(
    M, IRB.getInt64Ty(), false, GlobalVariable::ExternalLinkage, nullptr,
    "__msan_va_arg_overflow_size_tls", nullptr,
    GlobalVariable::InitialExecTLSModel);
  OriginTLS = new GlobalVariable(
    M, IRB.getInt32Ty(), false, GlobalVariable::ExternalLinkage, nullptr,
    "__msan_origin_tls", nullptr, GlobalVariable::InitialExecTLSModel);

  // We insert an empty inline asm after __msan_report* to avoid callback merge.
  EmptyAsm = InlineAsm::get(FunctionType::get(IRB.getVoidTy(), false),
                            StringRef(""), StringRef(""),
                            /*hasSideEffects=*/true);
}

/// \brief Module-level initialization.
///
/// inserts a call to __msan_init to the module's constructor list.
bool MemorySanitizer::doInitialization(Module &M) {
  auto &DL = M.getDataLayout();

  Triple TargetTriple(M.getTargetTriple());
  switch (TargetTriple.getOS()) {
    case Triple::FreeBSD:
      switch (TargetTriple.getArch()) {
        case Triple::x86_64:
          MapParams = FreeBSD_X86_MemoryMapParams.bits64;
          break;
        case Triple::x86:
          MapParams = FreeBSD_X86_MemoryMapParams.bits32;
          break;
        default:
          report_fatal_error("unsupported architecture");
      }
      break;
    case Triple::Linux:
      switch (TargetTriple.getArch()) {
        case Triple::x86_64:
          MapParams = Linux_X86_MemoryMapParams.bits64;
          break;
        case Triple::x86:
          MapParams = Linux_X86_MemoryMapParams.bits32;
          break;
        case Triple::mips64:
        case Triple::mips64el:
          MapParams = Linux_MIPS_MemoryMapParams.bits64;
          break;
        default:
          report_fatal_error("unsupported architecture");
      }
      break;
    default:
      report_fatal_error("unsupported operating system");
  }

  C = &(M.getContext());
  IRBuilder<> IRB(*C);
  IntptrTy = IRB.getIntPtrTy(DL);
  OriginTy = IRB.getInt32Ty();

  ColdCallWeights = MDBuilder(*C).createBranchWeights(1, 1000);
  OriginStoreWeights = MDBuilder(*C).createBranchWeights(1, 1000);

  // Insert a call to __msan_init/__msan_track_origins into the module's CTORs.
  appendToGlobalCtors(M, cast<Function>(M.getOrInsertFunction(
                      "__msan_init", IRB.getVoidTy(), nullptr)), 0);

  if (TrackOrigins)
    new GlobalVariable(M, IRB.getInt32Ty(), true, GlobalValue::WeakODRLinkage,
                       IRB.getInt32(TrackOrigins), "__msan_track_origins");

  if (ClKeepGoing)
    new GlobalVariable(M, IRB.getInt32Ty(), true, GlobalValue::WeakODRLinkage,
                       IRB.getInt32(ClKeepGoing), "__msan_keep_going");

  return true;
}

namespace {

/// \brief A helper class that handles instrumentation of VarArg
/// functions on a particular platform.
///
/// Implementations are expected to insert the instrumentation
/// necessary to propagate argument shadow through VarArg function
/// calls. Visit* methods are called during an InstVisitor pass over
/// the function, and should avoid creating new basic blocks. A new
/// instance of this class is created for each instrumented function.
struct VarArgHelper {
  /// \brief Visit a CallSite.
  virtual void visitCallSite(CallSite &CS, IRBuilder<> &IRB) = 0;

  /// \brief Visit a va_start call.
  virtual void visitVAStartInst(VAStartInst &I) = 0;

  /// \brief Visit a va_copy call.
  virtual void visitVACopyInst(VACopyInst &I) = 0;

  /// \brief Finalize function instrumentation.
  ///
  /// This method is called after visiting all interesting (see above)
  /// instructions in a function.
  virtual void finalizeInstrumentation() = 0;

  virtual ~VarArgHelper() {}
};

struct MemorySanitizerVisitor;

VarArgHelper*
CreateVarArgHelper(Function &Func, MemorySanitizer &Msan,
                   MemorySanitizerVisitor &Visitor);

unsigned TypeSizeToSizeIndex(unsigned TypeSize) {
  if (TypeSize <= 8) return 0;
  return Log2_32_Ceil(TypeSize / 8);
}

/// This class does all the work for a given function. Store and Load
/// instructions store and load corresponding shadow and origin
/// values. Most instructions propagate shadow from arguments to their
/// return values. Certain instructions (most importantly, BranchInst)
/// test their argument shadow and print reports (with a runtime call) if it's
/// non-zero.
struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
  Function &F;
  MemorySanitizer &MS;
  SmallVector<PHINode *, 16> ShadowPHINodes, OriginPHINodes;
  ValueMap<Value*, Value*> ShadowMap, OriginMap;
  std::unique_ptr<VarArgHelper> VAHelper;

  // The following flags disable parts of MSan instrumentation based on
  // blacklist contents and command-line options.
  bool InsertChecks;
  bool PropagateShadow;
  bool PoisonStack;
  bool PoisonUndef;
  bool CheckReturnValue;

  struct ShadowOriginAndInsertPoint {
    Value *Shadow;
    Value *Origin;
    Instruction *OrigIns;
    ShadowOriginAndInsertPoint(Value *S, Value *O, Instruction *I)
      : Shadow(S), Origin(O), OrigIns(I) { }
  };
  SmallVector<ShadowOriginAndInsertPoint, 16> InstrumentationList;
  SmallVector<Instruction*, 16> StoreList;

  MemorySanitizerVisitor(Function &F, MemorySanitizer &MS)
      : F(F), MS(MS), VAHelper(CreateVarArgHelper(F, MS, *this)) {
    bool SanitizeFunction = F.hasFnAttribute(Attribute::SanitizeMemory);
    InsertChecks = SanitizeFunction;
    PropagateShadow = SanitizeFunction;
    PoisonStack = SanitizeFunction && ClPoisonStack;
    PoisonUndef = SanitizeFunction && ClPoisonUndef;
    // FIXME: Consider using SpecialCaseList to specify a list of functions that
    // must always return fully initialized values. For now, we hardcode "main".
    CheckReturnValue = SanitizeFunction && (F.getName() == "main");

    DEBUG(if (!InsertChecks)
          dbgs() << "MemorySanitizer is not inserting checks into '"
                 << F.getName() << "'\n");
  }

  Value *updateOrigin(Value *V, IRBuilder<> &IRB) {
    if (MS.TrackOrigins <= 1) return V;
    return IRB.CreateCall(MS.MsanChainOriginFn, V);
  }

  Value *originToIntptr(IRBuilder<> &IRB, Value *Origin) {
    const DataLayout &DL = F.getParent()->getDataLayout();
    unsigned IntptrSize = DL.getTypeStoreSize(MS.IntptrTy);
    if (IntptrSize == kOriginSize) return Origin;
    assert(IntptrSize == kOriginSize * 2);
    Origin = IRB.CreateIntCast(Origin, MS.IntptrTy, /* isSigned */ false);
    return IRB.CreateOr(Origin, IRB.CreateShl(Origin, kOriginSize * 8));
  }

  /// \brief Fill memory range with the given origin value.
  void paintOrigin(IRBuilder<> &IRB, Value *Origin, Value *OriginPtr,
                   unsigned Size, unsigned Alignment) {
    const DataLayout &DL = F.getParent()->getDataLayout();
    unsigned IntptrAlignment = DL.getABITypeAlignment(MS.IntptrTy);
    unsigned IntptrSize = DL.getTypeStoreSize(MS.IntptrTy);
    assert(IntptrAlignment >= kMinOriginAlignment);
    assert(IntptrSize >= kOriginSize);

    unsigned Ofs = 0;
    unsigned CurrentAlignment = Alignment;
    if (Alignment >= IntptrAlignment && IntptrSize > kOriginSize) {
      Value *IntptrOrigin = originToIntptr(IRB, Origin);
      Value *IntptrOriginPtr =
          IRB.CreatePointerCast(OriginPtr, PointerType::get(MS.IntptrTy, 0));
      for (unsigned i = 0; i < Size / IntptrSize; ++i) {
        Value *Ptr =
            i ? IRB.CreateConstGEP1_32(IntptrOriginPtr, i) : IntptrOriginPtr;
        IRB.CreateAlignedStore(IntptrOrigin, Ptr, CurrentAlignment);
        Ofs += IntptrSize / kOriginSize;
        CurrentAlignment = IntptrAlignment;
      }
    }

    for (unsigned i = Ofs; i < (Size + kOriginSize - 1) / kOriginSize; ++i) {
      Value *GEP = i ? IRB.CreateConstGEP1_32(OriginPtr, i) : OriginPtr;
      IRB.CreateAlignedStore(Origin, GEP, CurrentAlignment);
      CurrentAlignment = kMinOriginAlignment;
    }
  }

  void storeOrigin(IRBuilder<> &IRB, Value *Addr, Value *Shadow, Value *Origin,
                   unsigned Alignment, bool AsCall) {
    const DataLayout &DL = F.getParent()->getDataLayout();
    unsigned OriginAlignment = std::max(kMinOriginAlignment, Alignment);
    unsigned StoreSize = DL.getTypeStoreSize(Shadow->getType());
    if (isa<StructType>(Shadow->getType())) {
      paintOrigin(IRB, updateOrigin(Origin, IRB),
                  getOriginPtr(Addr, IRB, Alignment), StoreSize,
                  OriginAlignment);
    } else {
      Value *ConvertedShadow = convertToShadowTyNoVec(Shadow, IRB);
      Constant *ConstantShadow = dyn_cast_or_null<Constant>(ConvertedShadow);
      if (ConstantShadow) {
        if (ClCheckConstantShadow && !ConstantShadow->isZeroValue())
          paintOrigin(IRB, updateOrigin(Origin, IRB),
                      getOriginPtr(Addr, IRB, Alignment), StoreSize,
                      OriginAlignment);
        return;
      }

      unsigned TypeSizeInBits =
          DL.getTypeSizeInBits(ConvertedShadow->getType());
      unsigned SizeIndex = TypeSizeToSizeIndex(TypeSizeInBits);
      if (AsCall && SizeIndex < kNumberOfAccessSizes) {
        Value *Fn = MS.MaybeStoreOriginFn[SizeIndex];
        Value *ConvertedShadow2 = IRB.CreateZExt(
            ConvertedShadow, IRB.getIntNTy(8 * (1 << SizeIndex)));
        IRB.CreateCall3(Fn, ConvertedShadow2,
                        IRB.CreatePointerCast(Addr, IRB.getInt8PtrTy()),
                        Origin);
      } else {
        Value *Cmp = IRB.CreateICmpNE(
            ConvertedShadow, getCleanShadow(ConvertedShadow), "_mscmp");
        Instruction *CheckTerm = SplitBlockAndInsertIfThen(
            Cmp, IRB.GetInsertPoint(), false, MS.OriginStoreWeights);
        IRBuilder<> IRBNew(CheckTerm);
        paintOrigin(IRBNew, updateOrigin(Origin, IRBNew),
                    getOriginPtr(Addr, IRBNew, Alignment), StoreSize,
                    OriginAlignment);
      }
    }
  }

  void materializeStores(bool InstrumentWithCalls) {
    for (auto Inst : StoreList) {
      StoreInst &SI = *dyn_cast<StoreInst>(Inst);

      IRBuilder<> IRB(&SI);
      Value *Val = SI.getValueOperand();
      Value *Addr = SI.getPointerOperand();
      Value *Shadow = SI.isAtomic() ? getCleanShadow(Val) : getShadow(Val);
      Value *ShadowPtr = getShadowPtr(Addr, Shadow->getType(), IRB);

      StoreInst *NewSI =
          IRB.CreateAlignedStore(Shadow, ShadowPtr, SI.getAlignment());
      DEBUG(dbgs() << "  STORE: " << *NewSI << "\n");
      (void)NewSI;

      if (ClCheckAccessAddress) insertShadowCheck(Addr, &SI);

      if (SI.isAtomic()) SI.setOrdering(addReleaseOrdering(SI.getOrdering()));

      if (MS.TrackOrigins && !SI.isAtomic())
        storeOrigin(IRB, Addr, Shadow, getOrigin(Val), SI.getAlignment(),
                    InstrumentWithCalls);
    }
  }

  void materializeOneCheck(Instruction *OrigIns, Value *Shadow, Value *Origin,
                           bool AsCall) {
    IRBuilder<> IRB(OrigIns);
    DEBUG(dbgs() << "  SHAD0 : " << *Shadow << "\n");
    Value *ConvertedShadow = convertToShadowTyNoVec(Shadow, IRB);
    DEBUG(dbgs() << "  SHAD1 : " << *ConvertedShadow << "\n");

    Constant *ConstantShadow = dyn_cast_or_null<Constant>(ConvertedShadow);
    if (ConstantShadow) {
      if (ClCheckConstantShadow && !ConstantShadow->isZeroValue()) {
        if (MS.TrackOrigins) {
          IRB.CreateStore(Origin ? (Value *)Origin : (Value *)IRB.getInt32(0),
                          MS.OriginTLS);
        }
        IRB.CreateCall(MS.WarningFn);
        IRB.CreateCall(MS.EmptyAsm);
        // FIXME: Insert UnreachableInst if !ClKeepGoing?
        // This may invalidate some of the following checks and needs to be done
        // at the very end.
      }
      return;
    }

    const DataLayout &DL = OrigIns->getModule()->getDataLayout();

    unsigned TypeSizeInBits = DL.getTypeSizeInBits(ConvertedShadow->getType());
    unsigned SizeIndex = TypeSizeToSizeIndex(TypeSizeInBits);
    if (AsCall && SizeIndex < kNumberOfAccessSizes) {
      Value *Fn = MS.MaybeWarningFn[SizeIndex];
      Value *ConvertedShadow2 =
          IRB.CreateZExt(ConvertedShadow, IRB.getIntNTy(8 * (1 << SizeIndex)));
      IRB.CreateCall2(Fn, ConvertedShadow2, MS.TrackOrigins && Origin
                                                ? Origin
                                                : (Value *)IRB.getInt32(0));
    } else {
      Value *Cmp = IRB.CreateICmpNE(ConvertedShadow,
                                    getCleanShadow(ConvertedShadow), "_mscmp");
      Instruction *CheckTerm = SplitBlockAndInsertIfThen(
          Cmp, OrigIns,
          /* Unreachable */ !ClKeepGoing, MS.ColdCallWeights);

      IRB.SetInsertPoint(CheckTerm);
      if (MS.TrackOrigins) {
        IRB.CreateStore(Origin ? (Value *)Origin : (Value *)IRB.getInt32(0),
                        MS.OriginTLS);
      }
      IRB.CreateCall(MS.WarningFn);
      IRB.CreateCall(MS.EmptyAsm);
      DEBUG(dbgs() << "  CHECK: " << *Cmp << "\n");
    }
  }

  void materializeChecks(bool InstrumentWithCalls) {
    for (const auto &ShadowData : InstrumentationList) {
      Instruction *OrigIns = ShadowData.OrigIns;
      Value *Shadow = ShadowData.Shadow;
      Value *Origin = ShadowData.Origin;
      materializeOneCheck(OrigIns, Shadow, Origin, InstrumentWithCalls);
    }
    DEBUG(dbgs() << "DONE:\n" << F);
  }

  /// \brief Add MemorySanitizer instrumentation to a function.
  bool runOnFunction() {
    MS.initializeCallbacks(*F.getParent());

    // In the presence of unreachable blocks, we may see Phi nodes with
    // incoming nodes from such blocks. Since InstVisitor skips unreachable
    // blocks, such nodes will not have any shadow value associated with them.
    // It's easier to remove unreachable blocks than deal with missing shadow.
    removeUnreachableBlocks(F);

    // Iterate all BBs in depth-first order and create shadow instructions
    // for all instructions (where applicable).
    // For PHI nodes we create dummy shadow PHIs which will be finalized later.
    for (BasicBlock *BB : depth_first(&F.getEntryBlock()))
      visit(*BB);


    // Finalize PHI nodes.
    for (PHINode *PN : ShadowPHINodes) {
      PHINode *PNS = cast<PHINode>(getShadow(PN));
      PHINode *PNO = MS.TrackOrigins ? cast<PHINode>(getOrigin(PN)) : nullptr;
      size_t NumValues = PN->getNumIncomingValues();
      for (size_t v = 0; v < NumValues; v++) {
        PNS->addIncoming(getShadow(PN, v), PN->getIncomingBlock(v));
        if (PNO) PNO->addIncoming(getOrigin(PN, v), PN->getIncomingBlock(v));
      }
    }

    VAHelper->finalizeInstrumentation();

    bool InstrumentWithCalls = ClInstrumentationWithCallThreshold >= 0 &&
                               InstrumentationList.size() + StoreList.size() >
                                   (unsigned)ClInstrumentationWithCallThreshold;

    // Delayed instrumentation of StoreInst.
    // This may add new checks to be inserted later.
    materializeStores(InstrumentWithCalls);

    // Insert shadow value checks.
    materializeChecks(InstrumentWithCalls);

    return true;
  }

  /// \brief Compute the shadow type that corresponds to a given Value.
  Type *getShadowTy(Value *V) {
    return getShadowTy(V->getType());
  }

  /// \brief Compute the shadow type that corresponds to a given Type.
  Type *getShadowTy(Type *OrigTy) {
    if (!OrigTy->isSized()) {
      return nullptr;
    }
    // For integer type, shadow is the same as the original type.
    // This may return weird-sized types like i1.
    if (IntegerType *IT = dyn_cast<IntegerType>(OrigTy))
      return IT;
    const DataLayout &DL = F.getParent()->getDataLayout();
    if (VectorType *VT = dyn_cast<VectorType>(OrigTy)) {
      uint32_t EltSize = DL.getTypeSizeInBits(VT->getElementType());
      return VectorType::get(IntegerType::get(*MS.C, EltSize),
                             VT->getNumElements());
    }
    if (ArrayType *AT = dyn_cast<ArrayType>(OrigTy)) {
      return ArrayType::get(getShadowTy(AT->getElementType()),
                            AT->getNumElements());
    }
    if (StructType *ST = dyn_cast<StructType>(OrigTy)) {
      SmallVector<Type*, 4> Elements;
      for (unsigned i = 0, n = ST->getNumElements(); i < n; i++)
        Elements.push_back(getShadowTy(ST->getElementType(i)));
      StructType *Res = StructType::get(*MS.C, Elements, ST->isPacked());
      DEBUG(dbgs() << "getShadowTy: " << *ST << " ===> " << *Res << "\n");
      return Res;
    }
    uint32_t TypeSize = DL.getTypeSizeInBits(OrigTy);
    return IntegerType::get(*MS.C, TypeSize);
  }

  /// \brief Flatten a vector type.
  Type *getShadowTyNoVec(Type *ty) {
    if (VectorType *vt = dyn_cast<VectorType>(ty))
      return IntegerType::get(*MS.C, vt->getBitWidth());
    return ty;
  }

  /// \brief Convert a shadow value to it's flattened variant.
  Value *convertToShadowTyNoVec(Value *V, IRBuilder<> &IRB) {
    Type *Ty = V->getType();
    Type *NoVecTy = getShadowTyNoVec(Ty);
    if (Ty == NoVecTy) return V;
    return IRB.CreateBitCast(V, NoVecTy);
  }

  /// \brief Compute the integer shadow offset that corresponds to a given
  /// application address.
  ///
  /// Offset = (Addr & ~AndMask) ^ XorMask
  Value *getShadowPtrOffset(Value *Addr, IRBuilder<> &IRB) {
    uint64_t AndMask = MS.MapParams->AndMask;
    assert(AndMask != 0 && "AndMask shall be specified");
    Value *OffsetLong =
      IRB.CreateAnd(IRB.CreatePointerCast(Addr, MS.IntptrTy),
                    ConstantInt::get(MS.IntptrTy, ~AndMask));

    uint64_t XorMask = MS.MapParams->XorMask;
    if (XorMask != 0)
      OffsetLong = IRB.CreateXor(OffsetLong,
                                 ConstantInt::get(MS.IntptrTy, XorMask));
    return OffsetLong;
  }

  /// \brief Compute the shadow address that corresponds to a given application
  /// address.
  ///
  /// Shadow = ShadowBase + Offset
  Value *getShadowPtr(Value *Addr, Type *ShadowTy,
                      IRBuilder<> &IRB) {
    Value *ShadowLong = getShadowPtrOffset(Addr, IRB);
    uint64_t ShadowBase = MS.MapParams->ShadowBase;
    if (ShadowBase != 0)
      ShadowLong =
        IRB.CreateAdd(ShadowLong,
                      ConstantInt::get(MS.IntptrTy, ShadowBase));
    return IRB.CreateIntToPtr(ShadowLong, PointerType::get(ShadowTy, 0));
  }

  /// \brief Compute the origin address that corresponds to a given application
  /// address.
  ///
  /// OriginAddr = (OriginBase + Offset) & ~3ULL
  Value *getOriginPtr(Value *Addr, IRBuilder<> &IRB, unsigned Alignment) {
    Value *OriginLong = getShadowPtrOffset(Addr, IRB);
    uint64_t OriginBase = MS.MapParams->OriginBase;
    if (OriginBase != 0)
      OriginLong =
        IRB.CreateAdd(OriginLong,
                      ConstantInt::get(MS.IntptrTy, OriginBase));
    if (Alignment < kMinOriginAlignment) {
      uint64_t Mask = kMinOriginAlignment - 1;
      OriginLong = IRB.CreateAnd(OriginLong,
                                 ConstantInt::get(MS.IntptrTy, ~Mask));
    }
    return IRB.CreateIntToPtr(OriginLong,
                              PointerType::get(IRB.getInt32Ty(), 0));
  }

  /// \brief Compute the shadow address for a given function argument.
  ///
  /// Shadow = ParamTLS+ArgOffset.
  Value *getShadowPtrForArgument(Value *A, IRBuilder<> &IRB,
                                 int ArgOffset) {
    Value *Base = IRB.CreatePointerCast(MS.ParamTLS, MS.IntptrTy);
    Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
    return IRB.CreateIntToPtr(Base, PointerType::get(getShadowTy(A), 0),
                              "_msarg");
  }

  /// \brief Compute the origin address for a given function argument.
  Value *getOriginPtrForArgument(Value *A, IRBuilder<> &IRB,
                                 int ArgOffset) {
    if (!MS.TrackOrigins) return nullptr;
    Value *Base = IRB.CreatePointerCast(MS.ParamOriginTLS, MS.IntptrTy);
    Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
    return IRB.CreateIntToPtr(Base, PointerType::get(MS.OriginTy, 0),
                              "_msarg_o");
  }

  /// \brief Compute the shadow address for a retval.
  Value *getShadowPtrForRetval(Value *A, IRBuilder<> &IRB) {
    Value *Base = IRB.CreatePointerCast(MS.RetvalTLS, MS.IntptrTy);
    return IRB.CreateIntToPtr(Base, PointerType::get(getShadowTy(A), 0),
                              "_msret");
  }

  /// \brief Compute the origin address for a retval.
  Value *getOriginPtrForRetval(IRBuilder<> &IRB) {
    // We keep a single origin for the entire retval. Might be too optimistic.
    return MS.RetvalOriginTLS;
  }

  /// \brief Set SV to be the shadow value for V.
  void setShadow(Value *V, Value *SV) {
    assert(!ShadowMap.count(V) && "Values may only have one shadow");
    ShadowMap[V] = PropagateShadow ? SV : getCleanShadow(V);
  }

  /// \brief Set Origin to be the origin value for V.
  void setOrigin(Value *V, Value *Origin) {
    if (!MS.TrackOrigins) return;
    assert(!OriginMap.count(V) && "Values may only have one origin");
    DEBUG(dbgs() << "ORIGIN: " << *V << "  ==> " << *Origin << "\n");
    OriginMap[V] = Origin;
  }

  /// \brief Create a clean shadow value for a given value.
  ///
  /// Clean shadow (all zeroes) means all bits of the value are defined
  /// (initialized).
  Constant *getCleanShadow(Value *V) {
    Type *ShadowTy = getShadowTy(V);
    if (!ShadowTy)
      return nullptr;
    return Constant::getNullValue(ShadowTy);
  }

  /// \brief Create a dirty shadow of a given shadow type.
  Constant *getPoisonedShadow(Type *ShadowTy) {
    assert(ShadowTy);
    if (isa<IntegerType>(ShadowTy) || isa<VectorType>(ShadowTy))
      return Constant::getAllOnesValue(ShadowTy);
    if (ArrayType *AT = dyn_cast<ArrayType>(ShadowTy)) {
      SmallVector<Constant *, 4> Vals(AT->getNumElements(),
                                      getPoisonedShadow(AT->getElementType()));
      return ConstantArray::get(AT, Vals);
    }
    if (StructType *ST = dyn_cast<StructType>(ShadowTy)) {
      SmallVector<Constant *, 4> Vals;
      for (unsigned i = 0, n = ST->getNumElements(); i < n; i++)
        Vals.push_back(getPoisonedShadow(ST->getElementType(i)));
      return ConstantStruct::get(ST, Vals);
    }
    llvm_unreachable("Unexpected shadow type");
  }

  /// \brief Create a dirty shadow for a given value.
  Constant *getPoisonedShadow(Value *V) {
    Type *ShadowTy = getShadowTy(V);
    if (!ShadowTy)
      return nullptr;
    return getPoisonedShadow(ShadowTy);
  }

  /// \brief Create a clean (zero) origin.
  Value *getCleanOrigin() {
    return Constant::getNullValue(MS.OriginTy);
  }

  /// \brief Get the shadow value for a given Value.
  ///
  /// This function either returns the value set earlier with setShadow,
  /// or extracts if from ParamTLS (for function arguments).
  Value *getShadow(Value *V) {
    if (!PropagateShadow) return getCleanShadow(V);
    if (Instruction *I = dyn_cast<Instruction>(V)) {
      // For instructions the shadow is already stored in the map.
      Value *Shadow = ShadowMap[V];
      if (!Shadow) {
        DEBUG(dbgs() << "No shadow: " << *V << "\n" << *(I->getParent()));
        (void)I;
        assert(Shadow && "No shadow for a value");
      }
      return Shadow;
    }
    if (UndefValue *U = dyn_cast<UndefValue>(V)) {
      Value *AllOnes = PoisonUndef ? getPoisonedShadow(V) : getCleanShadow(V);
      DEBUG(dbgs() << "Undef: " << *U << " ==> " << *AllOnes << "\n");
      (void)U;
      return AllOnes;
    }
    if (Argument *A = dyn_cast<Argument>(V)) {
      // For arguments we compute the shadow on demand and store it in the map.
      Value **ShadowPtr = &ShadowMap[V];
      if (*ShadowPtr)
        return *ShadowPtr;
      Function *F = A->getParent();
      IRBuilder<> EntryIRB(F->getEntryBlock().getFirstNonPHI());
      unsigned ArgOffset = 0;
      const DataLayout &DL = F->getParent()->getDataLayout();
      for (auto &FArg : F->args()) {
        if (!FArg.getType()->isSized()) {
          DEBUG(dbgs() << "Arg is not sized\n");
          continue;
        }
        unsigned Size =
            FArg.hasByValAttr()
                ? DL.getTypeAllocSize(FArg.getType()->getPointerElementType())
                : DL.getTypeAllocSize(FArg.getType());
        if (A == &FArg) {
          bool Overflow = ArgOffset + Size > kParamTLSSize;
          Value *Base = getShadowPtrForArgument(&FArg, EntryIRB, ArgOffset);
          if (FArg.hasByValAttr()) {
            // ByVal pointer itself has clean shadow. We copy the actual
            // argument shadow to the underlying memory.
            // Figure out maximal valid memcpy alignment.
            unsigned ArgAlign = FArg.getParamAlignment();
            if (ArgAlign == 0) {
              Type *EltType = A->getType()->getPointerElementType();
              ArgAlign = DL.getABITypeAlignment(EltType);
            }
            if (Overflow) {
              // ParamTLS overflow.
              EntryIRB.CreateMemSet(
                  getShadowPtr(V, EntryIRB.getInt8Ty(), EntryIRB),
                  Constant::getNullValue(EntryIRB.getInt8Ty()), Size, ArgAlign);
            } else {
              unsigned CopyAlign = std::min(ArgAlign, kShadowTLSAlignment);
              Value *Cpy = EntryIRB.CreateMemCpy(
                  getShadowPtr(V, EntryIRB.getInt8Ty(), EntryIRB), Base, Size,
                  CopyAlign);
              DEBUG(dbgs() << "  ByValCpy: " << *Cpy << "\n");
              (void)Cpy;
            }
            *ShadowPtr = getCleanShadow(V);
          } else {
            if (Overflow) {
              // ParamTLS overflow.
              *ShadowPtr = getCleanShadow(V);
            } else {
              *ShadowPtr =
                  EntryIRB.CreateAlignedLoad(Base, kShadowTLSAlignment);
            }
          }
          DEBUG(dbgs() << "  ARG:    "  << FArg << " ==> " <<
                **ShadowPtr << "\n");
          if (MS.TrackOrigins && !Overflow) {
            Value *OriginPtr =
                getOriginPtrForArgument(&FArg, EntryIRB, ArgOffset);
            setOrigin(A, EntryIRB.CreateLoad(OriginPtr));
          } else {
            setOrigin(A, getCleanOrigin());
          }
        }
        ArgOffset += RoundUpToAlignment(Size, kShadowTLSAlignment);
      }
      assert(*ShadowPtr && "Could not find shadow for an argument");
      return *ShadowPtr;
    }
    // For everything else the shadow is zero.
    return getCleanShadow(V);
  }

  /// \brief Get the shadow for i-th argument of the instruction I.
  Value *getShadow(Instruction *I, int i) {
    return getShadow(I->getOperand(i));
  }

  /// \brief Get the origin for a value.
  Value *getOrigin(Value *V) {
    if (!MS.TrackOrigins) return nullptr;
    if (!PropagateShadow) return getCleanOrigin();
    if (isa<Constant>(V)) return getCleanOrigin();
    assert((isa<Instruction>(V) || isa<Argument>(V)) &&
           "Unexpected value type in getOrigin()");
    Value *Origin = OriginMap[V];
    assert(Origin && "Missing origin");
    return Origin;
  }

  /// \brief Get the origin for i-th argument of the instruction I.
  Value *getOrigin(Instruction *I, int i) {
    return getOrigin(I->getOperand(i));
  }

  /// \brief Remember the place where a shadow check should be inserted.
  ///
  /// This location will be later instrumented with a check that will print a
  /// UMR warning in runtime if the shadow value is not 0.
  void insertShadowCheck(Value *Shadow, Value *Origin, Instruction *OrigIns) {
    assert(Shadow);
    if (!InsertChecks) return;
#ifndef NDEBUG
    Type *ShadowTy = Shadow->getType();
    assert((isa<IntegerType>(ShadowTy) || isa<VectorType>(ShadowTy)) &&
           "Can only insert checks for integer and vector shadow types");
#endif
    InstrumentationList.push_back(
        ShadowOriginAndInsertPoint(Shadow, Origin, OrigIns));
  }

  /// \brief Remember the place where a shadow check should be inserted.
  ///
  /// This location will be later instrumented with a check that will print a
  /// UMR warning in runtime if the value is not fully defined.
  void insertShadowCheck(Value *Val, Instruction *OrigIns) {
    assert(Val);
    Value *Shadow, *Origin;
    if (ClCheckConstantShadow) {
      Shadow = getShadow(Val);
      if (!Shadow) return;
      Origin = getOrigin(Val);
    } else {
      Shadow = dyn_cast_or_null<Instruction>(getShadow(Val));
      if (!Shadow) return;
      Origin = dyn_cast_or_null<Instruction>(getOrigin(Val));
    }
    insertShadowCheck(Shadow, Origin, OrigIns);
  }

  AtomicOrdering addReleaseOrdering(AtomicOrdering a) {
    switch (a) {
      case NotAtomic:
        return NotAtomic;
      case Unordered:
      case Monotonic:
      case Release:
        return Release;
      case Acquire:
      case AcquireRelease:
        return AcquireRelease;
      case SequentiallyConsistent:
        return SequentiallyConsistent;
    }
    llvm_unreachable("Unknown ordering");
  }

  AtomicOrdering addAcquireOrdering(AtomicOrdering a) {
    switch (a) {
      case NotAtomic:
        return NotAtomic;
      case Unordered:
      case Monotonic:
      case Acquire:
        return Acquire;
      case Release:
      case AcquireRelease:
        return AcquireRelease;
      case SequentiallyConsistent:
        return SequentiallyConsistent;
    }
    llvm_unreachable("Unknown ordering");
  }

  // ------------------- Visitors.

  /// \brief Instrument LoadInst
  ///
  /// Loads the corresponding shadow and (optionally) origin.
  /// Optionally, checks that the load address is fully defined.
  void visitLoadInst(LoadInst &I) {
    assert(I.getType()->isSized() && "Load type must have size");
    IRBuilder<> IRB(I.getNextNode());
    Type *ShadowTy = getShadowTy(&I);
    Value *Addr = I.getPointerOperand();
    if (PropagateShadow && !I.getMetadata("nosanitize")) {
      Value *ShadowPtr = getShadowPtr(Addr, ShadowTy, IRB);
      setShadow(&I,
                IRB.CreateAlignedLoad(ShadowPtr, I.getAlignment(), "_msld"));
    } else {
      setShadow(&I, getCleanShadow(&I));
    }

    if (ClCheckAccessAddress)
      insertShadowCheck(I.getPointerOperand(), &I);

    if (I.isAtomic())
      I.setOrdering(addAcquireOrdering(I.getOrdering()));

    if (MS.TrackOrigins) {
      if (PropagateShadow) {
        unsigned Alignment = I.getAlignment();
        unsigned OriginAlignment = std::max(kMinOriginAlignment, Alignment);
        setOrigin(&I, IRB.CreateAlignedLoad(getOriginPtr(Addr, IRB, Alignment),
                                            OriginAlignment));
      } else {
        setOrigin(&I, getCleanOrigin());
      }
    }
  }

  /// \brief Instrument StoreInst
  ///
  /// Stores the corresponding shadow and (optionally) origin.
  /// Optionally, checks that the store address is fully defined.
  void visitStoreInst(StoreInst &I) {
    StoreList.push_back(&I);
  }

  void handleCASOrRMW(Instruction &I) {
    assert(isa<AtomicRMWInst>(I) || isa<AtomicCmpXchgInst>(I));

    IRBuilder<> IRB(&I);
    Value *Addr = I.getOperand(0);
    Value *ShadowPtr = getShadowPtr(Addr, I.getType(), IRB);

    if (ClCheckAccessAddress)
      insertShadowCheck(Addr, &I);

    // Only test the conditional argument of cmpxchg instruction.
    // The other argument can potentially be uninitialized, but we can not
    // detect this situation reliably without possible false positives.
    if (isa<AtomicCmpXchgInst>(I))
      insertShadowCheck(I.getOperand(1), &I);

    IRB.CreateStore(getCleanShadow(&I), ShadowPtr);

    setShadow(&I, getCleanShadow(&I));
    setOrigin(&I, getCleanOrigin());
  }

  void visitAtomicRMWInst(AtomicRMWInst &I) {
    handleCASOrRMW(I);
    I.setOrdering(addReleaseOrdering(I.getOrdering()));
  }

  void visitAtomicCmpXchgInst(AtomicCmpXchgInst &I) {
    handleCASOrRMW(I);
    I.setSuccessOrdering(addReleaseOrdering(I.getSuccessOrdering()));
  }

  // Vector manipulation.
  void visitExtractElementInst(ExtractElementInst &I) {
    insertShadowCheck(I.getOperand(1), &I);
    IRBuilder<> IRB(&I);
    setShadow(&I, IRB.CreateExtractElement(getShadow(&I, 0), I.getOperand(1),
              "_msprop"));
    setOrigin(&I, getOrigin(&I, 0));
  }

  void visitInsertElementInst(InsertElementInst &I) {
    insertShadowCheck(I.getOperand(2), &I);
    IRBuilder<> IRB(&I);
    setShadow(&I, IRB.CreateInsertElement(getShadow(&I, 0), getShadow(&I, 1),
              I.getOperand(2), "_msprop"));
    setOriginForNaryOp(I);
  }

  void visitShuffleVectorInst(ShuffleVectorInst &I) {
    insertShadowCheck(I.getOperand(2), &I);
    IRBuilder<> IRB(&I);
    setShadow(&I, IRB.CreateShuffleVector(getShadow(&I, 0), getShadow(&I, 1),
              I.getOperand(2), "_msprop"));
    setOriginForNaryOp(I);
  }

  // Casts.
  void visitSExtInst(SExtInst &I) {
    IRBuilder<> IRB(&I);
    setShadow(&I, IRB.CreateSExt(getShadow(&I, 0), I.getType(), "_msprop"));
    setOrigin(&I, getOrigin(&I, 0));
  }

  void visitZExtInst(ZExtInst &I) {
    IRBuilder<> IRB(&I);
    setShadow(&I, IRB.CreateZExt(getShadow(&I, 0), I.getType(), "_msprop"));
    setOrigin(&I, getOrigin(&I, 0));
  }

  void visitTruncInst(TruncInst &I) {
    IRBuilder<> IRB(&I);
    setShadow(&I, IRB.CreateTrunc(getShadow(&I, 0), I.getType(), "_msprop"));
    setOrigin(&I, getOrigin(&I, 0));
  }

  void visitBitCastInst(BitCastInst &I) {
    IRBuilder<> IRB(&I);
    setShadow(&I, IRB.CreateBitCast(getShadow(&I, 0), getShadowTy(&I)));
    setOrigin(&I, getOrigin(&I, 0));
  }

  void visitPtrToIntInst(PtrToIntInst &I) {
    IRBuilder<> IRB(&I);
    setShadow(&I, IRB.CreateIntCast(getShadow(&I, 0), getShadowTy(&I), false,
             "_msprop_ptrtoint"));
    setOrigin(&I, getOrigin(&I, 0));
  }

  void visitIntToPtrInst(IntToPtrInst &I) {
    IRBuilder<> IRB(&I);
    setShadow(&I, IRB.CreateIntCast(getShadow(&I, 0), getShadowTy(&I), false,
             "_msprop_inttoptr"));
    setOrigin(&I, getOrigin(&I, 0));
  }

  void visitFPToSIInst(CastInst& I) { handleShadowOr(I); }
  void visitFPToUIInst(CastInst& I) { handleShadowOr(I); }
  void visitSIToFPInst(CastInst& I) { handleShadowOr(I); }
  void visitUIToFPInst(CastInst& I) { handleShadowOr(I); }
  void visitFPExtInst(CastInst& I) { handleShadowOr(I); }
  void visitFPTruncInst(CastInst& I) { handleShadowOr(I); }

  /// \brief Propagate shadow for bitwise AND.
  ///
  /// This code is exact, i.e. if, for example, a bit in the left argument
  /// is defined and 0, then neither the value not definedness of the
  /// corresponding bit in B don't affect the resulting shadow.
  void visitAnd(BinaryOperator &I) {
    IRBuilder<> IRB(&I);
    //  "And" of 0 and a poisoned value results in unpoisoned value.
    //  1&1 => 1;     0&1 => 0;     p&1 => p;
    //  1&0 => 0;     0&0 => 0;     p&0 => 0;
    //  1&p => p;     0&p => 0;     p&p => p;
    //  S = (S1 & S2) | (V1 & S2) | (S1 & V2)
    Value *S1 = getShadow(&I, 0);
    Value *S2 = getShadow(&I, 1);
    Value *V1 = I.getOperand(0);
    Value *V2 = I.getOperand(1);
    if (V1->getType() != S1->getType()) {
      V1 = IRB.CreateIntCast(V1, S1->getType(), false);
      V2 = IRB.CreateIntCast(V2, S2->getType(), false);
    }
    Value *S1S2 = IRB.CreateAnd(S1, S2);
    Value *V1S2 = IRB.CreateAnd(V1, S2);
    Value *S1V2 = IRB.CreateAnd(S1, V2);
    setShadow(&I, IRB.CreateOr(S1S2, IRB.CreateOr(V1S2, S1V2)));
    setOriginForNaryOp(I);
  }

  void visitOr(BinaryOperator &I) {
    IRBuilder<> IRB(&I);
    //  "Or" of 1 and a poisoned value results in unpoisoned value.
    //  1|1 => 1;     0|1 => 1;     p|1 => 1;
    //  1|0 => 1;     0|0 => 0;     p|0 => p;
    //  1|p => 1;     0|p => p;     p|p => p;
    //  S = (S1 & S2) | (~V1 & S2) | (S1 & ~V2)
    Value *S1 = getShadow(&I, 0);
    Value *S2 = getShadow(&I, 1);
    Value *V1 = IRB.CreateNot(I.getOperand(0));
    Value *V2 = IRB.CreateNot(I.getOperand(1));
    if (V1->getType() != S1->getType()) {
      V1 = IRB.CreateIntCast(V1, S1->getType(), false);
      V2 = IRB.CreateIntCast(V2, S2->getType(), false);
    }
    Value *S1S2 = IRB.CreateAnd(S1, S2);
    Value *V1S2 = IRB.CreateAnd(V1, S2);
    Value *S1V2 = IRB.CreateAnd(S1, V2);
    setShadow(&I, IRB.CreateOr(S1S2, IRB.CreateOr(V1S2, S1V2)));
    setOriginForNaryOp(I);
  }

  /// \brief Default propagation of shadow and/or origin.
  ///
  /// This class implements the general case of shadow propagation, used in all
  /// cases where we don't know and/or don't care about what the operation
  /// actually does. It converts all input shadow values to a common type
  /// (extending or truncating as necessary), and bitwise OR's them.
  ///
  /// This is much cheaper than inserting checks (i.e. requiring inputs to be
  /// fully initialized), and less prone to false positives.
  ///
  /// This class also implements the general case of origin propagation. For a
  /// Nary operation, result origin is set to the origin of an argument that is
  /// not entirely initialized. If there is more than one such arguments, the
  /// rightmost of them is picked. It does not matter which one is picked if all
  /// arguments are initialized.
  template <bool CombineShadow>
  class Combiner {
    Value *Shadow;
    Value *Origin;
    IRBuilder<> &IRB;
    MemorySanitizerVisitor *MSV;

  public:
    Combiner(MemorySanitizerVisitor *MSV, IRBuilder<> &IRB) :
      Shadow(nullptr), Origin(nullptr), IRB(IRB), MSV(MSV) {}

    /// \brief Add a pair of shadow and origin values to the mix.
    Combiner &Add(Value *OpShadow, Value *OpOrigin) {
      if (CombineShadow) {
        assert(OpShadow);
        if (!Shadow)
          Shadow = OpShadow;
        else {
          OpShadow = MSV->CreateShadowCast(IRB, OpShadow, Shadow->getType());
          Shadow = IRB.CreateOr(Shadow, OpShadow, "_msprop");
        }
      }

      if (MSV->MS.TrackOrigins) {
        assert(OpOrigin);
        if (!Origin) {
          Origin = OpOrigin;
        } else {
          Constant *ConstOrigin = dyn_cast<Constant>(OpOrigin);
          // No point in adding something that might result in 0 origin value.
          if (!ConstOrigin || !ConstOrigin->isNullValue()) {
            Value *FlatShadow = MSV->convertToShadowTyNoVec(OpShadow, IRB);
            Value *Cond =
                IRB.CreateICmpNE(FlatShadow, MSV->getCleanShadow(FlatShadow));
            Origin = IRB.CreateSelect(Cond, OpOrigin, Origin);
          }
        }
      }
      return *this;
    }

    /// \brief Add an application value to the mix.
    Combiner &Add(Value *V) {
      Value *OpShadow = MSV->getShadow(V);
      Value *OpOrigin = MSV->MS.TrackOrigins ? MSV->getOrigin(V) : nullptr;
      return Add(OpShadow, OpOrigin);
    }

    /// \brief Set the current combined values as the given instruction's shadow
    /// and origin.
    void Done(Instruction *I) {
      if (CombineShadow) {
        assert(Shadow);
        Shadow = MSV->CreateShadowCast(IRB, Shadow, MSV->getShadowTy(I));
        MSV->setShadow(I, Shadow);
      }
      if (MSV->MS.TrackOrigins) {
        assert(Origin);
        MSV->setOrigin(I, Origin);
      }
    }
  };

  typedef Combiner<true> ShadowAndOriginCombiner;
  typedef Combiner<false> OriginCombiner;

  /// \brief Propagate origin for arbitrary operation.
  void setOriginForNaryOp(Instruction &I) {
    if (!MS.TrackOrigins) return;
    IRBuilder<> IRB(&I);
    OriginCombiner OC(this, IRB);
    for (Instruction::op_iterator OI = I.op_begin(); OI != I.op_end(); ++OI)
      OC.Add(OI->get());
    OC.Done(&I);
  }

  size_t VectorOrPrimitiveTypeSizeInBits(Type *Ty) {
    assert(!(Ty->isVectorTy() && Ty->getScalarType()->isPointerTy()) &&
           "Vector of pointers is not a valid shadow type");
    return Ty->isVectorTy() ?
      Ty->getVectorNumElements() * Ty->getScalarSizeInBits() :
      Ty->getPrimitiveSizeInBits();
  }

  /// \brief Cast between two shadow types, extending or truncating as
  /// necessary.
  Value *CreateShadowCast(IRBuilder<> &IRB, Value *V, Type *dstTy,
                          bool Signed = false) {
    Type *srcTy = V->getType();
    if (dstTy->isIntegerTy() && srcTy->isIntegerTy())
      return IRB.CreateIntCast(V, dstTy, Signed);
    if (dstTy->isVectorTy() && srcTy->isVectorTy() &&
        dstTy->getVectorNumElements() == srcTy->getVectorNumElements())
      return IRB.CreateIntCast(V, dstTy, Signed);
    size_t srcSizeInBits = VectorOrPrimitiveTypeSizeInBits(srcTy);
    size_t dstSizeInBits = VectorOrPrimitiveTypeSizeInBits(dstTy);
    Value *V1 = IRB.CreateBitCast(V, Type::getIntNTy(*MS.C, srcSizeInBits));
    Value *V2 =
      IRB.CreateIntCast(V1, Type::getIntNTy(*MS.C, dstSizeInBits), Signed);
    return IRB.CreateBitCast(V2, dstTy);
    // TODO: handle struct types.
  }

  /// \brief Cast an application value to the type of its own shadow.
  Value *CreateAppToShadowCast(IRBuilder<> &IRB, Value *V) {
    Type *ShadowTy = getShadowTy(V);
    if (V->getType() == ShadowTy)
      return V;
    if (V->getType()->isPtrOrPtrVectorTy())
      return IRB.CreatePtrToInt(V, ShadowTy);
    else
      return IRB.CreateBitCast(V, ShadowTy);
  }

  /// \brief Propagate shadow for arbitrary operation.
  void handleShadowOr(Instruction &I) {
    IRBuilder<> IRB(&I);
    ShadowAndOriginCombiner SC(this, IRB);
    for (Instruction::op_iterator OI = I.op_begin(); OI != I.op_end(); ++OI)
      SC.Add(OI->get());
    SC.Done(&I);
  }

  // \brief Handle multiplication by constant.
  //
  // Handle a special case of multiplication by constant that may have one or
  // more zeros in the lower bits. This makes corresponding number of lower bits
  // of the result zero as well. We model it by shifting the other operand
  // shadow left by the required number of bits. Effectively, we transform
  // (X * (A * 2**B)) to ((X << B) * A) and instrument (X << B) as (Sx << B).
  // We use multiplication by 2**N instead of shift to cover the case of
  // multiplication by 0, which may occur in some elements of a vector operand.
  void handleMulByConstant(BinaryOperator &I, Constant *ConstArg,
                           Value *OtherArg) {
    Constant *ShadowMul;
    Type *Ty = ConstArg->getType();
    if (Ty->isVectorTy()) {
      unsigned NumElements = Ty->getVectorNumElements();
      Type *EltTy = Ty->getSequentialElementType();
      SmallVector<Constant *, 16> Elements;
      for (unsigned Idx = 0; Idx < NumElements; ++Idx) {
        ConstantInt *Elt =
            dyn_cast<ConstantInt>(ConstArg->getAggregateElement(Idx));
        APInt V = Elt->getValue();
        APInt V2 = APInt(V.getBitWidth(), 1) << V.countTrailingZeros();
        Elements.push_back(ConstantInt::get(EltTy, V2));
      }
      ShadowMul = ConstantVector::get(Elements);
    } else {
      ConstantInt *Elt = dyn_cast<ConstantInt>(ConstArg);
      APInt V = Elt->getValue();
      APInt V2 = APInt(V.getBitWidth(), 1) << V.countTrailingZeros();
      ShadowMul = ConstantInt::get(Elt->getType(), V2);
    }

    IRBuilder<> IRB(&I);
    setShadow(&I,
              IRB.CreateMul(getShadow(OtherArg), ShadowMul, "msprop_mul_cst"));
    setOrigin(&I, getOrigin(OtherArg));
  }

  void visitMul(BinaryOperator &I) {
    Constant *constOp0 = dyn_cast<Constant>(I.getOperand(0));
    Constant *constOp1 = dyn_cast<Constant>(I.getOperand(1));
    if (constOp0 && !constOp1)
      handleMulByConstant(I, constOp0, I.getOperand(1));
    else if (constOp1 && !constOp0)
      handleMulByConstant(I, constOp1, I.getOperand(0));
    else
      handleShadowOr(I);
  }

  void visitFAdd(BinaryOperator &I) { handleShadowOr(I); }
  void visitFSub(BinaryOperator &I) { handleShadowOr(I); }
  void visitFMul(BinaryOperator &I) { handleShadowOr(I); }
  void visitAdd(BinaryOperator &I) { handleShadowOr(I); }
  void visitSub(BinaryOperator &I) { handleShadowOr(I); }
  void visitXor(BinaryOperator &I) { handleShadowOr(I); }

  void handleDiv(Instruction &I) {
    IRBuilder<> IRB(&I);
    // Strict on the second argument.
    insertShadowCheck(I.getOperand(1), &I);
    setShadow(&I, getShadow(&I, 0));
    setOrigin(&I, getOrigin(&I, 0));
  }

  void visitUDiv(BinaryOperator &I) { handleDiv(I); }
  void visitSDiv(BinaryOperator &I) { handleDiv(I); }
  void visitFDiv(BinaryOperator &I) { handleDiv(I); }
  void visitURem(BinaryOperator &I) { handleDiv(I); }
  void visitSRem(BinaryOperator &I) { handleDiv(I); }
  void visitFRem(BinaryOperator &I) { handleDiv(I); }

  /// \brief Instrument == and != comparisons.
  ///
  /// Sometimes the comparison result is known even if some of the bits of the
  /// arguments are not.
  void handleEqualityComparison(ICmpInst &I) {
    IRBuilder<> IRB(&I);
    Value *A = I.getOperand(0);
    Value *B = I.getOperand(1);
    Value *Sa = getShadow(A);
    Value *Sb = getShadow(B);

    // Get rid of pointers and vectors of pointers.
    // For ints (and vectors of ints), types of A and Sa match,
    // and this is a no-op.
    A = IRB.CreatePointerCast(A, Sa->getType());
    B = IRB.CreatePointerCast(B, Sb->getType());

    // A == B  <==>  (C = A^B) == 0
    // A != B  <==>  (C = A^B) != 0
    // Sc = Sa | Sb
    Value *C = IRB.CreateXor(A, B);
    Value *Sc = IRB.CreateOr(Sa, Sb);
    // Now dealing with i = (C == 0) comparison (or C != 0, does not matter now)
    // Result is defined if one of the following is true
    // * there is a defined 1 bit in C
    // * C is fully defined
    // Si = !(C & ~Sc) && Sc
    Value *Zero = Constant::getNullValue(Sc->getType());
    Value *MinusOne = Constant::getAllOnesValue(Sc->getType());
    Value *Si =
      IRB.CreateAnd(IRB.CreateICmpNE(Sc, Zero),
                    IRB.CreateICmpEQ(
                      IRB.CreateAnd(IRB.CreateXor(Sc, MinusOne), C), Zero));
    Si->setName("_msprop_icmp");
    setShadow(&I, Si);
    setOriginForNaryOp(I);
  }

  /// \brief Build the lowest possible value of V, taking into account V's
  ///        uninitialized bits.
  Value *getLowestPossibleValue(IRBuilder<> &IRB, Value *A, Value *Sa,
                                bool isSigned) {
    if (isSigned) {
      // Split shadow into sign bit and other bits.
      Value *SaOtherBits = IRB.CreateLShr(IRB.CreateShl(Sa, 1), 1);
      Value *SaSignBit = IRB.CreateXor(Sa, SaOtherBits);
      // Maximise the undefined shadow bit, minimize other undefined bits.
      return
        IRB.CreateOr(IRB.CreateAnd(A, IRB.CreateNot(SaOtherBits)), SaSignBit);
    } else {
      // Minimize undefined bits.
      return IRB.CreateAnd(A, IRB.CreateNot(Sa));
    }
  }

  /// \brief Build the highest possible value of V, taking into account V's
  ///        uninitialized bits.
  Value *getHighestPossibleValue(IRBuilder<> &IRB, Value *A, Value *Sa,
                                bool isSigned) {
    if (isSigned) {
      // Split shadow into sign bit and other bits.
      Value *SaOtherBits = IRB.CreateLShr(IRB.CreateShl(Sa, 1), 1);
      Value *SaSignBit = IRB.CreateXor(Sa, SaOtherBits);
      // Minimise the undefined shadow bit, maximise other undefined bits.
      return
        IRB.CreateOr(IRB.CreateAnd(A, IRB.CreateNot(SaSignBit)), SaOtherBits);
    } else {
      // Maximize undefined bits.
      return IRB.CreateOr(A, Sa);
    }
  }

  /// \brief Instrument relational comparisons.
  ///
  /// This function does exact shadow propagation for all relational
  /// comparisons of integers, pointers and vectors of those.
  /// FIXME: output seems suboptimal when one of the operands is a constant
  void handleRelationalComparisonExact(ICmpInst &I) {
    IRBuilder<> IRB(&I);
    Value *A = I.getOperand(0);
    Value *B = I.getOperand(1);
    Value *Sa = getShadow(A);
    Value *Sb = getShadow(B);

    // Get rid of pointers and vectors of pointers.
    // For ints (and vectors of ints), types of A and Sa match,
    // and this is a no-op.
    A = IRB.CreatePointerCast(A, Sa->getType());
    B = IRB.CreatePointerCast(B, Sb->getType());

    // Let [a0, a1] be the interval of possible values of A, taking into account
    // its undefined bits. Let [b0, b1] be the interval of possible values of B.
    // Then (A cmp B) is defined iff (a0 cmp b1) == (a1 cmp b0).
    bool IsSigned = I.isSigned();
    Value *S1 = IRB.CreateICmp(I.getPredicate(),
                               getLowestPossibleValue(IRB, A, Sa, IsSigned),
                               getHighestPossibleValue(IRB, B, Sb, IsSigned));
    Value *S2 = IRB.CreateICmp(I.getPredicate(),
                               getHighestPossibleValue(IRB, A, Sa, IsSigned),
                               getLowestPossibleValue(IRB, B, Sb, IsSigned));
    Value *Si = IRB.CreateXor(S1, S2);
    setShadow(&I, Si);
    setOriginForNaryOp(I);
  }

  /// \brief Instrument signed relational comparisons.
  ///
  /// Handle (x<0) and (x>=0) comparisons (essentially, sign bit tests) by
  /// propagating the highest bit of the shadow. Everything else is delegated
  /// to handleShadowOr().
  void handleSignedRelationalComparison(ICmpInst &I) {
    Constant *constOp0 = dyn_cast<Constant>(I.getOperand(0));
    Constant *constOp1 = dyn_cast<Constant>(I.getOperand(1));
    Value* op = nullptr;
    CmpInst::Predicate pre = I.getPredicate();
    if (constOp0 && constOp0->isNullValue() &&
        (pre == CmpInst::ICMP_SGT || pre == CmpInst::ICMP_SLE)) {
      op = I.getOperand(1);
    } else if (constOp1 && constOp1->isNullValue() &&
               (pre == CmpInst::ICMP_SLT || pre == CmpInst::ICMP_SGE)) {
      op = I.getOperand(0);
    }
    if (op) {
      IRBuilder<> IRB(&I);
      Value* Shadow =
        IRB.CreateICmpSLT(getShadow(op), getCleanShadow(op), "_msprop_icmpslt");
      setShadow(&I, Shadow);
      setOrigin(&I, getOrigin(op));
    } else {
      handleShadowOr(I);
    }
  }

  void visitICmpInst(ICmpInst &I) {
    if (!ClHandleICmp) {
      handleShadowOr(I);
      return;
    }
    if (I.isEquality()) {
      handleEqualityComparison(I);
      return;
    }

    assert(I.isRelational());
    if (ClHandleICmpExact) {
      handleRelationalComparisonExact(I);
      return;
    }
    if (I.isSigned()) {
      handleSignedRelationalComparison(I);
      return;
    }

    assert(I.isUnsigned());
    if ((isa<Constant>(I.getOperand(0)) || isa<Constant>(I.getOperand(1)))) {
      handleRelationalComparisonExact(I);
      return;
    }

    handleShadowOr(I);
  }

  void visitFCmpInst(FCmpInst &I) {
    handleShadowOr(I);
  }

  void handleShift(BinaryOperator &I) {
    IRBuilder<> IRB(&I);
    // If any of the S2 bits are poisoned, the whole thing is poisoned.
    // Otherwise perform the same shift on S1.
    Value *S1 = getShadow(&I, 0);
    Value *S2 = getShadow(&I, 1);
    Value *S2Conv = IRB.CreateSExt(IRB.CreateICmpNE(S2, getCleanShadow(S2)),
                                   S2->getType());
    Value *V2 = I.getOperand(1);
    Value *Shift = IRB.CreateBinOp(I.getOpcode(), S1, V2);
    setShadow(&I, IRB.CreateOr(Shift, S2Conv));
    setOriginForNaryOp(I);
  }

  void visitShl(BinaryOperator &I) { handleShift(I); }
  void visitAShr(BinaryOperator &I) { handleShift(I); }
  void visitLShr(BinaryOperator &I) { handleShift(I); }

  /// \brief Instrument llvm.memmove
  ///
  /// At this point we don't know if llvm.memmove will be inlined or not.
  /// If we don't instrument it and it gets inlined,
  /// our interceptor will not kick in and we will lose the memmove.
  /// If we instrument the call here, but it does not get inlined,
  /// we will memove the shadow twice: which is bad in case
  /// of overlapping regions. So, we simply lower the intrinsic to a call.
  ///
  /// Similar situation exists for memcpy and memset.
  void visitMemMoveInst(MemMoveInst &I) {
    IRBuilder<> IRB(&I);
    IRB.CreateCall3(
      MS.MemmoveFn,
      IRB.CreatePointerCast(I.getArgOperand(0), IRB.getInt8PtrTy()),
      IRB.CreatePointerCast(I.getArgOperand(1), IRB.getInt8PtrTy()),
      IRB.CreateIntCast(I.getArgOperand(2), MS.IntptrTy, false));
    I.eraseFromParent();
  }

  // Similar to memmove: avoid copying shadow twice.
  // This is somewhat unfortunate as it may slowdown small constant memcpys.
  // FIXME: consider doing manual inline for small constant sizes and proper
  // alignment.
  void visitMemCpyInst(MemCpyInst &I) {
    IRBuilder<> IRB(&I);
    IRB.CreateCall3(
      MS.MemcpyFn,
      IRB.CreatePointerCast(I.getArgOperand(0), IRB.getInt8PtrTy()),
      IRB.CreatePointerCast(I.getArgOperand(1), IRB.getInt8PtrTy()),
      IRB.CreateIntCast(I.getArgOperand(2), MS.IntptrTy, false));
    I.eraseFromParent();
  }

  // Same as memcpy.
  void visitMemSetInst(MemSetInst &I) {
    IRBuilder<> IRB(&I);
    IRB.CreateCall3(
      MS.MemsetFn,
      IRB.CreatePointerCast(I.getArgOperand(0), IRB.getInt8PtrTy()),
      IRB.CreateIntCast(I.getArgOperand(1), IRB.getInt32Ty(), false),
      IRB.CreateIntCast(I.getArgOperand(2), MS.IntptrTy, false));
    I.eraseFromParent();
  }

  void visitVAStartInst(VAStartInst &I) {
    VAHelper->visitVAStartInst(I);
  }

  void visitVACopyInst(VACopyInst &I) {
    VAHelper->visitVACopyInst(I);
  }

  enum IntrinsicKind {
    IK_DoesNotAccessMemory,
    IK_OnlyReadsMemory,
    IK_WritesMemory
  };

  static IntrinsicKind getIntrinsicKind(Intrinsic::ID iid) {
    const int DoesNotAccessMemory = IK_DoesNotAccessMemory;
    const int OnlyReadsArgumentPointees = IK_OnlyReadsMemory;
    const int OnlyReadsMemory = IK_OnlyReadsMemory;
    const int OnlyAccessesArgumentPointees = IK_WritesMemory;
    const int UnknownModRefBehavior = IK_WritesMemory;
#define GET_INTRINSIC_MODREF_BEHAVIOR
#define ModRefBehavior IntrinsicKind
#include "llvm/IR/Intrinsics.gen"
#undef ModRefBehavior
#undef GET_INTRINSIC_MODREF_BEHAVIOR
  }

  /// \brief Handle vector store-like intrinsics.
  ///
  /// Instrument intrinsics that look like a simple SIMD store: writes memory,
  /// has 1 pointer argument and 1 vector argument, returns void.
  bool handleVectorStoreIntrinsic(IntrinsicInst &I) {
    IRBuilder<> IRB(&I);
    Value* Addr = I.getArgOperand(0);
    Value *Shadow = getShadow(&I, 1);
    Value *ShadowPtr = getShadowPtr(Addr, Shadow->getType(), IRB);

    // We don't know the pointer alignment (could be unaligned SSE store!).
    // Have to assume to worst case.
    IRB.CreateAlignedStore(Shadow, ShadowPtr, 1);

    if (ClCheckAccessAddress)
      insertShadowCheck(Addr, &I);

    // FIXME: use ClStoreCleanOrigin
    // FIXME: factor out common code from materializeStores
    if (MS.TrackOrigins)
      IRB.CreateStore(getOrigin(&I, 1), getOriginPtr(Addr, IRB, 1));
    return true;
  }

  /// \brief Handle vector load-like intrinsics.
  ///
  /// Instrument intrinsics that look like a simple SIMD load: reads memory,
  /// has 1 pointer argument, returns a vector.
  bool handleVectorLoadIntrinsic(IntrinsicInst &I) {
    IRBuilder<> IRB(&I);
    Value *Addr = I.getArgOperand(0);

    Type *ShadowTy = getShadowTy(&I);
    if (PropagateShadow) {
      Value *ShadowPtr = getShadowPtr(Addr, ShadowTy, IRB);
      // We don't know the pointer alignment (could be unaligned SSE load!).
      // Have to assume to worst case.
      setShadow(&I, IRB.CreateAlignedLoad(ShadowPtr, 1, "_msld"));
    } else {
      setShadow(&I, getCleanShadow(&I));
    }

    if (ClCheckAccessAddress)
      insertShadowCheck(Addr, &I);

    if (MS.TrackOrigins) {
      if (PropagateShadow)
        setOrigin(&I, IRB.CreateLoad(getOriginPtr(Addr, IRB, 1)));
      else
        setOrigin(&I, getCleanOrigin());
    }
    return true;
  }

  /// \brief Handle (SIMD arithmetic)-like intrinsics.
  ///
  /// Instrument intrinsics with any number of arguments of the same type,
  /// equal to the return type. The type should be simple (no aggregates or
  /// pointers; vectors are fine).
  /// Caller guarantees that this intrinsic does not access memory.
  bool maybeHandleSimpleNomemIntrinsic(IntrinsicInst &I) {
    Type *RetTy = I.getType();
    if (!(RetTy->isIntOrIntVectorTy() ||
          RetTy->isFPOrFPVectorTy() ||
          RetTy->isX86_MMXTy()))
      return false;

    unsigned NumArgOperands = I.getNumArgOperands();

    for (unsigned i = 0; i < NumArgOperands; ++i) {
      Type *Ty = I.getArgOperand(i)->getType();
      if (Ty != RetTy)
        return false;
    }

    IRBuilder<> IRB(&I);
    ShadowAndOriginCombiner SC(this, IRB);
    for (unsigned i = 0; i < NumArgOperands; ++i)
      SC.Add(I.getArgOperand(i));
    SC.Done(&I);

    return true;
  }

  /// \brief Heuristically instrument unknown intrinsics.
  ///
  /// The main purpose of this code is to do something reasonable with all
  /// random intrinsics we might encounter, most importantly - SIMD intrinsics.
  /// We recognize several classes of intrinsics by their argument types and
  /// ModRefBehaviour and apply special intrumentation when we are reasonably
  /// sure that we know what the intrinsic does.
  ///
  /// We special-case intrinsics where this approach fails. See llvm.bswap
  /// handling as an example of that.
  bool handleUnknownIntrinsic(IntrinsicInst &I) {
    unsigned NumArgOperands = I.getNumArgOperands();
    if (NumArgOperands == 0)
      return false;

    Intrinsic::ID iid = I.getIntrinsicID();
    IntrinsicKind IK = getIntrinsicKind(iid);
    bool OnlyReadsMemory = IK == IK_OnlyReadsMemory;
    bool WritesMemory = IK == IK_WritesMemory;
    assert(!(OnlyReadsMemory && WritesMemory));

    if (NumArgOperands == 2 &&
        I.getArgOperand(0)->getType()->isPointerTy() &&
        I.getArgOperand(1)->getType()->isVectorTy() &&
        I.getType()->isVoidTy() &&
        WritesMemory) {
      // This looks like a vector store.
      return handleVectorStoreIntrinsic(I);
    }

    if (NumArgOperands == 1 &&
        I.getArgOperand(0)->getType()->isPointerTy() &&
        I.getType()->isVectorTy() &&
        OnlyReadsMemory) {
      // This looks like a vector load.
      return handleVectorLoadIntrinsic(I);
    }

    if (!OnlyReadsMemory && !WritesMemory)
      if (maybeHandleSimpleNomemIntrinsic(I))
        return true;

    // FIXME: detect and handle SSE maskstore/maskload
    return false;
  }

  void handleBswap(IntrinsicInst &I) {
    IRBuilder<> IRB(&I);
    Value *Op = I.getArgOperand(0);
    Type *OpType = Op->getType();
    Function *BswapFunc = Intrinsic::getDeclaration(
      F.getParent(), Intrinsic::bswap, makeArrayRef(&OpType, 1));
    setShadow(&I, IRB.CreateCall(BswapFunc, getShadow(Op)));
    setOrigin(&I, getOrigin(Op));
  }

  // \brief Instrument vector convert instrinsic.
  //
  // This function instruments intrinsics like cvtsi2ss:
  // %Out = int_xxx_cvtyyy(%ConvertOp)
  // or
  // %Out = int_xxx_cvtyyy(%CopyOp, %ConvertOp)
  // Intrinsic converts \p NumUsedElements elements of \p ConvertOp to the same
  // number \p Out elements, and (if has 2 arguments) copies the rest of the
  // elements from \p CopyOp.
  // In most cases conversion involves floating-point value which may trigger a
  // hardware exception when not fully initialized. For this reason we require
  // \p ConvertOp[0:NumUsedElements] to be fully initialized and trap otherwise.
  // We copy the shadow of \p CopyOp[NumUsedElements:] to \p
  // Out[NumUsedElements:]. This means that intrinsics without \p CopyOp always
  // return a fully initialized value.
  void handleVectorConvertIntrinsic(IntrinsicInst &I, int NumUsedElements) {
    IRBuilder<> IRB(&I);
    Value *CopyOp, *ConvertOp;

    switch (I.getNumArgOperands()) {
    case 2:
      CopyOp = I.getArgOperand(0);
      ConvertOp = I.getArgOperand(1);
      break;
    case 1:
      ConvertOp = I.getArgOperand(0);
      CopyOp = nullptr;
      break;
    default:
      llvm_unreachable("Cvt intrinsic with unsupported number of arguments.");
    }

    // The first *NumUsedElements* elements of ConvertOp are converted to the
    // same number of output elements. The rest of the output is copied from
    // CopyOp, or (if not available) filled with zeroes.
    // Combine shadow for elements of ConvertOp that are used in this operation,
    // and insert a check.
    // FIXME: consider propagating shadow of ConvertOp, at least in the case of
    // int->any conversion.
    Value *ConvertShadow = getShadow(ConvertOp);
    Value *AggShadow = nullptr;
    if (ConvertOp->getType()->isVectorTy()) {
      AggShadow = IRB.CreateExtractElement(
          ConvertShadow, ConstantInt::get(IRB.getInt32Ty(), 0));
      for (int i = 1; i < NumUsedElements; ++i) {
        Value *MoreShadow = IRB.CreateExtractElement(
            ConvertShadow, ConstantInt::get(IRB.getInt32Ty(), i));
        AggShadow = IRB.CreateOr(AggShadow, MoreShadow);
      }
    } else {
      AggShadow = ConvertShadow;
    }
    assert(AggShadow->getType()->isIntegerTy());
    insertShadowCheck(AggShadow, getOrigin(ConvertOp), &I);

    // Build result shadow by zero-filling parts of CopyOp shadow that come from
    // ConvertOp.
    if (CopyOp) {
      assert(CopyOp->getType() == I.getType());
      assert(CopyOp->getType()->isVectorTy());
      Value *ResultShadow = getShadow(CopyOp);
      Type *EltTy = ResultShadow->getType()->getVectorElementType();
      for (int i = 0; i < NumUsedElements; ++i) {
        ResultShadow = IRB.CreateInsertElement(
            ResultShadow, ConstantInt::getNullValue(EltTy),
            ConstantInt::get(IRB.getInt32Ty(), i));
      }
      setShadow(&I, ResultShadow);
      setOrigin(&I, getOrigin(CopyOp));
    } else {
      setShadow(&I, getCleanShadow(&I));
      setOrigin(&I, getCleanOrigin());
    }
  }

  // Given a scalar or vector, extract lower 64 bits (or less), and return all
  // zeroes if it is zero, and all ones otherwise.
  Value *Lower64ShadowExtend(IRBuilder<> &IRB, Value *S, Type *T) {
    if (S->getType()->isVectorTy())
      S = CreateShadowCast(IRB, S, IRB.getInt64Ty(), /* Signed */ true);
    assert(S->getType()->getPrimitiveSizeInBits() <= 64);
    Value *S2 = IRB.CreateICmpNE(S, getCleanShadow(S));
    return CreateShadowCast(IRB, S2, T, /* Signed */ true);
  }

  Value *VariableShadowExtend(IRBuilder<> &IRB, Value *S) {
    Type *T = S->getType();
    assert(T->isVectorTy());
    Value *S2 = IRB.CreateICmpNE(S, getCleanShadow(S));
    return IRB.CreateSExt(S2, T);
  }

  // \brief Instrument vector shift instrinsic.
  //
  // This function instruments intrinsics like int_x86_avx2_psll_w.
  // Intrinsic shifts %In by %ShiftSize bits.
  // %ShiftSize may be a vector. In that case the lower 64 bits determine shift
  // size, and the rest is ignored. Behavior is defined even if shift size is
  // greater than register (or field) width.
  void handleVectorShiftIntrinsic(IntrinsicInst &I, bool Variable) {
    assert(I.getNumArgOperands() == 2);
    IRBuilder<> IRB(&I);
    // If any of the S2 bits are poisoned, the whole thing is poisoned.
    // Otherwise perform the same shift on S1.
    Value *S1 = getShadow(&I, 0);
    Value *S2 = getShadow(&I, 1);
    Value *S2Conv = Variable ? VariableShadowExtend(IRB, S2)
                             : Lower64ShadowExtend(IRB, S2, getShadowTy(&I));
    Value *V1 = I.getOperand(0);
    Value *V2 = I.getOperand(1);
    Value *Shift = IRB.CreateCall2(I.getCalledValue(),
                                   IRB.CreateBitCast(S1, V1->getType()), V2);
    Shift = IRB.CreateBitCast(Shift, getShadowTy(&I));
    setShadow(&I, IRB.CreateOr(Shift, S2Conv));
    setOriginForNaryOp(I);
  }

  // \brief Get an X86_MMX-sized vector type.
  Type *getMMXVectorTy(unsigned EltSizeInBits) {
    const unsigned X86_MMXSizeInBits = 64;
    return VectorType::get(IntegerType::get(*MS.C, EltSizeInBits),
                           X86_MMXSizeInBits / EltSizeInBits);
  }

  // \brief Returns a signed counterpart for an (un)signed-saturate-and-pack
  // intrinsic.
  Intrinsic::ID getSignedPackIntrinsic(Intrinsic::ID id) {
    switch (id) {
      case llvm::Intrinsic::x86_sse2_packsswb_128:
      case llvm::Intrinsic::x86_sse2_packuswb_128:
        return llvm::Intrinsic::x86_sse2_packsswb_128;

      case llvm::Intrinsic::x86_sse2_packssdw_128:
      case llvm::Intrinsic::x86_sse41_packusdw:
        return llvm::Intrinsic::x86_sse2_packssdw_128;

      case llvm::Intrinsic::x86_avx2_packsswb:
      case llvm::Intrinsic::x86_avx2_packuswb:
        return llvm::Intrinsic::x86_avx2_packsswb;

      case llvm::Intrinsic::x86_avx2_packssdw:
      case llvm::Intrinsic::x86_avx2_packusdw:
        return llvm::Intrinsic::x86_avx2_packssdw;

      case llvm::Intrinsic::x86_mmx_packsswb:
      case llvm::Intrinsic::x86_mmx_packuswb:
        return llvm::Intrinsic::x86_mmx_packsswb;

      case llvm::Intrinsic::x86_mmx_packssdw:
        return llvm::Intrinsic::x86_mmx_packssdw;
      default:
        llvm_unreachable("unexpected intrinsic id");
    }
  }

  // \brief Instrument vector pack instrinsic.
  //
  // This function instruments intrinsics like x86_mmx_packsswb, that
  // packs elements of 2 input vectors into half as many bits with saturation.
  // Shadow is propagated with the signed variant of the same intrinsic applied
  // to sext(Sa != zeroinitializer), sext(Sb != zeroinitializer).
  // EltSizeInBits is used only for x86mmx arguments.
  void handleVectorPackIntrinsic(IntrinsicInst &I, unsigned EltSizeInBits = 0) {
    assert(I.getNumArgOperands() == 2);
    bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
    IRBuilder<> IRB(&I);
    Value *S1 = getShadow(&I, 0);
    Value *S2 = getShadow(&I, 1);
    assert(isX86_MMX || S1->getType()->isVectorTy());

    // SExt and ICmpNE below must apply to individual elements of input vectors.
    // In case of x86mmx arguments, cast them to appropriate vector types and
    // back.
    Type *T = isX86_MMX ? getMMXVectorTy(EltSizeInBits) : S1->getType();
    if (isX86_MMX) {
      S1 = IRB.CreateBitCast(S1, T);
      S2 = IRB.CreateBitCast(S2, T);
    }
    Value *S1_ext = IRB.CreateSExt(
        IRB.CreateICmpNE(S1, llvm::Constant::getNullValue(T)), T);
    Value *S2_ext = IRB.CreateSExt(
        IRB.CreateICmpNE(S2, llvm::Constant::getNullValue(T)), T);
    if (isX86_MMX) {
      Type *X86_MMXTy = Type::getX86_MMXTy(*MS.C);
      S1_ext = IRB.CreateBitCast(S1_ext, X86_MMXTy);
      S2_ext = IRB.CreateBitCast(S2_ext, X86_MMXTy);
    }

    Function *ShadowFn = Intrinsic::getDeclaration(
        F.getParent(), getSignedPackIntrinsic(I.getIntrinsicID()));

    Value *S = IRB.CreateCall2(ShadowFn, S1_ext, S2_ext, "_msprop_vector_pack");
    if (isX86_MMX) S = IRB.CreateBitCast(S, getShadowTy(&I));
    setShadow(&I, S);
    setOriginForNaryOp(I);
  }

  // \brief Instrument sum-of-absolute-differencies intrinsic.
  void handleVectorSadIntrinsic(IntrinsicInst &I) {
    const unsigned SignificantBitsPerResultElement = 16;
    bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
    Type *ResTy = isX86_MMX ? IntegerType::get(*MS.C, 64) : I.getType();
    unsigned ZeroBitsPerResultElement =
        ResTy->getScalarSizeInBits() - SignificantBitsPerResultElement;

    IRBuilder<> IRB(&I);
    Value *S = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
    S = IRB.CreateBitCast(S, ResTy);
    S = IRB.CreateSExt(IRB.CreateICmpNE(S, Constant::getNullValue(ResTy)),
                       ResTy);
    S = IRB.CreateLShr(S, ZeroBitsPerResultElement);
    S = IRB.CreateBitCast(S, getShadowTy(&I));
    setShadow(&I, S);
    setOriginForNaryOp(I);
  }

  // \brief Instrument multiply-add intrinsic.
  void handleVectorPmaddIntrinsic(IntrinsicInst &I,
                                  unsigned EltSizeInBits = 0) {
    bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
    Type *ResTy = isX86_MMX ? getMMXVectorTy(EltSizeInBits * 2) : I.getType();
    IRBuilder<> IRB(&I);
    Value *S = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
    S = IRB.CreateBitCast(S, ResTy);
    S = IRB.CreateSExt(IRB.CreateICmpNE(S, Constant::getNullValue(ResTy)),
                       ResTy);
    S = IRB.CreateBitCast(S, getShadowTy(&I));
    setShadow(&I, S);
    setOriginForNaryOp(I);
  }

  void visitIntrinsicInst(IntrinsicInst &I) {
    switch (I.getIntrinsicID()) {
    case llvm::Intrinsic::bswap:
      handleBswap(I);
      break;
    case llvm::Intrinsic::x86_avx512_cvtsd2usi64:
    case llvm::Intrinsic::x86_avx512_cvtsd2usi:
    case llvm::Intrinsic::x86_avx512_cvtss2usi64:
    case llvm::Intrinsic::x86_avx512_cvtss2usi:
    case llvm::Intrinsic::x86_avx512_cvttss2usi64:
    case llvm::Intrinsic::x86_avx512_cvttss2usi:
    case llvm::Intrinsic::x86_avx512_cvttsd2usi64:
    case llvm::Intrinsic::x86_avx512_cvttsd2usi:
    case llvm::Intrinsic::x86_avx512_cvtusi2sd:
    case llvm::Intrinsic::x86_avx512_cvtusi2ss:
    case llvm::Intrinsic::x86_avx512_cvtusi642sd:
    case llvm::Intrinsic::x86_avx512_cvtusi642ss:
    case llvm::Intrinsic::x86_sse2_cvtsd2si64:
    case llvm::Intrinsic::x86_sse2_cvtsd2si:
    case llvm::Intrinsic::x86_sse2_cvtsd2ss:
    case llvm::Intrinsic::x86_sse2_cvtsi2sd:
    case llvm::Intrinsic::x86_sse2_cvtsi642sd:
    case llvm::Intrinsic::x86_sse2_cvtss2sd:
    case llvm::Intrinsic::x86_sse2_cvttsd2si64:
    case llvm::Intrinsic::x86_sse2_cvttsd2si:
    case llvm::Intrinsic::x86_sse_cvtsi2ss:
    case llvm::Intrinsic::x86_sse_cvtsi642ss:
    case llvm::Intrinsic::x86_sse_cvtss2si64:
    case llvm::Intrinsic::x86_sse_cvtss2si:
    case llvm::Intrinsic::x86_sse_cvttss2si64:
    case llvm::Intrinsic::x86_sse_cvttss2si:
      handleVectorConvertIntrinsic(I, 1);
      break;
    case llvm::Intrinsic::x86_sse2_cvtdq2pd:
    case llvm::Intrinsic::x86_sse2_cvtps2pd:
    case llvm::Intrinsic::x86_sse_cvtps2pi:
    case llvm::Intrinsic::x86_sse_cvttps2pi:
      handleVectorConvertIntrinsic(I, 2);
      break;
    case llvm::Intrinsic::x86_avx2_psll_w:
    case llvm::Intrinsic::x86_avx2_psll_d:
    case llvm::Intrinsic::x86_avx2_psll_q:
    case llvm::Intrinsic::x86_avx2_pslli_w:
    case llvm::Intrinsic::x86_avx2_pslli_d:
    case llvm::Intrinsic::x86_avx2_pslli_q:
    case llvm::Intrinsic::x86_avx2_psrl_w:
    case llvm::Intrinsic::x86_avx2_psrl_d:
    case llvm::Intrinsic::x86_avx2_psrl_q:
    case llvm::Intrinsic::x86_avx2_psra_w:
    case llvm::Intrinsic::x86_avx2_psra_d:
    case llvm::Intrinsic::x86_avx2_psrli_w:
    case llvm::Intrinsic::x86_avx2_psrli_d:
    case llvm::Intrinsic::x86_avx2_psrli_q:
    case llvm::Intrinsic::x86_avx2_psrai_w:
    case llvm::Intrinsic::x86_avx2_psrai_d:
    case llvm::Intrinsic::x86_sse2_psll_w:
    case llvm::Intrinsic::x86_sse2_psll_d:
    case llvm::Intrinsic::x86_sse2_psll_q:
    case llvm::Intrinsic::x86_sse2_pslli_w:
    case llvm::Intrinsic::x86_sse2_pslli_d:
    case llvm::Intrinsic::x86_sse2_pslli_q:
    case llvm::Intrinsic::x86_sse2_psrl_w:
    case llvm::Intrinsic::x86_sse2_psrl_d:
    case llvm::Intrinsic::x86_sse2_psrl_q:
    case llvm::Intrinsic::x86_sse2_psra_w:
    case llvm::Intrinsic::x86_sse2_psra_d:
    case llvm::Intrinsic::x86_sse2_psrli_w:
    case llvm::Intrinsic::x86_sse2_psrli_d:
    case llvm::Intrinsic::x86_sse2_psrli_q:
    case llvm::Intrinsic::x86_sse2_psrai_w:
    case llvm::Intrinsic::x86_sse2_psrai_d:
    case llvm::Intrinsic::x86_mmx_psll_w:
    case llvm::Intrinsic::x86_mmx_psll_d:
    case llvm::Intrinsic::x86_mmx_psll_q:
    case llvm::Intrinsic::x86_mmx_pslli_w:
    case llvm::Intrinsic::x86_mmx_pslli_d:
    case llvm::Intrinsic::x86_mmx_pslli_q:
    case llvm::Intrinsic::x86_mmx_psrl_w:
    case llvm::Intrinsic::x86_mmx_psrl_d:
    case llvm::Intrinsic::x86_mmx_psrl_q:
    case llvm::Intrinsic::x86_mmx_psra_w:
    case llvm::Intrinsic::x86_mmx_psra_d:
    case llvm::Intrinsic::x86_mmx_psrli_w:
    case llvm::Intrinsic::x86_mmx_psrli_d:
    case llvm::Intrinsic::x86_mmx_psrli_q:
    case llvm::Intrinsic::x86_mmx_psrai_w:
    case llvm::Intrinsic::x86_mmx_psrai_d:
      handleVectorShiftIntrinsic(I, /* Variable */ false);
      break;
    case llvm::Intrinsic::x86_avx2_psllv_d:
    case llvm::Intrinsic::x86_avx2_psllv_d_256:
    case llvm::Intrinsic::x86_avx2_psllv_q:
    case llvm::Intrinsic::x86_avx2_psllv_q_256:
    case llvm::Intrinsic::x86_avx2_psrlv_d:
    case llvm::Intrinsic::x86_avx2_psrlv_d_256:
    case llvm::Intrinsic::x86_avx2_psrlv_q:
    case llvm::Intrinsic::x86_avx2_psrlv_q_256:
    case llvm::Intrinsic::x86_avx2_psrav_d:
    case llvm::Intrinsic::x86_avx2_psrav_d_256:
      handleVectorShiftIntrinsic(I, /* Variable */ true);
      break;

    case llvm::Intrinsic::x86_sse2_packsswb_128:
    case llvm::Intrinsic::x86_sse2_packssdw_128:
    case llvm::Intrinsic::x86_sse2_packuswb_128:
    case llvm::Intrinsic::x86_sse41_packusdw:
    case llvm::Intrinsic::x86_avx2_packsswb:
    case llvm::Intrinsic::x86_avx2_packssdw:
    case llvm::Intrinsic::x86_avx2_packuswb:
    case llvm::Intrinsic::x86_avx2_packusdw:
      handleVectorPackIntrinsic(I);
      break;

    case llvm::Intrinsic::x86_mmx_packsswb:
    case llvm::Intrinsic::x86_mmx_packuswb:
      handleVectorPackIntrinsic(I, 16);
      break;

    case llvm::Intrinsic::x86_mmx_packssdw:
      handleVectorPackIntrinsic(I, 32);
      break;

    case llvm::Intrinsic::x86_mmx_psad_bw:
    case llvm::Intrinsic::x86_sse2_psad_bw:
    case llvm::Intrinsic::x86_avx2_psad_bw:
      handleVectorSadIntrinsic(I);
      break;

    case llvm::Intrinsic::x86_sse2_pmadd_wd:
    case llvm::Intrinsic::x86_avx2_pmadd_wd:
    case llvm::Intrinsic::x86_ssse3_pmadd_ub_sw_128:
    case llvm::Intrinsic::x86_avx2_pmadd_ub_sw:
      handleVectorPmaddIntrinsic(I);
      break;

    case llvm::Intrinsic::x86_ssse3_pmadd_ub_sw:
      handleVectorPmaddIntrinsic(I, 8);
      break;

    case llvm::Intrinsic::x86_mmx_pmadd_wd:
      handleVectorPmaddIntrinsic(I, 16);
      break;

    default:
      if (!handleUnknownIntrinsic(I))
        visitInstruction(I);
      break;
    }
  }

  void visitCallSite(CallSite CS) {
    Instruction &I = *CS.getInstruction();
    assert((CS.isCall() || CS.isInvoke()) && "Unknown type of CallSite");
    if (CS.isCall()) {
      CallInst *Call = cast<CallInst>(&I);

      // For inline asm, do the usual thing: check argument shadow and mark all
      // outputs as clean. Note that any side effects of the inline asm that are
      // not immediately visible in its constraints are not handled.
      if (Call->isInlineAsm()) {
        visitInstruction(I);
        return;
      }

      assert(!isa<IntrinsicInst>(&I) && "intrinsics are handled elsewhere");

      // We are going to insert code that relies on the fact that the callee
      // will become a non-readonly function after it is instrumented by us. To
      // prevent this code from being optimized out, mark that function
      // non-readonly in advance.
      if (Function *Func = Call->getCalledFunction()) {
        // Clear out readonly/readnone attributes.
        AttrBuilder B;
        B.addAttribute(Attribute::ReadOnly)
          .addAttribute(Attribute::ReadNone);
        Func->removeAttributes(AttributeSet::FunctionIndex,
                               AttributeSet::get(Func->getContext(),
                                                 AttributeSet::FunctionIndex,
                                                 B));
      }
    }
    IRBuilder<> IRB(&I);

    unsigned ArgOffset = 0;
    DEBUG(dbgs() << "  CallSite: " << I << "\n");
    for (CallSite::arg_iterator ArgIt = CS.arg_begin(), End = CS.arg_end();
         ArgIt != End; ++ArgIt) {
      Value *A = *ArgIt;
      unsigned i = ArgIt - CS.arg_begin();
      if (!A->getType()->isSized()) {
        DEBUG(dbgs() << "Arg " << i << " is not sized: " << I << "\n");
        continue;
      }
      unsigned Size = 0;
      Value *Store = nullptr;
      // Compute the Shadow for arg even if it is ByVal, because
      // in that case getShadow() will copy the actual arg shadow to
      // __msan_param_tls.
      Value *ArgShadow = getShadow(A);
      Value *ArgShadowBase = getShadowPtrForArgument(A, IRB, ArgOffset);
      DEBUG(dbgs() << "  Arg#" << i << ": " << *A <<
            " Shadow: " << *ArgShadow << "\n");
      bool ArgIsInitialized = false;
      const DataLayout &DL = F.getParent()->getDataLayout();
      if (CS.paramHasAttr(i + 1, Attribute::ByVal)) {
        assert(A->getType()->isPointerTy() &&
               "ByVal argument is not a pointer!");
        Size = DL.getTypeAllocSize(A->getType()->getPointerElementType());
        if (ArgOffset + Size > kParamTLSSize) break;
        unsigned ParamAlignment = CS.getParamAlignment(i + 1);
        unsigned Alignment = std::min(ParamAlignment, kShadowTLSAlignment);
        Store = IRB.CreateMemCpy(ArgShadowBase,
                                 getShadowPtr(A, Type::getInt8Ty(*MS.C), IRB),
                                 Size, Alignment);
      } else {
        Size = DL.getTypeAllocSize(A->getType());
        if (ArgOffset + Size > kParamTLSSize) break;
        Store = IRB.CreateAlignedStore(ArgShadow, ArgShadowBase,
                                       kShadowTLSAlignment);
        Constant *Cst = dyn_cast<Constant>(ArgShadow);
        if (Cst && Cst->isNullValue()) ArgIsInitialized = true;
      }
      if (MS.TrackOrigins && !ArgIsInitialized)
        IRB.CreateStore(getOrigin(A),
                        getOriginPtrForArgument(A, IRB, ArgOffset));
      (void)Store;
      assert(Size != 0 && Store != nullptr);
      DEBUG(dbgs() << "  Param:" << *Store << "\n");
      ArgOffset += RoundUpToAlignment(Size, 8);
    }
    DEBUG(dbgs() << "  done with call args\n");

    FunctionType *FT =
      cast<FunctionType>(CS.getCalledValue()->getType()->getContainedType(0));
    if (FT->isVarArg()) {
      VAHelper->visitCallSite(CS, IRB);
    }

    // Now, get the shadow for the RetVal.
    if (!I.getType()->isSized()) return;
    IRBuilder<> IRBBefore(&I);
    // Until we have full dynamic coverage, make sure the retval shadow is 0.
    Value *Base = getShadowPtrForRetval(&I, IRBBefore);
    IRBBefore.CreateAlignedStore(getCleanShadow(&I), Base, kShadowTLSAlignment);
    Instruction *NextInsn = nullptr;
    if (CS.isCall()) {
      NextInsn = I.getNextNode();
    } else {
      BasicBlock *NormalDest = cast<InvokeInst>(&I)->getNormalDest();
      if (!NormalDest->getSinglePredecessor()) {
        // FIXME: this case is tricky, so we are just conservative here.
        // Perhaps we need to split the edge between this BB and NormalDest,
        // but a naive attempt to use SplitEdge leads to a crash.
        setShadow(&I, getCleanShadow(&I));
        setOrigin(&I, getCleanOrigin());
        return;
      }
      NextInsn = NormalDest->getFirstInsertionPt();
      assert(NextInsn &&
             "Could not find insertion point for retval shadow load");
    }
    IRBuilder<> IRBAfter(NextInsn);
    Value *RetvalShadow =
      IRBAfter.CreateAlignedLoad(getShadowPtrForRetval(&I, IRBAfter),
                                 kShadowTLSAlignment, "_msret");
    setShadow(&I, RetvalShadow);
    if (MS.TrackOrigins)
      setOrigin(&I, IRBAfter.CreateLoad(getOriginPtrForRetval(IRBAfter)));
  }

  void visitReturnInst(ReturnInst &I) {
    IRBuilder<> IRB(&I);
    Value *RetVal = I.getReturnValue();
    if (!RetVal) return;
    Value *ShadowPtr = getShadowPtrForRetval(RetVal, IRB);
    if (CheckReturnValue) {
      insertShadowCheck(RetVal, &I);
      Value *Shadow = getCleanShadow(RetVal);
      IRB.CreateAlignedStore(Shadow, ShadowPtr, kShadowTLSAlignment);
    } else {
      Value *Shadow = getShadow(RetVal);
      IRB.CreateAlignedStore(Shadow, ShadowPtr, kShadowTLSAlignment);
      // FIXME: make it conditional if ClStoreCleanOrigin==0
      if (MS.TrackOrigins)
        IRB.CreateStore(getOrigin(RetVal), getOriginPtrForRetval(IRB));
    }
  }

  void visitPHINode(PHINode &I) {
    IRBuilder<> IRB(&I);
    if (!PropagateShadow) {
      setShadow(&I, getCleanShadow(&I));
      setOrigin(&I, getCleanOrigin());
      return;
    }

    ShadowPHINodes.push_back(&I);
    setShadow(&I, IRB.CreatePHI(getShadowTy(&I), I.getNumIncomingValues(),
                                "_msphi_s"));
    if (MS.TrackOrigins)
      setOrigin(&I, IRB.CreatePHI(MS.OriginTy, I.getNumIncomingValues(),
                                  "_msphi_o"));
  }

  void visitAllocaInst(AllocaInst &I) {
    setShadow(&I, getCleanShadow(&I));
    setOrigin(&I, getCleanOrigin());
    IRBuilder<> IRB(I.getNextNode());
    const DataLayout &DL = F.getParent()->getDataLayout();
    uint64_t Size = DL.getTypeAllocSize(I.getAllocatedType());
    if (PoisonStack && ClPoisonStackWithCall) {
      IRB.CreateCall2(MS.MsanPoisonStackFn,
                      IRB.CreatePointerCast(&I, IRB.getInt8PtrTy()),
                      ConstantInt::get(MS.IntptrTy, Size));
    } else {
      Value *ShadowBase = getShadowPtr(&I, Type::getInt8PtrTy(*MS.C), IRB);
      Value *PoisonValue = IRB.getInt8(PoisonStack ? ClPoisonStackPattern : 0);
      IRB.CreateMemSet(ShadowBase, PoisonValue, Size, I.getAlignment());
    }

    if (PoisonStack && MS.TrackOrigins) {
      SmallString<2048> StackDescriptionStorage;
      raw_svector_ostream StackDescription(StackDescriptionStorage);
      // We create a string with a description of the stack allocation and
      // pass it into __msan_set_alloca_origin.
      // It will be printed by the run-time if stack-originated UMR is found.
      // The first 4 bytes of the string are set to '----' and will be replaced
      // by __msan_va_arg_overflow_size_tls at the first call.
      StackDescription << "----" << I.getName() << "@" << F.getName();
      Value *Descr =
          createPrivateNonConstGlobalForString(*F.getParent(),
                                               StackDescription.str());

      IRB.CreateCall4(MS.MsanSetAllocaOrigin4Fn,
                      IRB.CreatePointerCast(&I, IRB.getInt8PtrTy()),
                      ConstantInt::get(MS.IntptrTy, Size),
                      IRB.CreatePointerCast(Descr, IRB.getInt8PtrTy()),
                      IRB.CreatePointerCast(&F, MS.IntptrTy));
    }
  }

  void visitSelectInst(SelectInst& I) {
    IRBuilder<> IRB(&I);
    // a = select b, c, d
    Value *B = I.getCondition();
    Value *C = I.getTrueValue();
    Value *D = I.getFalseValue();
    Value *Sb = getShadow(B);
    Value *Sc = getShadow(C);
    Value *Sd = getShadow(D);

    // Result shadow if condition shadow is 0.
    Value *Sa0 = IRB.CreateSelect(B, Sc, Sd);
    Value *Sa1;
    if (I.getType()->isAggregateType()) {
      // To avoid "sign extending" i1 to an arbitrary aggregate type, we just do
      // an extra "select". This results in much more compact IR.
      // Sa = select Sb, poisoned, (select b, Sc, Sd)
      Sa1 = getPoisonedShadow(getShadowTy(I.getType()));
    } else {
      // Sa = select Sb, [ (c^d) | Sc | Sd ], [ b ? Sc : Sd ]
      // If Sb (condition is poisoned), look for bits in c and d that are equal
      // and both unpoisoned.
      // If !Sb (condition is unpoisoned), simply pick one of Sc and Sd.

      // Cast arguments to shadow-compatible type.
      C = CreateAppToShadowCast(IRB, C);
      D = CreateAppToShadowCast(IRB, D);

      // Result shadow if condition shadow is 1.
      Sa1 = IRB.CreateOr(IRB.CreateXor(C, D), IRB.CreateOr(Sc, Sd));
    }
    Value *Sa = IRB.CreateSelect(Sb, Sa1, Sa0, "_msprop_select");
    setShadow(&I, Sa);
    if (MS.TrackOrigins) {
      // Origins are always i32, so any vector conditions must be flattened.
      // FIXME: consider tracking vector origins for app vectors?
      if (B->getType()->isVectorTy()) {
        Type *FlatTy = getShadowTyNoVec(B->getType());
        B = IRB.CreateICmpNE(IRB.CreateBitCast(B, FlatTy),
                                ConstantInt::getNullValue(FlatTy));
        Sb = IRB.CreateICmpNE(IRB.CreateBitCast(Sb, FlatTy),
                                      ConstantInt::getNullValue(FlatTy));
      }
      // a = select b, c, d
      // Oa = Sb ? Ob : (b ? Oc : Od)
      setOrigin(
          &I, IRB.CreateSelect(Sb, getOrigin(I.getCondition()),
                               IRB.CreateSelect(B, getOrigin(I.getTrueValue()),
                                                getOrigin(I.getFalseValue()))));
    }
  }

  void visitLandingPadInst(LandingPadInst &I) {
    // Do nothing.
    // See http://code.google.com/p/memory-sanitizer/issues/detail?id=1
    setShadow(&I, getCleanShadow(&I));
    setOrigin(&I, getCleanOrigin());
  }

  void visitGetElementPtrInst(GetElementPtrInst &I) {
    handleShadowOr(I);
  }

  void visitExtractValueInst(ExtractValueInst &I) {
    IRBuilder<> IRB(&I);
    Value *Agg = I.getAggregateOperand();
    DEBUG(dbgs() << "ExtractValue:  " << I << "\n");
    Value *AggShadow = getShadow(Agg);
    DEBUG(dbgs() << "   AggShadow:  " << *AggShadow << "\n");
    Value *ResShadow = IRB.CreateExtractValue(AggShadow, I.getIndices());
    DEBUG(dbgs() << "   ResShadow:  " << *ResShadow << "\n");
    setShadow(&I, ResShadow);
    setOriginForNaryOp(I);
  }

  void visitInsertValueInst(InsertValueInst &I) {
    IRBuilder<> IRB(&I);
    DEBUG(dbgs() << "InsertValue:  " << I << "\n");
    Value *AggShadow = getShadow(I.getAggregateOperand());
    Value *InsShadow = getShadow(I.getInsertedValueOperand());
    DEBUG(dbgs() << "   AggShadow:  " << *AggShadow << "\n");
    DEBUG(dbgs() << "   InsShadow:  " << *InsShadow << "\n");
    Value *Res = IRB.CreateInsertValue(AggShadow, InsShadow, I.getIndices());
    DEBUG(dbgs() << "   Res:        " << *Res << "\n");
    setShadow(&I, Res);
    setOriginForNaryOp(I);
  }

  void dumpInst(Instruction &I) {
    if (CallInst *CI = dyn_cast<CallInst>(&I)) {
      errs() << "ZZZ call " << CI->getCalledFunction()->getName() << "\n";
    } else {
      errs() << "ZZZ " << I.getOpcodeName() << "\n";
    }
    errs() << "QQQ " << I << "\n";
  }

  void visitResumeInst(ResumeInst &I) {
    DEBUG(dbgs() << "Resume: " << I << "\n");
    // Nothing to do here.
  }

  void visitInstruction(Instruction &I) {
    // Everything else: stop propagating and check for poisoned shadow.
    if (ClDumpStrictInstructions)
      dumpInst(I);
    DEBUG(dbgs() << "DEFAULT: " << I << "\n");
    for (size_t i = 0, n = I.getNumOperands(); i < n; i++)
      insertShadowCheck(I.getOperand(i), &I);
    setShadow(&I, getCleanShadow(&I));
    setOrigin(&I, getCleanOrigin());
  }
};

/// \brief AMD64-specific implementation of VarArgHelper.
struct VarArgAMD64Helper : public VarArgHelper {
  // An unfortunate workaround for asymmetric lowering of va_arg stuff.
  // See a comment in visitCallSite for more details.
  static const unsigned AMD64GpEndOffset = 48;  // AMD64 ABI Draft 0.99.6 p3.5.7
  static const unsigned AMD64FpEndOffset = 176;

  Function &F;
  MemorySanitizer &MS;
  MemorySanitizerVisitor &MSV;
  Value *VAArgTLSCopy;
  Value *VAArgOverflowSize;

  SmallVector<CallInst*, 16> VAStartInstrumentationList;

  VarArgAMD64Helper(Function &F, MemorySanitizer &MS,
                    MemorySanitizerVisitor &MSV)
    : F(F), MS(MS), MSV(MSV), VAArgTLSCopy(nullptr),
      VAArgOverflowSize(nullptr) {}

  enum ArgKind { AK_GeneralPurpose, AK_FloatingPoint, AK_Memory };

  ArgKind classifyArgument(Value* arg) {
    // A very rough approximation of X86_64 argument classification rules.
    Type *T = arg->getType();
    if (T->isFPOrFPVectorTy() || T->isX86_MMXTy())
      return AK_FloatingPoint;
    if (T->isIntegerTy() && T->getPrimitiveSizeInBits() <= 64)
      return AK_GeneralPurpose;
    if (T->isPointerTy())
      return AK_GeneralPurpose;
    return AK_Memory;
  }

  // For VarArg functions, store the argument shadow in an ABI-specific format
  // that corresponds to va_list layout.
  // We do this because Clang lowers va_arg in the frontend, and this pass
  // only sees the low level code that deals with va_list internals.
  // A much easier alternative (provided that Clang emits va_arg instructions)
  // would have been to associate each live instance of va_list with a copy of
  // MSanParamTLS, and extract shadow on va_arg() call in the argument list
  // order.
  void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {
    unsigned GpOffset = 0;
    unsigned FpOffset = AMD64GpEndOffset;
    unsigned OverflowOffset = AMD64FpEndOffset;
    const DataLayout &DL = F.getParent()->getDataLayout();
    for (CallSite::arg_iterator ArgIt = CS.arg_begin(), End = CS.arg_end();
         ArgIt != End; ++ArgIt) {
      Value *A = *ArgIt;
      unsigned ArgNo = CS.getArgumentNo(ArgIt);
      bool IsByVal = CS.paramHasAttr(ArgNo + 1, Attribute::ByVal);
      if (IsByVal) {
        // ByVal arguments always go to the overflow area.
        assert(A->getType()->isPointerTy());
        Type *RealTy = A->getType()->getPointerElementType();
        uint64_t ArgSize = DL.getTypeAllocSize(RealTy);
        Value *Base = getShadowPtrForVAArgument(RealTy, IRB, OverflowOffset);
        OverflowOffset += RoundUpToAlignment(ArgSize, 8);
        IRB.CreateMemCpy(Base, MSV.getShadowPtr(A, IRB.getInt8Ty(), IRB),
                         ArgSize, kShadowTLSAlignment);
      } else {
        ArgKind AK = classifyArgument(A);
        if (AK == AK_GeneralPurpose && GpOffset >= AMD64GpEndOffset)
          AK = AK_Memory;
        if (AK == AK_FloatingPoint && FpOffset >= AMD64FpEndOffset)
          AK = AK_Memory;
        Value *Base;
        switch (AK) {
          case AK_GeneralPurpose:
            Base = getShadowPtrForVAArgument(A->getType(), IRB, GpOffset);
            GpOffset += 8;
            break;
          case AK_FloatingPoint:
            Base = getShadowPtrForVAArgument(A->getType(), IRB, FpOffset);
            FpOffset += 16;
            break;
          case AK_Memory:
            uint64_t ArgSize = DL.getTypeAllocSize(A->getType());
            Base = getShadowPtrForVAArgument(A->getType(), IRB, OverflowOffset);
            OverflowOffset += RoundUpToAlignment(ArgSize, 8);
        }
        IRB.CreateAlignedStore(MSV.getShadow(A), Base, kShadowTLSAlignment);
      }
    }
    Constant *OverflowSize =
      ConstantInt::get(IRB.getInt64Ty(), OverflowOffset - AMD64FpEndOffset);
    IRB.CreateStore(OverflowSize, MS.VAArgOverflowSizeTLS);
  }

  /// \brief Compute the shadow address for a given va_arg.
  Value *getShadowPtrForVAArgument(Type *Ty, IRBuilder<> &IRB,
                                   int ArgOffset) {
    Value *Base = IRB.CreatePointerCast(MS.VAArgTLS, MS.IntptrTy);
    Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
    return IRB.CreateIntToPtr(Base, PointerType::get(MSV.getShadowTy(Ty), 0),
                              "_msarg");
  }

  void visitVAStartInst(VAStartInst &I) override {
    IRBuilder<> IRB(&I);
    VAStartInstrumentationList.push_back(&I);
    Value *VAListTag = I.getArgOperand(0);
    Value *ShadowPtr = MSV.getShadowPtr(VAListTag, IRB.getInt8Ty(), IRB);

    // Unpoison the whole __va_list_tag.
    // FIXME: magic ABI constants.
    IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
                     /* size */24, /* alignment */8, false);
  }

  void visitVACopyInst(VACopyInst &I) override {
    IRBuilder<> IRB(&I);
    Value *VAListTag = I.getArgOperand(0);
    Value *ShadowPtr = MSV.getShadowPtr(VAListTag, IRB.getInt8Ty(), IRB);

    // Unpoison the whole __va_list_tag.
    // FIXME: magic ABI constants.
    IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
                     /* size */24, /* alignment */8, false);
  }

  void finalizeInstrumentation() override {
    assert(!VAArgOverflowSize && !VAArgTLSCopy &&
           "finalizeInstrumentation called twice");
    if (!VAStartInstrumentationList.empty()) {
      // If there is a va_start in this function, make a backup copy of
      // va_arg_tls somewhere in the function entry block.
      IRBuilder<> IRB(F.getEntryBlock().getFirstNonPHI());
      VAArgOverflowSize = IRB.CreateLoad(MS.VAArgOverflowSizeTLS);
      Value *CopySize =
        IRB.CreateAdd(ConstantInt::get(MS.IntptrTy, AMD64FpEndOffset),
                      VAArgOverflowSize);
      VAArgTLSCopy = IRB.CreateAlloca(Type::getInt8Ty(*MS.C), CopySize);
      IRB.CreateMemCpy(VAArgTLSCopy, MS.VAArgTLS, CopySize, 8);
    }

    // Instrument va_start.
    // Copy va_list shadow from the backup copy of the TLS contents.
    for (size_t i = 0, n = VAStartInstrumentationList.size(); i < n; i++) {
      CallInst *OrigInst = VAStartInstrumentationList[i];
      IRBuilder<> IRB(OrigInst->getNextNode());
      Value *VAListTag = OrigInst->getArgOperand(0);

      Value *RegSaveAreaPtrPtr =
        IRB.CreateIntToPtr(
          IRB.CreateAdd(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
                        ConstantInt::get(MS.IntptrTy, 16)),
          Type::getInt64PtrTy(*MS.C));
      Value *RegSaveAreaPtr = IRB.CreateLoad(RegSaveAreaPtrPtr);
      Value *RegSaveAreaShadowPtr =
        MSV.getShadowPtr(RegSaveAreaPtr, IRB.getInt8Ty(), IRB);
      IRB.CreateMemCpy(RegSaveAreaShadowPtr, VAArgTLSCopy,
                       AMD64FpEndOffset, 16);

      Value *OverflowArgAreaPtrPtr =
        IRB.CreateIntToPtr(
          IRB.CreateAdd(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
                        ConstantInt::get(MS.IntptrTy, 8)),
          Type::getInt64PtrTy(*MS.C));
      Value *OverflowArgAreaPtr = IRB.CreateLoad(OverflowArgAreaPtrPtr);
      Value *OverflowArgAreaShadowPtr =
        MSV.getShadowPtr(OverflowArgAreaPtr, IRB.getInt8Ty(), IRB);
      Value *SrcPtr = IRB.CreateConstGEP1_32(VAArgTLSCopy, AMD64FpEndOffset);
      IRB.CreateMemCpy(OverflowArgAreaShadowPtr, SrcPtr, VAArgOverflowSize, 16);
    }
  }
};

/// \brief MIPS64-specific implementation of VarArgHelper.
struct VarArgMIPS64Helper : public VarArgHelper {
  Function &F;
  MemorySanitizer &MS;
  MemorySanitizerVisitor &MSV;
  Value *VAArgTLSCopy;
  Value *VAArgSize;

  SmallVector<CallInst*, 16> VAStartInstrumentationList;

  VarArgMIPS64Helper(Function &F, MemorySanitizer &MS,
                    MemorySanitizerVisitor &MSV)
    : F(F), MS(MS), MSV(MSV), VAArgTLSCopy(nullptr),
      VAArgSize(nullptr) {}

  void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {
    unsigned VAArgOffset = 0;
    const DataLayout &DL = F.getParent()->getDataLayout();
    for (CallSite::arg_iterator ArgIt = CS.arg_begin() + 1, End = CS.arg_end();
         ArgIt != End; ++ArgIt) {
      Value *A = *ArgIt;
      Value *Base;
      uint64_t ArgSize = DL.getTypeAllocSize(A->getType());
#if defined(__MIPSEB__) || defined(MIPSEB)
      // Adjusting the shadow for argument with size < 8 to match the placement
      // of bits in big endian system
      if (ArgSize < 8)
        VAArgOffset += (8 - ArgSize);
#endif
      Base = getShadowPtrForVAArgument(A->getType(), IRB, VAArgOffset);
      VAArgOffset += ArgSize;
      VAArgOffset = RoundUpToAlignment(VAArgOffset, 8);
      IRB.CreateAlignedStore(MSV.getShadow(A), Base, kShadowTLSAlignment);
    }

    Constant *TotalVAArgSize = ConstantInt::get(IRB.getInt64Ty(), VAArgOffset);
    // Here using VAArgOverflowSizeTLS as VAArgSizeTLS to avoid creation of
    // a new class member i.e. it is the total size of all VarArgs.
    IRB.CreateStore(TotalVAArgSize, MS.VAArgOverflowSizeTLS);
  }

  /// \brief Compute the shadow address for a given va_arg.
  Value *getShadowPtrForVAArgument(Type *Ty, IRBuilder<> &IRB,
                                   int ArgOffset) {
    Value *Base = IRB.CreatePointerCast(MS.VAArgTLS, MS.IntptrTy);
    Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
    return IRB.CreateIntToPtr(Base, PointerType::get(MSV.getShadowTy(Ty), 0),
                              "_msarg");
  }

  void visitVAStartInst(VAStartInst &I) override {
    IRBuilder<> IRB(&I);
    VAStartInstrumentationList.push_back(&I);
    Value *VAListTag = I.getArgOperand(0);
    Value *ShadowPtr = MSV.getShadowPtr(VAListTag, IRB.getInt8Ty(), IRB);
    IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
                     /* size */8, /* alignment */8, false);
  }

  void visitVACopyInst(VACopyInst &I) override {
    IRBuilder<> IRB(&I);
    Value *VAListTag = I.getArgOperand(0);
    Value *ShadowPtr = MSV.getShadowPtr(VAListTag, IRB.getInt8Ty(), IRB);
    // Unpoison the whole __va_list_tag.
    // FIXME: magic ABI constants.
    IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
                     /* size */8, /* alignment */8, false);
  }

  void finalizeInstrumentation() override {
    assert(!VAArgSize && !VAArgTLSCopy &&
           "finalizeInstrumentation called twice");
    IRBuilder<> IRB(F.getEntryBlock().getFirstNonPHI());
    VAArgSize = IRB.CreateLoad(MS.VAArgOverflowSizeTLS);
    Value *CopySize = IRB.CreateAdd(ConstantInt::get(MS.IntptrTy, 0),
                                    VAArgSize);

    if (!VAStartInstrumentationList.empty()) {
      // If there is a va_start in this function, make a backup copy of
      // va_arg_tls somewhere in the function entry block.
      VAArgTLSCopy = IRB.CreateAlloca(Type::getInt8Ty(*MS.C), CopySize);
      IRB.CreateMemCpy(VAArgTLSCopy, MS.VAArgTLS, CopySize, 8);
    }

    // Instrument va_start.
    // Copy va_list shadow from the backup copy of the TLS contents.
    for (size_t i = 0, n = VAStartInstrumentationList.size(); i < n; i++) {
      CallInst *OrigInst = VAStartInstrumentationList[i];
      IRBuilder<> IRB(OrigInst->getNextNode());
      Value *VAListTag = OrigInst->getArgOperand(0);
      Value *RegSaveAreaPtrPtr =
        IRB.CreateIntToPtr(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
                        Type::getInt64PtrTy(*MS.C));
      Value *RegSaveAreaPtr = IRB.CreateLoad(RegSaveAreaPtrPtr);
      Value *RegSaveAreaShadowPtr =
      MSV.getShadowPtr(RegSaveAreaPtr, IRB.getInt8Ty(), IRB);
      IRB.CreateMemCpy(RegSaveAreaShadowPtr, VAArgTLSCopy, CopySize, 8);
    }
  }
};

/// \brief A no-op implementation of VarArgHelper.
struct VarArgNoOpHelper : public VarArgHelper {
  VarArgNoOpHelper(Function &F, MemorySanitizer &MS,
                   MemorySanitizerVisitor &MSV) {}

  void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {}

  void visitVAStartInst(VAStartInst &I) override {}

  void visitVACopyInst(VACopyInst &I) override {}

  void finalizeInstrumentation() override {}
};

VarArgHelper *CreateVarArgHelper(Function &Func, MemorySanitizer &Msan,
                                 MemorySanitizerVisitor &Visitor) {
  // VarArg handling is only implemented on AMD64. False positives are possible
  // on other platforms.
  llvm::Triple TargetTriple(Func.getParent()->getTargetTriple());
  if (TargetTriple.getArch() == llvm::Triple::x86_64)
    return new VarArgAMD64Helper(Func, Msan, Visitor);
  else if (TargetTriple.getArch() == llvm::Triple::mips64 ||
           TargetTriple.getArch() == llvm::Triple::mips64el)
    return new VarArgMIPS64Helper(Func, Msan, Visitor);
  else
    return new VarArgNoOpHelper(Func, Msan, Visitor);
}

}  // namespace

bool MemorySanitizer::runOnFunction(Function &F) {
  MemorySanitizerVisitor Visitor(F, *this);

  // Clear out readonly/readnone attributes.
  AttrBuilder B;
  B.addAttribute(Attribute::ReadOnly)
    .addAttribute(Attribute::ReadNone);
  F.removeAttributes(AttributeSet::FunctionIndex,
                     AttributeSet::get(F.getContext(),
                                       AttributeSet::FunctionIndex, B));

  return Visitor.runOnFunction();
}