summaryrefslogtreecommitdiff
path: root/docs/WritingAnLLVMBackend.html
blob: 0ad472cb92310de73d403fd55ee9503bf9221b18 (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
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
                      "http://www.w3.org/TR/html4/strict.dtd">
<html>
<head>
  <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
  <title>Writing an LLVM Compiler Backend</title>
  <link rel="stylesheet" href="_static/llvm.css" type="text/css">
</head>

<body>

<h1>
  Writing an LLVM Compiler Backend
</h1>

<ol>
  <li><a href="#intro">Introduction</a>
  <ul>
    <li><a href="#Audience">Audience</a></li>
    <li><a href="#Prerequisite">Prerequisite Reading</a></li>
    <li><a href="#Basic">Basic Steps</a></li>
    <li><a href="#Preliminaries">Preliminaries</a></li>
  </ul>
  <li><a href="#TargetMachine">Target Machine</a></li>
  <li><a href="#TargetRegistration">Target Registration</a></li>
  <li><a href="#RegisterSet">Register Set and Register Classes</a>
  <ul>
    <li><a href="#RegisterDef">Defining a Register</a></li>
    <li><a href="#RegisterClassDef">Defining a Register Class</a></li>
    <li><a href="#implementRegister">Implement a subclass of TargetRegisterInfo</a></li>
  </ul></li>
  <li><a href="#InstructionSet">Instruction Set</a>
  <ul>  
    <li><a href="#operandMapping">Instruction Operand Mapping</a></li>
    <li><a href="#relationMapping">Instruction Relation Mapping</a></li>
    <li><a href="#implementInstr">Implement a subclass of TargetInstrInfo</a></li>
    <li><a href="#branchFolding">Branch Folding and If Conversion</a></li>
  </ul></li>
  <li><a href="#InstructionSelector">Instruction Selector</a>
  <ul>
    <li><a href="#LegalizePhase">The SelectionDAG Legalize Phase</a>
    <ul>
      <li><a href="#promote">Promote</a></li> 
      <li><a href="#expand">Expand</a></li> 
      <li><a href="#custom">Custom</a></li> 
      <li><a href="#legal">Legal</a></li>       
    </ul></li>
    <li><a href="#callingConventions">Calling Conventions</a></li>     
  </ul></li>
  <li><a href="#assemblyPrinter">Assembly Printer</a></li> 
  <li><a href="#subtargetSupport">Subtarget Support</a></li> 
  <li><a href="#jitSupport">JIT Support</a>
  <ul>  
    <li><a href="#mce">Machine Code Emitter</a></li>   
    <li><a href="#targetJITInfo">Target JIT Info</a></li>   
  </ul></li>
</ol>

<div class="doc_author">    
  <p>Written by <a href="http://www.woo.com">Mason Woo</a> and
                <a href="http://misha.brukman.net">Misha Brukman</a></p>
</div>

<!-- *********************************************************************** -->
<h2>
  <a name="intro">Introduction</a>
</h2>
<!-- *********************************************************************** -->

<div>

<p>
This document describes techniques for writing compiler backends that convert
the LLVM Intermediate Representation (IR) to code for a specified machine or
other languages. Code intended for a specific machine can take the form of
either assembly code or binary code (usable for a JIT compiler).
</p>

<p>
The backend of LLVM features a target-independent code generator that may create
output for several types of target CPUs &mdash; including X86, PowerPC, ARM,
and SPARC. The backend may also be used to generate code targeted at SPUs of the
Cell processor or GPUs to support the execution of compute kernels.
</p>

<p>
The document focuses on existing examples found in subdirectories
of <tt>llvm/lib/Target</tt> in a downloaded LLVM release. In particular, this
document focuses on the example of creating a static compiler (one that emits
text assembly) for a SPARC target, because SPARC has fairly standard
characteristics, such as a RISC instruction set and straightforward calling
conventions.
</p>

<h3>
  <a name="Audience">Audience</a>
</h3>

<div>

<p>
The audience for this document is anyone who needs to write an LLVM backend to
generate code for a specific hardware or software target.
</p>

</div>

<h3>
  <a name="Prerequisite">Prerequisite Reading</a>
</h3>

<div>  

<p>
These essential documents must be read before reading this document:
</p>

<ul>
<li><i><a href="LangRef.html">LLVM Language Reference
    Manual</a></i> &mdash; a reference manual for the LLVM assembly language.</li>

<li><i><a href="CodeGenerator.html">The LLVM
    Target-Independent Code Generator</a></i> &mdash; a guide to the components
    (classes and code generation algorithms) for translating the LLVM internal
    representation into machine code for a specified target.  Pay particular
    attention to the descriptions of code generation stages: Instruction
    Selection, Scheduling and Formation, SSA-based Optimization, Register
    Allocation, Prolog/Epilog Code Insertion, Late Machine Code Optimizations,
    and Code Emission.</li>

<li><i><a href="TableGenFundamentals.html">TableGen
    Fundamentals</a></i> &mdash;a document that describes the TableGen
    (<tt>tblgen</tt>) application that manages domain-specific information to
    support LLVM code generation. TableGen processes input from a target
    description file (<tt>.td</tt> suffix) and generates C++ code that can be
    used for code generation.</li>

<li><i><a href="WritingAnLLVMPass.html">Writing an LLVM
    Pass</a></i> &mdash; The assembly printer is a <tt>FunctionPass</tt>, as are
    several SelectionDAG processing steps.</li>
</ul>

<p>
To follow the SPARC examples in this document, have a copy of
<i><a href="http://www.sparc.org/standards/V8.pdf">The SPARC Architecture
Manual, Version 8</a></i> for reference. For details about the ARM instruction
set, refer to the <i><a href="http://infocenter.arm.com/">ARM Architecture
Reference Manual</a></i>. For more about the GNU Assembler format
(<tt>GAS</tt>), see
<i><a href="http://sourceware.org/binutils/docs/as/index.html">Using As</a></i>,
especially for the assembly printer. <i>Using As</i> contains a list of target
machine dependent features.
</p>

</div>

<h3>
  <a name="Basic">Basic Steps</a>
</h3>

<div>

<p>
To write a compiler backend for LLVM that converts the LLVM IR to code for a
specified target (machine or other language), follow these steps:
</p>

<ul>
<li>Create a subclass of the TargetMachine class that describes characteristics
    of your target machine. Copy existing examples of specific TargetMachine
    class and header files; for example, start with
    <tt>SparcTargetMachine.cpp</tt> and <tt>SparcTargetMachine.h</tt>, but
    change the file names for your target. Similarly, change code that
    references "Sparc" to reference your target. </li>

<li>Describe the register set of the target. Use TableGen to generate code for
    register definition, register aliases, and register classes from a
    target-specific <tt>RegisterInfo.td</tt> input file. You should also write
    additional code for a subclass of the TargetRegisterInfo class that
    represents the class register file data used for register allocation and
    also describes the interactions between registers.</li>

<li>Describe the instruction set of the target. Use TableGen to generate code
    for target-specific instructions from target-specific versions of
    <tt>TargetInstrFormats.td</tt> and <tt>TargetInstrInfo.td</tt>. You should
    write additional code for a subclass of the TargetInstrInfo class to
    represent machine instructions supported by the target machine. </li>

<li>Describe the selection and conversion of the LLVM IR from a Directed Acyclic
    Graph (DAG) representation of instructions to native target-specific
    instructions. Use TableGen to generate code that matches patterns and
    selects instructions based on additional information in a target-specific
    version of <tt>TargetInstrInfo.td</tt>. Write code
    for <tt>XXXISelDAGToDAG.cpp</tt>, where XXX identifies the specific target,
    to perform pattern matching and DAG-to-DAG instruction selection. Also write
    code in <tt>XXXISelLowering.cpp</tt> to replace or remove operations and
    data types that are not supported natively in a SelectionDAG. </li>

<li>Write code for an assembly printer that converts LLVM IR to a GAS format for
    your target machine.  You should add assembly strings to the instructions
    defined in your target-specific version of <tt>TargetInstrInfo.td</tt>. You
    should also write code for a subclass of AsmPrinter that performs the
    LLVM-to-assembly conversion and a trivial subclass of TargetAsmInfo.</li>

<li>Optionally, add support for subtargets (i.e., variants with different
    capabilities). You should also write code for a subclass of the
    TargetSubtarget class, which allows you to use the <tt>-mcpu=</tt>
    and <tt>-mattr=</tt> command-line options.</li>

<li>Optionally, add JIT support and create a machine code emitter (subclass of
    TargetJITInfo) that is used to emit binary code directly into memory. </li>
</ul>

<p>
In the <tt>.cpp</tt> and <tt>.h</tt>. files, initially stub up these methods and
then implement them later. Initially, you may not know which private members
that the class will need and which components will need to be subclassed.
</p>

</div>

<h3>
  <a name="Preliminaries">Preliminaries</a>
</h3>

<div>

<p>
To actually create your compiler backend, you need to create and modify a few
files. The absolute minimum is discussed here. But to actually use the LLVM
target-independent code generator, you must perform the steps described in
the <a href="CodeGenerator.html">LLVM
Target-Independent Code Generator</a> document.
</p>

<p>
First, you should create a subdirectory under <tt>lib/Target</tt> to hold all
the files related to your target. If your target is called "Dummy," create the
directory <tt>lib/Target/Dummy</tt>.
</p>

<p>
In this new
directory, create a <tt>Makefile</tt>. It is easiest to copy a
<tt>Makefile</tt> of another target and modify it. It should at least contain
the <tt>LEVEL</tt>, <tt>LIBRARYNAME</tt> and <tt>TARGET</tt> variables, and then
include <tt>$(LEVEL)/Makefile.common</tt>. The library can be
named <tt>LLVMDummy</tt> (for example, see the MIPS target). Alternatively, you
can split the library into <tt>LLVMDummyCodeGen</tt>
and <tt>LLVMDummyAsmPrinter</tt>, the latter of which should be implemented in a
subdirectory below <tt>lib/Target/Dummy</tt> (for example, see the PowerPC
target).
</p>

<p>
Note that these two naming schemes are hardcoded into <tt>llvm-config</tt>.
Using any other naming scheme will confuse <tt>llvm-config</tt> and produce a
lot of (seemingly unrelated) linker errors when linking <tt>llc</tt>.
</p>

<p>
To make your target actually do something, you need to implement a subclass of
<tt>TargetMachine</tt>. This implementation should typically be in the file
<tt>lib/Target/DummyTargetMachine.cpp</tt>, but any file in
the <tt>lib/Target</tt> directory will be built and should work. To use LLVM's
target independent code generator, you should do what all current machine
backends do: create a subclass of <tt>LLVMTargetMachine</tt>. (To create a
target from scratch, create a subclass of <tt>TargetMachine</tt>.)
</p>

<p>
To get LLVM to actually build and link your target, you need to add it to
the <tt>TARGETS_TO_BUILD</tt> variable. To do this, you modify the configure
script to know about your target when parsing the <tt>--enable-targets</tt>
option. Search the configure script for <tt>TARGETS_TO_BUILD</tt>, add your
target to the lists there (some creativity required), and then
reconfigure. Alternatively, you can change <tt>autotools/configure.ac</tt> and
regenerate configure by running <tt>./autoconf/AutoRegen.sh</tt>.
</p>

</div>

</div>

<!-- *********************************************************************** -->
<h2>
  <a name="TargetMachine">Target Machine</a>
</h2>
<!-- *********************************************************************** -->

<div>

<p>
<tt>LLVMTargetMachine</tt> is designed as a base class for targets implemented
with the LLVM target-independent code generator. The <tt>LLVMTargetMachine</tt>
class should be specialized by a concrete target class that implements the
various virtual methods. <tt>LLVMTargetMachine</tt> is defined as a subclass of
<tt>TargetMachine</tt> in <tt>include/llvm/Target/TargetMachine.h</tt>. The
<tt>TargetMachine</tt> class implementation (<tt>TargetMachine.cpp</tt>) also
processes numerous command-line options.
</p>

<p>
To create a concrete target-specific subclass of <tt>LLVMTargetMachine</tt>,
start by copying an existing <tt>TargetMachine</tt> class and header.  You
should name the files that you create to reflect your specific target. For
instance, for the SPARC target, name the files <tt>SparcTargetMachine.h</tt> and
<tt>SparcTargetMachine.cpp</tt>.
</p>

<p>
For a target machine <tt>XXX</tt>, the implementation of
<tt>XXXTargetMachine</tt> must have access methods to obtain objects that
represent target components.  These methods are named <tt>get*Info</tt>, and are
intended to obtain the instruction set (<tt>getInstrInfo</tt>), register set
(<tt>getRegisterInfo</tt>), stack frame layout (<tt>getFrameInfo</tt>), and
similar information. <tt>XXXTargetMachine</tt> must also implement the
<tt>getDataLayout</tt> method to access an object with target-specific data
characteristics, such as data type size and alignment requirements.
</p>

<p>
For instance, for the SPARC target, the header file
<tt>SparcTargetMachine.h</tt> declares prototypes for several <tt>get*Info</tt>
and <tt>getDataLayout</tt> methods that simply return a class member.
</p>

<div class="doc_code">
<pre>
namespace llvm {

class Module;

class SparcTargetMachine : public LLVMTargetMachine {
  const DataLayout DataLayout;       // Calculates type size &amp; alignment
  SparcSubtarget Subtarget;
  SparcInstrInfo InstrInfo;
  TargetFrameInfo FrameInfo;
  
protected:
  virtual const TargetAsmInfo *createTargetAsmInfo() const;
  
public:
  SparcTargetMachine(const Module &amp;M, const std::string &amp;FS);

  virtual const SparcInstrInfo *getInstrInfo() const {return &amp;InstrInfo; }
  virtual const TargetFrameInfo *getFrameInfo() const {return &amp;FrameInfo; }
  virtual const TargetSubtarget *getSubtargetImpl() const{return &amp;Subtarget; }
  virtual const TargetRegisterInfo *getRegisterInfo() const {
    return &amp;InstrInfo.getRegisterInfo();
  }
  virtual const DataLayout *getDataLayout() const { return &amp;DataLayout; }
  static unsigned getModuleMatchQuality(const Module &amp;M);

  // Pass Pipeline Configuration
  virtual bool addInstSelector(PassManagerBase &amp;PM, bool Fast);
  virtual bool addPreEmitPass(PassManagerBase &amp;PM, bool Fast);
};

} // end namespace llvm
</pre>
</div>

<ul>
<li><tt>getInstrInfo()</tt></li>
<li><tt>getRegisterInfo()</tt></li>
<li><tt>getFrameInfo()</tt></li>
<li><tt>getDataLayout()</tt></li>
<li><tt>getSubtargetImpl()</tt></li>
</ul>

<p>For some targets, you also need to support the following methods:</p>

<ul>
<li><tt>getTargetLowering()</tt></li>
<li><tt>getJITInfo()</tt></li>
</ul>

<p>
In addition, the <tt>XXXTargetMachine</tt> constructor should specify a
<tt>TargetDescription</tt> string that determines the data layout for the target
machine, including characteristics such as pointer size, alignment, and
endianness. For example, the constructor for SparcTargetMachine contains the
following:
</p>

<div class="doc_code">
<pre>
SparcTargetMachine::SparcTargetMachine(const Module &amp;M, const std::string &amp;FS)
  : DataLayout("E-p:32:32-f128:128:128"),
    Subtarget(M, FS), InstrInfo(Subtarget),
    FrameInfo(TargetFrameInfo::StackGrowsDown, 8, 0) {
}
</pre>
</div>

<p>Hyphens separate portions of the <tt>TargetDescription</tt> string.</p>

<ul>
<li>An upper-case "<tt>E</tt>" in the string indicates a big-endian target data
    model. a lower-case "<tt>e</tt>" indicates little-endian.</li>

<li>"<tt>p:</tt>" is followed by pointer information: size, ABI alignment, and
    preferred alignment. If only two figures follow "<tt>p:</tt>", then the
    first value is pointer size, and the second value is both ABI and preferred
    alignment.</li>

<li>Then a letter for numeric type alignment: "<tt>i</tt>", "<tt>f</tt>",
    "<tt>v</tt>", or "<tt>a</tt>" (corresponding to integer, floating point,
    vector, or aggregate). "<tt>i</tt>", "<tt>v</tt>", or "<tt>a</tt>" are
    followed by ABI alignment and preferred alignment. "<tt>f</tt>" is followed
    by three values: the first indicates the size of a long double, then ABI
    alignment, and then ABI preferred alignment.</li>
</ul>

</div>

<!-- *********************************************************************** -->
<h2>
  <a name="TargetRegistration">Target Registration</a>
</h2>
<!-- *********************************************************************** -->

<div>

<p>
You must also register your target with the <tt>TargetRegistry</tt>, which is
what other LLVM tools use to be able to lookup and use your target at
runtime. The <tt>TargetRegistry</tt> can be used directly, but for most targets
there are helper templates which should take care of the work for you.</p>

<p>
All targets should declare a global <tt>Target</tt> object which is used to
represent the target during registration. Then, in the target's TargetInfo
library, the target should define that object and use
the <tt>RegisterTarget</tt> template to register the target. For example, the Sparc registration code looks like this:
</p>

<div class="doc_code">
<pre>
Target llvm::TheSparcTarget;

extern "C" void LLVMInitializeSparcTargetInfo() { 
  RegisterTarget&lt;Triple::sparc, /*HasJIT=*/false&gt;
    X(TheSparcTarget, "sparc", "Sparc");
}
</pre>
</div>

<p>
This allows the <tt>TargetRegistry</tt> to look up the target by name or by
target triple. In addition, most targets will also register additional features
which are available in separate libraries. These registration steps are
separate, because some clients may wish to only link in some parts of the target
-- the JIT code generator does not require the use of the assembler printer, for
example. Here is an example of registering the Sparc assembly printer:
</p>

<div class="doc_code">
<pre>
extern "C" void LLVMInitializeSparcAsmPrinter() { 
  RegisterAsmPrinter&lt;SparcAsmPrinter&gt; X(TheSparcTarget);
}
</pre>
</div>

<p>
For more information, see
"<a href="/doxygen/TargetRegistry_8h-source.html">llvm/Target/TargetRegistry.h</a>".
</p>

</div>

<!-- *********************************************************************** -->
<h2>
  <a name="RegisterSet">Register Set and Register Classes</a>
</h2>
<!-- *********************************************************************** -->

<div>

<p>
You should describe a concrete target-specific class that represents the
register file of a target machine. This class is called <tt>XXXRegisterInfo</tt>
(where <tt>XXX</tt> identifies the target) and represents the class register
file data that is used for register allocation. It also describes the
interactions between registers.
</p>

<p>
You also need to define register classes to categorize related registers. A
register class should be added for groups of registers that are all treated the
same way for some instruction. Typical examples are register classes for
integer, floating-point, or vector registers. A register allocator allows an
instruction to use any register in a specified register class to perform the
instruction in a similar manner. Register classes allocate virtual registers to
instructions from these sets, and register classes let the target-independent
register allocator automatically choose the actual registers.
</p>

<p>
Much of the code for registers, including register definition, register aliases,
and register classes, is generated by TableGen from <tt>XXXRegisterInfo.td</tt>
input files and placed in <tt>XXXGenRegisterInfo.h.inc</tt> and
<tt>XXXGenRegisterInfo.inc</tt> output files. Some of the code in the
implementation of <tt>XXXRegisterInfo</tt> requires hand-coding.
</p>

<!-- ======================================================================= -->
<h3>
  <a name="RegisterDef">Defining a Register</a>
</h3>

<div>

<p>
The <tt>XXXRegisterInfo.td</tt> file typically starts with register definitions
for a target machine. The <tt>Register</tt> class (specified
in <tt>Target.td</tt>) is used to define an object for each register. The
specified string <tt>n</tt> becomes the <tt>Name</tt> of the register. The
basic <tt>Register</tt> object does not have any subregisters and does not
specify any aliases.
</p>

<div class="doc_code">
<pre>
class Register&lt;string n&gt; {
  string Namespace = "";
  string AsmName = n;
  string Name = n;
  int SpillSize = 0;
  int SpillAlignment = 0;
  list&lt;Register&gt; Aliases = [];
  list&lt;Register&gt; SubRegs = [];
  list&lt;int&gt; DwarfNumbers = [];
}
</pre>
</div>

<p>
For example, in the <tt>X86RegisterInfo.td</tt> file, there are register
definitions that utilize the Register class, such as:
</p>

<div class="doc_code">
<pre>
def AL : Register&lt;"AL"&gt;, DwarfRegNum&lt;[0, 0, 0]&gt;;
</pre>
</div>

<p>
This defines the register <tt>AL</tt> and assigns it values (with
<tt>DwarfRegNum</tt>) that are used by <tt>gcc</tt>, <tt>gdb</tt>, or a debug
information writer to identify a register. For register
<tt>AL</tt>, <tt>DwarfRegNum</tt> takes an array of 3 values representing 3
different modes: the first element is for X86-64, the second for exception
handling (EH) on X86-32, and the third is generic. -1 is a special Dwarf number
that indicates the gcc number is undefined, and -2 indicates the register number
is invalid for this mode.
</p>

<p>
From the previously described line in the <tt>X86RegisterInfo.td</tt> file,
TableGen generates this code in the <tt>X86GenRegisterInfo.inc</tt> file:
</p>

<div class="doc_code">
<pre>
static const unsigned GR8[] = { X86::AL, ... };

const unsigned AL_AliasSet[] = { X86::AX, X86::EAX, X86::RAX, 0 };

const TargetRegisterDesc RegisterDescriptors[] = { 
  ...
{ "AL", "AL", AL_AliasSet, Empty_SubRegsSet, Empty_SubRegsSet, AL_SuperRegsSet }, ...
</pre>
</div>

<p>
From the register info file, TableGen generates a <tt>TargetRegisterDesc</tt>
object for each register. <tt>TargetRegisterDesc</tt> is defined in
<tt>include/llvm/Target/TargetRegisterInfo.h</tt> with the following fields:
</p>

<div class="doc_code">
<pre>
struct TargetRegisterDesc {
  const char     *AsmName;      // Assembly language name for the register
  const char     *Name;         // Printable name for the reg (for debugging)
  const unsigned *AliasSet;     // Register Alias Set
  const unsigned *SubRegs;      // Sub-register set
  const unsigned *ImmSubRegs;   // Immediate sub-register set
  const unsigned *SuperRegs;    // Super-register set
};</pre>
</div>

<p>
TableGen uses the entire target description file (<tt>.td</tt>) to determine
text names for the register (in the <tt>AsmName</tt> and <tt>Name</tt> fields of
<tt>TargetRegisterDesc</tt>) and the relationships of other registers to the
defined register (in the other <tt>TargetRegisterDesc</tt> fields). In this
example, other definitions establish the registers "<tt>AX</tt>",
"<tt>EAX</tt>", and "<tt>RAX</tt>" as aliases for one another, so TableGen
generates a null-terminated array (<tt>AL_AliasSet</tt>) for this register alias
set.
</p>

<p>
The <tt>Register</tt> class is commonly used as a base class for more complex
classes. In <tt>Target.td</tt>, the <tt>Register</tt> class is the base for the
<tt>RegisterWithSubRegs</tt> class that is used to define registers that need to
specify subregisters in the <tt>SubRegs</tt> list, as shown here:
</p>

<div class="doc_code">
<pre>
class RegisterWithSubRegs&lt;string n,
list&lt;Register&gt; subregs&gt; : Register&lt;n&gt; {
  let SubRegs = subregs;
}
</pre>
</div>

<p>
In <tt>SparcRegisterInfo.td</tt>, additional register classes are defined for
SPARC: a Register subclass, SparcReg, and further subclasses: <tt>Ri</tt>,
<tt>Rf</tt>, and <tt>Rd</tt>. SPARC registers are identified by 5-bit ID
numbers, which is a feature common to these subclasses. Note the use of
'<tt>let</tt>' expressions to override values that are initially defined in a
superclass (such as <tt>SubRegs</tt> field in the <tt>Rd</tt> class).
</p>

<div class="doc_code">
<pre>
class SparcReg&lt;string n&gt; : Register&lt;n&gt; {
  field bits&lt;5&gt; Num;
  let Namespace = "SP";
}
// Ri - 32-bit integer registers
class Ri&lt;bits&lt;5&gt; num, string n&gt; :
SparcReg&lt;n&gt; {
  let Num = num;
}
// Rf - 32-bit floating-point registers
class Rf&lt;bits&lt;5&gt; num, string n&gt; :
SparcReg&lt;n&gt; {
  let Num = num;
}
// Rd - Slots in the FP register file for 64-bit
floating-point values.
class Rd&lt;bits&lt;5&gt; num, string n,
list&lt;Register&gt; subregs&gt; : SparcReg&lt;n&gt; {
  let Num = num;
  let SubRegs = subregs;
}
</pre>
</div>

<p>
In the <tt>SparcRegisterInfo.td</tt> file, there are register definitions that
utilize these subclasses of <tt>Register</tt>, such as:
</p>

<div class="doc_code">
<pre>
def G0 : Ri&lt; 0, "G0"&gt;,
DwarfRegNum&lt;[0]&gt;;
def G1 : Ri&lt; 1, "G1"&gt;, DwarfRegNum&lt;[1]&gt;;
...
def F0 : Rf&lt; 0, "F0"&gt;,
DwarfRegNum&lt;[32]&gt;;
def F1 : Rf&lt; 1, "F1"&gt;,
DwarfRegNum&lt;[33]&gt;;
...
def D0 : Rd&lt; 0, "F0", [F0, F1]&gt;,
DwarfRegNum&lt;[32]&gt;;
def D1 : Rd&lt; 2, "F2", [F2, F3]&gt;,
DwarfRegNum&lt;[34]&gt;;
</pre>
</div>

<p>
The last two registers shown above (<tt>D0</tt> and <tt>D1</tt>) are
double-precision floating-point registers that are aliases for pairs of
single-precision floating-point sub-registers. In addition to aliases, the
sub-register and super-register relationships of the defined register are in
fields of a register's TargetRegisterDesc.
</p>

</div>

<!-- ======================================================================= -->
<h3>
  <a name="RegisterClassDef">Defining a Register Class</a>
</h3>

<div>

<p>
The <tt>RegisterClass</tt> class (specified in <tt>Target.td</tt>) is used to
define an object that represents a group of related registers and also defines
the default allocation order of the registers. A target description file
<tt>XXXRegisterInfo.td</tt> that uses <tt>Target.td</tt> can construct register
classes using the following class:
</p>

<div class="doc_code">
<pre>
class RegisterClass&lt;string namespace,
list&lt;ValueType&gt; regTypes, int alignment, dag regList&gt; {
  string Namespace = namespace;
  list&lt;ValueType&gt; RegTypes = regTypes;
  int Size = 0;  // spill size, in bits; zero lets tblgen pick the size
  int Alignment = alignment;

  // CopyCost is the cost of copying a value between two registers
  // default value 1 means a single instruction
  // A negative value means copying is extremely expensive or impossible
  int CopyCost = 1;  
  dag MemberList = regList;
  
  // for register classes that are subregisters of this class
  list&lt;RegisterClass&gt; SubRegClassList = [];  
  
  code MethodProtos = [{}];  // to insert arbitrary code
  code MethodBodies = [{}];
}
</pre>
</div>

<p>To define a RegisterClass, use the following 4 arguments:</p>

<ul>
<li>The first argument of the definition is the name of the namespace.</li>

<li>The second argument is a list of <tt>ValueType</tt> register type values
    that are defined in <tt>include/llvm/CodeGen/ValueTypes.td</tt>. Defined
    values include integer types (such as <tt>i16</tt>, <tt>i32</tt>,
    and <tt>i1</tt> for Boolean), floating-point types
    (<tt>f32</tt>, <tt>f64</tt>), and vector types (for example, <tt>v8i16</tt>
    for an <tt>8 x i16</tt> vector). All registers in a <tt>RegisterClass</tt>
    must have the same <tt>ValueType</tt>, but some registers may store vector
    data in different configurations. For example a register that can process a
    128-bit vector may be able to handle 16 8-bit integer elements, 8 16-bit
    integers, 4 32-bit integers, and so on. </li>

<li>The third argument of the <tt>RegisterClass</tt> definition specifies the
    alignment required of the registers when they are stored or loaded to
    memory.</li>

<li>The final argument, <tt>regList</tt>, specifies which registers are in this
    class. If an alternative allocation order method is not specified, then
    <tt>regList</tt> also defines the order of allocation used by the register
    allocator. Besides simply listing registers with <tt>(add R0, R1, ...)</tt>,
    more advanced set operators are available. See
    <tt>include/llvm/Target/Target.td</tt> for more information.</li>
</ul>

<p>
In <tt>SparcRegisterInfo.td</tt>, three RegisterClass objects are defined:
<tt>FPRegs</tt>, <tt>DFPRegs</tt>, and <tt>IntRegs</tt>. For all three register
classes, the first argument defines the namespace with the string
'<tt>SP</tt>'. <tt>FPRegs</tt> defines a group of 32 single-precision
floating-point registers (<tt>F0</tt> to <tt>F31</tt>); <tt>DFPRegs</tt> defines
a group of 16 double-precision registers
(<tt>D0-D15</tt>).
</p>

<div class="doc_code">
<pre>
// F0, F1, F2, ..., F31
def FPRegs : RegisterClass&lt;"SP", [f32], 32, (sequence "F%u", 0, 31)&gt;;

def DFPRegs : RegisterClass&lt;"SP", [f64], 64,
                            (add D0, D1, D2, D3, D4, D5, D6, D7, D8,
                                 D9, D10, D11, D12, D13, D14, D15)&gt;;
&nbsp;
def IntRegs : RegisterClass&lt;"SP", [i32], 32,
    (add L0, L1, L2, L3, L4, L5, L6, L7,
         I0, I1, I2, I3, I4, I5,
         O0, O1, O2, O3, O4, O5, O7,
         G1,
         // Non-allocatable regs:
         G2, G3, G4,
         O6,        // stack ptr
         I6,        // frame ptr
         I7,        // return address
         G0,        // constant zero
         G5, G6, G7 // reserved for kernel
    )&gt;;
</pre>
</div>

<p>
Using <tt>SparcRegisterInfo.td</tt> with TableGen generates several output files
that are intended for inclusion in other source code that you write.
<tt>SparcRegisterInfo.td</tt> generates <tt>SparcGenRegisterInfo.h.inc</tt>,
which should be included in the header file for the implementation of the SPARC
register implementation that you write (<tt>SparcRegisterInfo.h</tt>). In
<tt>SparcGenRegisterInfo.h.inc</tt> a new structure is defined called
<tt>SparcGenRegisterInfo</tt> that uses <tt>TargetRegisterInfo</tt> as its
base. It also specifies types, based upon the defined register
classes: <tt>DFPRegsClass</tt>, <tt>FPRegsClass</tt>, and <tt>IntRegsClass</tt>.
</p>

<p>
<tt>SparcRegisterInfo.td</tt> also generates <tt>SparcGenRegisterInfo.inc</tt>,
which is included at the bottom of <tt>SparcRegisterInfo.cpp</tt>, the SPARC
register implementation. The code below shows only the generated integer
registers and associated register classes. The order of registers
in <tt>IntRegs</tt> reflects the order in the definition of <tt>IntRegs</tt> in
the target description file.
</p>

<div class="doc_code">
<pre>  // IntRegs Register Class...
  static const unsigned IntRegs[] = {
    SP::L0, SP::L1, SP::L2, SP::L3, SP::L4, SP::L5,
    SP::L6, SP::L7, SP::I0, SP::I1, SP::I2, SP::I3,
    SP::I4, SP::I5, SP::O0, SP::O1, SP::O2, SP::O3,
    SP::O4, SP::O5, SP::O7, SP::G1, SP::G2, SP::G3,
    SP::G4, SP::O6, SP::I6, SP::I7, SP::G0, SP::G5,
    SP::G6, SP::G7, 
  };

  // IntRegsVTs Register Class Value Types...
  static const MVT::ValueType IntRegsVTs[] = {
    MVT::i32, MVT::Other
  };

namespace SP {   // Register class instances
  DFPRegsClass&nbsp;&nbsp;&nbsp; DFPRegsRegClass;
  FPRegsClass&nbsp;&nbsp;&nbsp;&nbsp; FPRegsRegClass;
  IntRegsClass&nbsp;&nbsp;&nbsp; IntRegsRegClass;
...
  // IntRegs Sub-register Classess...
  static const TargetRegisterClass* const IntRegsSubRegClasses [] = {
    NULL
  };
...
  // IntRegs Super-register Classess...
  static const TargetRegisterClass* const IntRegsSuperRegClasses [] = {
    NULL
  };
...
  // IntRegs Register Class sub-classes...
  static const TargetRegisterClass* const IntRegsSubclasses [] = {
    NULL
  };
...
  // IntRegs Register Class super-classes...
  static const TargetRegisterClass* const IntRegsSuperclasses [] = {
    NULL
  };

  IntRegsClass::IntRegsClass() : TargetRegisterClass(IntRegsRegClassID, 
    IntRegsVTs, IntRegsSubclasses, IntRegsSuperclasses, IntRegsSubRegClasses, 
    IntRegsSuperRegClasses, 4, 4, 1, IntRegs, IntRegs + 32) {}
}
</pre>
</div>

<p>
The register allocators will avoid using reserved registers, and callee saved
registers are not used until all the volatile registers have been used.  That
is usually good enough, but in some cases it may be necessary to provide custom
allocation orders.
</p>

</div>

<!-- ======================================================================= -->
<h3>
  <a name="implementRegister">Implement a subclass of</a> 
  <a href="CodeGenerator.html#targetregisterinfo">TargetRegisterInfo</a>
</h3>

<div>

<p>
The final step is to hand code portions of <tt>XXXRegisterInfo</tt>, which
implements the interface described in <tt>TargetRegisterInfo.h</tt>. These
functions return <tt>0</tt>, <tt>NULL</tt>, or <tt>false</tt>, unless
overridden. Here is a list of functions that are overridden for the SPARC
implementation in <tt>SparcRegisterInfo.cpp</tt>:
</p>

<ul>
<li><tt>getCalleeSavedRegs</tt> &mdash; Returns a list of callee-saved registers
    in the order of the desired callee-save stack frame offset.</li>

<li><tt>getReservedRegs</tt> &mdash; Returns a bitset indexed by physical
    register numbers, indicating if a particular register is unavailable.</li>

<li><tt>hasFP</tt> &mdash; Return a Boolean indicating if a function should have
    a dedicated frame pointer register.</li>

<li><tt>eliminateCallFramePseudoInstr</tt> &mdash; If call frame setup or
    destroy pseudo instructions are used, this can be called to eliminate
    them.</li>

<li><tt>eliminateFrameIndex</tt> &mdash; Eliminate abstract frame indices from
    instructions that may use them.</li>

<li><tt>emitPrologue</tt> &mdash; Insert prologue code into the function.</li>

<li><tt>emitEpilogue</tt> &mdash; Insert epilogue code into the function.</li>
</ul>

</div>

</div>

<!-- *********************************************************************** -->
<h2>
  <a name="InstructionSet">Instruction Set</a>
</h2>

<!-- *********************************************************************** -->
<div>

<p>
During the early stages of code generation, the LLVM IR code is converted to a
<tt>SelectionDAG</tt> with nodes that are instances of the <tt>SDNode</tt> class
containing target instructions. An <tt>SDNode</tt> has an opcode, operands, type
requirements, and operation properties. For example, is an operation
commutative, does an operation load from memory. The various operation node
types are described in the <tt>include/llvm/CodeGen/SelectionDAGNodes.h</tt>
file (values of the <tt>NodeType</tt> enum in the <tt>ISD</tt> namespace).
</p>

<p>
TableGen uses the following target description (<tt>.td</tt>) input files to
generate much of the code for instruction definition:
</p>

<ul>
<li><tt>Target.td</tt> &mdash; Where the <tt>Instruction</tt>, <tt>Operand</tt>,
    <tt>InstrInfo</tt>, and other fundamental classes are defined.</li>

<li><tt>TargetSelectionDAG.td</tt>&mdash; Used by <tt>SelectionDAG</tt>
    instruction selection generators, contains <tt>SDTC*</tt> classes (selection
    DAG type constraint), definitions of <tt>SelectionDAG</tt> nodes (such as
    <tt>imm</tt>, <tt>cond</tt>, <tt>bb</tt>, <tt>add</tt>, <tt>fadd</tt>,
    <tt>sub</tt>), and pattern support (<tt>Pattern</tt>, <tt>Pat</tt>,
    <tt>PatFrag</tt>, <tt>PatLeaf</tt>, <tt>ComplexPattern</tt>.</li>

<li><tt>XXXInstrFormats.td</tt> &mdash; Patterns for definitions of
    target-specific instructions.</li>

<li><tt>XXXInstrInfo.td</tt> &mdash; Target-specific definitions of instruction
    templates, condition codes, and instructions of an instruction set. For
    architecture modifications, a different file name may be used. For example,
    for Pentium with SSE instruction, this file is <tt>X86InstrSSE.td</tt>, and
    for Pentium with MMX, this file is <tt>X86InstrMMX.td</tt>.</li>
</ul>

<p>
There is also a target-specific <tt>XXX.td</tt> file, where <tt>XXX</tt> is the
name of the target. The <tt>XXX.td</tt> file includes the other <tt>.td</tt>
input files, but its contents are only directly important for subtargets.
</p>

<p>
You should describe a concrete target-specific class <tt>XXXInstrInfo</tt> that
represents machine instructions supported by a target machine.
<tt>XXXInstrInfo</tt> contains an array of <tt>XXXInstrDescriptor</tt> objects,
each of which describes one instruction. An instruction descriptor defines:</p>

<ul>
<li>Opcode mnemonic</li>

<li>Number of operands</li>

<li>List of implicit register definitions and uses</li>

<li>Target-independent properties (such as memory access, is commutable)</li>

<li>Target-specific flags </li>
</ul>

<p>
The Instruction class (defined in <tt>Target.td</tt>) is mostly used as a base
for more complex instruction classes.
</p>

<div class="doc_code">
<pre>class Instruction {
  string Namespace = "";
  dag OutOperandList;       // An dag containing the MI def operand list.
  dag InOperandList;        // An dag containing the MI use operand list.
  string AsmString = "";    // The .s format to print the instruction with.
  list&lt;dag&gt; Pattern;  // Set to the DAG pattern for this instruction
  list&lt;Register&gt; Uses = []; 
  list&lt;Register&gt; Defs = [];
  list&lt;Predicate&gt; Predicates = [];  // predicates turned into isel match code
  ... remainder not shown for space ...
}
</pre>
</div>

<p>
A <tt>SelectionDAG</tt> node (<tt>SDNode</tt>) should contain an object
representing a target-specific instruction that is defined
in <tt>XXXInstrInfo.td</tt>. The instruction objects should represent
instructions from the architecture manual of the target machine (such as the
SPARC Architecture Manual for the SPARC target).
</p>

<p>
A single instruction from the architecture manual is often modeled as multiple
target instructions, depending upon its operands. For example, a manual might
describe an add instruction that takes a register or an immediate operand. An
LLVM target could model this with two instructions named <tt>ADDri</tt> and
<tt>ADDrr</tt>.
</p>

<p>
You should define a class for each instruction category and define each opcode
as a subclass of the category with appropriate parameters such as the fixed
binary encoding of opcodes and extended opcodes. You should map the register
bits to the bits of the instruction in which they are encoded (for the
JIT). Also you should specify how the instruction should be printed when the
automatic assembly printer is used.
</p>

<p>
As is described in the SPARC Architecture Manual, Version 8, there are three
major 32-bit formats for instructions. Format 1 is only for the <tt>CALL</tt>
instruction. Format 2 is for branch on condition codes and <tt>SETHI</tt> (set
high bits of a register) instructions.  Format 3 is for other instructions.
</p>

<p>
Each of these formats has corresponding classes in <tt>SparcInstrFormat.td</tt>.
<tt>InstSP</tt> is a base class for other instruction classes. Additional base
classes are specified for more precise formats: for example
in <tt>SparcInstrFormat.td</tt>, <tt>F2_1</tt> is for <tt>SETHI</tt>,
and <tt>F2_2</tt> is for branches. There are three other base
classes: <tt>F3_1</tt> for register/register operations, <tt>F3_2</tt> for
register/immediate operations, and <tt>F3_3</tt> for floating-point
operations. <tt>SparcInstrInfo.td</tt> also adds the base class Pseudo for
synthetic SPARC instructions.
</p>

<p>
<tt>SparcInstrInfo.td</tt> largely consists of operand and instruction
definitions for the SPARC target. In <tt>SparcInstrInfo.td</tt>, the following
target description file entry, <tt>LDrr</tt>, defines the Load Integer
instruction for a Word (the <tt>LD</tt> SPARC opcode) from a memory address to a
register. The first parameter, the value 3 (<tt>11<sub>2</sub></tt>), is the
operation value for this category of operation. The second parameter
(<tt>000000<sub>2</sub></tt>) is the specific operation value
for <tt>LD</tt>/Load Word. The third parameter is the output destination, which
is a register operand and defined in the <tt>Register</tt> target description
file (<tt>IntRegs</tt>).
</p>

<div class="doc_code">
<pre>def LDrr : F3_1 &lt;3, 0b000000, (outs IntRegs:$dst), (ins MEMrr:$addr),
                 "ld [$addr], $dst",
                 [(set IntRegs:$dst, (load ADDRrr:$addr))]&gt;;
</pre>
</div>

<p>
The fourth parameter is the input source, which uses the address
operand <tt>MEMrr</tt> that is defined earlier in <tt>SparcInstrInfo.td</tt>:
</p>

<div class="doc_code">
<pre>def MEMrr : Operand&lt;i32&gt; {
  let PrintMethod = "printMemOperand";
  let MIOperandInfo = (ops IntRegs, IntRegs);
}
</pre>
</div>

<p>
The fifth parameter is a string that is used by the assembly printer and can be
left as an empty string until the assembly printer interface is implemented. The
sixth and final parameter is the pattern used to match the instruction during
the SelectionDAG Select Phase described in
(<a href="CodeGenerator.html">The LLVM
Target-Independent Code Generator</a>).  This parameter is detailed in the next
section, <a href="#InstructionSelector">Instruction Selector</a>.
</p>

<p>
Instruction class definitions are not overloaded for different operand types, so
separate versions of instructions are needed for register, memory, or immediate
value operands. For example, to perform a Load Integer instruction for a Word
from an immediate operand to a register, the following instruction class is
defined:
</p>

<div class="doc_code">
<pre>def LDri : F3_2 &lt;3, 0b000000, (outs IntRegs:$dst), (ins MEMri:$addr),
                 "ld [$addr], $dst",
                 [(set IntRegs:$dst, (load ADDRri:$addr))]&gt;;
</pre>
</div>

<p>
Writing these definitions for so many similar instructions can involve a lot of
cut and paste. In td files, the <tt>multiclass</tt> directive enables the
creation of templates to define several instruction classes at once (using
the <tt>defm</tt> directive). For example in <tt>SparcInstrInfo.td</tt>, the
<tt>multiclass</tt> pattern <tt>F3_12</tt> is defined to create 2 instruction
classes each time <tt>F3_12</tt> is invoked:
</p>

<div class="doc_code">
<pre>multiclass F3_12 &lt;string OpcStr, bits&lt;6&gt; Op3Val, SDNode OpNode&gt; {
  def rr  : F3_1 &lt;2, Op3Val, 
                 (outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
                 !strconcat(OpcStr, " $b, $c, $dst"),
                 [(set IntRegs:$dst, (OpNode IntRegs:$b, IntRegs:$c))]&gt;;
  def ri  : F3_2 &lt;2, Op3Val,
                 (outs IntRegs:$dst), (ins IntRegs:$b, i32imm:$c),
                 !strconcat(OpcStr, " $b, $c, $dst"),
                 [(set IntRegs:$dst, (OpNode IntRegs:$b, simm13:$c))]&gt;;
}
</pre>
</div>

<p>
So when the <tt>defm</tt> directive is used for the <tt>XOR</tt>
and <tt>ADD</tt> instructions, as seen below, it creates four instruction
objects: <tt>XORrr</tt>, <tt>XORri</tt>, <tt>ADDrr</tt>, and <tt>ADDri</tt>.
</p>

<div class="doc_code">
<pre>
defm XOR   : F3_12&lt;"xor", 0b000011, xor&gt;;
defm ADD   : F3_12&lt;"add", 0b000000, add&gt;;
</pre>
</div>

<p>
<tt>SparcInstrInfo.td</tt> also includes definitions for condition codes that
are referenced by branch instructions. The following definitions
in <tt>SparcInstrInfo.td</tt> indicate the bit location of the SPARC condition
code. For example, the 10<sup>th</sup> bit represents the 'greater than'
condition for integers, and the 22<sup>nd</sup> bit represents the 'greater
than' condition for floats.
</p>

<div class="doc_code">
<pre>
def ICC_NE  : ICC_VAL&lt; 9&gt;;  // Not Equal
def ICC_E   : ICC_VAL&lt; 1&gt;;  // Equal
def ICC_G   : ICC_VAL&lt;10&gt;;  // Greater
...
def FCC_U   : FCC_VAL&lt;23&gt;;  // Unordered
def FCC_G   : FCC_VAL&lt;22&gt;;  // Greater
def FCC_UG  : FCC_VAL&lt;21&gt;;  // Unordered or Greater
...
</pre>
</div>

<p>
(Note that <tt>Sparc.h</tt> also defines enums that correspond to the same SPARC
condition codes. Care must be taken to ensure the values in <tt>Sparc.h</tt>
correspond to the values in <tt>SparcInstrInfo.td</tt>. I.e.,
<tt>SPCC::ICC_NE = 9</tt>, <tt>SPCC::FCC_U = 23</tt> and so on.)
</p>

<!-- ======================================================================= -->
<h3>
  <a name="operandMapping">Instruction Operand Mapping</a>
</h3>

<div>

<p>
The code generator backend maps instruction operands to fields in the
instruction.  Operands are assigned to unbound fields in the instruction in the
order they are defined. Fields are bound when they are assigned a value.  For
example, the Sparc target defines the <tt>XNORrr</tt> instruction as
a <tt>F3_1</tt> format instruction having three operands.
</p>

<div class="doc_code">
<pre>
def XNORrr  : F3_1&lt;2, 0b000111,
                   (outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
                   "xnor $b, $c, $dst",
                   [(set IntRegs:$dst, (not (xor IntRegs:$b, IntRegs:$c)))]&gt;;
</pre>
</div>

<p>
The instruction templates in <tt>SparcInstrFormats.td</tt> show the base class
for <tt>F3_1</tt> is <tt>InstSP</tt>.
</p>

<div class="doc_code">
<pre>
class InstSP&lt;dag outs, dag ins, string asmstr, list&lt;dag&gt; pattern&gt; : Instruction {
  field bits&lt;32&gt; Inst;
  let Namespace = "SP";
  bits&lt;2&gt; op;
  let Inst{31-30} = op;       
  dag OutOperandList = outs;
  dag InOperandList = ins;
  let AsmString   = asmstr;
  let Pattern = pattern;
}
</pre>
</div>

<p><tt>InstSP</tt> leaves the <tt>op</tt> field unbound.</p>

<div class="doc_code">
<pre>
class F3&lt;dag outs, dag ins, string asmstr, list&lt;dag&gt; pattern&gt;
    : InstSP&lt;outs, ins, asmstr, pattern&gt; {
  bits&lt;5&gt; rd;
  bits&lt;6&gt; op3;
  bits&lt;5&gt; rs1;
  let op{1} = 1;   // Op = 2 or 3
  let Inst{29-25} = rd;
  let Inst{24-19} = op3;
  let Inst{18-14} = rs1;
}
</pre>
</div>

<p>
<tt>F3</tt> binds the <tt>op</tt> field and defines the <tt>rd</tt>,
<tt>op3</tt>, and <tt>rs1</tt> fields.  <tt>F3</tt> format instructions will
bind the operands <tt>rd</tt>, <tt>op3</tt>, and <tt>rs1</tt> fields.
</p>

<div class="doc_code">
<pre>
class F3_1&lt;bits&lt;2&gt; opVal, bits&lt;6&gt; op3val, dag outs, dag ins,
           string asmstr, list&lt;dag&gt; pattern&gt; : F3&lt;outs, ins, asmstr, pattern&gt; {
  bits&lt;8&gt; asi = 0; // asi not currently used
  bits&lt;5&gt; rs2;
  let op         = opVal;
  let op3        = op3val;
  let Inst{13}   = 0;     // i field = 0
  let Inst{12-5} = asi;   // address space identifier
  let Inst{4-0}  = rs2;
}
</pre>
</div>

<p>
<tt>F3_1</tt> binds the <tt>op3</tt> field and defines the <tt>rs2</tt>
fields.  <tt>F3_1</tt> format instructions will bind the operands to the <tt>rd</tt>,
<tt>rs1</tt>, and <tt>rs2</tt> fields. This results in the <tt>XNORrr</tt>
instruction binding <tt>$dst</tt>, <tt>$b</tt>, and <tt>$c</tt> operands to
the <tt>rd</tt>, <tt>rs1</tt>, and <tt>rs2</tt> fields respectively.
</p>

</div>

<!-- ======================================================================= -->
<h3>
  <a name="relationMapping">Instruction Relation Mapping</a>
</h3>

<div>

<p>
This TableGen feature is used to relate instructions with each other. It is
particularly useful when you have multiple instruction formats and need to
switch between them after instruction selection. This entire feature is driven
by relation models which can be defined in <tt>XXXInstrInfo.td</tt> files
according to the target-specific instruction set. Relation models are defined
using <tt>InstrMapping</tt> class as a base. TableGen parses all the models
and generates instruction relation maps using the specified information.
Relation maps are emitted as tables in the <tt>XXXGenInstrInfo.inc</tt> file
along with the functions to query them. For the detailed information on how to
use this feature, please refer to
<a href="HowToUseInstrMappings.html">How to add Instruction Mappings</a>
document.
</p>
</div>

<!-- ======================================================================= -->
<h3>
  <a name="implementInstr">Implement a subclass of </a>
  <a href="CodeGenerator.html#targetinstrinfo">TargetInstrInfo</a>
</h3>

<div>

<p>
The final step is to hand code portions of <tt>XXXInstrInfo</tt>, which
implements the interface described in <tt>TargetInstrInfo.h</tt>. These
functions return <tt>0</tt> or a Boolean or they assert, unless
overridden. Here's a list of functions that are overridden for the SPARC
implementation in <tt>SparcInstrInfo.cpp</tt>:
</p>

<ul>
<li><tt>isLoadFromStackSlot</tt> &mdash; If the specified machine instruction is
    a direct load from a stack slot, return the register number of the
    destination and the <tt>FrameIndex</tt> of the stack slot.</li>

<li><tt>isStoreToStackSlot</tt> &mdash; If the specified machine instruction is
    a direct store to a stack slot, return the register number of the
    destination and the <tt>FrameIndex</tt> of the stack slot.</li>

<li><tt>copyPhysReg</tt> &mdash; Copy values between a pair of physical
    registers.</li>

<li><tt>storeRegToStackSlot</tt> &mdash; Store a register value to a stack
    slot.</li>

<li><tt>loadRegFromStackSlot</tt> &mdash; Load a register value from a stack
    slot.</li>

<li><tt>storeRegToAddr</tt> &mdash; Store a register value to memory.</li>

<li><tt>loadRegFromAddr</tt> &mdash; Load a register value from memory.</li>

<li><tt>foldMemoryOperand</tt> &mdash; Attempt to combine instructions of any
    load or store instruction for the specified operand(s).</li>
</ul>

</div>

<!-- ======================================================================= -->
<h3>
  <a name="branchFolding">Branch Folding and If Conversion</a>
</h3>
<div>

<p>
Performance can be improved by combining instructions or by eliminating
instructions that are never reached. The <tt>AnalyzeBranch</tt> method
in <tt>XXXInstrInfo</tt> may be implemented to examine conditional instructions
and remove unnecessary instructions. <tt>AnalyzeBranch</tt> looks at the end of
a machine basic block (MBB) for opportunities for improvement, such as branch
folding and if conversion. The <tt>BranchFolder</tt> and <tt>IfConverter</tt>
machine function passes (see the source files <tt>BranchFolding.cpp</tt> and
<tt>IfConversion.cpp</tt> in the <tt>lib/CodeGen</tt> directory) call
<tt>AnalyzeBranch</tt> to improve the control flow graph that represents the
instructions.
</p>

<p>
Several implementations of <tt>AnalyzeBranch</tt> (for ARM, Alpha, and X86) can
be examined as models for your own <tt>AnalyzeBranch</tt> implementation. Since
SPARC does not implement a useful <tt>AnalyzeBranch</tt>, the ARM target
implementation is shown below.
</p>

<p><tt>AnalyzeBranch</tt> returns a Boolean value and takes four parameters:</p>

<ul>
<li><tt>MachineBasicBlock &amp;MBB</tt> &mdash; The incoming block to be
    examined.</li>

<li><tt>MachineBasicBlock *&amp;TBB</tt> &mdash; A destination block that is
    returned. For a conditional branch that evaluates to true, <tt>TBB</tt> is
    the destination.</li>

<li><tt>MachineBasicBlock *&amp;FBB</tt> &mdash; For a conditional branch that
    evaluates to false, <tt>FBB</tt> is returned as the destination.</li>

<li><tt>std::vector&lt;MachineOperand&gt; &amp;Cond</tt> &mdash; List of
    operands to evaluate a condition for a conditional branch.</li>
</ul>

<p>
In the simplest case, if a block ends without a branch, then it falls through to
the successor block. No destination blocks are specified for either <tt>TBB</tt>
or <tt>FBB</tt>, so both parameters return <tt>NULL</tt>. The start of
the <tt>AnalyzeBranch</tt> (see code below for the ARM target) shows the
function parameters and the code for the simplest case.
</p>

<div class="doc_code">
<pre>bool ARMInstrInfo::AnalyzeBranch(MachineBasicBlock &amp;MBB,
        MachineBasicBlock *&amp;TBB, MachineBasicBlock *&amp;FBB,
        std::vector&lt;MachineOperand&gt; &amp;Cond) const
{
  MachineBasicBlock::iterator I = MBB.end();
  if (I == MBB.begin() || !isUnpredicatedTerminator(--I))
    return false;
</pre>
</div>

<p>
If a block ends with a single unconditional branch instruction, then
<tt>AnalyzeBranch</tt> (shown below) should return the destination of that
branch in the <tt>TBB</tt> parameter.
</p>

<div class="doc_code">
<pre>
  if (LastOpc == ARM::B || LastOpc == ARM::tB) {
    TBB = LastInst-&gt;getOperand(0).getMBB();
    return false;
  }
</pre>
</div>

<p>
If a block ends with two unconditional branches, then the second branch is never
reached. In that situation, as shown below, remove the last branch instruction
and return the penultimate branch in the <tt>TBB</tt> parameter.
</p>

<div class="doc_code">
<pre>
  if ((SecondLastOpc == ARM::B || SecondLastOpc==ARM::tB) &amp;&amp;
      (LastOpc == ARM::B || LastOpc == ARM::tB)) {
    TBB = SecondLastInst-&gt;getOperand(0).getMBB();
    I = LastInst;
    I-&gt;eraseFromParent();
    return false;
  }
</pre>
</div>

<p>
A block may end with a single conditional branch instruction that falls through
to successor block if the condition evaluates to false. In that case,
<tt>AnalyzeBranch</tt> (shown below) should return the destination of that
conditional branch in the <tt>TBB</tt> parameter and a list of operands in
the <tt>Cond</tt> parameter to evaluate the condition.
</p>

<div class="doc_code">
<pre>
  if (LastOpc == ARM::Bcc || LastOpc == ARM::tBcc) {
    // Block ends with fall-through condbranch.
    TBB = LastInst-&gt;getOperand(0).getMBB();
    Cond.push_back(LastInst-&gt;getOperand(1));
    Cond.push_back(LastInst-&gt;getOperand(2));
    return false;
  }
</pre>
</div>

<p>
If a block ends with both a conditional branch and an ensuing unconditional
branch, then <tt>AnalyzeBranch</tt> (shown below) should return the conditional
branch destination (assuming it corresponds to a conditional evaluation of
'<tt>true</tt>') in the <tt>TBB</tt> parameter and the unconditional branch
destination in the <tt>FBB</tt> (corresponding to a conditional evaluation of
'<tt>false</tt>').  A list of operands to evaluate the condition should be
returned in the <tt>Cond</tt> parameter.
</p>

<div class="doc_code">
<pre>
  unsigned SecondLastOpc = SecondLastInst-&gt;getOpcode();

  if ((SecondLastOpc == ARM::Bcc &amp;&amp; LastOpc == ARM::B) ||
      (SecondLastOpc == ARM::tBcc &amp;&amp; LastOpc == ARM::tB)) {
    TBB =  SecondLastInst-&gt;getOperand(0).getMBB();
    Cond.push_back(SecondLastInst-&gt;getOperand(1));
    Cond.push_back(SecondLastInst-&gt;getOperand(2));
    FBB = LastInst-&gt;getOperand(0).getMBB();
    return false;
  }
</pre>
</div>

<p>
For the last two cases (ending with a single conditional branch or ending with
one conditional and one unconditional branch), the operands returned in
the <tt>Cond</tt> parameter can be passed to methods of other instructions to
create new branches or perform other operations. An implementation
of <tt>AnalyzeBranch</tt> requires the helper methods <tt>RemoveBranch</tt>
and <tt>InsertBranch</tt> to manage subsequent operations.
</p>

<p>
<tt>AnalyzeBranch</tt> should return false indicating success in most circumstances.
<tt>AnalyzeBranch</tt> should only return true when the method is stumped about what to
do, for example, if a block has three terminating branches. <tt>AnalyzeBranch</tt> may
return true if it encounters a terminator it cannot handle, such as an indirect
branch.
</p>

</div>

</div>

<!-- *********************************************************************** -->
<h2>
  <a name="InstructionSelector">Instruction Selector</a>
</h2>
<!-- *********************************************************************** -->

<div>

<p>
LLVM uses a <tt>SelectionDAG</tt> to represent LLVM IR instructions, and nodes
of the <tt>SelectionDAG</tt> ideally represent native target
instructions. During code generation, instruction selection passes are performed
to convert non-native DAG instructions into native target-specific
instructions. The pass described in <tt>XXXISelDAGToDAG.cpp</tt> is used to
match patterns and perform DAG-to-DAG instruction selection. Optionally, a pass
may be defined (in <tt>XXXBranchSelector.cpp</tt>) to perform similar DAG-to-DAG
operations for branch instructions. Later, the code in
<tt>XXXISelLowering.cpp</tt> replaces or removes operations and data types not
supported natively (legalizes) in a <tt>SelectionDAG</tt>.
</p>

<p>
TableGen generates code for instruction selection using the following target
description input files:
</p>

<ul>
<li><tt>XXXInstrInfo.td</tt> &mdash; Contains definitions of instructions in a
    target-specific instruction set, generates <tt>XXXGenDAGISel.inc</tt>, which
    is included in <tt>XXXISelDAGToDAG.cpp</tt>.</li>

<li><tt>XXXCallingConv.td</tt> &mdash; Contains the calling and return value
    conventions for the target architecture, and it generates
    <tt>XXXGenCallingConv.inc</tt>, which is included in
    <tt>XXXISelLowering.cpp</tt>.</li>
</ul>

<p>
The implementation of an instruction selection pass must include a header that
declares the <tt>FunctionPass</tt> class or a subclass of <tt>FunctionPass</tt>. In
<tt>XXXTargetMachine.cpp</tt>, a Pass Manager (PM) should add each instruction
selection pass into the queue of passes to run.
</p>

<p>
The LLVM static compiler (<tt>llc</tt>) is an excellent tool for visualizing the
contents of DAGs. To display the <tt>SelectionDAG</tt> before or after specific
processing phases, use the command line options for <tt>llc</tt>, described
at <a href="CodeGenerator.html#selectiondag_process">
SelectionDAG Instruction Selection Process</a>.
</p>

<p>
To describe instruction selector behavior, you should add patterns for lowering
LLVM code into a <tt>SelectionDAG</tt> as the last parameter of the instruction
definitions in <tt>XXXInstrInfo.td</tt>. For example, in
<tt>SparcInstrInfo.td</tt>, this entry defines a register store operation, and
the last parameter describes a pattern with the store DAG operator.
</p>

<div class="doc_code">
<pre>
def STrr  : F3_1&lt; 3, 0b000100, (outs), (ins MEMrr:$addr, IntRegs:$src),
                 "st $src, [$addr]", [(store IntRegs:$src, ADDRrr:$addr)]&gt;;
</pre>
</div>

<p>
<tt>ADDRrr</tt> is a memory mode that is also defined in
<tt>SparcInstrInfo.td</tt>:
</p>

<div class="doc_code">
<pre>
def ADDRrr : ComplexPattern&lt;i32, 2, "SelectADDRrr", [], []&gt;;
</pre>
</div>

<p>
The definition of <tt>ADDRrr</tt> refers to <tt>SelectADDRrr</tt>, which is a
function defined in an implementation of the Instructor Selector (such
as <tt>SparcISelDAGToDAG.cpp</tt>).
</p>

<p>
In <tt>lib/Target/TargetSelectionDAG.td</tt>, the DAG operator for store is
defined below:
</p>

<div class="doc_code">
<pre>
def store : PatFrag&lt;(ops node:$val, node:$ptr),
                    (st node:$val, node:$ptr), [{
  if (StoreSDNode *ST = dyn_cast&lt;StoreSDNode&gt;(N))
    return !ST-&gt;isTruncatingStore() &amp;&amp; 
           ST-&gt;getAddressingMode() == ISD::UNINDEXED;
  return false;
}]&gt;;
</pre>
</div>

<p>
<tt>XXXInstrInfo.td</tt> also generates (in <tt>XXXGenDAGISel.inc</tt>) the
<tt>SelectCode</tt> method that is used to call the appropriate processing
method for an instruction. In this example, <tt>SelectCode</tt>
calls <tt>Select_ISD_STORE</tt> for the <tt>ISD::STORE</tt> opcode.
</p>

<div class="doc_code">
<pre>
SDNode *SelectCode(SDValue N) {
  ... 
  MVT::ValueType NVT = N.getNode()-&gt;getValueType(0);
  switch (N.getOpcode()) {
  case ISD::STORE: {
    switch (NVT) {
    default:
      return Select_ISD_STORE(N);
      break;
    }
    break;
  }
  ...
</pre>
</div>

<p>
The pattern for <tt>STrr</tt> is matched, so elsewhere in
<tt>XXXGenDAGISel.inc</tt>, code for <tt>STrr</tt> is created for
<tt>Select_ISD_STORE</tt>. The <tt>Emit_22</tt> method is also generated
in <tt>XXXGenDAGISel.inc</tt> to complete the processing of this
instruction.
</p>

<div class="doc_code">
<pre>
SDNode *Select_ISD_STORE(const SDValue &amp;N) {
  SDValue Chain = N.getOperand(0);
  if (Predicate_store(N.getNode())) {
    SDValue N1 = N.getOperand(1);
    SDValue N2 = N.getOperand(2);
    SDValue CPTmp0;
    SDValue CPTmp1;

    // Pattern: (st:void IntRegs:i32:$src, 
    //           ADDRrr:i32:$addr)&lt;&lt;P:Predicate_store&gt;&gt;
    // Emits: (STrr:void ADDRrr:i32:$addr, IntRegs:i32:$src)
    // Pattern complexity = 13  cost = 1  size = 0
    if (SelectADDRrr(N, N2, CPTmp0, CPTmp1) &amp;&amp;
        N1.getNode()-&gt;getValueType(0) == MVT::i32 &amp;&amp;
        N2.getNode()-&gt;getValueType(0) == MVT::i32) {
      return Emit_22(N, SP::STrr, CPTmp0, CPTmp1);
    }
...
</pre>
</div>

<!-- ======================================================================= -->
<h3>
  <a name="LegalizePhase">The SelectionDAG Legalize Phase</a>
</h3>

<div>

<p>
The Legalize phase converts a DAG to use types and operations that are natively
supported by the target. For natively unsupported types and operations, you need
to add code to the target-specific XXXTargetLowering implementation to convert
unsupported types and operations to supported ones.
</p>

<p>
In the constructor for the <tt>XXXTargetLowering</tt> class, first use the
<tt>addRegisterClass</tt> method to specify which types are supports and which
register classes are associated with them. The code for the register classes are
generated by TableGen from <tt>XXXRegisterInfo.td</tt> and placed
in <tt>XXXGenRegisterInfo.h.inc</tt>. For example, the implementation of the
constructor for the SparcTargetLowering class (in
<tt>SparcISelLowering.cpp</tt>) starts with the following code:
</p>

<div class="doc_code">
<pre>
addRegisterClass(MVT::i32, SP::IntRegsRegisterClass);
addRegisterClass(MVT::f32, SP::FPRegsRegisterClass);
addRegisterClass(MVT::f64, SP::DFPRegsRegisterClass); 
</pre>
</div>

<p>
You should examine the node types in the <tt>ISD</tt> namespace
(<tt>include/llvm/CodeGen/SelectionDAGNodes.h</tt>) and determine which
operations the target natively supports. For operations that do <b>not</b> have
native support, add a callback to the constructor for the XXXTargetLowering
class, so the instruction selection process knows what to do. The TargetLowering
class callback methods (declared in <tt>llvm/Target/TargetLowering.h</tt>) are:
</p>

<ul>
<li><tt>setOperationAction</tt> &mdash; General operation.</li>

<li><tt>setLoadExtAction</tt> &mdash; Load with extension.</li>

<li><tt>setTruncStoreAction</tt> &mdash; Truncating store.</li>

<li><tt>setIndexedLoadAction</tt> &mdash; Indexed load.</li>

<li><tt>setIndexedStoreAction</tt> &mdash; Indexed store.</li>

<li><tt>setConvertAction</tt> &mdash; Type conversion.</li>

<li><tt>setCondCodeAction</tt> &mdash; Support for a given condition code.</li>
</ul>

<p>
Note: on older releases, <tt>setLoadXAction</tt> is used instead
of <tt>setLoadExtAction</tt>.  Also, on older releases,
<tt>setCondCodeAction</tt> may not be supported. Examine your release
to see what methods are specifically supported.
</p>

<p>
These callbacks are used to determine that an operation does or does not work
with a specified type (or types). And in all cases, the third parameter is
a <tt>LegalAction</tt> type enum value: <tt>Promote</tt>, <tt>Expand</tt>,
<tt>Custom</tt>, or <tt>Legal</tt>. <tt>SparcISelLowering.cpp</tt>
contains examples of all four <tt>LegalAction</tt> values.
</p>

<!-- _______________________________________________________________________ -->
<h4>
  <a name="promote">Promote</a>
</h4>

<div>

<p>
For an operation without native support for a given type, the specified type may
be promoted to a larger type that is supported. For example, SPARC does not
support a sign-extending load for Boolean values (<tt>i1</tt> type), so
in <tt>SparcISelLowering.cpp</tt> the third parameter below, <tt>Promote</tt>,
changes <tt>i1</tt> type values to a large type before loading.
</p>

<div class="doc_code">
<pre>
setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
</pre>
</div>

</div>

<!-- _______________________________________________________________________ -->
<h4>
  <a name="expand">Expand</a>
</h4>

<div>

<p>
For a type without native support, a value may need to be broken down further,
rather than promoted. For an operation without native support, a combination of
other operations may be used to similar effect. In SPARC, the floating-point
sine and cosine trig operations are supported by expansion to other operations,
as indicated by the third parameter, <tt>Expand</tt>, to
<tt>setOperationAction</tt>:
</p>

<div class="doc_code">
<pre>
setOperationAction(ISD::FSIN, MVT::f32, Expand);
setOperationAction(ISD::FCOS, MVT::f32, Expand);
</pre>
</div>

</div>

<!-- _______________________________________________________________________ -->
<h4>
  <a name="custom">Custom</a>
</h4>

<div>

<p>
For some operations, simple type promotion or operation expansion may be
insufficient. In some cases, a special intrinsic function must be implemented.
</p>

<p>
For example, a constant value may require special treatment, or an operation may
require spilling and restoring registers in the stack and working with register
allocators.
</p>

<p>
As seen in <tt>SparcISelLowering.cpp</tt> code below, to perform a type
conversion from a floating point value to a signed integer, first the
<tt>setOperationAction</tt> should be called with <tt>Custom</tt> as the third
parameter:
</p>

<div class="doc_code">
<pre>
setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
</pre>
</div>    

<p>
In the <tt>LowerOperation</tt> method, for each <tt>Custom</tt> operation, a
case statement should be added to indicate what function to call. In the
following code, an <tt>FP_TO_SINT</tt> opcode will call
the <tt>LowerFP_TO_SINT</tt> method:
</p>

<div class="doc_code">
<pre>
SDValue SparcTargetLowering::LowerOperation(SDValue Op, SelectionDAG &amp;DAG) {
  switch (Op.getOpcode()) {
  case ISD::FP_TO_SINT: return LowerFP_TO_SINT(Op, DAG);
  ...
  }
}
</pre>
</div>

<p>
Finally, the <tt>LowerFP_TO_SINT</tt> method is implemented, using an FP
register to convert the floating-point value to an integer.
</p>

<div class="doc_code">
<pre>
static SDValue LowerFP_TO_SINT(SDValue Op, SelectionDAG &amp;DAG) {
  assert(Op.getValueType() == MVT::i32);
  Op = DAG.getNode(SPISD::FTOI, MVT::f32, Op.getOperand(0));
  return DAG.getNode(ISD::BITCAST, MVT::i32, Op);
}
</pre>
</div>    

</div>

<!-- _______________________________________________________________________ -->
<h4>
  <a name="legal">Legal</a>
</h4>

<div>

<p>
The <tt>Legal</tt> LegalizeAction enum value simply indicates that an
operation <b>is</b> natively supported. <tt>Legal</tt> represents the default
condition, so it is rarely used. In <tt>SparcISelLowering.cpp</tt>, the action
for <tt>CTPOP</tt> (an operation to count the bits set in an integer) is
natively supported only for SPARC v9. The following code enables
the <tt>Expand</tt> conversion technique for non-v9 SPARC implementations.
</p>

<div class="doc_code">
<pre>
setOperationAction(ISD::CTPOP, MVT::i32, Expand);
...
if (TM.getSubtarget&lt;SparcSubtarget&gt;().isV9())
  setOperationAction(ISD::CTPOP, MVT::i32, Legal);
  case ISD::SETULT: return SPCC::ICC_CS;
  case ISD::SETULE: return SPCC::ICC_LEU;
  case ISD::SETUGT: return SPCC::ICC_GU;
  case ISD::SETUGE: return SPCC::ICC_CC;
  }
}
</pre>
</div>

</div>

</div>

<!-- ======================================================================= -->
<h3>
  <a name="callingConventions">Calling Conventions</a>
</h3>

<div>

<p>
To support target-specific calling conventions, <tt>XXXGenCallingConv.td</tt>
uses interfaces (such as CCIfType and CCAssignToReg) that are defined in
<tt>lib/Target/TargetCallingConv.td</tt>. TableGen can take the target
descriptor file <tt>XXXGenCallingConv.td</tt> and generate the header
file <tt>XXXGenCallingConv.inc</tt>, which is typically included
in <tt>XXXISelLowering.cpp</tt>. You can use the interfaces in
<tt>TargetCallingConv.td</tt> to specify:
</p>

<ul>
<li>The order of parameter allocation.</li>

<li>Where parameters and return values are placed (that is, on the stack or in
    registers).</li>

<li>Which registers may be used.</li>

<li>Whether the caller or callee unwinds the stack.</li>
</ul>

<p>
The following example demonstrates the use of the <tt>CCIfType</tt> and
<tt>CCAssignToReg</tt> interfaces. If the <tt>CCIfType</tt> predicate is true
(that is, if the current argument is of type <tt>f32</tt> or <tt>f64</tt>), then
the action is performed. In this case, the <tt>CCAssignToReg</tt> action assigns
the argument value to the first available register: either <tt>R0</tt>
or <tt>R1</tt>.
</p>

<div class="doc_code">
<pre>
CCIfType&lt;[f32,f64], CCAssignToReg&lt;[R0, R1]&gt;&gt;
</pre>
</div>

<p>
<tt>SparcCallingConv.td</tt> contains definitions for a target-specific
return-value calling convention (RetCC_Sparc32) and a basic 32-bit C calling
convention (<tt>CC_Sparc32</tt>). The definition of <tt>RetCC_Sparc32</tt>
(shown below) indicates which registers are used for specified scalar return
types. A single-precision float is returned to register <tt>F0</tt>, and a
double-precision float goes to register <tt>D0</tt>. A 32-bit integer is
returned in register <tt>I0</tt> or <tt>I1</tt>.
</p>

<div class="doc_code">
<pre>
def RetCC_Sparc32 : CallingConv&lt;[
  CCIfType&lt;[i32], CCAssignToReg&lt;[I0, I1]&gt;&gt;,
  CCIfType&lt;[f32], CCAssignToReg&lt;[F0]&gt;&gt;,
  CCIfType&lt;[f64], CCAssignToReg&lt;[D0]&gt;&gt;
]&gt;;
</pre>
</div>

<p>
The definition of <tt>CC_Sparc32</tt> in <tt>SparcCallingConv.td</tt> introduces
<tt>CCAssignToStack</tt>, which assigns the value to a stack slot with the
specified size and alignment. In the example below, the first parameter, 4,
indicates the size of the slot, and the second parameter, also 4, indicates the
stack alignment along 4-byte units. (Special cases: if size is zero, then the
ABI size is used; if alignment is zero, then the ABI alignment is used.)
</p>

<div class="doc_code">
<pre>
def CC_Sparc32 : CallingConv&lt;[
  // All arguments get passed in integer registers if there is space.
  CCIfType&lt;[i32, f32, f64], CCAssignToReg&lt;[I0, I1, I2, I3, I4, I5]&gt;&gt;,
  CCAssignToStack&lt;4, 4&gt;
]&gt;;
</pre>
</div>

<p>
<tt>CCDelegateTo</tt> is another commonly used interface, which tries to find a
specified sub-calling convention, and, if a match is found, it is invoked. In
the following example (in <tt>X86CallingConv.td</tt>), the definition of
<tt>RetCC_X86_32_C</tt> ends with <tt>CCDelegateTo</tt>. After the current value
is assigned to the register <tt>ST0</tt> or <tt>ST1</tt>,
the <tt>RetCC_X86Common</tt> is invoked.
</p>

<div class="doc_code">
<pre>
def RetCC_X86_32_C : CallingConv&lt;[
  CCIfType&lt;[f32], CCAssignToReg&lt;[ST0, ST1]&gt;&gt;,
  CCIfType&lt;[f64], CCAssignToReg&lt;[ST0, ST1]&gt;&gt;,
  CCDelegateTo&lt;RetCC_X86Common&gt;
]&gt;;
</pre>
</div>

<p>
<tt>CCIfCC</tt> is an interface that attempts to match the given name to the
current calling convention. If the name identifies the current calling
convention, then a specified action is invoked. In the following example (in
<tt>X86CallingConv.td</tt>), if the <tt>Fast</tt> calling convention is in use,
then <tt>RetCC_X86_32_Fast</tt> is invoked. If the <tt>SSECall</tt> calling
convention is in use, then <tt>RetCC_X86_32_SSE</tt> is invoked.
</p>

<div class="doc_code">
<pre>
def RetCC_X86_32 : CallingConv&lt;[
  CCIfCC&lt;"CallingConv::Fast", CCDelegateTo&lt;RetCC_X86_32_Fast&gt;&gt;,
  CCIfCC&lt;"CallingConv::X86_SSECall", CCDelegateTo&lt;RetCC_X86_32_SSE&gt;&gt;,
  CCDelegateTo&lt;RetCC_X86_32_C&gt;
]&gt;;
</pre>
</div>

<p>Other calling convention interfaces include:</p>

<ul>
<li><tt>CCIf &lt;predicate, action&gt;</tt> &mdash; If the predicate matches,
    apply the action.</li>

<li><tt>CCIfInReg &lt;action&gt;</tt> &mdash; If the argument is marked with the
    '<tt>inreg</tt>' attribute, then apply the action.</li>

<li><tt>CCIfNest &lt;action&gt;</tt> &mdash; Inf the argument is marked with the
    '<tt>nest</tt>' attribute, then apply the action.</li>

<li><tt>CCIfNotVarArg &lt;action&gt;</tt> &mdash; If the current function does
    not take a variable number of arguments, apply the action.</li>

<li><tt>CCAssignToRegWithShadow &lt;registerList, shadowList&gt;</tt> &mdash;
    similar to <tt>CCAssignToReg</tt>, but with a shadow list of registers.</li>

<li><tt>CCPassByVal &lt;size, align&gt;</tt> &mdash; Assign value to a stack
    slot with the minimum specified size and alignment.</li>

<li><tt>CCPromoteToType &lt;type&gt;</tt> &mdash; Promote the current value to
    the specified type.</li>

<li><tt>CallingConv &lt;[actions]&gt;</tt> &mdash; Define each calling
    convention that is supported.</li>
</ul>

</div>

</div>

<!-- *********************************************************************** -->
<h2>
  <a name="assemblyPrinter">Assembly Printer</a>
</h2>
<!-- *********************************************************************** -->

<div>

<p>
During the code emission stage, the code generator may utilize an LLVM pass to
produce assembly output. To do this, you want to implement the code for a
printer that converts LLVM IR to a GAS-format assembly language for your target
machine, using the following steps:
</p>

<ul>
<li>Define all the assembly strings for your target, adding them to the
    instructions defined in the <tt>XXXInstrInfo.td</tt> file.
    (See <a href="#InstructionSet">Instruction Set</a>.)  TableGen will produce
    an output file (<tt>XXXGenAsmWriter.inc</tt>) with an implementation of
    the <tt>printInstruction</tt> method for the XXXAsmPrinter class.</li>

<li>Write <tt>XXXTargetAsmInfo.h</tt>, which contains the bare-bones declaration
    of the <tt>XXXTargetAsmInfo</tt> class (a subclass
    of <tt>TargetAsmInfo</tt>).</li>

<li>Write <tt>XXXTargetAsmInfo.cpp</tt>, which contains target-specific values
    for <tt>TargetAsmInfo</tt> properties and sometimes new implementations for
    methods.</li>

<li>Write <tt>XXXAsmPrinter.cpp</tt>, which implements the <tt>AsmPrinter</tt>
    class that performs the LLVM-to-assembly conversion.</li>
</ul>

<p>
The code in <tt>XXXTargetAsmInfo.h</tt> is usually a trivial declaration of the
<tt>XXXTargetAsmInfo</tt> class for use in <tt>XXXTargetAsmInfo.cpp</tt>.
Similarly, <tt>XXXTargetAsmInfo.cpp</tt> usually has a few declarations of
<tt>XXXTargetAsmInfo</tt> replacement values that override the default values
in <tt>TargetAsmInfo.cpp</tt>. For example in <tt>SparcTargetAsmInfo.cpp</tt>:
</p>

<div class="doc_code">
<pre>
SparcTargetAsmInfo::SparcTargetAsmInfo(const SparcTargetMachine &amp;TM) {
  Data16bitsDirective = "\t.half\t";
  Data32bitsDirective = "\t.word\t";
  Data64bitsDirective = 0;  // .xword is only supported by V9.
  ZeroDirective = "\t.skip\t";
  CommentString = "!";
  ConstantPoolSection = "\t.section \".rodata\",#alloc\n";
}
</pre>
</div>

<p>
The X86 assembly printer implementation (<tt>X86TargetAsmInfo</tt>) is an
example where the target specific <tt>TargetAsmInfo</tt> class uses an 
overridden methods: <tt>ExpandInlineAsm</tt>.
</p>

<p>
A target-specific implementation of AsmPrinter is written in
<tt>XXXAsmPrinter.cpp</tt>, which implements the <tt>AsmPrinter</tt> class that
converts the LLVM to printable assembly. The implementation must include the
following headers that have declarations for the <tt>AsmPrinter</tt> and
<tt>MachineFunctionPass</tt> classes. The <tt>MachineFunctionPass</tt> is a
subclass of <tt>FunctionPass</tt>.
</p>

<div class="doc_code">
<pre>
#include "llvm/CodeGen/AsmPrinter.h"
#include "llvm/CodeGen/MachineFunctionPass.h" 
</pre>
</div>

<p>
As a <tt>FunctionPass</tt>, <tt>AsmPrinter</tt> first
calls <tt>doInitialization</tt> to set up the <tt>AsmPrinter</tt>. In
<tt>SparcAsmPrinter</tt>, a <tt>Mangler</tt> object is instantiated to process
variable names.
</p>

<p>
In <tt>XXXAsmPrinter.cpp</tt>, the <tt>runOnMachineFunction</tt> method
(declared in <tt>MachineFunctionPass</tt>) must be implemented
for <tt>XXXAsmPrinter</tt>. In <tt>MachineFunctionPass</tt>,
the <tt>runOnFunction</tt> method invokes <tt>runOnMachineFunction</tt>.
Target-specific implementations of <tt>runOnMachineFunction</tt> differ, but
generally do the following to process each machine function:
</p>

<ul>
<li>Call <tt>SetupMachineFunction</tt> to perform initialization.</li>

<li>Call <tt>EmitConstantPool</tt> to print out (to the output stream) constants
    which have been spilled to memory.</li>

<li>Call <tt>EmitJumpTableInfo</tt> to print out jump tables used by the current
    function.</li>

<li>Print out the label for the current function.</li>

<li>Print out the code for the function, including basic block labels and the
    assembly for the instruction (using <tt>printInstruction</tt>)</li>
</ul>

<p>
The <tt>XXXAsmPrinter</tt> implementation must also include the code generated
by TableGen that is output in the <tt>XXXGenAsmWriter.inc</tt> file. The code
in <tt>XXXGenAsmWriter.inc</tt> contains an implementation of the
<tt>printInstruction</tt> method that may call these methods:
</p>

<ul>
<li><tt>printOperand</tt></li>

<li><tt>printMemOperand</tt></li>

<li><tt>printCCOperand (for conditional statements)</tt></li>

<li><tt>printDataDirective</tt></li>

<li><tt>printDeclare</tt></li>

<li><tt>printImplicitDef</tt></li>

<li><tt>printInlineAsm</tt></li>
</ul>

<p>
The implementations of <tt>printDeclare</tt>, <tt>printImplicitDef</tt>,
<tt>printInlineAsm</tt>, and <tt>printLabel</tt> in <tt>AsmPrinter.cpp</tt> are
generally adequate for printing assembly and do not need to be
overridden.
</p>

<p>
The <tt>printOperand</tt> method is implemented with a long switch/case
statement for the type of operand: register, immediate, basic block, external
symbol, global address, constant pool index, or jump table index. For an
instruction with a memory address operand, the <tt>printMemOperand</tt> method
should be implemented to generate the proper output. Similarly,
<tt>printCCOperand</tt> should be used to print a conditional operand.
</p>

<p><tt>doFinalization</tt> should be overridden in <tt>XXXAsmPrinter</tt>, and
it should be called to shut down the assembly printer. During
<tt>doFinalization</tt>, global variables and constants are printed to
output.
</p>

</div>

<!-- *********************************************************************** -->
<h2>
  <a name="subtargetSupport">Subtarget Support</a>
</h2>
<!-- *********************************************************************** -->

<div>

<p>
Subtarget support is used to inform the code generation process of instruction
set variations for a given chip set.  For example, the LLVM SPARC implementation
provided covers three major versions of the SPARC microprocessor architecture:
Version 8 (V8, which is a 32-bit architecture), Version 9 (V9, a 64-bit
architecture), and the UltraSPARC architecture. V8 has 16 double-precision
floating-point registers that are also usable as either 32 single-precision or 8
quad-precision registers.  V8 is also purely big-endian. V9 has 32
double-precision floating-point registers that are also usable as 16
quad-precision registers, but cannot be used as single-precision registers. The
UltraSPARC architecture combines V9 with UltraSPARC Visual Instruction Set
extensions.
</p>

<p>
If subtarget support is needed, you should implement a target-specific
XXXSubtarget class for your architecture. This class should process the
command-line options <tt>-mcpu=</tt> and <tt>-mattr=</tt>.
</p>

<p>
TableGen uses definitions in the <tt>Target.td</tt> and <tt>Sparc.td</tt> files
to generate code in <tt>SparcGenSubtarget.inc</tt>. In <tt>Target.td</tt>, shown
below, the <tt>SubtargetFeature</tt> interface is defined. The first 4 string
parameters of the <tt>SubtargetFeature</tt> interface are a feature name, an
attribute set by the feature, the value of the attribute, and a description of
the feature. (The fifth parameter is a list of features whose presence is
implied, and its default value is an empty array.)
</p>

<div class="doc_code">
<pre>
class SubtargetFeature&lt;string n, string a,  string v, string d,
                       list&lt;SubtargetFeature&gt; i = []&gt; {
  string Name = n;
  string Attribute = a;
  string Value = v;
  string Desc = d;
  list&lt;SubtargetFeature&gt; Implies = i;
}
</pre>
</div>

<p>
In the <tt>Sparc.td</tt> file, the SubtargetFeature is used to define the
following features.
</p>

<div class="doc_code">
<pre>
def FeatureV9 : SubtargetFeature&lt;"v9", "IsV9", "true",
                     "Enable SPARC-V9 instructions"&gt;;
def FeatureV8Deprecated : SubtargetFeature&lt;"deprecated-v8", 
                     "V8DeprecatedInsts", "true",
                     "Enable deprecated V8 instructions in V9 mode"&gt;;
def FeatureVIS : SubtargetFeature&lt;"vis", "IsVIS", "true",
                     "Enable UltraSPARC Visual Instruction Set extensions"&gt;;
</pre>
</div>

<p>
Elsewhere in <tt>Sparc.td</tt>, the Proc class is defined and then is used to
define particular SPARC processor subtypes that may have the previously
described features.
</p>

<div class="doc_code">
<pre>
class Proc&lt;string Name, list&lt;SubtargetFeature&gt; Features&gt;
  : Processor&lt;Name, NoItineraries, Features&gt;;
&nbsp;
def : Proc&lt;"generic",         []&gt;;
def : Proc&lt;"v8",              []&gt;;
def : Proc&lt;"supersparc",      []&gt;;
def : Proc&lt;"sparclite",       []&gt;;
def : Proc&lt;"f934",            []&gt;;
def : Proc&lt;"hypersparc",      []&gt;;
def : Proc&lt;"sparclite86x",    []&gt;;
def : Proc&lt;"sparclet",        []&gt;;
def : Proc&lt;"tsc701",          []&gt;;
def : Proc&lt;"v9",              [FeatureV9]&gt;;
def : Proc&lt;"ultrasparc",      [FeatureV9, FeatureV8Deprecated]&gt;;
def : Proc&lt;"ultrasparc3",     [FeatureV9, FeatureV8Deprecated]&gt;;
def : Proc&lt;"ultrasparc3-vis", [FeatureV9, FeatureV8Deprecated, FeatureVIS]&gt;;
</pre>
</div>

<p>
From <tt>Target.td</tt> and <tt>Sparc.td</tt> files, the resulting
SparcGenSubtarget.inc specifies enum values to identify the features, arrays of
constants to represent the CPU features and CPU subtypes, and the
ParseSubtargetFeatures method that parses the features string that sets
specified subtarget options. The generated <tt>SparcGenSubtarget.inc</tt> file
should be included in the <tt>SparcSubtarget.cpp</tt>. The target-specific
implementation of the XXXSubtarget method should follow this pseudocode:
</p>

<div class="doc_code">
<pre>
XXXSubtarget::XXXSubtarget(const Module &amp;M, const std::string &amp;FS) {
  // Set the default features
  // Determine default and user specified characteristics of the CPU
  // Call ParseSubtargetFeatures(FS, CPU) to parse the features string
  // Perform any additional operations
}
</pre>
</div>

</div>

<!-- *********************************************************************** -->
<h2>
  <a name="jitSupport">JIT Support</a>
</h2>
<!-- *********************************************************************** -->

<div>

<p>
The implementation of a target machine optionally includes a Just-In-Time (JIT)
code generator that emits machine code and auxiliary structures as binary output
that can be written directly to memory.  To do this, implement JIT code
generation by performing the following steps:
</p>

<ul>
<li>Write an <tt>XXXCodeEmitter.cpp</tt> file that contains a machine function
    pass that transforms target-machine instructions into relocatable machine
    code.</li>

<li>Write an <tt>XXXJITInfo.cpp</tt> file that implements the JIT interfaces for
    target-specific code-generation activities, such as emitting machine code
    and stubs.</li>

<li>Modify <tt>XXXTargetMachine</tt> so that it provides a
    <tt>TargetJITInfo</tt> object through its <tt>getJITInfo</tt> method.</li>
</ul>

<p>
There are several different approaches to writing the JIT support code. For
instance, TableGen and target descriptor files may be used for creating a JIT
code generator, but are not mandatory. For the Alpha and PowerPC target
machines, TableGen is used to generate <tt>XXXGenCodeEmitter.inc</tt>, which
contains the binary coding of machine instructions and the
<tt>getBinaryCodeForInstr</tt> method to access those codes. Other JIT
implementations do not.
</p>

<p>
Both <tt>XXXJITInfo.cpp</tt> and <tt>XXXCodeEmitter.cpp</tt> must include the
<tt>llvm/CodeGen/MachineCodeEmitter.h</tt> header file that defines the
<tt>MachineCodeEmitter</tt> class containing code for several callback functions
that write data (in bytes, words, strings, etc.) to the output stream.
</p>

<!-- ======================================================================= -->
<h3>
  <a name="mce">Machine Code Emitter</a>
</h3>

<div>

<p>
In <tt>XXXCodeEmitter.cpp</tt>, a target-specific of the <tt>Emitter</tt> class
is implemented as a function pass (subclass
of <tt>MachineFunctionPass</tt>). The target-specific implementation
of <tt>runOnMachineFunction</tt> (invoked by
<tt>runOnFunction</tt> in <tt>MachineFunctionPass</tt>) iterates through the
<tt>MachineBasicBlock</tt> calls <tt>emitInstruction</tt> to process each
instruction and emit binary code. <tt>emitInstruction</tt> is largely
implemented with case statements on the instruction types defined in
<tt>XXXInstrInfo.h</tt>. For example, in <tt>X86CodeEmitter.cpp</tt>,
the <tt>emitInstruction</tt> method is built around the following switch/case
statements:
</p>

<div class="doc_code">
<pre>
switch (Desc-&gt;TSFlags &amp; X86::FormMask) {
case X86II::Pseudo:  // for not yet implemented instructions 
   ...               // or pseudo-instructions
   break;
case X86II::RawFrm:  // for instructions with a fixed opcode value
   ...
   break;
case X86II::AddRegFrm: // for instructions that have one register operand 
   ...                 // added to their opcode
   break;
case X86II::MRMDestReg:// for instructions that use the Mod/RM byte
   ...                 // to specify a destination (register)
   break;
case X86II::MRMDestMem:// for instructions that use the Mod/RM byte
   ...                 // to specify a destination (memory)
   break;
case X86II::MRMSrcReg: // for instructions that use the Mod/RM byte
   ...                 // to specify a source (register)
   break;
case X86II::MRMSrcMem: // for instructions that use the Mod/RM byte
   ...                 // to specify a source (memory)
   break;
case X86II::MRM0r: case X86II::MRM1r:  // for instructions that operate on 
case X86II::MRM2r: case X86II::MRM3r:  // a REGISTER r/m operand and
case X86II::MRM4r: case X86II::MRM5r:  // use the Mod/RM byte and a field
case X86II::MRM6r: case X86II::MRM7r:  // to hold extended opcode data
   ...  
   break;
case X86II::MRM0m: case X86II::MRM1m:  // for instructions that operate on
case X86II::MRM2m: case X86II::MRM3m:  // a MEMORY r/m operand and
case X86II::MRM4m: case X86II::MRM5m:  // use the Mod/RM byte and a field
case X86II::MRM6m: case X86II::MRM7m:  // to hold extended opcode data
   ...  
   break;
case X86II::MRMInitReg: // for instructions whose source and
   ...                  // destination are the same register
   break;
}
</pre>
</div>

<p>
The implementations of these case statements often first emit the opcode and
then get the operand(s). Then depending upon the operand, helper methods may be
called to process the operand(s). For example, in <tt>X86CodeEmitter.cpp</tt>,
for the <tt>X86II::AddRegFrm</tt> case, the first data emitted
(by <tt>emitByte</tt>) is the opcode added to the register operand. Then an
object representing the machine operand, <tt>MO1</tt>, is extracted. The helper
methods such as <tt>isImmediate</tt>,
<tt>isGlobalAddress</tt>, <tt>isExternalSymbol</tt>, <tt>isConstantPoolIndex</tt>, and 
<tt>isJumpTableIndex</tt> determine the operand
type. (<tt>X86CodeEmitter.cpp</tt> also has private methods such
as <tt>emitConstant</tt>, <tt>emitGlobalAddress</tt>,
<tt>emitExternalSymbolAddress</tt>, <tt>emitConstPoolAddress</tt>,
and <tt>emitJumpTableAddress</tt> that emit the data into the output stream.)
</p>

<div class="doc_code">
<pre>
case X86II::AddRegFrm:
  MCE.emitByte(BaseOpcode + getX86RegNum(MI.getOperand(CurOp++).getReg()));
  
  if (CurOp != NumOps) {
    const MachineOperand &amp;MO1 = MI.getOperand(CurOp++);
    unsigned Size = X86InstrInfo::sizeOfImm(Desc);
    if (MO1.isImmediate())
      emitConstant(MO1.getImm(), Size);
    else {
      unsigned rt = Is64BitMode ? X86::reloc_pcrel_word
        : (IsPIC ? X86::reloc_picrel_word : X86::reloc_absolute_word);
      if (Opcode == X86::MOV64ri) 
        rt = X86::reloc_absolute_dword;  // FIXME: add X86II flag?
      if (MO1.isGlobalAddress()) {
        bool NeedStub = isa&lt;Function&gt;(MO1.getGlobal());
        bool isLazy = gvNeedsLazyPtr(MO1.getGlobal());
        emitGlobalAddress(MO1.getGlobal(), rt, MO1.getOffset(), 0,
                          NeedStub, isLazy);
      } else if (MO1.isExternalSymbol())
        emitExternalSymbolAddress(MO1.getSymbolName(), rt);
      else if (MO1.isConstantPoolIndex())
        emitConstPoolAddress(MO1.getIndex(), rt);
      else if (MO1.isJumpTableIndex())
        emitJumpTableAddress(MO1.getIndex(), rt);
    }
  }
  break;
</pre>
</div>

<p>
In the previous example, <tt>XXXCodeEmitter.cpp</tt> uses the
variable <tt>rt</tt>, which is a RelocationType enum that may be used to
relocate addresses (for example, a global address with a PIC base offset). The
<tt>RelocationType</tt> enum for that target is defined in the short
target-specific <tt>XXXRelocations.h</tt> file. The <tt>RelocationType</tt> is used by
the <tt>relocate</tt> method defined in <tt>XXXJITInfo.cpp</tt> to rewrite
addresses for referenced global symbols.
</p>

<p>
For example, <tt>X86Relocations.h</tt> specifies the following relocation types
for the X86 addresses. In all four cases, the relocated value is added to the
value already in memory. For <tt>reloc_pcrel_word</tt>
and <tt>reloc_picrel_word</tt>, there is an additional initial adjustment.
</p>

<div class="doc_code">
<pre>
enum RelocationType {
  reloc_pcrel_word = 0,    // add reloc value after adjusting for the PC loc
  reloc_picrel_word = 1,   // add reloc value after adjusting for the PIC base
  reloc_absolute_word = 2, // absolute relocation; no additional adjustment 
  reloc_absolute_dword = 3 // absolute relocation; no additional adjustment
};
</pre>
</div>

</div>

<!-- ======================================================================= -->
<h3>
  <a name="targetJITInfo">Target JIT Info</a>
</h3>

<div>

<p>
<tt>XXXJITInfo.cpp</tt> implements the JIT interfaces for target-specific
code-generation activities, such as emitting machine code and stubs. At minimum,
a target-specific version of <tt>XXXJITInfo</tt> implements the following:
</p>

<ul>
<li><tt>getLazyResolverFunction</tt> &mdash; Initializes the JIT, gives the
    target a function that is used for compilation.</li>

<li><tt>emitFunctionStub</tt> &mdash; Returns a native function with a specified
    address for a callback function.</li>

<li><tt>relocate</tt> &mdash; Changes the addresses of referenced globals, based
    on relocation types.</li>

<li>Callback function that are wrappers to a function stub that is used when the
    real target is not initially known.</li>
</ul>

<p>
<tt>getLazyResolverFunction</tt> is generally trivial to implement. It makes the
incoming parameter as the global <tt>JITCompilerFunction</tt> and returns the
callback function that will be used a function wrapper. For the Alpha target
(in <tt>AlphaJITInfo.cpp</tt>), the <tt>getLazyResolverFunction</tt>
implementation is simply:
</p>

<div class="doc_code">
<pre>
TargetJITInfo::LazyResolverFn AlphaJITInfo::getLazyResolverFunction(  
                                            JITCompilerFn F) {
  JITCompilerFunction = F;
  return AlphaCompilationCallback;
}
</pre>
</div>

<p>
For the X86 target, the <tt>getLazyResolverFunction</tt> implementation is a
little more complication, because it returns a different callback function for
processors with SSE instructions and XMM registers.
</p>

<p>
The callback function initially saves and later restores the callee register
values, incoming arguments, and frame and return address. The callback function
needs low-level access to the registers or stack, so it is typically implemented
with assembler.
</p>

</div>

</div>

<!-- *********************************************************************** -->

<hr>
<address>
  <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
  src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
  <a href="http://validator.w3.org/check/referer"><img
  src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>

  <a href="http://www.woo.com">Mason Woo</a> and <a href="http://misha.brukman.net">Misha Brukman</a><br>
  <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a>
  <br>
  Last modified: $Date$
</address>

</body>
</html>