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
|
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
"http://www.w3.org/TR/html4/strict.dtd">
<html>
<head>
<title>Writing an LLVM Compiler Backend</title>
<link rel="stylesheet" href="llvm.css" type="text/css">
</head>
<body>
<div class="doc_title">
Writing an LLVM Compiler Backend
</div>
<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="#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="#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>
<!-- *********************************************************************** -->
<div class="doc_section">
<a name="intro">Introduction</a>
</div>
<!-- *********************************************************************** -->
<div class="doc_text">
<p>This document describes techniques for writing compiler backends
that convert the LLVM IR (intermediate representation) 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, including X86,
PowerPC, Alpha, 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>
</div>
<div class="doc_subsection">
<a name="Audience">Audience</a>
</div>
<div class="doc_text">
<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>
<div class="doc_subsection">
<a name="Prerequisite">Prerequisite Reading</a>
</div>
<div class="doc_text">
These essential documents must be read before reading this document:
<ul>
<li>
<i><a href="http://www.llvm.org/docs/LangRef.html">LLVM Language Reference Manual</a></i> -
a reference manual for the LLVM assembly language
</li>
<li>
<it><a href="http://www.llvm.org/docs/CodeGenerator.html">The LLVM Target-Independent Code Generator </a></it> -
a guide to the components (classes and code generation algorithms) for translating
the LLVM internal representation to the 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>
<it><a href="http://www.llvm.org/docs/TableGenFundamentals.html">TableGen Fundamentals</a></it> -
a document that describes the TableGen (tblgen) application that manages domain-specific
information to support LLVM code generation. TableGen processes input from a
target description file (.td suffix) and generates C++ code that can be used
for code generation.
</li>
<li>
<it><a href="http://www.llvm.org/docs/WritingAnLLVMPass.html">Writing an LLVM Pass</a></it> -
The assembly printer is a FunctionPass, as are several SelectionDAG processing steps.
</li>
</ul>
To follow the SPARC examples in this document, have a copy of
<it><a href="http://www.sparc.org/standards/V8.pdf">The SPARC Architecture Manual, Version 8</a></it>
for reference. For details about the ARM instruction set, refer to the
<it><a href="http://infocenter.arm.com/">ARM Architecture Reference Manual</a></it>
For more about the GNU Assembler format (GAS), see
<it><a href="http://sourceware.org/binutils/docs/as/index.html">Using As</a></it>
especially for the assembly printer. <it>Using As</it> contains lists of target machine dependent features.
</div>
<div class="doc_subsection">
<a name="Basic">Basic Steps</a>
</div>
<div class="doc_text">
<p>To write a compiler
backend for LLVM that converts the LLVM IR (intermediate representation)
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 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 DAG (directed
acyclic graph) 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 (that is, 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 .cpp and .h 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>
<div class="doc_subsection">
<a name="Preliminaries">Preliminaries</a>
</div>
<div class="doc_text">
<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="http://www.llvm.org/docs/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 LLVMDummy (for example, see the MIPS target). Alternatively, you can
split the library into LLVMDummyCodeGen and LLVMDummyAsmPrinter, 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 lots 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 TargetMachine. 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 LLVMTargetMachine. (To create a target from scratch, create a subclass of
TargetMachine.)</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 class="doc_section">
<a name="TargetMachine">Target Machine</a>
</div>
<!-- *********************************************************************** -->
<div class="doc_text">
<p>LLVMTargetMachine is designed as a base class for targets
implemented with the LLVM target-independent code generator. The
LLVMTargetMachine class should be specialized by a concrete target class that
implements the various virtual methods. LLVMTargetMachine is defined as a
subclass of TargetMachine in <tt>include/llvm/Target/TargetMachine.h</tt>. The
TargetMachine class implementation (<tt>TargetMachine.cpp</tt>) also processes numerous
command-line options. </p>
<p>To create a concrete target-specific subclass of
LLVMTargetMachine, start by copying an existing TargetMachine 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 XXX, the implementation of XXXTargetMachine
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. XXXTargetMachine must also implement
the <tt>getTargetData</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>getTargetData</tt> methods that simply
return a class member. </p>
</div>
<div class="doc_code">
<pre>namespace llvm {
class Module;
class SparcTargetMachine : public LLVMTargetMachine {
const TargetData DataLayout; // Calculates type size & alignment
SparcSubtarget Subtarget;
SparcInstrInfo InstrInfo;
TargetFrameInfo FrameInfo;
protected:
virtual const TargetAsmInfo *createTargetAsmInfo()
const;
public:
SparcTargetMachine(const Module &M, const std::string &FS);
virtual const SparcInstrInfo *getInstrInfo() const {return &InstrInfo; }
virtual const TargetFrameInfo *getFrameInfo() const {return &FrameInfo; }
virtual const TargetSubtarget *getSubtargetImpl() const{return &Subtarget; }
virtual const TargetRegisterInfo *getRegisterInfo() const {
return &InstrInfo.getRegisterInfo();
}
virtual const TargetData *getTargetData() const { return &DataLayout; }
static unsigned getModuleMatchQuality(const Module &M);
// Pass Pipeline Configuration
virtual bool addInstSelector(PassManagerBase &PM, bool Fast);
virtual bool addPreEmitPass(PassManagerBase &PM, bool Fast);
virtual bool addAssemblyEmitter(PassManagerBase &PM, bool Fast,
std::ostream &Out);
};
} // end namespace llvm
</pre>
</div>
<div class="doc_text">
<ul>
<li><tt>getInstrInfo </tt></li>
<li><tt>getRegisterInfo</tt></li>
<li><tt>getFrameInfo</tt></li>
<li><tt>getTargetData</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 XXXTargetMachine constructor should specify a
TargetDescription 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>
<div class="doc_code">
<pre>
SparcTargetMachine::SparcTargetMachine(const Module &M, const std::string &FS)
: DataLayout("E-p:32:32-f128:128:128"),
Subtarget(M, FS), InstrInfo(Subtarget),
FrameInfo(TargetFrameInfo::StackGrowsDown, 8, 0) {
}
</pre>
</div>
<div class="doc_text">
<p>Hyphens separate portions of the TargetDescription string. </p>
<ul>
<li>The "E" in the string indicates a big-endian target data model; a
lower-case "e" would indicate little-endian. </li>
<li>"p:" is followed by pointer information: size, ABI alignment, and
preferred alignment. If only two figures follow "p:", 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: "i", "f", "v", or "a"
(corresponding to integer, floating point, vector, or aggregate). "i", "v", or
"a" are followed by ABI alignment and preferred alignment. "f" is followed by
three values, the first indicates the size of a long double, then ABI alignment
and preferred alignment.</li>
</ul>
<p>You must also register your target using the RegisterTarget
template. (See the TargetMachineRegistry class.) For example, in <tt>SparcTargetMachine.cpp</tt>,
the target is registered with:</p>
</div>
<div class="doc_code">
<pre>
namespace {
// Register the target.
RegisterTarget<SparcTargetMachine>X("sparc", "SPARC");
}
</pre>
</div>
<!-- *********************************************************************** -->
<div class="doc_section">
<a name="RegisterSet">Register Set and Register Classes</a>
</div>
<!-- *********************************************************************** -->
<div class="doc_text">
<p>You should describe
a concrete target-specific class
that represents the register file of a target machine. This class is
called XXXRegisterInfo (where XXX identifies the target) and represents the
class register file data that is used for register allocation and 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 that include 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
XXXRegisterInfo requires hand-coding. </p>
</div>
<!-- ======================================================================= -->
<div class="doc_subsection">
<a name="RegisterDef">Defining a Register</a>
</div>
<div class="doc_text">
<p>The <tt>XXXRegisterInfo.td</tt> file typically starts with register definitions
for a target machine. The Register class (specified in <tt>Target.td</tt>) is used to
define an object for each register. The specified string n becomes the Name of
the register. The basic Register object does not have any subregisters and does
not specify any aliases.</p>
</div>
<div class="doc_code">
<pre>
class Register<string n> {
string Namespace = "";
string AsmName = n;
string Name = n;
int SpillSize = 0;
int SpillAlignment = 0;
list<Register> Aliases = [];
list<Register> SubRegs = [];
list<int> DwarfNumbers = [];
}
</pre>
</div>
<div class="doc_text">
<p>For example, in the <tt>X86RegisterInfo.td</tt> file, there are register
definitions that utilize the Register class, such as:</p>
</div>
<div class="doc_code">
<pre>
def AL : Register<"AL">,
DwarfRegNum<[0, 0, 0]>;
</pre>
</div>
<div class="doc_text">
<p>This defines the register AL and assigns it values (with
DwarfRegNum) that are used by <tt>gcc</tt>, <tt>gdb</tt>, or a debug information writer (such as
DwarfWriter in <tt>llvm/lib/CodeGen</tt>) to identify a register. For register AL,
DwarfRegNum takes an array of 3 values, representing 3 different modes: the
first element is for X86-64, the second for EH (exception handling) 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>
<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>
<div class="doc_text">
<p>From the register info file, TableGen generates a
TargetRegisterDesc object for each register. TargetRegisterDesc is defined in
<tt>include/llvm/Target/TargetRegisterInfo.h</tt> with the following fields:</p>
</div>
<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>
<div class="doc_text">
<p>TableGen uses the entire target description file (<tt>.td</tt>) to
determine text names for the register (in the AsmName and Name fields of
TargetRegisterDesc) and the relationships of other registers to the defined
register (in the other TargetRegisterDesc fields). In this example, other
definitions establish the registers "AX", "EAX", and "RAX" as aliases for one
another, so TableGen generates a null-terminated array (AL_AliasSet) for this
register alias set. </p>
<p>The Register class is commonly used as a base class for more
complex classes. In <tt>Target.td</tt>, the Register class is the base for the
RegisterWithSubRegs class that is used to define registers that need to specify
subregisters in the SubRegs list, as shown here:</p>
</div>
<div class="doc_code">
<pre>
class RegisterWithSubRegs<string n,
list<Register> subregs> : Register<n> {
let SubRegs = subregs;
}</pre>
</div>
<div class="doc_text">
<p>In <tt>SparcRegisterInfo.td</tt>, additional register classes are defined
for SPARC: a Register subclass, SparcReg, and further subclasses: Ri, Rf, and
Rd. SPARC registers are identified by 5-bit ID numbers, which is a feature
common to these subclasses. Note the use of ‘let’ expressions to override values
that are initially defined in a superclass (such as SubRegs field in the Rd
class). </p>
</div>
<div class="doc_code">
<pre>
class SparcReg<string n> : Register<n> {
field bits<5> Num;
let Namespace = "SP";
}
// Ri - 32-bit integer registers
class Ri<bits<5> num, string n> :
SparcReg<n> {
let Num = num;
}
// Rf - 32-bit floating-point registers
class Rf<bits<5> num, string n> :
SparcReg<n> {
let Num = num;
}
// Rd - Slots in the FP register file for 64-bit
floating-point values.
class Rd<bits<5> num, string n,
list<Register> subregs> : SparcReg<n> {
let Num = num;
let SubRegs = subregs;
}</pre>
</div>
<div class="doc_text">
<p>In the <tt>SparcRegisterInfo.td</tt> file, there are register definitions
that utilize these subclasses of Register, such as:</p>
</div>
<div class="doc_code">
<pre>
def G0 : Ri< 0, "G0">,
DwarfRegNum<[0]>;
def G1 : Ri< 1, "G1">, DwarfRegNum<[1]>;
...
def F0 : Rf< 0, "F0">,
DwarfRegNum<[32]>;
def F1 : Rf< 1, "F1">,
DwarfRegNum<[33]>;
...
def D0 : Rd< 0, "F0", [F0, F1]>,
DwarfRegNum<[32]>;
def D1 : Rd< 2, "F2", [F2, F3]>,
DwarfRegNum<[34]>;
</pre>
</div>
<div class="doc_text">
<p>The last two registers shown above (D0 and D1) 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>
<!-- ======================================================================= -->
<div class="doc_subsection">
<a name="RegisterClassDef">Defining a Register Class</a>
</div>
<div class="doc_text">
<p>The RegisterClass 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>
<div class="doc_code">
<pre>
class RegisterClass<string namespace,
list<ValueType> regTypes, int alignment,
list<Register> regList> {
string Namespace = namespace;
list<ValueType> 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;
list<Register> MemberList = regList;
// for register classes that are subregisters of this class
list<RegisterClass> SubRegClassList = [];
code MethodProtos = [{}]; // to insert arbitrary code
code MethodBodies = [{}];
}</pre>
</div>
<div class="doc_text">
<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 ValueType register type values
that are defined in <tt>include/llvm/CodeGen/ValueTypes.td</tt>. Defined values include
integer types (such as i16, i32, and i1 for Boolean), floating-point types
(f32, f64), and vector types (for example, v8i16 for an 8 x i16 vector). All
registers in a RegisterClass must have the same ValueType, 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 RegisterClass 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 <tt>allocation_order_*</tt> method is not specified, then <tt>regList</tt> also
defines the order of allocation used by the register allocator.</li>
</ul>
<p>In <tt>SparcRegisterInfo.td</tt>, three RegisterClass objects are defined:
FPRegs, DFPRegs, and IntRegs. For all three register classes, the first
argument defines the namespace with the string “SP”. FPRegs defines a group of 32
single-precision floating-point registers (F0 to F31); DFPRegs defines a group
of 16 double-precision registers (D0-D15). For IntRegs, the MethodProtos and
MethodBodies methods are used by TableGen to insert the specified code into generated
output.</p>
</div>
<div class="doc_code">
<pre>
def FPRegs : RegisterClass<"SP", [f32], 32, [F0, F1, F2, F3, F4, F5, F6, F7,
F8, F9, F10, F11, F12, F13, F14, F15, F16, F17, F18, F19, F20, F21, F22,
F23, F24, F25, F26, F27, F28, F29, F30, F31]>;
def DFPRegs : RegisterClass<"SP", [f64], 64, [D0, D1, D2, D3, D4, D5, D6, D7,
D8, D9, D10, D11, D12, D13, D14, D15]>;
def IntRegs : RegisterClass<"SP", [i32], 32, [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
]> {
let MethodProtos = [{
iterator allocation_order_end(const MachineFunction &MF) const;
}];
let MethodBodies = [{
IntRegsClass::iterator
IntRegsClass::allocation_order_end(const MachineFunction &MF) const {
return end()-10 // Don't allocate special registers
-1;
}
}];
}
</pre>
</div>
<div class="doc_text">
<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
SparcGenRegisterInfo that uses TargetRegisterInfo as its base. It also
specifies types, based upon the defined register classes: DFPRegsClass, FPRegsClass,
and IntRegsClass. </p>
<p><tt>SparcRegisterInfo.td</tt> also generates SparcGenRegisterInfo.inc,
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 IntRegs reflects the
order in the definition of IntRegs in the target description file. Take special
note of the use of MethodBodies in <tt>SparcRegisterInfo.td</tt> to create code in
<tt>SparcGenRegisterInfo.inc</tt>. MethodProtos generates similar code in
<tt>SparcGenRegisterInfo.h.inc</tt>.</p>
</div>
<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 DFPRegsRegClass;
FPRegsClass FPRegsRegClass;
IntRegsClass 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::iterator
IntRegsClass::allocation_order_end(const MachineFunction &MF) const {
return end()-10 // Don't allocate special registers
-1;
}
IntRegsClass::IntRegsClass() : TargetRegisterClass(IntRegsRegClassID,
IntRegsVTs, IntRegsSubclasses, IntRegsSuperclasses, IntRegsSubRegClasses,
IntRegsSuperRegClasses, 4, 4, 1, IntRegs, IntRegs + 32) {}
}
</pre>
</div>
<!-- ======================================================================= -->
<div class="doc_subsection">
<a name="implementRegister">Implement a subclass of
<a href="http://www.llvm.org/docs/CodeGenerator.html#targetregisterinfo">TargetRegisterInfo</a></a>
</div>
<div class="doc_text">
<p>The final step is to hand code portions of XXXRegisterInfo, which
implements the interface described in <tt>TargetRegisterInfo.h</tt>. These functions
return 0, NULL, or false, unless overridden. Here’s a list of functions that
are overridden for the SPARC implementation in <tt>SparcRegisterInfo.cpp</tt>:</p>
<ul>
<li><tt>getCalleeSavedRegs</tt> (returns a list of callee-saved registers in
the order of the desired callee-save stack frame offset)</li>
<li><tt>getCalleeSavedRegClasses</tt> (returns a list of preferred register
classes with which to spill each callee saved register)</li>
<li><tt>getReservedRegs</tt> (returns a bitset indexed by physical register
numbers, indicating if a particular register is unavailable)</li>
<li><tt>hasFP</tt> (return a Boolean indicating if a function should have a
dedicated frame pointer register)</li>
<li><tt>eliminateCallFramePseudoInstr</tt> (if call frame setup or destroy
pseudo instructions are used, this can be called to eliminate them)</li>
<li><tt>eliminateFrameIndex</tt> (eliminate abstract frame indices from
instructions that may use them)</li>
<li><tt>emitPrologue</tt> (insert prologue code into the function)</li>
<li><tt>emitEpilogue</tt> (insert epilogue code into the function)</li>
</ul>
</div>
<!-- *********************************************************************** -->
<div class="doc_section">
<a name="InstructionSet">Instruction Set</a>
</div>
<!-- *********************************************************************** -->
<div class="doc_text">
<p>During the early stages of code generation, the LLVM IR code is
converted to a SelectionDAG with nodes that are instances of the SDNode class
containing target instructions. An SDNode 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 NodeType enum in the ISD namespace).</p>
<p>TableGen uses the following target description (.td) input files
to generate much of the code for instruction definition:</p>
<ul>
<li><tt>Target.td</tt>, where the Instruction, Operand, InstrInfo, and other
fundamental classes are defined</li>
<li><tt>TargetSelectionDAG.td</tt>, used by SelectionDAG instruction selection
generators, contains SDTC* classes (selection DAG type constraint), definitions
of SelectionDAG nodes (such as imm, cond, bb, add, fadd, sub), and pattern
support (Pattern, Pat, PatFrag, PatLeaf, ComplexPattern)</li>
<li><tt>XXXInstrFormats.td</tt>, patterns for definitions of target-specific
instructions</li>
<li><tt>XXXInstrInfo.td</tt>, 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 XXX is the
name of the target. The <tt>XXX.td</tt> file includes the other .td input files, but its
contents are only directly important for subtargets.</p>
<p>You should describe
a concrete target-specific class
XXXInstrInfo that represents machine
instructions supported by a target machine. XXXInstrInfo contains an array of
XXXInstrDescriptor 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>
<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<dag> Pattern; // Set to the DAG pattern for this instruction
list<Register> Uses = [];
list<Register> Defs = [];
list<Predicate> Predicates = []; // predicates turned into isel match code
... remainder not shown for space ...
}
</pre>
</div>
<div class="doc_text">
<p>A SelectionDAG node (SDNode) 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 ADDri and ADDrr.</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 CALL instruction. Format 2 is for
branch on condition codes and SETHI (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>. InstSP is a base
class for other instruction classes. Additional base classes are specified for
more precise formats: for example in <tt>SparcInstrFormat.td</tt>, F2_1 is for SETHI,
and F2_2 is for branches. There are three other base classes: F3_1 for
register/register operations, F3_2 for register/immediate operations, and F3_3 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, LDrr, defines
the Load Integer instruction for a Word (the LD SPARC opcode) from a memory
address to a register. The first parameter, the value 3 (11<sub>2</sub>), is
the operation value for this category of operation. The second parameter
(000000<sub>2</sub>) is the specific operation value for LD/Load Word. The
third parameter is the output destination, which is a register operand and
defined in the Register target description file (IntRegs). </p>
</div>
<div class="doc_code">
<pre>def LDrr : F3_1 <3, 0b000000, (outs IntRegs:$dst), (ins MEMrr:$addr),
"ld [$addr], $dst",
[(set IntRegs:$dst, (load ADDRrr:$addr))]>;
</pre>
</div>
<div class="doc_text">
<p>The fourth
parameter is the input source, which uses the address operand MEMrr that is
defined earlier in <tt>SparcInstrInfo.td</tt>:</p>
</div>
<div class="doc_code">
<pre>def MEMrr : Operand<i32> {
let PrintMethod = "printMemOperand";
let MIOperandInfo = (ops IntRegs, IntRegs);
}
</pre>
</div>
<div class="doc_text">
<p>The fifth parameter is a string that is used by the assembly
printer and can be left as an empty string until the assembly printer interface
is implemented. The sixth and final parameter is the pattern used to match the
instruction during the SelectionDAG Select Phase described in
(<a href="http://www.llvm.org/docs/CodeGenerator.html">The LLVM Target-Independent Code Generator</a>).
This parameter is detailed in the next section, <a href="#InstructionSelector">Instruction Selector</a>.</p>
<p>Instruction class definitions are not overloaded for different
operand types, so separate versions of instructions are needed for register,
memory, or immediate value operands. For example, to perform a
Load Integer instruction for a Word
from an immediate operand to a register, the following instruction class is
defined: </p>
</div>
<div class="doc_code">
<pre>def LDri : F3_2 <3, 0b000000, (outs IntRegs:$dst), (ins MEMri:$addr),
"ld [$addr], $dst",
[(set IntRegs:$dst, (load ADDRri:$addr))]>;
</pre>
</div>
<div class="doc_text">
<p>Writing these definitions for so many similar instructions can
involve a lot of cut and paste. In td files, the <tt>multiclass</tt> directive enables
the creation of templates to define several instruction classes at once (using
the <tt>defm</tt> directive). For example in
<tt>SparcInstrInfo.td</tt>, the <tt>multiclass</tt> pattern F3_12 is defined to create 2
instruction classes each time F3_12 is invoked: </p>
</div>
<div class="doc_code">
<pre>multiclass F3_12 <string OpcStr, bits<6> Op3Val, SDNode OpNode> {
def rr : F3_1 <2, Op3Val,
(outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
!strconcat(OpcStr, " $b, $c, $dst"),
[(set IntRegs:$dst, (OpNode IntRegs:$b, IntRegs:$c))]>;
def ri : F3_2 <2, Op3Val,
(outs IntRegs:$dst), (ins IntRegs:$b, i32imm:$c),
!strconcat(OpcStr, " $b, $c, $dst"),
[(set IntRegs:$dst, (OpNode IntRegs:$b, simm13:$c))]>;
}
</pre>
</div>
<div class="doc_text">
<p>So when the <tt>defm</tt> directive is used for the XOR and ADD
instructions, as seen below, it creates four instruction objects: XORrr, XORri,
ADDrr, and ADDri.</p>
</div>
<div class="doc_code">
<pre>defm XOR : F3_12<"xor", 0b000011, xor>;
defm ADD : F3_12<"add", 0b000000, add>;
</pre>
</div>
<div class="doc_text">
<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>
<div class="doc_code">
<pre>def ICC_NE : ICC_VAL< 9>; // Not Equal
def ICC_E : ICC_VAL< 1>; // Equal
def ICC_G : ICC_VAL<10>; // Greater
...
def FCC_U : FCC_VAL<23>; // Unordered
def FCC_G : FCC_VAL<22>; // Greater
def FCC_UG : FCC_VAL<21>; // Unordered or Greater
...
</pre>
</div>
<div class="doc_text">
<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>; that is, <tt>SPCC::ICC_NE = 9</tt>, <tt>SPCC::FCC_U = 23</tt> and so on.)</p>
</div>
<!-- ======================================================================= -->
<div class="doc_subsection">
<a name="implementInstr">Implement a subclass of
<a href="http://www.llvm.org/docs/CodeGenerator.html#targetinstrinfo">TargetInstrInfo</a></a>
</div>
<div class="doc_text">
<p>The final step is to hand code portions of XXXInstrInfo, which
implements the interface described in <tt>TargetInstrInfo.h</tt>. These functions return
0 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>isMoveInstr</tt> (return true if the instruction is a register to
register move; false, otherwise)</li>
<li><tt>isLoadFromStackSlot</tt> (if the specified machine instruction is a
direct load from a stack slot, return the register number of the destination
and the FrameIndex of the stack slot)</li>
<li><tt>isStoreToStackSlot</tt> (if the specified machine instruction is a
direct store to a stack slot, return the register number of the destination and
the FrameIndex of the stack slot)</li>
<li><tt>copyRegToReg</tt> (copy values between a pair of registers)</li>
<li><tt>storeRegToStackSlot</tt> (store a register value to a stack slot)</li>
<li><tt>loadRegFromStackSlot</tt> (load a register value from a stack slot)</li>
<li><tt>storeRegToAddr</tt> (store a register value to memory)</li>
<li><tt>loadRegFromAddr</tt> (load a register value from memory)</li>
<li><tt>foldMemoryOperand</tt> (attempt to combine instructions of any load or
store instruction for the specified operand(s))</li>
</ul>
</div>
<!-- ======================================================================= -->
<div class="doc_subsection">
<a name="branchFolding">Branch Folding and If Conversion</a>
</div>
<div class="doc_text">
<p>Performance can be improved by combining instructions or by eliminating
instructions that are never reached. The <tt>AnalyzeBranch</tt> method in XXXInstrInfo 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>MachineBasicBlock &MBB – the incoming block to be
examined</li>
<li>MachineBasicBlock *&TBB – a destination block that is
returned; for a conditional branch that evaluates to true, TBB is the
destination </li>
<li>MachineBasicBlock *&FBB – for a conditional branch that
evaluates to false, FBB is returned as the destination</li>
<li>std::vector<MachineOperand> &Cond – 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 TBB or FBB, so both parameters return NULL. 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>
<div class="doc_code">
<pre>bool ARMInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,
MachineBasicBlock *&TBB, MachineBasicBlock *&FBB,
std::vector<MachineOperand> &Cond) const
{
MachineBasicBlock::iterator I = MBB.end();
if (I == MBB.begin() || !isUnpredicatedTerminator(--I))
return false;
</pre>
</div>
<div class="doc_text">
<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 TBB parameter. </p>
</div>
<div class="doc_code">
<pre>if (LastOpc == ARM::B || LastOpc == ARM::tB) {
TBB = LastInst->getOperand(0).getMBB();
return false;
}
</pre>
</div>
<div class="doc_text">
<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 TBB parameter. </p>
</div>
<div class="doc_code">
<pre>if ((SecondLastOpc == ARM::B || SecondLastOpc==ARM::tB) &&
(LastOpc == ARM::B || LastOpc == ARM::tB)) {
TBB = SecondLastInst->getOperand(0).getMBB();
I = LastInst;
I->eraseFromParent();
return false;
}
</pre>
</div>
<div class="doc_text">
<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 TBB parameter and a list of operands in the <tt>Cond</tt>
parameter to evaluate the condition. </p>
</div>
<div class="doc_code">
<pre>if (LastOpc == ARM::Bcc || LastOpc == ARM::tBcc) {
// Block ends with fall-through condbranch.
TBB = LastInst->getOperand(0).getMBB();
Cond.push_back(LastInst->getOperand(1));
Cond.push_back(LastInst->getOperand(2));
return false;
}
</pre>
</div>
<div class="doc_text">
<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 ‘true’) in the TBB parameter and the unconditional branch
destination in the FBB (corresponding to a conditional evaluation of ‘false’).
A list of operands to evaluate the condition should be returned in the <tt>Cond</tt>
parameter.</p>
</div>
<div class="doc_code">
<pre>unsigned SecondLastOpc = SecondLastInst->getOpcode();
if ((SecondLastOpc == ARM::Bcc && LastOpc == ARM::B) ||
(SecondLastOpc == ARM::tBcc && LastOpc == ARM::tB)) {
TBB = SecondLastInst->getOperand(0).getMBB();
Cond.push_back(SecondLastInst->getOperand(1));
Cond.push_back(SecondLastInst->getOperand(2));
FBB = LastInst->getOperand(0).getMBB();
return false;
}
</pre>
</div>
<div class="doc_text">
<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 class="doc_section">
<a name="InstructionSelector">Instruction Selector</a>
</div>
<!-- *********************************************************************** -->
<div class="doc_text">
<p>LLVM uses a SelectionDAG to represent LLVM IR instructions, and nodes
of the SelectionDAG 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 Selection DAG. </p>
<p>TableGen generates code for instruction selection using the
following target description input files:</p>
<ul>
<li><tt>XXXInstrInfo.td</tt> 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> 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 FunctionPass class or a subclass of FunctionPass. 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 SelectionDAG before or after specific processing phases, use the command
line options for <tt>llc</tt>, described at <a
href="http://llvm.org/docs/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 SelectionDAG 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>
<div class="doc_code">
<pre>def STrr : F3_1< 3, 0b000100, (outs), (ins MEMrr:$addr, IntRegs:$src),
"st $src, [$addr]", [(store IntRegs:$src, ADDRrr:$addr)]>;
</pre>
</div>
<div class="doc_text">
<p>ADDRrr is a memory mode that is also defined in <tt>SparcInstrInfo.td</tt>:</p>
</div>
<div class="doc_code">
<pre>def ADDRrr : ComplexPattern<i32, 2, "SelectADDRrr", [], []>;
</pre>
</div>
<div class="doc_text">
<p>The definition of ADDRrr refers to SelectADDRrr, 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>
<div class="doc_code">
<pre>def store : PatFrag<(ops node:$val, node:$ptr),
(st node:$val, node:$ptr), [{
if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N))
return !ST->isTruncatingStore() &&
ST->getAddressingMode() == ISD::UNINDEXED;
return false;
}]>;
</pre>
</div>
<div class="doc_text">
<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
ISD::STORE opcode.</p>
</div>
<div class="doc_code">
<pre>SDNode *SelectCode(SDOperand N) {
...
MVT::ValueType NVT = N.Val->getValueType(0);
switch (N.getOpcode()) {
case ISD::STORE: {
switch (NVT) {
default:
return Select_ISD_STORE(N);
break;
}
break;
}
...
</pre>
</div>
<div class="doc_text">
<p>The pattern for STrr is matched, so elsewhere in
<tt>XXXGenDAGISel.inc</tt>, code for STrr 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>
<div class="doc_code">
<pre>SDNode *Select_ISD_STORE(const SDOperand &N) {
SDOperand Chain = N.getOperand(0);
if (Predicate_store(N.Val)) {
SDOperand N1 = N.getOperand(1);
SDOperand N2 = N.getOperand(2);
SDOperand CPTmp0;
SDOperand CPTmp1;
// Pattern: (st:void IntRegs:i32:$src,
// ADDRrr:i32:$addr)<<P:Predicate_store>>
// Emits: (STrr:void ADDRrr:i32:$addr, IntRegs:i32:$src)
// Pattern complexity = 13 cost = 1 size = 0
if (SelectADDRrr(N, N2, CPTmp0, CPTmp1) &&
N1.Val->getValueType(0) == MVT::i32 &&
N2.Val->getValueType(0) == MVT::i32) {
return Emit_22(N, SP::STrr, CPTmp0, CPTmp1);
}
...
</pre>
</div>
<!-- ======================================================================= -->
<div class="doc_subsection">
<a name="LegalizePhase">The SelectionDAG Legalize Phase</a>
</div>
<div class="doc_text">
<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 XXXTargetLowering 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>
<div class="doc_code">
<pre>addRegisterClass(MVT::i32, SP::IntRegsRegisterClass);
addRegisterClass(MVT::f32, SP::FPRegsRegisterClass);
addRegisterClass(MVT::f64, SP::DFPRegsRegisterClass);
</pre>
</div>
<div class="doc_text">
<p>You should examine the node types in the ISD namespace
(<tt>include/llvm/CodeGen/SelectionDAGNodes.h</tt>)
and determine which operations the target natively supports. For operations
that do <u>not</u> 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> (general operation)</li>
<li><tt>setLoadExtAction</tt> (load with extension)</li>
<li><tt>setTruncStoreAction</tt> (truncating store)</li>
<li><tt>setIndexedLoadAction</tt> (indexed load)</li>
<li><tt>setIndexedStoreAction</tt> (indexed store)</li>
<li><tt>setConvertAction</tt> (type conversion)</li>
<li><tt>setCondCodeAction</tt> (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 LegalAction 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 LegalAction values.</p>
</div>
<!-- _______________________________________________________________________ -->
<div class="doc_subsubsection">
<a name="promote">Promote</a>
</div>
<div class="doc_text">
<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>
<div class="doc_code">
<pre>setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
</pre>
</div>
<!-- _______________________________________________________________________ -->
<div class="doc_subsubsection">
<a name="expand">Expand</a>
</div>
<div class="doc_text">
<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>
<div class="doc_code">
<pre>setOperationAction(ISD::FSIN, MVT::f32, Expand);
setOperationAction(ISD::FCOS, MVT::f32, Expand);
</pre>
</div>
<!-- _______________________________________________________________________ -->
<div class="doc_subsubsection">
<a name="custom">Custom</a>
</div>
<div class="doc_text">
<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>
<div class="doc_code">
<pre>setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
</pre>
</div>
<div class="doc_text">
<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 FP_TO_SINT opcode will call the <tt>LowerFP_TO_SINT</tt> method:</p>
</div>
<div class="doc_code">
<pre>SDOperand SparcTargetLowering::LowerOperation(
SDOperand Op, SelectionDAG &DAG) {
switch (Op.getOpcode()) {
case ISD::FP_TO_SINT: return LowerFP_TO_SINT(Op, DAG);
...
}
}
</pre>
</div>
<div class="doc_text">
<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>
<div class="doc_code">
<pre>static SDOperand LowerFP_TO_SINT(SDOperand Op, SelectionDAG &DAG) {
assert(Op.getValueType() == MVT::i32);
Op = DAG.getNode(SPISD::FTOI, MVT::f32, Op.getOperand(0));
return DAG.getNode(ISD::BIT_CONVERT, MVT::i32, Op);
}
</pre>
</div>
<!-- _______________________________________________________________________ -->
<div class="doc_subsubsection">
<a name="legal">Legal</a>
</div>
<div class="doc_text">
<p>The <tt>Legal</tt> LegalizeAction enum value simply indicates that an
operation <u>is</u> natively supported. <tt>Legal</tt> represents the default condition,
so it is rarely used. In <tt>SparcISelLowering.cpp</tt>, the action for CTPOP (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>
<div class="doc_code">
<pre>setOperationAction(ISD::CTPOP, MVT::i32, Expand);
...
if (TM.getSubtarget<SparcSubtarget>().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 class="doc_subsection">
<a name="callingConventions">Calling Conventions</a>
</div>
<div class="doc_text">
<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 CCIfType and
CCAssignToReg interfaces. If the CCIfType predicate is true (that is, if the
current argument is of type f32 or f64), then the action is performed. In this
case, the CCAssignToReg action assigns the argument value to the first
available register: either R0 or R1. </p>
</div>
<div class="doc_code">
<pre>CCIfType<[f32,f64], CCAssignToReg<[R0, R1]>>
</pre>
</div>
<div class="doc_text">
<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
(CC_Sparc32). The definition of RetCC_Sparc32 (shown below) indicates which
registers are used for specified scalar return types. A single-precision float
is returned to register F0, and a double-precision float goes to register D0. A
32-bit integer is returned in register I0 or I1. </p>
</div>
<div class="doc_code">
<pre>def RetCC_Sparc32 : CallingConv<[
CCIfType<[i32], CCAssignToReg<[I0, I1]>>,
CCIfType<[f32], CCAssignToReg<[F0]>>,
CCIfType<[f64], CCAssignToReg<[D0]>>
]>;
</pre>
</div>
<div class="doc_text">
<p>The definition of CC_Sparc32 in <tt>SparcCallingConv.td</tt> introduces
CCAssignToStack, 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>
<div class="doc_code">
<pre>def CC_Sparc32 : CallingConv<[
// All arguments get passed in integer registers if there is space.
CCIfType<[i32, f32, f64], CCAssignToReg<[I0, I1, I2, I3, I4, I5]>>,
CCAssignToStack<4, 4>
]>;
</pre>
</div>
<div class="doc_text">
<p>CCDelegateTo 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 RetCC_X86_32_C
ends with CCDelegateTo. After the current value is assigned to the register ST0
or ST1, the RetCC_X86Common is invoked.</p>
</div>
<div class="doc_code">
<pre>def RetCC_X86_32_C : CallingConv<[
CCIfType<[f32], CCAssignToReg<[ST0, ST1]>>,
CCIfType<[f64], CCAssignToReg<[ST0, ST1]>>,
CCDelegateTo<RetCC_X86Common>
]>;
</pre>
</div>
<div class="doc_text">
<p>CCIfCC 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 Fast calling convention is in use, then RetCC_X86_32_Fast
is invoked. If the SSECall calling convention is in use, then RetCC_X86_32_SSE
is invoked. </p>
</div>
<div class="doc_code">
<pre>def RetCC_X86_32 : CallingConv<[
CCIfCC<"CallingConv::Fast", CCDelegateTo<RetCC_X86_32_Fast>>,
CCIfCC<"CallingConv::X86_SSECall", CCDelegateTo<RetCC_X86_32_SSE>>,
CCDelegateTo<RetCC_X86_32_C>
]>;
</pre>
</div>
<div class="doc_text">
<p>Other calling convention interfaces include:</p>
<ul>
<li>CCIf <predicate, action> - if the predicate matches, apply
the action</li>
<li>CCIfInReg <action> - if the argument is marked with the
‘inreg’ attribute, then apply the action </li>
<li>CCIfNest <action> - if the argument is marked with the
‘nest’ attribute, then apply the action</li>
<li>CCIfNotVarArg <action> - if the current function does not
take a variable number of arguments, apply the action</li>
<li>CCAssignToRegWithShadow <registerList, shadowList> -
similar to CCAssignToReg, but with a shadow list of registers</li>
<li>CCPassByVal <size, align> - assign value to a stack slot
with the minimum specified size and alignment </li>
<li>CCPromoteToType <type> - promote the current value to the specified
type</li>
<li>CallingConv <[actions]> - define each calling convention
that is supported</li>
</ul>
</div>
<!-- *********************************************************************** -->
<div class="doc_section">
<a name="assemblyPrinter">Assembly Printer</a>
</div>
<!-- *********************************************************************** -->
<div class="doc_text">
<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 XXXTargetAsmInfo class (a subclass of TargetAsmInfo). </li>
<li>Write <tt>XXXTargetAsmInfo.cpp</tt>, which contains target-specific values
for TargetAsmInfo properties and sometimes new implementations for methods</li>
<li>Write <tt>XXXAsmPrinter.cpp</tt>, which implements the AsmPrinter 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 XXXTargetAsmInfo class for use in <tt>XXXTargetAsmInfo.cpp</tt>. Similarly,
<tt>XXXTargetAsmInfo.cpp</tt> usually has a few declarations of XXXTargetAsmInfo replacement
values that override the default values in <tt>TargetAsmInfo.cpp</tt>. For example in
<tt>SparcTargetAsmInfo.cpp</tt>, </p>
</div>
<div class="doc_code">
<pre>SparcTargetAsmInfo::SparcTargetAsmInfo(const SparcTargetMachine &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>
<div class="doc_text">
<p>The X86 assembly printer implementation (X86TargetAsmInfo) is an
example where the target specific TargetAsmInfo class uses overridden methods:
<tt>ExpandInlineAsm</tt> and <tt>PreferredEHDataFormat</tt>. </p>
<p>A target-specific implementation of AsmPrinter is written in
<tt>XXXAsmPrinter.cpp</tt>, which implements the AsmPrinter class that converts the LLVM
to printable assembly. The implementation must include the following headers
that have declarations for the AsmPrinter and MachineFunctionPass classes. The
MachineFunctionPass is a subclass of FunctionPass. </p>
</div>
<div class="doc_code">
<pre>#include "llvm/CodeGen/AsmPrinter.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
</pre>
</div>
<div class="doc_text">
<p>As a FunctionPass, AsmPrinter first calls <tt>doInitialization</tt> to set
up the AsmPrinter. In SparcAsmPrinter, a Mangler object is instantiated to
process variable names.</p>
<p>In <tt>XXXAsmPrinter.cpp</tt>, the <tt>runOnMachineFunction</tt> method (declared
in MachineFunctionPass) must be implemented for XXXAsmPrinter. In
MachineFunctionPass, 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 XXXAsmPrinter 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>
<li><tt>printLabel</tt></li>
<li><tt>printPICJumpTableEntry</tt></li>
<li><tt>printPICJumpTableSetLabel</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. (<tt>printBasicBlockLabel</tt> is
another method that is implemented in <tt>AsmPrinter.cpp</tt> that may be directly used
in an implementation of XXXAsmPrinter.)</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 XXXAsmPrinter, 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>
<!-- *********************************************************************** -->
<div class="doc_section">
<a name="subtargetSupport">Subtarget Support</a>
</div>
<!-- *********************************************************************** -->
<div class="doc_text">
<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
SubtargetFeature interface is defined. The first 4 string parameters of the
SubtargetFeature 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>
<div class="doc_code">
<pre>class SubtargetFeature<string n, string a, string v, string d,
list<SubtargetFeature> i = []> {
string Name = n;
string Attribute = a;
string Value = v;
string Desc = d;
list<SubtargetFeature> Implies = i;
}
</pre>
</div>
<div class="doc_text">
<p>In the <tt>Sparc.td</tt> file, the SubtargetFeature is used to define the
following features. </p>
</div>
<div class="doc_code">
<pre>def FeatureV9 : SubtargetFeature<"v9", "IsV9", "true",
"Enable SPARC-V9 instructions">;
def FeatureV8Deprecated : SubtargetFeature<"deprecated-v8",
"V8DeprecatedInsts", "true",
"Enable deprecated V8 instructions in V9 mode">;
def FeatureVIS : SubtargetFeature<"vis", "IsVIS", "true",
"Enable UltraSPARC Visual Instruction Set extensions">;
</pre>
</div>
<div class="doc_text">
<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>
<div class="doc_code">
<pre>class Proc<string Name, list<SubtargetFeature> Features>
: Processor<Name, NoItineraries, Features>;
def : Proc<"generic", []>;
def : Proc<"v8", []>;
def : Proc<"supersparc", []>;
def : Proc<"sparclite", []>;
def : Proc<"f934", []>;
def : Proc<"hypersparc", []>;
def : Proc<"sparclite86x", []>;
def : Proc<"sparclet", []>;
def : Proc<"tsc701", []>;
def : Proc<"v9", [FeatureV9]>;
def : Proc<"ultrasparc", [FeatureV9, FeatureV8Deprecated]>;
def : Proc<"ultrasparc3", [FeatureV9, FeatureV8Deprecated]>;
def : Proc<"ultrasparc3-vis", [FeatureV9, FeatureV8Deprecated, FeatureVIS]>;
</pre>
</div>
<div class="doc_text">
<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>
<div class="doc_code">
<pre>XXXSubtarget::XXXSubtarget(const Module &M, const std::string &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 class="doc_section">
<a name="jitSupport">JIT Support</a>
</div>
<!-- *********************************************************************** -->
<div class="doc_text">
<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 XXXTargetMachine so that it provides a TargetJITInfo
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 MachineCodeEmitter
class containing code for several callback functions that write data (in bytes,
words, strings, etc.) to the output stream.</p>
</div>
<!-- ======================================================================= -->
<div class="doc_subsection">
<a name="mce">Machine Code Emitter</a>
</div>
<div class="doc_text">
<p>In <tt>XXXCodeEmitter.cpp</tt>, a target-specific of the Emitter class is
implemented as a function pass (subclass of MachineFunctionPass). The
target-specific implementation of <tt>runOnMachineFunction</tt> (invoked by
<tt>runOnFunction</tt> in MachineFunctionPass) iterates through the MachineBasicBlock
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>
<div class="doc_code">
<pre>switch (Desc->TSFlags & 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>
<div class="doc_text">
<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, MO1, 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>
<div class="doc_code">
<pre>case X86II::AddRegFrm:
MCE.emitByte(BaseOpcode + getX86RegNum(MI.getOperand(CurOp++).getReg()));
if (CurOp != NumOps) {
const MachineOperand &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<Function>(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>
<div class="doc_text">
<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 RelocationType enum for
that target is defined in the short target-specific <tt>XXXRelocations.h</tt> file. The
RelocationType 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>
<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 class="doc_subsection">
<a name="targetJITInfo">Target JIT Info</a>
</div>
<div class="doc_text">
<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 XXXJITInfo implements the following:</p>
<ul>
<li><tt>getLazyResolverFunction</tt> – initializes the JIT, gives the
target a function that is used for compilation </li>
<li><tt>emitFunctionStub</tt> – returns a native function with a
specified address for a callback function</li>
<li><tt>relocate</tt> – 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 JITCompilerFunction 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>
<div class="doc_code">
<pre>TargetJITInfo::LazyResolverFn AlphaJITInfo::getLazyResolverFunction(
JITCompilerFn F)
{
JITCompilerFunction = F;
return AlphaCompilationCallback;
}
</pre>
</div>
<div class="doc_text">
<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>
<!-- *********************************************************************** -->
<hr>
<address>
<a href="http://jigsaw.w3.org/css-validator/check/referer"><img
src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
<a href="http://validator.w3.org/check/referer"><img
src="http://www.w3.org/Icons/valid-html401" 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>
|
