| blob | e220764ac6e0711de16be312a54f6eb782535388 |
3 <html>
4 <head>
11 </head>
13 <body>
16 <ol>
21 <ol>
31 </ol>
32 </li>
34 <ol>
36 <ol>
38 </ol>
39 </li>
41 <ol>
49 </ol>
50 </li>
51 </ol>
52 </li>
54 <ol>
60 </ol>
61 </li>
63 <ol>
65 </ol>
66 </li>
68 <ol>
70 <ol>
77 </ol>
78 </li>
80 <ol>
90 </ol>
91 </li>
93 <ol>
100 </ol>
101 </li>
103 <ol>
107 </ol>
108 </li>
110 <ol>
117 </ol>
118 </li>
120 <ol>
133 </ol>
135 <ol>
142 </ol>
143 </li>
144 </ol>
145 </li>
147 <ol>
149 <ol>
153 </ol>
154 </li>
156 <ol>
160 </ol>
161 </li>
163 <ol>
171 </ol>
172 </li>
174 <ol>
180 </ol>
181 </li>
183 <ol>
190 </ol>
191 </li>
194 </ol>
195 </li>
196 </ol>
201 </div>
203 <!-- *********************************************************************** -->
205 <!-- *********************************************************************** -->
208 <p>This document is a reference manual for the LLVM assembly language.
209 LLVM is an SSA based representation that provides type safety,
210 low-level operations, flexibility, and the capability of representing
211 'all' high-level languages cleanly. It is the common code
212 representation used throughout all phases of the LLVM compilation
213 strategy.</p>
214 </div>
216 <!-- *********************************************************************** -->
218 <!-- *********************************************************************** -->
222 <p>The LLVM code representation is designed to be used in three
223 different forms: as an in-memory compiler IR, as an on-disk bytecode
224 representation (suitable for fast loading by a Just-In-Time compiler),
225 and as a human readable assembly language representation. This allows
226 LLVM to provide a powerful intermediate representation for efficient
227 compiler transformations and analysis, while providing a natural means
228 to debug and visualize the transformations. The three different forms
229 of LLVM are all equivalent. This document describes the human readable
230 representation and notation.</p>
232 <p>The LLVM representation aims to be light-weight and low-level
233 while being expressive, typed, and extensible at the same time. It
234 aims to be a "universal IR" of sorts, by being at a low enough level
235 that high-level ideas may be cleanly mapped to it (similar to how
236 microprocessors are "universal IR's", allowing many source languages to
237 be mapped to them). By providing type information, LLVM can be used as
238 the target of optimizations: for example, through pointer analysis, it
239 can be proven that a C automatic variable is never accessed outside of
240 the current function... allowing it to be promoted to a simple SSA
241 value instead of a memory location.</p>
243 </div>
245 <!-- _______________________________________________________________________ -->
250 <p>It is important to note that this document describes 'well formed'
251 LLVM assembly language. There is a difference between what the parser
252 accepts and what is considered 'well formed'. For example, the
253 following instruction is syntactically okay, but not well formed:</p>
255 <pre>
257 </pre>
260 its uses. The LLVM infrastructure provides a verification pass that may
261 be used to verify that an LLVM module is well formed. This pass is
262 automatically run by the parser after parsing input assembly and by
263 the optimizer before it outputs bytecode. The violations pointed out
264 by the verifier pass indicate bugs in transformation passes or input to
265 the parser.</p>
269 <!-- *********************************************************************** -->
271 <!-- *********************************************************************** -->
275 <p>LLVM uses three different forms of identifiers, for different
276 purposes:</p>
278 <ol>
279 <li>Named values are represented as a string of characters with a '%' prefix.
280 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
282 Identifiers which require other characters in their names can be surrounded
284 in a name.</li>
286 <li>Unnamed values are represented as an unsigned numeric value with a '%'
291 </ol>
293 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
294 don't need to worry about name clashes with reserved words, and the set of
295 reserved words may be expanded in the future without penalty. Additionally,
296 unnamed identifiers allow a compiler to quickly come up with a temporary
297 variable without having to avoid symbol table conflicts.</p>
299 <p>Reserved words in LLVM are very similar to reserved words in other
300 languages. There are keywords for different opcodes
305 and others. These reserved words cannot conflict with variable names, because
306 none of them start with a '%' character.</p>
308 <p>Here is an example of LLVM code to multiply the integer variable
313 <pre>
315 </pre>
319 <pre>
321 </pre>
325 <pre>
329 </pre>
332 important lexical features of LLVM:</p>
334 <ol>
337 line.</li>
339 <li>Unnamed temporaries are created when the result of a computation is not
340 assigned to a named value.</li>
344 </ol>
346 <p>...and it also shows a convention that we follow in this document. When
347 demonstrating instructions, we will follow an instruction with a comment that
348 defines the type and name of value produced. Comments are shown in italic
349 text.</p>
351 </div>
353 <!-- *********************************************************************** -->
355 <!-- *********************************************************************** -->
357 <!-- ======================================================================= -->
359 </div>
364 translation unit of the input programs. Each module consists of
365 functions, global variables, and symbol table entries. Modules may be
366 combined together with the LLVM linker, which merges function (and
367 global variable) definitions, resolves forward declarations, and merges
372 href="#globalvars">constant</a> <a href="#t_array">[13 x i8 ]</a> c"hello world\0A\00" <i>; [13 x i8 ]*</i>
380 %cast210 = <a
384 <a
386 <a
394 <p>In general, a module is made up of a list of global values,
395 where both functions and global variables are global values. Global values are
396 represented by a pointer to a memory location (in this case, a pointer to an
397 array of char, and a pointer to a function), and have one of the following <a
400 </div>
402 <!-- ======================================================================= -->
405 </div>
409 <p>
410 All Global Variables and Functions have one of the following types of linkage:
411 </p>
413 <dl>
417 <dd>Global values with internal linkage are only directly accessible by
418 objects in the current module. In particular, linking code into a module with
419 an internal global value may cause the internal to be renamed as necessary to
420 avoid collisions. Because the symbol is internal to the module, all
421 references can be updated. This corresponds to the notion of the
423 </dd>
428 the same name when linkage occurs. This is typically used to implement
429 inline functions, templates, or other code which must be generated in each
431 allowed to be discarded.
432 </dd>
438 used for globals that may be emitted in multiple translation units, but that
439 are not guaranteed to be emitted into every translation unit that uses them.
440 One example of this are common globals in C, such as "<tt>int X;</tt>" at
441 global scope.
442 </dd>
447 pointer to array type. When two global variables with appending linkage are
448 linked together, the two global arrays are appended together. This is the
449 LLVM, typesafe, equivalent of having the system linker append together
450 "sections" with identical names when .o files are linked.
451 </dd>
454 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
455 until linked, if not linked, the symbol becomes null instead of being an
456 undefined reference.
457 </dd>
461 <dd>If none of the above identifiers are used, the global is externally
462 visible, meaning that it participates in linkage and can be used to resolve
463 external symbol references.
464 </dd>
465 </dl>
467 <p>
468 The next two types of linkage are targeted for Microsoft Windows platform
469 only. They are designed to support importing (exporting) symbols from (to)
470 DLLs.
471 </p>
473 <dl>
477 or variable via a global pointer to a pointer that is set up by the DLL
478 exporting the symbol. On Microsoft Windows targets, the pointer name is
480 </dd>
485 pointer to a pointer in a DLL, so that it can be referenced with the
488 name.
489 </dd>
491 </dl>
494 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
495 variable and was linked with this one, one of the two would be renamed,
497 external (i.e., lacking any linkage declarations), they are accessible
498 outside of the current module.</p>
503 linkages.
504 </div>
506 <!-- ======================================================================= -->
509 </div>
515 specified for the call. The calling convention of any pair of dynamic
516 caller/callee must match, or the behavior of the program is undefined. The
517 following calling conventions are supported by LLVM, and more may be added in
518 the future:</p>
520 <dl>
523 <dd>This calling convention (the default if no other calling convention is
524 specified) matches the target C calling conventions. This calling convention
525 supports varargs function calls and tolerates some mismatch in the declared
526 prototype and implemented declaration of the function (as does normal C).
527 </dd>
531 <dd>This calling convention attempts to make calls as fast as possible
532 (e.g. by passing things in registers). This calling convention allows the
533 target to use whatever tricks it wants to produce fast code for the target,
534 without having to conform to an externally specified ABI. Implementations of
535 this convention should allow arbitrary tail call optimization to be supported.
536 This calling convention does not support varargs and requires the prototype of
537 all callees to exactly match the prototype of the function definition.
538 </dd>
542 <dd>This calling convention attempts to make code in the caller as efficient
543 as possible under the assumption that the call is not commonly executed. As
544 such, these calls often preserve all registers so that the call does not break
545 any live ranges in the caller side. This calling convention does not support
546 varargs and requires the prototype of all callees to exactly match the
547 prototype of the function definition.
548 </dd>
552 <dd>Any calling convention may be specified by number, allowing
553 target-specific calling conventions to be used. Target specific calling
554 conventions start at 64.
555 </dd>
556 </dl>
558 <p>More calling conventions can be added/defined on an as-needed basis, to
559 support pascal conventions or any other well-known target-independent
560 convention.</p>
562 </div>
564 <!-- ======================================================================= -->
567 </div>
571 <p>
572 All Global Variables and Functions have one of the following visibility styles:
573 </p>
575 <dl>
578 <dd>On ELF, default visibility means that the declaration is visible to other
579 modules and, in shared libraries, means that the declared entity may be
580 overridden. On Darwin, default visibility means that the declaration is
581 visible to other modules. Default visibility corresponds to "external
582 linkage" in the language.
583 </dd>
587 <dd>Two declarations of an object with hidden visibility refer to the same
588 object if they are in the same shared object. Usually, hidden visibility
589 indicates that the symbol will not be placed into the dynamic symbol table,
590 so no other module (executable or shared library) can reference it
591 directly.
592 </dd>
596 <dd>On ELF, protected visibility indicates that the symbol will be placed in
597 the dynamic symbol table, but that references within the defining module will
598 bind to the local symbol. That is, the symbol cannot be overridden by another
599 module.
600 </dd>
601 </dl>
603 </div>
605 <!-- ======================================================================= -->
608 </div>
612 <p>Global variables define regions of memory allocated at compilation time
613 instead of run-time. Global variables may optionally be initialized, may have
614 an explicit section to be placed in, and may have an optional explicit alignment
615 specified. A variable may be defined as "thread_local", which means that it
616 will not be shared by threads (each thread will have a separated copy of the
617 variable). A variable may be defined as a global "constant," which indicates
619 optimization, allowing the global data to be placed in the read-only section of
620 an executable, etc). Note that variables that need runtime initialization
623 <p>
625 constant, even if the final definition of the global is not. This capability
626 can be used to enable slightly better optimization of the program, but requires
627 the language definition to guarantee that optimizations based on the
628 'constantness' are valid for the translation units that do not include the
629 definition.
630 </p>
632 <p>As SSA values, global variables define pointer values that are in
633 scope (i.e. they dominate) all basic blocks in the program. Global
634 variables always define a pointer to their "content" type because they
635 describe a region of memory, and all memory objects in LLVM are
636 accessed through pointers.</p>
638 <p>LLVM allows an explicit section to be specified for globals. If the target
639 supports it, it will emit globals to the section specified.</p>
641 <p>An explicit alignment may be specified for a global. If not present, or if
642 the alignment is set to zero, the alignment of the global is set by the target
643 to whatever it feels convenient. If an explicit alignment is specified, the
644 global is forced to have at least that much alignment. All alignments must be
647 <p>For example, the following defines a global with an initializer, section,
648 and alignment:</p>
650 <pre>
652 </pre>
654 </div>
657 <!-- ======================================================================= -->
660 </div>
669 name, a (possibly empty) argument list (each with optional
671 optional alignment, an opening curly brace, a list of basic blocks, and a
672 closing curly brace.
674 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
679 name, a possibly empty list of arguments, and an optional alignment.</p>
681 <p>A function definition contains a list of basic blocks, forming the CFG for
682 the function. Each basic block may optionally start with a label (giving the
683 basic block a symbol table entry), contains a list of instructions, and ends
685 function return).</p>
687 <p>The first basic block in a program is special in two ways: it is immediately
688 executed on entrance to the function, and it is not allowed to have predecessor
689 basic blocks (i.e. there can not be any branches to the entry block of a
690 function). Because the block can have no predecessors, it also cannot have any
693 <p>LLVM functions are identified by their name and type signature. Hence, two
694 functions with the same name but different parameter lists or return values are
695 considered different functions, and LLVM will resolve references to each
696 appropriately.</p>
698 <p>LLVM allows an explicit section to be specified for functions. If the target
699 supports it, it will emit functions to the section specified.</p>
701 <p>An explicit alignment may be specified for a function. If not present, or if
702 the alignment is set to zero, the alignment of the function is set by the target
703 to whatever it feels convenient. If an explicit alignment is specified, the
704 function is forced to have at least that much alignment. All alignments must be
707 </div>
710 <!-- ======================================================================= -->
713 </div>
716 function or global variable or bitcast of global value). Aliases may have an
722 <pre>
724 </pre>
726 </div>
730 <!-- ======================================================================= -->
733 <p>The return type and each parameter of a function type may have a set of
735 used to communicate additional information about the result or parameters of
736 a function. Parameter attributes are considered to be part of the function
737 type so two functions types that differ only by the parameter attributes
738 are different function types.</p>
740 <p>Parameter attributes are simple keywords that follow the type specified. If
741 multiple parameter attributes are needed, they are space separated. For
742 example:</p><pre>
743 %someFunc = i16 (i8 sext %someParam) zext
744 %someFunc = i16 (i8 zext %someParam) zext</pre>
745 <p>Note that the two function types above are unique because the parameter has
746 a different attribute (sext in the first one, zext in the second). Also note
747 that the attribute for the function result (zext) comes immediately after the
748 argument list.</p>
751 <dl>
753 <dd>This indicates that the parameter should be zero extended just before
754 a call to this function.</dd>
756 <dd>This indicates that the parameter should be sign extended just before
757 a call to this function.</dd>
759 <dd>This indicates that the parameter should be placed in register (if
760 possible) during assembling function call. Support for this attribute is
761 target-specific</dd>
763 <dd>This indicates that the parameter specifies the address of a structure
764 that is the return value of the function in the source program.</dd>
766 <dd>This function attribute indicates that the function never returns. This
767 indicates to LLVM that every call to this function should be treated as if
770 <dd>This function attribute indicates that the function type does not use
771 the unwind instruction and does not allow stack unwinding to propagate
772 through it.</dd>
773 </dl>
775 </div>
777 <!-- ======================================================================= -->
780 </div>
783 <p>
784 Modules may contain "module-level inline asm" blocks, which corresponds to the
785 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
786 LLVM and treated as a single unit, but may be separated in the .ll file if
787 desired. The syntax is very simple:
788 </p>
791 module asm "inline asm code goes here"
792 module asm "more can go here"
793 </pre></div>
795 <p>The strings can contain any character by escaping non-printable characters.
797 for the number.
798 </p>
800 <p>
801 The inline asm code is simply printed to the machine code .s file when
802 assembly code is generated.
803 </p>
804 </div>
806 <!-- ======================================================================= -->
809 </div>
812 <p>A module may specify a target specific data layout string that specifies how
813 data is to be laid out in memory. The syntax for the data layout is simply:</p>
816 separated by the minus sign character ('-'). Each specification starts with a
817 letter and may include other information after the letter to define some
818 aspect of the data layout. The specifications accepted are as follows: </p>
819 <dl>
821 <dd>Specifies that the target lays out data in big-endian form. That is, the
822 bits with the most significance have the lowest address location.</dd>
824 <dd>Specifies that hte target lays out data in little-endian form. That is,
825 the bits with the least significance have the lowest address location.</dd>
830 too.</dd>
832 <dd>This specifies the alignment for an integer type of a given bit
835 <dd>This specifies the alignment for a vector type of a given bit
838 <dd>This specifies the alignment for a floating point type of a given bit
840 (double).</dd>
842 <dd>This specifies the alignment for an aggregate type of a given bit
844 </dl>
845 <p>When constructing the data layout for a given target, LLVM starts with a
846 default set of specifications which are then (possibly) overriden by the
848 are given in this list:</p>
849 <ul>
863 </ul>
864 <p>When llvm is determining the alignment for a given type, it uses the
865 following rules:
866 <ol>
867 <li>If the type sought is an exact match for one of the specifications, that
868 specification is used.</li>
869 <li>If no match is found, and the type sought is an integer type, then the
870 smallest integer type that is larger than the bitwidth of the sought type is
871 used. If none of the specifications are larger than the bitwidth then the the
872 largest integer type is used. For example, given the default specifications
873 above, the i7 type will use the alignment of i8 (next largest) while both
874 i65 and i256 will use the alignment of i64 (largest specified).</li>
875 <li>If no match is found, and the type sought is a vector type, then the
876 largest vector type that is smaller than the sought vector type will be used
879 </ol>
880 </div>
882 <!-- *********************************************************************** -->
884 <!-- *********************************************************************** -->
888 <p>The LLVM type system is one of the most important features of the
889 intermediate representation. Being typed enables a number of
890 optimizations to be performed on the IR directly, without having to do
891 extra analyses on the side before the transformation. A strong type
892 system makes it easier to read the generated code and enables novel
893 analyses and transformations that are not feasible to perform on normal
894 three address code representations.</p>
896 </div>
898 <!-- ======================================================================= -->
901 <p>The primitive types are the fundamental building blocks of the LLVM
902 system. The current set of primitive types is as follows:</p>
907 <table>
908 <tbody>
912 </tbody>
913 </table>
914 </td>
916 <table>
917 <tbody>
921 </tbody>
922 </table>
923 </td>
924 </tr>
925 </table>
926 </div>
928 <!-- _______________________________________________________________________ -->
930 Classifications</a> </div>
932 <p>These different primitive types fall into a few useful
933 classifications:</p>
936 <tbody>
938 <tr>
941 </tr>
942 <tr>
945 </tr>
946 <tr>
950 </td>
951 </tr>
952 </tbody>
953 </table>
956 most important. Values of these types are the only ones which can be
957 produced by instructions, passed as arguments, or used as operands to
958 instructions. This means that all structures and arrays must be
959 manipulated either by pointer or by component.</p>
960 </div>
962 <!-- ======================================================================= -->
967 <p>The real power in LLVM comes from the derived types in the system.
968 This is what allows a programmer to represent arrays, functions,
969 pointers, and other useful types. Note that these derived types may be
970 recursive: For example, it is possible to have a two dimensional array.</p>
972 </div>
974 <!-- _______________________________________________________________________ -->
980 <p>The integer type is a very simple derived type that simply specifies an
981 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
986 <pre>
987 iN
988 </pre>
991 value.</p>
1005 </td>
1012 An integer whose bit width is the answer. <br/>
1015 </td>
1016 </tr>
1017 </table>
1019 <!-- _______________________________________________________________________ -->
1026 <p>The array type is a very simple derived type that arranges elements
1027 sequentially in memory. The array type requires a size (number of
1028 elements) and an underlying data type.</p>
1032 <pre>
1034 </pre>
1036 <p>The number of elements is a constant integer value; elementtype may
1037 be any type with a size.</p>
1046 </td>
1051 </td>
1052 </tr>
1053 </table>
1061 </td>
1066 </td>
1067 </tr>
1068 </table>
1070 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1071 length array. Normally, accesses past the end of an array are undefined in
1073 As a special case, however, zero length arrays are recognized to be variable
1074 length. This allows implementation of 'pascal style arrays' with the LLVM
1077 </div>
1079 <!-- _______________________________________________________________________ -->
1083 <p>The function type can be thought of as a function signature. It
1084 consists of a return type and a list of formal parameter types.
1085 Function types are usually used to build virtual function tables
1086 (which are structures of pointers to functions), for indirect function
1087 calls, and when defining a function.</p>
1088 <p>
1089 The return type of a function type cannot be an aggregate type.
1090 </p>
1095 which indicates that the function takes a variable number of arguments.
1096 Variable argument functions can access their arguments with the <a
1103 </td>
1106 </tt></td>
1111 </td>
1117 LLVM.
1118 </td>
1119 </tr>
1120 </table>
1122 </div>
1123 <!-- _______________________________________________________________________ -->
1127 <p>The structure type is used to represent a collection of data members
1128 together in memory. The packing of the field types is defined to match
1129 the ABI of the underlying processor. The elements of a structure may
1130 be any type that has a size.</p>
1134 instruction.</p>
1148 </tr>
1149 </table>
1150 </div>
1152 <!-- _______________________________________________________________________ -->
1154 </div>
1157 <p>The packed structure type is used to represent a collection of data members
1158 together in memory. There is no padding between fields. Further, the alignment
1159 of a packed structure is 1 byte. The elements of a packed structure may
1160 be any type that has a size.</p>
1164 instruction.</p>
1173 <td class="left"><tt>< { float, i32 (i32) * } ></tt></td>
1178 </tr>
1179 </table>
1180 </div>
1182 <!-- _______________________________________________________________________ -->
1186 <p>As in many languages, the pointer type represents a pointer or
1187 reference to another object, which must live in memory.</p>
1196 </td>
1203 </td>
1204 </tr>
1205 </table>
1206 </div>
1208 <!-- _______________________________________________________________________ -->
1214 <p>A vector type is a simple derived type that represents a vector
1215 of elements. Vector types are used when multiple primitive data
1216 are operated in parallel using a single instruction (SIMD).
1217 A vector type requires a size (number of
1218 elements) and an underlying primitive data type. Vectors must have a power
1224 <pre>
1226 </pre>
1228 <p>The number of elements is a constant integer value; elementtype may
1229 be any integer or floating point type.</p>
1239 </td>
1244 </td>
1245 </tr>
1246 </table>
1247 </div>
1249 <!-- _______________________________________________________________________ -->
1255 <p>Opaque types are used to represent unknown types in the system. This
1256 corresponds (for example) to the C notion of a foward declared structure type.
1257 In LLVM, opaque types can eventually be resolved to any type (not just a
1258 structure type).</p>
1262 <pre>
1263 opaque
1264 </pre>
1272 </td>
1274 An opaque type.<br/>
1275 </td>
1276 </tr>
1277 </table>
1278 </div>
1281 <!-- *********************************************************************** -->
1283 <!-- *********************************************************************** -->
1287 <p>LLVM has several different basic types of constants. This section describes
1288 them all and their syntax.</p>
1290 </div>
1292 <!-- ======================================================================= -->
1297 <dl>
1302 </dd>
1308 integer types.
1309 </dd>
1314 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1315 notation (see below). Floating point constants must have a <a
1323 </dl>
1325 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1326 of floating point constants. For example, the form '<tt>double
1329 (and the only time that they are generated by the disassembler) is when a
1330 floating point constant must be emitted but it cannot be represented as a
1331 decimal floating point number. For example, NaN's, infinities, and other
1332 special values are represented in their IEEE hexadecimal format so that
1333 assembly and disassembly do not cause any bits to change in the constants.</p>
1335 </div>
1337 <!-- ======================================================================= -->
1339 </div>
1342 <p>Aggregate constants arise from aggregation of simple constants
1343 and smaller aggregate constants.</p>
1345 <dl>
1348 <dd>Structure constants are represented with notation similar to structure
1349 type definitions (a comma separated list of elements, surrounded by braces
1353 types of elements must match those specified by the type.
1354 </dd>
1358 <dd>Array constants are represented with notation similar to array type
1359 definitions (a comma separated list of elements, surrounded by square brackets
1362 types of elements must match those specified by the type.
1363 </dd>
1367 <dd>Vector constants are represented with notation similar to vector type
1368 definitions (a comma separated list of elements, surrounded by
1372 match those specified by the type.
1373 </dd>
1379 This is often used to avoid having to print large zero initializers (e.g. for
1380 large arrays) and is always exactly equivalent to using explicit zero
1381 initializers.
1382 </dd>
1383 </dl>
1385 </div>
1387 <!-- ======================================================================= -->
1390 </div>
1396 constants. These constants are explicitly referenced when the <a
1399 file:</p>
1401 <pre>
1402 %X = global i32 17
1403 %Y = global i32 42
1404 %Z = global [2 x i32*] [ i32* %X, i32* %Y ]
1405 </pre>
1407 </div>
1409 <!-- ======================================================================= -->
1413 no specific value. Undefined values may be of any type and be used anywhere
1414 a constant is permitted.</p>
1416 <p>Undefined values indicate to the compiler that the program is well defined
1417 no matter what value is used, giving the compiler more freedom to optimize.
1418 </p>
1419 </div>
1421 <!-- ======================================================================= -->
1423 </div>
1427 <p>Constant expressions are used to allow expressions involving other constants
1428 to be used as constants. Constant expressions may be of any <a
1430 that does not have side effects (e.g. load and call are not supported). The
1431 following is the syntax for constant expressions:</p>
1433 <dl>
1435 <dd>Truncate a constant to another type. The bit size of CST must be larger
1436 than the bit size of TYPE. Both types must be integers.</dd>
1439 <dd>Zero extend a constant to another type. The bit size of CST must be
1440 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1443 <dd>Sign extend a constant to another type. The bit size of CST must be
1444 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1447 <dd>Truncate a floating point constant to another floating point type. The
1448 size of CST must be larger than the size of TYPE. Both types must be
1449 floating point.</dd>
1452 <dd>Floating point extend a constant to another type. The size of CST must be
1453 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1456 <dd>Convert a floating point constant to the corresponding unsigned integer
1457 constant. TYPE must be an integer type. CST must be floating point. If the
1458 value won't fit in the integer type, the results are undefined.</dd>
1461 <dd>Convert a floating point constant to the corresponding signed integer
1462 constant. TYPE must be an integer type. CST must be floating point. If the
1463 value won't fit in the integer type, the results are undefined.</dd>
1466 <dd>Convert an unsigned integer constant to the corresponding floating point
1467 constant. TYPE must be floating point. CST must be of integer type. If the
1468 value won't fit in the floating point type, the results are undefined.</dd>
1471 <dd>Convert a signed integer constant to the corresponding floating point
1472 constant. TYPE must be floating point. CST must be of integer type. If the
1473 value won't fit in the floating point type, the results are undefined.</dd>
1476 <dd>Convert a pointer typed constant to the corresponding integer constant
1477 TYPE must be an integer type. CST must be of pointer type. The CST value is
1478 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1481 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1482 pointer type. CST must be of integer type. The CST value is zero extended,
1483 truncated, or unchanged to make it fit in a pointer size. This one is
1487 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1488 identical (same number of bits). The conversion is done as if the CST value
1489 was stored to memory and read back as TYPE. In other words, no bits change
1490 with this operator, just the type. This can be used for conversion of
1491 vector types to any other type, as long as they have the same bit width. For
1492 pointers it is only valid to cast to another pointer type.
1493 </dd>
1499 instruction, the index list may have zero or more indexes, which are required
1505 constants.</dd>
1516 operation</a> on constants.
1531 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1533 binary</a> operations. The constraints on operands are the same as those for
1534 the corresponding instruction (e.g. no bitwise operations on floating point
1535 values are allowed).</dd>
1536 </dl>
1537 </div>
1539 <!-- *********************************************************************** -->
1541 <!-- *********************************************************************** -->
1543 <!-- ======================================================================= -->
1546 </div>
1550 <p>
1552 Module-Level Inline Assembly</a>) through the use of a special value. This
1553 value represents the inline assembler as a string (containing the instructions
1554 to emit), a list of operand constraints (stored as a string), and a flag that
1555 indicates whether or not the inline asm expression has side effects. An example
1556 inline assembler expression is:
1557 </p>
1559 <pre>
1561 </pre>
1563 <p>
1566 </p>
1568 <pre>
1570 </pre>
1572 <p>
1573 Inline asms with side effects not visible in the constraint list must be marked
1574 as having side effects. This is done through the use of the
1576 </p>
1578 <pre>
1580 </pre>
1582 <p>TODO: The format of the asm and constraints string still need to be
1583 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1584 need to be documented).
1585 </p>
1587 </div>
1589 <!-- *********************************************************************** -->
1591 <!-- *********************************************************************** -->
1595 <p>The LLVM instruction set consists of several different
1602 </div>
1604 <!-- ======================================================================= -->
1606 Instructions</a> </div>
1611 basic block in a program ends with a "Terminator" instruction, which
1612 indicates which block should be executed after the current block is
1614 value: they produce control flow, not values (the one exception being
1623 </div>
1625 <!-- _______________________________________________________________________ -->
1627 Instruction</a> </div>
1632 </pre>
1635 value) from a function back to the caller.</p>
1637 returns a value and then causes control flow, and one that just causes
1638 control flow to occur.</p>
1643 instruction inside of the function that returns a value that does not
1644 match the return type of the function.</p>
1647 returns back to the calling function's context. If the caller is a "<a
1649 the instruction after the call. If the caller was an "<a
1651 at the beginning of the "normal" destination block. If the instruction
1652 returns a value, that value shall set the call or invoke instruction's
1653 return value.</p>
1657 </pre>
1658 </div>
1659 <!-- _______________________________________________________________________ -->
1663 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1664 </pre>
1667 transfer to a different basic block in the current function. There are
1668 two forms of this instruction, corresponding to a conditional branch
1669 and an unconditional branch.</p>
1681 <pre>Test:<br> %cond = <a href="#i_icmp">icmp</a> eq, i32 %a, %b<br> br i1 %cond, label %IfEqual, label %IfUnequal<br>IfEqual:<br> <a
1683 </div>
1684 <!-- _______________________________________________________________________ -->
1687 </div>
1692 <pre>
1693 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1694 </pre>
1700 instruction, allowing a branch to occur to one of many possible
1701 destinations.</p>
1709 table is not allowed to contain duplicate constant entries.</p>
1715 table is searched for the given value. If the value is found, control flow is
1716 transfered to the corresponding destination; otherwise, control flow is
1717 transfered to the default destination.</p>
1721 <p>Depending on properties of the target machine and the particular
1723 ways. For example, it could be generated as a series of chained conditional
1724 branches or with a lookup table.</p>
1728 <pre>
1731 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1734 switch i32 0, label %dest [ ]
1737 switch i32 %val, label %otherwise [ i32 0, label %onzero
1738 i32 1, label %onone
1739 i32 2, label %ontwo ]
1740 </pre>
1741 </div>
1743 <!-- _______________________________________________________________________ -->
1746 </div>
1752 <pre>
1753 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1755 </pre>
1760 function, with the possibility of control flow transfer to either the
1772 <ol>
1773 <li>
1775 convention</a> the call should use. If none is specified, the call defaults
1776 to using C calling conventions.
1777 </li>
1779 function value being invoked. In most cases, this is a direct function
1781 an arbitrary pointer to function value.
1782 </li>
1785 function to be invoked. </li>
1788 signature argument types. If the function signature indicates the function
1789 accepts a variable number of arguments, the extra arguments can be
1790 specified. </li>
1798 </ol>
1804 difference is that it establishes an association with a label, which is used by
1805 the runtime library to unwind the stack.</p>
1807 <p>This instruction is used in languages with destructors to ensure that proper
1809 exception. Additionally, this is important for implementation of
1813 <pre>
1814 %retval = invoke i32 %Test(i32 15) to label %Continue
1818 </pre>
1819 </div>
1822 <!-- _______________________________________________________________________ -->
1825 Instruction</a> </div>
1830 <pre>
1831 unwind
1832 </pre>
1837 at the first callee in the dynamic call stack which used an <a
1839 primarily used to implement exception handling.</p>
1844 immediately halt. The dynamic call stack is then searched for the first <a
1846 execution continues at the "exceptional" destination block specified by the
1848 dynamic call chain, undefined behavior results.</p>
1849 </div>
1851 <!-- _______________________________________________________________________ -->
1854 Instruction</a> </div>
1859 <pre>
1860 unreachable
1861 </pre>
1866 instruction is used to inform the optimizer that a particular portion of the
1867 code is not reachable. This can be used to indicate that the code after a
1868 no-return function cannot be reached, and other facts.</p>
1873 </div>
1877 <!-- ======================================================================= -->
1880 <p>Binary operators are used to do most of the computation in a
1881 program. They require two operands, execute an operation on them, and
1882 produce a single value. The operands might represent
1884 The result value of a binary operator is not
1885 necessarily the same type as its operands.</p>
1887 </div>
1888 <!-- _______________________________________________________________________ -->
1890 Instruction</a> </div>
1894 </pre>
1901 Both arguments must have identical types.</p>
1903 <p>The value produced is the integer or floating point sum of the two
1904 operands.</p>
1907 </pre>
1908 </div>
1909 <!-- _______________________________________________________________________ -->
1911 Instruction</a> </div>
1915 </pre>
1918 operands.</p>
1920 instruction present in most other intermediate representations.</p>
1924 values.
1926 Both arguments must have identical types.</p>
1928 <p>The value produced is the integer or floating point difference of
1929 the two operands.</p>
1933 </pre>
1934 </div>
1935 <!-- _______________________________________________________________________ -->
1937 Instruction</a> </div>
1941 </pre>
1944 operands.</p>
1948 values.
1950 Both arguments must have identical types.</p>
1952 <p>The value produced is the integer or floating point product of the
1953 two operands.</p>
1954 <p>Because the operands are the same width, the result of an integer
1955 multiplication is the same whether the operands should be deemed unsigned or
1956 signed.</p>
1959 </pre>
1960 </div>
1961 <!-- _______________________________________________________________________ -->
1963 </a></div>
1967 </pre>
1970 operands.</p>
1975 of the values in which case the elements must be integers.</p>
1977 <p>The value produced is the unsigned integer quotient of the two operands. This
1978 instruction always performs an unsigned division operation, regardless of
1979 whether the arguments are unsigned or not.</p>
1982 </pre>
1983 </div>
1984 <!-- _______________________________________________________________________ -->
1986 </a> </div>
1990 </pre>
1993 operands.</p>
1998 of the values in which case the elements must be integers.</p>
2000 <p>The value produced is the signed integer quotient of the two operands. This
2001 instruction always performs a signed division operation, regardless of whether
2002 the arguments are signed or not.</p>
2005 </pre>
2006 </div>
2007 <!-- _______________________________________________________________________ -->
2009 Instruction</a> </div>
2013 </pre>
2016 operands.</p>
2021 versions of floating point values.</p>
2026 </pre>
2027 </div>
2028 <!-- _______________________________________________________________________ -->
2030 </div>
2034 </pre>
2037 unsigned division of its two arguments.</p>
2041 types.</p>
2044 This instruction always performs an unsigned division to get the remainder,
2045 regardless of whether the arguments are unsigned or not.</p>
2048 </pre>
2050 </div>
2051 <!-- _______________________________________________________________________ -->
2053 Instruction</a> </div>
2057 </pre>
2060 signed division of its two operands.</p>
2064 types.</p>
2069 a value. For more information about the difference, see <a
2071 Math Forum</a>. For a table of how this is implemented in various languages,
2076 </pre>
2078 </div>
2079 <!-- _______________________________________________________________________ -->
2081 Instruction</a> </div>
2085 </pre>
2088 division of its two operands.</p>
2092 identical types.</p>
2097 </pre>
2098 </div>
2100 <!-- ======================================================================= -->
2102 Operations</a> </div>
2104 <p>Bitwise binary operators are used to do various forms of
2105 bit-twiddling in a program. They are generally very efficient
2106 instructions and can commonly be strength reduced from other
2107 instructions. They require two operands, execute an operation on them,
2108 and produce a single value. The resulting value of the bitwise binary
2109 operators is always the same type as its first operand.</p>
2110 </div>
2112 <!-- _______________________________________________________________________ -->
2114 Instruction</a> </div>
2118 </pre>
2121 the left a specified number of bits.</p>
2131 </pre>
2132 </div>
2133 <!-- _______________________________________________________________________ -->
2135 Instruction</a> </div>
2139 </pre>
2143 operand shifted to the right a specified number of bits with zero fill.</p>
2150 <p>This instruction always performs a logical shift right operation. The most
2151 significant bits of the result will be filled with zero bits after the
2152 shift.</p>
2155 <pre>
2160 </pre>
2161 </div>
2163 <!-- _______________________________________________________________________ -->
2165 Instruction</a> </div>
2170 </pre>
2174 operand shifted to the right a specified number of bits with sign extension.</p>
2181 <p>This instruction always performs an arithmetic shift right operation,
2182 The most significant bits of the result will be filled with the sign bit
2186 <pre>
2191 </pre>
2192 </div>
2194 <!-- _______________________________________________________________________ -->
2196 Instruction</a> </div>
2200 </pre>
2203 its two operands.</p>
2207 identical types.</p>
2210 <p> </p>
2213 <tbody>
2214 <tr>
2218 </tr>
2219 <tr>
2223 </tr>
2224 <tr>
2228 </tr>
2229 <tr>
2233 </tr>
2234 <tr>
2238 </tr>
2239 </tbody>
2240 </table>
2241 </div>
2246 </pre>
2247 </div>
2248 <!-- _______________________________________________________________________ -->
2253 </pre>
2256 or of its two operands.</p>
2260 identical types.</p>
2263 <p> </p>
2266 <tbody>
2267 <tr>
2271 </tr>
2272 <tr>
2276 </tr>
2277 <tr>
2281 </tr>
2282 <tr>
2286 </tr>
2287 <tr>
2291 </tr>
2292 </tbody>
2293 </table>
2294 </div>
2299 </pre>
2300 </div>
2301 <!-- _______________________________________________________________________ -->
2303 Instruction</a> </div>
2307 </pre>
2315 identical types.</p>
2318 <p> </p>
2321 <tbody>
2322 <tr>
2326 </tr>
2327 <tr>
2331 </tr>
2332 <tr>
2336 </tr>
2337 <tr>
2341 </tr>
2342 <tr>
2346 </tr>
2347 </tbody>
2348 </table>
2349 </div>
2350 <p> </p>
2356 </pre>
2357 </div>
2359 <!-- ======================================================================= -->
2362 </div>
2366 <p>LLVM supports several instructions to represent vector operations in a
2367 target-independent manner. These instructions cover the element-access and
2368 vector-specific operations needed to process vectors effectively. While LLVM
2369 does directly support these vector operations, many sophisticated algorithms
2370 will want to use target-specific intrinsics to take full advantage of a specific
2371 target.</p>
2373 </div>
2375 <!-- _______________________________________________________________________ -->
2378 </div>
2384 <pre>
2385 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2386 </pre>
2390 <p>
2392 element from a vector at a specified index.
2393 </p>
2398 <p>
2401 an index indicating the position from which to extract the element.
2402 The index may be a variable.</p>
2406 <p>
2407 The result is a scalar of the same type as the element type of
2410 results are undefined.
2411 </p>
2415 <pre>
2417 </pre>
2418 </div>
2421 <!-- _______________________________________________________________________ -->
2424 </div>
2430 <pre>
2431 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2432 </pre>
2436 <p>
2438 element into a vector at a specified index.
2439 </p>
2444 <p>
2447 scalar value whose type must equal the element type of the first
2448 operand. The third operand is an index indicating the position at
2449 which to insert the value. The index may be a variable.</p>
2453 <p>
2458 </p>
2462 <pre>
2464 </pre>
2465 </div>
2467 <!-- _______________________________________________________________________ -->
2470 </div>
2476 <pre>
2477 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2478 </pre>
2482 <p>
2484 from two input vectors, returning a vector of the same type.
2485 </p>
2489 <p>
2491 with types that match each other and types that match the result of the
2492 instruction. The third argument is a shuffle mask, which has the same number
2493 of elements as the other vector type, but whose element type is always 'i32'.
2494 </p>
2496 <p>
2497 The shuffle mask operand is required to be a constant vector with either
2498 constant integer or undef values.
2499 </p>
2503 <p>
2504 The elements of the two input vectors are numbered from left to right across
2505 both of the vectors. The shuffle mask operand specifies, for each element of
2506 the result vector, which element of the two input registers the result element
2507 gets. The element selector may be undef (meaning "don't care") and the second
2508 operand may be undef if performing a shuffle from only one vector.
2509 </p>
2513 <pre>
2517 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2518 </pre>
2519 </div>
2522 <!-- ======================================================================= -->
2525 </div>
2529 <p>A key design point of an SSA-based representation is how it
2530 represents memory. In LLVM, no memory locations are in SSA form, which
2531 makes things very simple. This section describes how to read, write,
2532 allocate, and free memory in LLVM.</p>
2534 </div>
2536 <!-- _______________________________________________________________________ -->
2539 </div>
2545 <pre>
2546 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2547 </pre>
2552 heap and returns a pointer to it.</p>
2558 bytes of memory from the operating system and returns a pointer of the
2559 appropriate type to the program. If "NumElements" is specified, it is the
2560 number of elements allocated. If an alignment is specified, the value result
2561 of the allocation is guaranteed to be aligned to at least that boundary. If
2562 not specified, or if zero, the target can choose to align the allocation on any
2563 convenient boundary.</p>
2570 a pointer is returned.</p>
2574 <pre>
2582 </pre>
2583 </div>
2585 <!-- _______________________________________________________________________ -->
2588 </div>
2594 <pre>
2596 </pre>
2601 memory heap to be reallocated in the future.</p>
2607 instruction.</p>
2611 <p>Access to the memory pointed to by the pointer is no longer defined
2612 after this instruction executes.</p>
2616 <pre>
2618 free [4 x i8]* %array
2619 </pre>
2620 </div>
2622 <!-- _______________________________________________________________________ -->
2625 </div>
2631 <pre>
2632 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2633 </pre>
2638 currently executing function, to be automatically released when this function
2639 returns to its caller.</p>
2644 bytes of memory on the runtime stack, returning a pointer of the
2645 appropriate type to the program. If "NumElements" is specified, it is the
2646 number of elements allocated. If an alignment is specified, the value result
2647 of the allocation is guaranteed to be aligned to at least that boundary. If
2648 not specified, or if zero, the target can choose to align the allocation on any
2649 convenient boundary.</p>
2657 instruction is commonly used to represent automatic variables that must
2658 have an address available. When the function returns (either with the <tt><a
2660 instructions), the memory is reclaimed.</p>
2664 <pre>
2669 </pre>
2670 </div>
2672 <!-- _______________________________________________________________________ -->
2674 Instruction</a> </div>
2677 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
2682 address from which to load. The pointer must point to a <a
2687 instructions. </p>
2692 <a
2695 </pre>
2696 </div>
2697 <!-- _______________________________________________________________________ -->
2699 Instruction</a> </div>
2702 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2703 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2704 </pre>
2712 optimizer is not allowed to modify the number or order of execution of
2720 <a
2723 </pre>
2724 </div>
2726 <!-- _______________________________________________________________________ -->
2729 </div>
2733 <pre>
2735 </pre>
2739 <p>
2741 subelement of an aggregate data structure.</p>
2745 <p>This instruction takes a list of integer operands that indicate what
2746 elements of the aggregate object to index to. The actual types of the arguments
2747 provided depend on the type of the first pointer argument. The
2749 levels of a structure or to a specific index in an array. When indexing into a
2754 <p>For example, let's consider a C code fragment and how it gets
2755 compiled to LLVM:</p>
2757 <pre>
2758 struct RT {
2759 char A;
2761 char C;
2762 };
2763 struct ST {
2764 i32 X;
2765 double Y;
2766 struct RT Z;
2767 };
2769 define i32 *foo(struct ST *s) {
2771 }
2772 </pre>
2776 <pre>
2778 %ST = type { i32, double, %RT }
2780 define i32* %foo(%ST* %s) {
2781 entry:
2783 ret i32* %reg
2784 }
2785 </pre>
2798 }</tt>' type, a structure. The second index indexes into the third element of
2800 i8 }</tt>' type, another structure. The third index indexes into the second
2802 array. The two dimensions of the array are subscripted into, yielding an
2806 <p>Note that it is perfectly legal to index partially through a
2807 structure, returning a pointer to an inner element. Because of this,
2808 the LLVM code for the given testcase is equivalent to:</p>
2810 <pre>
2811 define i32* %foo(%ST* %s) {
2817 ret i32* %t5
2818 }
2819 </pre>
2821 <p>Note that it is undefined to access an array out of bounds: array and
2822 pointer indexes must always be within the defined bounds of the array type.
2823 The one exception for this rules is zero length arrays. These arrays are
2824 defined to be accessible as variable length arrays, which requires access
2825 beyond the zero'th element.</p>
2827 <p>The getelementptr instruction is often confusing. For some more insight
2833 <pre>
2836 </pre>
2837 </div>
2839 <!-- ======================================================================= -->
2841 </div>
2843 <p>The instructions in this category are the conversion instructions (casting)
2844 which all take a single operand and a type. They perform various bit conversions
2845 on the operand.</p>
2846 </div>
2848 <!-- _______________________________________________________________________ -->
2851 </div>
2855 <pre>
2857 </pre>
2860 <p>
2862 </p>
2865 <p>
2873 <p>
2877 It will always truncate bits.</p>
2880 <pre>
2884 </pre>
2885 </div>
2887 <!-- _______________________________________________________________________ -->
2890 </div>
2894 <pre>
2896 </pre>
2913 the operand and the type are the same size, no bit filling is done and the
2915 changes).</p>
2920 <pre>
2923 </pre>
2924 </div>
2926 <!-- _______________________________________________________________________ -->
2929 </div>
2933 <pre>
2935 </pre>
2941 <p>
2949 <p>
2954 no bits change (only the type changes).</p>
2959 <pre>
2962 </pre>
2963 </div>
2965 <!-- _______________________________________________________________________ -->
2968 </div>
2974 <pre>
2976 </pre>
2997 <pre>
3000 </pre>
3001 </div>
3003 <!-- _______________________________________________________________________ -->
3006 </div>
3010 <pre>
3012 </pre>
3016 floating point value.</p>
3022 type must be smaller than the destination type.</p>
3032 <pre>
3035 </pre>
3036 </div>
3038 <!-- _______________________________________________________________________ -->
3041 </div>
3045 <pre>
3047 </pre>
3052 </p>
3063 the results are undefined.</p>
3065 <p>When converting to i1, the conversion is done as a comparison against
3070 <pre>
3074 </pre>
3075 </div>
3077 <!-- _______________________________________________________________________ -->
3080 </div>
3084 <pre>
3086 </pre>
3091 </p>
3103 the results are undefined.</p>
3105 <p>When converting to i1, the conversion is done as a comparison against
3110 <pre>
3114 </pre>
3115 </div>
3117 <!-- _______________________________________________________________________ -->
3120 </div>
3124 <pre>
3126 </pre>
3140 integer quantity and converts it to the corresponding floating point value. If
3141 the value cannot fit in the floating point value, the results are undefined.</p>
3145 <pre>
3148 </pre>
3149 </div>
3151 <!-- _______________________________________________________________________ -->
3154 </div>
3158 <pre>
3160 </pre>
3173 integer quantity and converts it to the corresponding floating point value. If
3174 the value cannot fit in the floating point value, the results are undefined.</p>
3177 <pre>
3180 </pre>
3181 </div>
3183 <!-- _______________________________________________________________________ -->
3186 </div>
3190 <pre>
3192 </pre>
3206 truncating or zero extending that value to the size of the integer type. If
3210 change.</p>
3213 <pre>
3216 </pre>
3217 </div>
3219 <!-- _______________________________________________________________________ -->
3222 </div>
3226 <pre>
3228 </pre>
3236 value to cast, and a type to cast it to, which must be a
3244 the size of a pointer then a zero extension is done. If they are the same size,
3248 <pre>
3252 </pre>
3253 </div>
3255 <!-- _______________________________________________________________________ -->
3258 </div>
3262 <pre>
3264 </pre>
3272 a first class value, and a type to cast it to, which must also be a <a
3275 type is a pointer, the destination type must also be a pointer.</p>
3282 converted to other pointer types with this instruction. To convert pointers to
3287 <pre>
3291 </pre>
3292 </div>
3294 <!-- ======================================================================= -->
3298 instructions, which defy better classification.</p>
3299 </div>
3301 <!-- _______________________________________________________________________ -->
3303 </div>
3306 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3307 </pre>
3310 of its two integer operands.</p>
3313 the condition code indicating the kind of comparison to perform. It is not
3314 a value, just a keyword. The possible condition code are:
3315 <ol>
3326 </ol>
3333 <ol>
3336 </li>
3355 </ol>
3357 values are compared as if they were integers.</p>
3366 </pre>
3367 </div>
3369 <!-- _______________________________________________________________________ -->
3371 </div>
3374 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3375 </pre>
3378 of its floating point operands.</p>
3381 the condition code indicating the kind of comparison to perform. It is not
3382 a value, just a keyword. The possible condition code are:
3383 <ol>
3400 </ol>
3405 types.</p>
3410 <ol>
3439 </ol>
3446 </pre>
3447 </div>
3449 <!-- _______________________________________________________________________ -->
3451 Instruction</a> </div>
3457 the SSA graph representing the function.</p>
3459 <p>The type of the incoming values is specified with the first type
3461 as arguments, with one pair for each predecessor basic block of the
3463 type may be used as the value arguments to the PHI node. Only labels
3464 may be used as the label arguments.</p>
3465 <p>There must be no non-phi instructions between the start of a basic
3466 block and the PHI instructions: i.e. PHI instructions must be first in
3467 a basic block.</p>
3470 specified by the pair corresponding to the predecessor basic block that executed
3471 just prior to the current block.</p>
3473 <pre>Loop: ; Infinite loop that counts from 0 on up...<br> %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]<br> %nextindvar = add i32 %indvar, 1<br> br label %Loop<br></pre>
3474 </div>
3476 <!-- _______________________________________________________________________ -->
3479 </div>
3485 <pre>
3486 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3487 </pre>
3491 <p>
3493 condition, without branching.
3494 </p>
3499 <p>
3500 The '<tt>select</tt>' instruction requires a boolean value indicating the condition, and two values of the same <a href="#t_firstclass">first class</a> type.
3501 </p>
3505 <p>
3506 If the boolean condition evaluates to true, the instruction returns the first
3507 value argument; otherwise, it returns the second value argument.
3508 </p>
3512 <pre>
3514 </pre>
3515 </div>
3518 <!-- _______________________________________________________________________ -->
3521 </div>
3526 <pre>
3527 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
3528 </pre>
3538 <ol>
3539 <li>
3541 any allocas or varargs in the caller. If the "tail" marker is present, the
3542 function call is eligible for tail call optimization. Note that calls may
3545 </li>
3546 <li>
3548 convention</a> the call should use. If none is specified, the call defaults
3549 to using C calling conventions.
3550 </li>
3551 <li>
3553 being invoked. The argument types must match the types implied by this
3554 signature. This type can be omitted if the function is not varargs and
3555 if the function type does not return a pointer to a function.</p>
3556 </li>
3557 <li>
3559 be invoked. In most cases, this is a direct function invocation, but
3561 to function value.</p>
3562 </li>
3563 <li>
3565 function signature argument types. All arguments must be of
3567 indicates the function accepts a variable number of arguments, the extra
3568 arguments can be specified.</p>
3569 </li>
3570 </ol>
3575 transfer to a specified function, with its incoming arguments bound to
3577 instruction in the called function, control flow continues with the
3578 instruction after the function call, and the return value of the
3579 function is bound to the result argument. This is a simpler case of
3584 <pre>
3585 %retval = call i32 %test(i32 %argc)
3587 %X = tail call i32 %foo()
3589 </pre>
3591 </div>
3593 <!-- _______________________________________________________________________ -->
3596 </div>
3602 <pre>
3604 </pre>
3609 the "variable argument" area of a function call. It is used to implement the
3615 the argument. It returns a value of the specified argument type and
3627 <p>It is legal for this instruction to be called in a function which does not
3629 function.</p>
3633 argument.</p>
3639 </div>
3641 <!-- *********************************************************************** -->
3643 <!-- *********************************************************************** -->
3648 well known names and semantics and are required to follow certain restrictions.
3649 Overall, these intrinsics represent an extension mechanism for the LLVM
3650 language that does not require changing all of the transformations in LLVM when
3651 adding to the language (or the bytecode reader/writer, the parser, etc...).</p>
3654 prefix is reserved in LLVM for intrinsic names; thus, function names may not
3655 begin with this prefix. Intrinsic functions must always be external functions:
3656 you cannot define the body of intrinsic functions. Intrinsic functions may
3657 only be used in call or invoke instructions: it is illegal to take the address
3658 of an intrinsic function. Additionally, because intrinsic functions are part
3659 of the LLVM language, it is required if any are added that they be documented
3660 here.</p>
3662 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3663 a family of functions that perform the same operation but on different data
3664 types. This is most frequent with the integer types. Since LLVM can represent
3665 over 8 million different integer types, there is a way to declare an intrinsic
3666 that can be overloaded based on its arguments. Such an intrinsic will have the
3667 names of its argument types encoded into its function name, each
3669 integer of any width. This leads to a family of functions such as
3671 </p>
3674 <p>To learn how to add an intrinsic function, please see the
3676 </p>
3678 </div>
3680 <!-- ======================================================================= -->
3683 </div>
3689 intrinsic functions. These functions are related to the similarly
3692 <p>All of these functions operate on arguments that use a
3693 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3694 language reference manual does not define what this type is, so all
3695 transformations should be prepared to handle these functions regardless of
3696 the type used.</p>
3699 instruction and the variable argument handling intrinsic functions are
3700 used.</p>
3702 <pre>
3703 define i32 @test(i32 %X, ...) {
3704 ; Initialize variable argument processing
3705 %ap = alloca i8*
3706 %ap2 = bitcast i8** %ap to i8*
3707 call void @llvm.va_start(i8* %ap2)
3709 ; Read a single integer argument
3710 %tmp = va_arg i8** %ap, i32
3712 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3713 %aq = alloca i8*
3714 %aq2 = bitcast i8** %aq to i8*
3715 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
3716 call void @llvm.va_end(i8* %aq2)
3718 ; Stop processing of arguments.
3719 call void @llvm.va_end(i8* %ap2)
3720 ret i32 %tmp
3721 }
3723 declare void @llvm.va_start(i8*)
3724 declare void @llvm.va_copy(i8*, i8*)
3725 declare void @llvm.va_end(i8*)
3726 </pre>
3727 </div>
3729 <!-- _______________________________________________________________________ -->
3732 </div>
3750 macro available in C. In a target-dependent way, it initializes the
3754 last argument of the function as the compiler can figure that out.</p>
3756 </div>
3758 <!-- _______________________________________________________________________ -->
3761 </div>
3779 macro available in C. In a target-dependent way, it destroys the
3785 </div>
3787 <!-- _______________________________________________________________________ -->
3790 </div>
3796 <pre>
3798 </pre>
3803 from the source argument list to the destination argument list.</p>
3814 macro available in C. In a target-dependent way, it copies the source
3817 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
3818 example, memory allocation.</p>
3820 </div>
3822 <!-- ======================================================================= -->
3825 </div>
3829 <p>
3831 Collection</a> requires the implementation and generation of these intrinsics.
3835 Front-ends for type-safe garbage collected languages should generate these
3836 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
3838 </p>
3839 </div>
3841 <!-- _______________________________________________________________________ -->
3844 </div>
3850 <pre>
3852 </pre>
3857 the code generator, and allows some metadata to be associated with it.</p>
3861 <p>The first argument specifies the address of a stack object that contains the
3862 root pointer. The second pointer (which must be either a constant or a global
3863 value address) contains the meta-data to be associated with the root.</p>
3868 location. At compile-time, the code generator generates information to allow
3869 the runtime to find the pointer at GC safe points.
3870 </p>
3872 </div>
3875 <!-- _______________________________________________________________________ -->
3878 </div>
3884 <pre>
3885 declare i8 * @llvm.gcread(i8 * %ObjPtr, i8 ** %Ptr)
3886 </pre>
3891 locations, allowing garbage collector implementations that require read
3892 barriers.</p>
3896 <p>The second argument is the address to read from, which should be an address
3897 allocated from the garbage collector. The first object is a pointer to the
3898 start of the referenced object, if needed by the language runtime (otherwise
3899 null).</p>
3904 instruction, but may be replaced with substantially more complex code by the
3905 garbage collector runtime, as needed.</p>
3907 </div>
3910 <!-- _______________________________________________________________________ -->
3913 </div>
3919 <pre>
3920 declare void @llvm.gcwrite(i8 * %P1, i8 * %Obj, i8 ** %P2)
3921 </pre>
3926 locations, allowing garbage collector implementations that require write
3927 barriers (such as generational or reference counting collectors).</p>
3931 <p>The first argument is the reference to store, the second is the start of the
3932 object to store it to, and the third is the address of the field of Obj to
3933 store to. If the runtime does not require a pointer to the object, Obj may be
3934 null.</p>
3939 instruction, but may be replaced with substantially more complex code by the
3940 garbage collector runtime, as needed.</p>
3942 </div>
3946 <!-- ======================================================================= -->
3949 </div>
3952 <p>
3953 These intrinsics are provided by LLVM to expose special features that may only
3954 be implemented with code generator support.
3955 </p>
3957 </div>
3959 <!-- _______________________________________________________________________ -->
3962 </div>
3967 <pre>
3969 </pre>
3973 <p>
3975 target-specific value indicating the return address of the current function
3976 or one of its callers.
3977 </p>
3981 <p>
3982 The argument to this intrinsic indicates which function to return the address
3983 for. Zero indicates the calling function, one indicates its caller, etc. The
3985 </p>
3989 <p>
3991 the return address of the specified call frame, or zero if it cannot be
3992 identified. The value returned by this intrinsic is likely to be incorrect or 0
3993 for arguments other than zero, so it should only be used for debugging purposes.
3994 </p>
3996 <p>
3997 Note that calling this intrinsic does not prevent function inlining or other
3998 aggressive transformations, so the value returned may not be that of the obvious
3999 source-language caller.
4000 </p>
4001 </div>
4004 <!-- _______________________________________________________________________ -->
4007 </div>
4012 <pre>
4014 </pre>
4018 <p>
4020 target-specific frame pointer value for the specified stack frame.
4021 </p>
4025 <p>
4026 The argument to this intrinsic indicates which function to return the frame
4027 pointer for. Zero indicates the calling function, one indicates its caller,
4029 </p>
4033 <p>
4035 the frame address of the specified call frame, or zero if it cannot be
4036 identified. The value returned by this intrinsic is likely to be incorrect or 0
4037 for arguments other than zero, so it should only be used for debugging purposes.
4038 </p>
4040 <p>
4041 Note that calling this intrinsic does not prevent function inlining or other
4042 aggressive transformations, so the value returned may not be that of the obvious
4043 source-language caller.
4044 </p>
4045 </div>
4047 <!-- _______________________________________________________________________ -->
4050 </div>
4055 <pre>
4056 declare i8 *@llvm.stacksave()
4057 </pre>
4061 <p>
4065 features like scoped automatic variable sized arrays in C99.
4066 </p>
4070 <p>
4071 This intrinsic returns a opaque pointer value that can be passed to <a
4078 </p>
4080 </div>
4082 <!-- _______________________________________________________________________ -->
4085 </div>
4090 <pre>
4091 declare void @llvm.stackrestore(i8 * %ptr)
4092 </pre>
4096 <p>
4098 the function stack to the state it was in when the corresponding <a
4100 useful for implementing language features like scoped automatic variable sized
4101 arrays in C99.
4102 </p>
4106 <p>
4108 </p>
4110 </div>
4113 <!-- _______________________________________________________________________ -->
4116 </div>
4121 <pre>
4124 </pre>
4129 <p>
4131 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4132 no
4133 effect on the behavior of the program but can change its performance
4134 characteristics.
4135 </p>
4139 <p>
4145 </p>
4149 <p>
4150 This intrinsic does not modify the behavior of the program. In particular,
4151 prefetches cannot trap and do not produce a value. On targets that support this
4152 intrinsic, the prefetch can provide hints to the processor cache for better
4153 performance.
4154 </p>
4156 </div>
4158 <!-- _______________________________________________________________________ -->
4161 </div>
4166 <pre>
4168 </pre>
4173 <p>
4175 (PC) in a region of
4176 code to simulators and other tools. The method is target specific, but it is
4177 expected that the marker will use exported symbols to transmit the PC of the marker.
4178 The marker makes no guarantees that it will remain with any specific instruction
4179 after optimizations. It is possible that the presence of a marker will inhibit
4180 optimizations. The intended use is to be inserted after optimizations to allow
4181 correlations of simulation runs.
4182 </p>
4186 <p>
4188 </p>
4192 <p>
4193 This intrinsic does not modify the behavior of the program. Backends that do not
4194 support this intrinisic may ignore it.
4195 </p>
4197 </div>
4199 <!-- _______________________________________________________________________ -->
4202 </div>
4207 <pre>
4208 declare i64 @llvm.readcyclecounter( )
4209 </pre>
4214 <p>
4216 counter register (or similar low latency, high accuracy clocks) on those targets
4217 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4218 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4219 should only be used for small timings.
4220 </p>
4224 <p>
4225 When directly supported, reading the cycle counter should not modify any memory.
4226 Implementations are allowed to either return a application specific value or a
4227 system wide value. On backends without support, this is lowered to a constant 0.
4228 </p>
4230 </div>
4232 <!-- ======================================================================= -->
4235 </div>
4238 <p>
4239 LLVM provides intrinsics for a few important standard C library functions.
4240 These intrinsics allow source-language front-ends to pass information about the
4241 alignment of the pointer arguments to the code generator, providing opportunity
4242 for more efficient code generation.
4243 </p>
4245 </div>
4247 <!-- _______________________________________________________________________ -->
4250 </div>
4255 <pre>
4260 </pre>
4264 <p>
4266 location to the destination location.
4267 </p>
4269 <p>
4271 intrinsics do not return a value, and takes an extra alignment argument.
4272 </p>
4276 <p>
4277 The first argument is a pointer to the destination, the second is a pointer to
4278 the source. The third argument is an integer argument
4279 specifying the number of bytes to copy, and the fourth argument is the alignment
4280 of the source and destination locations.
4281 </p>
4283 <p>
4285 the caller guarantees that both the source and destination pointers are aligned
4286 to that boundary.
4287 </p>
4291 <p>
4293 location to the destination location, which are not allowed to overlap. It
4294 copies "len" bytes of memory over. If the argument is known to be aligned to
4295 some boundary, this can be specified as the fourth argument, otherwise it should
4297 </p>
4298 </div>
4301 <!-- _______________________________________________________________________ -->
4304 </div>
4309 <pre>
4314 </pre>
4318 <p>
4320 location to the destination location. It is similar to the
4322 </p>
4324 <p>
4326 intrinsics do not return a value, and takes an extra alignment argument.
4327 </p>
4331 <p>
4332 The first argument is a pointer to the destination, the second is a pointer to
4333 the source. The third argument is an integer argument
4334 specifying the number of bytes to copy, and the fourth argument is the alignment
4335 of the source and destination locations.
4336 </p>
4338 <p>
4340 the caller guarantees that the source and destination pointers are aligned to
4341 that boundary.
4342 </p>
4346 <p>
4348 location to the destination location, which may overlap. It
4349 copies "len" bytes of memory over. If the argument is known to be aligned to
4350 some boundary, this can be specified as the fourth argument, otherwise it should
4352 </p>
4353 </div>
4356 <!-- _______________________________________________________________________ -->
4359 </div>
4364 <pre>
4369 </pre>
4373 <p>
4375 byte value.
4376 </p>
4378 <p>
4380 does not return a value, and takes an extra alignment argument.
4381 </p>
4385 <p>
4386 The first argument is a pointer to the destination to fill, the second is the
4387 byte value to fill it with, the third argument is an integer
4388 argument specifying the number of bytes to fill, and the fourth argument is the
4389 known alignment of destination location.
4390 </p>
4392 <p>
4394 the caller guarantees that the destination pointer is aligned to that boundary.
4395 </p>
4399 <p>
4401 the
4402 destination location. If the argument is known to be aligned to some boundary,
4403 this can be specified as the fourth argument, otherwise it should be set to 0 or
4404 1.
4405 </p>
4406 </div>
4409 <!-- _______________________________________________________________________ -->
4412 </div>
4417 <pre>
4418 declare float @llvm.sqrt.f32(float %Val)
4419 declare double @llvm.sqrt.f64(double %Val)
4420 </pre>
4424 <p>
4428 negative numbers (which allows for better optimization).
4429 </p>
4433 <p>
4434 The argument and return value are floating point numbers of the same type.
4435 </p>
4439 <p>
4440 This function returns the sqrt of the specified operand if it is a positive
4441 floating point number.
4442 </p>
4443 </div>
4445 <!-- _______________________________________________________________________ -->
4448 </div>
4453 <pre>
4454 declare float @llvm.powi.f32(float %Val, i32 %power)
4455 declare double @llvm.powi.f64(double %Val, i32 %power)
4456 </pre>
4460 <p>
4462 specified (positive or negative) power. The order of evaluation of
4463 multiplications is not defined.
4464 </p>
4468 <p>
4469 The second argument is an integer power, and the first is a value to raise to
4470 that power.
4471 </p>
4475 <p>
4476 This function returns the first value raised to the second power with an
4477 unspecified sequence of rounding operations.</p>
4478 </div>
4481 <!-- ======================================================================= -->
4484 </div>
4487 <p>
4488 LLVM provides intrinsics for a few important bit manipulation operations.
4489 These allow efficient code generation for some algorithms.
4490 </p>
4492 </div>
4494 <!-- _______________________________________________________________________ -->
4497 </div>
4502 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4504 that includes the type for the result and the operand.
4505 <pre>
4509 </pre>
4513 <p>
4515 values with an even number of bytes (positive multiple of 16 bits). These are
4516 useful for performing operations on data that is not in the target's native
4517 byte order.
4518 </p>
4522 <p>
4525 intrinsic returns an i32 value that has the four bytes of the input i32
4530 </p>
4532 </div>
4534 <!-- _______________________________________________________________________ -->
4537 </div>
4542 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4543 width. Not all targets support all bit widths however.
4544 <pre>
4550 </pre>
4554 <p>
4556 value.
4557 </p>
4561 <p>
4562 The only argument is the value to be counted. The argument may be of any
4563 integer type. The return type must match the argument type.
4564 </p>
4568 <p>
4570 </p>
4571 </div>
4573 <!-- _______________________________________________________________________ -->
4576 </div>
4582 integer bit width. Not all targets support all bit widths however.
4583 <pre>
4589 </pre>
4593 <p>
4595 leading zeros in a variable.
4596 </p>
4600 <p>
4601 The only argument is the value to be counted. The argument may be of any
4602 integer type. The return type must match the argument type.
4603 </p>
4607 <p>
4609 in a variable. If the src == 0 then the result is the size in bits of the type
4611 </p>
4612 </div>
4616 <!-- _______________________________________________________________________ -->
4619 </div>
4625 integer bit width. Not all targets support all bit widths however.
4626 <pre>
4632 </pre>
4636 <p>
4638 trailing zeros.
4639 </p>
4643 <p>
4644 The only argument is the value to be counted. The argument may be of any
4645 integer type. The return type must match the argument type.
4646 </p>
4650 <p>
4652 in a variable. If the src == 0 then the result is the size in bits of the type
4654 </p>
4655 </div>
4657 <!-- _______________________________________________________________________ -->
4660 </div>
4666 on any integer bit width.
4667 <pre>
4668 declare i17 @llvm.part.select.i17.i17 (i17 %val, i32 %loBit, i32 %hiBit)
4669 declare i29 @llvm.part.select.i29.i29 (i29 %val, i32 %loBit, i32 %hiBit)
4670 </pre>
4674 range of bits from an integer value and returns them in the same bit width as
4675 the original value.</p>
4679 any bit width but they must have the same bit width. The second and third
4686 operates in forward mode.</p>
4690 <ol>
4694 to determine the number of bits to retain.</li>
4697 </ol>
4698 <p>In reverse mode, a similar computation is made except that the bits are
4703 </div>
4707 </div>
4713 on any integer bit width.
4714 <pre>
4715 declare i17 @llvm.part.set.i17.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
4716 declare i29 @llvm.part.set.i29.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
4717 </pre>
4721 of bits in an integer value with another integer value. It returns the integer
4722 with the replaced bits.</p>
4729 type since they specify only a bit index.</p>
4735 operates in forward mode.</p>
4737 truncating it down to the size of the replacement area or zero extending it
4738 up to that size.</p>
4743 <p>In reverse mode, a similar computation is made except that the bits are
4747 <pre>
4753 </pre>
4754 </div>
4756 <!-- ======================================================================= -->
4759 </div>
4762 <p>
4764 are described in the <a
4766 Debugging</a> document.
4767 </p>
4768 </div>
4771 <!-- ======================================================================= -->
4774 </div>
4777 <p> The LLVM exception handling intrinsics (which all start with
4781 </div>
4784 <!-- *********************************************************************** -->
4785 <hr>
4786 <address>
4794 Last modified: $Date$
4795 </address>
4796 </body>
4797 </html>