| blob | 1c587f88e62ebeace2e9d5d896a4303be31ee3cd |
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>
256 <pre>
258 </pre>
259 </div>
262 its uses. The LLVM infrastructure provides a verification pass that may
263 be used to verify that an LLVM module is well formed. This pass is
264 automatically run by the parser after parsing input assembly and by
265 the optimizer before it outputs bytecode. The violations pointed out
266 by the verifier pass indicate bugs in transformation passes or input to
267 the parser.</p>
268 </div>
272 <!-- *********************************************************************** -->
274 <!-- *********************************************************************** -->
278 <p>LLVM uses three different forms of identifiers, for different
279 purposes:</p>
281 <ol>
282 <li>Named values are represented as a string of characters with a '%' prefix.
283 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
285 Identifiers which require other characters in their names can be surrounded
287 in a name.</li>
289 <li>Unnamed values are represented as an unsigned numeric value with a '%'
294 </ol>
296 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
297 don't need to worry about name clashes with reserved words, and the set of
298 reserved words may be expanded in the future without penalty. Additionally,
299 unnamed identifiers allow a compiler to quickly come up with a temporary
300 variable without having to avoid symbol table conflicts.</p>
302 <p>Reserved words in LLVM are very similar to reserved words in other
303 languages. There are keywords for different opcodes
308 and others. These reserved words cannot conflict with variable names, because
309 none of them start with a '%' character.</p>
311 <p>Here is an example of LLVM code to multiply the integer variable
317 <pre>
319 </pre>
320 </div>
325 <pre>
327 </pre>
328 </div>
333 <pre>
337 </pre>
338 </div>
341 important lexical features of LLVM:</p>
343 <ol>
346 line.</li>
348 <li>Unnamed temporaries are created when the result of a computation is not
349 assigned to a named value.</li>
353 </ol>
355 <p>...and it also shows a convention that we follow in this document. When
356 demonstrating instructions, we will follow an instruction with a comment that
357 defines the type and name of value produced. Comments are shown in italic
358 text.</p>
360 </div>
362 <!-- *********************************************************************** -->
364 <!-- *********************************************************************** -->
366 <!-- ======================================================================= -->
368 </div>
373 translation unit of the input programs. Each module consists of
374 functions, global variables, and symbol table entries. Modules may be
375 combined together with the LLVM linker, which merges function (and
376 global variable) definitions, resolves forward declarations, and merges
382 href="#globalvars">constant</a> <a href="#t_array">[13 x i8 ]</a> c"hello world\0A\00" <i>; [13 x i8 ]*</i>
390 %cast210 = <a
394 <a
396 <a
398 </pre>
399 </div>
406 <p>In general, a module is made up of a list of global values,
407 where both functions and global variables are global values. Global values are
408 represented by a pointer to a memory location (in this case, a pointer to an
409 array of char, and a pointer to a function), and have one of the following <a
412 </div>
414 <!-- ======================================================================= -->
417 </div>
421 <p>
422 All Global Variables and Functions have one of the following types of linkage:
423 </p>
425 <dl>
429 <dd>Global values with internal linkage are only directly accessible by
430 objects in the current module. In particular, linking code into a module with
431 an internal global value may cause the internal to be renamed as necessary to
432 avoid collisions. Because the symbol is internal to the module, all
433 references can be updated. This corresponds to the notion of the
435 </dd>
440 the same name when linkage occurs. This is typically used to implement
441 inline functions, templates, or other code which must be generated in each
443 allowed to be discarded.
444 </dd>
450 used for globals that may be emitted in multiple translation units, but that
451 are not guaranteed to be emitted into every translation unit that uses them.
452 One example of this are common globals in C, such as "<tt>int X;</tt>" at
453 global scope.
454 </dd>
459 pointer to array type. When two global variables with appending linkage are
460 linked together, the two global arrays are appended together. This is the
461 LLVM, typesafe, equivalent of having the system linker append together
462 "sections" with identical names when .o files are linked.
463 </dd>
466 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
467 until linked, if not linked, the symbol becomes null instead of being an
468 undefined reference.
469 </dd>
473 <dd>If none of the above identifiers are used, the global is externally
474 visible, meaning that it participates in linkage and can be used to resolve
475 external symbol references.
476 </dd>
477 </dl>
479 <p>
480 The next two types of linkage are targeted for Microsoft Windows platform
481 only. They are designed to support importing (exporting) symbols from (to)
482 DLLs.
483 </p>
485 <dl>
489 or variable via a global pointer to a pointer that is set up by the DLL
490 exporting the symbol. On Microsoft Windows targets, the pointer name is
492 </dd>
497 pointer to a pointer in a DLL, so that it can be referenced with the
500 name.
501 </dd>
503 </dl>
506 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
507 variable and was linked with this one, one of the two would be renamed,
509 external (i.e., lacking any linkage declarations), they are accessible
510 outside of the current module.</p>
515 linkages.
516 </div>
518 <!-- ======================================================================= -->
521 </div>
527 specified for the call. The calling convention of any pair of dynamic
528 caller/callee must match, or the behavior of the program is undefined. The
529 following calling conventions are supported by LLVM, and more may be added in
530 the future:</p>
532 <dl>
535 <dd>This calling convention (the default if no other calling convention is
536 specified) matches the target C calling conventions. This calling convention
537 supports varargs function calls and tolerates some mismatch in the declared
538 prototype and implemented declaration of the function (as does normal C).
539 </dd>
543 <dd>This calling convention attempts to make calls as fast as possible
544 (e.g. by passing things in registers). This calling convention allows the
545 target to use whatever tricks it wants to produce fast code for the target,
546 without having to conform to an externally specified ABI. Implementations of
547 this convention should allow arbitrary tail call optimization to be supported.
548 This calling convention does not support varargs and requires the prototype of
549 all callees to exactly match the prototype of the function definition.
550 </dd>
554 <dd>This calling convention attempts to make code in the caller as efficient
555 as possible under the assumption that the call is not commonly executed. As
556 such, these calls often preserve all registers so that the call does not break
557 any live ranges in the caller side. This calling convention does not support
558 varargs and requires the prototype of all callees to exactly match the
559 prototype of the function definition.
560 </dd>
564 <dd>Any calling convention may be specified by number, allowing
565 target-specific calling conventions to be used. Target specific calling
566 conventions start at 64.
567 </dd>
568 </dl>
570 <p>More calling conventions can be added/defined on an as-needed basis, to
571 support pascal conventions or any other well-known target-independent
572 convention.</p>
574 </div>
576 <!-- ======================================================================= -->
579 </div>
583 <p>
584 All Global Variables and Functions have one of the following visibility styles:
585 </p>
587 <dl>
590 <dd>On ELF, default visibility means that the declaration is visible to other
591 modules and, in shared libraries, means that the declared entity may be
592 overridden. On Darwin, default visibility means that the declaration is
593 visible to other modules. Default visibility corresponds to "external
594 linkage" in the language.
595 </dd>
599 <dd>Two declarations of an object with hidden visibility refer to the same
600 object if they are in the same shared object. Usually, hidden visibility
601 indicates that the symbol will not be placed into the dynamic symbol table,
602 so no other module (executable or shared library) can reference it
603 directly.
604 </dd>
608 <dd>On ELF, protected visibility indicates that the symbol will be placed in
609 the dynamic symbol table, but that references within the defining module will
610 bind to the local symbol. That is, the symbol cannot be overridden by another
611 module.
612 </dd>
613 </dl>
615 </div>
617 <!-- ======================================================================= -->
620 </div>
624 <p>Global variables define regions of memory allocated at compilation time
625 instead of run-time. Global variables may optionally be initialized, may have
626 an explicit section to be placed in, and may have an optional explicit alignment
627 specified. A variable may be defined as "thread_local", which means that it
628 will not be shared by threads (each thread will have a separated copy of the
629 variable). A variable may be defined as a global "constant," which indicates
631 optimization, allowing the global data to be placed in the read-only section of
632 an executable, etc). Note that variables that need runtime initialization
635 <p>
637 constant, even if the final definition of the global is not. This capability
638 can be used to enable slightly better optimization of the program, but requires
639 the language definition to guarantee that optimizations based on the
640 'constantness' are valid for the translation units that do not include the
641 definition.
642 </p>
644 <p>As SSA values, global variables define pointer values that are in
645 scope (i.e. they dominate) all basic blocks in the program. Global
646 variables always define a pointer to their "content" type because they
647 describe a region of memory, and all memory objects in LLVM are
648 accessed through pointers.</p>
650 <p>LLVM allows an explicit section to be specified for globals. If the target
651 supports it, it will emit globals to the section specified.</p>
653 <p>An explicit alignment may be specified for a global. If not present, or if
654 the alignment is set to zero, the alignment of the global is set by the target
655 to whatever it feels convenient. If an explicit alignment is specified, the
656 global is forced to have at least that much alignment. All alignments must be
659 <p>For example, the following defines a global with an initializer, section,
660 and alignment:</p>
663 <pre>
665 </pre>
666 </div>
668 </div>
671 <!-- ======================================================================= -->
674 </div>
683 name, a (possibly empty) argument list (each with optional
685 optional alignment, an opening curly brace, a list of basic blocks, and a
686 closing curly brace.
688 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
693 name, a possibly empty list of arguments, and an optional alignment.</p>
695 <p>A function definition contains a list of basic blocks, forming the CFG for
696 the function. Each basic block may optionally start with a label (giving the
697 basic block a symbol table entry), contains a list of instructions, and ends
699 function return).</p>
701 <p>The first basic block in a program is special in two ways: it is immediately
702 executed on entrance to the function, and it is not allowed to have predecessor
703 basic blocks (i.e. there can not be any branches to the entry block of a
704 function). Because the block can have no predecessors, it also cannot have any
707 <p>LLVM functions are identified by their name and type signature. Hence, two
708 functions with the same name but different parameter lists or return values are
709 considered different functions, and LLVM will resolve references to each
710 appropriately.</p>
712 <p>LLVM allows an explicit section to be specified for functions. If the target
713 supports it, it will emit functions to the section specified.</p>
715 <p>An explicit alignment may be specified for a function. If not present, or if
716 the alignment is set to zero, the alignment of the function is set by the target
717 to whatever it feels convenient. If an explicit alignment is specified, the
718 function is forced to have at least that much alignment. All alignments must be
721 </div>
724 <!-- ======================================================================= -->
727 </div>
730 function or global variable or bitcast of global value). Aliases may have an
737 <pre>
739 </pre>
740 </div>
742 </div>
746 <!-- ======================================================================= -->
749 <p>The return type and each parameter of a function type may have a set of
751 used to communicate additional information about the result or parameters of
752 a function. Parameter attributes are considered to be part of the function
753 type so two functions types that differ only by the parameter attributes
754 are different function types.</p>
756 <p>Parameter attributes are simple keywords that follow the type specified. If
757 multiple parameter attributes are needed, they are space separated. For
758 example:</p>
761 <pre>
762 %someFunc = i16 (i8 sext %someParam) zext
763 %someFunc = i16 (i8 zext %someParam) zext
764 </pre>
765 </div>
767 <p>Note that the two function types above are unique because the parameter has
768 a different attribute (sext in the first one, zext in the second). Also note
769 that the attribute for the function result (zext) comes immediately after the
770 argument list.</p>
773 <dl>
775 <dd>This indicates that the parameter should be zero extended just before
776 a call to this function.</dd>
778 <dd>This indicates that the parameter should be sign extended just before
779 a call to this function.</dd>
781 <dd>This indicates that the parameter should be placed in register (if
782 possible) during assembling function call. Support for this attribute is
783 target-specific</dd>
785 <dd>This indicates that the parameter specifies the address of a structure
786 that is the return value of the function in the source program.</dd>
788 <dd>This indicates that the parameter not alias any other object or any
789 other "noalias" objects during the function call.
791 <dd>This function attribute indicates that the function never returns. This
792 indicates to LLVM that every call to this function should be treated as if
795 <dd>This function attribute indicates that the function type does not use
796 the unwind instruction and does not allow stack unwinding to propagate
797 through it.</dd>
798 </dl>
800 </div>
802 <!-- ======================================================================= -->
805 </div>
808 <p>
809 Modules may contain "module-level inline asm" blocks, which corresponds to the
810 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
811 LLVM and treated as a single unit, but may be separated in the .ll file if
812 desired. The syntax is very simple:
813 </p>
816 <pre>
817 module asm "inline asm code goes here"
818 module asm "more can go here"
819 </pre>
820 </div>
822 <p>The strings can contain any character by escaping non-printable characters.
824 for the number.
825 </p>
827 <p>
828 The inline asm code is simply printed to the machine code .s file when
829 assembly code is generated.
830 </p>
831 </div>
833 <!-- ======================================================================= -->
836 </div>
839 <p>A module may specify a target specific data layout string that specifies how
840 data is to be laid out in memory. The syntax for the data layout is simply:</p>
843 separated by the minus sign character ('-'). Each specification starts with a
844 letter and may include other information after the letter to define some
845 aspect of the data layout. The specifications accepted are as follows: </p>
846 <dl>
848 <dd>Specifies that the target lays out data in big-endian form. That is, the
849 bits with the most significance have the lowest address location.</dd>
851 <dd>Specifies that hte target lays out data in little-endian form. That is,
852 the bits with the least significance have the lowest address location.</dd>
857 too.</dd>
859 <dd>This specifies the alignment for an integer type of a given bit
862 <dd>This specifies the alignment for a vector type of a given bit
865 <dd>This specifies the alignment for a floating point type of a given bit
867 (double).</dd>
869 <dd>This specifies the alignment for an aggregate type of a given bit
871 </dl>
872 <p>When constructing the data layout for a given target, LLVM starts with a
873 default set of specifications which are then (possibly) overriden by the
875 are given in this list:</p>
876 <ul>
890 </ul>
891 <p>When llvm is determining the alignment for a given type, it uses the
892 following rules:
893 <ol>
894 <li>If the type sought is an exact match for one of the specifications, that
895 specification is used.</li>
896 <li>If no match is found, and the type sought is an integer type, then the
897 smallest integer type that is larger than the bitwidth of the sought type is
898 used. If none of the specifications are larger than the bitwidth then the the
899 largest integer type is used. For example, given the default specifications
900 above, the i7 type will use the alignment of i8 (next largest) while both
901 i65 and i256 will use the alignment of i64 (largest specified).</li>
902 <li>If no match is found, and the type sought is a vector type, then the
903 largest vector type that is smaller than the sought vector type will be used
906 </ol>
907 </div>
909 <!-- *********************************************************************** -->
911 <!-- *********************************************************************** -->
915 <p>The LLVM type system is one of the most important features of the
916 intermediate representation. Being typed enables a number of
917 optimizations to be performed on the IR directly, without having to do
918 extra analyses on the side before the transformation. A strong type
919 system makes it easier to read the generated code and enables novel
920 analyses and transformations that are not feasible to perform on normal
921 three address code representations.</p>
923 </div>
925 <!-- ======================================================================= -->
928 <p>The primitive types are the fundamental building blocks of the LLVM
929 system. The current set of primitive types is as follows:</p>
934 <table>
935 <tbody>
939 </tbody>
940 </table>
941 </td>
943 <table>
944 <tbody>
948 </tbody>
949 </table>
950 </td>
951 </tr>
952 </table>
953 </div>
955 <!-- _______________________________________________________________________ -->
957 Classifications</a> </div>
959 <p>These different primitive types fall into a few useful
960 classifications:</p>
963 <tbody>
965 <tr>
968 </tr>
969 <tr>
972 </tr>
973 <tr>
977 </td>
978 </tr>
979 </tbody>
980 </table>
983 most important. Values of these types are the only ones which can be
984 produced by instructions, passed as arguments, or used as operands to
985 instructions. This means that all structures and arrays must be
986 manipulated either by pointer or by component.</p>
987 </div>
989 <!-- ======================================================================= -->
994 <p>The real power in LLVM comes from the derived types in the system.
995 This is what allows a programmer to represent arrays, functions,
996 pointers, and other useful types. Note that these derived types may be
997 recursive: For example, it is possible to have a two dimensional array.</p>
999 </div>
1001 <!-- _______________________________________________________________________ -->
1007 <p>The integer type is a very simple derived type that simply specifies an
1008 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1013 <pre>
1014 iN
1015 </pre>
1018 value.</p>
1032 </td>
1039 An integer whose bit width is the answer. <br/>
1042 </td>
1043 </tr>
1044 </table>
1045 </div>
1047 <!-- _______________________________________________________________________ -->
1054 <p>The array type is a very simple derived type that arranges elements
1055 sequentially in memory. The array type requires a size (number of
1056 elements) and an underlying data type.</p>
1060 <pre>
1062 </pre>
1064 <p>The number of elements is a constant integer value; elementtype may
1065 be any type with a size.</p>
1074 </td>
1079 </td>
1080 </tr>
1081 </table>
1089 </td>
1094 </td>
1095 </tr>
1096 </table>
1098 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1099 length array. Normally, accesses past the end of an array are undefined in
1101 As a special case, however, zero length arrays are recognized to be variable
1102 length. This allows implementation of 'pascal style arrays' with the LLVM
1105 </div>
1107 <!-- _______________________________________________________________________ -->
1111 <p>The function type can be thought of as a function signature. It
1112 consists of a return type and a list of formal parameter types.
1113 Function types are usually used to build virtual function tables
1114 (which are structures of pointers to functions), for indirect function
1115 calls, and when defining a function.</p>
1116 <p>
1117 The return type of a function type cannot be an aggregate type.
1118 </p>
1123 which indicates that the function takes a variable number of arguments.
1124 Variable argument functions can access their arguments with the <a
1131 </td>
1134 </tt></td>
1139 </td>
1145 LLVM.
1146 </td>
1147 </tr>
1148 </table>
1150 </div>
1151 <!-- _______________________________________________________________________ -->
1155 <p>The structure type is used to represent a collection of data members
1156 together in memory. The packing of the field types is defined to match
1157 the ABI of the underlying processor. The elements of a structure may
1158 be any type that has a size.</p>
1162 instruction.</p>
1176 </tr>
1177 </table>
1178 </div>
1180 <!-- _______________________________________________________________________ -->
1182 </div>
1185 <p>The packed structure type is used to represent a collection of data members
1186 together in memory. There is no padding between fields. Further, the alignment
1187 of a packed structure is 1 byte. The elements of a packed structure may
1188 be any type that has a size.</p>
1192 instruction.</p>
1201 <td class="left"><tt>< { float, i32 (i32) * } ></tt></td>
1206 </tr>
1207 </table>
1208 </div>
1210 <!-- _______________________________________________________________________ -->
1214 <p>As in many languages, the pointer type represents a pointer or
1215 reference to another object, which must live in memory.</p>
1224 </td>
1231 </td>
1232 </tr>
1233 </table>
1234 </div>
1236 <!-- _______________________________________________________________________ -->
1242 <p>A vector type is a simple derived type that represents a vector
1243 of elements. Vector types are used when multiple primitive data
1244 are operated in parallel using a single instruction (SIMD).
1245 A vector type requires a size (number of
1246 elements) and an underlying primitive data type. Vectors must have a power
1252 <pre>
1254 </pre>
1256 <p>The number of elements is a constant integer value; elementtype may
1257 be any integer or floating point type.</p>
1267 </td>
1272 </td>
1273 </tr>
1274 </table>
1275 </div>
1277 <!-- _______________________________________________________________________ -->
1283 <p>Opaque types are used to represent unknown types in the system. This
1284 corresponds (for example) to the C notion of a foward declared structure type.
1285 In LLVM, opaque types can eventually be resolved to any type (not just a
1286 structure type).</p>
1290 <pre>
1291 opaque
1292 </pre>
1300 </td>
1302 An opaque type.<br/>
1303 </td>
1304 </tr>
1305 </table>
1306 </div>
1309 <!-- *********************************************************************** -->
1311 <!-- *********************************************************************** -->
1315 <p>LLVM has several different basic types of constants. This section describes
1316 them all and their syntax.</p>
1318 </div>
1320 <!-- ======================================================================= -->
1325 <dl>
1330 </dd>
1336 integer types.
1337 </dd>
1342 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1343 notation (see below). Floating point constants must have a <a
1351 </dl>
1353 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1354 of floating point constants. For example, the form '<tt>double
1357 (and the only time that they are generated by the disassembler) is when a
1358 floating point constant must be emitted but it cannot be represented as a
1359 decimal floating point number. For example, NaN's, infinities, and other
1360 special values are represented in their IEEE hexadecimal format so that
1361 assembly and disassembly do not cause any bits to change in the constants.</p>
1363 </div>
1365 <!-- ======================================================================= -->
1367 </div>
1370 <p>Aggregate constants arise from aggregation of simple constants
1371 and smaller aggregate constants.</p>
1373 <dl>
1376 <dd>Structure constants are represented with notation similar to structure
1377 type definitions (a comma separated list of elements, surrounded by braces
1381 types of elements must match those specified by the type.
1382 </dd>
1386 <dd>Array constants are represented with notation similar to array type
1387 definitions (a comma separated list of elements, surrounded by square brackets
1390 types of elements must match those specified by the type.
1391 </dd>
1395 <dd>Vector constants are represented with notation similar to vector type
1396 definitions (a comma separated list of elements, surrounded by
1400 match those specified by the type.
1401 </dd>
1407 This is often used to avoid having to print large zero initializers (e.g. for
1408 large arrays) and is always exactly equivalent to using explicit zero
1409 initializers.
1410 </dd>
1411 </dl>
1413 </div>
1415 <!-- ======================================================================= -->
1418 </div>
1424 constants. These constants are explicitly referenced when the <a
1427 file:</p>
1430 <pre>
1431 @X = global i32 17
1432 @Y = global i32 42
1433 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1434 </pre>
1435 </div>
1437 </div>
1439 <!-- ======================================================================= -->
1443 no specific value. Undefined values may be of any type and be used anywhere
1444 a constant is permitted.</p>
1446 <p>Undefined values indicate to the compiler that the program is well defined
1447 no matter what value is used, giving the compiler more freedom to optimize.
1448 </p>
1449 </div>
1451 <!-- ======================================================================= -->
1453 </div>
1457 <p>Constant expressions are used to allow expressions involving other constants
1458 to be used as constants. Constant expressions may be of any <a
1460 that does not have side effects (e.g. load and call are not supported). The
1461 following is the syntax for constant expressions:</p>
1463 <dl>
1465 <dd>Truncate a constant to another type. The bit size of CST must be larger
1466 than the bit size of TYPE. Both types must be integers.</dd>
1469 <dd>Zero extend a constant to another type. The bit size of CST must be
1470 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1473 <dd>Sign extend a constant to another type. The bit size of CST must be
1474 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1477 <dd>Truncate a floating point constant to another floating point type. The
1478 size of CST must be larger than the size of TYPE. Both types must be
1479 floating point.</dd>
1482 <dd>Floating point extend a constant to another type. The size of CST must be
1483 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1486 <dd>Convert a floating point constant to the corresponding unsigned integer
1487 constant. TYPE must be an integer type. CST must be floating point. If the
1488 value won't fit in the integer type, the results are undefined.</dd>
1491 <dd>Convert a floating point constant to the corresponding signed integer
1492 constant. TYPE must be an integer type. CST must be floating point. If the
1493 value won't fit in the integer type, the results are undefined.</dd>
1496 <dd>Convert an unsigned integer constant to the corresponding floating point
1497 constant. TYPE must be floating point. CST must be of integer type. If the
1498 value won't fit in the floating point type, the results are undefined.</dd>
1501 <dd>Convert a signed integer constant to the corresponding floating point
1502 constant. TYPE must be floating point. CST must be of integer type. If the
1503 value won't fit in the floating point type, the results are undefined.</dd>
1506 <dd>Convert a pointer typed constant to the corresponding integer constant
1507 TYPE must be an integer type. CST must be of pointer type. The CST value is
1508 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1511 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1512 pointer type. CST must be of integer type. The CST value is zero extended,
1513 truncated, or unchanged to make it fit in a pointer size. This one is
1517 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1518 identical (same number of bits). The conversion is done as if the CST value
1519 was stored to memory and read back as TYPE. In other words, no bits change
1520 with this operator, just the type. This can be used for conversion of
1521 vector types to any other type, as long as they have the same bit width. For
1522 pointers it is only valid to cast to another pointer type.
1523 </dd>
1529 instruction, the index list may have zero or more indexes, which are required
1535 constants.</dd>
1546 operation</a> on constants.
1561 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1563 binary</a> operations. The constraints on operands are the same as those for
1564 the corresponding instruction (e.g. no bitwise operations on floating point
1565 values are allowed).</dd>
1566 </dl>
1567 </div>
1569 <!-- *********************************************************************** -->
1571 <!-- *********************************************************************** -->
1573 <!-- ======================================================================= -->
1576 </div>
1580 <p>
1582 Module-Level Inline Assembly</a>) through the use of a special value. This
1583 value represents the inline assembler as a string (containing the instructions
1584 to emit), a list of operand constraints (stored as a string), and a flag that
1585 indicates whether or not the inline asm expression has side effects. An example
1586 inline assembler expression is:
1587 </p>
1590 <pre>
1592 </pre>
1593 </div>
1595 <p>
1598 </p>
1601 <pre>
1603 </pre>
1604 </div>
1606 <p>
1607 Inline asms with side effects not visible in the constraint list must be marked
1608 as having side effects. This is done through the use of the
1610 </p>
1613 <pre>
1615 </pre>
1616 </div>
1618 <p>TODO: The format of the asm and constraints string still need to be
1619 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1620 need to be documented).
1621 </p>
1623 </div>
1625 <!-- *********************************************************************** -->
1627 <!-- *********************************************************************** -->
1631 <p>The LLVM instruction set consists of several different
1638 </div>
1640 <!-- ======================================================================= -->
1642 Instructions</a> </div>
1647 basic block in a program ends with a "Terminator" instruction, which
1648 indicates which block should be executed after the current block is
1650 value: they produce control flow, not values (the one exception being
1659 </div>
1661 <!-- _______________________________________________________________________ -->
1663 Instruction</a> </div>
1668 </pre>
1671 value) from a function back to the caller.</p>
1673 returns a value and then causes control flow, and one that just causes
1674 control flow to occur.</p>
1679 instruction inside of the function that returns a value that does not
1680 match the return type of the function.</p>
1683 returns back to the calling function's context. If the caller is a "<a
1685 the instruction after the call. If the caller was an "<a
1687 at the beginning of the "normal" destination block. If the instruction
1688 returns a value, that value shall set the call or invoke instruction's
1689 return value.</p>
1693 </pre>
1694 </div>
1695 <!-- _______________________________________________________________________ -->
1699 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1700 </pre>
1703 transfer to a different basic block in the current function. There are
1704 two forms of this instruction, corresponding to a conditional branch
1705 and an unconditional branch.</p>
1717 <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
1719 </div>
1720 <!-- _______________________________________________________________________ -->
1723 </div>
1728 <pre>
1729 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1730 </pre>
1736 instruction, allowing a branch to occur to one of many possible
1737 destinations.</p>
1745 table is not allowed to contain duplicate constant entries.</p>
1751 table is searched for the given value. If the value is found, control flow is
1752 transfered to the corresponding destination; otherwise, control flow is
1753 transfered to the default destination.</p>
1757 <p>Depending on properties of the target machine and the particular
1759 ways. For example, it could be generated as a series of chained conditional
1760 branches or with a lookup table.</p>
1764 <pre>
1767 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1770 switch i32 0, label %dest [ ]
1773 switch i32 %val, label %otherwise [ i32 0, label %onzero
1774 i32 1, label %onone
1775 i32 2, label %ontwo ]
1776 </pre>
1777 </div>
1779 <!-- _______________________________________________________________________ -->
1782 </div>
1788 <pre>
1789 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1791 </pre>
1796 function, with the possibility of control flow transfer to either the
1808 <ol>
1809 <li>
1811 convention</a> the call should use. If none is specified, the call defaults
1812 to using C calling conventions.
1813 </li>
1815 function value being invoked. In most cases, this is a direct function
1817 an arbitrary pointer to function value.
1818 </li>
1821 function to be invoked. </li>
1824 signature argument types. If the function signature indicates the function
1825 accepts a variable number of arguments, the extra arguments can be
1826 specified. </li>
1834 </ol>
1840 difference is that it establishes an association with a label, which is used by
1841 the runtime library to unwind the stack.</p>
1843 <p>This instruction is used in languages with destructors to ensure that proper
1845 exception. Additionally, this is important for implementation of
1849 <pre>
1850 %retval = invoke i32 %Test(i32 15) to label %Continue
1854 </pre>
1855 </div>
1858 <!-- _______________________________________________________________________ -->
1861 Instruction</a> </div>
1866 <pre>
1867 unwind
1868 </pre>
1873 at the first callee in the dynamic call stack which used an <a
1875 primarily used to implement exception handling.</p>
1880 immediately halt. The dynamic call stack is then searched for the first <a
1882 execution continues at the "exceptional" destination block specified by the
1884 dynamic call chain, undefined behavior results.</p>
1885 </div>
1887 <!-- _______________________________________________________________________ -->
1890 Instruction</a> </div>
1895 <pre>
1896 unreachable
1897 </pre>
1902 instruction is used to inform the optimizer that a particular portion of the
1903 code is not reachable. This can be used to indicate that the code after a
1904 no-return function cannot be reached, and other facts.</p>
1909 </div>
1913 <!-- ======================================================================= -->
1916 <p>Binary operators are used to do most of the computation in a
1917 program. They require two operands, execute an operation on them, and
1918 produce a single value. The operands might represent
1920 The result value of a binary operator is not
1921 necessarily the same type as its operands.</p>
1923 </div>
1924 <!-- _______________________________________________________________________ -->
1926 Instruction</a> </div>
1930 </pre>
1937 Both arguments must have identical types.</p>
1939 <p>The value produced is the integer or floating point sum of the two
1940 operands.</p>
1943 </pre>
1944 </div>
1945 <!-- _______________________________________________________________________ -->
1947 Instruction</a> </div>
1951 </pre>
1954 operands.</p>
1956 instruction present in most other intermediate representations.</p>
1960 values.
1962 Both arguments must have identical types.</p>
1964 <p>The value produced is the integer or floating point difference of
1965 the two operands.</p>
1967 <pre>
1970 </pre>
1971 </div>
1972 <!-- _______________________________________________________________________ -->
1974 Instruction</a> </div>
1978 </pre>
1981 operands.</p>
1985 values.
1987 Both arguments must have identical types.</p>
1989 <p>The value produced is the integer or floating point product of the
1990 two operands.</p>
1991 <p>Because the operands are the same width, the result of an integer
1992 multiplication is the same whether the operands should be deemed unsigned or
1993 signed.</p>
1996 </pre>
1997 </div>
1998 <!-- _______________________________________________________________________ -->
2000 </a></div>
2004 </pre>
2007 operands.</p>
2012 of the values in which case the elements must be integers.</p>
2014 <p>The value produced is the unsigned integer quotient of the two operands. This
2015 instruction always performs an unsigned division operation, regardless of
2016 whether the arguments are unsigned or not.</p>
2019 </pre>
2020 </div>
2021 <!-- _______________________________________________________________________ -->
2023 </a> </div>
2027 </pre>
2030 operands.</p>
2035 of the values in which case the elements must be integers.</p>
2037 <p>The value produced is the signed integer quotient of the two operands. This
2038 instruction always performs a signed division operation, regardless of whether
2039 the arguments are signed or not.</p>
2042 </pre>
2043 </div>
2044 <!-- _______________________________________________________________________ -->
2046 Instruction</a> </div>
2050 </pre>
2053 operands.</p>
2058 versions of floating point values.</p>
2063 </pre>
2064 </div>
2065 <!-- _______________________________________________________________________ -->
2067 </div>
2071 </pre>
2074 unsigned division of its two arguments.</p>
2078 types.</p>
2081 This instruction always performs an unsigned division to get the remainder,
2082 regardless of whether the arguments are unsigned or not.</p>
2085 </pre>
2087 </div>
2088 <!-- _______________________________________________________________________ -->
2090 Instruction</a> </div>
2094 </pre>
2097 signed division of its two operands.</p>
2101 types.</p>
2106 a value. For more information about the difference, see <a
2108 Math Forum</a>. For a table of how this is implemented in various languages,
2113 </pre>
2115 </div>
2116 <!-- _______________________________________________________________________ -->
2118 Instruction</a> </div>
2122 </pre>
2125 division of its two operands.</p>
2129 identical types.</p>
2134 </pre>
2135 </div>
2137 <!-- ======================================================================= -->
2139 Operations</a> </div>
2141 <p>Bitwise binary operators are used to do various forms of
2142 bit-twiddling in a program. They are generally very efficient
2143 instructions and can commonly be strength reduced from other
2144 instructions. They require two operands, execute an operation on them,
2145 and produce a single value. The resulting value of the bitwise binary
2146 operators is always the same type as its first operand.</p>
2147 </div>
2149 <!-- _______________________________________________________________________ -->
2151 Instruction</a> </div>
2155 </pre>
2158 the left a specified number of bits.</p>
2168 </pre>
2169 </div>
2170 <!-- _______________________________________________________________________ -->
2172 Instruction</a> </div>
2176 </pre>
2180 operand shifted to the right a specified number of bits with zero fill.</p>
2187 <p>This instruction always performs a logical shift right operation. The most
2188 significant bits of the result will be filled with zero bits after the
2189 shift.</p>
2192 <pre>
2197 </pre>
2198 </div>
2200 <!-- _______________________________________________________________________ -->
2202 Instruction</a> </div>
2207 </pre>
2211 operand shifted to the right a specified number of bits with sign extension.</p>
2218 <p>This instruction always performs an arithmetic shift right operation,
2219 The most significant bits of the result will be filled with the sign bit
2223 <pre>
2228 </pre>
2229 </div>
2231 <!-- _______________________________________________________________________ -->
2233 Instruction</a> </div>
2237 </pre>
2240 its two operands.</p>
2244 identical types.</p>
2247 <p> </p>
2250 <tbody>
2251 <tr>
2255 </tr>
2256 <tr>
2260 </tr>
2261 <tr>
2265 </tr>
2266 <tr>
2270 </tr>
2271 <tr>
2275 </tr>
2276 </tbody>
2277 </table>
2278 </div>
2283 </pre>
2284 </div>
2285 <!-- _______________________________________________________________________ -->
2290 </pre>
2293 or of its two operands.</p>
2297 identical types.</p>
2300 <p> </p>
2303 <tbody>
2304 <tr>
2308 </tr>
2309 <tr>
2313 </tr>
2314 <tr>
2318 </tr>
2319 <tr>
2323 </tr>
2324 <tr>
2328 </tr>
2329 </tbody>
2330 </table>
2331 </div>
2336 </pre>
2337 </div>
2338 <!-- _______________________________________________________________________ -->
2340 Instruction</a> </div>
2344 </pre>
2352 identical types.</p>
2355 <p> </p>
2358 <tbody>
2359 <tr>
2363 </tr>
2364 <tr>
2368 </tr>
2369 <tr>
2373 </tr>
2374 <tr>
2378 </tr>
2379 <tr>
2383 </tr>
2384 </tbody>
2385 </table>
2386 </div>
2387 <p> </p>
2393 </pre>
2394 </div>
2396 <!-- ======================================================================= -->
2399 </div>
2403 <p>LLVM supports several instructions to represent vector operations in a
2404 target-independent manner. These instructions cover the element-access and
2405 vector-specific operations needed to process vectors effectively. While LLVM
2406 does directly support these vector operations, many sophisticated algorithms
2407 will want to use target-specific intrinsics to take full advantage of a specific
2408 target.</p>
2410 </div>
2412 <!-- _______________________________________________________________________ -->
2415 </div>
2421 <pre>
2422 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2423 </pre>
2427 <p>
2429 element from a vector at a specified index.
2430 </p>
2435 <p>
2438 an index indicating the position from which to extract the element.
2439 The index may be a variable.</p>
2443 <p>
2444 The result is a scalar of the same type as the element type of
2447 results are undefined.
2448 </p>
2452 <pre>
2454 </pre>
2455 </div>
2458 <!-- _______________________________________________________________________ -->
2461 </div>
2467 <pre>
2468 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2469 </pre>
2473 <p>
2475 element into a vector at a specified index.
2476 </p>
2481 <p>
2484 scalar value whose type must equal the element type of the first
2485 operand. The third operand is an index indicating the position at
2486 which to insert the value. The index may be a variable.</p>
2490 <p>
2495 </p>
2499 <pre>
2501 </pre>
2502 </div>
2504 <!-- _______________________________________________________________________ -->
2507 </div>
2513 <pre>
2514 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2515 </pre>
2519 <p>
2521 from two input vectors, returning a vector of the same type.
2522 </p>
2526 <p>
2528 with types that match each other and types that match the result of the
2529 instruction. The third argument is a shuffle mask, which has the same number
2530 of elements as the other vector type, but whose element type is always 'i32'.
2531 </p>
2533 <p>
2534 The shuffle mask operand is required to be a constant vector with either
2535 constant integer or undef values.
2536 </p>
2540 <p>
2541 The elements of the two input vectors are numbered from left to right across
2542 both of the vectors. The shuffle mask operand specifies, for each element of
2543 the result vector, which element of the two input registers the result element
2544 gets. The element selector may be undef (meaning "don't care") and the second
2545 operand may be undef if performing a shuffle from only one vector.
2546 </p>
2550 <pre>
2554 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2555 </pre>
2556 </div>
2559 <!-- ======================================================================= -->
2562 </div>
2566 <p>A key design point of an SSA-based representation is how it
2567 represents memory. In LLVM, no memory locations are in SSA form, which
2568 makes things very simple. This section describes how to read, write,
2569 allocate, and free memory in LLVM.</p>
2571 </div>
2573 <!-- _______________________________________________________________________ -->
2576 </div>
2582 <pre>
2583 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2584 </pre>
2589 heap and returns a pointer to it.</p>
2595 bytes of memory from the operating system and returns a pointer of the
2596 appropriate type to the program. If "NumElements" is specified, it is the
2597 number of elements allocated. If an alignment is specified, the value result
2598 of the allocation is guaranteed to be aligned to at least that boundary. If
2599 not specified, or if zero, the target can choose to align the allocation on any
2600 convenient boundary.</p>
2607 a pointer is returned.</p>
2611 <pre>
2619 </pre>
2620 </div>
2622 <!-- _______________________________________________________________________ -->
2625 </div>
2631 <pre>
2633 </pre>
2638 memory heap to be reallocated in the future.</p>
2644 instruction.</p>
2648 <p>Access to the memory pointed to by the pointer is no longer defined
2649 after this instruction executes.</p>
2653 <pre>
2655 free [4 x i8]* %array
2656 </pre>
2657 </div>
2659 <!-- _______________________________________________________________________ -->
2662 </div>
2668 <pre>
2669 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2670 </pre>
2675 currently executing function, to be automatically released when this function
2676 returns to its caller.</p>
2681 bytes of memory on the runtime stack, returning a pointer of the
2682 appropriate type to the program. If "NumElements" is specified, it is the
2683 number of elements allocated. If an alignment is specified, the value result
2684 of the allocation is guaranteed to be aligned to at least that boundary. If
2685 not specified, or if zero, the target can choose to align the allocation on any
2686 convenient boundary.</p>
2694 instruction is commonly used to represent automatic variables that must
2695 have an address available. When the function returns (either with the <tt><a
2697 instructions), the memory is reclaimed.</p>
2701 <pre>
2706 </pre>
2707 </div>
2709 <!-- _______________________________________________________________________ -->
2711 Instruction</a> </div>
2714 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
2719 address from which to load. The pointer must point to a <a
2724 instructions. </p>
2729 <a
2732 </pre>
2733 </div>
2734 <!-- _______________________________________________________________________ -->
2736 Instruction</a> </div>
2739 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2740 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2741 </pre>
2749 optimizer is not allowed to modify the number or order of execution of
2757 <a
2760 </pre>
2761 </div>
2763 <!-- _______________________________________________________________________ -->
2766 </div>
2770 <pre>
2772 </pre>
2776 <p>
2778 subelement of an aggregate data structure.</p>
2782 <p>This instruction takes a list of integer operands that indicate what
2783 elements of the aggregate object to index to. The actual types of the arguments
2784 provided depend on the type of the first pointer argument. The
2786 levels of a structure or to a specific index in an array. When indexing into a
2791 <p>For example, let's consider a C code fragment and how it gets
2792 compiled to LLVM:</p>
2795 <pre>
2796 struct RT {
2797 char A;
2799 char C;
2800 };
2801 struct ST {
2802 int X;
2803 double Y;
2804 struct RT Z;
2805 };
2807 int *foo(struct ST *s) {
2809 }
2810 </pre>
2811 </div>
2816 <pre>
2818 %ST = type { i32, double, %RT }
2820 define i32* %foo(%ST* %s) {
2821 entry:
2823 ret i32* %reg
2824 }
2825 </pre>
2826 </div>
2839 }</tt>' type, a structure. The second index indexes into the third element of
2841 i8 }</tt>' type, another structure. The third index indexes into the second
2843 array. The two dimensions of the array are subscripted into, yielding an
2847 <p>Note that it is perfectly legal to index partially through a
2848 structure, returning a pointer to an inner element. Because of this,
2849 the LLVM code for the given testcase is equivalent to:</p>
2851 <pre>
2852 define i32* %foo(%ST* %s) {
2858 ret i32* %t5
2859 }
2860 </pre>
2862 <p>Note that it is undefined to access an array out of bounds: array and
2863 pointer indexes must always be within the defined bounds of the array type.
2864 The one exception for this rules is zero length arrays. These arrays are
2865 defined to be accessible as variable length arrays, which requires access
2866 beyond the zero'th element.</p>
2868 <p>The getelementptr instruction is often confusing. For some more insight
2874 <pre>
2877 </pre>
2878 </div>
2880 <!-- ======================================================================= -->
2882 </div>
2884 <p>The instructions in this category are the conversion instructions (casting)
2885 which all take a single operand and a type. They perform various bit conversions
2886 on the operand.</p>
2887 </div>
2889 <!-- _______________________________________________________________________ -->
2892 </div>
2896 <pre>
2898 </pre>
2901 <p>
2903 </p>
2906 <p>
2914 <p>
2918 It will always truncate bits.</p>
2921 <pre>
2925 </pre>
2926 </div>
2928 <!-- _______________________________________________________________________ -->
2931 </div>
2935 <pre>
2937 </pre>
2958 <pre>
2961 </pre>
2962 </div>
2964 <!-- _______________________________________________________________________ -->
2967 </div>
2971 <pre>
2973 </pre>
2979 <p>
2987 <p>
2995 <pre>
2998 </pre>
2999 </div>
3001 <!-- _______________________________________________________________________ -->
3004 </div>
3010 <pre>
3012 </pre>
3033 <pre>
3036 </pre>
3037 </div>
3039 <!-- _______________________________________________________________________ -->
3042 </div>
3046 <pre>
3048 </pre>
3052 floating point value.</p>
3058 type must be smaller than the destination type.</p>
3068 <pre>
3071 </pre>
3072 </div>
3074 <!-- _______________________________________________________________________ -->
3077 </div>
3081 <pre>
3083 </pre>
3088 </p>
3099 the results are undefined.</p>
3101 <p>When converting to i1, the conversion is done as a comparison against
3106 <pre>
3110 </pre>
3111 </div>
3113 <!-- _______________________________________________________________________ -->
3116 </div>
3120 <pre>
3122 </pre>
3127 </p>
3139 the results are undefined.</p>
3141 <p>When converting to i1, the conversion is done as a comparison against
3146 <pre>
3150 </pre>
3151 </div>
3153 <!-- _______________________________________________________________________ -->
3156 </div>
3160 <pre>
3162 </pre>
3176 integer quantity and converts it to the corresponding floating point value. If
3177 the value cannot fit in the floating point value, the results are undefined.</p>
3181 <pre>
3184 </pre>
3185 </div>
3187 <!-- _______________________________________________________________________ -->
3190 </div>
3194 <pre>
3196 </pre>
3209 integer quantity and converts it to the corresponding floating point value. If
3210 the value cannot fit in the floating point value, the results are undefined.</p>
3213 <pre>
3216 </pre>
3217 </div>
3219 <!-- _______________________________________________________________________ -->
3222 </div>
3226 <pre>
3228 </pre>
3242 truncating or zero extending that value to the size of the integer type. If
3246 change.</p>
3249 <pre>
3252 </pre>
3253 </div>
3255 <!-- _______________________________________________________________________ -->
3258 </div>
3262 <pre>
3264 </pre>
3272 value to cast, and a type to cast it to, which must be a
3280 the size of a pointer then a zero extension is done. If they are the same size,
3284 <pre>
3288 </pre>
3289 </div>
3291 <!-- _______________________________________________________________________ -->
3294 </div>
3298 <pre>
3300 </pre>
3308 a first class value, and a type to cast it to, which must also be a <a
3311 type is a pointer, the destination type must also be a pointer.</p>
3318 converted to other pointer types with this instruction. To convert pointers to
3323 <pre>
3327 </pre>
3328 </div>
3330 <!-- ======================================================================= -->
3334 instructions, which defy better classification.</p>
3335 </div>
3337 <!-- _______________________________________________________________________ -->
3339 </div>
3342 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3343 </pre>
3346 of its two integer operands.</p>
3349 the condition code indicating the kind of comparison to perform. It is not
3350 a value, just a keyword. The possible condition code are:
3351 <ol>
3362 </ol>
3369 <ol>
3372 </li>
3391 </ol>
3393 values are compared as if they were integers.</p>
3402 </pre>
3403 </div>
3405 <!-- _______________________________________________________________________ -->
3407 </div>
3410 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3411 </pre>
3414 of its floating point operands.</p>
3417 the condition code indicating the kind of comparison to perform. It is not
3418 a value, just a keyword. The possible condition code are:
3419 <ol>
3436 </ol>
3441 types.</p>
3446 <ol>
3475 </ol>
3482 </pre>
3483 </div>
3485 <!-- _______________________________________________________________________ -->
3487 Instruction</a> </div>
3493 the SSA graph representing the function.</p>
3495 <p>The type of the incoming values is specified with the first type
3497 as arguments, with one pair for each predecessor basic block of the
3499 type may be used as the value arguments to the PHI node. Only labels
3500 may be used as the label arguments.</p>
3501 <p>There must be no non-phi instructions between the start of a basic
3502 block and the PHI instructions: i.e. PHI instructions must be first in
3503 a basic block.</p>
3506 specified by the pair corresponding to the predecessor basic block that executed
3507 just prior to the current block.</p>
3509 <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>
3510 </div>
3512 <!-- _______________________________________________________________________ -->
3515 </div>
3521 <pre>
3522 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3523 </pre>
3527 <p>
3529 condition, without branching.
3530 </p>
3535 <p>
3536 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.
3537 </p>
3541 <p>
3542 If the boolean condition evaluates to true, the instruction returns the first
3543 value argument; otherwise, it returns the second value argument.
3544 </p>
3548 <pre>
3550 </pre>
3551 </div>
3554 <!-- _______________________________________________________________________ -->
3557 </div>
3562 <pre>
3563 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
3564 </pre>
3574 <ol>
3575 <li>
3577 any allocas or varargs in the caller. If the "tail" marker is present, the
3578 function call is eligible for tail call optimization. Note that calls may
3581 </li>
3582 <li>
3584 convention</a> the call should use. If none is specified, the call defaults
3585 to using C calling conventions.
3586 </li>
3587 <li>
3589 being invoked. The argument types must match the types implied by this
3590 signature. This type can be omitted if the function is not varargs and
3591 if the function type does not return a pointer to a function.</p>
3592 </li>
3593 <li>
3595 be invoked. In most cases, this is a direct function invocation, but
3597 to function value.</p>
3598 </li>
3599 <li>
3601 function signature argument types. All arguments must be of
3603 indicates the function accepts a variable number of arguments, the extra
3604 arguments can be specified.</p>
3605 </li>
3606 </ol>
3611 transfer to a specified function, with its incoming arguments bound to
3613 instruction in the called function, control flow continues with the
3614 instruction after the function call, and the return value of the
3615 function is bound to the result argument. This is a simpler case of
3620 <pre>
3621 %retval = call i32 %test(i32 %argc)
3623 %X = tail call i32 %foo()
3625 </pre>
3627 </div>
3629 <!-- _______________________________________________________________________ -->
3632 </div>
3638 <pre>
3640 </pre>
3645 the "variable argument" area of a function call. It is used to implement the
3651 the argument. It returns a value of the specified argument type and
3663 <p>It is legal for this instruction to be called in a function which does not
3665 function.</p>
3669 argument.</p>
3675 </div>
3677 <!-- *********************************************************************** -->
3679 <!-- *********************************************************************** -->
3684 well known names and semantics and are required to follow certain restrictions.
3685 Overall, these intrinsics represent an extension mechanism for the LLVM
3686 language that does not require changing all of the transformations in LLVM when
3687 adding to the language (or the bytecode reader/writer, the parser, etc...).</p>
3690 prefix is reserved in LLVM for intrinsic names; thus, function names may not
3691 begin with this prefix. Intrinsic functions must always be external functions:
3692 you cannot define the body of intrinsic functions. Intrinsic functions may
3693 only be used in call or invoke instructions: it is illegal to take the address
3694 of an intrinsic function. Additionally, because intrinsic functions are part
3695 of the LLVM language, it is required if any are added that they be documented
3696 here.</p>
3698 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3699 a family of functions that perform the same operation but on different data
3700 types. This is most frequent with the integer types. Since LLVM can represent
3701 over 8 million different integer types, there is a way to declare an intrinsic
3702 that can be overloaded based on its arguments. Such an intrinsic will have the
3703 names of its argument types encoded into its function name, each
3705 integer of any width. This leads to a family of functions such as
3707 </p>
3710 <p>To learn how to add an intrinsic function, please see the
3712 </p>
3714 </div>
3716 <!-- ======================================================================= -->
3719 </div>
3725 intrinsic functions. These functions are related to the similarly
3728 <p>All of these functions operate on arguments that use a
3729 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3730 language reference manual does not define what this type is, so all
3731 transformations should be prepared to handle these functions regardless of
3732 the type used.</p>
3735 instruction and the variable argument handling intrinsic functions are
3736 used.</p>
3739 <pre>
3740 define i32 @test(i32 %X, ...) {
3741 ; Initialize variable argument processing
3742 %ap = alloca i8*
3743 %ap2 = bitcast i8** %ap to i8*
3744 call void @llvm.va_start(i8* %ap2)
3746 ; Read a single integer argument
3747 %tmp = va_arg i8** %ap, i32
3749 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3750 %aq = alloca i8*
3751 %aq2 = bitcast i8** %aq to i8*
3752 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
3753 call void @llvm.va_end(i8* %aq2)
3755 ; Stop processing of arguments.
3756 call void @llvm.va_end(i8* %ap2)
3757 ret i32 %tmp
3758 }
3760 declare void @llvm.va_start(i8*)
3761 declare void @llvm.va_copy(i8*, i8*)
3762 declare void @llvm.va_end(i8*)
3763 </pre>
3764 </div>
3766 </div>
3768 <!-- _______________________________________________________________________ -->
3771 </div>
3789 macro available in C. In a target-dependent way, it initializes the
3793 last argument of the function as the compiler can figure that out.</p>
3795 </div>
3797 <!-- _______________________________________________________________________ -->
3800 </div>
3818 macro available in C. In a target-dependent way, it destroys the
3824 </div>
3826 <!-- _______________________________________________________________________ -->
3829 </div>
3835 <pre>
3837 </pre>
3842 from the source argument list to the destination argument list.</p>
3853 macro available in C. In a target-dependent way, it copies the source
3856 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
3857 example, memory allocation.</p>
3859 </div>
3861 <!-- ======================================================================= -->
3864 </div>
3868 <p>
3870 Collection</a> requires the implementation and generation of these intrinsics.
3874 Front-ends for type-safe garbage collected languages should generate these
3875 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
3877 </p>
3878 </div>
3880 <!-- _______________________________________________________________________ -->
3883 </div>
3889 <pre>
3891 </pre>
3896 the code generator, and allows some metadata to be associated with it.</p>
3900 <p>The first argument specifies the address of a stack object that contains the
3901 root pointer. The second pointer (which must be either a constant or a global
3902 value address) contains the meta-data to be associated with the root.</p>
3907 location. At compile-time, the code generator generates information to allow
3908 the runtime to find the pointer at GC safe points.
3909 </p>
3911 </div>
3914 <!-- _______________________________________________________________________ -->
3917 </div>
3923 <pre>
3924 declare i8 * @llvm.gcread(i8 * %ObjPtr, i8 ** %Ptr)
3925 </pre>
3930 locations, allowing garbage collector implementations that require read
3931 barriers.</p>
3935 <p>The second argument is the address to read from, which should be an address
3936 allocated from the garbage collector. The first object is a pointer to the
3937 start of the referenced object, if needed by the language runtime (otherwise
3938 null).</p>
3943 instruction, but may be replaced with substantially more complex code by the
3944 garbage collector runtime, as needed.</p>
3946 </div>
3949 <!-- _______________________________________________________________________ -->
3952 </div>
3958 <pre>
3959 declare void @llvm.gcwrite(i8 * %P1, i8 * %Obj, i8 ** %P2)
3960 </pre>
3965 locations, allowing garbage collector implementations that require write
3966 barriers (such as generational or reference counting collectors).</p>
3970 <p>The first argument is the reference to store, the second is the start of the
3971 object to store it to, and the third is the address of the field of Obj to
3972 store to. If the runtime does not require a pointer to the object, Obj may be
3973 null.</p>
3978 instruction, but may be replaced with substantially more complex code by the
3979 garbage collector runtime, as needed.</p>
3981 </div>
3985 <!-- ======================================================================= -->
3988 </div>
3991 <p>
3992 These intrinsics are provided by LLVM to expose special features that may only
3993 be implemented with code generator support.
3994 </p>
3996 </div>
3998 <!-- _______________________________________________________________________ -->
4001 </div>
4006 <pre>
4008 </pre>
4012 <p>
4014 target-specific value indicating the return address of the current function
4015 or one of its callers.
4016 </p>
4020 <p>
4021 The argument to this intrinsic indicates which function to return the address
4022 for. Zero indicates the calling function, one indicates its caller, etc. The
4024 </p>
4028 <p>
4030 the return address of the specified call frame, or zero if it cannot be
4031 identified. The value returned by this intrinsic is likely to be incorrect or 0
4032 for arguments other than zero, so it should only be used for debugging purposes.
4033 </p>
4035 <p>
4036 Note that calling this intrinsic does not prevent function inlining or other
4037 aggressive transformations, so the value returned may not be that of the obvious
4038 source-language caller.
4039 </p>
4040 </div>
4043 <!-- _______________________________________________________________________ -->
4046 </div>
4051 <pre>
4053 </pre>
4057 <p>
4059 target-specific frame pointer value for the specified stack frame.
4060 </p>
4064 <p>
4065 The argument to this intrinsic indicates which function to return the frame
4066 pointer for. Zero indicates the calling function, one indicates its caller,
4068 </p>
4072 <p>
4074 the frame address of the specified call frame, or zero if it cannot be
4075 identified. The value returned by this intrinsic is likely to be incorrect or 0
4076 for arguments other than zero, so it should only be used for debugging purposes.
4077 </p>
4079 <p>
4080 Note that calling this intrinsic does not prevent function inlining or other
4081 aggressive transformations, so the value returned may not be that of the obvious
4082 source-language caller.
4083 </p>
4084 </div>
4086 <!-- _______________________________________________________________________ -->
4089 </div>
4094 <pre>
4095 declare i8 *@llvm.stacksave()
4096 </pre>
4100 <p>
4104 features like scoped automatic variable sized arrays in C99.
4105 </p>
4109 <p>
4110 This intrinsic returns a opaque pointer value that can be passed to <a
4117 </p>
4119 </div>
4121 <!-- _______________________________________________________________________ -->
4124 </div>
4129 <pre>
4130 declare void @llvm.stackrestore(i8 * %ptr)
4131 </pre>
4135 <p>
4137 the function stack to the state it was in when the corresponding <a
4139 useful for implementing language features like scoped automatic variable sized
4140 arrays in C99.
4141 </p>
4145 <p>
4147 </p>
4149 </div>
4152 <!-- _______________________________________________________________________ -->
4155 </div>
4160 <pre>
4163 </pre>
4168 <p>
4170 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4171 no
4172 effect on the behavior of the program but can change its performance
4173 characteristics.
4174 </p>
4178 <p>
4184 </p>
4188 <p>
4189 This intrinsic does not modify the behavior of the program. In particular,
4190 prefetches cannot trap and do not produce a value. On targets that support this
4191 intrinsic, the prefetch can provide hints to the processor cache for better
4192 performance.
4193 </p>
4195 </div>
4197 <!-- _______________________________________________________________________ -->
4200 </div>
4205 <pre>
4207 </pre>
4212 <p>
4214 (PC) in a region of
4215 code to simulators and other tools. The method is target specific, but it is
4216 expected that the marker will use exported symbols to transmit the PC of the marker.
4217 The marker makes no guarantees that it will remain with any specific instruction
4218 after optimizations. It is possible that the presence of a marker will inhibit
4219 optimizations. The intended use is to be inserted after optimizations to allow
4220 correlations of simulation runs.
4221 </p>
4225 <p>
4227 </p>
4231 <p>
4232 This intrinsic does not modify the behavior of the program. Backends that do not
4233 support this intrinisic may ignore it.
4234 </p>
4236 </div>
4238 <!-- _______________________________________________________________________ -->
4241 </div>
4246 <pre>
4247 declare i64 @llvm.readcyclecounter( )
4248 </pre>
4253 <p>
4255 counter register (or similar low latency, high accuracy clocks) on those targets
4256 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4257 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4258 should only be used for small timings.
4259 </p>
4263 <p>
4264 When directly supported, reading the cycle counter should not modify any memory.
4265 Implementations are allowed to either return a application specific value or a
4266 system wide value. On backends without support, this is lowered to a constant 0.
4267 </p>
4269 </div>
4271 <!-- ======================================================================= -->
4274 </div>
4277 <p>
4278 LLVM provides intrinsics for a few important standard C library functions.
4279 These intrinsics allow source-language front-ends to pass information about the
4280 alignment of the pointer arguments to the code generator, providing opportunity
4281 for more efficient code generation.
4282 </p>
4284 </div>
4286 <!-- _______________________________________________________________________ -->
4289 </div>
4294 <pre>
4299 </pre>
4303 <p>
4305 location to the destination location.
4306 </p>
4308 <p>
4310 intrinsics do not return a value, and takes an extra alignment argument.
4311 </p>
4315 <p>
4316 The first argument is a pointer to the destination, the second is a pointer to
4317 the source. The third argument is an integer argument
4318 specifying the number of bytes to copy, and the fourth argument is the alignment
4319 of the source and destination locations.
4320 </p>
4322 <p>
4324 the caller guarantees that both the source and destination pointers are aligned
4325 to that boundary.
4326 </p>
4330 <p>
4332 location to the destination location, which are not allowed to overlap. It
4333 copies "len" bytes of memory over. If the argument is known to be aligned to
4334 some boundary, this can be specified as the fourth argument, otherwise it should
4336 </p>
4337 </div>
4340 <!-- _______________________________________________________________________ -->
4343 </div>
4348 <pre>
4353 </pre>
4357 <p>
4359 location to the destination location. It is similar to the
4361 </p>
4363 <p>
4365 intrinsics do not return a value, and takes an extra alignment argument.
4366 </p>
4370 <p>
4371 The first argument is a pointer to the destination, the second is a pointer to
4372 the source. The third argument is an integer argument
4373 specifying the number of bytes to copy, and the fourth argument is the alignment
4374 of the source and destination locations.
4375 </p>
4377 <p>
4379 the caller guarantees that the source and destination pointers are aligned to
4380 that boundary.
4381 </p>
4385 <p>
4387 location to the destination location, which may overlap. It
4388 copies "len" bytes of memory over. If the argument is known to be aligned to
4389 some boundary, this can be specified as the fourth argument, otherwise it should
4391 </p>
4392 </div>
4395 <!-- _______________________________________________________________________ -->
4398 </div>
4403 <pre>
4408 </pre>
4412 <p>
4414 byte value.
4415 </p>
4417 <p>
4419 does not return a value, and takes an extra alignment argument.
4420 </p>
4424 <p>
4425 The first argument is a pointer to the destination to fill, the second is the
4426 byte value to fill it with, the third argument is an integer
4427 argument specifying the number of bytes to fill, and the fourth argument is the
4428 known alignment of destination location.
4429 </p>
4431 <p>
4433 the caller guarantees that the destination pointer is aligned to that boundary.
4434 </p>
4438 <p>
4440 the
4441 destination location. If the argument is known to be aligned to some boundary,
4442 this can be specified as the fourth argument, otherwise it should be set to 0 or
4443 1.
4444 </p>
4445 </div>
4448 <!-- _______________________________________________________________________ -->
4451 </div>
4456 <pre>
4457 declare float @llvm.sqrt.f32(float %Val)
4458 declare double @llvm.sqrt.f64(double %Val)
4459 </pre>
4463 <p>
4467 negative numbers (which allows for better optimization).
4468 </p>
4472 <p>
4473 The argument and return value are floating point numbers of the same type.
4474 </p>
4478 <p>
4479 This function returns the sqrt of the specified operand if it is a positive
4480 floating point number.
4481 </p>
4482 </div>
4484 <!-- _______________________________________________________________________ -->
4487 </div>
4492 <pre>
4493 declare float @llvm.powi.f32(float %Val, i32 %power)
4494 declare double @llvm.powi.f64(double %Val, i32 %power)
4495 </pre>
4499 <p>
4501 specified (positive or negative) power. The order of evaluation of
4502 multiplications is not defined.
4503 </p>
4507 <p>
4508 The second argument is an integer power, and the first is a value to raise to
4509 that power.
4510 </p>
4514 <p>
4515 This function returns the first value raised to the second power with an
4516 unspecified sequence of rounding operations.</p>
4517 </div>
4520 <!-- ======================================================================= -->
4523 </div>
4526 <p>
4527 LLVM provides intrinsics for a few important bit manipulation operations.
4528 These allow efficient code generation for some algorithms.
4529 </p>
4531 </div>
4533 <!-- _______________________________________________________________________ -->
4536 </div>
4541 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4543 that includes the type for the result and the operand.
4544 <pre>
4548 </pre>
4552 <p>
4554 values with an even number of bytes (positive multiple of 16 bits). These are
4555 useful for performing operations on data that is not in the target's native
4556 byte order.
4557 </p>
4561 <p>
4564 intrinsic returns an i32 value that has the four bytes of the input i32
4569 </p>
4571 </div>
4573 <!-- _______________________________________________________________________ -->
4576 </div>
4581 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4582 width. Not all targets support all bit widths however.
4583 <pre>
4589 </pre>
4593 <p>
4595 value.
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 </p>
4610 </div>
4612 <!-- _______________________________________________________________________ -->
4615 </div>
4621 integer bit width. Not all targets support all bit widths however.
4622 <pre>
4628 </pre>
4632 <p>
4634 leading zeros in a variable.
4635 </p>
4639 <p>
4640 The only argument is the value to be counted. The argument may be of any
4641 integer type. The return type must match the argument type.
4642 </p>
4646 <p>
4648 in a variable. If the src == 0 then the result is the size in bits of the type
4650 </p>
4651 </div>
4655 <!-- _______________________________________________________________________ -->
4658 </div>
4664 integer bit width. Not all targets support all bit widths however.
4665 <pre>
4671 </pre>
4675 <p>
4677 trailing zeros.
4678 </p>
4682 <p>
4683 The only argument is the value to be counted. The argument may be of any
4684 integer type. The return type must match the argument type.
4685 </p>
4689 <p>
4691 in a variable. If the src == 0 then the result is the size in bits of the type
4693 </p>
4694 </div>
4696 <!-- _______________________________________________________________________ -->
4699 </div>
4705 on any integer bit width.
4706 <pre>
4707 declare i17 @llvm.part.select.i17.i17 (i17 %val, i32 %loBit, i32 %hiBit)
4708 declare i29 @llvm.part.select.i29.i29 (i29 %val, i32 %loBit, i32 %hiBit)
4709 </pre>
4713 range of bits from an integer value and returns them in the same bit width as
4714 the original value.</p>
4718 any bit width but they must have the same bit width. The second and third
4725 operates in forward mode.</p>
4729 <ol>
4733 to determine the number of bits to retain.</li>
4736 </ol>
4737 <p>In reverse mode, a similar computation is made except that the bits are
4742 </div>
4746 </div>
4752 on any integer bit width.
4753 <pre>
4754 declare i17 @llvm.part.set.i17.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
4755 declare i29 @llvm.part.set.i29.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
4756 </pre>
4760 of bits in an integer value with another integer value. It returns the integer
4761 with the replaced bits.</p>
4768 type since they specify only a bit index.</p>
4774 operates in forward mode.</p>
4776 truncating it down to the size of the replacement area or zero extending it
4777 up to that size.</p>
4782 <p>In reverse mode, a similar computation is made except that the bits are
4786 <pre>
4792 </pre>
4793 </div>
4795 <!-- ======================================================================= -->
4798 </div>
4801 <p>
4803 are described in the <a
4805 Debugging</a> document.
4806 </p>
4807 </div>
4810 <!-- ======================================================================= -->
4813 </div>
4816 <p> The LLVM exception handling intrinsics (which all start with
4820 </div>
4823 <!-- *********************************************************************** -->
4824 <hr>
4825 <address>
4833 Last modified: $Date$
4834 </address>
4835 </body>
4836 </html>