| blob | 5aa5157251e425449e9ce2b163bc7dc1601c8ef2 |
3 <html>
4 <head>
11 </head>
13 <body>
16 <ol>
21 <ol>
32 </ol>
33 </li>
35 <ol>
38 <ol>
42 </ol>
43 </li>
45 <ol>
54 </ol>
55 </li>
56 </ol>
57 </li>
59 <ol>
65 </ol>
66 </li>
68 <ol>
70 </ol>
71 </li>
73 <ol>
75 <ol>
82 </ol>
83 </li>
85 <ol>
95 </ol>
96 </li>
98 <ol>
105 </ol>
106 </li>
108 <ol>
112 </ol>
113 </li>
115 <ol>
122 </ol>
123 </li>
125 <ol>
138 </ol>
140 <ol>
147 </ol>
148 </li>
149 </ol>
150 </li>
152 <ol>
154 <ol>
158 </ol>
159 </li>
161 <ol>
165 </ol>
166 </li>
168 <ol>
176 </ol>
177 </li>
179 <ol>
188 </ol>
189 </li>
191 <ol>
198 </ol>
199 </li>
203 <ol>
205 </ol>
206 </li>
208 <ol>
215 </ol>
216 </li>
217 </ol>
218 </li>
219 </ol>
224 </div>
226 <!-- *********************************************************************** -->
228 <!-- *********************************************************************** -->
231 <p>This document is a reference manual for the LLVM assembly language.
232 LLVM is an SSA based representation that provides type safety,
233 low-level operations, flexibility, and the capability of representing
234 'all' high-level languages cleanly. It is the common code
235 representation used throughout all phases of the LLVM compilation
236 strategy.</p>
237 </div>
239 <!-- *********************************************************************** -->
241 <!-- *********************************************************************** -->
245 <p>The LLVM code representation is designed to be used in three
246 different forms: as an in-memory compiler IR, as an on-disk bitcode
247 representation (suitable for fast loading by a Just-In-Time compiler),
248 and as a human readable assembly language representation. This allows
249 LLVM to provide a powerful intermediate representation for efficient
250 compiler transformations and analysis, while providing a natural means
251 to debug and visualize the transformations. The three different forms
252 of LLVM are all equivalent. This document describes the human readable
253 representation and notation.</p>
255 <p>The LLVM representation aims to be light-weight and low-level
256 while being expressive, typed, and extensible at the same time. It
257 aims to be a "universal IR" of sorts, by being at a low enough level
258 that high-level ideas may be cleanly mapped to it (similar to how
259 microprocessors are "universal IR's", allowing many source languages to
260 be mapped to them). By providing type information, LLVM can be used as
261 the target of optimizations: for example, through pointer analysis, it
262 can be proven that a C automatic variable is never accessed outside of
263 the current function... allowing it to be promoted to a simple SSA
264 value instead of a memory location.</p>
266 </div>
268 <!-- _______________________________________________________________________ -->
273 <p>It is important to note that this document describes 'well formed'
274 LLVM assembly language. There is a difference between what the parser
275 accepts and what is considered 'well formed'. For example, the
276 following instruction is syntactically okay, but not well formed:</p>
279 <pre>
281 </pre>
282 </div>
285 its uses. The LLVM infrastructure provides a verification pass that may
286 be used to verify that an LLVM module is well formed. This pass is
287 automatically run by the parser after parsing input assembly and by
288 the optimizer before it outputs bitcode. The violations pointed out
289 by the verifier pass indicate bugs in transformation passes or input to
290 the parser.</p>
291 </div>
293 <!-- Describe the typesetting conventions here. -->
295 <!-- *********************************************************************** -->
297 <!-- *********************************************************************** -->
301 <p>LLVM identifiers come in two basic types: global and local. Global
302 identifiers (functions, global variables) begin with the @ character. Local
303 identifiers (register names, types) begin with the % character. Additionally,
304 there are three different formats for identifiers, for different purposes:
306 <ol>
307 <li>Named values are represented as a string of characters with their prefix.
308 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
310 Identifiers which require other characters in their names can be surrounded
312 be used in a named value.</li>
314 <li>Unnamed values are represented as an unsigned numeric value with their
319 </ol>
321 <p>LLVM requires that values start with a prefix for two reasons: Compilers
322 don't need to worry about name clashes with reserved words, and the set of
323 reserved words may be expanded in the future without penalty. Additionally,
324 unnamed identifiers allow a compiler to quickly come up with a temporary
325 variable without having to avoid symbol table conflicts.</p>
327 <p>Reserved words in LLVM are very similar to reserved words in other
328 languages. There are keywords for different opcodes
333 and others. These reserved words cannot conflict with variable names, because
334 none of them start with a prefix character ('%' or '@').</p>
336 <p>Here is an example of LLVM code to multiply the integer variable
342 <pre>
344 </pre>
345 </div>
350 <pre>
352 </pre>
353 </div>
358 <pre>
362 </pre>
363 </div>
366 important lexical features of LLVM:</p>
368 <ol>
371 line.</li>
373 <li>Unnamed temporaries are created when the result of a computation is not
374 assigned to a named value.</li>
378 </ol>
380 <p>...and it also shows a convention that we follow in this document. When
381 demonstrating instructions, we will follow an instruction with a comment that
382 defines the type and name of value produced. Comments are shown in italic
383 text.</p>
385 </div>
387 <!-- *********************************************************************** -->
389 <!-- *********************************************************************** -->
391 <!-- ======================================================================= -->
393 </div>
398 translation unit of the input programs. Each module consists of
399 functions, global variables, and symbol table entries. Modules may be
400 combined together with the LLVM linker, which merges function (and
401 global variable) definitions, resolves forward declarations, and merges
407 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
415 %cast210 = <a
419 <a
421 <a
423 </pre>
424 </div>
431 <p>In general, a module is made up of a list of global values,
432 where both functions and global variables are global values. Global values are
433 represented by a pointer to a memory location (in this case, a pointer to an
434 array of char, and a pointer to a function), and have one of the following <a
437 </div>
439 <!-- ======================================================================= -->
442 </div>
446 <p>
447 All Global Variables and Functions have one of the following types of linkage:
448 </p>
450 <dl>
454 <dd>Global values with internal linkage are only directly accessible by
455 objects in the current module. In particular, linking code into a module with
456 an internal global value may cause the internal to be renamed as necessary to
457 avoid collisions. Because the symbol is internal to the module, all
458 references can be updated. This corresponds to the notion of the
460 </dd>
465 the same name when linkage occurs. This is typically used to implement
466 inline functions, templates, or other code which must be generated in each
468 allowed to be discarded.
469 </dd>
475 used for globals that may be emitted in multiple translation units, but that
476 are not guaranteed to be emitted into every translation unit that uses them.
477 One example of this are common globals in C, such as "<tt>int X;</tt>" at
478 global scope.
479 </dd>
484 pointer to array type. When two global variables with appending linkage are
485 linked together, the two global arrays are appended together. This is the
486 LLVM, typesafe, equivalent of having the system linker append together
487 "sections" with identical names when .o files are linked.
488 </dd>
491 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
492 until linked, if not linked, the symbol becomes null instead of being an
493 undefined reference.
494 </dd>
498 <dd>If none of the above identifiers are used, the global is externally
499 visible, meaning that it participates in linkage and can be used to resolve
500 external symbol references.
501 </dd>
502 </dl>
504 <p>
505 The next two types of linkage are targeted for Microsoft Windows platform
506 only. They are designed to support importing (exporting) symbols from (to)
507 DLLs.
508 </p>
510 <dl>
514 or variable via a global pointer to a pointer that is set up by the DLL
515 exporting the symbol. On Microsoft Windows targets, the pointer name is
517 </dd>
522 pointer to a pointer in a DLL, so that it can be referenced with the
525 name.
526 </dd>
528 </dl>
531 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
532 variable and was linked with this one, one of the two would be renamed,
534 external (i.e., lacking any linkage declarations), they are accessible
535 outside of the current module.</p>
540 linkages.
541 </div>
543 <!-- ======================================================================= -->
546 </div>
552 specified for the call. The calling convention of any pair of dynamic
553 caller/callee must match, or the behavior of the program is undefined. The
554 following calling conventions are supported by LLVM, and more may be added in
555 the future:</p>
557 <dl>
560 <dd>This calling convention (the default if no other calling convention is
561 specified) matches the target C calling conventions. This calling convention
562 supports varargs function calls and tolerates some mismatch in the declared
563 prototype and implemented declaration of the function (as does normal C).
564 </dd>
568 <dd>This calling convention attempts to make calls as fast as possible
569 (e.g. by passing things in registers). This calling convention allows the
570 target to use whatever tricks it wants to produce fast code for the target,
571 without having to conform to an externally specified ABI. Implementations of
572 this convention should allow arbitrary tail call optimization to be supported.
573 This calling convention does not support varargs and requires the prototype of
574 all callees to exactly match the prototype of the function definition.
575 </dd>
579 <dd>This calling convention attempts to make code in the caller as efficient
580 as possible under the assumption that the call is not commonly executed. As
581 such, these calls often preserve all registers so that the call does not break
582 any live ranges in the caller side. This calling convention does not support
583 varargs and requires the prototype of all callees to exactly match the
584 prototype of the function definition.
585 </dd>
589 <dd>Any calling convention may be specified by number, allowing
590 target-specific calling conventions to be used. Target specific calling
591 conventions start at 64.
592 </dd>
593 </dl>
595 <p>More calling conventions can be added/defined on an as-needed basis, to
596 support pascal conventions or any other well-known target-independent
597 convention.</p>
599 </div>
601 <!-- ======================================================================= -->
604 </div>
608 <p>
609 All Global Variables and Functions have one of the following visibility styles:
610 </p>
612 <dl>
615 <dd>On ELF, default visibility means that the declaration is visible to other
616 modules and, in shared libraries, means that the declared entity may be
617 overridden. On Darwin, default visibility means that the declaration is
618 visible to other modules. Default visibility corresponds to "external
619 linkage" in the language.
620 </dd>
624 <dd>Two declarations of an object with hidden visibility refer to the same
625 object if they are in the same shared object. Usually, hidden visibility
626 indicates that the symbol will not be placed into the dynamic symbol table,
627 so no other module (executable or shared library) can reference it
628 directly.
629 </dd>
633 <dd>On ELF, protected visibility indicates that the symbol will be placed in
634 the dynamic symbol table, but that references within the defining module will
635 bind to the local symbol. That is, the symbol cannot be overridden by another
636 module.
637 </dd>
638 </dl>
640 </div>
642 <!-- ======================================================================= -->
645 </div>
649 <p>Global variables define regions of memory allocated at compilation time
650 instead of run-time. Global variables may optionally be initialized, may have
651 an explicit section to be placed in, and may have an optional explicit alignment
652 specified. A variable may be defined as "thread_local", which means that it
653 will not be shared by threads (each thread will have a separated copy of the
654 variable). A variable may be defined as a global "constant," which indicates
656 optimization, allowing the global data to be placed in the read-only section of
657 an executable, etc). Note that variables that need runtime initialization
660 <p>
662 constant, even if the final definition of the global is not. This capability
663 can be used to enable slightly better optimization of the program, but requires
664 the language definition to guarantee that optimizations based on the
665 'constantness' are valid for the translation units that do not include the
666 definition.
667 </p>
669 <p>As SSA values, global variables define pointer values that are in
670 scope (i.e. they dominate) all basic blocks in the program. Global
671 variables always define a pointer to their "content" type because they
672 describe a region of memory, and all memory objects in LLVM are
673 accessed through pointers.</p>
675 <p>A global variable may be declared to reside in a target-specifc numbered
676 address space. For targets that support them, address spaces may affect how
677 optimizations are performed and/or what target instructions are used to access
678 the variable. The default address space is zero. The address space qualifier
679 must precede any other attributes.</p>
681 <p>LLVM allows an explicit section to be specified for globals. If the target
682 supports it, it will emit globals to the section specified.</p>
684 <p>An explicit alignment may be specified for a global. If not present, or if
685 the alignment is set to zero, the alignment of the global is set by the target
686 to whatever it feels convenient. If an explicit alignment is specified, the
687 global is forced to have at least that much alignment. All alignments must be
690 <p>For example, the following defines a global in a numbered address space with
691 an initializer, section, and alignment:</p>
694 <pre>
696 </pre>
697 </div>
699 </div>
702 <!-- ======================================================================= -->
705 </div>
714 name, a (possibly empty) argument list (each with optional
717 opening curly brace, a list of basic blocks, and a closing curly brace.
719 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
724 name, a possibly empty list of arguments, an optional alignment, and an optional
727 <p>A function definition contains a list of basic blocks, forming the CFG for
728 the function. Each basic block may optionally start with a label (giving the
729 basic block a symbol table entry), contains a list of instructions, and ends
731 function return).</p>
733 <p>The first basic block in a function is special in two ways: it is immediately
734 executed on entrance to the function, and it is not allowed to have predecessor
735 basic blocks (i.e. there can not be any branches to the entry block of a
736 function). Because the block can have no predecessors, it also cannot have any
739 <p>LLVM allows an explicit section to be specified for functions. If the target
740 supports it, it will emit functions to the section specified.</p>
742 <p>An explicit alignment may be specified for a function. If not present, or if
743 the alignment is set to zero, the alignment of the function is set by the target
744 to whatever it feels convenient. If an explicit alignment is specified, the
745 function is forced to have at least that much alignment. All alignments must be
748 </div>
751 <!-- ======================================================================= -->
754 </div>
757 function or global variable or bitcast of global value). Aliases may have an
764 <pre>
766 </pre>
767 </div>
769 </div>
773 <!-- ======================================================================= -->
776 <p>The return type and each parameter of a function type may have a set of
778 used to communicate additional information about the result or parameters of
779 a function. Parameter attributes are considered to be part of the function,
780 not of the function type, so functions with different parameter attributes
781 can have the same function type.</p>
783 <p>Parameter attributes are simple keywords that follow the type specified. If
784 multiple parameter attributes are needed, they are space separated. For
785 example:</p>
788 <pre>
789 declare i32 @printf(i8* noalias , ...) nounwind
790 declare i32 @atoi(i8*) nounwind readonly
791 </pre>
792 </div>
798 <dl>
800 <dd>This indicates that the parameter should be zero extended just before
801 a call to this function.</dd>
804 <dd>This indicates that the parameter should be sign extended just before
805 a call to this function.</dd>
808 <dd>This indicates that the parameter should be placed in register (if
809 possible) during assembling function call. Support for this attribute is
810 target-specific</dd>
813 <dd>This indicates that the pointer parameter should really be passed by
814 value to the function. The attribute implies that a hidden copy of the
815 pointee is made between the caller and the callee, so the callee is unable
816 to modify the value in the callee. This attribute is only valid on llvm
817 pointer arguments. It is generally used to pass structs and arrays by
818 value, but is also valid on scalars (even though this is silly).</dd>
821 <dd>This indicates that the parameter specifies the address of a structure
822 that is the return value of the function in the source program.</dd>
825 <dd>This indicates that the parameter not alias any other object or any
826 other "noalias" objects during the function call.
829 <dd>This function attribute indicates that the function never returns. This
830 indicates to LLVM that every call to this function should be treated as if
834 <dd>This function attribute indicates that the function type does not use
835 the unwind instruction and does not allow stack unwinding to propagate
836 through it.</dd>
839 <dd>This indicates that the parameter can be excised using the
842 <dd>This function attribute indicates that the function has no side-effects
843 except for producing a return value or throwing an exception. The value
844 returned must only depend on the function arguments and/or global variables.
845 It may use values obtained by dereferencing pointers.</dd>
848 function, but in addition it is not allowed to dereference any pointer arguments
849 or global variables.
850 </dl>
852 </div>
854 <!-- ======================================================================= -->
857 </div>
860 <p>Each function may specify a garbage collector name, which is simply a
861 string.</p>
867 collector which will cause the compiler to alter its output in order to support
868 the named garbage collection algorithm.</p>
869 </div>
871 <!-- ======================================================================= -->
874 </div>
877 <p>
878 Modules may contain "module-level inline asm" blocks, which corresponds to the
879 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
880 LLVM and treated as a single unit, but may be separated in the .ll file if
881 desired. The syntax is very simple:
882 </p>
885 <pre>
886 module asm "inline asm code goes here"
887 module asm "more can go here"
888 </pre>
889 </div>
891 <p>The strings can contain any character by escaping non-printable characters.
893 for the number.
894 </p>
896 <p>
897 The inline asm code is simply printed to the machine code .s file when
898 assembly code is generated.
899 </p>
900 </div>
902 <!-- ======================================================================= -->
905 </div>
908 <p>A module may specify a target specific data layout string that specifies how
909 data is to be laid out in memory. The syntax for the data layout is simply:</p>
912 separated by the minus sign character ('-'). Each specification starts with a
913 letter and may include other information after the letter to define some
914 aspect of the data layout. The specifications accepted are as follows: </p>
915 <dl>
917 <dd>Specifies that the target lays out data in big-endian form. That is, the
918 bits with the most significance have the lowest address location.</dd>
920 <dd>Specifies that hte target lays out data in little-endian form. That is,
921 the bits with the least significance have the lowest address location.</dd>
926 too.</dd>
928 <dd>This specifies the alignment for an integer type of a given bit
931 <dd>This specifies the alignment for a vector type of a given bit
934 <dd>This specifies the alignment for a floating point type of a given bit
936 (double).</dd>
938 <dd>This specifies the alignment for an aggregate type of a given bit
940 </dl>
941 <p>When constructing the data layout for a given target, LLVM starts with a
942 default set of specifications which are then (possibly) overriden by the
944 are given in this list:</p>
945 <ul>
959 </ul>
960 <p>When llvm is determining the alignment for a given type, it uses the
961 following rules:
962 <ol>
963 <li>If the type sought is an exact match for one of the specifications, that
964 specification is used.</li>
965 <li>If no match is found, and the type sought is an integer type, then the
966 smallest integer type that is larger than the bitwidth of the sought type is
967 used. If none of the specifications are larger than the bitwidth then the the
968 largest integer type is used. For example, given the default specifications
969 above, the i7 type will use the alignment of i8 (next largest) while both
970 i65 and i256 will use the alignment of i64 (largest specified).</li>
971 <li>If no match is found, and the type sought is a vector type, then the
972 largest vector type that is smaller than the sought vector type will be used
975 </ol>
976 </div>
978 <!-- *********************************************************************** -->
980 <!-- *********************************************************************** -->
984 <p>The LLVM type system is one of the most important features of the
985 intermediate representation. Being typed enables a number of
986 optimizations to be performed on the IR directly, without having to do
987 extra analyses on the side before the transformation. A strong type
988 system makes it easier to read the generated code and enables novel
989 analyses and transformations that are not feasible to perform on normal
990 three address code representations.</p>
992 </div>
994 <!-- ======================================================================= -->
996 Classifications</a> </div>
998 <p>The types fall into a few useful
999 classifications:</p>
1002 <tbody>
1004 <tr>
1007 </tr>
1008 <tr>
1011 </tr>
1012 <tr>
1018 </td>
1019 </tr>
1020 <tr>
1026 </tr>
1027 <tr>
1037 </tr>
1038 </tbody>
1039 </table>
1042 most important. Values of these types are the only ones which can be
1043 produced by instructions, passed as arguments, or used as operands to
1044 instructions. This means that all structures and arrays must be
1045 manipulated either by pointer or by component.</p>
1046 </div>
1048 <!-- ======================================================================= -->
1052 <p>The primitive types are the fundamental building blocks of the LLVM
1053 system.</p>
1055 </div>
1057 <!-- _______________________________________________________________________ -->
1061 <table>
1062 <tbody>
1069 </tbody>
1070 </table>
1071 </div>
1073 <!-- _______________________________________________________________________ -->
1082 <pre>
1083 void
1084 </pre>
1085 </div>
1087 <!-- _______________________________________________________________________ -->
1096 <pre>
1097 label
1098 </pre>
1099 </div>
1102 <!-- ======================================================================= -->
1107 <p>The real power in LLVM comes from the derived types in the system.
1108 This is what allows a programmer to represent arrays, functions,
1109 pointers, and other useful types. Note that these derived types may be
1110 recursive: For example, it is possible to have a two dimensional array.</p>
1112 </div>
1114 <!-- _______________________________________________________________________ -->
1120 <p>The integer type is a very simple derived type that simply specifies an
1121 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1126 <pre>
1127 iN
1128 </pre>
1131 value.</p>
1135 <tbody>
1136 <tr>
1139 </tr><tr>
1142 </tr><tr>
1145 </tr>
1146 </tbody>
1147 </table>
1148 </div>
1150 <!-- _______________________________________________________________________ -->
1157 <p>The array type is a very simple derived type that arranges elements
1158 sequentially in memory. The array type requires a size (number of
1159 elements) and an underlying data type.</p>
1163 <pre>
1165 </pre>
1167 <p>The number of elements is a constant integer value; elementtype may
1168 be any type with a size.</p>
1175 </tr>
1179 </tr>
1183 </tr>
1184 </table>
1190 </tr>
1194 </tr>
1198 </tr>
1199 </table>
1201 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1202 length array. Normally, accesses past the end of an array are undefined in
1204 As a special case, however, zero length arrays are recognized to be variable
1205 length. This allows implementation of 'pascal style arrays' with the LLVM
1208 </div>
1210 <!-- _______________________________________________________________________ -->
1214 <p>The function type can be thought of as a function signature. It
1215 consists of a return type and a list of formal parameter types.
1216 Function types are usually used to build virtual function tables
1217 (which are structures of pointers to functions), for indirect function
1218 calls, and when defining a function.</p>
1219 <p>
1220 The return type of a function type cannot be an aggregate type.
1221 </p>
1226 which indicates that the function takes a variable number of arguments.
1227 Variable argument functions can access their arguments with the <a
1234 </td>
1237 </tt></td>
1242 </td>
1248 LLVM.
1249 </td>
1250 </tr>
1251 </table>
1253 </div>
1254 <!-- _______________________________________________________________________ -->
1258 <p>The structure type is used to represent a collection of data members
1259 together in memory. The packing of the field types is defined to match
1260 the ABI of the underlying processor. The elements of a structure may
1261 be any type that has a size.</p>
1265 instruction.</p>
1279 </tr>
1280 </table>
1281 </div>
1283 <!-- _______________________________________________________________________ -->
1285 </div>
1288 <p>The packed structure type is used to represent a collection of data members
1289 together in memory. There is no padding between fields. Further, the alignment
1290 of a packed structure is 1 byte. The elements of a packed structure may
1291 be any type that has a size.</p>
1295 instruction.</p>
1309 </tr>
1310 </table>
1311 </div>
1313 <!-- _______________________________________________________________________ -->
1317 <p>As in many languages, the pointer type represents a pointer or
1318 reference to another object, which must live in memory. Pointer types may have
1319 an optional address space attribute defining the target-specific numbered
1320 address space where the pointed-to object resides. The default address space is
1321 zero.</p>
1330 </tr>
1336 </tr>
1341 </tr>
1342 </table>
1343 </div>
1345 <!-- _______________________________________________________________________ -->
1351 <p>A vector type is a simple derived type that represents a vector
1352 of elements. Vector types are used when multiple primitive data
1353 are operated in parallel using a single instruction (SIMD).
1354 A vector type requires a size (number of
1355 elements) and an underlying primitive data type. Vectors must have a power
1361 <pre>
1363 </pre>
1365 <p>The number of elements is a constant integer value; elementtype may
1366 be any integer or floating point type.</p>
1374 </tr>
1378 </tr>
1382 </tr>
1383 </table>
1384 </div>
1386 <!-- _______________________________________________________________________ -->
1392 <p>Opaque types are used to represent unknown types in the system. This
1393 corresponds (for example) to the C notion of a forward declared structure type.
1394 In LLVM, opaque types can eventually be resolved to any type (not just a
1395 structure type).</p>
1399 <pre>
1400 opaque
1401 </pre>
1409 </tr>
1410 </table>
1411 </div>
1414 <!-- *********************************************************************** -->
1416 <!-- *********************************************************************** -->
1420 <p>LLVM has several different basic types of constants. This section describes
1421 them all and their syntax.</p>
1423 </div>
1425 <!-- ======================================================================= -->
1430 <dl>
1435 </dd>
1441 integer types.
1442 </dd>
1447 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1448 notation (see below). Floating point constants must have a <a
1456 </dl>
1458 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1459 of floating point constants. For example, the form '<tt>double
1462 (and the only time that they are generated by the disassembler) is when a
1463 floating point constant must be emitted but it cannot be represented as a
1464 decimal floating point number. For example, NaN's, infinities, and other
1465 special values are represented in their IEEE hexadecimal format so that
1466 assembly and disassembly do not cause any bits to change in the constants.</p>
1468 </div>
1470 <!-- ======================================================================= -->
1472 </div>
1475 <p>Aggregate constants arise from aggregation of simple constants
1476 and smaller aggregate constants.</p>
1478 <dl>
1481 <dd>Structure constants are represented with notation similar to structure
1482 type definitions (a comma separated list of elements, surrounded by braces
1486 types of elements must match those specified by the type.
1487 </dd>
1491 <dd>Array constants are represented with notation similar to array type
1492 definitions (a comma separated list of elements, surrounded by square brackets
1495 types of elements must match those specified by the type.
1496 </dd>
1500 <dd>Vector constants are represented with notation similar to vector type
1501 definitions (a comma separated list of elements, surrounded by
1505 match those specified by the type.
1506 </dd>
1512 This is often used to avoid having to print large zero initializers (e.g. for
1513 large arrays) and is always exactly equivalent to using explicit zero
1514 initializers.
1515 </dd>
1516 </dl>
1518 </div>
1520 <!-- ======================================================================= -->
1523 </div>
1529 constants. These constants are explicitly referenced when the <a
1532 file:</p>
1535 <pre>
1536 @X = global i32 17
1537 @Y = global i32 42
1538 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1539 </pre>
1540 </div>
1542 </div>
1544 <!-- ======================================================================= -->
1548 no specific value. Undefined values may be of any type and be used anywhere
1549 a constant is permitted.</p>
1551 <p>Undefined values indicate to the compiler that the program is well defined
1552 no matter what value is used, giving the compiler more freedom to optimize.
1553 </p>
1554 </div>
1556 <!-- ======================================================================= -->
1558 </div>
1562 <p>Constant expressions are used to allow expressions involving other constants
1563 to be used as constants. Constant expressions may be of any <a
1565 that does not have side effects (e.g. load and call are not supported). The
1566 following is the syntax for constant expressions:</p>
1568 <dl>
1570 <dd>Truncate a constant to another type. The bit size of CST must be larger
1571 than the bit size of TYPE. Both types must be integers.</dd>
1574 <dd>Zero extend a constant to another type. The bit size of CST must be
1575 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1578 <dd>Sign extend a constant to another type. The bit size of CST must be
1579 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1582 <dd>Truncate a floating point constant to another floating point type. The
1583 size of CST must be larger than the size of TYPE. Both types must be
1584 floating point.</dd>
1587 <dd>Floating point extend a constant to another type. The size of CST must be
1588 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1591 <dd>Convert a floating point constant to the corresponding unsigned integer
1592 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1593 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1594 of the same number of elements. If the value won't fit in the integer type,
1595 the results are undefined.</dd>
1598 <dd>Convert a floating point constant to the corresponding signed integer
1599 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1600 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1601 of the same number of elements. If the value won't fit in the integer type,
1602 the results are undefined.</dd>
1605 <dd>Convert an unsigned integer constant to the corresponding floating point
1606 constant. TYPE must be a scalar or vector floating point type. CST must be of
1607 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1608 of the same number of elements. If the value won't fit in the floating point
1609 type, the results are undefined.</dd>
1612 <dd>Convert a signed integer constant to the corresponding floating point
1613 constant. TYPE must be a scalar or vector floating point type. CST must be of
1614 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1615 of the same number of elements. If the value won't fit in the floating point
1616 type, the results are undefined.</dd>
1619 <dd>Convert a pointer typed constant to the corresponding integer constant
1620 TYPE must be an integer type. CST must be of pointer type. The CST value is
1621 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1624 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1625 pointer type. CST must be of integer type. The CST value is zero extended,
1626 truncated, or unchanged to make it fit in a pointer size. This one is
1630 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1631 identical (same number of bits). The conversion is done as if the CST value
1632 was stored to memory and read back as TYPE. In other words, no bits change
1633 with this operator, just the type. This can be used for conversion of
1634 vector types to any other type, as long as they have the same bit width. For
1635 pointers it is only valid to cast to another pointer type.
1636 </dd>
1642 instruction, the index list may have zero or more indexes, which are required
1648 constants.</dd>
1659 operation</a> on constants.
1674 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1676 binary</a> operations. The constraints on operands are the same as those for
1677 the corresponding instruction (e.g. no bitwise operations on floating point
1678 values are allowed).</dd>
1679 </dl>
1680 </div>
1682 <!-- *********************************************************************** -->
1684 <!-- *********************************************************************** -->
1686 <!-- ======================================================================= -->
1689 </div>
1693 <p>
1695 Module-Level Inline Assembly</a>) through the use of a special value. This
1696 value represents the inline assembler as a string (containing the instructions
1697 to emit), a list of operand constraints (stored as a string), and a flag that
1698 indicates whether or not the inline asm expression has side effects. An example
1699 inline assembler expression is:
1700 </p>
1703 <pre>
1705 </pre>
1706 </div>
1708 <p>
1711 </p>
1714 <pre>
1716 </pre>
1717 </div>
1719 <p>
1720 Inline asms with side effects not visible in the constraint list must be marked
1721 as having side effects. This is done through the use of the
1723 </p>
1726 <pre>
1728 </pre>
1729 </div>
1731 <p>TODO: The format of the asm and constraints string still need to be
1732 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1733 need to be documented).
1734 </p>
1736 </div>
1738 <!-- *********************************************************************** -->
1740 <!-- *********************************************************************** -->
1744 <p>The LLVM instruction set consists of several different
1751 </div>
1753 <!-- ======================================================================= -->
1755 Instructions</a> </div>
1760 basic block in a program ends with a "Terminator" instruction, which
1761 indicates which block should be executed after the current block is
1763 value: they produce control flow, not values (the one exception being
1772 </div>
1774 <!-- _______________________________________________________________________ -->
1776 Instruction</a> </div>
1781 </pre>
1784 value) from a function back to the caller.</p>
1786 returns a value and then causes control flow, and one that just causes
1787 control flow to occur.</p>
1792 instruction inside of the function that returns a value that does not
1793 match the return type of the function.</p>
1796 returns back to the calling function's context. If the caller is a "<a
1798 the instruction after the call. If the caller was an "<a
1800 at the beginning of the "normal" destination block. If the instruction
1801 returns a value, that value shall set the call or invoke instruction's
1802 return value.</p>
1806 </pre>
1807 </div>
1808 <!-- _______________________________________________________________________ -->
1812 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1813 </pre>
1816 transfer to a different basic block in the current function. There are
1817 two forms of this instruction, corresponding to a conditional branch
1818 and an unconditional branch.</p>
1830 <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
1832 </div>
1833 <!-- _______________________________________________________________________ -->
1836 </div>
1841 <pre>
1842 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1843 </pre>
1849 instruction, allowing a branch to occur to one of many possible
1850 destinations.</p>
1858 table is not allowed to contain duplicate constant entries.</p>
1864 table is searched for the given value. If the value is found, control flow is
1865 transfered to the corresponding destination; otherwise, control flow is
1866 transfered to the default destination.</p>
1870 <p>Depending on properties of the target machine and the particular
1872 ways. For example, it could be generated as a series of chained conditional
1873 branches or with a lookup table.</p>
1877 <pre>
1880 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1883 switch i32 0, label %dest [ ]
1886 switch i32 %val, label %otherwise [ i32 0, label %onzero
1887 i32 1, label %onone
1888 i32 2, label %ontwo ]
1889 </pre>
1890 </div>
1892 <!-- _______________________________________________________________________ -->
1895 </div>
1901 <pre>
1902 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1904 </pre>
1909 function, with the possibility of control flow transfer to either the
1921 <ol>
1922 <li>
1924 convention</a> the call should use. If none is specified, the call defaults
1925 to using C calling conventions.
1926 </li>
1928 function value being invoked. In most cases, this is a direct function
1930 an arbitrary pointer to function value.
1931 </li>
1934 function to be invoked. </li>
1937 signature argument types. If the function signature indicates the function
1938 accepts a variable number of arguments, the extra arguments can be
1939 specified. </li>
1947 </ol>
1953 difference is that it establishes an association with a label, which is used by
1954 the runtime library to unwind the stack.</p>
1956 <p>This instruction is used in languages with destructors to ensure that proper
1958 exception. Additionally, this is important for implementation of
1962 <pre>
1963 %retval = invoke i32 %Test(i32 15) to label %Continue
1967 </pre>
1968 </div>
1971 <!-- _______________________________________________________________________ -->
1974 Instruction</a> </div>
1979 <pre>
1980 unwind
1981 </pre>
1986 at the first callee in the dynamic call stack which used an <a
1988 primarily used to implement exception handling.</p>
1993 immediately halt. The dynamic call stack is then searched for the first <a
1995 execution continues at the "exceptional" destination block specified by the
1997 dynamic call chain, undefined behavior results.</p>
1998 </div>
2000 <!-- _______________________________________________________________________ -->
2003 Instruction</a> </div>
2008 <pre>
2009 unreachable
2010 </pre>
2015 instruction is used to inform the optimizer that a particular portion of the
2016 code is not reachable. This can be used to indicate that the code after a
2017 no-return function cannot be reached, and other facts.</p>
2022 </div>
2026 <!-- ======================================================================= -->
2029 <p>Binary operators are used to do most of the computation in a
2030 program. They require two operands, execute an operation on them, and
2031 produce a single value. The operands might represent
2033 The result value of a binary operator is not
2034 necessarily the same type as its operands.</p>
2036 </div>
2037 <!-- _______________________________________________________________________ -->
2039 Instruction</a> </div>
2043 </pre>
2050 Both arguments must have identical types.</p>
2052 <p>The value produced is the integer or floating point sum of the two
2053 operands.</p>
2056 </pre>
2057 </div>
2058 <!-- _______________________________________________________________________ -->
2060 Instruction</a> </div>
2064 </pre>
2067 operands.</p>
2069 instruction present in most other intermediate representations.</p>
2073 values.
2075 Both arguments must have identical types.</p>
2077 <p>The value produced is the integer or floating point difference of
2078 the two operands.</p>
2080 <pre>
2083 </pre>
2084 </div>
2085 <!-- _______________________________________________________________________ -->
2087 Instruction</a> </div>
2091 </pre>
2094 operands.</p>
2098 values.
2100 Both arguments must have identical types.</p>
2102 <p>The value produced is the integer or floating point product of the
2103 two operands.</p>
2104 <p>Because the operands are the same width, the result of an integer
2105 multiplication is the same whether the operands should be deemed unsigned or
2106 signed.</p>
2109 </pre>
2110 </div>
2111 <!-- _______________________________________________________________________ -->
2113 </a></div>
2117 </pre>
2120 operands.</p>
2125 of the values in which case the elements must be integers.</p>
2127 <p>The value produced is the unsigned integer quotient of the two operands. This
2128 instruction always performs an unsigned division operation, regardless of
2129 whether the arguments are unsigned or not.</p>
2132 </pre>
2133 </div>
2134 <!-- _______________________________________________________________________ -->
2136 </a> </div>
2140 </pre>
2143 operands.</p>
2148 of the values in which case the elements must be integers.</p>
2150 <p>The value produced is the signed integer quotient of the two operands. This
2151 instruction always performs a signed division operation, regardless of whether
2152 the arguments are signed or not.</p>
2155 </pre>
2156 </div>
2157 <!-- _______________________________________________________________________ -->
2159 Instruction</a> </div>
2163 </pre>
2166 operands.</p>
2171 versions of floating point values.</p>
2176 </pre>
2177 </div>
2178 <!-- _______________________________________________________________________ -->
2180 </div>
2184 </pre>
2187 unsigned division of its two arguments.</p>
2192 of the values in which case the elements must be integers.</p>
2195 This instruction always performs an unsigned division to get the remainder,
2196 regardless of whether the arguments are unsigned or not.</p>
2199 </pre>
2201 </div>
2202 <!-- _______________________________________________________________________ -->
2204 Instruction</a> </div>
2208 </pre>
2211 signed division of its two operands. This instruction can also take
2213 the elements must be integers.</p>
2218 types.</p>
2223 a value. For more information about the difference, see <a
2225 Math Forum</a>. For a table of how this is implemented in various languages,
2230 </pre>
2232 </div>
2233 <!-- _______________________________________________________________________ -->
2235 Instruction</a> </div>
2239 </pre>
2242 division of its two operands.</p>
2247 versions of floating point values.</p>
2252 </pre>
2253 </div>
2255 <!-- ======================================================================= -->
2257 Operations</a> </div>
2259 <p>Bitwise binary operators are used to do various forms of
2260 bit-twiddling in a program. They are generally very efficient
2261 instructions and can commonly be strength reduced from other
2262 instructions. They require two operands, execute an operation on them,
2263 and produce a single value. The resulting value of the bitwise binary
2264 operators is always the same type as its first operand.</p>
2265 </div>
2267 <!-- _______________________________________________________________________ -->
2269 Instruction</a> </div>
2273 </pre>
2278 the left a specified number of bits.</p>
2296 </pre>
2297 </div>
2298 <!-- _______________________________________________________________________ -->
2300 Instruction</a> </div>
2304 </pre>
2308 operand shifted to the right a specified number of bits with zero fill.</p>
2316 <p>This instruction always performs a logical shift right operation. The most
2317 significant bits of the result will be filled with zero bits after the
2322 <pre>
2328 </pre>
2329 </div>
2331 <!-- _______________________________________________________________________ -->
2333 Instruction</a> </div>
2338 </pre>
2342 operand shifted to the right a specified number of bits with sign extension.</p>
2349 <p>This instruction always performs an arithmetic shift right operation,
2350 The most significant bits of the result will be filled with the sign bit
2353 </p>
2356 <pre>
2362 </pre>
2363 </div>
2365 <!-- _______________________________________________________________________ -->
2367 Instruction</a> </div>
2371 </pre>
2374 its two operands.</p>
2378 identical types.</p>
2381 <p> </p>
2384 <tbody>
2385 <tr>
2389 </tr>
2390 <tr>
2394 </tr>
2395 <tr>
2399 </tr>
2400 <tr>
2404 </tr>
2405 <tr>
2409 </tr>
2410 </tbody>
2411 </table>
2412 </div>
2417 </pre>
2418 </div>
2419 <!-- _______________________________________________________________________ -->
2424 </pre>
2427 or of its two operands.</p>
2431 identical types.</p>
2434 <p> </p>
2437 <tbody>
2438 <tr>
2442 </tr>
2443 <tr>
2447 </tr>
2448 <tr>
2452 </tr>
2453 <tr>
2457 </tr>
2458 <tr>
2462 </tr>
2463 </tbody>
2464 </table>
2465 </div>
2470 </pre>
2471 </div>
2472 <!-- _______________________________________________________________________ -->
2474 Instruction</a> </div>
2478 </pre>
2486 identical types.</p>
2489 <p> </p>
2492 <tbody>
2493 <tr>
2497 </tr>
2498 <tr>
2502 </tr>
2503 <tr>
2507 </tr>
2508 <tr>
2512 </tr>
2513 <tr>
2517 </tr>
2518 </tbody>
2519 </table>
2520 </div>
2521 <p> </p>
2527 </pre>
2528 </div>
2530 <!-- ======================================================================= -->
2533 </div>
2537 <p>LLVM supports several instructions to represent vector operations in a
2538 target-independent manner. These instructions cover the element-access and
2539 vector-specific operations needed to process vectors effectively. While LLVM
2540 does directly support these vector operations, many sophisticated algorithms
2541 will want to use target-specific intrinsics to take full advantage of a specific
2542 target.</p>
2544 </div>
2546 <!-- _______________________________________________________________________ -->
2549 </div>
2555 <pre>
2556 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2557 </pre>
2561 <p>
2563 element from a vector at a specified index.
2564 </p>
2569 <p>
2572 an index indicating the position from which to extract the element.
2573 The index may be a variable.</p>
2577 <p>
2578 The result is a scalar of the same type as the element type of
2581 results are undefined.
2582 </p>
2586 <pre>
2588 </pre>
2589 </div>
2592 <!-- _______________________________________________________________________ -->
2595 </div>
2601 <pre>
2602 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2603 </pre>
2607 <p>
2609 element into a vector at a specified index.
2610 </p>
2615 <p>
2618 scalar value whose type must equal the element type of the first
2619 operand. The third operand is an index indicating the position at
2620 which to insert the value. The index may be a variable.</p>
2624 <p>
2629 </p>
2633 <pre>
2635 </pre>
2636 </div>
2638 <!-- _______________________________________________________________________ -->
2641 </div>
2647 <pre>
2648 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2649 </pre>
2653 <p>
2655 from two input vectors, returning a vector of the same type.
2656 </p>
2660 <p>
2662 with types that match each other and types that match the result of the
2663 instruction. The third argument is a shuffle mask, which has the same number
2664 of elements as the other vector type, but whose element type is always 'i32'.
2665 </p>
2667 <p>
2668 The shuffle mask operand is required to be a constant vector with either
2669 constant integer or undef values.
2670 </p>
2674 <p>
2675 The elements of the two input vectors are numbered from left to right across
2676 both of the vectors. The shuffle mask operand specifies, for each element of
2677 the result vector, which element of the two input registers the result element
2678 gets. The element selector may be undef (meaning "don't care") and the second
2679 operand may be undef if performing a shuffle from only one vector.
2680 </p>
2684 <pre>
2688 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2689 </pre>
2690 </div>
2693 <!-- ======================================================================= -->
2696 </div>
2700 <p>A key design point of an SSA-based representation is how it
2701 represents memory. In LLVM, no memory locations are in SSA form, which
2702 makes things very simple. This section describes how to read, write,
2703 allocate, and free memory in LLVM.</p>
2705 </div>
2707 <!-- _______________________________________________________________________ -->
2710 </div>
2716 <pre>
2717 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2718 </pre>
2723 heap and returns a pointer to it. The object is always allocated in the generic
2724 address space (address space zero).</p>
2730 bytes of memory from the operating system and returns a pointer of the
2731 appropriate type to the program. If "NumElements" is specified, it is the
2732 number of elements allocated. If an alignment is specified, the value result
2733 of the allocation is guaranteed to be aligned to at least that boundary. If
2734 not specified, or if zero, the target can choose to align the allocation on any
2735 convenient boundary.</p>
2742 a pointer is returned.</p>
2746 <pre>
2754 </pre>
2755 </div>
2757 <!-- _______________________________________________________________________ -->
2760 </div>
2766 <pre>
2768 </pre>
2773 memory heap to be reallocated in the future.</p>
2779 instruction.</p>
2783 <p>Access to the memory pointed to by the pointer is no longer defined
2784 after this instruction executes.</p>
2788 <pre>
2790 free [4 x i8]* %array
2791 </pre>
2792 </div>
2794 <!-- _______________________________________________________________________ -->
2797 </div>
2803 <pre>
2804 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2805 </pre>
2810 currently executing function, to be automatically released when this function
2811 returns to its caller. The object is always allocated in the generic address
2812 space (address space zero).</p>
2817 bytes of memory on the runtime stack, returning a pointer of the
2818 appropriate type to the program. If "NumElements" is specified, it is the
2819 number of elements allocated. If an alignment is specified, the value result
2820 of the allocation is guaranteed to be aligned to at least that boundary. If
2821 not specified, or if zero, the target can choose to align the allocation on any
2822 convenient boundary.</p>
2830 instruction is commonly used to represent automatic variables that must
2831 have an address available. When the function returns (either with the <tt><a
2833 instructions), the memory is reclaimed.</p>
2837 <pre>
2842 </pre>
2843 </div>
2845 <!-- _______________________________________________________________________ -->
2847 Instruction</a> </div>
2850 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
2855 address from which to load. The pointer must point to a <a
2860 instructions. </p>
2861 <p>
2862 The optional "align" argument specifies the alignment of the operation
2863 (that is, the alignment of the memory address). A value of 0 or an
2864 omitted "align" argument means that the operation has the preferential
2865 alignment for the target. It is the responsibility of the code emitter
2866 to ensure that the alignment information is correct. Overestimating
2867 the alignment results in an undefined behavior. Underestimating the
2868 alignment may produce less efficient code. An alignment of 1 is always
2869 safe.
2870 </p>
2875 <a
2878 </pre>
2879 </div>
2880 <!-- _______________________________________________________________________ -->
2882 Instruction</a> </div>
2885 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2886 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2887 </pre>
2895 optimizer is not allowed to modify the number or order of execution of
2898 <p>
2899 The optional "align" argument specifies the alignment of the operation
2900 (that is, the alignment of the memory address). A value of 0 or an
2901 omitted "align" argument means that the operation has the preferential
2902 alignment for the target. It is the responsibility of the code emitter
2903 to ensure that the alignment information is correct. Overestimating
2904 the alignment results in an undefined behavior. Underestimating the
2905 alignment may produce less efficient code. An alignment of 1 is always
2906 safe.
2907 </p>
2915 </pre>
2916 </div>
2918 <!-- _______________________________________________________________________ -->
2921 </div>
2925 <pre>
2927 </pre>
2931 <p>
2933 subelement of an aggregate data structure.</p>
2937 <p>This instruction takes a list of integer operands that indicate what
2938 elements of the aggregate object to index to. The actual types of the arguments
2939 provided depend on the type of the first pointer argument. The
2941 levels of a structure or to a specific index in an array. When indexing into a
2946 <p>For example, let's consider a C code fragment and how it gets
2947 compiled to LLVM:</p>
2950 <pre>
2951 struct RT {
2952 char A;
2954 char C;
2955 };
2956 struct ST {
2957 int X;
2958 double Y;
2959 struct RT Z;
2960 };
2962 int *foo(struct ST *s) {
2964 }
2965 </pre>
2966 </div>
2971 <pre>
2973 %ST = type { i32, double, %RT }
2975 define i32* %foo(%ST* %s) {
2976 entry:
2978 ret i32* %reg
2979 }
2980 </pre>
2981 </div>
2994 }</tt>' type, a structure. The second index indexes into the third element of
2996 i8 }</tt>' type, another structure. The third index indexes into the second
2998 array. The two dimensions of the array are subscripted into, yielding an
3002 <p>Note that it is perfectly legal to index partially through a
3003 structure, returning a pointer to an inner element. Because of this,
3004 the LLVM code for the given testcase is equivalent to:</p>
3006 <pre>
3007 define i32* %foo(%ST* %s) {
3013 ret i32* %t5
3014 }
3015 </pre>
3017 <p>Note that it is undefined to access an array out of bounds: array and
3018 pointer indexes must always be within the defined bounds of the array type.
3019 The one exception for this rules is zero length arrays. These arrays are
3020 defined to be accessible as variable length arrays, which requires access
3021 beyond the zero'th element.</p>
3023 <p>The getelementptr instruction is often confusing. For some more insight
3029 <pre>
3032 </pre>
3033 </div>
3035 <!-- ======================================================================= -->
3037 </div>
3039 <p>The instructions in this category are the conversion instructions (casting)
3040 which all take a single operand and a type. They perform various bit conversions
3041 on the operand.</p>
3042 </div>
3044 <!-- _______________________________________________________________________ -->
3047 </div>
3051 <pre>
3053 </pre>
3056 <p>
3058 </p>
3061 <p>
3069 <p>
3073 It will always truncate bits.</p>
3076 <pre>
3080 </pre>
3081 </div>
3083 <!-- _______________________________________________________________________ -->
3086 </div>
3090 <pre>
3092 </pre>
3113 <pre>
3116 </pre>
3117 </div>
3119 <!-- _______________________________________________________________________ -->
3122 </div>
3126 <pre>
3128 </pre>
3134 <p>
3142 <p>
3150 <pre>
3153 </pre>
3154 </div>
3156 <!-- _______________________________________________________________________ -->
3159 </div>
3165 <pre>
3167 </pre>
3188 <pre>
3191 </pre>
3192 </div>
3194 <!-- _______________________________________________________________________ -->
3197 </div>
3201 <pre>
3203 </pre>
3207 floating point value.</p>
3213 type must be smaller than the destination type.</p>
3223 <pre>
3226 </pre>
3227 </div>
3229 <!-- _______________________________________________________________________ -->
3232 </div>
3236 <pre>
3238 </pre>
3243 </p>
3256 the results are undefined.</p>
3259 <pre>
3263 </pre>
3264 </div>
3266 <!-- _______________________________________________________________________ -->
3269 </div>
3273 <pre>
3275 </pre>
3280 </p>
3293 the results are undefined.</p>
3296 <pre>
3300 </pre>
3301 </div>
3303 <!-- _______________________________________________________________________ -->
3306 </div>
3310 <pre>
3312 </pre>
3327 integer quantity and converts it to the corresponding floating point value. If
3328 the value cannot fit in the floating point value, the results are undefined.</p>
3331 <pre>
3334 </pre>
3335 </div>
3337 <!-- _______________________________________________________________________ -->
3340 </div>
3344 <pre>
3346 </pre>
3361 integer quantity and converts it to the corresponding floating point value. If
3362 the value cannot fit in the floating point value, the results are undefined.</p>
3365 <pre>
3368 </pre>
3369 </div>
3371 <!-- _______________________________________________________________________ -->
3374 </div>
3378 <pre>
3380 </pre>
3394 truncating or zero extending that value to the size of the integer type. If
3398 change.</p>
3401 <pre>
3404 </pre>
3405 </div>
3407 <!-- _______________________________________________________________________ -->
3410 </div>
3414 <pre>
3416 </pre>
3424 value to cast, and a type to cast it to, which must be a
3432 the size of a pointer then a zero extension is done. If they are the same size,
3436 <pre>
3440 </pre>
3441 </div>
3443 <!-- _______________________________________________________________________ -->
3446 </div>
3450 <pre>
3452 </pre>
3460 a first class value, and a type to cast it to, which must also be a <a
3463 type is a pointer, the destination type must also be a pointer.</p>
3470 converted to other pointer types with this instruction. To convert pointers to
3475 <pre>
3479 </pre>
3480 </div>
3482 <!-- ======================================================================= -->
3486 instructions, which defy better classification.</p>
3487 </div>
3489 <!-- _______________________________________________________________________ -->
3491 </div>
3494 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3495 </pre>
3498 of its two integer operands.</p>
3501 the condition code indicating the kind of comparison to perform. It is not
3502 a value, just a keyword. The possible condition code are:
3503 <ol>
3514 </ol>
3521 <ol>
3524 </li>
3543 </ol>
3545 values are compared as if they were integers.</p>
3554 </pre>
3555 </div>
3557 <!-- _______________________________________________________________________ -->
3559 </div>
3562 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3563 </pre>
3566 of its floating point operands.</p>
3569 the condition code indicating the kind of comparison to perform. It is not
3570 a value, just a keyword. The possible condition code are:
3571 <ol>
3588 </ol>
3593 types.</p>
3598 <ol>
3627 </ol>
3634 </pre>
3635 </div>
3637 <!-- _______________________________________________________________________ -->
3639 Instruction</a> </div>
3645 the SSA graph representing the function.</p>
3647 <p>The type of the incoming values is specified with the first type
3649 as arguments, with one pair for each predecessor basic block of the
3651 type may be used as the value arguments to the PHI node. Only labels
3652 may be used as the label arguments.</p>
3653 <p>There must be no non-phi instructions between the start of a basic
3654 block and the PHI instructions: i.e. PHI instructions must be first in
3655 a basic block.</p>
3658 specified by the pair corresponding to the predecessor basic block that executed
3659 just prior to the current block.</p>
3661 <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>
3662 </div>
3664 <!-- _______________________________________________________________________ -->
3667 </div>
3673 <pre>
3674 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3675 </pre>
3679 <p>
3681 condition, without branching.
3682 </p>
3687 <p>
3688 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.
3689 </p>
3693 <p>
3694 If the boolean condition evaluates to true, the instruction returns the first
3695 value argument; otherwise, it returns the second value argument.
3696 </p>
3700 <pre>
3702 </pre>
3703 </div>
3706 <!-- _______________________________________________________________________ -->
3709 </div>
3714 <pre>
3715 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
3716 </pre>
3726 <ol>
3727 <li>
3729 any allocas or varargs in the caller. If the "tail" marker is present, the
3730 function call is eligible for tail call optimization. Note that calls may
3733 </li>
3734 <li>
3736 convention</a> the call should use. If none is specified, the call defaults
3737 to using C calling conventions.
3738 </li>
3739 <li>
3741 the type of the return value. Functions that return no value are marked
3743 </li>
3744 <li>
3746 value being invoked. The argument types must match the types implied by
3747 this signature. This type can be omitted if the function is not varargs
3748 and if the function type does not return a pointer to a function.</p>
3749 </li>
3750 <li>
3752 be invoked. In most cases, this is a direct function invocation, but
3754 to function value.</p>
3755 </li>
3756 <li>
3758 function signature argument types. All arguments must be of
3760 indicates the function accepts a variable number of arguments, the extra
3761 arguments can be specified.</p>
3762 </li>
3763 </ol>
3768 transfer to a specified function, with its incoming arguments bound to
3770 instruction in the called function, control flow continues with the
3771 instruction after the function call, and the return value of the
3772 function is bound to the result argument. This is a simpler case of
3777 <pre>
3778 %retval = call i32 @test(i32 %argc)
3780 %X = tail call i32 @foo()
3782 %Z = call void %foo(i8 97 signext)
3783 </pre>
3785 </div>
3787 <!-- _______________________________________________________________________ -->
3790 </div>
3796 <pre>
3798 </pre>
3803 the "variable argument" area of a function call. It is used to implement the
3809 the argument. It returns a value of the specified argument type and
3821 <p>It is legal for this instruction to be called in a function which does not
3823 function.</p>
3827 argument.</p>
3833 </div>
3835 <!-- *********************************************************************** -->
3837 <!-- *********************************************************************** -->
3842 well known names and semantics and are required to follow certain restrictions.
3843 Overall, these intrinsics represent an extension mechanism for the LLVM
3844 language that does not require changing all of the transformations in LLVM when
3845 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
3848 prefix is reserved in LLVM for intrinsic names; thus, function names may not
3849 begin with this prefix. Intrinsic functions must always be external functions:
3850 you cannot define the body of intrinsic functions. Intrinsic functions may
3851 only be used in call or invoke instructions: it is illegal to take the address
3852 of an intrinsic function. Additionally, because intrinsic functions are part
3853 of the LLVM language, it is required if any are added that they be documented
3854 here.</p>
3856 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3857 a family of functions that perform the same operation but on different data
3858 types. Because LLVM can represent over 8 million different integer types,
3859 overloading is used commonly to allow an intrinsic function to operate on any
3860 integer type. One or more of the argument types or the result type can be
3861 overloaded to accept any integer type. Argument types may also be defined as
3862 exactly matching a previous argument's type or the result type. This allows an
3863 intrinsic function which accepts multiple arguments, but needs all of them to
3864 be of the same type, to only be overloaded with respect to a single argument or
3865 the result.</p>
3867 <p>Overloaded intrinsics will have the names of its overloaded argument types
3868 encoded into its function name, each preceded by a period. Only those types
3869 which are overloaded result in a name suffix. Arguments whose type is matched
3871 take an integer of any width and returns an integer of exactly the same integer
3872 width. This leads to a family of functions such as
3874 Only one type, the return type, is overloaded, and only one type suffix is
3875 required. Because the argument's type is matched against the return type, it
3876 does not require its own name suffix.</p>
3878 <p>To learn how to add an intrinsic function, please see the
3880 </p>
3882 </div>
3884 <!-- ======================================================================= -->
3887 </div>
3893 intrinsic functions. These functions are related to the similarly
3896 <p>All of these functions operate on arguments that use a
3897 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3898 language reference manual does not define what this type is, so all
3899 transformations should be prepared to handle these functions regardless of
3900 the type used.</p>
3903 instruction and the variable argument handling intrinsic functions are
3904 used.</p>
3907 <pre>
3908 define i32 @test(i32 %X, ...) {
3909 ; Initialize variable argument processing
3910 %ap = alloca i8*
3911 %ap2 = bitcast i8** %ap to i8*
3912 call void @llvm.va_start(i8* %ap2)
3914 ; Read a single integer argument
3915 %tmp = va_arg i8** %ap, i32
3917 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3918 %aq = alloca i8*
3919 %aq2 = bitcast i8** %aq to i8*
3920 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
3921 call void @llvm.va_end(i8* %aq2)
3923 ; Stop processing of arguments.
3924 call void @llvm.va_end(i8* %ap2)
3925 ret i32 %tmp
3926 }
3928 declare void @llvm.va_start(i8*)
3929 declare void @llvm.va_copy(i8*, i8*)
3930 declare void @llvm.va_end(i8*)
3931 </pre>
3932 </div>
3934 </div>
3936 <!-- _______________________________________________________________________ -->
3939 </div>
3957 macro available in C. In a target-dependent way, it initializes the
3961 last argument of the function as the compiler can figure that out.</p>
3963 </div>
3965 <!-- _______________________________________________________________________ -->
3968 </div>
3986 macro available in C. In a target-dependent way, it destroys the
3992 </div>
3994 <!-- _______________________________________________________________________ -->
3997 </div>
4003 <pre>
4005 </pre>
4010 from the source argument list to the destination argument list.</p>
4021 macro available in C. In a target-dependent way, it copies the source
4024 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4025 example, memory allocation.</p>
4027 </div>
4029 <!-- ======================================================================= -->
4032 </div>
4036 <p>
4038 Collection</a> requires the implementation and generation of these intrinsics.
4042 Front-ends for type-safe garbage collected languages should generate these
4043 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4045 </p>
4047 <p>The garbage collection intrinsics only operate on objects in the generic
4048 address space (address space zero).</p>
4050 </div>
4052 <!-- _______________________________________________________________________ -->
4055 </div>
4061 <pre>
4062 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4063 </pre>
4068 the code generator, and allows some metadata to be associated with it.</p>
4072 <p>The first argument specifies the address of a stack object that contains the
4073 root pointer. The second pointer (which must be either a constant or a global
4074 value address) contains the meta-data to be associated with the root.</p>
4079 location. At compile-time, the code generator generates information to allow
4084 </div>
4087 <!-- _______________________________________________________________________ -->
4090 </div>
4096 <pre>
4097 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4098 </pre>
4103 locations, allowing garbage collector implementations that require read
4104 barriers.</p>
4108 <p>The second argument is the address to read from, which should be an address
4109 allocated from the garbage collector. The first object is a pointer to the
4110 start of the referenced object, if needed by the language runtime (otherwise
4111 null).</p>
4116 instruction, but may be replaced with substantially more complex code by the
4121 </div>
4124 <!-- _______________________________________________________________________ -->
4127 </div>
4133 <pre>
4134 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4135 </pre>
4140 locations, allowing garbage collector implementations that require write
4141 barriers (such as generational or reference counting collectors).</p>
4145 <p>The first argument is the reference to store, the second is the start of the
4146 object to store it to, and the third is the address of the field of Obj to
4147 store to. If the runtime does not require a pointer to the object, Obj may be
4148 null.</p>
4153 instruction, but may be replaced with substantially more complex code by the
4158 </div>
4162 <!-- ======================================================================= -->
4165 </div>
4168 <p>
4169 These intrinsics are provided by LLVM to expose special features that may only
4170 be implemented with code generator support.
4171 </p>
4173 </div>
4175 <!-- _______________________________________________________________________ -->
4178 </div>
4183 <pre>
4185 </pre>
4189 <p>
4191 target-specific value indicating the return address of the current function
4192 or one of its callers.
4193 </p>
4197 <p>
4198 The argument to this intrinsic indicates which function to return the address
4199 for. Zero indicates the calling function, one indicates its caller, etc. The
4201 </p>
4205 <p>
4207 the return address of the specified call frame, or zero if it cannot be
4208 identified. The value returned by this intrinsic is likely to be incorrect or 0
4209 for arguments other than zero, so it should only be used for debugging purposes.
4210 </p>
4212 <p>
4213 Note that calling this intrinsic does not prevent function inlining or other
4214 aggressive transformations, so the value returned may not be that of the obvious
4215 source-language caller.
4216 </p>
4217 </div>
4220 <!-- _______________________________________________________________________ -->
4223 </div>
4228 <pre>
4230 </pre>
4234 <p>
4236 target-specific frame pointer value for the specified stack frame.
4237 </p>
4241 <p>
4242 The argument to this intrinsic indicates which function to return the frame
4243 pointer for. Zero indicates the calling function, one indicates its caller,
4245 </p>
4249 <p>
4251 the frame address of the specified call frame, or zero if it cannot be
4252 identified. The value returned by this intrinsic is likely to be incorrect or 0
4253 for arguments other than zero, so it should only be used for debugging purposes.
4254 </p>
4256 <p>
4257 Note that calling this intrinsic does not prevent function inlining or other
4258 aggressive transformations, so the value returned may not be that of the obvious
4259 source-language caller.
4260 </p>
4261 </div>
4263 <!-- _______________________________________________________________________ -->
4266 </div>
4271 <pre>
4272 declare i8 *@llvm.stacksave()
4273 </pre>
4277 <p>
4281 features like scoped automatic variable sized arrays in C99.
4282 </p>
4286 <p>
4287 This intrinsic returns a opaque pointer value that can be passed to <a
4294 </p>
4296 </div>
4298 <!-- _______________________________________________________________________ -->
4301 </div>
4306 <pre>
4307 declare void @llvm.stackrestore(i8 * %ptr)
4308 </pre>
4312 <p>
4314 the function stack to the state it was in when the corresponding <a
4316 useful for implementing language features like scoped automatic variable sized
4317 arrays in C99.
4318 </p>
4322 <p>
4324 </p>
4326 </div>
4329 <!-- _______________________________________________________________________ -->
4332 </div>
4337 <pre>
4339 </pre>
4344 <p>
4346 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4347 no
4348 effect on the behavior of the program but can change its performance
4349 characteristics.
4350 </p>
4354 <p>
4360 </p>
4364 <p>
4365 This intrinsic does not modify the behavior of the program. In particular,
4366 prefetches cannot trap and do not produce a value. On targets that support this
4367 intrinsic, the prefetch can provide hints to the processor cache for better
4368 performance.
4369 </p>
4371 </div>
4373 <!-- _______________________________________________________________________ -->
4376 </div>
4381 <pre>
4383 </pre>
4388 <p>
4390 (PC) in a region of
4391 code to simulators and other tools. The method is target specific, but it is
4392 expected that the marker will use exported symbols to transmit the PC of the marker.
4393 The marker makes no guarantees that it will remain with any specific instruction
4394 after optimizations. It is possible that the presence of a marker will inhibit
4395 optimizations. The intended use is to be inserted after optimizations to allow
4396 correlations of simulation runs.
4397 </p>
4401 <p>
4403 </p>
4407 <p>
4408 This intrinsic does not modify the behavior of the program. Backends that do not
4409 support this intrinisic may ignore it.
4410 </p>
4412 </div>
4414 <!-- _______________________________________________________________________ -->
4417 </div>
4422 <pre>
4423 declare i64 @llvm.readcyclecounter( )
4424 </pre>
4429 <p>
4431 counter register (or similar low latency, high accuracy clocks) on those targets
4432 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4433 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4434 should only be used for small timings.
4435 </p>
4439 <p>
4440 When directly supported, reading the cycle counter should not modify any memory.
4441 Implementations are allowed to either return a application specific value or a
4442 system wide value. On backends without support, this is lowered to a constant 0.
4443 </p>
4445 </div>
4447 <!-- ======================================================================= -->
4450 </div>
4453 <p>
4454 LLVM provides intrinsics for a few important standard C library functions.
4455 These intrinsics allow source-language front-ends to pass information about the
4456 alignment of the pointer arguments to the code generator, providing opportunity
4457 for more efficient code generation.
4458 </p>
4460 </div>
4462 <!-- _______________________________________________________________________ -->
4465 </div>
4470 <pre>
4475 </pre>
4479 <p>
4481 location to the destination location.
4482 </p>
4484 <p>
4486 intrinsics do not return a value, and takes an extra alignment argument.
4487 </p>
4491 <p>
4492 The first argument is a pointer to the destination, the second is a pointer to
4493 the source. The third argument is an integer argument
4494 specifying the number of bytes to copy, and the fourth argument is the alignment
4495 of the source and destination locations.
4496 </p>
4498 <p>
4500 the caller guarantees that both the source and destination pointers are aligned
4501 to that boundary.
4502 </p>
4506 <p>
4508 location to the destination location, which are not allowed to overlap. It
4509 copies "len" bytes of memory over. If the argument is known to be aligned to
4510 some boundary, this can be specified as the fourth argument, otherwise it should
4512 </p>
4513 </div>
4516 <!-- _______________________________________________________________________ -->
4519 </div>
4524 <pre>
4529 </pre>
4533 <p>
4535 location to the destination location. It is similar to the
4537 </p>
4539 <p>
4541 intrinsics do not return a value, and takes an extra alignment argument.
4542 </p>
4546 <p>
4547 The first argument is a pointer to the destination, the second is a pointer to
4548 the source. The third argument is an integer argument
4549 specifying the number of bytes to copy, and the fourth argument is the alignment
4550 of the source and destination locations.
4551 </p>
4553 <p>
4555 the caller guarantees that the source and destination pointers are aligned to
4556 that boundary.
4557 </p>
4561 <p>
4563 location to the destination location, which may overlap. It
4564 copies "len" bytes of memory over. If the argument is known to be aligned to
4565 some boundary, this can be specified as the fourth argument, otherwise it should
4567 </p>
4568 </div>
4571 <!-- _______________________________________________________________________ -->
4574 </div>
4579 <pre>
4584 </pre>
4588 <p>
4590 byte value.
4591 </p>
4593 <p>
4595 does not return a value, and takes an extra alignment argument.
4596 </p>
4600 <p>
4601 The first argument is a pointer to the destination to fill, the second is the
4602 byte value to fill it with, the third argument is an integer
4603 argument specifying the number of bytes to fill, and the fourth argument is the
4604 known alignment of destination location.
4605 </p>
4607 <p>
4609 the caller guarantees that the destination pointer is aligned to that boundary.
4610 </p>
4614 <p>
4616 the
4617 destination location. If the argument is known to be aligned to some boundary,
4618 this can be specified as the fourth argument, otherwise it should be set to 0 or
4619 1.
4620 </p>
4621 </div>
4624 <!-- _______________________________________________________________________ -->
4627 </div>
4633 floating point or vector of floating point type. Not all targets support all
4634 types however.
4635 <pre>
4636 declare float @llvm.sqrt.f32(float %Val)
4637 declare double @llvm.sqrt.f64(double %Val)
4638 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
4639 declare fp128 @llvm.sqrt.f128(fp128 %Val)
4640 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
4641 </pre>
4645 <p>
4649 negative numbers (which allows for better optimization).
4650 </p>
4654 <p>
4655 The argument and return value are floating point numbers of the same type.
4656 </p>
4660 <p>
4661 This function returns the sqrt of the specified operand if it is a nonnegative
4662 floating point number.
4663 </p>
4664 </div>
4666 <!-- _______________________________________________________________________ -->
4669 </div>
4675 floating point or vector of floating point type. Not all targets support all
4676 types however.
4677 <pre>
4678 declare float @llvm.powi.f32(float %Val, i32 %power)
4679 declare double @llvm.powi.f64(double %Val, i32 %power)
4680 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
4681 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
4682 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
4683 </pre>
4687 <p>
4689 specified (positive or negative) power. The order of evaluation of
4690 multiplications is not defined. When a vector of floating point type is
4691 used, the second argument remains a scalar integer value.
4692 </p>
4696 <p>
4697 The second argument is an integer power, and the first is a value to raise to
4698 that power.
4699 </p>
4703 <p>
4704 This function returns the first value raised to the second power with an
4705 unspecified sequence of rounding operations.</p>
4706 </div>
4708 <!-- _______________________________________________________________________ -->
4711 </div>
4717 floating point or vector of floating point type. Not all targets support all
4718 types however.
4719 <pre>
4720 declare float @llvm.sin.f32(float %Val)
4721 declare double @llvm.sin.f64(double %Val)
4722 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
4723 declare fp128 @llvm.sin.f128(fp128 %Val)
4724 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
4725 </pre>
4729 <p>
4731 </p>
4735 <p>
4736 The argument and return value are floating point numbers of the same type.
4737 </p>
4741 <p>
4742 This function returns the sine of the specified operand, returning the
4744 conditions in the same way.</p>
4745 </div>
4747 <!-- _______________________________________________________________________ -->
4750 </div>
4756 floating point or vector of floating point type. Not all targets support all
4757 types however.
4758 <pre>
4759 declare float @llvm.cos.f32(float %Val)
4760 declare double @llvm.cos.f64(double %Val)
4761 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
4762 declare fp128 @llvm.cos.f128(fp128 %Val)
4763 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
4764 </pre>
4768 <p>
4770 </p>
4774 <p>
4775 The argument and return value are floating point numbers of the same type.
4776 </p>
4780 <p>
4781 This function returns the cosine of the specified operand, returning the
4783 conditions in the same way.</p>
4784 </div>
4786 <!-- _______________________________________________________________________ -->
4789 </div>
4795 floating point or vector of floating point type. Not all targets support all
4796 types however.
4797 <pre>
4798 declare float @llvm.pow.f32(float %Val, float %Power)
4799 declare double @llvm.pow.f64(double %Val, double %Power)
4800 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
4801 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
4802 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
4803 </pre>
4807 <p>
4809 specified (positive or negative) power.
4810 </p>
4814 <p>
4815 The second argument is a floating point power, and the first is a value to
4816 raise to that power.
4817 </p>
4821 <p>
4822 This function returns the first value raised to the second power,
4823 returning the
4825 conditions in the same way.</p>
4826 </div>
4829 <!-- ======================================================================= -->
4832 </div>
4835 <p>
4836 LLVM provides intrinsics for a few important bit manipulation operations.
4837 These allow efficient code generation for some algorithms.
4838 </p>
4840 </div>
4842 <!-- _______________________________________________________________________ -->
4845 </div>
4850 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4852 <pre>
4856 </pre>
4860 <p>
4862 values with an even number of bytes (positive multiple of 16 bits). These are
4863 useful for performing operations on data that is not in the target's native
4864 byte order.
4865 </p>
4869 <p>
4872 intrinsic returns an i32 value that has the four bytes of the input i32
4877 </p>
4879 </div>
4881 <!-- _______________________________________________________________________ -->
4884 </div>
4889 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4890 width. Not all targets support all bit widths however.
4891 <pre>
4897 </pre>
4901 <p>
4903 value.
4904 </p>
4908 <p>
4909 The only argument is the value to be counted. The argument may be of any
4910 integer type. The return type must match the argument type.
4911 </p>
4915 <p>
4917 </p>
4918 </div>
4920 <!-- _______________________________________________________________________ -->
4923 </div>
4929 integer bit width. Not all targets support all bit widths however.
4930 <pre>
4936 </pre>
4940 <p>
4942 leading zeros in a variable.
4943 </p>
4947 <p>
4948 The only argument is the value to be counted. The argument may be of any
4949 integer type. The return type must match the argument type.
4950 </p>
4954 <p>
4956 in a variable. If the src == 0 then the result is the size in bits of the type
4958 </p>
4959 </div>
4963 <!-- _______________________________________________________________________ -->
4966 </div>
4972 integer bit width. Not all targets support all bit widths however.
4973 <pre>
4979 </pre>
4983 <p>
4985 trailing zeros.
4986 </p>
4990 <p>
4991 The only argument is the value to be counted. The argument may be of any
4992 integer type. The return type must match the argument type.
4993 </p>
4997 <p>
4999 in a variable. If the src == 0 then the result is the size in bits of the type
5001 </p>
5002 </div>
5004 <!-- _______________________________________________________________________ -->
5007 </div>
5013 on any integer bit width.
5014 <pre>
5015 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5016 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5017 </pre>
5021 range of bits from an integer value and returns them in the same bit width as
5022 the original value.</p>
5026 any bit width but they must have the same bit width. The second and third
5033 operates in forward mode.</p>
5037 <ol>
5041 to determine the number of bits to retain.</li>
5044 </ol>
5045 <p>In reverse mode, a similar computation is made except that the bits are
5050 </div>
5054 </div>
5060 on any integer bit width.
5061 <pre>
5062 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5063 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5064 </pre>
5068 of bits in an integer value with another integer value. It returns the integer
5069 with the replaced bits.</p>
5076 type since they specify only a bit index.</p>
5082 operates in forward mode.</p>
5084 truncating it down to the size of the replacement area or zero extending it
5085 up to that size.</p>
5090 <p>In reverse mode, a similar computation is made except that the bits are
5094 <pre>
5100 </pre>
5101 </div>
5103 <!-- ======================================================================= -->
5106 </div>
5109 <p>
5111 are described in the <a
5113 Debugging</a> document.
5114 </p>
5115 </div>
5118 <!-- ======================================================================= -->
5121 </div>
5124 <p> The LLVM exception handling intrinsics (which all start with
5128 </div>
5130 <!-- ======================================================================= -->
5133 </div>
5136 <p>
5137 This intrinsic makes it possible to excise one parameter, marked with
5139 function pointer lacking the nest parameter - the caller does not need
5140 to provide a value for it. Instead, the value to use is stored in
5141 advance in a "trampoline", a block of memory usually allocated
5142 on the stack, which also contains code to splice the nest value into the
5143 argument list. This is used to implement the GCC nested function address
5144 extension.
5145 </p>
5146 <p>
5147 For example, if the function is
5150 <pre>
5153 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5154 %fp = bitcast i8* %p to i32 (i32, i32)*
5155 </pre>
5158 </div>
5160 <!-- _______________________________________________________________________ -->
5163 </div>
5166 <pre>
5168 </pre>
5170 <p>
5172 and returns a function pointer suitable for executing it.
5173 </p>
5175 <p>
5178 and sufficiently aligned block of memory; this memory is written to by the
5179 intrinsic. Note that the size and the alignment are target-specific - LLVM
5180 currently provides no portable way of determining them, so a front-end that
5181 generates this intrinsic needs to have some target-specific knowledge.
5183 </p>
5185 <p>
5187 dependent code, turning it into a function. A pointer to this function is
5188 returned, but needs to be bitcast to an
5190 before being called. The new function's signature is the same as that of
5193 of pointer type. Calling the new function is equivalent to calling
5197 modified, then the effect of any later call to the returned function pointer is
5198 undefined.
5199 </p>
5200 </div>
5202 <!-- ======================================================================= -->
5205 </div>
5208 <p> This class of intrinsics is designed to be generic and has
5209 no specific purpose. </p>
5210 </div>
5212 <!-- _______________________________________________________________________ -->
5215 </div>
5220 <pre>
5221 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5222 </pre>
5226 <p>
5228 </p>
5232 <p>
5233 The first argument is a pointer to a value, the second is a pointer to a
5234 global string, the third is a pointer to a global string which is the source
5235 file name, and the last argument is the line number.
5236 </p>
5240 <p>
5241 This intrinsic allows annotation of local variables with arbitrary strings.
5242 This can be useful for special purpose optimizations that want to look for these
5243 annotations. These have no other defined use, they are ignored by code
5244 generation and optimization.
5245 </p>
5246 </div>
5248 <!-- _______________________________________________________________________ -->
5251 </div>
5257 any integer bit width.
5258 </p>
5259 <pre>
5260 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
5261 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
5262 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
5263 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
5264 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
5265 </pre>
5269 <p>
5271 </p>
5275 <p>
5276 The first argument is an integer value (result of some expression),
5277 the second is a pointer to a global string, the third is a pointer to a global
5278 string which is the source file name, and the last argument is the line number.
5279 It returns the value of the first argument.
5280 </p>
5284 <p>
5285 This intrinsic allows annotations to be put on arbitrary expressions
5286 with arbitrary strings. This can be useful for special purpose optimizations
5287 that want to look for these annotations. These have no other defined use, they
5288 are ignored by code generation and optimization.
5289 </div>
5291 <!-- _______________________________________________________________________ -->
5294 </div>
5299 <pre>
5300 declare void @llvm.trap()
5301 </pre>
5305 <p>
5307 </p>
5311 <p>
5312 None
5313 </p>
5317 <p>
5318 This intrinsics is lowered to the target dependent trap instruction. If the
5319 target does not have a trap instruction, this intrinsic will be lowered to the
5320 call of the abort() function.
5321 </p>
5322 </div>
5324 <!-- *********************************************************************** -->
5325 <hr>
5326 <address>
5334 Last modified: $Date$
5335 </address>
5337 </body>
5338 </html>