| blob | 2d2f945df96b62b4a38c5216e4bf6db7aeea2e18 |
3 <html>
4 <head>
11 </head>
13 <body>
16 <ol>
21 <ol>
31 </ol>
32 </li>
34 <ol>
36 <ol>
38 </ol>
39 </li>
41 <ol>
49 </ol>
50 </li>
51 </ol>
52 </li>
54 <ol>
60 </ol>
61 </li>
63 <ol>
65 </ol>
66 </li>
68 <ol>
70 <ol>
77 </ol>
78 </li>
80 <ol>
90 </ol>
91 </li>
93 <ol>
100 </ol>
101 </li>
103 <ol>
107 </ol>
108 </li>
110 <ol>
117 </ol>
118 </li>
120 <ol>
133 </ol>
135 <ol>
142 </ol>
143 </li>
144 </ol>
145 </li>
147 <ol>
149 <ol>
153 </ol>
154 </li>
156 <ol>
160 </ol>
161 </li>
163 <ol>
171 </ol>
172 </li>
174 <ol>
180 </ol>
181 </li>
183 <ol>
190 </ol>
191 </li>
195 <ol>
201 </ol>
202 </li>
204 <ol>
206 </ol>
207 </li>
209 <ol>
212 </ol>
213 <ol>
216 </ol>
217 </li>
218 </ol>
219 </li>
220 </ol>
225 </div>
227 <!-- *********************************************************************** -->
229 <!-- *********************************************************************** -->
232 <p>This document is a reference manual for the LLVM assembly language.
233 LLVM is an SSA based representation that provides type safety,
234 low-level operations, flexibility, and the capability of representing
235 'all' high-level languages cleanly. It is the common code
236 representation used throughout all phases of the LLVM compilation
237 strategy.</p>
238 </div>
240 <!-- *********************************************************************** -->
242 <!-- *********************************************************************** -->
246 <p>The LLVM code representation is designed to be used in three
247 different forms: as an in-memory compiler IR, as an on-disk bitcode
248 representation (suitable for fast loading by a Just-In-Time compiler),
249 and as a human readable assembly language representation. This allows
250 LLVM to provide a powerful intermediate representation for efficient
251 compiler transformations and analysis, while providing a natural means
252 to debug and visualize the transformations. The three different forms
253 of LLVM are all equivalent. This document describes the human readable
254 representation and notation.</p>
256 <p>The LLVM representation aims to be light-weight and low-level
257 while being expressive, typed, and extensible at the same time. It
258 aims to be a "universal IR" of sorts, by being at a low enough level
259 that high-level ideas may be cleanly mapped to it (similar to how
260 microprocessors are "universal IR's", allowing many source languages to
261 be mapped to them). By providing type information, LLVM can be used as
262 the target of optimizations: for example, through pointer analysis, it
263 can be proven that a C automatic variable is never accessed outside of
264 the current function... allowing it to be promoted to a simple SSA
265 value instead of a memory location.</p>
267 </div>
269 <!-- _______________________________________________________________________ -->
274 <p>It is important to note that this document describes 'well formed'
275 LLVM assembly language. There is a difference between what the parser
276 accepts and what is considered 'well formed'. For example, the
277 following instruction is syntactically okay, but not well formed:</p>
280 <pre>
282 </pre>
283 </div>
286 its uses. The LLVM infrastructure provides a verification pass that may
287 be used to verify that an LLVM module is well formed. This pass is
288 automatically run by the parser after parsing input assembly and by
289 the optimizer before it outputs bitcode. The violations pointed out
290 by the verifier pass indicate bugs in transformation passes or input to
291 the parser.</p>
292 </div>
294 <!-- Describe the typesetting conventions here. -->
296 <!-- *********************************************************************** -->
298 <!-- *********************************************************************** -->
302 <p>LLVM identifiers come in two basic types: global and local. Global
303 identifiers (functions, global variables) begin with the @ character. Local
304 identifiers (register names, types) begin with the % character. Additionally,
305 there are three different formats for identifiers, for different purposes:
307 <ol>
308 <li>Named values are represented as a string of characters with their prefix.
309 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
311 Identifiers which require other characters in their names can be surrounded
313 be used in a named value.</li>
315 <li>Unnamed values are represented as an unsigned numeric value with their
320 </ol>
322 <p>LLVM requires that values start with a prefix for two reasons: Compilers
323 don't need to worry about name clashes with reserved words, and the set of
324 reserved words may be expanded in the future without penalty. Additionally,
325 unnamed identifiers allow a compiler to quickly come up with a temporary
326 variable without having to avoid symbol table conflicts.</p>
328 <p>Reserved words in LLVM are very similar to reserved words in other
329 languages. There are keywords for different opcodes
334 and others. These reserved words cannot conflict with variable names, because
335 none of them start with a prefix character ('%' or '@').</p>
337 <p>Here is an example of LLVM code to multiply the integer variable
343 <pre>
345 </pre>
346 </div>
351 <pre>
353 </pre>
354 </div>
359 <pre>
363 </pre>
364 </div>
367 important lexical features of LLVM:</p>
369 <ol>
372 line.</li>
374 <li>Unnamed temporaries are created when the result of a computation is not
375 assigned to a named value.</li>
379 </ol>
381 <p>...and it also shows a convention that we follow in this document. When
382 demonstrating instructions, we will follow an instruction with a comment that
383 defines the type and name of value produced. Comments are shown in italic
384 text.</p>
386 </div>
388 <!-- *********************************************************************** -->
390 <!-- *********************************************************************** -->
392 <!-- ======================================================================= -->
394 </div>
399 translation unit of the input programs. Each module consists of
400 functions, global variables, and symbol table entries. Modules may be
401 combined together with the LLVM linker, which merges function (and
402 global variable) definitions, resolves forward declarations, and merges
408 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
416 %cast210 = <a
420 <a
422 <a
424 </pre>
425 </div>
432 <p>In general, a module is made up of a list of global values,
433 where both functions and global variables are global values. Global values are
434 represented by a pointer to a memory location (in this case, a pointer to an
435 array of char, and a pointer to a function), and have one of the following <a
438 </div>
440 <!-- ======================================================================= -->
443 </div>
447 <p>
448 All Global Variables and Functions have one of the following types of linkage:
449 </p>
451 <dl>
455 <dd>Global values with internal linkage are only directly accessible by
456 objects in the current module. In particular, linking code into a module with
457 an internal global value may cause the internal to be renamed as necessary to
458 avoid collisions. Because the symbol is internal to the module, all
459 references can be updated. This corresponds to the notion of the
461 </dd>
466 the same name when linkage occurs. This is typically used to implement
467 inline functions, templates, or other code which must be generated in each
469 allowed to be discarded.
470 </dd>
476 used for globals that may be emitted in multiple translation units, but that
477 are not guaranteed to be emitted into every translation unit that uses them.
478 One example of this are common globals in C, such as "<tt>int X;</tt>" at
479 global scope.
480 </dd>
485 pointer to array type. When two global variables with appending linkage are
486 linked together, the two global arrays are appended together. This is the
487 LLVM, typesafe, equivalent of having the system linker append together
488 "sections" with identical names when .o files are linked.
489 </dd>
492 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
493 until linked, if not linked, the symbol becomes null instead of being an
494 undefined reference.
495 </dd>
499 <dd>If none of the above identifiers are used, the global is externally
500 visible, meaning that it participates in linkage and can be used to resolve
501 external symbol references.
502 </dd>
503 </dl>
505 <p>
506 The next two types of linkage are targeted for Microsoft Windows platform
507 only. They are designed to support importing (exporting) symbols from (to)
508 DLLs.
509 </p>
511 <dl>
515 or variable via a global pointer to a pointer that is set up by the DLL
516 exporting the symbol. On Microsoft Windows targets, the pointer name is
518 </dd>
523 pointer to a pointer in a DLL, so that it can be referenced with the
526 name.
527 </dd>
529 </dl>
532 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
533 variable and was linked with this one, one of the two would be renamed,
535 external (i.e., lacking any linkage declarations), they are accessible
536 outside of the current module.</p>
541 linkages.
542 </div>
544 <!-- ======================================================================= -->
547 </div>
553 specified for the call. The calling convention of any pair of dynamic
554 caller/callee must match, or the behavior of the program is undefined. The
555 following calling conventions are supported by LLVM, and more may be added in
556 the future:</p>
558 <dl>
561 <dd>This calling convention (the default if no other calling convention is
562 specified) matches the target C calling conventions. This calling convention
563 supports varargs function calls and tolerates some mismatch in the declared
564 prototype and implemented declaration of the function (as does normal C).
565 </dd>
569 <dd>This calling convention attempts to make calls as fast as possible
570 (e.g. by passing things in registers). This calling convention allows the
571 target to use whatever tricks it wants to produce fast code for the target,
572 without having to conform to an externally specified ABI. Implementations of
573 this convention should allow arbitrary tail call optimization to be supported.
574 This calling convention does not support varargs and requires the prototype of
575 all callees to exactly match the prototype of the function definition.
576 </dd>
580 <dd>This calling convention attempts to make code in the caller as efficient
581 as possible under the assumption that the call is not commonly executed. As
582 such, these calls often preserve all registers so that the call does not break
583 any live ranges in the caller side. This calling convention does not support
584 varargs and requires the prototype of all callees to exactly match the
585 prototype of the function definition.
586 </dd>
590 <dd>Any calling convention may be specified by number, allowing
591 target-specific calling conventions to be used. Target specific calling
592 conventions start at 64.
593 </dd>
594 </dl>
596 <p>More calling conventions can be added/defined on an as-needed basis, to
597 support pascal conventions or any other well-known target-independent
598 convention.</p>
600 </div>
602 <!-- ======================================================================= -->
605 </div>
609 <p>
610 All Global Variables and Functions have one of the following visibility styles:
611 </p>
613 <dl>
616 <dd>On ELF, default visibility means that the declaration is visible to other
617 modules and, in shared libraries, means that the declared entity may be
618 overridden. On Darwin, default visibility means that the declaration is
619 visible to other modules. Default visibility corresponds to "external
620 linkage" in the language.
621 </dd>
625 <dd>Two declarations of an object with hidden visibility refer to the same
626 object if they are in the same shared object. Usually, hidden visibility
627 indicates that the symbol will not be placed into the dynamic symbol table,
628 so no other module (executable or shared library) can reference it
629 directly.
630 </dd>
634 <dd>On ELF, protected visibility indicates that the symbol will be placed in
635 the dynamic symbol table, but that references within the defining module will
636 bind to the local symbol. That is, the symbol cannot be overridden by another
637 module.
638 </dd>
639 </dl>
641 </div>
643 <!-- ======================================================================= -->
646 </div>
650 <p>Global variables define regions of memory allocated at compilation time
651 instead of run-time. Global variables may optionally be initialized, may have
652 an explicit section to be placed in, and may have an optional explicit alignment
653 specified. A variable may be defined as "thread_local", which means that it
654 will not be shared by threads (each thread will have a separated copy of the
655 variable). A variable may be defined as a global "constant," which indicates
657 optimization, allowing the global data to be placed in the read-only section of
658 an executable, etc). Note that variables that need runtime initialization
661 <p>
663 constant, even if the final definition of the global is not. This capability
664 can be used to enable slightly better optimization of the program, but requires
665 the language definition to guarantee that optimizations based on the
666 'constantness' are valid for the translation units that do not include the
667 definition.
668 </p>
670 <p>As SSA values, global variables define pointer values that are in
671 scope (i.e. they dominate) all basic blocks in the program. Global
672 variables always define a pointer to their "content" type because they
673 describe a region of memory, and all memory objects in LLVM are
674 accessed through pointers.</p>
676 <p>LLVM allows an explicit section to be specified for globals. If the target
677 supports it, it will emit globals to the section specified.</p>
679 <p>An explicit alignment may be specified for a global. If not present, or if
680 the alignment is set to zero, the alignment of the global is set by the target
681 to whatever it feels convenient. If an explicit alignment is specified, the
682 global is forced to have at least that much alignment. All alignments must be
685 <p>For example, the following defines a global with an initializer, section,
686 and alignment:</p>
689 <pre>
691 </pre>
692 </div>
694 </div>
697 <!-- ======================================================================= -->
700 </div>
709 name, a (possibly empty) argument list (each with optional
711 optional alignment, an opening curly brace, a list of basic blocks, and a
712 closing curly brace.
714 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
719 name, a possibly empty list of arguments, and an optional alignment.</p>
721 <p>A function definition contains a list of basic blocks, forming the CFG for
722 the function. Each basic block may optionally start with a label (giving the
723 basic block a symbol table entry), contains a list of instructions, and ends
725 function return).</p>
727 <p>The first basic block in a function is special in two ways: it is immediately
728 executed on entrance to the function, and it is not allowed to have predecessor
729 basic blocks (i.e. there can not be any branches to the entry block of a
730 function). Because the block can have no predecessors, it also cannot have any
733 <p>LLVM allows an explicit section to be specified for functions. If the target
734 supports it, it will emit functions to the section specified.</p>
736 <p>An explicit alignment may be specified for a function. If not present, or if
737 the alignment is set to zero, the alignment of the function is set by the target
738 to whatever it feels convenient. If an explicit alignment is specified, the
739 function is forced to have at least that much alignment. All alignments must be
742 </div>
745 <!-- ======================================================================= -->
748 </div>
751 function or global variable or bitcast of global value). Aliases may have an
758 <pre>
760 </pre>
761 </div>
763 </div>
767 <!-- ======================================================================= -->
770 <p>The return type and each parameter of a function type may have a set of
772 used to communicate additional information about the result or parameters of
773 a function. Parameter attributes are considered to be part of the function
774 type so two functions types that differ only by the parameter attributes
775 are different function types.</p>
777 <p>Parameter attributes are simple keywords that follow the type specified. If
778 multiple parameter attributes are needed, they are space separated. For
779 example:</p>
782 <pre>
783 %someFunc = i16 (i8 signext %someParam) zeroext
784 %someFunc = i16 (i8 zeroext %someParam) zeroext
785 </pre>
786 </div>
788 <p>Note that the two function types above are unique because the parameter has
790 the second). Also note that the attribute for the function result
794 <dl>
796 <dd>This indicates that the parameter should be zero extended just before
797 a call to this function.</dd>
799 <dd>This indicates that the parameter should be sign extended just before
800 a call to this function.</dd>
802 <dd>This indicates that the parameter should be placed in register (if
803 possible) during assembling function call. Support for this attribute is
804 target-specific</dd>
806 <dd>This indicates that the parameter specifies the address of a structure
807 that is the return value of the function in the source program.</dd>
809 <dd>This indicates that the parameter not alias any other object or any
810 other "noalias" objects during the function call.
812 <dd>This function attribute indicates that the function never returns. This
813 indicates to LLVM that every call to this function should be treated as if
816 <dd>This function attribute indicates that the function type does not use
817 the unwind instruction and does not allow stack unwinding to propagate
818 through it.</dd>
820 <dd>This indicates that the parameter can be excised using the
822 </dl>
824 </div>
826 <!-- ======================================================================= -->
829 </div>
832 <p>
833 Modules may contain "module-level inline asm" blocks, which corresponds to the
834 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
835 LLVM and treated as a single unit, but may be separated in the .ll file if
836 desired. The syntax is very simple:
837 </p>
840 <pre>
841 module asm "inline asm code goes here"
842 module asm "more can go here"
843 </pre>
844 </div>
846 <p>The strings can contain any character by escaping non-printable characters.
848 for the number.
849 </p>
851 <p>
852 The inline asm code is simply printed to the machine code .s file when
853 assembly code is generated.
854 </p>
855 </div>
857 <!-- ======================================================================= -->
860 </div>
863 <p>A module may specify a target specific data layout string that specifies how
864 data is to be laid out in memory. The syntax for the data layout is simply:</p>
867 separated by the minus sign character ('-'). Each specification starts with a
868 letter and may include other information after the letter to define some
869 aspect of the data layout. The specifications accepted are as follows: </p>
870 <dl>
872 <dd>Specifies that the target lays out data in big-endian form. That is, the
873 bits with the most significance have the lowest address location.</dd>
875 <dd>Specifies that hte target lays out data in little-endian form. That is,
876 the bits with the least significance have the lowest address location.</dd>
881 too.</dd>
883 <dd>This specifies the alignment for an integer type of a given bit
886 <dd>This specifies the alignment for a vector type of a given bit
889 <dd>This specifies the alignment for a floating point type of a given bit
891 (double).</dd>
893 <dd>This specifies the alignment for an aggregate type of a given bit
895 </dl>
896 <p>When constructing the data layout for a given target, LLVM starts with a
897 default set of specifications which are then (possibly) overriden by the
899 are given in this list:</p>
900 <ul>
914 </ul>
915 <p>When llvm is determining the alignment for a given type, it uses the
916 following rules:
917 <ol>
918 <li>If the type sought is an exact match for one of the specifications, that
919 specification is used.</li>
920 <li>If no match is found, and the type sought is an integer type, then the
921 smallest integer type that is larger than the bitwidth of the sought type is
922 used. If none of the specifications are larger than the bitwidth then the the
923 largest integer type is used. For example, given the default specifications
924 above, the i7 type will use the alignment of i8 (next largest) while both
925 i65 and i256 will use the alignment of i64 (largest specified).</li>
926 <li>If no match is found, and the type sought is a vector type, then the
927 largest vector type that is smaller than the sought vector type will be used
930 </ol>
931 </div>
933 <!-- *********************************************************************** -->
935 <!-- *********************************************************************** -->
939 <p>The LLVM type system is one of the most important features of the
940 intermediate representation. Being typed enables a number of
941 optimizations to be performed on the IR directly, without having to do
942 extra analyses on the side before the transformation. A strong type
943 system makes it easier to read the generated code and enables novel
944 analyses and transformations that are not feasible to perform on normal
945 three address code representations.</p>
947 </div>
949 <!-- ======================================================================= -->
952 <p>The primitive types are the fundamental building blocks of the LLVM
953 system. The current set of primitive types is as follows:</p>
958 <table>
959 <tbody>
963 </tbody>
964 </table>
965 </td>
967 <table>
968 <tbody>
972 </tbody>
973 </table>
974 </td>
975 </tr>
976 </table>
977 </div>
979 <!-- _______________________________________________________________________ -->
981 Classifications</a> </div>
983 <p>These different primitive types fall into a few useful
984 classifications:</p>
987 <tbody>
989 <tr>
992 </tr>
993 <tr>
996 </tr>
997 <tr>
1001 </td>
1002 </tr>
1003 </tbody>
1004 </table>
1007 most important. Values of these types are the only ones which can be
1008 produced by instructions, passed as arguments, or used as operands to
1009 instructions. This means that all structures and arrays must be
1010 manipulated either by pointer or by component.</p>
1011 </div>
1013 <!-- ======================================================================= -->
1018 <p>The real power in LLVM comes from the derived types in the system.
1019 This is what allows a programmer to represent arrays, functions,
1020 pointers, and other useful types. Note that these derived types may be
1021 recursive: For example, it is possible to have a two dimensional array.</p>
1023 </div>
1025 <!-- _______________________________________________________________________ -->
1031 <p>The integer type is a very simple derived type that simply specifies an
1032 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1037 <pre>
1038 iN
1039 </pre>
1042 value.</p>
1056 </td>
1063 An integer whose bit width is the answer. <br/>
1066 </td>
1067 </tr>
1068 </table>
1069 </div>
1071 <!-- _______________________________________________________________________ -->
1078 <p>The array type is a very simple derived type that arranges elements
1079 sequentially in memory. The array type requires a size (number of
1080 elements) and an underlying data type.</p>
1084 <pre>
1086 </pre>
1088 <p>The number of elements is a constant integer value; elementtype may
1089 be any type with a size.</p>
1098 </td>
1103 </td>
1104 </tr>
1105 </table>
1113 </td>
1118 </td>
1119 </tr>
1120 </table>
1122 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1123 length array. Normally, accesses past the end of an array are undefined in
1125 As a special case, however, zero length arrays are recognized to be variable
1126 length. This allows implementation of 'pascal style arrays' with the LLVM
1129 </div>
1131 <!-- _______________________________________________________________________ -->
1135 <p>The function type can be thought of as a function signature. It
1136 consists of a return type and a list of formal parameter types.
1137 Function types are usually used to build virtual function tables
1138 (which are structures of pointers to functions), for indirect function
1139 calls, and when defining a function.</p>
1140 <p>
1141 The return type of a function type cannot be an aggregate type.
1142 </p>
1147 which indicates that the function takes a variable number of arguments.
1148 Variable argument functions can access their arguments with the <a
1155 </td>
1158 </tt></td>
1163 </td>
1169 LLVM.
1170 </td>
1171 </tr>
1172 </table>
1174 </div>
1175 <!-- _______________________________________________________________________ -->
1179 <p>The structure type is used to represent a collection of data members
1180 together in memory. The packing of the field types is defined to match
1181 the ABI of the underlying processor. The elements of a structure may
1182 be any type that has a size.</p>
1186 instruction.</p>
1200 </tr>
1201 </table>
1202 </div>
1204 <!-- _______________________________________________________________________ -->
1206 </div>
1209 <p>The packed structure type is used to represent a collection of data members
1210 together in memory. There is no padding between fields. Further, the alignment
1211 of a packed structure is 1 byte. The elements of a packed structure may
1212 be any type that has a size.</p>
1216 instruction.</p>
1225 <td class="left"><tt>< { float, i32 (i32) * } ></tt></td>
1230 </tr>
1231 </table>
1232 </div>
1234 <!-- _______________________________________________________________________ -->
1238 <p>As in many languages, the pointer type represents a pointer or
1239 reference to another object, which must live in memory.</p>
1248 </td>
1255 </td>
1256 </tr>
1257 </table>
1258 </div>
1260 <!-- _______________________________________________________________________ -->
1266 <p>A vector type is a simple derived type that represents a vector
1267 of elements. Vector types are used when multiple primitive data
1268 are operated in parallel using a single instruction (SIMD).
1269 A vector type requires a size (number of
1270 elements) and an underlying primitive data type. Vectors must have a power
1276 <pre>
1278 </pre>
1280 <p>The number of elements is a constant integer value; elementtype may
1281 be any integer or floating point type.</p>
1291 </td>
1296 </td>
1297 </tr>
1298 </table>
1299 </div>
1301 <!-- _______________________________________________________________________ -->
1307 <p>Opaque types are used to represent unknown types in the system. This
1308 corresponds (for example) to the C notion of a foward declared structure type.
1309 In LLVM, opaque types can eventually be resolved to any type (not just a
1310 structure type).</p>
1314 <pre>
1315 opaque
1316 </pre>
1324 </td>
1326 An opaque type.<br/>
1327 </td>
1328 </tr>
1329 </table>
1330 </div>
1333 <!-- *********************************************************************** -->
1335 <!-- *********************************************************************** -->
1339 <p>LLVM has several different basic types of constants. This section describes
1340 them all and their syntax.</p>
1342 </div>
1344 <!-- ======================================================================= -->
1349 <dl>
1354 </dd>
1360 integer types.
1361 </dd>
1366 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1367 notation (see below). Floating point constants must have a <a
1375 </dl>
1377 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1378 of floating point constants. For example, the form '<tt>double
1381 (and the only time that they are generated by the disassembler) is when a
1382 floating point constant must be emitted but it cannot be represented as a
1383 decimal floating point number. For example, NaN's, infinities, and other
1384 special values are represented in their IEEE hexadecimal format so that
1385 assembly and disassembly do not cause any bits to change in the constants.</p>
1387 </div>
1389 <!-- ======================================================================= -->
1391 </div>
1394 <p>Aggregate constants arise from aggregation of simple constants
1395 and smaller aggregate constants.</p>
1397 <dl>
1400 <dd>Structure constants are represented with notation similar to structure
1401 type definitions (a comma separated list of elements, surrounded by braces
1405 types of elements must match those specified by the type.
1406 </dd>
1410 <dd>Array constants are represented with notation similar to array type
1411 definitions (a comma separated list of elements, surrounded by square brackets
1414 types of elements must match those specified by the type.
1415 </dd>
1419 <dd>Vector constants are represented with notation similar to vector type
1420 definitions (a comma separated list of elements, surrounded by
1424 match those specified by the type.
1425 </dd>
1431 This is often used to avoid having to print large zero initializers (e.g. for
1432 large arrays) and is always exactly equivalent to using explicit zero
1433 initializers.
1434 </dd>
1435 </dl>
1437 </div>
1439 <!-- ======================================================================= -->
1442 </div>
1448 constants. These constants are explicitly referenced when the <a
1451 file:</p>
1454 <pre>
1455 @X = global i32 17
1456 @Y = global i32 42
1457 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1458 </pre>
1459 </div>
1461 </div>
1463 <!-- ======================================================================= -->
1467 no specific value. Undefined values may be of any type and be used anywhere
1468 a constant is permitted.</p>
1470 <p>Undefined values indicate to the compiler that the program is well defined
1471 no matter what value is used, giving the compiler more freedom to optimize.
1472 </p>
1473 </div>
1475 <!-- ======================================================================= -->
1477 </div>
1481 <p>Constant expressions are used to allow expressions involving other constants
1482 to be used as constants. Constant expressions may be of any <a
1484 that does not have side effects (e.g. load and call are not supported). The
1485 following is the syntax for constant expressions:</p>
1487 <dl>
1489 <dd>Truncate a constant to another type. The bit size of CST must be larger
1490 than the bit size of TYPE. Both types must be integers.</dd>
1493 <dd>Zero extend a constant to another type. The bit size of CST must be
1494 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1497 <dd>Sign extend a constant to another type. The bit size of CST must be
1498 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1501 <dd>Truncate a floating point constant to another floating point type. The
1502 size of CST must be larger than the size of TYPE. Both types must be
1503 floating point.</dd>
1506 <dd>Floating point extend a constant to another type. The size of CST must be
1507 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1510 <dd>Convert a floating point constant to the corresponding unsigned integer
1511 constant. TYPE must be an integer type. CST must be floating point. If the
1512 value won't fit in the integer type, the results are undefined.</dd>
1515 <dd>Convert a floating point constant to the corresponding signed integer
1516 constant. TYPE must be an integer type. CST must be floating point. If the
1517 value won't fit in the integer type, the results are undefined.</dd>
1520 <dd>Convert an unsigned integer constant to the corresponding floating point
1521 constant. TYPE must be floating point. CST must be of integer type. If the
1522 value won't fit in the floating point type, the results are undefined.</dd>
1525 <dd>Convert a signed integer constant to the corresponding floating point
1526 constant. TYPE must be floating point. CST must be of integer type. If the
1527 value won't fit in the floating point type, the results are undefined.</dd>
1530 <dd>Convert a pointer typed constant to the corresponding integer constant
1531 TYPE must be an integer type. CST must be of pointer type. The CST value is
1532 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1535 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1536 pointer type. CST must be of integer type. The CST value is zero extended,
1537 truncated, or unchanged to make it fit in a pointer size. This one is
1541 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1542 identical (same number of bits). The conversion is done as if the CST value
1543 was stored to memory and read back as TYPE. In other words, no bits change
1544 with this operator, just the type. This can be used for conversion of
1545 vector types to any other type, as long as they have the same bit width. For
1546 pointers it is only valid to cast to another pointer type.
1547 </dd>
1553 instruction, the index list may have zero or more indexes, which are required
1559 constants.</dd>
1570 operation</a> on constants.
1585 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1587 binary</a> operations. The constraints on operands are the same as those for
1588 the corresponding instruction (e.g. no bitwise operations on floating point
1589 values are allowed).</dd>
1590 </dl>
1591 </div>
1593 <!-- *********************************************************************** -->
1595 <!-- *********************************************************************** -->
1597 <!-- ======================================================================= -->
1600 </div>
1604 <p>
1606 Module-Level Inline Assembly</a>) through the use of a special value. This
1607 value represents the inline assembler as a string (containing the instructions
1608 to emit), a list of operand constraints (stored as a string), and a flag that
1609 indicates whether or not the inline asm expression has side effects. An example
1610 inline assembler expression is:
1611 </p>
1614 <pre>
1616 </pre>
1617 </div>
1619 <p>
1622 </p>
1625 <pre>
1627 </pre>
1628 </div>
1630 <p>
1631 Inline asms with side effects not visible in the constraint list must be marked
1632 as having side effects. This is done through the use of the
1634 </p>
1637 <pre>
1639 </pre>
1640 </div>
1642 <p>TODO: The format of the asm and constraints string still need to be
1643 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1644 need to be documented).
1645 </p>
1647 </div>
1649 <!-- *********************************************************************** -->
1651 <!-- *********************************************************************** -->
1655 <p>The LLVM instruction set consists of several different
1662 </div>
1664 <!-- ======================================================================= -->
1666 Instructions</a> </div>
1671 basic block in a program ends with a "Terminator" instruction, which
1672 indicates which block should be executed after the current block is
1674 value: they produce control flow, not values (the one exception being
1683 </div>
1685 <!-- _______________________________________________________________________ -->
1687 Instruction</a> </div>
1692 </pre>
1695 value) from a function back to the caller.</p>
1697 returns a value and then causes control flow, and one that just causes
1698 control flow to occur.</p>
1703 instruction inside of the function that returns a value that does not
1704 match the return type of the function.</p>
1707 returns back to the calling function's context. If the caller is a "<a
1709 the instruction after the call. If the caller was an "<a
1711 at the beginning of the "normal" destination block. If the instruction
1712 returns a value, that value shall set the call or invoke instruction's
1713 return value.</p>
1717 </pre>
1718 </div>
1719 <!-- _______________________________________________________________________ -->
1723 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1724 </pre>
1727 transfer to a different basic block in the current function. There are
1728 two forms of this instruction, corresponding to a conditional branch
1729 and an unconditional branch.</p>
1741 <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
1743 </div>
1744 <!-- _______________________________________________________________________ -->
1747 </div>
1752 <pre>
1753 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1754 </pre>
1760 instruction, allowing a branch to occur to one of many possible
1761 destinations.</p>
1769 table is not allowed to contain duplicate constant entries.</p>
1775 table is searched for the given value. If the value is found, control flow is
1776 transfered to the corresponding destination; otherwise, control flow is
1777 transfered to the default destination.</p>
1781 <p>Depending on properties of the target machine and the particular
1783 ways. For example, it could be generated as a series of chained conditional
1784 branches or with a lookup table.</p>
1788 <pre>
1791 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1794 switch i32 0, label %dest [ ]
1797 switch i32 %val, label %otherwise [ i32 0, label %onzero
1798 i32 1, label %onone
1799 i32 2, label %ontwo ]
1800 </pre>
1801 </div>
1803 <!-- _______________________________________________________________________ -->
1806 </div>
1812 <pre>
1813 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1815 </pre>
1820 function, with the possibility of control flow transfer to either the
1832 <ol>
1833 <li>
1835 convention</a> the call should use. If none is specified, the call defaults
1836 to using C calling conventions.
1837 </li>
1839 function value being invoked. In most cases, this is a direct function
1841 an arbitrary pointer to function value.
1842 </li>
1845 function to be invoked. </li>
1848 signature argument types. If the function signature indicates the function
1849 accepts a variable number of arguments, the extra arguments can be
1850 specified. </li>
1858 </ol>
1864 difference is that it establishes an association with a label, which is used by
1865 the runtime library to unwind the stack.</p>
1867 <p>This instruction is used in languages with destructors to ensure that proper
1869 exception. Additionally, this is important for implementation of
1873 <pre>
1874 %retval = invoke i32 %Test(i32 15) to label %Continue
1878 </pre>
1879 </div>
1882 <!-- _______________________________________________________________________ -->
1885 Instruction</a> </div>
1890 <pre>
1891 unwind
1892 </pre>
1897 at the first callee in the dynamic call stack which used an <a
1899 primarily used to implement exception handling.</p>
1904 immediately halt. The dynamic call stack is then searched for the first <a
1906 execution continues at the "exceptional" destination block specified by the
1908 dynamic call chain, undefined behavior results.</p>
1909 </div>
1911 <!-- _______________________________________________________________________ -->
1914 Instruction</a> </div>
1919 <pre>
1920 unreachable
1921 </pre>
1926 instruction is used to inform the optimizer that a particular portion of the
1927 code is not reachable. This can be used to indicate that the code after a
1928 no-return function cannot be reached, and other facts.</p>
1933 </div>
1937 <!-- ======================================================================= -->
1940 <p>Binary operators are used to do most of the computation in a
1941 program. They require two operands, execute an operation on them, and
1942 produce a single value. The operands might represent
1944 The result value of a binary operator is not
1945 necessarily the same type as its operands.</p>
1947 </div>
1948 <!-- _______________________________________________________________________ -->
1950 Instruction</a> </div>
1954 </pre>
1961 Both arguments must have identical types.</p>
1963 <p>The value produced is the integer or floating point sum of the two
1964 operands.</p>
1967 </pre>
1968 </div>
1969 <!-- _______________________________________________________________________ -->
1971 Instruction</a> </div>
1975 </pre>
1978 operands.</p>
1980 instruction present in most other intermediate representations.</p>
1984 values.
1986 Both arguments must have identical types.</p>
1988 <p>The value produced is the integer or floating point difference of
1989 the two operands.</p>
1991 <pre>
1994 </pre>
1995 </div>
1996 <!-- _______________________________________________________________________ -->
1998 Instruction</a> </div>
2002 </pre>
2005 operands.</p>
2009 values.
2011 Both arguments must have identical types.</p>
2013 <p>The value produced is the integer or floating point product of the
2014 two operands.</p>
2015 <p>Because the operands are the same width, the result of an integer
2016 multiplication is the same whether the operands should be deemed unsigned or
2017 signed.</p>
2020 </pre>
2021 </div>
2022 <!-- _______________________________________________________________________ -->
2024 </a></div>
2028 </pre>
2031 operands.</p>
2036 of the values in which case the elements must be integers.</p>
2038 <p>The value produced is the unsigned integer quotient of the two operands. This
2039 instruction always performs an unsigned division operation, regardless of
2040 whether the arguments are unsigned or not.</p>
2043 </pre>
2044 </div>
2045 <!-- _______________________________________________________________________ -->
2047 </a> </div>
2051 </pre>
2054 operands.</p>
2059 of the values in which case the elements must be integers.</p>
2061 <p>The value produced is the signed integer quotient of the two operands. This
2062 instruction always performs a signed division operation, regardless of whether
2063 the arguments are signed or not.</p>
2066 </pre>
2067 </div>
2068 <!-- _______________________________________________________________________ -->
2070 Instruction</a> </div>
2074 </pre>
2077 operands.</p>
2082 versions of floating point values.</p>
2087 </pre>
2088 </div>
2089 <!-- _______________________________________________________________________ -->
2091 </div>
2095 </pre>
2098 unsigned division of its two arguments.</p>
2102 types.</p>
2105 This instruction always performs an unsigned division to get the remainder,
2106 regardless of whether the arguments are unsigned or not.</p>
2109 </pre>
2111 </div>
2112 <!-- _______________________________________________________________________ -->
2114 Instruction</a> </div>
2118 </pre>
2121 signed division of its two operands.</p>
2125 types.</p>
2130 a value. For more information about the difference, see <a
2132 Math Forum</a>. For a table of how this is implemented in various languages,
2137 </pre>
2139 </div>
2140 <!-- _______________________________________________________________________ -->
2142 Instruction</a> </div>
2146 </pre>
2149 division of its two operands.</p>
2153 identical types.</p>
2158 </pre>
2159 </div>
2161 <!-- ======================================================================= -->
2163 Operations</a> </div>
2165 <p>Bitwise binary operators are used to do various forms of
2166 bit-twiddling in a program. They are generally very efficient
2167 instructions and can commonly be strength reduced from other
2168 instructions. They require two operands, execute an operation on them,
2169 and produce a single value. The resulting value of the bitwise binary
2170 operators is always the same type as its first operand.</p>
2171 </div>
2173 <!-- _______________________________________________________________________ -->
2175 Instruction</a> </div>
2179 </pre>
2182 the left a specified number of bits.</p>
2192 </pre>
2193 </div>
2194 <!-- _______________________________________________________________________ -->
2196 Instruction</a> </div>
2200 </pre>
2204 operand shifted to the right a specified number of bits with zero fill.</p>
2211 <p>This instruction always performs a logical shift right operation. The most
2212 significant bits of the result will be filled with zero bits after the
2213 shift.</p>
2216 <pre>
2221 </pre>
2222 </div>
2224 <!-- _______________________________________________________________________ -->
2226 Instruction</a> </div>
2231 </pre>
2235 operand shifted to the right a specified number of bits with sign extension.</p>
2242 <p>This instruction always performs an arithmetic shift right operation,
2243 The most significant bits of the result will be filled with the sign bit
2247 <pre>
2252 </pre>
2253 </div>
2255 <!-- _______________________________________________________________________ -->
2257 Instruction</a> </div>
2261 </pre>
2264 its two operands.</p>
2268 identical types.</p>
2271 <p> </p>
2274 <tbody>
2275 <tr>
2279 </tr>
2280 <tr>
2284 </tr>
2285 <tr>
2289 </tr>
2290 <tr>
2294 </tr>
2295 <tr>
2299 </tr>
2300 </tbody>
2301 </table>
2302 </div>
2307 </pre>
2308 </div>
2309 <!-- _______________________________________________________________________ -->
2314 </pre>
2317 or of its two operands.</p>
2321 identical types.</p>
2324 <p> </p>
2327 <tbody>
2328 <tr>
2332 </tr>
2333 <tr>
2337 </tr>
2338 <tr>
2342 </tr>
2343 <tr>
2347 </tr>
2348 <tr>
2352 </tr>
2353 </tbody>
2354 </table>
2355 </div>
2360 </pre>
2361 </div>
2362 <!-- _______________________________________________________________________ -->
2364 Instruction</a> </div>
2368 </pre>
2376 identical types.</p>
2379 <p> </p>
2382 <tbody>
2383 <tr>
2387 </tr>
2388 <tr>
2392 </tr>
2393 <tr>
2397 </tr>
2398 <tr>
2402 </tr>
2403 <tr>
2407 </tr>
2408 </tbody>
2409 </table>
2410 </div>
2411 <p> </p>
2417 </pre>
2418 </div>
2420 <!-- ======================================================================= -->
2423 </div>
2427 <p>LLVM supports several instructions to represent vector operations in a
2428 target-independent manner. These instructions cover the element-access and
2429 vector-specific operations needed to process vectors effectively. While LLVM
2430 does directly support these vector operations, many sophisticated algorithms
2431 will want to use target-specific intrinsics to take full advantage of a specific
2432 target.</p>
2434 </div>
2436 <!-- _______________________________________________________________________ -->
2439 </div>
2445 <pre>
2446 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2447 </pre>
2451 <p>
2453 element from a vector at a specified index.
2454 </p>
2459 <p>
2462 an index indicating the position from which to extract the element.
2463 The index may be a variable.</p>
2467 <p>
2468 The result is a scalar of the same type as the element type of
2471 results are undefined.
2472 </p>
2476 <pre>
2478 </pre>
2479 </div>
2482 <!-- _______________________________________________________________________ -->
2485 </div>
2491 <pre>
2492 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2493 </pre>
2497 <p>
2499 element into a vector at a specified index.
2500 </p>
2505 <p>
2508 scalar value whose type must equal the element type of the first
2509 operand. The third operand is an index indicating the position at
2510 which to insert the value. The index may be a variable.</p>
2514 <p>
2519 </p>
2523 <pre>
2525 </pre>
2526 </div>
2528 <!-- _______________________________________________________________________ -->
2531 </div>
2537 <pre>
2538 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2539 </pre>
2543 <p>
2545 from two input vectors, returning a vector of the same type.
2546 </p>
2550 <p>
2552 with types that match each other and types that match the result of the
2553 instruction. The third argument is a shuffle mask, which has the same number
2554 of elements as the other vector type, but whose element type is always 'i32'.
2555 </p>
2557 <p>
2558 The shuffle mask operand is required to be a constant vector with either
2559 constant integer or undef values.
2560 </p>
2564 <p>
2565 The elements of the two input vectors are numbered from left to right across
2566 both of the vectors. The shuffle mask operand specifies, for each element of
2567 the result vector, which element of the two input registers the result element
2568 gets. The element selector may be undef (meaning "don't care") and the second
2569 operand may be undef if performing a shuffle from only one vector.
2570 </p>
2574 <pre>
2578 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2579 </pre>
2580 </div>
2583 <!-- ======================================================================= -->
2586 </div>
2590 <p>A key design point of an SSA-based representation is how it
2591 represents memory. In LLVM, no memory locations are in SSA form, which
2592 makes things very simple. This section describes how to read, write,
2593 allocate, and free memory in LLVM.</p>
2595 </div>
2597 <!-- _______________________________________________________________________ -->
2600 </div>
2606 <pre>
2607 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2608 </pre>
2613 heap and returns a pointer to it.</p>
2619 bytes of memory from the operating system and returns a pointer of the
2620 appropriate type to the program. If "NumElements" is specified, it is the
2621 number of elements allocated. If an alignment is specified, the value result
2622 of the allocation is guaranteed to be aligned to at least that boundary. If
2623 not specified, or if zero, the target can choose to align the allocation on any
2624 convenient boundary.</p>
2631 a pointer is returned.</p>
2635 <pre>
2643 </pre>
2644 </div>
2646 <!-- _______________________________________________________________________ -->
2649 </div>
2655 <pre>
2657 </pre>
2662 memory heap to be reallocated in the future.</p>
2668 instruction.</p>
2672 <p>Access to the memory pointed to by the pointer is no longer defined
2673 after this instruction executes.</p>
2677 <pre>
2679 free [4 x i8]* %array
2680 </pre>
2681 </div>
2683 <!-- _______________________________________________________________________ -->
2686 </div>
2692 <pre>
2693 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2694 </pre>
2699 currently executing function, to be automatically released when this function
2700 returns to its caller.</p>
2705 bytes of memory on the runtime stack, returning a pointer of the
2706 appropriate type to the program. If "NumElements" is specified, it is the
2707 number of elements allocated. If an alignment is specified, the value result
2708 of the allocation is guaranteed to be aligned to at least that boundary. If
2709 not specified, or if zero, the target can choose to align the allocation on any
2710 convenient boundary.</p>
2718 instruction is commonly used to represent automatic variables that must
2719 have an address available. When the function returns (either with the <tt><a
2721 instructions), the memory is reclaimed.</p>
2725 <pre>
2730 </pre>
2731 </div>
2733 <!-- _______________________________________________________________________ -->
2735 Instruction</a> </div>
2738 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
2743 address from which to load. The pointer must point to a <a
2748 instructions. </p>
2753 <a
2756 </pre>
2757 </div>
2758 <!-- _______________________________________________________________________ -->
2760 Instruction</a> </div>
2763 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2764 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2765 </pre>
2773 optimizer is not allowed to modify the number or order of execution of
2781 <a
2784 </pre>
2785 </div>
2787 <!-- _______________________________________________________________________ -->
2790 </div>
2794 <pre>
2796 </pre>
2800 <p>
2802 subelement of an aggregate data structure.</p>
2806 <p>This instruction takes a list of integer operands that indicate what
2807 elements of the aggregate object to index to. The actual types of the arguments
2808 provided depend on the type of the first pointer argument. The
2810 levels of a structure or to a specific index in an array. When indexing into a
2815 <p>For example, let's consider a C code fragment and how it gets
2816 compiled to LLVM:</p>
2819 <pre>
2820 struct RT {
2821 char A;
2823 char C;
2824 };
2825 struct ST {
2826 int X;
2827 double Y;
2828 struct RT Z;
2829 };
2831 int *foo(struct ST *s) {
2833 }
2834 </pre>
2835 </div>
2840 <pre>
2842 %ST = type { i32, double, %RT }
2844 define i32* %foo(%ST* %s) {
2845 entry:
2847 ret i32* %reg
2848 }
2849 </pre>
2850 </div>
2863 }</tt>' type, a structure. The second index indexes into the third element of
2865 i8 }</tt>' type, another structure. The third index indexes into the second
2867 array. The two dimensions of the array are subscripted into, yielding an
2871 <p>Note that it is perfectly legal to index partially through a
2872 structure, returning a pointer to an inner element. Because of this,
2873 the LLVM code for the given testcase is equivalent to:</p>
2875 <pre>
2876 define i32* %foo(%ST* %s) {
2882 ret i32* %t5
2883 }
2884 </pre>
2886 <p>Note that it is undefined to access an array out of bounds: array and
2887 pointer indexes must always be within the defined bounds of the array type.
2888 The one exception for this rules is zero length arrays. These arrays are
2889 defined to be accessible as variable length arrays, which requires access
2890 beyond the zero'th element.</p>
2892 <p>The getelementptr instruction is often confusing. For some more insight
2898 <pre>
2901 </pre>
2902 </div>
2904 <!-- ======================================================================= -->
2906 </div>
2908 <p>The instructions in this category are the conversion instructions (casting)
2909 which all take a single operand and a type. They perform various bit conversions
2910 on the operand.</p>
2911 </div>
2913 <!-- _______________________________________________________________________ -->
2916 </div>
2920 <pre>
2922 </pre>
2925 <p>
2927 </p>
2930 <p>
2938 <p>
2942 It will always truncate bits.</p>
2945 <pre>
2949 </pre>
2950 </div>
2952 <!-- _______________________________________________________________________ -->
2955 </div>
2959 <pre>
2961 </pre>
2982 <pre>
2985 </pre>
2986 </div>
2988 <!-- _______________________________________________________________________ -->
2991 </div>
2995 <pre>
2997 </pre>
3003 <p>
3011 <p>
3019 <pre>
3022 </pre>
3023 </div>
3025 <!-- _______________________________________________________________________ -->
3028 </div>
3034 <pre>
3036 </pre>
3057 <pre>
3060 </pre>
3061 </div>
3063 <!-- _______________________________________________________________________ -->
3066 </div>
3070 <pre>
3072 </pre>
3076 floating point value.</p>
3082 type must be smaller than the destination type.</p>
3092 <pre>
3095 </pre>
3096 </div>
3098 <!-- _______________________________________________________________________ -->
3101 </div>
3105 <pre>
3107 </pre>
3112 </p>
3123 the results are undefined.</p>
3125 <p>When converting to i1, the conversion is done as a comparison against
3130 <pre>
3134 </pre>
3135 </div>
3137 <!-- _______________________________________________________________________ -->
3140 </div>
3144 <pre>
3146 </pre>
3151 </p>
3163 the results are undefined.</p>
3165 <p>When converting to i1, the conversion is done as a comparison against
3170 <pre>
3174 </pre>
3175 </div>
3177 <!-- _______________________________________________________________________ -->
3180 </div>
3184 <pre>
3186 </pre>
3200 integer quantity and converts it to the corresponding floating point value. If
3201 the value cannot fit in the floating point value, the results are undefined.</p>
3205 <pre>
3208 </pre>
3209 </div>
3211 <!-- _______________________________________________________________________ -->
3214 </div>
3218 <pre>
3220 </pre>
3233 integer quantity and converts it to the corresponding floating point value. If
3234 the value cannot fit in the floating point value, the results are undefined.</p>
3237 <pre>
3240 </pre>
3241 </div>
3243 <!-- _______________________________________________________________________ -->
3246 </div>
3250 <pre>
3252 </pre>
3266 truncating or zero extending that value to the size of the integer type. If
3270 change.</p>
3273 <pre>
3276 </pre>
3277 </div>
3279 <!-- _______________________________________________________________________ -->
3282 </div>
3286 <pre>
3288 </pre>
3296 value to cast, and a type to cast it to, which must be a
3304 the size of a pointer then a zero extension is done. If they are the same size,
3308 <pre>
3312 </pre>
3313 </div>
3315 <!-- _______________________________________________________________________ -->
3318 </div>
3322 <pre>
3324 </pre>
3332 a first class value, and a type to cast it to, which must also be a <a
3335 type is a pointer, the destination type must also be a pointer.</p>
3342 converted to other pointer types with this instruction. To convert pointers to
3347 <pre>
3351 </pre>
3352 </div>
3354 <!-- ======================================================================= -->
3358 instructions, which defy better classification.</p>
3359 </div>
3361 <!-- _______________________________________________________________________ -->
3363 </div>
3366 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3367 </pre>
3370 of its two integer operands.</p>
3373 the condition code indicating the kind of comparison to perform. It is not
3374 a value, just a keyword. The possible condition code are:
3375 <ol>
3386 </ol>
3393 <ol>
3396 </li>
3415 </ol>
3417 values are compared as if they were integers.</p>
3426 </pre>
3427 </div>
3429 <!-- _______________________________________________________________________ -->
3431 </div>
3434 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3435 </pre>
3438 of its floating point operands.</p>
3441 the condition code indicating the kind of comparison to perform. It is not
3442 a value, just a keyword. The possible condition code are:
3443 <ol>
3460 </ol>
3465 types.</p>
3470 <ol>
3499 </ol>
3506 </pre>
3507 </div>
3509 <!-- _______________________________________________________________________ -->
3511 Instruction</a> </div>
3517 the SSA graph representing the function.</p>
3519 <p>The type of the incoming values is specified with the first type
3521 as arguments, with one pair for each predecessor basic block of the
3523 type may be used as the value arguments to the PHI node. Only labels
3524 may be used as the label arguments.</p>
3525 <p>There must be no non-phi instructions between the start of a basic
3526 block and the PHI instructions: i.e. PHI instructions must be first in
3527 a basic block.</p>
3530 specified by the pair corresponding to the predecessor basic block that executed
3531 just prior to the current block.</p>
3533 <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>
3534 </div>
3536 <!-- _______________________________________________________________________ -->
3539 </div>
3545 <pre>
3546 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3547 </pre>
3551 <p>
3553 condition, without branching.
3554 </p>
3559 <p>
3560 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.
3561 </p>
3565 <p>
3566 If the boolean condition evaluates to true, the instruction returns the first
3567 value argument; otherwise, it returns the second value argument.
3568 </p>
3572 <pre>
3574 </pre>
3575 </div>
3578 <!-- _______________________________________________________________________ -->
3581 </div>
3586 <pre>
3587 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
3588 </pre>
3598 <ol>
3599 <li>
3601 any allocas or varargs in the caller. If the "tail" marker is present, the
3602 function call is eligible for tail call optimization. Note that calls may
3605 </li>
3606 <li>
3608 convention</a> the call should use. If none is specified, the call defaults
3609 to using C calling conventions.
3610 </li>
3611 <li>
3613 the type of the return value. Functions that return no value are marked
3615 </li>
3616 <li>
3618 value being invoked. The argument types must match the types implied by
3619 this signature. This type can be omitted if the function is not varargs
3620 and if the function type does not return a pointer to a function.</p>
3621 </li>
3622 <li>
3624 be invoked. In most cases, this is a direct function invocation, but
3626 to function value.</p>
3627 </li>
3628 <li>
3630 function signature argument types. All arguments must be of
3632 indicates the function accepts a variable number of arguments, the extra
3633 arguments can be specified.</p>
3634 </li>
3635 </ol>
3640 transfer to a specified function, with its incoming arguments bound to
3642 instruction in the called function, control flow continues with the
3643 instruction after the function call, and the return value of the
3644 function is bound to the result argument. This is a simpler case of
3649 <pre>
3650 %retval = call i32 @test(i32 %argc)
3652 %X = tail call i32 @foo()
3654 %Z = call void %foo(i8 97 signext)
3655 </pre>
3657 </div>
3659 <!-- _______________________________________________________________________ -->
3662 </div>
3668 <pre>
3670 </pre>
3675 the "variable argument" area of a function call. It is used to implement the
3681 the argument. It returns a value of the specified argument type and
3693 <p>It is legal for this instruction to be called in a function which does not
3695 function.</p>
3699 argument.</p>
3705 </div>
3707 <!-- *********************************************************************** -->
3709 <!-- *********************************************************************** -->
3714 well known names and semantics and are required to follow certain restrictions.
3715 Overall, these intrinsics represent an extension mechanism for the LLVM
3716 language that does not require changing all of the transformations in LLVM when
3717 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
3720 prefix is reserved in LLVM for intrinsic names; thus, function names may not
3721 begin with this prefix. Intrinsic functions must always be external functions:
3722 you cannot define the body of intrinsic functions. Intrinsic functions may
3723 only be used in call or invoke instructions: it is illegal to take the address
3724 of an intrinsic function. Additionally, because intrinsic functions are part
3725 of the LLVM language, it is required if any are added that they be documented
3726 here.</p>
3728 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3729 a family of functions that perform the same operation but on different data
3730 types. Because LLVM can represent over 8 million different integer types,
3731 overloading is used commonly to allow an intrinsic function to operate on any
3732 integer type. One or more of the argument types or the result type can be
3733 overloaded to accept any integer type. Argument types may also be defined as
3734 exactly matching a previous argument's type or the result type. This allows an
3735 intrinsic function which accepts multiple arguments, but needs all of them to
3736 be of the same type, to only be overloaded with respect to a single argument or
3737 the result.</p>
3739 <p>Overloaded intrinsics will have the names of its overloaded argument types
3740 encoded into its function name, each preceded by a period. Only those types
3741 which are overloaded result in a name suffix. Arguments whose type is matched
3743 take an integer of any width and returns an integer of exactly the same integer
3744 width. This leads to a family of functions such as
3746 Only one type, the return type, is overloaded, and only one type suffix is
3747 required. Because the argument's type is matched against the return type, it
3748 does not require its own name suffix.</p>
3750 <p>To learn how to add an intrinsic function, please see the
3752 </p>
3754 </div>
3756 <!-- ======================================================================= -->
3759 </div>
3765 intrinsic functions. These functions are related to the similarly
3768 <p>All of these functions operate on arguments that use a
3769 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3770 language reference manual does not define what this type is, so all
3771 transformations should be prepared to handle these functions regardless of
3772 the type used.</p>
3775 instruction and the variable argument handling intrinsic functions are
3776 used.</p>
3779 <pre>
3780 define i32 @test(i32 %X, ...) {
3781 ; Initialize variable argument processing
3782 %ap = alloca i8*
3783 %ap2 = bitcast i8** %ap to i8*
3784 call void @llvm.va_start(i8* %ap2)
3786 ; Read a single integer argument
3787 %tmp = va_arg i8** %ap, i32
3789 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3790 %aq = alloca i8*
3791 %aq2 = bitcast i8** %aq to i8*
3792 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
3793 call void @llvm.va_end(i8* %aq2)
3795 ; Stop processing of arguments.
3796 call void @llvm.va_end(i8* %ap2)
3797 ret i32 %tmp
3798 }
3800 declare void @llvm.va_start(i8*)
3801 declare void @llvm.va_copy(i8*, i8*)
3802 declare void @llvm.va_end(i8*)
3803 </pre>
3804 </div>
3806 </div>
3808 <!-- _______________________________________________________________________ -->
3811 </div>
3829 macro available in C. In a target-dependent way, it initializes the
3833 last argument of the function as the compiler can figure that out.</p>
3835 </div>
3837 <!-- _______________________________________________________________________ -->
3840 </div>
3858 macro available in C. In a target-dependent way, it destroys the
3864 </div>
3866 <!-- _______________________________________________________________________ -->
3869 </div>
3875 <pre>
3877 </pre>
3882 from the source argument list to the destination argument list.</p>
3893 macro available in C. In a target-dependent way, it copies the source
3896 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
3897 example, memory allocation.</p>
3899 </div>
3901 <!-- ======================================================================= -->
3904 </div>
3908 <p>
3910 Collection</a> requires the implementation and generation of these intrinsics.
3914 Front-ends for type-safe garbage collected languages should generate these
3915 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
3917 </p>
3918 </div>
3920 <!-- _______________________________________________________________________ -->
3923 </div>
3929 <pre>
3930 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
3931 </pre>
3936 the code generator, and allows some metadata to be associated with it.</p>
3940 <p>The first argument specifies the address of a stack object that contains the
3941 root pointer. The second pointer (which must be either a constant or a global
3942 value address) contains the meta-data to be associated with the root.</p>
3947 location. At compile-time, the code generator generates information to allow
3948 the runtime to find the pointer at GC safe points.
3949 </p>
3951 </div>
3954 <!-- _______________________________________________________________________ -->
3957 </div>
3963 <pre>
3964 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
3965 </pre>
3970 locations, allowing garbage collector implementations that require read
3971 barriers.</p>
3975 <p>The second argument is the address to read from, which should be an address
3976 allocated from the garbage collector. The first object is a pointer to the
3977 start of the referenced object, if needed by the language runtime (otherwise
3978 null).</p>
3983 instruction, but may be replaced with substantially more complex code by the
3984 garbage collector runtime, as needed.</p>
3986 </div>
3989 <!-- _______________________________________________________________________ -->
3992 </div>
3998 <pre>
3999 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4000 </pre>
4005 locations, allowing garbage collector implementations that require write
4006 barriers (such as generational or reference counting collectors).</p>
4010 <p>The first argument is the reference to store, the second is the start of the
4011 object to store it to, and the third is the address of the field of Obj to
4012 store to. If the runtime does not require a pointer to the object, Obj may be
4013 null.</p>
4018 instruction, but may be replaced with substantially more complex code by the
4019 garbage collector runtime, as needed.</p>
4021 </div>
4025 <!-- ======================================================================= -->
4028 </div>
4031 <p>
4032 These intrinsics are provided by LLVM to expose special features that may only
4033 be implemented with code generator support.
4034 </p>
4036 </div>
4038 <!-- _______________________________________________________________________ -->
4041 </div>
4046 <pre>
4048 </pre>
4052 <p>
4054 target-specific value indicating the return address of the current function
4055 or one of its callers.
4056 </p>
4060 <p>
4061 The argument to this intrinsic indicates which function to return the address
4062 for. Zero indicates the calling function, one indicates its caller, etc. The
4064 </p>
4068 <p>
4070 the return address of the specified call frame, or zero if it cannot be
4071 identified. The value returned by this intrinsic is likely to be incorrect or 0
4072 for arguments other than zero, so it should only be used for debugging purposes.
4073 </p>
4075 <p>
4076 Note that calling this intrinsic does not prevent function inlining or other
4077 aggressive transformations, so the value returned may not be that of the obvious
4078 source-language caller.
4079 </p>
4080 </div>
4083 <!-- _______________________________________________________________________ -->
4086 </div>
4091 <pre>
4093 </pre>
4097 <p>
4099 target-specific frame pointer value for the specified stack frame.
4100 </p>
4104 <p>
4105 The argument to this intrinsic indicates which function to return the frame
4106 pointer for. Zero indicates the calling function, one indicates its caller,
4108 </p>
4112 <p>
4114 the frame address of the specified call frame, or zero if it cannot be
4115 identified. The value returned by this intrinsic is likely to be incorrect or 0
4116 for arguments other than zero, so it should only be used for debugging purposes.
4117 </p>
4119 <p>
4120 Note that calling this intrinsic does not prevent function inlining or other
4121 aggressive transformations, so the value returned may not be that of the obvious
4122 source-language caller.
4123 </p>
4124 </div>
4126 <!-- _______________________________________________________________________ -->
4129 </div>
4134 <pre>
4135 declare i8 *@llvm.stacksave()
4136 </pre>
4140 <p>
4144 features like scoped automatic variable sized arrays in C99.
4145 </p>
4149 <p>
4150 This intrinsic returns a opaque pointer value that can be passed to <a
4157 </p>
4159 </div>
4161 <!-- _______________________________________________________________________ -->
4164 </div>
4169 <pre>
4170 declare void @llvm.stackrestore(i8 * %ptr)
4171 </pre>
4175 <p>
4177 the function stack to the state it was in when the corresponding <a
4179 useful for implementing language features like scoped automatic variable sized
4180 arrays in C99.
4181 </p>
4185 <p>
4187 </p>
4189 </div>
4192 <!-- _______________________________________________________________________ -->
4195 </div>
4200 <pre>
4202 </pre>
4207 <p>
4209 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4210 no
4211 effect on the behavior of the program but can change its performance
4212 characteristics.
4213 </p>
4217 <p>
4223 </p>
4227 <p>
4228 This intrinsic does not modify the behavior of the program. In particular,
4229 prefetches cannot trap and do not produce a value. On targets that support this
4230 intrinsic, the prefetch can provide hints to the processor cache for better
4231 performance.
4232 </p>
4234 </div>
4236 <!-- _______________________________________________________________________ -->
4239 </div>
4244 <pre>
4246 </pre>
4251 <p>
4253 (PC) in a region of
4254 code to simulators and other tools. The method is target specific, but it is
4255 expected that the marker will use exported symbols to transmit the PC of the marker.
4256 The marker makes no guarantees that it will remain with any specific instruction
4257 after optimizations. It is possible that the presence of a marker will inhibit
4258 optimizations. The intended use is to be inserted after optimizations to allow
4259 correlations of simulation runs.
4260 </p>
4264 <p>
4266 </p>
4270 <p>
4271 This intrinsic does not modify the behavior of the program. Backends that do not
4272 support this intrinisic may ignore it.
4273 </p>
4275 </div>
4277 <!-- _______________________________________________________________________ -->
4280 </div>
4285 <pre>
4286 declare i64 @llvm.readcyclecounter( )
4287 </pre>
4292 <p>
4294 counter register (or similar low latency, high accuracy clocks) on those targets
4295 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4296 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4297 should only be used for small timings.
4298 </p>
4302 <p>
4303 When directly supported, reading the cycle counter should not modify any memory.
4304 Implementations are allowed to either return a application specific value or a
4305 system wide value. On backends without support, this is lowered to a constant 0.
4306 </p>
4308 </div>
4310 <!-- ======================================================================= -->
4313 </div>
4316 <p>
4317 LLVM provides intrinsics for a few important standard C library functions.
4318 These intrinsics allow source-language front-ends to pass information about the
4319 alignment of the pointer arguments to the code generator, providing opportunity
4320 for more efficient code generation.
4321 </p>
4323 </div>
4325 <!-- _______________________________________________________________________ -->
4328 </div>
4333 <pre>
4338 </pre>
4342 <p>
4344 location to the destination location.
4345 </p>
4347 <p>
4349 intrinsics do not return a value, and takes an extra alignment argument.
4350 </p>
4354 <p>
4355 The first argument is a pointer to the destination, the second is a pointer to
4356 the source. The third argument is an integer argument
4357 specifying the number of bytes to copy, and the fourth argument is the alignment
4358 of the source and destination locations.
4359 </p>
4361 <p>
4363 the caller guarantees that both the source and destination pointers are aligned
4364 to that boundary.
4365 </p>
4369 <p>
4371 location to the destination location, which are not allowed to overlap. It
4372 copies "len" bytes of memory over. If the argument is known to be aligned to
4373 some boundary, this can be specified as the fourth argument, otherwise it should
4375 </p>
4376 </div>
4379 <!-- _______________________________________________________________________ -->
4382 </div>
4387 <pre>
4392 </pre>
4396 <p>
4398 location to the destination location. It is similar to the
4400 </p>
4402 <p>
4404 intrinsics do not return a value, and takes an extra alignment argument.
4405 </p>
4409 <p>
4410 The first argument is a pointer to the destination, the second is a pointer to
4411 the source. The third argument is an integer argument
4412 specifying the number of bytes to copy, and the fourth argument is the alignment
4413 of the source and destination locations.
4414 </p>
4416 <p>
4418 the caller guarantees that the source and destination pointers are aligned to
4419 that boundary.
4420 </p>
4424 <p>
4426 location to the destination location, which may overlap. It
4427 copies "len" bytes of memory over. If the argument is known to be aligned to
4428 some boundary, this can be specified as the fourth argument, otherwise it should
4430 </p>
4431 </div>
4434 <!-- _______________________________________________________________________ -->
4437 </div>
4442 <pre>
4447 </pre>
4451 <p>
4453 byte value.
4454 </p>
4456 <p>
4458 does not return a value, and takes an extra alignment argument.
4459 </p>
4463 <p>
4464 The first argument is a pointer to the destination to fill, the second is the
4465 byte value to fill it with, the third argument is an integer
4466 argument specifying the number of bytes to fill, and the fourth argument is the
4467 known alignment of destination location.
4468 </p>
4470 <p>
4472 the caller guarantees that the destination pointer is aligned to that boundary.
4473 </p>
4477 <p>
4479 the
4480 destination location. If the argument is known to be aligned to some boundary,
4481 this can be specified as the fourth argument, otherwise it should be set to 0 or
4482 1.
4483 </p>
4484 </div>
4487 <!-- _______________________________________________________________________ -->
4490 </div>
4495 <pre>
4496 declare float @llvm.sqrt.f32(float %Val)
4497 declare double @llvm.sqrt.f64(double %Val)
4498 </pre>
4502 <p>
4506 negative numbers (which allows for better optimization).
4507 </p>
4511 <p>
4512 The argument and return value are floating point numbers of the same type.
4513 </p>
4517 <p>
4518 This function returns the sqrt of the specified operand if it is a nonnegative
4519 floating point number.
4520 </p>
4521 </div>
4523 <!-- _______________________________________________________________________ -->
4526 </div>
4531 <pre>
4532 declare float @llvm.powi.f32(float %Val, i32 %power)
4533 declare double @llvm.powi.f64(double %Val, i32 %power)
4534 </pre>
4538 <p>
4540 specified (positive or negative) power. The order of evaluation of
4541 multiplications is not defined.
4542 </p>
4546 <p>
4547 The second argument is an integer power, and the first is a value to raise to
4548 that power.
4549 </p>
4553 <p>
4554 This function returns the first value raised to the second power with an
4555 unspecified sequence of rounding operations.</p>
4556 </div>
4559 <!-- ======================================================================= -->
4562 </div>
4565 <p>
4566 LLVM provides intrinsics for a few important bit manipulation operations.
4567 These allow efficient code generation for some algorithms.
4568 </p>
4570 </div>
4572 <!-- _______________________________________________________________________ -->
4575 </div>
4580 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4582 <pre>
4586 </pre>
4590 <p>
4592 values with an even number of bytes (positive multiple of 16 bits). These are
4593 useful for performing operations on data that is not in the target's native
4594 byte order.
4595 </p>
4599 <p>
4602 intrinsic returns an i32 value that has the four bytes of the input i32
4607 </p>
4609 </div>
4611 <!-- _______________________________________________________________________ -->
4614 </div>
4619 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4620 width. Not all targets support all bit widths however.
4621 <pre>
4627 </pre>
4631 <p>
4633 value.
4634 </p>
4638 <p>
4639 The only argument is the value to be counted. The argument may be of any
4640 integer type. The return type must match the argument type.
4641 </p>
4645 <p>
4647 </p>
4648 </div>
4650 <!-- _______________________________________________________________________ -->
4653 </div>
4659 integer bit width. Not all targets support all bit widths however.
4660 <pre>
4666 </pre>
4670 <p>
4672 leading zeros in a variable.
4673 </p>
4677 <p>
4678 The only argument is the value to be counted. The argument may be of any
4679 integer type. The return type must match the argument type.
4680 </p>
4684 <p>
4686 in a variable. If the src == 0 then the result is the size in bits of the type
4688 </p>
4689 </div>
4693 <!-- _______________________________________________________________________ -->
4696 </div>
4702 integer bit width. Not all targets support all bit widths however.
4703 <pre>
4709 </pre>
4713 <p>
4715 trailing zeros.
4716 </p>
4720 <p>
4721 The only argument is the value to be counted. The argument may be of any
4722 integer type. The return type must match the argument type.
4723 </p>
4727 <p>
4729 in a variable. If the src == 0 then the result is the size in bits of the type
4731 </p>
4732 </div>
4734 <!-- _______________________________________________________________________ -->
4737 </div>
4743 on any integer bit width.
4744 <pre>
4745 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
4746 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
4747 </pre>
4751 range of bits from an integer value and returns them in the same bit width as
4752 the original value.</p>
4756 any bit width but they must have the same bit width. The second and third
4763 operates in forward mode.</p>
4767 <ol>
4771 to determine the number of bits to retain.</li>
4774 </ol>
4775 <p>In reverse mode, a similar computation is made except that the bits are
4780 </div>
4784 </div>
4790 on any integer bit width.
4791 <pre>
4792 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
4793 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
4794 </pre>
4798 of bits in an integer value with another integer value. It returns the integer
4799 with the replaced bits.</p>
4806 type since they specify only a bit index.</p>
4812 operates in forward mode.</p>
4814 truncating it down to the size of the replacement area or zero extending it
4815 up to that size.</p>
4820 <p>In reverse mode, a similar computation is made except that the bits are
4824 <pre>
4830 </pre>
4831 </div>
4833 <!-- ======================================================================= -->
4836 </div>
4839 <p>
4841 are described in the <a
4843 Debugging</a> document.
4844 </p>
4845 </div>
4848 <!-- ======================================================================= -->
4851 </div>
4854 <p> The LLVM exception handling intrinsics (which all start with
4858 </div>
4860 <!-- ======================================================================= -->
4863 </div>
4866 <p>
4867 These intrinsic functions expand the "universal IR" of LLVM to represent
4868 hardware constructs for atomic operations and memory synchronization. This
4869 provides an interface to the hardware, not an interface to the programmer. It
4870 is aimed at a low enough level to allow any programming models or APIs which
4871 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
4872 hardware behavior. Just as hardware provides a "universal IR" for source
4873 languages, it also provides a starting point for developing a "universal"
4874 atomic operation and synchronization IR.
4875 </p>
4876 <p>
4878 software transaction memory systems, atomic primitives, and intrinsic
4879 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
4880 application libraries. The hardware interface provided by LLVM should allow
4881 a clean implementation of all of these APIs and parallel programming models.
4882 No one model or paradigm should be selected above others unless the hardware
4883 itself ubiquitously does so.
4884 </p>
4885 </div>
4887 <!-- _______________________________________________________________________ -->
4890 </div>
4893 <p>
4895 integer bit width. Not all targets support all bit widths however.</p>
4896 <pre>
4898 declare i16 @llvm.atomic.lcs.i16.i16p.i16.i16( i16* <ptr>, i16 <cmp>, i16 <val> )
4899 declare i32 @llvm.atomic.lcs.i32.i32p.i32.i32( i32* <ptr>, i32 <cmp>, i32 <val> )
4900 declare i64 @llvm.atomic.lcs.i64.i64p.i64.i64( i64* <ptr>, i64 <cmp>, i64 <val> )
4901 </pre>
4903 <p>
4904 This loads a value in memory and compares it to a given value. If they are
4905 equal, it stores a new value into the memory.
4906 </p>
4908 <p>
4912 this integer type. While any bit width integer may be used, targets may only
4913 lower representations they support in hardware.
4914 </p>
4916 <p>
4917 This entire intrinsic must be executed atomically. It first loads the value
4920 loaded value is yielded in all cases. This provides the equivalent of an
4921 atomic compare-and-swap operation within the SSA framework.
4922 </p>
4924 <pre>
4925 %ptr = malloc i32
4926 store i32 4, %ptr
4929 %result1 = call i32 @llvm.atomic.lcs( i32* %ptr, i32 4, %val1 )
4935 %result2 = call i32 @llvm.atomic.lcs( i32* %ptr, i32 5, %val2 )
4939 </pre>
4940 </div>
4942 <!-- _______________________________________________________________________ -->
4945 </div>
4948 <p>
4950 integer bit width. Not all targets support all bit widths however.</p>
4951 <pre>
4956 </pre>
4958 <p>
4962 </p>
4964 <p>
4968 integer type. The targets may only lower integer representations they
4969 support.
4970 </p>
4972 <p>
4975 equivalent of an atomic swap operation within the SSA framework.
4976 </p>
4978 <pre>
4979 %ptr = malloc i32
4980 store i32 4, %ptr
4983 %result1 = call i32 @llvm.atomic.ls( i32* %ptr, i32 %val1 )
4989 %result2 = call i32 @llvm.atomic.ls( i32* %ptr, i32 %val2 )
4993 </pre>
4994 </div>
4996 <!-- _______________________________________________________________________ -->
4999 </div>
5002 <p>
5004 integer bit width. Not all targets support all bit widths however.</p>
5005 <pre>
5010 </pre>
5012 <p>
5015 </p>
5017 <p>
5018 The intrinsic takes two arguments, the first a pointer to an integer value
5019 and the second an integer value. The result is also an integer value. These
5020 integer types can have any bit width, but they must all have the same bit
5021 width. The targets may only lower integer representations they support.
5022 </p>
5024 <p>
5025 This intrinsic does a series of operations atomically. It first loads the
5028 </p>
5030 <pre>
5031 %ptr = malloc i32
5032 store i32 4, %ptr
5033 %result1 = call i32 @llvm.atomic.las( i32* %ptr, i32 4 )
5035 %result2 = call i32 @llvm.atomic.las( i32* %ptr, i32 2 )
5037 %result3 = call i32 @llvm.atomic.las( i32* %ptr, i32 5 )
5040 </pre>
5041 </div>
5043 <!-- _______________________________________________________________________ -->
5046 </div>
5049 <p>
5051 integer bit width. Not all targets support all bit widths however.</p>
5052 <pre>
5057 </pre>
5059 <p>
5062 </p>
5064 <p>
5065 The intrinsic takes two arguments, the first a pointer to an integer value
5066 and the second an integer value. The result is also an integer value. These
5067 integer types can have any bit width, but they must all have the same bit
5068 width. The targets may only lower integer representations they support.
5069 </p>
5071 <p>
5072 This intrinsic does a series of operations atomically. It first loads the
5076 </p>
5078 <pre>
5079 %ptr = malloc i32
5080 store i32 32, %ptr
5081 %result1 = call i32 @llvm.atomic.lss( i32* %ptr, i32 4 )
5083 %result2 = call i32 @llvm.atomic.lss( i32* %ptr, i32 2 )
5085 %result3 = call i32 @llvm.atomic.lss( i32* %ptr, i32 5 )
5088 </pre>
5089 </div>
5091 <!-- _______________________________________________________________________ -->
5094 </div>
5097 <pre>
5098 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss> )
5099 </pre>
5101 <p>
5103 specific pairs of memory access types.
5104 </p>
5106 <p>
5108 Each argument enables a specific barrier as listed below.
5109 </p>
5110 <ul>
5115 </ul>
5117 <p>
5118 This intrinsic causes the system to enforce some ordering constraints upon
5119 the loads and stores of the program. This barrier does not indicate
5121 which they occur. For any of the specified pairs of load and store operations
5122 (f.ex. load-load, or store-load), all of the first operations preceding the
5123 barrier will complete before any of the second operations succeeding the
5124 barrier begin. Specifically the semantics for each pairing is as follows:
5125 </p>
5126 <ul>
5128 after the barrier begins.</li>
5130 store after the barrier begins.</li>
5132 store after the barrier begins.</li>
5134 load after the barrier begins.</li>
5135 </ul>
5136 <p>
5137 These semantics are applied with a logical "and" behavior when more than one
5138 is enabled in a single memory barrier intrinsic.
5139 </p>
5141 <pre>
5142 %ptr = malloc i32
5143 store i32 4, %ptr
5146 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5149 </pre>
5150 </div>
5152 <!-- ======================================================================= -->
5155 </div>
5158 <p>
5159 This intrinsic makes it possible to excise one parameter, marked with
5161 function pointer lacking the nest parameter - the caller does not need
5162 to provide a value for it. Instead, the value to use is stored in
5163 advance in a "trampoline", a block of memory usually allocated
5164 on the stack, which also contains code to splice the nest value into the
5165 argument list. This is used to implement the GCC nested function address
5166 extension.
5167 </p>
5168 <p>
5169 For example, if the function is
5172 <pre>
5175 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5176 %fp = bitcast i8* %p to i32 (i32, i32)*
5177 </pre>
5180 </p>
5181 </div>
5183 <!-- _______________________________________________________________________ -->
5186 </div>
5189 <pre>
5191 </pre>
5193 <p>
5195 and returns a function pointer suitable for executing it.
5196 </p>
5198 <p>
5201 and sufficiently aligned block of memory; this memory is written to by the
5202 intrinsic. Note that the size and the alignment are target-specific - LLVM
5203 currently provides no portable way of determining them, so a front-end that
5204 generates this intrinsic needs to have some target-specific knowledge.
5206 </p>
5208 <p>
5210 dependent code, turning it into a function. A pointer to this function is
5211 returned, but needs to be bitcast to an
5213 before being called. The new function's signature is the same as that of
5216 of pointer type. Calling the new function is equivalent to calling
5220 modified, then the effect of any later call to the returned function pointer is
5221 undefined.
5222 </p>
5223 </div>
5225 <!-- ======================================================================= -->
5228 </div>
5231 <p> This class of intrinsics is designed to be generic and has
5232 no specific purpose. </p>
5233 </div>
5235 <!-- _______________________________________________________________________ -->
5238 </div>
5243 <pre>
5244 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5245 </pre>
5249 <p>
5251 </p>
5255 <p>
5256 The first argument is a pointer to a value, the second is a pointer to a
5257 global string, the third is a pointer to a global string which is the source
5258 file name, and the last argument is the line number.
5259 </p>
5263 <p>
5264 This intrinsic allows annotation of local variables with arbitrary strings.
5265 This can be useful for special purpose optimizations that want to look for these
5266 annotations. These have no other defined use, they are ignored by code
5267 generation and optimization.
5268 </div>
5270 <!-- _______________________________________________________________________ -->
5273 </div>
5279 any integer bit width.
5280 </p>
5281 <pre>
5282 declare i8 @llvm.annotation(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
5283 declare i16 @llvm.annotation(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
5284 declare i32 @llvm.annotation(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
5285 declare i64 @llvm.annotation(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
5286 declare i256 @llvm.annotation(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
5287 </pre>
5291 <p>
5293 </p>
5297 <p>
5298 The first argument is an integer value (result of some expression),
5299 the second is a pointer to a global string, the third is a pointer to a global
5300 string which is the source file name, and the last argument is the line number.
5301 It returns the value of the first argument.
5302 </p>
5306 <p>
5307 This intrinsic allows annotations to be put on arbitrary expressions
5308 with arbitrary strings. This can be useful for special purpose optimizations
5309 that want to look for these annotations. These have no other defined use, they
5310 are ignored by code generation and optimization.
5311 </div>
5313 <!-- *********************************************************************** -->
5314 <hr>
5315 <address>
5323 Last modified: $Date$
5324 </address>
5325 </body>
5326 </html>