| blob | f07fc31083ea28d1b4f2eca7c79106df44353960 |
3 <html>
4 <head>
11 </head>
13 <body>
16 <ol>
21 <ol>
34 </ol>
35 </li>
37 <ol>
40 <ol>
44 </ol>
45 </li>
47 <ol>
56 </ol>
57 </li>
59 </ol>
60 </li>
62 <ol>
69 </ol>
70 </li>
72 <ol>
74 </ol>
75 </li>
77 <ol>
79 <ol>
86 </ol>
87 </li>
89 <ol>
99 </ol>
100 </li>
102 <ol>
109 </ol>
110 </li>
112 <ol>
116 </ol>
117 </li>
119 <ol>
122 </ol>
123 </li>
125 <ol>
132 </ol>
133 </li>
135 <ol>
148 </ol>
149 </li>
151 <ol>
160 </ol>
161 </li>
162 </ol>
163 </li>
165 <ol>
167 <ol>
171 </ol>
172 </li>
174 <ol>
178 </ol>
179 </li>
181 <ol>
189 </ol>
190 </li>
192 <ol>
201 </ol>
202 </li>
204 <ol>
211 </ol>
212 </li>
214 <ol>
221 </ol>
222 </li>
226 <ol>
228 </ol>
229 </li>
231 <ol>
245 </ol>
246 </li>
248 <ol>
257 </ol>
258 </li>
259 </ol>
260 </li>
261 </ol>
266 </div>
268 <!-- *********************************************************************** -->
270 <!-- *********************************************************************** -->
273 <p>This document is a reference manual for the LLVM assembly language.
274 LLVM is a Static Single Assignment (SSA) based representation that provides
275 type safety, low-level operations, flexibility, and the capability of
276 representing 'all' high-level languages cleanly. It is the common code
277 representation used throughout all phases of the LLVM compilation
278 strategy.</p>
279 </div>
281 <!-- *********************************************************************** -->
283 <!-- *********************************************************************** -->
287 <p>The LLVM code representation is designed to be used in three
288 different forms: as an in-memory compiler IR, as an on-disk bitcode
289 representation (suitable for fast loading by a Just-In-Time compiler),
290 and as a human readable assembly language representation. This allows
291 LLVM to provide a powerful intermediate representation for efficient
292 compiler transformations and analysis, while providing a natural means
293 to debug and visualize the transformations. The three different forms
294 of LLVM are all equivalent. This document describes the human readable
295 representation and notation.</p>
297 <p>The LLVM representation aims to be light-weight and low-level
298 while being expressive, typed, and extensible at the same time. It
299 aims to be a "universal IR" of sorts, by being at a low enough level
300 that high-level ideas may be cleanly mapped to it (similar to how
301 microprocessors are "universal IR's", allowing many source languages to
302 be mapped to them). By providing type information, LLVM can be used as
303 the target of optimizations: for example, through pointer analysis, it
304 can be proven that a C automatic variable is never accessed outside of
305 the current function... allowing it to be promoted to a simple SSA
306 value instead of a memory location.</p>
308 </div>
310 <!-- _______________________________________________________________________ -->
315 <p>It is important to note that this document describes 'well formed'
316 LLVM assembly language. There is a difference between what the parser
317 accepts and what is considered 'well formed'. For example, the
318 following instruction is syntactically okay, but not well formed:</p>
321 <pre>
323 </pre>
324 </div>
327 its uses. The LLVM infrastructure provides a verification pass that may
328 be used to verify that an LLVM module is well formed. This pass is
329 automatically run by the parser after parsing input assembly and by
330 the optimizer before it outputs bitcode. The violations pointed out
331 by the verifier pass indicate bugs in transformation passes or input to
332 the parser.</p>
333 </div>
335 <!-- Describe the typesetting conventions here. -->
337 <!-- *********************************************************************** -->
339 <!-- *********************************************************************** -->
343 <p>LLVM identifiers come in two basic types: global and local. Global
344 identifiers (functions, global variables) begin with the @ character. Local
345 identifiers (register names, types) begin with the % character. Additionally,
346 there are three different formats for identifiers, for different purposes:</p>
348 <ol>
349 <li>Named values are represented as a string of characters with their prefix.
350 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
352 Identifiers which require other characters in their names can be surrounded
353 with quotes. Special characters may be escaped using "\xx" where xx is the
354 ASCII code for the character in hexadecimal. In this way, any character can
355 be used in a name value, even quotes themselves.
357 <li>Unnamed values are represented as an unsigned numeric value with their
362 </ol>
364 <p>LLVM requires that values start with a prefix for two reasons: Compilers
365 don't need to worry about name clashes with reserved words, and the set of
366 reserved words may be expanded in the future without penalty. Additionally,
367 unnamed identifiers allow a compiler to quickly come up with a temporary
368 variable without having to avoid symbol table conflicts.</p>
370 <p>Reserved words in LLVM are very similar to reserved words in other
371 languages. There are keywords for different opcodes
376 and others. These reserved words cannot conflict with variable names, because
377 none of them start with a prefix character ('%' or '@').</p>
379 <p>Here is an example of LLVM code to multiply the integer variable
385 <pre>
387 </pre>
388 </div>
393 <pre>
395 </pre>
396 </div>
401 <pre>
405 </pre>
406 </div>
409 important lexical features of LLVM:</p>
411 <ol>
414 line.</li>
416 <li>Unnamed temporaries are created when the result of a computation is not
417 assigned to a named value.</li>
421 </ol>
423 <p>...and it also shows a convention that we follow in this document. When
424 demonstrating instructions, we will follow an instruction with a comment that
425 defines the type and name of value produced. Comments are shown in italic
426 text.</p>
428 </div>
430 <!-- *********************************************************************** -->
432 <!-- *********************************************************************** -->
434 <!-- ======================================================================= -->
436 </div>
441 translation unit of the input programs. Each module consists of
442 functions, global variables, and symbol table entries. Modules may be
443 combined together with the LLVM linker, which merges function (and
444 global variable) definitions, resolves forward declarations, and merges
450 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
458 %cast210 = <a
462 <a
464 <a
466 </pre>
467 </div>
474 <p>In general, a module is made up of a list of global values,
475 where both functions and global variables are global values. Global values are
476 represented by a pointer to a memory location (in this case, a pointer to an
477 array of char, and a pointer to a function), and have one of the following <a
480 </div>
482 <!-- ======================================================================= -->
485 </div>
489 <p>
490 All Global Variables and Functions have one of the following types of linkage:
491 </p>
493 <dl>
497 <dd>Global values with private linkage are only directly accessible by
498 objects in the current module. In particular, linking code into a module with
499 an private global value may cause the private to be renamed as necessary to
500 avoid collisions. Because the symbol is private to the module, all
501 references can be updated. This doesn't show up in any symbol table in the
502 object file.
503 </dd>
507 <dd> Similar to private, but the value shows as a local symbol (STB_LOCAL in
508 the case of ELF) in the object file. This corresponds to the notion of the
510 </dd>
513 </dt>
516 into the object file corresponding to the LLVM module. They exist to
517 allow inlining and other optimizations to take place given knowledge of the
518 definition of the global, which is known to be somewhere outside the module.
521 type is only allowed on definitions, not declarations.</dd>
526 the same name when linkage occurs. This is typically used to implement
527 inline functions, templates, or other code which must be generated in each
529 allowed to be discarded.
530 </dd>
536 discarded. This is used for globals that may be emitted in multiple
537 translation units, but that are not guaranteed to be emitted into every
538 translation unit that uses them. One example of this is tentative
539 definitions in C, such as "<tt>int X;</tt>" at global scope.
540 </dd>
545 that some targets may choose to emit different assembly sequences for them
546 for target-dependent reasons. This is used for globals that are declared
547 "weak" in C source code.
548 </dd>
553 pointer to array type. When two global variables with appending linkage are
554 linked together, the two global arrays are appended together. This is the
555 LLVM, typesafe, equivalent of having the system linker append together
556 "sections" with identical names when .o files are linked.
557 </dd>
561 <dd>The semantics of this linkage follow the ELF object file model: the
562 symbol is weak until linked, if not linked, the symbol becomes null instead
563 of being an undefined reference.
564 </dd>
568 <dd>Some languages allow differing globals to be merged, such as two
570 ensure that only equivalent globals are ever merged (the "one definition
573 be merged with equivalent globals. These linkage types are otherwise the
575 </dd>
579 <dd>If none of the above identifiers are used, the global is externally
580 visible, meaning that it participates in linkage and can be used to resolve
581 external symbol references.
582 </dd>
583 </dl>
585 <p>
586 The next two types of linkage are targeted for Microsoft Windows platform
587 only. They are designed to support importing (exporting) symbols from (to)
588 DLLs (Dynamic Link Libraries).
589 </p>
591 <dl>
595 or variable via a global pointer to a pointer that is set up by the DLL
596 exporting the symbol. On Microsoft Windows targets, the pointer name is
598 </dd>
603 pointer to a pointer in a DLL, so that it can be referenced with the
606 name.
607 </dd>
609 </dl>
612 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
613 variable and was linked with this one, one of the two would be renamed,
615 external (i.e., lacking any linkage declarations), they are accessible
616 outside of the current module.</p>
622 </div>
624 <!-- ======================================================================= -->
627 </div>
633 specified for the call. The calling convention of any pair of dynamic
634 caller/callee must match, or the behavior of the program is undefined. The
635 following calling conventions are supported by LLVM, and more may be added in
636 the future:</p>
638 <dl>
641 <dd>This calling convention (the default if no other calling convention is
642 specified) matches the target C calling conventions. This calling convention
643 supports varargs function calls and tolerates some mismatch in the declared
644 prototype and implemented declaration of the function (as does normal C).
645 </dd>
649 <dd>This calling convention attempts to make calls as fast as possible
650 (e.g. by passing things in registers). This calling convention allows the
651 target to use whatever tricks it wants to produce fast code for the target,
652 without having to conform to an externally specified ABI (Application Binary
653 Interface). Implementations of this convention should allow arbitrary
655 supported. This calling convention does not support varargs and requires the
656 prototype of all callees to exactly match the prototype of the function
657 definition.
658 </dd>
662 <dd>This calling convention attempts to make code in the caller as efficient
663 as possible under the assumption that the call is not commonly executed. As
664 such, these calls often preserve all registers so that the call does not break
665 any live ranges in the caller side. This calling convention does not support
666 varargs and requires the prototype of all callees to exactly match the
667 prototype of the function definition.
668 </dd>
672 <dd>Any calling convention may be specified by number, allowing
673 target-specific calling conventions to be used. Target specific calling
674 conventions start at 64.
675 </dd>
676 </dl>
678 <p>More calling conventions can be added/defined on an as-needed basis, to
679 support pascal conventions or any other well-known target-independent
680 convention.</p>
682 </div>
684 <!-- ======================================================================= -->
687 </div>
691 <p>
692 All Global Variables and Functions have one of the following visibility styles:
693 </p>
695 <dl>
698 <dd>On targets that use the ELF object file format, default visibility means
699 that the declaration is visible to other
700 modules and, in shared libraries, means that the declared entity may be
701 overridden. On Darwin, default visibility means that the declaration is
702 visible to other modules. Default visibility corresponds to "external
703 linkage" in the language.
704 </dd>
708 <dd>Two declarations of an object with hidden visibility refer to the same
709 object if they are in the same shared object. Usually, hidden visibility
710 indicates that the symbol will not be placed into the dynamic symbol table,
711 so no other module (executable or shared library) can reference it
712 directly.
713 </dd>
717 <dd>On ELF, protected visibility indicates that the symbol will be placed in
718 the dynamic symbol table, but that references within the defining module will
719 bind to the local symbol. That is, the symbol cannot be overridden by another
720 module.
721 </dd>
722 </dl>
724 </div>
726 <!-- ======================================================================= -->
729 </div>
733 <p>LLVM IR allows you to specify name aliases for certain types. This can make
734 it easier to read the IR and make the IR more condensed (particularly when
735 recursive types are involved). An example of a name specification is:
736 </p>
739 <pre>
740 %mytype = type { %mytype*, i32 }
741 </pre>
742 </div>
748 <p>Note that type names are aliases for the structural type that they indicate,
749 and that you can therefore specify multiple names for the same type. This often
750 leads to confusing behavior when dumping out a .ll file. Since LLVM IR uses
751 structural typing, the name is not part of the type. When printing out LLVM IR,
753 shape. This means that if you have code where two different source types end up
754 having the same LLVM type, that the dumper will sometimes print the "wrong" or
755 unexpected type. This is an important design point and isn't going to
756 change.</p>
758 </div>
760 <!-- ======================================================================= -->
763 </div>
767 <p>Global variables define regions of memory allocated at compilation time
768 instead of run-time. Global variables may optionally be initialized, may have
769 an explicit section to be placed in, and may have an optional explicit alignment
770 specified. A variable may be defined as "thread_local", which means that it
771 will not be shared by threads (each thread will have a separated copy of the
772 variable). A variable may be defined as a global "constant," which indicates
774 optimization, allowing the global data to be placed in the read-only section of
775 an executable, etc). Note that variables that need runtime initialization
778 <p>
780 constant, even if the final definition of the global is not. This capability
781 can be used to enable slightly better optimization of the program, but requires
782 the language definition to guarantee that optimizations based on the
783 'constantness' are valid for the translation units that do not include the
784 definition.
785 </p>
787 <p>As SSA values, global variables define pointer values that are in
788 scope (i.e. they dominate) all basic blocks in the program. Global
789 variables always define a pointer to their "content" type because they
790 describe a region of memory, and all memory objects in LLVM are
791 accessed through pointers.</p>
793 <p>A global variable may be declared to reside in a target-specifc numbered
794 address space. For targets that support them, address spaces may affect how
795 optimizations are performed and/or what target instructions are used to access
796 the variable. The default address space is zero. The address space qualifier
797 must precede any other attributes.</p>
799 <p>LLVM allows an explicit section to be specified for globals. If the target
800 supports it, it will emit globals to the section specified.</p>
802 <p>An explicit alignment may be specified for a global. If not present, or if
803 the alignment is set to zero, the alignment of the global is set by the target
804 to whatever it feels convenient. If an explicit alignment is specified, the
805 global is forced to have at least that much alignment. All alignments must be
808 <p>For example, the following defines a global in a numbered address space with
809 an initializer, section, and alignment:</p>
812 <pre>
814 </pre>
815 </div>
817 </div>
820 <!-- ======================================================================= -->
823 </div>
832 name, a (possibly empty) argument list (each with optional
836 an opening curly brace, a list of basic blocks, and a closing curly brace.
838 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
843 name, a possibly empty list of arguments, an optional alignment, and an optional
846 <p>A function definition contains a list of basic blocks, forming the CFG
847 (Control Flow Graph) for
848 the function. Each basic block may optionally start with a label (giving the
849 basic block a symbol table entry), contains a list of instructions, and ends
851 function return).</p>
853 <p>The first basic block in a function is special in two ways: it is immediately
854 executed on entrance to the function, and it is not allowed to have predecessor
855 basic blocks (i.e. there can not be any branches to the entry block of a
856 function). Because the block can have no predecessors, it also cannot have any
859 <p>LLVM allows an explicit section to be specified for functions. If the target
860 supports it, it will emit functions to the section specified.</p>
862 <p>An explicit alignment may be specified for a function. If not present, or if
863 the alignment is set to zero, the alignment of the function is set by the target
864 to whatever it feels convenient. If an explicit alignment is specified, the
865 function is forced to have at least that much alignment. All alignments must be
871 <tt>
877 </tt>
878 </div>
880 </div>
883 <!-- ======================================================================= -->
886 </div>
889 function, global variable, another alias or bitcast of global value). Aliases
896 <pre>
898 </pre>
899 </div>
901 </div>
905 <!-- ======================================================================= -->
908 <p>The return type and each parameter of a function type may have a set of
910 used to communicate additional information about the result or parameters of
911 a function. Parameter attributes are considered to be part of the function,
912 not of the function type, so functions with different parameter attributes
913 can have the same function type.</p>
915 <p>Parameter attributes are simple keywords that follow the type specified. If
916 multiple parameter attributes are needed, they are space separated. For
917 example:</p>
920 <pre>
921 declare i32 @printf(i8* noalias nocapture, ...)
922 declare i32 @atoi(i8 zeroext)
923 declare signext i8 @returns_signed_char()
924 </pre>
925 </div>
931 <dl>
933 <dd>This indicates to the code generator that the parameter or return value
934 should be zero-extended to a 32-bit value by the caller (for a parameter)
935 or the callee (for a return value).</dd>
938 <dd>This indicates to the code generator that the parameter or return value
939 should be sign-extended to a 32-bit value by the caller (for a parameter)
940 or the callee (for a return value).</dd>
943 <dd>This indicates that this parameter or return value should be treated
944 in a special target-dependent fashion during while emitting code for a
945 function call or return (usually, by putting it in a register as opposed
946 to memory, though some targets use it to distinguish between two different
947 kinds of registers). Use of this attribute is target-specific.</dd>
950 <dd>This indicates that the pointer parameter should really be passed by
951 value to the function. The attribute implies that a hidden copy of the
952 pointee is made between the caller and the callee, so the callee is unable
953 to modify the value in the callee. This attribute is only valid on LLVM
954 pointer arguments. It is generally used to pass structs and arrays by
955 value, but is also valid on pointers to scalars. The copy is considered to
956 belong to the caller not the callee (for example,
959 values. The byval attribute also supports specifying an alignment with the
960 align attribute. This has a target-specific effect on the code generator
961 that usually indicates a desired alignment for the synthesized stack
962 slot.</dd>
965 <dd>This indicates that the pointer parameter specifies the address of a
966 structure that is the return value of the function in the source program.
967 This pointer must be guaranteed by the caller to be valid: loads and stores
968 to the structure may be assumed by the callee to not to trap. This may only
969 be applied to the first parameter. This is not a valid attribute for
970 return values. </dd>
973 <dd>This indicates that the pointer does not alias any global or any other
974 parameter. The caller is responsible for ensuring that this is the
976 that the pointer does not alias any other pointers visible to the
977 caller. For further details, please see the discussion of the NoAlias
978 response in
983 <dd>This indicates that the callee does not make any copies of the pointer
984 that outlive the callee itself. This is not a valid attribute for return
985 values.</dd>
988 <dd>This indicates that the pointer parameter can be excised using the
990 attribute for return values.</dd>
991 </dl>
993 </div>
995 <!-- ======================================================================= -->
998 </div>
1001 <p>Each function may specify a garbage collector name, which is simply a
1002 string.</p>
1008 collector which will cause the compiler to alter its output in order to support
1009 the named garbage collection algorithm.</p>
1010 </div>
1012 <!-- ======================================================================= -->
1015 </div>
1019 <p>Function attributes are set to communicate additional information about
1020 a function. Function attributes are considered to be part of the function,
1021 not of the function type, so functions with different parameter attributes
1022 can have the same function type.</p>
1024 <p>Function attributes are simple keywords that follow the type specified. If
1025 multiple attributes are needed, they are space separated. For
1026 example:</p>
1029 <pre>
1030 define void @f() noinline { ... }
1031 define void @f() alwaysinline { ... }
1032 define void @f() alwaysinline optsize { ... }
1033 define void @f() optsize
1034 </pre>
1035 </div>
1037 <dl>
1039 <dd>This attribute indicates that the inliner should attempt to inline this
1040 function into callers whenever possible, ignoring any active inlining size
1041 threshold for this caller.</dd>
1044 <dd>This attribute indicates that the inliner should never inline this function
1045 in any situation. This attribute may not be used together with the
1049 <dd>This attribute suggests that optimization passes and code generator passes
1050 make choices that keep the code size of this function low, and otherwise do
1051 optimizations specifically to reduce code size.</dd>
1054 <dd>This function attribute indicates that the function never returns normally.
1055 This produces undefined behavior at runtime if the function ever does
1056 dynamically return.</dd>
1059 <dd>This function attribute indicates that the function never returns with an
1060 unwind or exceptional control flow. If the function does unwind, its runtime
1061 behavior is undefined.</dd>
1064 <dd>This attribute indicates that the function computes its result (or the
1065 exception it throws) based strictly on its arguments, without dereferencing any
1066 pointer arguments or otherwise accessing any mutable state (e.g. memory, control
1067 registers, etc) visible to caller functions. It does not write through any
1069 never changes any state visible to callers. readnone functions may not throw
1070 an exception that escapes into the caller.</dd>
1073 <dd>This attribute indicates that the function does not write through any
1075 or otherwise modify any state (e.g. memory, control registers, etc) visible to
1076 caller functions. It may dereference pointer arguments and read state that may
1077 be set in the caller. A readonly function always returns the same value when
1078 called with the same set of arguments and global
1079 state. readonly functions may not throw an exception that escapes into the
1080 caller.</dd>
1083 <dd>This attribute indicates that the function should emit a stack smashing
1085 stack before the local variables that's checked upon return from the function to
1086 see if it has been overwritten. A heuristic is used to determine if a function
1087 needs stack protectors or not.
1096 function attribute.
1102 </dl>
1104 </div>
1106 <!-- ======================================================================= -->
1109 </div>
1112 <p>
1113 Modules may contain "module-level inline asm" blocks, which corresponds to the
1114 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
1115 LLVM and treated as a single unit, but may be separated in the .ll file if
1116 desired. The syntax is very simple:
1117 </p>
1120 <pre>
1121 module asm "inline asm code goes here"
1122 module asm "more can go here"
1123 </pre>
1124 </div>
1126 <p>The strings can contain any character by escaping non-printable characters.
1128 for the number.
1129 </p>
1131 <p>
1132 The inline asm code is simply printed to the machine code .s file when
1133 assembly code is generated.
1134 </p>
1135 </div>
1137 <!-- ======================================================================= -->
1140 </div>
1143 <p>A module may specify a target specific data layout string that specifies how
1144 data is to be laid out in memory. The syntax for the data layout is simply:</p>
1147 separated by the minus sign character ('-'). Each specification starts with a
1148 letter and may include other information after the letter to define some
1149 aspect of the data layout. The specifications accepted are as follows: </p>
1150 <dl>
1152 <dd>Specifies that the target lays out data in big-endian form. That is, the
1153 bits with the most significance have the lowest address location.</dd>
1155 <dd>Specifies that the target lays out data in little-endian form. That is,
1156 the bits with the least significance have the lowest address location.</dd>
1161 too.</dd>
1163 <dd>This specifies the alignment for an integer type of a given bit
1166 <dd>This specifies the alignment for a vector type of a given bit
1169 <dd>This specifies the alignment for a floating point type of a given bit
1171 (double).</dd>
1173 <dd>This specifies the alignment for an aggregate type of a given bit
1175 </dl>
1176 <p>When constructing the data layout for a given target, LLVM starts with a
1177 default set of specifications which are then (possibly) overriden by the
1179 are given in this list:</p>
1180 <ul>
1194 </ul>
1195 <p>When LLVM is determining the alignment for a given type, it uses the
1196 following rules:</p>
1197 <ol>
1198 <li>If the type sought is an exact match for one of the specifications, that
1199 specification is used.</li>
1200 <li>If no match is found, and the type sought is an integer type, then the
1201 smallest integer type that is larger than the bitwidth of the sought type is
1202 used. If none of the specifications are larger than the bitwidth then the the
1203 largest integer type is used. For example, given the default specifications
1204 above, the i7 type will use the alignment of i8 (next largest) while both
1205 i65 and i256 will use the alignment of i64 (largest specified).</li>
1206 <li>If no match is found, and the type sought is a vector type, then the
1207 largest vector type that is smaller than the sought vector type will be used
1210 </ol>
1211 </div>
1213 <!-- *********************************************************************** -->
1215 <!-- *********************************************************************** -->
1219 <p>The LLVM type system is one of the most important features of the
1220 intermediate representation. Being typed enables a number of
1221 optimizations to be performed on the intermediate representation directly,
1222 without having to do
1223 extra analyses on the side before the transformation. A strong type
1224 system makes it easier to read the generated code and enables novel
1225 analyses and transformations that are not feasible to perform on normal
1226 three address code representations.</p>
1228 </div>
1230 <!-- ======================================================================= -->
1232 Classifications</a> </div>
1234 <p>The types fall into a few useful
1235 classifications:</p>
1238 <tbody>
1240 <tr>
1243 </tr>
1244 <tr>
1247 </tr>
1248 <tr>
1257 </td>
1258 </tr>
1259 <tr>
1264 </tr>
1265 <tr>
1275 </td>
1276 </tr>
1277 </tbody>
1278 </table>
1281 most important. Values of these types are the only ones which can be
1282 produced by instructions, passed as arguments, or used as operands to
1283 instructions.</p>
1284 </div>
1286 <!-- ======================================================================= -->
1290 <p>The primitive types are the fundamental building blocks of the LLVM
1291 system.</p>
1293 </div>
1295 <!-- _______________________________________________________________________ -->
1299 <table>
1300 <tbody>
1307 </tbody>
1308 </table>
1309 </div>
1311 <!-- _______________________________________________________________________ -->
1320 <pre>
1321 void
1322 </pre>
1323 </div>
1325 <!-- _______________________________________________________________________ -->
1334 <pre>
1335 label
1336 </pre>
1337 </div>
1340 <!-- ======================================================================= -->
1345 <p>The real power in LLVM comes from the derived types in the system.
1346 This is what allows a programmer to represent arrays, functions,
1347 pointers, and other useful types. Note that these derived types may be
1348 recursive: For example, it is possible to have a two dimensional array.</p>
1350 </div>
1352 <!-- _______________________________________________________________________ -->
1358 <p>The integer type is a very simple derived type that simply specifies an
1359 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1364 <pre>
1365 iN
1366 </pre>
1369 value.</p>
1373 <tbody>
1374 <tr>
1377 </tr><tr>
1380 </tr><tr>
1383 </tr>
1384 </tbody>
1385 </table>
1387 <p>Note that the code generator does not yet support large integer types
1388 to be used as function return types. The specific limit on how large a
1389 return type the code generator can currently handle is target-dependent;
1391 targets.</p>
1393 </div>
1395 <!-- _______________________________________________________________________ -->
1402 <p>The array type is a very simple derived type that arranges elements
1403 sequentially in memory. The array type requires a size (number of
1404 elements) and an underlying data type.</p>
1408 <pre>
1410 </pre>
1412 <p>The number of elements is a constant integer value; elementtype may
1413 be any type with a size.</p>
1420 </tr>
1424 </tr>
1428 </tr>
1429 </table>
1435 </tr>
1439 </tr>
1443 </tr>
1444 </table>
1446 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1447 length array. Normally, accesses past the end of an array are undefined in
1449 As a special case, however, zero length arrays are recognized to be variable
1450 length. This allows implementation of 'pascal style arrays' with the LLVM
1453 <p>Note that the code generator does not yet support large aggregate types
1454 to be used as function return types. The specific limit on how large an
1455 aggregate return type the code generator can currently handle is
1456 target-dependent, and also dependent on the aggregate element types.</p>
1458 </div>
1460 <!-- _______________________________________________________________________ -->
1466 <p>The function type can be thought of as a function signature. It
1467 consists of a return type and a list of formal parameter types. The
1468 return type of a function type is a scalar type, a void type, or a struct type.
1469 If the return type is a struct type then all struct elements must be of first
1470 class types, and the struct must have at least one element.</p>
1474 <pre>
1476 </pre>
1480 which indicates that the function takes a variable number of arguments.
1481 Variable argument functions can access their arguments with the <a
1491 </td>
1494 </tt></td>
1499 </td>
1505 LLVM.
1506 </td>
1511 </td>
1512 </tr>
1513 </table>
1515 </div>
1516 <!-- _______________________________________________________________________ -->
1520 <p>The structure type is used to represent a collection of data members
1521 together in memory. The packing of the field types is defined to match
1522 the ABI of the underlying processor. The elements of a structure may
1523 be any type that has a size.</p>
1527 instruction.</p>
1541 </tr>
1542 </table>
1544 <p>Note that the code generator does not yet support large aggregate types
1545 to be used as function return types. The specific limit on how large an
1546 aggregate return type the code generator can currently handle is
1547 target-dependent, and also dependent on the aggregate element types.</p>
1549 </div>
1551 <!-- _______________________________________________________________________ -->
1553 </div>
1556 <p>The packed structure type is used to represent a collection of data members
1557 together in memory. There is no padding between fields. Further, the alignment
1558 of a packed structure is 1 byte. The elements of a packed structure may
1559 be any type that has a size.</p>
1563 instruction.</p>
1578 </tr>
1579 </table>
1580 </div>
1582 <!-- _______________________________________________________________________ -->
1586 <p>As in many languages, the pointer type represents a pointer or
1587 reference to another object, which must live in memory. Pointer types may have
1588 an optional address space attribute defining the target-specific numbered
1589 address space where the pointed-to object resides. The default address space is
1590 zero.</p>
1603 </tr>
1609 </tr>
1614 </tr>
1615 </table>
1616 </div>
1618 <!-- _______________________________________________________________________ -->
1624 <p>A vector type is a simple derived type that represents a vector
1625 of elements. Vector types are used when multiple primitive data
1626 are operated in parallel using a single instruction (SIMD).
1627 A vector type requires a size (number of
1628 elements) and an underlying primitive data type. Vectors must have a power
1634 <pre>
1636 </pre>
1638 <p>The number of elements is a constant integer value; elementtype may
1639 be any integer or floating point type.</p>
1647 </tr>
1651 </tr>
1655 </tr>
1656 </table>
1658 <p>Note that the code generator does not yet support large vector types
1659 to be used as function return types. The specific limit on how large a
1660 vector return type codegen can currently handle is target-dependent;
1661 currently it's often a few times longer than a hardware vector register.</p>
1663 </div>
1665 <!-- _______________________________________________________________________ -->
1671 <p>Opaque types are used to represent unknown types in the system. This
1672 corresponds (for example) to the C notion of a forward declared structure type.
1673 In LLVM, opaque types can eventually be resolved to any type (not just a
1674 structure type).</p>
1678 <pre>
1679 opaque
1680 </pre>
1688 </tr>
1689 </table>
1690 </div>
1692 <!-- ======================================================================= -->
1695 </div>
1699 <p>
1700 An "up reference" allows you to refer to a lexically enclosing type without
1701 requiring it to have a name. For instance, a structure declaration may contain a
1702 pointer to any of the types it is lexically a member of. Example of up
1703 references (with their equivalent as named type declarations) include:</p>
1705 <pre>
1706 { \2 * } %x = type { %x* }
1707 { \2 }* %y = type { %y }*
1708 \1* %z = type %z*
1709 </pre>
1711 <p>
1712 An up reference is needed by the asmprinter for printing out cyclic types when
1713 there is no declared name for a type in the cycle. Because the asmprinter does
1714 not want to print out an infinite type string, it needs a syntax to handle
1715 recursive types that have no names (all names are optional in llvm IR).
1716 </p>
1719 <pre>
1721 </pre>
1723 <p>
1724 The level is the count of the lexical type that is being referred to.
1725 </p>
1733 </tr>
1737 structure.</td>
1738 </tr>
1739 </table>
1740 </div>
1743 <!-- *********************************************************************** -->
1745 <!-- *********************************************************************** -->
1749 <p>LLVM has several different basic types of constants. This section describes
1750 them all and their syntax.</p>
1752 </div>
1754 <!-- ======================================================================= -->
1759 <dl>
1764 </dd>
1770 integer types.
1771 </dd>
1776 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1777 notation (see below). The assembler requires the exact decimal value of
1778 a floating-point constant. For example, the assembler accepts 1.25 but
1787 </dl>
1789 <p>The one non-intuitive notation for constants is the hexadecimal form
1790 of floating point constants. For example, the form '<tt>double
1793 (and the only time that they are generated by the disassembler) is when a
1794 floating point constant must be emitted but it cannot be represented as a
1795 decimal floating point number in a reasonable number of digits. For example,
1796 NaN's, infinities, and other
1797 special values are represented in their IEEE hexadecimal format so that
1798 assembly and disassembly do not cause any bits to change in the constants.</p>
1799 <p>When using the hexadecimal form, constants of types float and double are
1800 represented using the 16-digit form shown above (which matches the IEEE754
1801 representation for double); float values must, however, be exactly representable
1802 as IEE754 single precision.
1803 Hexadecimal format is always used for long
1804 double, and there are three forms of long double. The 80-bit
1806 followed by 20 hexadecimal digits.
1807 The 128-bit format used by PowerPC (two adjacent doubles) is represented
1809 format is represented
1811 target uses this format. Long doubles will only work if they match
1812 the long double format on your target. All hexadecimal formats are big-endian
1813 (sign bit at the left).</p>
1814 </div>
1816 <!-- ======================================================================= -->
1820 </div>
1823 <p>Complex constants are a (potentially recursive) combination of simple
1824 constants and smaller complex constants.</p>
1826 <dl>
1829 <dd>Structure constants are represented with notation similar to structure
1830 type definitions (a comma separated list of elements, surrounded by braces
1834 types of elements must match those specified by the type.
1835 </dd>
1839 <dd>Array constants are represented with notation similar to array type
1840 definitions (a comma separated list of elements, surrounded by square brackets
1843 types of elements must match those specified by the type.
1844 </dd>
1848 <dd>Vector constants are represented with notation similar to vector type
1849 definitions (a comma separated list of elements, surrounded by
1853 match those specified by the type.
1854 </dd>
1860 This is often used to avoid having to print large zero initializers (e.g. for
1861 large arrays) and is always exactly equivalent to using explicit zero
1862 initializers.
1863 </dd>
1867 <dd>A metadata node is a structure-like constant with the type of an empty
1869 constants that are meant to be interpreted as part of the instruction stream,
1870 metadata is a place to attach additional information such as debug info.
1871 </dd>
1872 </dl>
1874 </div>
1876 <!-- ======================================================================= -->
1879 </div>
1885 constants. These constants are explicitly referenced when the <a
1888 file:</p>
1891 <pre>
1892 @X = global i32 17
1893 @Y = global i32 42
1894 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1895 </pre>
1896 </div>
1898 </div>
1900 <!-- ======================================================================= -->
1904 no specific value. Undefined values may be of any type and be used anywhere
1905 a constant is permitted.</p>
1907 <p>Undefined values indicate to the compiler that the program is well defined
1908 no matter what value is used, giving the compiler more freedom to optimize.
1909 </p>
1910 </div>
1912 <!-- ======================================================================= -->
1914 </div>
1918 <p>Constant expressions are used to allow expressions involving other constants
1919 to be used as constants. Constant expressions may be of any <a
1921 that does not have side effects (e.g. load and call are not supported). The
1922 following is the syntax for constant expressions:</p>
1924 <dl>
1926 <dd>Truncate a constant to another type. The bit size of CST must be larger
1927 than the bit size of TYPE. Both types must be integers.</dd>
1930 <dd>Zero extend a constant to another type. The bit size of CST must be
1931 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1934 <dd>Sign extend a constant to another type. The bit size of CST must be
1935 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1938 <dd>Truncate a floating point constant to another floating point type. The
1939 size of CST must be larger than the size of TYPE. Both types must be
1940 floating point.</dd>
1943 <dd>Floating point extend a constant to another type. The size of CST must be
1944 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1947 <dd>Convert a floating point constant to the corresponding unsigned integer
1948 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1949 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1950 of the same number of elements. If the value won't fit in the integer type,
1951 the results are undefined.</dd>
1954 <dd>Convert a floating point constant to the corresponding signed integer
1955 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1956 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1957 of the same number of elements. If the value won't fit in the integer type,
1958 the results are undefined.</dd>
1961 <dd>Convert an unsigned integer constant to the corresponding floating point
1962 constant. TYPE must be a scalar or vector floating point type. CST must be of
1963 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1964 of the same number of elements. If the value won't fit in the floating point
1965 type, the results are undefined.</dd>
1968 <dd>Convert a signed integer constant to the corresponding floating point
1969 constant. TYPE must be a scalar or vector floating point type. CST must be of
1970 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1971 of the same number of elements. If the value won't fit in the floating point
1972 type, the results are undefined.</dd>
1975 <dd>Convert a pointer typed constant to the corresponding integer constant
1976 TYPE must be an integer type. CST must be of pointer type. The CST value is
1977 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1980 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1981 pointer type. CST must be of integer type. The CST value is zero extended,
1982 truncated, or unchanged to make it fit in a pointer size. This one is
1986 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
1994 instruction, the index list may have zero or more indexes, which are required
2000 constants.</dd>
2032 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2034 binary</a> operations. The constraints on operands are the same as those for
2035 the corresponding instruction (e.g. no bitwise operations on floating point
2036 values are allowed).</dd>
2037 </dl>
2038 </div>
2040 <!-- ======================================================================= -->
2042 </div>
2046 <p>Embedded metadata provides a way to attach arbitrary data to the
2047 instruction stream without affecting the behaviour of the program. There are
2048 two metadata primitives, strings and nodes. All metadata has the type of an
2049 empty struct and is identified in syntax by a preceding exclamation point
2051 </p>
2053 <p>A metadata string is a string surrounded by double quotes. It can contain
2056 </p>
2058 <p>Metadata nodes are represented with notation similar to structure constants
2059 (a comma separated list of elements, surrounded by braces and preceeded by an
2061 </p>
2063 <p>Optimizations may rely on metadata to provide additional information about
2064 the program that isn't available in the instructions, or that isn't easily
2065 computable. Similarly, the code generator may expect a certain metadata format
2066 to be used to express debugging information.</p>
2067 </div>
2069 <!-- *********************************************************************** -->
2071 <!-- *********************************************************************** -->
2073 <!-- ======================================================================= -->
2076 </div>
2080 <p>
2082 Module-Level Inline Assembly</a>) through the use of a special value. This
2083 value represents the inline assembler as a string (containing the instructions
2084 to emit), a list of operand constraints (stored as a string), and a flag that
2085 indicates whether or not the inline asm expression has side effects. An example
2086 inline assembler expression is:
2087 </p>
2090 <pre>
2092 </pre>
2093 </div>
2095 <p>
2098 </p>
2101 <pre>
2103 </pre>
2104 </div>
2106 <p>
2107 Inline asms with side effects not visible in the constraint list must be marked
2108 as having side effects. This is done through the use of the
2110 </p>
2113 <pre>
2115 </pre>
2116 </div>
2118 <p>TODO: The format of the asm and constraints string still need to be
2119 documented here. Constraints on what can be done (e.g. duplication, moving, etc
2120 need to be documented). This is probably best done by reference to another
2121 document that covers inline asm from a holistic perspective.
2122 </p>
2124 </div>
2126 <!-- *********************************************************************** -->
2128 <!-- *********************************************************************** -->
2132 <p>The LLVM instruction set consists of several different
2139 </div>
2141 <!-- ======================================================================= -->
2143 Instructions</a> </div>
2148 basic block in a program ends with a "Terminator" instruction, which
2149 indicates which block should be executed after the current block is
2151 value: they produce control flow, not values (the one exception being
2160 </div>
2162 <!-- _______________________________________________________________________ -->
2164 Instruction</a> </div>
2167 <pre>
2170 </pre>
2175 optionally a value) from a function back to the caller.</p>
2177 returns a value and then causes control flow, and one that just causes
2178 control flow to occur.</p>
2183 the return value. The type of the return value must be a
2188 instruction with no return value or a return value with a type that
2189 does not match its type, or if it has a void return type and contains
2195 returns back to the calling function's context. If the caller is a "<a
2197 the instruction after the call. If the caller was an "<a
2199 at the beginning of the "normal" destination block. If the instruction
2200 returns a value, that value shall set the call or invoke instruction's
2201 return value.</p>
2205 <pre>
2209 </pre>
2211 <p>Note that the code generator does not yet fully support large
2212 return values. The specific sizes that are currently supported are
2213 dependent on the target. For integers, on 32-bit targets the limit
2215 For aggregate types, the current limits are dependent on the element
2216 types; for example targets are often limited to 2 total integer
2219 </div>
2220 <!-- _______________________________________________________________________ -->
2224 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2225 </pre>
2228 transfer to a different basic block in the current function. There are
2229 two forms of this instruction, corresponding to a conditional branch
2230 and an unconditional branch.</p>
2242 <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
2244 </div>
2245 <!-- _______________________________________________________________________ -->
2248 </div>
2253 <pre>
2254 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2255 </pre>
2261 instruction, allowing a branch to occur to one of many possible
2262 destinations.</p>
2270 table is not allowed to contain duplicate constant entries.</p>
2276 table is searched for the given value. If the value is found, control flow is
2277 transfered to the corresponding destination; otherwise, control flow is
2278 transfered to the default destination.</p>
2282 <p>Depending on properties of the target machine and the particular
2284 ways. For example, it could be generated as a series of chained conditional
2285 branches or with a lookup table.</p>
2289 <pre>
2292 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2295 switch i32 0, label %dest [ ]
2298 switch i32 %val, label %otherwise [ i32 0, label %onzero
2299 i32 1, label %onone
2300 i32 2, label %ontwo ]
2301 </pre>
2302 </div>
2304 <!-- _______________________________________________________________________ -->
2307 </div>
2313 <pre>
2314 <result> = invoke [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] <ptr to function ty> <function ptr val>(<function args>) [<a href="#fnattrs">fn attrs</a>]
2316 </pre>
2321 function, with the possibility of control flow transfer to either the
2333 <ol>
2334 <li>
2336 convention</a> the call should use. If none is specified, the call defaults
2337 to using C calling conventions.
2338 </li>
2345 function value being invoked. In most cases, this is a direct function
2347 an arbitrary pointer to function value.
2348 </li>
2351 function to be invoked. </li>
2354 signature argument types. If the function signature indicates the function
2355 accepts a variable number of arguments, the extra arguments can be
2356 specified. </li>
2367 </ol>
2373 difference is that it establishes an association with a label, which is used by
2374 the runtime library to unwind the stack.</p>
2376 <p>This instruction is used in languages with destructors to ensure that proper
2378 exception. Additionally, this is important for implementation of
2382 <pre>
2383 %retval = invoke i32 @Test(i32 15) to label %Continue
2387 </pre>
2388 </div>
2391 <!-- _______________________________________________________________________ -->
2394 Instruction</a> </div>
2399 <pre>
2400 unwind
2401 </pre>
2406 at the first callee in the dynamic call stack which used an <a
2408 primarily used to implement exception handling.</p>
2413 immediately halt. The dynamic call stack is then searched for the first <a
2415 execution continues at the "exceptional" destination block specified by the
2417 dynamic call chain, undefined behavior results.</p>
2418 </div>
2420 <!-- _______________________________________________________________________ -->
2423 Instruction</a> </div>
2428 <pre>
2429 unreachable
2430 </pre>
2435 instruction is used to inform the optimizer that a particular portion of the
2436 code is not reachable. This can be used to indicate that the code after a
2437 no-return function cannot be reached, and other facts.</p>
2442 </div>
2446 <!-- ======================================================================= -->
2449 <p>Binary operators are used to do most of the computation in a
2450 program. They require two operands of the same type, execute an operation on them, and
2451 produce a single value. The operands might represent
2453 The result value has the same type as its operands.</p>
2455 </div>
2456 <!-- _______________________________________________________________________ -->
2459 </div>
2465 <pre>
2467 </pre>
2478 types.</p>
2482 <p>The value produced is the integer or floating point sum of the two
2483 operands.</p>
2485 <p>If an integer sum has unsigned overflow, the result returned is the
2487 the result.</p>
2489 <p>Because LLVM integers use a two's complement representation, this
2490 instruction is appropriate for both signed and unsigned integers.</p>
2494 <pre>
2496 </pre>
2497 </div>
2498 <!-- _______________________________________________________________________ -->
2501 </div>
2507 <pre>
2509 </pre>
2514 operands.</p>
2518 representations.</p>
2525 types.</p>
2529 <p>The value produced is the integer or floating point difference of
2530 the two operands.</p>
2532 <p>If an integer difference has unsigned overflow, the result returned is the
2534 the result.</p>
2536 <p>Because LLVM integers use a two's complement representation, this
2537 instruction is appropriate for both signed and unsigned integers.</p>
2540 <pre>
2543 </pre>
2544 </div>
2546 <!-- _______________________________________________________________________ -->
2549 </div>
2555 </pre>
2558 operands.</p>
2565 types.</p>
2569 <p>The value produced is the integer or floating point product of the
2570 two operands.</p>
2572 <p>If the result of an integer multiplication has unsigned overflow,
2573 the result returned is the mathematical result modulo
2575 <p>Because LLVM integers use a two's complement representation, and the
2576 result is the same width as the operands, this instruction returns the
2577 correct result for both signed and unsigned integers. If a full product
2579 should be sign-extended or zero-extended as appropriate to the
2580 width of the full product.</p>
2583 </pre>
2584 </div>
2586 <!-- _______________________________________________________________________ -->
2588 </a></div>
2592 </pre>
2595 operands.</p>
2601 values. Both arguments must have identical types.</p>
2606 <p>Note that unsigned integer division and signed integer division are distinct
2611 </pre>
2612 </div>
2613 <!-- _______________________________________________________________________ -->
2615 </a> </div>
2618 <pre>
2620 </pre>
2625 operands.</p>
2631 values. Both arguments must have identical types.</p>
2634 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2635 <p>Note that signed integer division and unsigned integer division are distinct
2637 <p>Division by zero leads to undefined behavior. Overflow also leads to
2638 undefined behavior; this is a rare case, but can occur, for example,
2642 </pre>
2643 </div>
2644 <!-- _______________________________________________________________________ -->
2646 Instruction</a> </div>
2649 <pre>
2651 </pre>
2655 operands.</p>
2661 of floating point values. Both arguments must have identical types.</p>
2669 <pre>
2671 </pre>
2672 </div>
2674 <!-- _______________________________________________________________________ -->
2676 </div>
2680 </pre>
2683 unsigned division of its two arguments.</p>
2687 values. Both arguments must have identical types.</p>
2690 This instruction always performs an unsigned division to get the remainder.</p>
2691 <p>Note that unsigned integer remainder and signed integer remainder are
2696 </pre>
2698 </div>
2699 <!-- _______________________________________________________________________ -->
2702 </div>
2708 <pre>
2710 </pre>
2715 signed division of its two operands. This instruction can also take
2717 the elements must be integers.</p>
2723 values. Both arguments must have identical types.</p>
2730 a value. For more information about the difference, see <a
2732 Math Forum</a>. For a table of how this is implemented in various languages,
2735 <p>Note that signed integer remainder and unsigned integer remainder are
2737 <p>Taking the remainder of a division by zero leads to undefined behavior.
2738 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2740 (The remainder doesn't actually overflow, but this rule lets srem be
2741 implemented using instructions that return both the result of the division
2742 and the remainder.)</p>
2745 </pre>
2747 </div>
2748 <!-- _______________________________________________________________________ -->
2756 </pre>
2759 division of its two operands.</p>
2763 of floating point values. Both arguments must have identical types.</p>
2768 The remainder has the same sign as the dividend.</p>
2772 <pre>
2774 </pre>
2775 </div>
2777 <!-- ======================================================================= -->
2779 Operations</a> </div>
2781 <p>Bitwise binary operators are used to do various forms of
2782 bit-twiddling in a program. They are generally very efficient
2783 instructions and can commonly be strength reduced from other
2784 instructions. They require two operands of the same type, execute an operation on them,
2785 and produce a single value. The resulting value is the same type as its operands.</p>
2786 </div>
2788 <!-- _______________________________________________________________________ -->
2790 Instruction</a> </div>
2794 </pre>
2799 the left a specified number of bits.</p>
2820 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
2821 </pre>
2822 </div>
2823 <!-- _______________________________________________________________________ -->
2825 Instruction</a> </div>
2829 </pre>
2833 operand shifted to the right a specified number of bits with zero fill.</p>
2842 <p>This instruction always performs a logical shift right operation. The most
2843 significant bits of the result will be filled with zero bits after the
2850 <pre>
2856 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
2857 </pre>
2858 </div>
2860 <!-- _______________________________________________________________________ -->
2862 Instruction</a> </div>
2867 </pre>
2871 operand shifted to the right a specified number of bits with sign extension.</p>
2879 <p>This instruction always performs an arithmetic shift right operation,
2880 The most significant bits of the result will be filled with the sign bit
2887 <pre>
2893 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
2894 </pre>
2895 </div>
2897 <!-- _______________________________________________________________________ -->
2899 Instruction</a> </div>
2905 <pre>
2907 </pre>
2912 its two operands.</p>
2918 values. Both arguments must have identical types.</p>
2922 <p> </p>
2923 <div>
2925 <tbody>
2926 <tr>
2930 </tr>
2931 <tr>
2935 </tr>
2936 <tr>
2940 </tr>
2941 <tr>
2945 </tr>
2946 <tr>
2950 </tr>
2951 </tbody>
2952 </table>
2953 </div>
2955 <pre>
2959 </pre>
2960 </div>
2961 <!-- _______________________________________________________________________ -->
2966 </pre>
2969 or of its two operands.</p>
2974 values. Both arguments must have identical types.</p>
2977 <p> </p>
2978 <div>
2980 <tbody>
2981 <tr>
2985 </tr>
2986 <tr>
2990 </tr>
2991 <tr>
2995 </tr>
2996 <tr>
3000 </tr>
3001 <tr>
3005 </tr>
3006 </tbody>
3007 </table>
3008 </div>
3013 </pre>
3014 </div>
3015 <!-- _______________________________________________________________________ -->
3017 Instruction</a> </div>
3021 </pre>
3029 values. Both arguments must have identical types.</p>
3034 <p> </p>
3035 <div>
3037 <tbody>
3038 <tr>
3042 </tr>
3043 <tr>
3047 </tr>
3048 <tr>
3052 </tr>
3053 <tr>
3057 </tr>
3058 <tr>
3062 </tr>
3063 </tbody>
3064 </table>
3065 </div>
3066 <p> </p>
3072 </pre>
3073 </div>
3075 <!-- ======================================================================= -->
3078 </div>
3082 <p>LLVM supports several instructions to represent vector operations in a
3083 target-independent manner. These instructions cover the element-access and
3084 vector-specific operations needed to process vectors effectively. While LLVM
3085 does directly support these vector operations, many sophisticated algorithms
3086 will want to use target-specific intrinsics to take full advantage of a specific
3087 target.</p>
3089 </div>
3091 <!-- _______________________________________________________________________ -->
3094 </div>
3100 <pre>
3101 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3102 </pre>
3106 <p>
3108 element from a vector at a specified index.
3109 </p>
3114 <p>
3117 an index indicating the position from which to extract the element.
3118 The index may be a variable.</p>
3122 <p>
3123 The result is a scalar of the same type as the element type of
3126 results are undefined.
3127 </p>
3131 <pre>
3133 </pre>
3134 </div>
3137 <!-- _______________________________________________________________________ -->
3140 </div>
3146 <pre>
3147 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3148 </pre>
3152 <p>
3154 element into a vector at a specified index.
3155 </p>
3160 <p>
3163 scalar value whose type must equal the element type of the first
3164 operand. The third operand is an index indicating the position at
3165 which to insert the value. The index may be a variable.</p>
3169 <p>
3174 </p>
3178 <pre>
3180 </pre>
3181 </div>
3183 <!-- _______________________________________________________________________ -->
3186 </div>
3192 <pre>
3193 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3194 </pre>
3198 <p>
3200 from two input vectors, returning a vector with the same element type as
3201 the input and length that is the same as the shuffle mask.
3202 </p>
3206 <p>
3208 with types that match each other. The third argument is a shuffle mask whose
3209 element type is always 'i32'. The result of the instruction is a vector whose
3210 length is the same as the shuffle mask and whose element type is the same as
3211 the element type of the first two operands.
3212 </p>
3214 <p>
3215 The shuffle mask operand is required to be a constant vector with either
3216 constant integer or undef values.
3217 </p>
3221 <p>
3222 The elements of the two input vectors are numbered from left to right across
3223 both of the vectors. The shuffle mask operand specifies, for each element of
3224 the result vector, which element of the two input vectors the result element
3225 gets. The element selector may be undef (meaning "don't care") and the second
3226 operand may be undef if performing a shuffle from only one vector.
3227 </p>
3231 <pre>
3235 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3239 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > <i>; yields <8 x i32></i>
3240 </pre>
3241 </div>
3244 <!-- ======================================================================= -->
3247 </div>
3251 <p>LLVM supports several instructions for working with aggregate values.
3252 </p>
3254 </div>
3256 <!-- _______________________________________________________________________ -->
3259 </div>
3265 <pre>
3267 </pre>
3271 <p>
3273 or array element from an aggregate value.
3274 </p>
3279 <p>
3282 type. The operands are constant indices to specify which value to extract
3283 in a similar manner as indices in a
3285 </p>
3289 <p>
3290 The result is the value at the position in the aggregate specified by
3291 the index operands.
3292 </p>
3296 <pre>
3298 </pre>
3299 </div>
3302 <!-- _______________________________________________________________________ -->
3305 </div>
3311 <pre>
3312 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3313 </pre>
3317 <p>
3319 into a struct field or array element in an aggregate.
3320 </p>
3325 <p>
3328 The second operand is a first-class value to insert.
3329 The following operands are constant indices
3330 indicating the position at which to insert the value in a similar manner as
3331 indices in a
3333 The value to insert must have the same type as the value identified
3334 by the indices.
3335 </p>
3339 <p>
3343 </p>
3347 <pre>
3349 </pre>
3350 </div>
3353 <!-- ======================================================================= -->
3356 </div>
3360 <p>A key design point of an SSA-based representation is how it
3361 represents memory. In LLVM, no memory locations are in SSA form, which
3362 makes things very simple. This section describes how to read, write,
3363 allocate, and free memory in LLVM.</p>
3365 </div>
3367 <!-- _______________________________________________________________________ -->
3370 </div>
3376 <pre>
3377 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3378 </pre>
3383 heap and returns a pointer to it. The object is always allocated in the generic
3384 address space (address space zero).</p>
3390 bytes of memory from the operating system and returns a pointer of the
3391 appropriate type to the program. If "NumElements" is specified, it is the
3392 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3393 If a constant alignment is specified, the value result of the allocation is guaranteed to
3394 be aligned to at least that boundary. If not specified, or if zero, the target can
3395 choose to align the allocation on any convenient boundary.</p>
3402 a pointer is returned. The result of a zero byte allocation is undefined. The
3403 result is null if there is insufficient memory available.</p>
3407 <pre>
3415 </pre>
3417 <p>Note that the code generator does not yet respect the
3418 alignment value.</p>
3420 </div>
3422 <!-- _______________________________________________________________________ -->
3425 </div>
3431 <pre>
3433 </pre>
3438 memory heap to be reallocated in the future.</p>
3444 instruction.</p>
3448 <p>Access to the memory pointed to by the pointer is no longer defined
3449 after this instruction executes. If the pointer is null, the operation
3450 is a noop.</p>
3454 <pre>
3456 free [4 x i8]* %array
3457 </pre>
3458 </div>
3460 <!-- _______________________________________________________________________ -->
3463 </div>
3469 <pre>
3470 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3471 </pre>
3476 currently executing function, to be automatically released when this function
3477 returns to its caller. The object is always allocated in the generic address
3478 space (address space zero).</p>
3483 bytes of memory on the runtime stack, returning a pointer of the
3484 appropriate type to the program. If "NumElements" is specified, it is the
3485 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3486 If a constant alignment is specified, the value result of the allocation is guaranteed
3487 to be aligned to at least that boundary. If not specified, or if zero, the target
3488 can choose to align the allocation on any convenient boundary.</p>
3494 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3497 instruction is commonly used to represent automatic variables that must
3498 have an address available. When the function returns (either with the <tt><a
3500 instructions), the memory is reclaimed. Allocating zero bytes
3501 is legal, but the result is undefined.</p>
3505 <pre>
3510 </pre>
3511 </div>
3513 <!-- _______________________________________________________________________ -->
3515 Instruction</a> </div>
3518 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3523 address from which to load. The pointer must point to a <a
3528 instructions. </p>
3529 <p>
3530 The optional constant "align" argument specifies the alignment of the operation
3531 (that is, the alignment of the memory address). A value of 0 or an
3532 omitted "align" argument means that the operation has the preferential
3533 alignment for the target. It is the responsibility of the code emitter
3534 to ensure that the alignment information is correct. Overestimating
3535 the alignment results in an undefined behavior. Underestimating the
3536 alignment may produce less efficient code. An alignment of 1 is always
3537 safe.
3538 </p>
3540 <p>The location of memory pointed to is loaded. If the value being loaded
3541 is of scalar type then the number of bytes read does not exceed the minimum
3542 number of bytes needed to hold all bits of the type. For example, loading an
3545 is undefined if the value was not originally written using a store of the
3546 same type.</p>
3549 <a
3552 </pre>
3553 </div>
3554 <!-- _______________________________________________________________________ -->
3556 Instruction</a> </div>
3559 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3560 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3561 </pre>
3570 optimizer is not allowed to modify the number or order of execution of
3573 <p>
3574 The optional constant "align" argument specifies the alignment of the operation
3575 (that is, the alignment of the memory address). A value of 0 or an
3576 omitted "align" argument means that the operation has the preferential
3577 alignment for the target. It is the responsibility of the code emitter
3578 to ensure that the alignment information is correct. Overestimating
3579 the alignment results in an undefined behavior. Underestimating the
3580 alignment may produce less efficient code. An alignment of 1 is always
3581 safe.
3582 </p>
3587 written does not exceed the minimum number of bytes needed to hold all
3590 size that is not an integral number of bytes, it is unspecified what
3591 happens to the extra bits that do not belong to the type, but they will
3592 typically be overwritten.</p>
3597 </pre>
3598 </div>
3600 <!-- _______________________________________________________________________ -->
3603 </div>
3607 <pre>
3609 </pre>
3613 <p>
3615 subelement of an aggregate data structure. It performs address calculation only
3616 and does not access memory.</p>
3620 <p>The first argument is always a pointer, and forms the basis of the
3621 calculation. The remaining arguments are indices, that indicate which of the
3622 elements of the aggregate object are indexed. The interpretation of each index
3623 is dependent on the type being indexed into. The first index always indexes the
3624 pointer value given as the first argument, the second index indexes a value of
3625 the type pointed to (not necessarily the value directly pointed to, since the
3626 first index can be non-zero), etc. The first type indexed into must be a pointer
3627 value, subsequent types can be arrays, vectors and structs. Note that subsequent
3628 types being indexed into can never be pointers, since that would require loading
3629 the pointer before continuing calculation.</p>
3631 <p>The type of each index argument depends on the type it is indexing into.
3634 integers of any width are allowed (also non-constants).</p>
3636 <p>For example, let's consider a C code fragment and how it gets
3637 compiled to LLVM:</p>
3640 <pre>
3641 struct RT {
3642 char A;
3644 char C;
3645 };
3646 struct ST {
3647 int X;
3648 double Y;
3649 struct RT Z;
3650 };
3652 int *foo(struct ST *s) {
3654 }
3655 </pre>
3656 </div>
3661 <pre>
3665 define i32* %foo(%ST* %s) {
3666 entry:
3668 ret i32* %reg
3669 }
3670 </pre>
3671 </div>
3677 }</tt>' type, a structure. The second index indexes into the third element of
3679 i8 }</tt>' type, another structure. The third index indexes into the second
3681 array. The two dimensions of the array are subscripted into, yielding an
3685 <p>Note that it is perfectly legal to index partially through a
3686 structure, returning a pointer to an inner element. Because of this,
3687 the LLVM code for the given testcase is equivalent to:</p>
3689 <pre>
3690 define i32* %foo(%ST* %s) {
3696 ret i32* %t5
3697 }
3698 </pre>
3700 <p>Note that it is undefined to access an array out of bounds: array
3701 and pointer indexes must always be within the defined bounds of the
3702 array type when accessed with an instruction that dereferences the
3703 pointer (e.g. a load or store instruction). The one exception for
3704 this rule is zero length arrays. These arrays are defined to be
3705 accessible as variable length arrays, which requires access beyond the
3706 zero'th element.</p>
3708 <p>The getelementptr instruction is often confusing. For some more insight
3714 <pre>
3723 </pre>
3724 </div>
3726 <!-- ======================================================================= -->
3728 </div>
3730 <p>The instructions in this category are the conversion instructions (casting)
3731 which all take a single operand and a type. They perform various bit conversions
3732 on the operand.</p>
3733 </div>
3735 <!-- _______________________________________________________________________ -->
3738 </div>
3742 <pre>
3744 </pre>
3747 <p>
3749 </p>
3752 <p>
3760 <p>
3764 It will always truncate bits.</p>
3767 <pre>
3771 </pre>
3772 </div>
3774 <!-- _______________________________________________________________________ -->
3777 </div>
3781 <pre>
3783 </pre>
3804 <pre>
3807 </pre>
3808 </div>
3810 <!-- _______________________________________________________________________ -->
3813 </div>
3817 <pre>
3819 </pre>
3825 <p>
3833 <p>
3841 <pre>
3844 </pre>
3845 </div>
3847 <!-- _______________________________________________________________________ -->
3850 </div>
3856 <pre>
3858 </pre>
3879 <pre>
3882 </pre>
3883 </div>
3885 <!-- _______________________________________________________________________ -->
3888 </div>
3892 <pre>
3894 </pre>
3898 floating point value.</p>
3904 type must be smaller than the destination type.</p>
3914 <pre>
3917 </pre>
3918 </div>
3920 <!-- _______________________________________________________________________ -->
3923 </div>
3927 <pre>
3929 </pre>
3934 </p>
3947 the results are undefined.</p>
3950 <pre>
3954 </pre>
3955 </div>
3957 <!-- _______________________________________________________________________ -->
3960 </div>
3964 <pre>
3966 </pre>
3971 </p>
3984 the results are undefined.</p>
3987 <pre>
3991 </pre>
3992 </div>
3994 <!-- _______________________________________________________________________ -->
3997 </div>
4001 <pre>
4003 </pre>
4018 integer quantity and converts it to the corresponding floating point value. If
4019 the value cannot fit in the floating point value, the results are undefined.</p>
4022 <pre>
4025 </pre>
4026 </div>
4028 <!-- _______________________________________________________________________ -->
4031 </div>
4035 <pre>
4037 </pre>
4052 integer quantity and converts it to the corresponding floating point value. If
4053 the value cannot fit in the floating point value, the results are undefined.</p>
4056 <pre>
4059 </pre>
4060 </div>
4062 <!-- _______________________________________________________________________ -->
4065 </div>
4069 <pre>
4071 </pre>
4085 truncating or zero extending that value to the size of the integer type. If
4089 change.</p>
4092 <pre>
4095 </pre>
4096 </div>
4098 <!-- _______________________________________________________________________ -->
4101 </div>
4105 <pre>
4107 </pre>
4115 value to cast, and a type to cast it to, which must be a
4123 the size of a pointer then a zero extension is done. If they are the same size,
4127 <pre>
4131 </pre>
4132 </div>
4134 <!-- _______________________________________________________________________ -->
4137 </div>
4141 <pre>
4143 </pre>
4153 a non-aggregate first class value, and a type to cast it to, which must also be
4157 type is a pointer, the destination type must also be a pointer. This
4158 instruction supports bitwise conversion of vectors to integers and to vectors
4159 of other types (as long as they have the same size).</p>
4166 converted to other pointer types with this instruction. To convert pointers to
4171 <pre>
4175 </pre>
4176 </div>
4178 <!-- ======================================================================= -->
4182 instructions, which defy better classification.</p>
4183 </div>
4185 <!-- _______________________________________________________________________ -->
4187 </div>
4190 <pre> <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4191 </pre>
4194 a vector of boolean values based on comparison
4195 of its two integer, integer vector, or pointer operands.</p>
4198 the condition code indicating the kind of comparison to perform. It is not
4199 a value, just a keyword. The possible condition code are:
4200 </p>
4201 <ol>
4212 </ol>
4216 They must also be identical types.</p>
4220 yields either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt> result, as follows:
4221 </p>
4222 <ol>
4225 </li>
4244 </ol>
4246 values are compared as if they were integers.</p>
4247 <p>If the operands are integer vectors, then they are compared
4249 the same number of elements as the values being compared.
4251 </p>
4260 </pre>
4262 <p>Note that the code generator does not yet support vector types with
4265 </div>
4267 <!-- _______________________________________________________________________ -->
4269 </div>
4272 <pre> <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4273 </pre>
4276 or vector of boolean values based on comparison
4277 of its operands.</p>
4278 <p>
4279 If the operands are floating point scalars, then the result
4281 </p>
4282 <p>If the operands are floating point vectors, then the result type
4283 is a vector of boolean with the same number of elements as the
4284 operands being compared.</p>
4287 the condition code indicating the kind of comparison to perform. It is not
4288 a value, just a keyword. The possible condition code are:</p>
4289 <ol>
4306 </ol>
4312 They must have identical types.</p>
4316 If the operands are vectors, then the vectors are compared
4317 element by element.
4318 Each comparison performed
4320 <ol>
4349 </ol>
4356 </pre>
4358 <p>Note that the code generator does not yet support vector types with
4361 </div>
4363 <!-- _______________________________________________________________________ -->
4366 </div>
4369 <pre> <result> = vicmp <cond> <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4370 </pre>
4373 element-wise comparison of its two integer vector operands.</p>
4376 the condition code indicating the kind of comparison to perform. It is not
4377 a value, just a keyword. The possible condition code are:</p>
4378 <ol>
4389 </ol>
4396 identical type as the values being compared. The most significant bit in each
4398 otherwise. All other bits of the result are undefined. The condition codes
4403 <pre>
4404 <result> = vicmp eq <2 x i32> < i32 4, i32 0>, < i32 5, i32 0> <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
4405 <result> = vicmp ult <2 x i8 > < i8 1, i8 2>, < i8 2, i8 2 > <i>; yields: result=<2 x i8> < i8 -1, i8 0 ></i>
4406 </pre>
4407 </div>
4409 <!-- _______________________________________________________________________ -->
4412 </div>
4418 element-wise comparison of its two floating point vector operands. The output
4419 elements have the same width as the input elements.</p>
4422 the condition code indicating the kind of comparison to perform. It is not
4423 a value, just a keyword. The possible condition code are:</p>
4424 <ol>
4441 </ol>
4444 types.</p>
4449 an identical number of elements as the values being compared, and each element
4450 having identical with to the width of the floating point elements. The most
4451 significant bit in each element is 1 if the element-wise comparison evaluates to
4452 true, and is 0 otherwise. All other bits of the result are undefined. The
4453 condition codes are evaluated identically to the
4457 <pre>
4459 <result> = vfcmp oeq <2 x float> < float 4, float 0 >, < float 5, float 0 >
4462 <result> = vfcmp ult <2 x double> < double 1, double 2 >, < double 2, double 2>
4463 </pre>
4464 </div>
4466 <!-- _______________________________________________________________________ -->
4469 </div>
4478 the SSA graph representing the function.</p>
4481 <p>The type of the incoming values is specified with the first type
4483 as arguments, with one pair for each predecessor basic block of the
4485 type may be used as the value arguments to the PHI node. Only labels
4486 may be used as the label arguments.</p>
4488 <p>There must be no non-phi instructions between the start of a basic
4489 block and the PHI instructions: i.e. PHI instructions must be first in
4490 a basic block.</p>
4495 specified by the pair corresponding to the predecessor basic block that executed
4496 just prior to the current block.</p>
4499 <pre>
4500 Loop: ; Infinite loop that counts from 0 on up...
4501 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4502 %nextindvar = add i32 %indvar, 1
4503 br label %Loop
4504 </pre>
4505 </div>
4507 <!-- _______________________________________________________________________ -->
4510 </div>
4516 <pre>
4517 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4520 </pre>
4524 <p>
4526 condition, without branching.
4527 </p>
4532 <p>
4534 a vector of 'i1' values indicating the
4536 type. If the val1/val2 are vectors and
4537 the condition is a scalar, then entire vectors are selected, not
4538 individual elements.
4539 </p>
4543 <p>
4544 If the condition is an i1 and it evaluates to 1, the instruction returns the first
4545 value argument; otherwise, it returns the second value argument.
4546 </p>
4547 <p>
4548 If the condition is a vector of i1, then the value arguments must
4549 be vectors of the same size, and the selection is done element
4550 by element.
4551 </p>
4555 <pre>
4557 </pre>
4559 <p>Note that the code generator does not yet support conditions
4560 with vector type.</p>
4562 </div>
4565 <!-- _______________________________________________________________________ -->
4568 </div>
4573 <pre>
4574 <result> = [tail] call [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] <ty> [<fnty>*] <fnptrval>(<function args>) [<a href="#fnattrs">fn attrs</a>]
4575 </pre>
4585 <ol>
4586 <li>
4588 any allocas or varargs in the caller. If the "tail" marker is present, the
4589 function call is eligible for tail call optimization. Note that calls may
4592 </li>
4593 <li>
4595 convention</a> the call should use. If none is specified, the call defaults
4596 to using C calling conventions.</p>
4597 </li>
4599 <li>
4603 </li>
4605 <li>
4607 the type of the return value. Functions that return no value are marked
4609 </li>
4610 <li>
4612 value being invoked. The argument types must match the types implied by
4613 this signature. This type can be omitted if the function is not varargs
4614 and if the function type does not return a pointer to a function.</p>
4615 </li>
4616 <li>
4618 be invoked. In most cases, this is a direct function invocation, but
4620 to function value.</p>
4621 </li>
4622 <li>
4624 function signature argument types. All arguments must be of
4626 indicates the function accepts a variable number of arguments, the extra
4627 arguments can be specified.</p>
4628 </li>
4629 <li>
4633 </li>
4634 </ol>
4639 transfer to a specified function, with its incoming arguments bound to
4641 instruction in the called function, control flow continues with the
4642 instruction after the function call, and the return value of the
4643 function is bound to the result argument.</p>
4647 <pre>
4648 %retval = call i32 @test(i32 %argc)
4652 call void %foo(i8 97 signext)
4654 %struct.A = type { i32, i8 }
4660 </pre>
4662 </div>
4664 <!-- _______________________________________________________________________ -->
4667 </div>
4673 <pre>
4675 </pre>
4680 the "variable argument" area of a function call. It is used to implement the
4686 the argument. It returns a value of the specified argument type and
4698 <p>It is legal for this instruction to be called in a function which does not
4700 function.</p>
4704 argument.</p>
4710 <p>Note that the code generator does not yet fully support va_arg
4711 on many targets. Also, it does not currently support va_arg with
4712 aggregate types on any target.</p>
4714 </div>
4716 <!-- *********************************************************************** -->
4718 <!-- *********************************************************************** -->
4723 well known names and semantics and are required to follow certain restrictions.
4724 Overall, these intrinsics represent an extension mechanism for the LLVM
4725 language that does not require changing all of the transformations in LLVM when
4726 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4729 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4730 begin with this prefix. Intrinsic functions must always be external functions:
4731 you cannot define the body of intrinsic functions. Intrinsic functions may
4732 only be used in call or invoke instructions: it is illegal to take the address
4733 of an intrinsic function. Additionally, because intrinsic functions are part
4734 of the LLVM language, it is required if any are added that they be documented
4735 here.</p>
4737 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4738 a family of functions that perform the same operation but on different data
4739 types. Because LLVM can represent over 8 million different integer types,
4740 overloading is used commonly to allow an intrinsic function to operate on any
4741 integer type. One or more of the argument types or the result type can be
4742 overloaded to accept any integer type. Argument types may also be defined as
4743 exactly matching a previous argument's type or the result type. This allows an
4744 intrinsic function which accepts multiple arguments, but needs all of them to
4745 be of the same type, to only be overloaded with respect to a single argument or
4746 the result.</p>
4748 <p>Overloaded intrinsics will have the names of its overloaded argument types
4749 encoded into its function name, each preceded by a period. Only those types
4750 which are overloaded result in a name suffix. Arguments whose type is matched
4752 take an integer of any width and returns an integer of exactly the same integer
4753 width. This leads to a family of functions such as
4755 Only one type, the return type, is overloaded, and only one type suffix is
4756 required. Because the argument's type is matched against the return type, it
4757 does not require its own name suffix.</p>
4759 <p>To learn how to add an intrinsic function, please see the
4761 </p>
4763 </div>
4765 <!-- ======================================================================= -->
4768 </div>
4774 intrinsic functions. These functions are related to the similarly
4777 <p>All of these functions operate on arguments that use a
4778 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4779 language reference manual does not define what this type is, so all
4780 transformations should be prepared to handle these functions regardless of
4781 the type used.</p>
4784 instruction and the variable argument handling intrinsic functions are
4785 used.</p>
4788 <pre>
4789 define i32 @test(i32 %X, ...) {
4790 ; Initialize variable argument processing
4791 %ap = alloca i8*
4792 %ap2 = bitcast i8** %ap to i8*
4793 call void @llvm.va_start(i8* %ap2)
4795 ; Read a single integer argument
4796 %tmp = va_arg i8** %ap, i32
4798 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4799 %aq = alloca i8*
4800 %aq2 = bitcast i8** %aq to i8*
4801 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4802 call void @llvm.va_end(i8* %aq2)
4804 ; Stop processing of arguments.
4805 call void @llvm.va_end(i8* %ap2)
4806 ret i32 %tmp
4807 }
4809 declare void @llvm.va_start(i8*)
4810 declare void @llvm.va_copy(i8*, i8*)
4811 declare void @llvm.va_end(i8*)
4812 </pre>
4813 </div>
4815 </div>
4817 <!-- _______________________________________________________________________ -->
4820 </div>
4838 macro available in C. In a target-dependent way, it initializes the
4842 last argument of the function as the compiler can figure that out.</p>
4844 </div>
4846 <!-- _______________________________________________________________________ -->
4849 </div>
4867 macro available in C. In a target-dependent way, it destroys the
4873 </div>
4875 <!-- _______________________________________________________________________ -->
4878 </div>
4884 <pre>
4886 </pre>
4891 from the source argument list to the destination argument list.</p>
4902 macro available in C. In a target-dependent way, it copies the source
4905 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4906 example, memory allocation.</p>
4908 </div>
4910 <!-- ======================================================================= -->
4913 </div>
4917 <p>
4919 Collection</a> (GC) requires the implementation and generation of these
4920 intrinsics.
4924 Front-ends for type-safe garbage collected languages should generate these
4925 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4927 </p>
4929 <p>The garbage collection intrinsics only operate on objects in the generic
4930 address space (address space zero).</p>
4932 </div>
4934 <!-- _______________________________________________________________________ -->
4937 </div>
4943 <pre>
4944 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4945 </pre>
4950 the code generator, and allows some metadata to be associated with it.</p>
4954 <p>The first argument specifies the address of a stack object that contains the
4955 root pointer. The second pointer (which must be either a constant or a global
4956 value address) contains the meta-data to be associated with the root.</p>
4961 location. At compile-time, the code generator generates information to allow
4966 </div>
4969 <!-- _______________________________________________________________________ -->
4972 </div>
4978 <pre>
4979 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4980 </pre>
4985 locations, allowing garbage collector implementations that require read
4986 barriers.</p>
4990 <p>The second argument is the address to read from, which should be an address
4991 allocated from the garbage collector. The first object is a pointer to the
4992 start of the referenced object, if needed by the language runtime (otherwise
4993 null).</p>
4998 instruction, but may be replaced with substantially more complex code by the
5003 </div>
5006 <!-- _______________________________________________________________________ -->
5009 </div>
5015 <pre>
5016 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5017 </pre>
5022 locations, allowing garbage collector implementations that require write
5023 barriers (such as generational or reference counting collectors).</p>
5027 <p>The first argument is the reference to store, the second is the start of the
5028 object to store it to, and the third is the address of the field of Obj to
5029 store to. If the runtime does not require a pointer to the object, Obj may be
5030 null.</p>
5035 instruction, but may be replaced with substantially more complex code by the
5040 </div>
5044 <!-- ======================================================================= -->
5047 </div>
5050 <p>
5051 These intrinsics are provided by LLVM to expose special features that may only
5052 be implemented with code generator support.
5053 </p>
5055 </div>
5057 <!-- _______________________________________________________________________ -->
5060 </div>
5065 <pre>
5067 </pre>
5071 <p>
5073 target-specific value indicating the return address of the current function
5074 or one of its callers.
5075 </p>
5079 <p>
5080 The argument to this intrinsic indicates which function to return the address
5081 for. Zero indicates the calling function, one indicates its caller, etc. The
5083 </p>
5087 <p>
5089 the return address of the specified call frame, or zero if it cannot be
5090 identified. The value returned by this intrinsic is likely to be incorrect or 0
5091 for arguments other than zero, so it should only be used for debugging purposes.
5092 </p>
5094 <p>
5095 Note that calling this intrinsic does not prevent function inlining or other
5096 aggressive transformations, so the value returned may not be that of the obvious
5097 source-language caller.
5098 </p>
5099 </div>
5102 <!-- _______________________________________________________________________ -->
5105 </div>
5110 <pre>
5112 </pre>
5116 <p>
5118 target-specific frame pointer value for the specified stack frame.
5119 </p>
5123 <p>
5124 The argument to this intrinsic indicates which function to return the frame
5125 pointer for. Zero indicates the calling function, one indicates its caller,
5127 </p>
5131 <p>
5133 the frame address of the specified call frame, or zero if it cannot be
5134 identified. The value returned by this intrinsic is likely to be incorrect or 0
5135 for arguments other than zero, so it should only be used for debugging purposes.
5136 </p>
5138 <p>
5139 Note that calling this intrinsic does not prevent function inlining or other
5140 aggressive transformations, so the value returned may not be that of the obvious
5141 source-language caller.
5142 </p>
5143 </div>
5145 <!-- _______________________________________________________________________ -->
5148 </div>
5153 <pre>
5154 declare i8 *@llvm.stacksave()
5155 </pre>
5159 <p>
5163 features like scoped automatic variable sized arrays in C99.
5164 </p>
5168 <p>
5169 This intrinsic returns a opaque pointer value that can be passed to <a
5176 </p>
5178 </div>
5180 <!-- _______________________________________________________________________ -->
5183 </div>
5188 <pre>
5189 declare void @llvm.stackrestore(i8 * %ptr)
5190 </pre>
5194 <p>
5196 the function stack to the state it was in when the corresponding <a
5198 useful for implementing language features like scoped automatic variable sized
5199 arrays in C99.
5200 </p>
5204 <p>
5206 </p>
5208 </div>
5211 <!-- _______________________________________________________________________ -->
5214 </div>
5219 <pre>
5221 </pre>
5226 <p>
5228 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
5229 no
5230 effect on the behavior of the program but can change its performance
5231 characteristics.
5232 </p>
5236 <p>
5242 </p>
5246 <p>
5247 This intrinsic does not modify the behavior of the program. In particular,
5248 prefetches cannot trap and do not produce a value. On targets that support this
5249 intrinsic, the prefetch can provide hints to the processor cache for better
5250 performance.
5251 </p>
5253 </div>
5255 <!-- _______________________________________________________________________ -->
5258 </div>
5263 <pre>
5265 </pre>
5270 <p>
5272 (PC) in a region of
5273 code to simulators and other tools. The method is target specific, but it is
5274 expected that the marker will use exported symbols to transmit the PC of the
5275 marker.
5276 The marker makes no guarantees that it will remain with any specific instruction
5277 after optimizations. It is possible that the presence of a marker will inhibit
5278 optimizations. The intended use is to be inserted after optimizations to allow
5279 correlations of simulation runs.
5280 </p>
5284 <p>
5286 </p>
5290 <p>
5291 This intrinsic does not modify the behavior of the program. Backends that do not
5292 support this intrinisic may ignore it.
5293 </p>
5295 </div>
5297 <!-- _______________________________________________________________________ -->
5300 </div>
5305 <pre>
5306 declare i64 @llvm.readcyclecounter( )
5307 </pre>
5312 <p>
5314 counter register (or similar low latency, high accuracy clocks) on those targets
5315 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
5316 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
5317 should only be used for small timings.
5318 </p>
5322 <p>
5323 When directly supported, reading the cycle counter should not modify any memory.
5324 Implementations are allowed to either return a application specific value or a
5325 system wide value. On backends without support, this is lowered to a constant 0.
5326 </p>
5328 </div>
5330 <!-- ======================================================================= -->
5333 </div>
5336 <p>
5337 LLVM provides intrinsics for a few important standard C library functions.
5338 These intrinsics allow source-language front-ends to pass information about the
5339 alignment of the pointer arguments to the code generator, providing opportunity
5340 for more efficient code generation.
5341 </p>
5343 </div>
5345 <!-- _______________________________________________________________________ -->
5348 </div>
5353 <p>This is an overloaded intrinsic. You can use llvm.memcpy on any integer bit
5354 width. Not all targets support all bit widths however.</p>
5355 <pre>
5364 </pre>
5368 <p>
5370 location to the destination location.
5371 </p>
5373 <p>
5375 intrinsics do not return a value, and takes an extra alignment argument.
5376 </p>
5380 <p>
5381 The first argument is a pointer to the destination, the second is a pointer to
5382 the source. The third argument is an integer argument
5383 specifying the number of bytes to copy, and the fourth argument is the alignment
5384 of the source and destination locations.
5385 </p>
5387 <p>
5389 the caller guarantees that both the source and destination pointers are aligned
5390 to that boundary.
5391 </p>
5395 <p>
5397 location to the destination location, which are not allowed to overlap. It
5398 copies "len" bytes of memory over. If the argument is known to be aligned to
5399 some boundary, this can be specified as the fourth argument, otherwise it should
5401 </p>
5402 </div>
5405 <!-- _______________________________________________________________________ -->
5408 </div>
5413 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5414 width. Not all targets support all bit widths however.</p>
5415 <pre>
5424 </pre>
5428 <p>
5430 location to the destination location. It is similar to the
5432 </p>
5434 <p>
5436 intrinsics do not return a value, and takes an extra alignment argument.
5437 </p>
5441 <p>
5442 The first argument is a pointer to the destination, the second is a pointer to
5443 the source. The third argument is an integer argument
5444 specifying the number of bytes to copy, and the fourth argument is the alignment
5445 of the source and destination locations.
5446 </p>
5448 <p>
5450 the caller guarantees that the source and destination pointers are aligned to
5451 that boundary.
5452 </p>
5456 <p>
5458 location to the destination location, which may overlap. It
5459 copies "len" bytes of memory over. If the argument is known to be aligned to
5460 some boundary, this can be specified as the fourth argument, otherwise it should
5462 </p>
5463 </div>
5466 <!-- _______________________________________________________________________ -->
5469 </div>
5474 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5475 width. Not all targets support all bit widths however.</p>
5476 <pre>
5485 </pre>
5489 <p>
5491 byte value.
5492 </p>
5494 <p>
5496 does not return a value, and takes an extra alignment argument.
5497 </p>
5501 <p>
5502 The first argument is a pointer to the destination to fill, the second is the
5503 byte value to fill it with, the third argument is an integer
5504 argument specifying the number of bytes to fill, and the fourth argument is the
5505 known alignment of destination location.
5506 </p>
5508 <p>
5510 the caller guarantees that the destination pointer is aligned to that boundary.
5511 </p>
5515 <p>
5517 the
5518 destination location. If the argument is known to be aligned to some boundary,
5519 this can be specified as the fourth argument, otherwise it should be set to 0 or
5520 1.
5521 </p>
5522 </div>
5525 <!-- _______________________________________________________________________ -->
5528 </div>
5534 floating point or vector of floating point type. Not all targets support all
5535 types however.</p>
5536 <pre>
5537 declare float @llvm.sqrt.f32(float %Val)
5538 declare double @llvm.sqrt.f64(double %Val)
5539 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5540 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5541 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5542 </pre>
5546 <p>
5550 negative numbers other than -0.0 (which allows for better optimization, because
5552 defined to return -0.0 like IEEE sqrt.
5553 </p>
5557 <p>
5558 The argument and return value are floating point numbers of the same type.
5559 </p>
5563 <p>
5564 This function returns the sqrt of the specified operand if it is a nonnegative
5565 floating point number.
5566 </p>
5567 </div>
5569 <!-- _______________________________________________________________________ -->
5572 </div>
5578 floating point or vector of floating point type. Not all targets support all
5579 types however.</p>
5580 <pre>
5581 declare float @llvm.powi.f32(float %Val, i32 %power)
5582 declare double @llvm.powi.f64(double %Val, i32 %power)
5583 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5584 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5585 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5586 </pre>
5590 <p>
5592 specified (positive or negative) power. The order of evaluation of
5593 multiplications is not defined. When a vector of floating point type is
5594 used, the second argument remains a scalar integer value.
5595 </p>
5599 <p>
5600 The second argument is an integer power, and the first is a value to raise to
5601 that power.
5602 </p>
5606 <p>
5607 This function returns the first value raised to the second power with an
5608 unspecified sequence of rounding operations.</p>
5609 </div>
5611 <!-- _______________________________________________________________________ -->
5614 </div>
5620 floating point or vector of floating point type. Not all targets support all
5621 types however.</p>
5622 <pre>
5623 declare float @llvm.sin.f32(float %Val)
5624 declare double @llvm.sin.f64(double %Val)
5625 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5626 declare fp128 @llvm.sin.f128(fp128 %Val)
5627 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5628 </pre>
5632 <p>
5634 </p>
5638 <p>
5639 The argument and return value are floating point numbers of the same type.
5640 </p>
5644 <p>
5645 This function returns the sine of the specified operand, returning the
5647 conditions in the same way.</p>
5648 </div>
5650 <!-- _______________________________________________________________________ -->
5653 </div>
5659 floating point or vector of floating point type. Not all targets support all
5660 types however.</p>
5661 <pre>
5662 declare float @llvm.cos.f32(float %Val)
5663 declare double @llvm.cos.f64(double %Val)
5664 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5665 declare fp128 @llvm.cos.f128(fp128 %Val)
5666 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5667 </pre>
5671 <p>
5673 </p>
5677 <p>
5678 The argument and return value are floating point numbers of the same type.
5679 </p>
5683 <p>
5684 This function returns the cosine of the specified operand, returning the
5686 conditions in the same way.</p>
5687 </div>
5689 <!-- _______________________________________________________________________ -->
5692 </div>
5698 floating point or vector of floating point type. Not all targets support all
5699 types however.</p>
5700 <pre>
5701 declare float @llvm.pow.f32(float %Val, float %Power)
5702 declare double @llvm.pow.f64(double %Val, double %Power)
5703 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5704 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5705 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5706 </pre>
5710 <p>
5712 specified (positive or negative) power.
5713 </p>
5717 <p>
5718 The second argument is a floating point power, and the first is a value to
5719 raise to that power.
5720 </p>
5724 <p>
5725 This function returns the first value raised to the second power,
5726 returning the
5728 conditions in the same way.</p>
5729 </div>
5732 <!-- ======================================================================= -->
5735 </div>
5738 <p>
5739 LLVM provides intrinsics for a few important bit manipulation operations.
5740 These allow efficient code generation for some algorithms.
5741 </p>
5743 </div>
5745 <!-- _______________________________________________________________________ -->
5748 </div>
5753 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5755 <pre>
5759 </pre>
5763 <p>
5765 values with an even number of bytes (positive multiple of 16 bits). These are
5766 useful for performing operations on data that is not in the target's native
5767 byte order.
5768 </p>
5772 <p>
5775 intrinsic returns an i32 value that has the four bytes of the input i32
5780 </p>
5782 </div>
5784 <!-- _______________________________________________________________________ -->
5787 </div>
5792 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5793 width. Not all targets support all bit widths however.</p>
5794 <pre>
5800 </pre>
5804 <p>
5806 value.
5807 </p>
5811 <p>
5812 The only argument is the value to be counted. The argument may be of any
5813 integer type. The return type must match the argument type.
5814 </p>
5818 <p>
5820 </p>
5821 </div>
5823 <!-- _______________________________________________________________________ -->
5826 </div>
5832 integer bit width. Not all targets support all bit widths however.</p>
5833 <pre>
5839 </pre>
5843 <p>
5845 leading zeros in a variable.
5846 </p>
5850 <p>
5851 The only argument is the value to be counted. The argument may be of any
5852 integer type. The return type must match the argument type.
5853 </p>
5857 <p>
5859 in a variable. If the src == 0 then the result is the size in bits of the type
5861 </p>
5862 </div>
5866 <!-- _______________________________________________________________________ -->
5869 </div>
5875 integer bit width. Not all targets support all bit widths however.</p>
5876 <pre>
5882 </pre>
5886 <p>
5888 trailing zeros.
5889 </p>
5893 <p>
5894 The only argument is the value to be counted. The argument may be of any
5895 integer type. The return type must match the argument type.
5896 </p>
5900 <p>
5902 in a variable. If the src == 0 then the result is the size in bits of the type
5904 </p>
5905 </div>
5907 <!-- _______________________________________________________________________ -->
5910 </div>
5916 on any integer bit width.</p>
5917 <pre>
5918 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5919 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5920 </pre>
5924 range of bits from an integer value and returns them in the same bit width as
5925 the original value.</p>
5929 any bit width but they must have the same bit width. The second and third
5936 operates in forward mode.</p>
5940 <ol>
5944 to determine the number of bits to retain.</li>
5947 </ol>
5948 <p>In reverse mode, a similar computation is made except that the bits are
5953 </div>
5957 </div>
5963 on any integer bit width.</p>
5964 <pre>
5965 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5966 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5967 </pre>
5971 of bits in an integer value with another integer value. It returns the integer
5972 with the replaced bits.</p>
5979 type since they specify only a bit index.</p>
5985 operates in forward mode.</p>
5988 truncating it down to the size of the replacement area or zero extending it
5989 up to that size.</p>
5996 <p>In reverse mode, a similar computation is made except that the bits are
6002 <pre>
6008 </pre>
6010 </div>
6012 <!-- ======================================================================= -->
6015 </div>
6018 <p>
6019 LLVM provides intrinsics for some arithmetic with overflow operations.
6020 </p>
6022 </div>
6024 <!-- _______________________________________________________________________ -->
6027 </div>
6034 on any integer bit width.</p>
6036 <pre>
6037 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6038 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6039 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6040 </pre>
6045 a signed addition of the two arguments, and indicate whether an overflow
6046 occurred during the signed summation.</p>
6050 <p>The arguments (%a and %b) and the first element of the result structure may
6051 be of integer types of any bit width, but they must have the same bit width. The
6058 a signed addition of the two variables. They return a structure — the
6059 first element of which is the signed summation, and the second element of which
6060 is a bit specifying if the signed summation resulted in an overflow.</p>
6063 <pre>
6064 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6065 %sum = extractvalue {i32, i1} %res, 0
6066 %obit = extractvalue {i32, i1} %res, 1
6067 br i1 %obit, label %overflow, label %normal
6068 </pre>
6070 </div>
6072 <!-- _______________________________________________________________________ -->
6075 </div>
6082 on any integer bit width.</p>
6084 <pre>
6085 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6086 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6087 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6088 </pre>
6093 an unsigned addition of the two arguments, and indicate whether a carry occurred
6094 during the unsigned summation.</p>
6098 <p>The arguments (%a and %b) and the first element of the result structure may
6099 be of integer types of any bit width, but they must have the same bit width. The
6106 an unsigned addition of the two arguments. They return a structure — the
6107 first element of which is the sum, and the second element of which is a bit
6108 specifying if the unsigned summation resulted in a carry.</p>
6111 <pre>
6112 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6113 %sum = extractvalue {i32, i1} %res, 0
6114 %obit = extractvalue {i32, i1} %res, 1
6115 br i1 %obit, label %carry, label %normal
6116 </pre>
6118 </div>
6120 <!-- _______________________________________________________________________ -->
6123 </div>
6130 on any integer bit width.</p>
6132 <pre>
6133 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6134 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6135 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6136 </pre>
6141 a signed subtraction of the two arguments, and indicate whether an overflow
6142 occurred during the signed subtraction.</p>
6146 <p>The arguments (%a and %b) and the first element of the result structure may
6147 be of integer types of any bit width, but they must have the same bit width. The
6154 a signed subtraction of the two arguments. They return a structure — the
6155 first element of which is the subtraction, and the second element of which is a bit
6156 specifying if the signed subtraction resulted in an overflow.</p>
6159 <pre>
6160 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6161 %sum = extractvalue {i32, i1} %res, 0
6162 %obit = extractvalue {i32, i1} %res, 1
6163 br i1 %obit, label %overflow, label %normal
6164 </pre>
6166 </div>
6168 <!-- _______________________________________________________________________ -->
6171 </div>
6178 on any integer bit width.</p>
6180 <pre>
6181 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6182 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6183 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6184 </pre>
6189 an unsigned subtraction of the two arguments, and indicate whether an overflow
6190 occurred during the unsigned subtraction.</p>
6194 <p>The arguments (%a and %b) and the first element of the result structure may
6195 be of integer types of any bit width, but they must have the same bit width. The
6202 an unsigned subtraction of the two arguments. They return a structure — the
6203 first element of which is the subtraction, and the second element of which is a bit
6204 specifying if the unsigned subtraction resulted in an overflow.</p>
6207 <pre>
6208 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6209 %sum = extractvalue {i32, i1} %res, 0
6210 %obit = extractvalue {i32, i1} %res, 1
6211 br i1 %obit, label %overflow, label %normal
6212 </pre>
6214 </div>
6216 <!-- _______________________________________________________________________ -->
6219 </div>
6226 on any integer bit width.</p>
6228 <pre>
6229 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6230 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6231 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6232 </pre>
6237 a signed multiplication of the two arguments, and indicate whether an overflow
6238 occurred during the signed multiplication.</p>
6242 <p>The arguments (%a and %b) and the first element of the result structure may
6243 be of integer types of any bit width, but they must have the same bit width. The
6250 a signed multiplication of the two arguments. They return a structure —
6251 the first element of which is the multiplication, and the second element of
6252 which is a bit specifying if the signed multiplication resulted in an
6253 overflow.</p>
6256 <pre>
6257 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6258 %sum = extractvalue {i32, i1} %res, 0
6259 %obit = extractvalue {i32, i1} %res, 1
6260 br i1 %obit, label %overflow, label %normal
6261 </pre>
6263 </div>
6265 <!-- _______________________________________________________________________ -->
6268 </div>
6275 on any integer bit width.</p>
6277 <pre>
6278 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6279 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6280 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6281 </pre>
6286 actively being fixed, but it should not currently be used!</i></p>
6289 a unsigned multiplication of the two arguments, and indicate whether an overflow
6290 occurred during the unsigned multiplication.</p>
6294 <p>The arguments (%a and %b) and the first element of the result structure may
6295 be of integer types of any bit width, but they must have the same bit width. The
6298 multiplication.</p>
6303 an unsigned multiplication of the two arguments. They return a structure —
6304 the first element of which is the multiplication, and the second element of
6305 which is a bit specifying if the unsigned multiplication resulted in an
6306 overflow.</p>
6309 <pre>
6310 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6311 %sum = extractvalue {i32, i1} %res, 0
6312 %obit = extractvalue {i32, i1} %res, 1
6313 br i1 %obit, label %overflow, label %normal
6314 </pre>
6316 </div>
6318 <!-- ======================================================================= -->
6321 </div>
6324 <p>
6326 are described in the <a
6328 Debugging</a> document.
6329 </p>
6330 </div>
6333 <!-- ======================================================================= -->
6336 </div>
6339 <p> The LLVM exception handling intrinsics (which all start with
6343 </div>
6345 <!-- ======================================================================= -->
6348 </div>
6351 <p>
6352 This intrinsic makes it possible to excise one parameter, marked with
6354 function pointer lacking the nest parameter - the caller does not need
6355 to provide a value for it. Instead, the value to use is stored in
6356 advance in a "trampoline", a block of memory usually allocated
6357 on the stack, which also contains code to splice the nest value into the
6358 argument list. This is used to implement the GCC nested function address
6359 extension.
6360 </p>
6361 <p>
6362 For example, if the function is
6365 <pre>
6368 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6369 %fp = bitcast i8* %p to i32 (i32, i32)*
6370 </pre>
6373 </div>
6375 <!-- _______________________________________________________________________ -->
6378 </div>
6381 <pre>
6383 </pre>
6385 <p>
6387 and returns a function pointer suitable for executing it.
6388 </p>
6390 <p>
6393 and sufficiently aligned block of memory; this memory is written to by the
6394 intrinsic. Note that the size and the alignment are target-specific - LLVM
6395 currently provides no portable way of determining them, so a front-end that
6396 generates this intrinsic needs to have some target-specific knowledge.
6398 </p>
6400 <p>
6402 dependent code, turning it into a function. A pointer to this function is
6403 returned, but needs to be bitcast to an
6405 before being called. The new function's signature is the same as that of
6408 of pointer type. Calling the new function is equivalent to calling
6412 modified, then the effect of any later call to the returned function pointer is
6413 undefined.
6414 </p>
6415 </div>
6417 <!-- ======================================================================= -->
6420 </div>
6423 <p>
6424 These intrinsic functions expand the "universal IR" of LLVM to represent
6425 hardware constructs for atomic operations and memory synchronization. This
6426 provides an interface to the hardware, not an interface to the programmer. It
6427 is aimed at a low enough level to allow any programming models or APIs
6428 (Application Programming Interfaces) which
6429 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
6430 hardware behavior. Just as hardware provides a "universal IR" for source
6431 languages, it also provides a starting point for developing a "universal"
6432 atomic operation and synchronization IR.
6433 </p>
6434 <p>
6436 software transaction memory systems, atomic primitives, and intrinsic
6437 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6438 application libraries. The hardware interface provided by LLVM should allow
6439 a clean implementation of all of these APIs and parallel programming models.
6440 No one model or paradigm should be selected above others unless the hardware
6441 itself ubiquitously does so.
6443 </p>
6444 </div>
6446 <!-- _______________________________________________________________________ -->
6449 </div>
6452 <pre>
6456 </pre>
6458 <p>
6460 specific pairs of memory access types.
6461 </p>
6463 <p>
6465 The first four arguments enables a specific barrier as listed below. The fith
6466 argument specifies that the barrier applies to io or device or uncached memory.
6468 </p>
6469 <ul>
6475 </ul>
6477 <p>
6478 This intrinsic causes the system to enforce some ordering constraints upon
6479 the loads and stores of the program. This barrier does not indicate
6481 which they occur. For any of the specified pairs of load and store operations
6482 (f.ex. load-load, or store-load), all of the first operations preceding the
6483 barrier will complete before any of the second operations succeeding the
6484 barrier begin. Specifically the semantics for each pairing is as follows:
6485 </p>
6486 <ul>
6488 after the barrier begins.</li>
6491 store after the barrier begins.</li>
6493 store after the barrier begins.</li>
6495 load after the barrier begins.</li>
6496 </ul>
6497 <p>
6498 These semantics are applied with a logical "and" behavior when more than one
6499 is enabled in a single memory barrier intrinsic.
6500 </p>
6501 <p>
6502 Backends may implement stronger barriers than those requested when they do not
6503 support as fine grained a barrier as requested. Some architectures do not
6504 need all types of barriers and on such architectures, these become noops.
6505 </p>
6507 <pre>
6508 %ptr = malloc i32
6509 store i32 4, %ptr
6512 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6515 </pre>
6516 </div>
6518 <!-- _______________________________________________________________________ -->
6521 </div>
6524 <p>
6526 any integer bit width and for different address spaces. Not all targets
6527 support all bit widths however.</p>
6529 <pre>
6531 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6532 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6533 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6535 </pre>
6537 <p>
6538 This loads a value in memory and compares it to a given value. If they are
6539 equal, it stores a new value into the memory.
6540 </p>
6542 <p>
6546 this integer type. While any bit width integer may be used, targets may only
6547 lower representations they support in hardware.
6549 </p>
6551 <p>
6552 This entire intrinsic must be executed atomically. It first loads the value
6555 loaded value is yielded in all cases. This provides the equivalent of an
6556 atomic compare-and-swap operation within the SSA framework.
6557 </p>
6560 <pre>
6561 %ptr = malloc i32
6562 store i32 4, %ptr
6565 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6571 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6576 </pre>
6577 </div>
6579 <!-- _______________________________________________________________________ -->
6582 </div>
6586 <p>
6588 integer bit width. Not all targets support all bit widths however.</p>
6589 <pre>
6595 </pre>
6597 <p>
6601 </p>
6604 <p>
6608 integer type. The targets may only lower integer representations they
6609 support.
6610 </p>
6612 <p>
6615 equivalent of an atomic swap operation within the SSA framework.
6617 </p>
6619 <pre>
6620 %ptr = malloc i32
6621 store i32 4, %ptr
6624 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6630 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6635 </pre>
6636 </div>
6638 <!-- _______________________________________________________________________ -->
6642 </div>
6645 <p>
6647 integer bit width. Not all targets support all bit widths however.</p>
6648 <pre>
6654 </pre>
6656 <p>
6659 </p>
6661 <p>
6663 The intrinsic takes two arguments, the first a pointer to an integer value
6664 and the second an integer value. The result is also an integer value. These
6665 integer types can have any bit width, but they must all have the same bit
6666 width. The targets may only lower integer representations they support.
6667 </p>
6669 <p>
6670 This intrinsic does a series of operations atomically. It first loads the
6673 </p>
6676 <pre>
6677 %ptr = malloc i32
6678 store i32 4, %ptr
6679 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6681 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6683 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6686 </pre>
6687 </div>
6689 <!-- _______________________________________________________________________ -->
6693 </div>
6696 <p>
6698 any integer bit width and for different address spaces. Not all targets
6699 support all bit widths however.</p>
6700 <pre>
6706 </pre>
6708 <p>
6711 </p>
6713 <p>
6715 The intrinsic takes two arguments, the first a pointer to an integer value
6716 and the second an integer value. The result is also an integer value. These
6717 integer types can have any bit width, but they must all have the same bit
6718 width. The targets may only lower integer representations they support.
6719 </p>
6721 <p>
6722 This intrinsic does a series of operations atomically. It first loads the
6725 </p>
6728 <pre>
6729 %ptr = malloc i32
6730 store i32 8, %ptr
6731 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6733 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6735 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6738 </pre>
6739 </div>
6741 <!-- _______________________________________________________________________ -->
6748 </div>
6751 <p>
6755 address spaces. Not all targets support all bit widths however.</p>
6756 <pre>
6762 </pre>
6764 <pre>
6770 </pre>
6772 <pre>
6778 </pre>
6780 <pre>
6786 </pre>
6788 <p>
6792 </p>
6794 <p>
6796 These intrinsics take two arguments, the first a pointer to an integer value
6797 and the second an integer value. The result is also an integer value. These
6798 integer types can have any bit width, but they must all have the same bit
6799 width. The targets may only lower integer representations they support.
6800 </p>
6802 <p>
6803 These intrinsics does a series of operations atomically. They first load the
6807 </p>
6810 <pre>
6811 %ptr = malloc i32
6812 store i32 0x0F0F, %ptr
6813 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6815 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6817 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6819 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6822 </pre>
6823 </div>
6826 <!-- _______________________________________________________________________ -->
6833 </div>
6836 <p>
6840 address spaces. Not all targets
6841 support all bit widths however.</p>
6842 <pre>
6848 </pre>
6850 <pre>
6856 </pre>
6858 <pre>
6864 </pre>
6866 <pre>
6872 </pre>
6874 <p>
6875 These intrinsics takes the signed or unsigned minimum or maximum of
6878 </p>
6880 <p>
6882 These intrinsics take two arguments, the first a pointer to an integer value
6883 and the second an integer value. The result is also an integer value. These
6884 integer types can have any bit width, but they must all have the same bit
6885 width. The targets may only lower integer representations they support.
6886 </p>
6888 <p>
6889 These intrinsics does a series of operations atomically. They first load the
6893 </p>
6896 <pre>
6897 %ptr = malloc i32
6898 store i32 7, %ptr
6899 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6901 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6903 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6905 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6908 </pre>
6909 </div>
6911 <!-- ======================================================================= -->
6914 </div>
6917 <p> This class of intrinsics is designed to be generic and has
6918 no specific purpose. </p>
6919 </div>
6921 <!-- _______________________________________________________________________ -->
6924 </div>
6929 <pre>
6930 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6931 </pre>
6935 <p>
6937 </p>
6941 <p>
6942 The first argument is a pointer to a value, the second is a pointer to a
6943 global string, the third is a pointer to a global string which is the source
6944 file name, and the last argument is the line number.
6945 </p>
6949 <p>
6950 This intrinsic allows annotation of local variables with arbitrary strings.
6951 This can be useful for special purpose optimizations that want to look for these
6952 annotations. These have no other defined use, they are ignored by code
6953 generation and optimization.
6954 </p>
6955 </div>
6957 <!-- _______________________________________________________________________ -->
6960 </div>
6966 any integer bit width.
6967 </p>
6968 <pre>
6969 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6970 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6971 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6972 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6973 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6974 </pre>
6978 <p>
6980 </p>
6984 <p>
6985 The first argument is an integer value (result of some expression),
6986 the second is a pointer to a global string, the third is a pointer to a global
6987 string which is the source file name, and the last argument is the line number.
6988 It returns the value of the first argument.
6989 </p>
6993 <p>
6994 This intrinsic allows annotations to be put on arbitrary expressions
6995 with arbitrary strings. This can be useful for special purpose optimizations
6996 that want to look for these annotations. These have no other defined use, they
6997 are ignored by code generation and optimization.
6998 </p>
6999 </div>
7001 <!-- _______________________________________________________________________ -->
7004 </div>
7009 <pre>
7010 declare void @llvm.trap()
7011 </pre>
7015 <p>
7017 </p>
7021 <p>
7022 None
7023 </p>
7027 <p>
7028 This intrinsics is lowered to the target dependent trap instruction. If the
7029 target does not have a trap instruction, this intrinsic will be lowered to the
7030 call of the abort() function.
7031 </p>
7032 </div>
7034 <!-- _______________________________________________________________________ -->
7037 </div>
7040 <pre>
7043 </pre>
7045 <p>
7048 it is placed on the stack before local variables.
7049 </p>
7051 <p>
7053 first argument is the value loaded from the stack guard
7055 has enough space to hold the value of the guard.
7056 </p>
7058 <p>
7059 This intrinsic causes the prologue/epilogue inserter to force the position of
7061 stack. This is to ensure that if a local variable on the stack is overwritten,
7062 it will destroy the value of the guard. When the function exits, the guard on
7063 the stack is checked against the original guard. If they're different, then
7065 </p>
7066 </div>
7068 <!-- *********************************************************************** -->
7069 <hr>
7070 <address>
7078 Last modified: $Date$
7079 </address>
7081 </body>
7082 </html>