| blob | 5a73f6afd8167b829132fea937ed9bddffbde270 |
3 <html>
4 <head>
11 </head>
13 <body>
16 <ol>
21 <ol>
24 <ol>
39 </ol>
40 </li>
54 </ol>
55 </li>
57 <ol>
60 <ol>
66 </ol>
67 </li>
69 <ol>
71 <ol>
77 </ol>
78 </li>
82 </ol>
83 </li>
85 </ol>
86 </li>
88 <ol>
96 </ol>
97 </li>
99 <ol>
102 </ol>
103 </li>
105 <ol>
108 Global Variable</a></li>
110 Global Variable</a></li>
112 Global Variable</a></li>
113 </ol>
114 </li>
116 <ol>
118 <ol>
126 </ol>
127 </li>
129 <ol>
142 </ol>
143 </li>
145 <ol>
152 </ol>
153 </li>
155 <ol>
159 </ol>
160 </li>
162 <ol>
165 </ol>
166 </li>
168 <ol>
173 </ol>
174 </li>
176 <ol>
189 </ol>
190 </li>
192 <ol>
199 </ol>
200 </li>
201 </ol>
202 </li>
204 <ol>
206 <ol>
210 </ol>
211 </li>
213 <ol>
217 </ol>
218 </li>
220 <ol>
228 </ol>
229 </li>
231 <ol>
240 </ol>
241 </li>
243 <ol>
248 </ol>
249 </li>
251 <ol>
258 </ol>
259 </li>
261 <ol>
264 </ol>
265 </li>
269 <ol>
271 </ol>
272 </li>
274 <ol>
288 </ol>
289 </li>
291 <ol>
296 </ol>
297 </li>
299 <ol>
310 </ol>
311 </li>
312 </ol>
313 </li>
314 </ol>
319 </div>
321 <!-- *********************************************************************** -->
323 <!-- *********************************************************************** -->
327 <p>This document is a reference manual for the LLVM assembly language. LLVM is
328 a Static Single Assignment (SSA) based representation that provides type
329 safety, low-level operations, flexibility, and the capability of representing
330 'all' high-level languages cleanly. It is the common code representation
331 used throughout all phases of the LLVM compilation strategy.</p>
333 </div>
335 <!-- *********************************************************************** -->
337 <!-- *********************************************************************** -->
341 <p>The LLVM code representation is designed to be used in three different forms:
342 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
343 for fast loading by a Just-In-Time compiler), and as a human readable
344 assembly language representation. This allows LLVM to provide a powerful
345 intermediate representation for efficient compiler transformations and
346 analysis, while providing a natural means to debug and visualize the
347 transformations. The three different forms of LLVM are all equivalent. This
348 document describes the human readable representation and notation.</p>
350 <p>The LLVM representation aims to be light-weight and low-level while being
351 expressive, typed, and extensible at the same time. It aims to be a
352 "universal IR" of sorts, by being at a low enough level that high-level ideas
353 may be cleanly mapped to it (similar to how microprocessors are "universal
354 IR's", allowing many source languages to be mapped to them). By providing
355 type information, LLVM can be used as the target of optimizations: for
356 example, through pointer analysis, it can be proven that a C automatic
357 variable is never accessed outside of the current function, allowing it to
358 be promoted to a simple SSA value instead of a memory location.</p>
360 </div>
362 <!-- _______________________________________________________________________ -->
367 <p>It is important to note that this document describes 'well formed' LLVM
368 assembly language. There is a difference between what the parser accepts and
369 what is considered 'well formed'. For example, the following instruction is
370 syntactically okay, but not well formed:</p>
373 <pre>
375 </pre>
376 </div>
379 LLVM infrastructure provides a verification pass that may be used to verify
380 that an LLVM module is well formed. This pass is automatically run by the
381 parser after parsing input assembly and by the optimizer before it outputs
382 bitcode. The violations pointed out by the verifier pass indicate bugs in
383 transformation passes or input to the parser.</p>
385 </div>
387 <!-- Describe the typesetting conventions here. -->
389 <!-- *********************************************************************** -->
391 <!-- *********************************************************************** -->
395 <p>LLVM identifiers come in two basic types: global and local. Global
397 character. Local identifiers (register names, types) begin with
399 for identifiers, for different purposes:</p>
401 <ol>
402 <li>Named values are represented as a string of characters with their prefix.
406 other characters in their names can be surrounded with quotes. Special
408 ASCII code for the character in hexadecimal. In this way, any character
409 can be used in a name value, even quotes themselves.</li>
411 <li>Unnamed values are represented as an unsigned numeric value with their
416 </ol>
418 <p>LLVM requires that values start with a prefix for two reasons: Compilers
419 don't need to worry about name clashes with reserved words, and the set of
420 reserved words may be expanded in the future without penalty. Additionally,
421 unnamed identifiers allow a compiler to quickly come up with a temporary
422 variable without having to avoid symbol table conflicts.</p>
424 <p>Reserved words in LLVM are very similar to reserved words in other
425 languages. There are keywords for different opcodes
431 reserved words cannot conflict with variable names, because none of them
434 <p>Here is an example of LLVM code to multiply the integer variable
440 <pre>
442 </pre>
443 </div>
448 <pre>
450 </pre>
451 </div>
456 <pre>
460 </pre>
461 </div>
464 lexical features of LLVM:</p>
466 <ol>
468 line.</li>
470 <li>Unnamed temporaries are created when the result of a computation is not
471 assigned to a named value.</li>
474 </ol>
476 <p>It also shows a convention that we follow in this document. When
477 demonstrating instructions, we will follow an instruction with a comment that
478 defines the type and name of value produced. Comments are shown in italic
479 text.</p>
481 </div>
483 <!-- *********************************************************************** -->
485 <!-- *********************************************************************** -->
487 <!-- ======================================================================= -->
489 </div>
494 of the input programs. Each module consists of functions, global variables,
495 and symbol table entries. Modules may be combined together with the LLVM
496 linker, which merges function (and global variable) definitions, resolves
497 forward declarations, and merges symbol table entries. Here is an example of
501 <pre>
503 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
511 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
519 !foo = !{!1, null}
520 </pre>
521 </div>
529 <p>In general, a module is made up of a list of global values, where both
530 functions and global variables are global values. Global values are
531 represented by a pointer to a memory location (in this case, a pointer to an
532 array of char, and a pointer to a function), and have one of the
535 </div>
537 <!-- ======================================================================= -->
540 </div>
544 <p>All Global Variables and Functions have one of the following types of
545 linkage:</p>
547 <dl>
549 <dd>Global values with private linkage are only directly accessible by objects
550 in the current module. In particular, linking code into a module with an
551 private global value may cause the private to be renamed as necessary to
552 avoid collisions. Because the symbol is private to the module, all
553 references can be updated. This doesn't show up in any symbol table in the
554 object file.</dd>
557 <dd>Similar to private, but the symbol is passed through the assembler and
558 removed by the linker after evaluation. Note that (unlike private
559 symbols) linker_private symbols are subject to coalescing by the linker:
560 weak symbols get merged and redefinitions are rejected. However, unlike
561 normal strong symbols, they are removed by the linker from the final
562 linked image (executable or dynamic library).</dd>
565 <dd>Similar to private, but the value shows as a local symbol
571 into the object file corresponding to the LLVM module. They exist to
572 allow inlining and other optimizations to take place given knowledge of
573 the definition of the global, which is known to be somewhere outside the
576 This linkage type is only allowed on definitions, not declarations.</dd>
580 the same name when linkage occurs. This can be used to implement
581 some forms of inline functions, templates, or other code which must be
582 generated in each translation unit that uses it, but where the body may
583 be overridden with a more definitive definition later. Unreferenced
586 inline the body of this function into callers because it doesn't know if
587 this definition of the function is the definitive definition within the
588 program or whether it will be overridden by a stronger definition.
589 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
590 linkage.</dd>
600 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
601 global scope.
602 Symbols with "<tt>common</tt>" linkage are merged in the same way as
605 must have a zero initializer, and may not be marked '<a
607 have common linkage.</dd>
612 pointer to array type. When two global variables with appending linkage
613 are linked together, the two global arrays are appended together. This is
614 the LLVM, typesafe, equivalent of having the system linker append together
618 <dd>The semantics of this linkage follow the ELF object file model: the symbol
619 is weak until linked, if not linked, the symbol becomes null instead of
620 being an undefined reference.</dd>
624 <dd>Some languages allow differing globals to be merged, such as two functions
626 that only equivalent globals are ever merged (the "one definition rule" -
629 be merged with equivalent globals. These linkage types are otherwise the
633 <dd>If none of the above identifiers are used, the global is externally
634 visible, meaning that it participates in linkage and can be used to
635 resolve external symbol references.</dd>
636 </dl>
638 <p>The next two types of linkage are targeted for Microsoft Windows platform
639 only. They are designed to support importing (exporting) symbols from (to)
640 DLLs (Dynamic Link Libraries).</p>
642 <dl>
645 or variable via a global pointer to a pointer that is set up by the DLL
646 exporting the symbol. On Microsoft Windows targets, the pointer name is
648 name.</dd>
652 pointer to a pointer in a DLL, so that it can be referenced with the
655 variable name.</dd>
656 </dl>
659 another module defined a "<tt>.LC0</tt>" variable and was linked with this
660 one, one of the two would be renamed, preventing a collision. Since
662 declarations), they are accessible outside of the current module.</p>
671 </div>
673 <!-- ======================================================================= -->
676 </div>
682 convention specified for the call. The calling convention of any pair of
683 dynamic caller/callee must match, or the behavior of the program is
684 undefined. The following calling conventions are supported by LLVM, and more
685 may be added in the future:</p>
687 <dl>
689 <dd>This calling convention (the default if no other calling convention is
690 specified) matches the target C calling conventions. This calling
691 convention supports varargs function calls and tolerates some mismatch in
692 the declared prototype and implemented declaration of the function (as
693 does normal C).</dd>
696 <dd>This calling convention attempts to make calls as fast as possible
697 (e.g. by passing things in registers). This calling convention allows the
698 target to use whatever tricks it wants to produce fast code for the
699 target, without having to conform to an externally specified ABI
700 (Application Binary Interface).
702 when this or the GHC convention is used.</a> This calling convention
703 does not support varargs and requires the prototype of all callees to
704 exactly match the prototype of the function definition.</dd>
707 <dd>This calling convention attempts to make code in the caller as efficient
708 as possible under the assumption that the call is not commonly executed.
709 As such, these calls often preserve all registers so that the call does
710 not break any live ranges in the caller side. This calling convention
711 does not support varargs and requires the prototype of all callees to
712 exactly match the prototype of the function definition.</dd>
715 <dd>This calling convention has been implemented specifically for use by the
717 It passes everything in registers, going to extremes to achieve this by
718 disabling callee save registers. This calling convention should not be
719 used lightly but only for specific situations such as an alternative to
721 implementing functional programming languages.At the moment only X86
722 supports this convention and it has the following limitations:
723 <ul>
725 floating point types are supported.</li>
728 </ul>
729 This calling convention supports
731 requires both the caller and callee are using it.
732 </dd>
735 <dd>Any calling convention may be specified by number, allowing
736 target-specific calling conventions to be used. Target specific calling
738 </dl>
740 <p>More calling conventions can be added/defined on an as-needed basis, to
741 support Pascal conventions or any other well-known target-independent
742 convention.</p>
744 </div>
746 <!-- ======================================================================= -->
749 </div>
753 <p>All Global Variables and Functions have one of the following visibility
754 styles:</p>
756 <dl>
758 <dd>On targets that use the ELF object file format, default visibility means
759 that the declaration is visible to other modules and, in shared libraries,
760 means that the declared entity may be overridden. On Darwin, default
761 visibility means that the declaration is visible to other modules. Default
765 <dd>Two declarations of an object with hidden visibility refer to the same
766 object if they are in the same shared object. Usually, hidden visibility
767 indicates that the symbol will not be placed into the dynamic symbol
768 table, so no other module (executable or shared library) can reference it
769 directly.</dd>
772 <dd>On ELF, protected visibility indicates that the symbol will be placed in
773 the dynamic symbol table, but that references within the defining module
774 will bind to the local symbol. That is, the symbol cannot be overridden by
775 another module.</dd>
776 </dl>
778 </div>
780 <!-- ======================================================================= -->
783 </div>
787 <p>LLVM IR allows you to specify name aliases for certain types. This can make
788 it easier to read the IR and make the IR more condensed (particularly when
789 recursive types are involved). An example of a name specification is:</p>
792 <pre>
793 %mytype = type { %mytype*, i32 }
794 </pre>
795 </div>
801 <p>Note that type names are aliases for the structural type that they indicate,
802 and that you can therefore specify multiple names for the same type. This
803 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
804 uses structural typing, the name is not part of the type. When printing out
806 particular shape. This means that if you have code where two different
807 source types end up having the same LLVM type, that the dumper will sometimes
808 print the "wrong" or unexpected type. This is an important design point and
809 isn't going to change.</p>
811 </div>
813 <!-- ======================================================================= -->
816 </div>
820 <p>Global variables define regions of memory allocated at compilation time
821 instead of run-time. Global variables may optionally be initialized, may
822 have an explicit section to be placed in, and may have an optional explicit
823 alignment specified. A variable may be defined as "thread_local", which
824 means that it will not be shared by threads (each thread will have a
825 separated copy of the variable). A variable may be defined as a global
826 "constant," which indicates that the contents of the variable
828 global data to be placed in the read-only section of an executable, etc).
829 Note that variables that need runtime initialization cannot be marked
833 constant, even if the final definition of the global is not. This capability
834 can be used to enable slightly better optimization of the program, but
835 requires the language definition to guarantee that optimizations based on the
836 'constantness' are valid for the translation units that do not include the
837 definition.</p>
839 <p>As SSA values, global variables define pointer values that are in scope
840 (i.e. they dominate) all basic blocks in the program. Global variables
841 always define a pointer to their "content" type because they describe a
842 region of memory, and all memory objects in LLVM are accessed through
843 pointers.</p>
845 <p>A global variable may be declared to reside in a target-specific numbered
846 address space. For targets that support them, address spaces may affect how
847 optimizations are performed and/or what target instructions are used to
848 access the variable. The default address space is zero. The address space
849 qualifier must precede any other attributes.</p>
851 <p>LLVM allows an explicit section to be specified for globals. If the target
852 supports it, it will emit globals to the section specified.</p>
854 <p>An explicit alignment may be specified for a global, which must be a power
855 of 2. If not present, or if the alignment is set to zero, the alignment of
856 the global is set by the target to whatever it feels convenient. If an
857 explicit alignment is specified, the global is forced to have exactly that
858 alignment. Targets and optimizers are not allowed to over-align the global
859 if the global has an assigned section. In this case, the extra alignment
860 could be observable: for example, code could assume that the globals are
861 densely packed in their section and try to iterate over them as an array,
862 alignment padding would break this iteration.</p>
864 <p>For example, the following defines a global in a numbered address space with
865 an initializer, section, and alignment:</p>
868 <pre>
870 </pre>
871 </div>
873 </div>
876 <!-- ======================================================================= -->
879 </div>
888 name, a (possibly empty) argument list (each with optional
892 curly brace, a list of basic blocks, and a closing curly brace.</p>
899 name, a possibly empty list of arguments, an optional alignment, and an
902 <p>A function definition contains a list of basic blocks, forming the CFG
903 (Control Flow Graph) for the function. Each basic block may optionally start
904 with a label (giving the basic block a symbol table entry), contains a list
906 instruction (such as a branch or function return).</p>
908 <p>The first basic block in a function is special in two ways: it is immediately
909 executed on entrance to the function, and it is not allowed to have
910 predecessor basic blocks (i.e. there can not be any branches to the entry
911 block of a function). Because the block can have no predecessors, it also
914 <p>LLVM allows an explicit section to be specified for functions. If the target
915 supports it, it will emit functions to the section specified.</p>
917 <p>An explicit alignment may be specified for a function. If not present, or if
918 the alignment is set to zero, the alignment of the function is set by the
919 target to whatever it feels convenient. If an explicit alignment is
920 specified, the function is forced to have at least that much alignment. All
925 <pre>
931 </pre>
932 </div>
934 </div>
936 <!-- ======================================================================= -->
939 </div>
944 function, global variable, another alias or bitcast of global value). Aliases
950 <pre>
952 </pre>
953 </div>
955 </div>
957 <!-- ======================================================================= -->
960 </div>
965 nodes</a> (but not metadata strings) and null are the only valid operands for
966 a named metadata.</p>
970 <pre>
972 !name = !{null, !1}
973 </pre>
974 </div>
976 </div>
978 <!-- ======================================================================= -->
983 <p>The return type and each parameter of a function type may have a set of
985 used to communicate additional information about the result or parameters of
986 a function. Parameter attributes are considered to be part of the function,
987 not of the function type, so functions with different parameter attributes
988 can have the same function type.</p>
990 <p>Parameter attributes are simple keywords that follow the type specified. If
991 multiple parameter attributes are needed, they are space separated. For
992 example:</p>
995 <pre>
996 declare i32 @printf(i8* noalias nocapture, ...)
997 declare i32 @atoi(i8 zeroext)
998 declare signext i8 @returns_signed_char()
999 </pre>
1000 </div>
1007 <dl>
1009 <dd>This indicates to the code generator that the parameter or return value
1010 should be zero-extended to a 32-bit value by the caller (for a parameter)
1011 or the callee (for a return value).</dd>
1014 <dd>This indicates to the code generator that the parameter or return value
1015 should be sign-extended to a 32-bit value by the caller (for a parameter)
1016 or the callee (for a return value).</dd>
1019 <dd>This indicates that this parameter or return value should be treated in a
1020 special target-dependent fashion during while emitting code for a function
1021 call or return (usually, by putting it in a register as opposed to memory,
1022 though some targets use it to distinguish between two different kinds of
1023 registers). Use of this attribute is target-specific.</dd>
1026 <dd>This indicates that the pointer parameter should really be passed by value
1027 to the function. The attribute implies that a hidden copy of the pointee
1028 is made between the caller and the callee, so the callee is unable to
1029 modify the value in the callee. This attribute is only valid on LLVM
1030 pointer arguments. It is generally used to pass structs and arrays by
1031 value, but is also valid on pointers to scalars. The copy is considered
1032 to belong to the caller not the callee (for example,
1035 values. The byval attribute also supports specifying an alignment with
1036 the align attribute. This has a target-specific effect on the code
1037 generator that usually indicates a desired alignment for the synthesized
1038 stack slot.</dd>
1041 <dd>This indicates that the pointer parameter specifies the address of a
1042 structure that is the return value of the function in the source program.
1043 This pointer must be guaranteed by the caller to be valid: loads and
1044 stores to the structure may be assumed by the callee to not to trap. This
1045 may only be applied to the first parameter. This is not a valid attribute
1046 for return values. </dd>
1049 <dd>This indicates that the pointer does not alias any global or any other
1050 parameter. The caller is responsible for ensuring that this is the
1052 that the pointer does not alias any other pointers visible to the
1053 caller. For further details, please see the discussion of the NoAlias
1054 response in
1059 <dd>This indicates that the callee does not make any copies of the pointer
1060 that outlive the callee itself. This is not a valid attribute for return
1061 values.</dd>
1064 <dd>This indicates that the pointer parameter can be excised using the
1066 attribute for return values.</dd>
1067 </dl>
1069 </div>
1071 <!-- ======================================================================= -->
1074 </div>
1078 <p>Each function may specify a garbage collector name, which is simply a
1079 string:</p>
1082 <pre>
1083 define void @f() gc "name" { ... }
1084 </pre>
1085 </div>
1088 collector which will cause the compiler to alter its output in order to
1089 support the named garbage collection algorithm.</p>
1091 </div>
1093 <!-- ======================================================================= -->
1096 </div>
1100 <p>Function attributes are set to communicate additional information about a
1101 function. Function attributes are considered to be part of the function, not
1102 of the function type, so functions with different parameter attributes can
1103 have the same function type.</p>
1105 <p>Function attributes are simple keywords that follow the type specified. If
1106 multiple attributes are needed, they are space separated. For example:</p>
1109 <pre>
1110 define void @f() noinline { ... }
1111 define void @f() alwaysinline { ... }
1112 define void @f() alwaysinline optsize { ... }
1113 define void @f() optsize { ... }
1114 </pre>
1115 </div>
1117 <dl>
1119 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1120 the backend should forcibly align the stack pointer. Specify the
1121 desired alignment, which must be a power of two, in parentheses.
1124 <dd>This attribute indicates that the inliner should attempt to inline this
1125 function into callers whenever possible, ignoring any active inlining size
1126 threshold for this caller.</dd>
1129 <dd>This attribute indicates that the source code contained a hint that inlining
1130 this function is desirable (such as the "inline" keyword in C/C++). It
1131 is just a hint; it imposes no requirements on the inliner.</dd>
1134 <dd>This attribute indicates that the inliner should never inline this
1135 function in any situation. This attribute may not be used together with
1139 <dd>This attribute suggests that optimization passes and code generator passes
1140 make choices that keep the code size of this function low, and otherwise
1141 do optimizations specifically to reduce code size.</dd>
1144 <dd>This function attribute indicates that the function never returns
1145 normally. This produces undefined behavior at runtime if the function
1146 ever does dynamically return.</dd>
1149 <dd>This function attribute indicates that the function never returns with an
1150 unwind or exceptional control flow. If the function does unwind, its
1151 runtime behavior is undefined.</dd>
1154 <dd>This attribute indicates that the function computes its result (or decides
1155 to unwind an exception) based strictly on its arguments, without
1156 dereferencing any pointer arguments or otherwise accessing any mutable
1157 state (e.g. memory, control registers, etc) visible to caller functions.
1158 It does not write through any pointer arguments
1160 changes any state visible to callers. This means that it cannot unwind
1165 <dd>This attribute indicates that the function does not write through any
1167 arguments) or otherwise modify any state (e.g. memory, control registers,
1168 etc) visible to caller functions. It may dereference pointer arguments
1169 and read state that may be set in the caller. A readonly function always
1170 returns the same value (or unwinds an exception identically) when called
1171 with the same set of arguments and global state. It cannot unwind an
1176 <dd>This attribute indicates that the function should emit a stack smashing
1178 the stack before the local variables that's checked upon return from the
1179 function to see if it has been overwritten. A heuristic is used to
1180 determine if a function needs stack protectors or not.<br>
1181 <br>
1188 stack smashing protector. This overrides
1190 <br>
1197 <dd>This attribute indicates that the code generator should not use a red
1198 zone, even if the target-specific ABI normally permits it.</dd>
1204 <dd>This attribute disables prologue / epilogue emission for the function.
1205 This can have very system-specific consequences.</dd>
1206 </dl>
1208 </div>
1210 <!-- ======================================================================= -->
1213 </div>
1218 the GCC "file scope inline asm" blocks. These blocks are internally
1219 concatenated by LLVM and treated as a single unit, but may be separated in
1223 <pre>
1224 module asm "inline asm code goes here"
1225 module asm "more can go here"
1226 </pre>
1227 </div>
1229 <p>The strings can contain any character by escaping non-printable characters.
1231 for the number.</p>
1233 <p>The inline asm code is simply printed to the machine code .s file when
1234 assembly code is generated.</p>
1236 </div>
1238 <!-- ======================================================================= -->
1241 </div>
1245 <p>A module may specify a target specific data layout string that specifies how
1246 data is to be laid out in memory. The syntax for the data layout is
1247 simply:</p>
1250 <pre>
1251 target datalayout = "<i>layout specification</i>"
1252 </pre>
1253 </div>
1256 separated by the minus sign character ('-'). Each specification starts with
1257 a letter and may include other information after the letter to define some
1258 aspect of the data layout. The specifications accepted are as follows:</p>
1260 <dl>
1262 <dd>Specifies that the target lays out data in big-endian form. That is, the
1263 bits with the most significance have the lowest address location.</dd>
1266 <dd>Specifies that the target lays out data in little-endian form. That is,
1267 the bits with the least significance have the lowest address
1268 location.</dd>
1277 <dd>This specifies the alignment for an integer type of a given bit
1281 <dd>This specifies the alignment for a vector type of a given bit
1285 <dd>This specifies the alignment for a floating point type of a given bit
1287 (double).</dd>
1290 <dd>This specifies the alignment for an aggregate type of a given bit
1294 <dd>This specifies the alignment for a stack object of a given bit
1298 <dd>This specifies a set of native integer widths for the target CPU
1301 this set are considered to support most general arithmetic
1302 operations efficiently.</dd>
1303 </dl>
1305 <p>When constructing the data layout for a given target, LLVM starts with a
1306 default set of specifications which are then (possibly) overridden by the
1308 are given in this list:</p>
1310 <ul>
1325 </ul>
1327 <p>When LLVM is determining the alignment for a given type, it uses the
1328 following rules:</p>
1330 <ol>
1331 <li>If the type sought is an exact match for one of the specifications, that
1332 specification is used.</li>
1334 <li>If no match is found, and the type sought is an integer type, then the
1335 smallest integer type that is larger than the bitwidth of the sought type
1336 is used. If none of the specifications are larger than the bitwidth then
1337 the the largest integer type is used. For example, given the default
1338 specifications above, the i7 type will use the alignment of i8 (next
1339 largest) while both i65 and i256 will use the alignment of i64 (largest
1340 specified).</li>
1342 <li>If no match is found, and the type sought is a vector type, then the
1343 largest vector type that is smaller than the sought vector type will be
1346 </ol>
1348 </div>
1350 <!-- ======================================================================= -->
1353 </div>
1357 <p>Any memory access must be done through a pointer value associated
1358 with an address range of the memory access, otherwise the behavior
1359 is undefined. Pointer values are associated with address ranges
1360 according to the following rules:</p>
1362 <ul>
1363 <li>A pointer value formed from a
1365 is associated with the addresses associated with the first operand
1367 <li>An address of a global variable is associated with the address
1368 range of the variable's storage.</li>
1369 <li>The result value of an allocation instruction is associated with
1370 the address range of the allocated storage.</li>
1371 <li>A null pointer in the default address-space is associated with
1372 no address.</li>
1373 <li>A pointer value formed by an
1375 address ranges of all pointer values that contribute (directly or
1376 indirectly) to the computation of the pointer's value.</li>
1377 <li>The result value of a
1380 <li>An integer constant other than zero or a pointer value returned
1381 from a function not defined within LLVM may be associated with address
1382 ranges allocated through mechanisms other than those provided by
1383 LLVM. Such ranges shall not overlap with any ranges of addresses
1384 allocated by mechanisms provided by LLVM.</li>
1385 </ul>
1387 <p>LLVM IR does not associate types with memory. The result type of a
1389 alignment of the memory from which to load, as well as the
1390 interpretation of the value. The first operand of a
1392 and alignment of the store.</p>
1394 <p>Consequently, type-based alias analysis, aka TBAA, aka
1397 additional information which specialized optimization passes may use
1398 to implement type-based alias analysis.</p>
1400 </div>
1402 <!-- ======================================================================= -->
1405 </div>
1412 The optimizers must not change the number of volatile operations or change their
1413 order of execution relative to other volatile operations. The optimizers
1415 operations. This is not Java's "volatile" and has no cross-thread
1416 synchronization behavior.</p>
1418 </div>
1420 <!-- *********************************************************************** -->
1422 <!-- *********************************************************************** -->
1426 <p>The LLVM type system is one of the most important features of the
1427 intermediate representation. Being typed enables a number of optimizations
1428 to be performed on the intermediate representation directly, without having
1429 to do extra analyses on the side before the transformation. A strong type
1430 system makes it easier to read the generated code and enables novel analyses
1431 and transformations that are not feasible to perform on normal three address
1432 code representations.</p>
1434 </div>
1436 <!-- ======================================================================= -->
1438 Classifications</a> </div>
1445 <tbody>
1447 <tr>
1450 </tr>
1451 <tr>
1454 </tr>
1455 <tr>
1466 </td>
1467 </tr>
1468 <tr>
1474 </tr>
1475 <tr>
1485 </td>
1486 </tr>
1487 </tbody>
1488 </table>
1491 important. Values of these types are the only ones which can be produced by
1492 instructions.</p>
1494 </div>
1496 <!-- ======================================================================= -->
1501 <p>The primitive types are the fundamental building blocks of the LLVM
1502 system.</p>
1504 </div>
1506 <!-- _______________________________________________________________________ -->
1512 <p>The integer type is a very simple type that simply specifies an arbitrary
1513 bit width for the integer type desired. Any bit width from 1 bit to
1517 <pre>
1518 iN
1519 </pre>
1522 value.</p>
1529 </tr>
1533 </tr>
1537 </tr>
1538 </table>
1540 </div>
1542 <!-- _______________________________________________________________________ -->
1547 <table>
1548 <tbody>
1555 </tbody>
1556 </table>
1558 </div>
1560 <!-- _______________________________________________________________________ -->
1569 <pre>
1570 void
1571 </pre>
1573 </div>
1575 <!-- _______________________________________________________________________ -->
1584 <pre>
1585 label
1586 </pre>
1588 </div>
1590 <!-- _______________________________________________________________________ -->
1596 <p>The metadata type represents embedded metadata. No derived types may be
1598 arguments.
1601 <pre>
1602 metadata
1603 </pre>
1605 </div>
1608 <!-- ======================================================================= -->
1613 <p>The real power in LLVM comes from the derived types in the system. This is
1614 what allows a programmer to represent arrays, functions, pointers, and other
1615 useful types. Each of these types contain one or more element types which
1616 may be a primitive type, or another derived type. For example, it is
1617 possible to have a two dimensional array, using an array as the element type
1618 of another array.</p>
1621 </div>
1623 <!-- _______________________________________________________________________ -->
1628 <p>Aggregate Types are a subset of derived types that can contain multiple
1633 </div>
1635 </div>
1637 <!-- _______________________________________________________________________ -->
1643 <p>The array type is a very simple derived type that arranges elements
1644 sequentially in memory. The array type requires a size (number of elements)
1645 and an underlying data type.</p>
1648 <pre>
1650 </pre>
1653 be any type with a size.</p>
1660 </tr>
1664 </tr>
1668 </tr>
1669 </table>
1675 </tr>
1679 </tr>
1683 </tr>
1684 </table>
1686 <p>There is no restriction on indexing beyond the end of the array implied by
1687 a static type (though there are restrictions on indexing beyond the bounds
1688 of an allocated object in some cases). This means that single-dimension
1689 'variable sized array' addressing can be implemented in LLVM with a zero
1690 length array type. An implementation of 'pascal style arrays' in LLVM could
1693 </div>
1695 <!-- _______________________________________________________________________ -->
1701 <p>The function type can be thought of as a function signature. It consists of
1702 a return type and a list of formal parameter types. The return type of a
1703 function type is a scalar type, a void type, a struct type, or a union
1704 type. If the return type is a struct type then all struct elements must be
1705 of first class types, and the struct must have at least one element.</p>
1708 <pre>
1710 </pre>
1714 which indicates that the function takes a variable number of arguments.
1715 Variable argument functions can access their arguments with
1725 </td>
1728 </tt></td>
1732 </td>
1738 LLVM.
1739 </td>
1744 </td>
1745 </tr>
1746 </table>
1748 </div>
1750 <!-- _______________________________________________________________________ -->
1756 <p>The structure type is used to represent a collection of data members together
1757 in memory. The packing of the field types is defined to match the ABI of the
1758 underlying processor. The elements of a structure may be any type that has a
1759 size.</p>
1764 Structures in registers are accessed using the
1768 <pre>
1770 </pre>
1783 </tr>
1784 </table>
1786 </div>
1788 <!-- _______________________________________________________________________ -->
1790 </div>
1795 <p>The packed structure type is used to represent a collection of data members
1796 together in memory. There is no padding between fields. Further, the
1797 alignment of a packed structure is 1 byte. The elements of a packed
1798 structure may be any type that has a size.</p>
1805 <pre>
1807 </pre>
1821 </tr>
1822 </table>
1824 </div>
1826 <!-- _______________________________________________________________________ -->
1832 <p>A union type describes an object with size and alignment suitable for
1833 an object of any one of a given set of types (also known as an "untagged"
1834 union). It is similar in concept and usage to a
1836 have an offset of zero. The elements of a union may be any type that has a
1837 size. Unions must have at least one member - empty unions are not allowed.
1838 </p>
1840 <p>The size of the union as a whole will be the size of its largest member,
1841 and the alignment requirements of the union as a whole will be the largest
1842 alignment requirement of any member.</p>
1847 Since all members are at offset zero, the getelementptr instruction does
1848 not affect the address, only the type of the resulting pointer.</p>
1851 <pre>
1853 </pre>
1868 </tr>
1869 </table>
1871 </div>
1873 <!-- _______________________________________________________________________ -->
1879 <p>The pointer type is used to specify memory locations.
1880 Pointers are commonly used to reference objects in memory.</p>
1882 <p>Pointer types may have an optional address space attribute defining the
1883 numbered address space where the pointed-to object resides. The default
1884 address space is number zero. The semantics of non-zero address
1885 spaces are target-specific.</p>
1891 <pre>
1893 </pre>
1901 </tr>
1907 </tr>
1912 </tr>
1913 </table>
1915 </div>
1917 <!-- _______________________________________________________________________ -->
1923 <p>A vector type is a simple derived type that represents a vector of elements.
1924 Vector types are used when multiple primitive data are operated in parallel
1925 using a single instruction (SIMD). A vector type requires a size (number of
1926 elements) and an underlying primitive data type. Vector types are considered
1930 <pre>
1932 </pre>
1934 <p>The number of elements is a constant integer value; elementtype may be any
1935 integer or floating point type.</p>
1942 </tr>
1946 </tr>
1950 </tr>
1951 </table>
1953 </div>
1955 <!-- _______________________________________________________________________ -->
1960 <p>Opaque types are used to represent unknown types in the system. This
1961 corresponds (for example) to the C notion of a forward declared structure
1962 type. In LLVM, opaque types can eventually be resolved to any type (not just
1963 a structure type).</p>
1966 <pre>
1967 opaque
1968 </pre>
1975 </tr>
1976 </table>
1978 </div>
1980 <!-- ======================================================================= -->
1983 </div>
1989 requiring it to have a name. For instance, a structure declaration may
1990 contain a pointer to any of the types it is lexically a member of. Example
1991 of up references (with their equivalent as named type declarations)
1992 include:</p>
1994 <pre>
1995 { \2 * } %x = type { %x* }
1996 { \2 }* %y = type { %y }*
1997 \1* %z = type %z*
1998 </pre>
2000 <p>An up reference is needed by the asmprinter for printing out cyclic types
2001 when there is no declared name for a type in the cycle. Because the
2002 asmprinter does not want to print out an infinite type string, it needs a
2003 syntax to handle recursive types that have no names (all names are optional
2004 in llvm IR).</p>
2007 <pre>
2009 </pre>
2018 </tr>
2022 structure.</td>
2023 </tr>
2024 </table>
2026 </div>
2028 <!-- *********************************************************************** -->
2030 <!-- *********************************************************************** -->
2034 <p>LLVM has several different basic types of constants. This section describes
2035 them all and their syntax.</p>
2037 </div>
2039 <!-- ======================================================================= -->
2044 <dl>
2052 with integer types.</dd>
2056 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2057 notation (see below). The assembler requires the exact decimal value of a
2058 floating-point constant. For example, the assembler accepts 1.25 but
2065 </dl>
2067 <p>The one non-intuitive notation for constants is the hexadecimal form of
2068 floating point constants. For example, the form '<tt>double
2071 constants are required (and the only time that they are generated by the
2072 disassembler) is when a floating point constant must be emitted but it cannot
2073 be represented as a decimal floating point number in a reasonable number of
2074 digits. For example, NaN's, infinities, and other special values are
2075 represented in their IEEE hexadecimal format so that assembly and disassembly
2076 do not cause any bits to change in the constants.</p>
2078 <p>When using the hexadecimal form, constants of types float and double are
2079 represented using the 16-digit form shown above (which matches the IEEE754
2080 representation for double); float values must, however, be exactly
2081 representable as IEE754 single precision. Hexadecimal format is always used
2082 for long double, and there are three forms of long double. The 80-bit format
2084 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2087 currently supported target uses this format. Long doubles will only work if
2088 they match the long double format on your target. All hexadecimal formats
2089 are big-endian (sign bit at the left).</p>
2091 </div>
2093 <!-- ======================================================================= -->
2097 </div>
2101 <p>Complex constants are a (potentially recursive) combination of simple
2102 constants and smaller complex constants.</p>
2104 <dl>
2106 <dd>Structure constants are represented with notation similar to structure
2107 type definitions (a comma separated list of elements, surrounded by braces
2111 the number and types of elements must match those specified by the
2112 type.</dd>
2115 <dd>Union constants are represented with notation similar to a structure with
2116 a single element - that is, a single typed element surrounded
2119 struct as long as the type of the struct element matches the type of
2120 one of the union members.</dd>
2123 <dd>Array constants are represented with notation similar to array type
2124 definitions (a comma separated list of elements, surrounded by square
2127 the number and types of elements must match those specified by the
2128 type.</dd>
2131 <dd>Vector constants are represented with notation similar to vector type
2132 definitions (a comma separated list of elements, surrounded by
2134 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2136 elements must match those specified by the type.</dd>
2142 This is often used to avoid having to print large zero initializers
2143 (e.g. for large arrays) and is always exactly equivalent to using explicit
2144 zero initializers.</dd>
2147 <dd>A metadata node is a structure-like constant with
2150 be interpreted as part of the instruction stream, metadata is a place to
2151 attach additional information such as debug info.</dd>
2152 </dl>
2154 </div>
2156 <!-- ======================================================================= -->
2159 </div>
2165 (link-time) constants. These constants are explicitly referenced when
2168 legal LLVM file:</p>
2171 <pre>
2172 @X = global i32 17
2173 @Y = global i32 42
2174 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2175 </pre>
2176 </div>
2178 </div>
2180 <!-- ======================================================================= -->
2185 indicates that the user of the value may receive an unspecified bit-pattern.
2186 Undefined values may be of any type (other than label or void) and be used
2187 anywhere a constant is permitted.</p>
2189 <p>Undefined values are useful because they indicate to the compiler that the
2190 program is well defined no matter what value is used. This gives the
2191 compiler more freedom to optimize. Here are some examples of (potentially
2192 surprising) transformations that are valid (in pseudo IR):</p>
2196 <pre>
2197 %A = add %X, undef
2198 %B = sub %X, undef
2199 %C = xor %X, undef
2200 Safe:
2201 %A = undef
2202 %B = undef
2203 %C = undef
2204 </pre>
2205 </div>
2207 <p>This is safe because all of the output bits are affected by the undef bits.
2208 Any output bit can have a zero or one depending on the input bits.</p>
2211 <pre>
2212 %A = or %X, undef
2213 %B = and %X, undef
2214 Safe:
2215 %A = -1
2216 %B = 0
2217 Unsafe:
2218 %A = undef
2219 %B = undef
2220 </pre>
2221 </div>
2223 <p>These logical operations have bits that are not always affected by the input.
2224 For example, if "%X" has a zero bit, then the output of the 'and' operation will
2225 always be a zero, no matter what the corresponding bit from the undef is. As
2226 such, it is unsafe to optimize or assume that the result of the and is undef.
2227 However, it is safe to assume that all bits of the undef could be 0, and
2228 optimize the and to 0. Likewise, it is safe to assume that all the bits of
2229 the undef operand to the or could be set, allowing the or to be folded to
2233 <pre>
2234 %A = select undef, %X, %Y
2235 %B = select undef, 42, %Y
2236 %C = select %X, %Y, undef
2237 Safe:
2238 %A = %X (or %Y)
2239 %B = 42 (or %Y)
2240 %C = %Y
2241 Unsafe:
2242 %A = undef
2243 %B = undef
2244 %C = undef
2245 </pre>
2246 </div>
2248 <p>This set of examples show that undefined select (and conditional branch)
2249 conditions can go "either way" but they have to come from one of the two
2250 operands. In the %A example, if %X and %Y were both known to have a clear low
2251 bit, then %A would have to have a cleared low bit. However, in the %C example,
2252 the optimizer is allowed to assume that the undef operand could be the same as
2253 %Y, allowing the whole select to be eliminated.</p>
2257 <pre>
2258 %A = xor undef, undef
2260 %B = undef
2261 %C = xor %B, %B
2263 %D = undef
2264 %E = icmp lt %D, 4
2265 %F = icmp gte %D, 4
2267 Safe:
2268 %A = undef
2269 %B = undef
2270 %C = undef
2271 %D = undef
2272 %E = undef
2273 %F = undef
2274 </pre>
2275 </div>
2277 <p>This example points out that two undef operands are not necessarily the same.
2278 This can be surprising to people (and also matches C semantics) where they
2279 assume that "X^X" is always zero, even if X is undef. This isn't true for a
2280 number of reasons, but the short answer is that an undef "variable" can
2281 arbitrarily change its value over its "live range". This is true because the
2283 logically read from arbitrary registers that happen to be around when needed,
2284 so the value is not necessarily consistent over time. In fact, %A and %C need
2285 to have the same semantics or the core LLVM "replace all uses with" concept
2286 would not hold.</p>
2289 <pre>
2290 %A = fdiv undef, %X
2291 %B = fdiv %X, undef
2292 Safe:
2293 %A = undef
2294 b: unreachable
2295 </pre>
2296 </div>
2300 allowed to have an arbitrary bit-pattern. This means that the %A operation
2301 can be constant folded to undef because the undef could be an SNaN, and fdiv is
2302 not (currently) defined on SNaN's. However, in the second example, we can make
2303 a more aggressive assumption: because the undef is allowed to be an arbitrary
2304 value, we are allowed to assume that it could be zero. Since a divide by zero
2306 does not execute at all. This allows us to delete the divide and all code after
2307 it: since the undefined operation "can't happen", the optimizer can assume that
2308 it occurs in dead code.
2309 </p>
2312 <pre>
2313 a: store undef -> %X
2314 b: store %X -> undef
2315 Safe:
2317 b: unreachable
2318 </pre>
2319 </div>
2322 can be assumed to not have any effect: we can assume that the value is
2323 overwritten with bits that happen to match what was already there. However, a
2324 store "to" an undefined location could clobber arbitrary memory, therefore, it
2325 has undefined behavior.</p>
2327 </div>
2329 <!-- ======================================================================= -->
2334 instead of representing an unspecified bit pattern, they represent the
2335 fact that an instruction or constant expression which cannot evoke side
2336 effects has nevertheless detected a condition which results in undefined
2337 behavior.</p>
2339 <p>There is currently no way of representing a trap value in the IR; they
2340 only exist when produced by operations such as
2345 <p>
2346 <ul>
2348 their operands.</li>
2351 to their dynamic predecessor basic block.</li>
2353 <li>Function arguments depend on the corresponding actual argument values in
2354 the dynamic callers of their functions.</li>
2358 control back to them.</li>
2362 or exception-throwing call instructions that dynamically transfer control
2363 back to them.</li>
2365 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2366 referenced memory addresses, following the order in the IR
2367 (including loads and stores implied by intrinsics such as
2370 <!-- TODO: In the case of multiple threads, this only applies if the store
2371 "happens-before" the load or store. -->
2373 <!-- TODO: floating-point exception state -->
2375 <li>An instruction with externally visible side effects depends on the most
2376 recent preceding instruction with externally visible side effects, following
2377 the order in the IR. (This includes volatile loads and stores.)</li>
2381 if the terminator instruction has multiple successors and the instruction
2382 is always executed when control transfers to one of the successors, and
2383 may not be executed when control is transfered to another.</li>
2387 </ul>
2388 </p>
2390 <p>Whenever a trap value is generated, all values which depend on it evaluate
2391 to trap. If they have side effects, the evoke their side effects as if each
2392 operand with a trap value were undef. If they have externally-visible side
2393 effects, the behavior is undefined.</p>
2398 <pre>
2399 entry:
2402 %trap_yet_again = getelementptr i32* @h, i32 %still_trap
2403 store i32 0, i32* %trap_yet_again ; undefined behavior
2405 store i32 %trap, i32* @g ; Trap value conceptually stored to memory.
2406 %trap2 = load i32* @g ; Returns a trap value, not just undef.
2408 volatile store i32 %trap, i32* @g ; External observation; undefined behavior.
2410 %narrowaddr = bitcast i32* @g to i16*
2411 %wideaddr = bitcast i32* @g to i64*
2412 %trap3 = load 16* %narrowaddr ; Returns a trap value.
2413 %trap4 = load i64* %widaddr ; Returns a trap value.
2415 %cmp = icmp i32 slt %trap, 0 ; Returns a trap value.
2416 %br i1 %cmp, %true, %end ; Branch to either destination.
2418 true:
2419 volatile store i32 0, i32* @g ; This is control-dependent on %cmp, so
2420 ; it has undefined behavior.
2421 br label %end
2423 end:
2425 ; Both edges into this PHI are
2426 ; control-dependent on %cmp, so this
2427 ; always results in a trap value.
2429 volatile store i32 0, i32* @g ; %end is control-equivalent to %entry
2430 ; so this is defined (ignoring earlier
2431 ; undefined behavior in this example).
2432 </pre>
2433 </div>
2435 </div>
2437 <!-- ======================================================================= -->
2439 Blocks</a></div>
2445 basic block in the specified function, and always has an i8* type. Taking
2446 the address of the entry block is illegal.</p>
2448 <p>This value only has defined behavior when used as an operand to the
2450 against null. Pointer equality tests between labels addresses is undefined
2451 behavior - though, again, comparison against null is ok, and no label is
2452 equal to the null pointer. This may also be passed around as an opaque
2453 pointer sized value as long as the bits are not inspected. This allows
2457 <p>Finally, some targets may provide defined semantics when
2458 using the value as the operand to an inline assembly, but that is target
2459 specific.
2460 </p>
2462 </div>
2465 <!-- ======================================================================= -->
2467 </div>
2471 <p>Constant expressions are used to allow expressions involving other constants
2472 to be used as constants. Constant expressions may be of
2474 operation that does not have side effects (e.g. load and call are not
2475 supported). The following is the syntax for constant expressions:</p>
2477 <dl>
2479 <dd>Truncate a constant to another type. The bit size of CST must be larger
2480 than the bit size of TYPE. Both types must be integers.</dd>
2483 <dd>Zero extend a constant to another type. The bit size of CST must be
2484 smaller or equal to the bit size of TYPE. Both types must be
2485 integers.</dd>
2488 <dd>Sign extend a constant to another type. The bit size of CST must be
2489 smaller or equal to the bit size of TYPE. Both types must be
2490 integers.</dd>
2493 <dd>Truncate a floating point constant to another floating point type. The
2494 size of CST must be larger than the size of TYPE. Both types must be
2495 floating point.</dd>
2498 <dd>Floating point extend a constant to another type. The size of CST must be
2499 smaller or equal to the size of TYPE. Both types must be floating
2500 point.</dd>
2503 <dd>Convert a floating point constant to the corresponding unsigned integer
2504 constant. TYPE must be a scalar or vector integer type. CST must be of
2505 scalar or vector floating point type. Both CST and TYPE must be scalars,
2506 or vectors of the same number of elements. If the value won't fit in the
2507 integer type, the results are undefined.</dd>
2510 <dd>Convert a floating point constant to the corresponding signed integer
2511 constant. TYPE must be a scalar or vector integer type. CST must be of
2512 scalar or vector floating point type. Both CST and TYPE must be scalars,
2513 or vectors of the same number of elements. If the value won't fit in the
2514 integer type, the results are undefined.</dd>
2517 <dd>Convert an unsigned integer constant to the corresponding floating point
2518 constant. TYPE must be a scalar or vector floating point type. CST must be
2519 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2520 vectors of the same number of elements. If the value won't fit in the
2521 floating point type, the results are undefined.</dd>
2524 <dd>Convert a signed integer constant to the corresponding floating point
2525 constant. TYPE must be a scalar or vector floating point type. CST must be
2526 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2527 vectors of the same number of elements. If the value won't fit in the
2528 floating point type, the results are undefined.</dd>
2531 <dd>Convert a pointer typed constant to the corresponding integer constant
2537 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2538 type. CST must be of integer type. The CST value is zero extended,
2539 truncated, or unchanged to make it fit in a pointer size. This one is
2543 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2551 instruction, the index list may have zero or more indexes, which are
2565 constants.</dd>
2569 constants.</dd>
2573 constants.</dd>
2576 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2579 on operands are the same as those for the corresponding instruction
2580 (e.g. no bitwise operations on floating point values are allowed).</dd>
2581 </dl>
2583 </div>
2585 <!-- *********************************************************************** -->
2587 <!-- *********************************************************************** -->
2589 <!-- ======================================================================= -->
2592 </div>
2596 <p>LLVM supports inline assembler expressions (as opposed
2598 a special value. This value represents the inline assembler as a string
2599 (containing the instructions to emit), a list of operand constraints (stored
2600 as a string), a flag that indicates whether or not the inline asm
2601 expression has side effects, and a flag indicating whether the function
2602 containing the asm needs to align its stack conservatively. An example
2603 inline assembler expression is:</p>
2606 <pre>
2608 </pre>
2609 </div>
2613 have:</p>
2616 <pre>
2618 </pre>
2619 </div>
2621 <p>Inline asms with side effects not visible in the constraint list must be
2622 marked as having side effects. This is done through the use of the
2626 <pre>
2628 </pre>
2629 </div>
2631 <p>In some cases inline asms will contain code that will not work unless the
2632 stack is aligned in some way, such as calls or SSE instructions on x86,
2633 yet will not contain code that does that alignment within the asm.
2634 The compiler should make conservative assumptions about what the asm might
2635 contain and should generate its usual stack alignment code in the prologue
2639 <pre>
2641 </pre>
2642 </div>
2645 first.</p>
2647 <p>TODO: The format of the asm and constraints string still need to be
2648 documented here. Constraints on what can be done (e.g. duplication, moving,
2649 etc need to be documented). This is probably best done by reference to
2650 another document that covers inline asm from a holistic perspective.</p>
2651 </div>
2655 </div>
2660 attached to it that contains a constant integer. If present, the code
2661 generator will use the integer as the location cookie value when report
2662 errors through the LLVMContext error reporting mechanisms. This allows a
2663 front-end to correlate backend errors that occur with inline asm back to the
2664 source code that produced it. For example:</p>
2667 <pre>
2669 ...
2671 </pre>
2672 </div>
2674 <p>It is up to the front-end to make sense of the magic numbers it places in the
2675 IR.</p>
2677 </div>
2679 <!-- ======================================================================= -->
2681 Strings</a>
2682 </div>
2686 <p>LLVM IR allows metadata to be attached to instructions in the program that
2687 can convey extra information about the code to the optimizers and code
2688 generator. One example application of metadata is source-level debug
2689 information. There are two metadata primitives: strings and nodes. All
2693 <p>A metadata string is a string surrounded by double quotes. It can contain
2697 <p>Metadata nodes are represented with notation similar to structure constants
2698 (a comma separated list of elements, surrounded by braces and preceded by an
2703 metadata nodes, which can be looked up in the module symbol table. For
2704 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2707 function is using two metadata arguments.
2710 <pre>
2712 </pre>
2713 </div></p>
2719 <pre>
2721 </pre>
2722 </div></p>
2723 </div>
2726 <!-- *********************************************************************** -->
2729 </div>
2730 <!-- *********************************************************************** -->
2733 code generation or other IR semantics. These are documented here. All globals
2734 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2735 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2736 by LLVM.</p>
2738 <!-- ======================================================================= -->
2741 </div>
2747 pointers to global variables and functions which may optionally have a pointer
2748 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2750 <pre>
2751 @X = global i8 4
2752 @Y = global i32 123
2754 @llvm.used = appending global [2 x i8*] [
2755 i8* @X,
2756 i8* bitcast (i32* @Y to i8*)
2757 ], section "llvm.metadata"
2758 </pre>
2761 compiler, assembler, and linker are required to treat the symbol as if there is
2762 a reference to the global that it cannot see. For example, if a variable has
2764 list, it cannot be deleted. This is commonly used to represent references from
2765 inline asms and other things the compiler cannot "see", and corresponds to
2768 <p>On some targets, the code generator must emit a directive to the assembler or
2769 object file to prevent the assembler and linker from molesting the symbol.</p>
2771 </div>
2773 <!-- ======================================================================= -->
2776 </div>
2782 touching the symbol. On targets that support it, this allows an intelligent
2783 linker to optimize references to the symbol without being impeded as it would be
2786 <p>This is a rare construct that should only be used in rare circumstances, and
2787 should not be exposed to source languages.</p>
2789 </div>
2791 <!-- ======================================================================= -->
2794 </div>
2797 <pre>
2798 %0 = type { i32, void ()* }
2800 </pre>
2801 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor functions and associated priorities. The functions referenced by this array will be called in ascending order of priority (i.e. lowest first) when the module is loaded. The order of functions with the same priority is not defined.
2802 </p>
2804 </div>
2806 <!-- ======================================================================= -->
2809 </div>
2812 <pre>
2813 %0 = type { i32, void ()* }
2815 </pre>
2817 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions and associated priorities. The functions referenced by this array will be called in descending order of priority (i.e. highest first) when the module is loaded. The order of functions with the same priority is not defined.
2818 </p>
2820 </div>
2823 <!-- *********************************************************************** -->
2825 <!-- *********************************************************************** -->
2829 <p>The LLVM instruction set consists of several different classifications of
2836 </div>
2838 <!-- ======================================================================= -->
2840 Instructions</a> </div>
2845 in a program ends with a "Terminator" instruction, which indicates which
2846 block should be executed after the current block is finished. These
2848 control flow, not values (the one exception being the
2851 <p>There are seven different terminator instructions: the
2860 </div>
2862 <!-- _______________________________________________________________________ -->
2864 Instruction</a> </div>
2869 <pre>
2872 </pre>
2876 a value) from a function back to the caller.</p>
2879 value and then causes control flow, and one that just causes control flow to
2880 occur.</p>
2884 return value. The type of the return value must be a
2889 value or a return value with a type that does not match its type, or if it
2891 return value.</p>
2895 the calling function's context. If the caller is a
2897 instruction after the call. If the caller was an
2899 the beginning of the "normal" destination block. If the instruction returns
2900 a value, that value shall set the call or invoke instruction's return
2901 value.</p>
2904 <pre>
2908 </pre>
2910 </div>
2911 <!-- _______________________________________________________________________ -->
2917 <pre>
2918 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2919 </pre>
2923 different basic block in the current function. There are two forms of this
2924 instruction, corresponding to a conditional branch and an unconditional
2925 branch.</p>
2931 target.</p>
2940 <pre>
2941 Test:
2943 br i1 %cond, label %IfEqual, label %IfUnequal
2944 IfEqual:
2946 IfUnequal:
2948 </pre>
2950 </div>
2952 <!-- _______________________________________________________________________ -->
2955 </div>
2960 <pre>
2961 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2962 </pre>
2967 instruction, allowing a branch to occur to one of many possible
2968 destinations.</p>
2974 The table is not allowed to contain duplicate constant entries.</p>
2979 is searched for the given value. If the value is found, control flow is
2980 transferred to the corresponding destination; otherwise, control flow is
2981 transferred to the default destination.</p>
2984 <p>Depending on properties of the target machine and the particular
2986 different ways. For example, it could be generated as a series of chained
2987 conditional branches or with a lookup table.</p>
2990 <pre>
2993 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2996 switch i32 0, label %dest [ ]
2999 switch i32 %val, label %otherwise [ i32 0, label %onzero
3000 i32 1, label %onone
3001 i32 2, label %ontwo ]
3002 </pre>
3004 </div>
3007 <!-- _______________________________________________________________________ -->
3010 </div>
3015 <pre>
3017 </pre>
3022 within the current function, whose address is specified by
3029 rest of the arguments indicate the full set of possible destinations that the
3030 address may point to. Blocks are allowed to occur multiple times in the
3031 destination list, though this isn't particularly useful.</p>
3033 <p>This destination list is required so that dataflow analysis has an accurate
3034 understanding of the CFG.</p>
3038 <p>Control transfers to the block specified in the address argument. All
3039 possible destination blocks must be listed in the label list, otherwise this
3040 instruction has undefined behavior. This implies that jumps to labels
3041 defined in other functions have undefined behavior as well.</p>
3048 <pre>
3049 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3050 </pre>
3052 </div>
3055 <!-- _______________________________________________________________________ -->
3058 </div>
3063 <pre>
3064 <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>]
3066 </pre>
3070 function, with the possibility of control flow transfer to either the
3073 control flow will return to the "normal" label. If the callee (or any
3075 instruction, control is interrupted and continued at the dynamically nearest
3081 <ol>
3083 convention</a> the call should use. If none is specified, the call
3084 defaults to using C calling conventions.</li>
3091 function value being invoked. In most cases, this is a direct function
3093 off an arbitrary pointer to function value.</li>
3096 function to be invoked. </li>
3099 signature argument types and parameter attributes. All arguments must be
3101 signature indicates the function accepts a variable number of arguments,
3102 the extra arguments can be specified.</li>
3113 </ol>
3116 <p>This instruction is designed to operate as a standard
3118 primary difference is that it establishes an association with a label, which
3119 is used by the runtime library to unwind the stack.</p>
3121 <p>This instruction is used in languages with destructors to ensure that proper
3123 exception. Additionally, this is important for implementation of
3126 <p>For the purposes of the SSA form, the definition of the value returned by the
3128 block to the "normal" label. If the callee unwinds then no return value is
3129 available.</p>
3131 <p>Note that the code generator does not yet completely support unwind, and
3132 that the invoke/unwind semantics are likely to change in future versions.</p>
3135 <pre>
3136 %retval = invoke i32 @Test(i32 15) to label %Continue
3140 </pre>
3142 </div>
3144 <!-- _______________________________________________________________________ -->
3147 Instruction</a> </div>
3152 <pre>
3153 unwind
3154 </pre>
3158 at the first callee in the dynamic call stack which used
3160 This is primarily used to implement exception handling.</p>
3164 immediately halt. The dynamic call stack is then searched for the
3166 Once found, execution continues at the "exceptional" destination block
3168 instruction in the dynamic call chain, undefined behavior results.</p>
3170 <p>Note that the code generator does not yet completely support unwind, and
3171 that the invoke/unwind semantics are likely to change in future versions.</p>
3173 </div>
3175 <!-- _______________________________________________________________________ -->
3178 Instruction</a> </div>
3183 <pre>
3184 unreachable
3185 </pre>
3189 instruction is used to inform the optimizer that a particular portion of the
3190 code is not reachable. This can be used to indicate that the code after a
3191 no-return function cannot be reached, and other facts.</p>
3196 </div>
3198 <!-- ======================================================================= -->
3203 <p>Binary operators are used to do most of the computation in a program. They
3204 require two operands of the same type, execute an operation on them, and
3205 produce a single value. The operands might represent multiple data, as is
3207 has the same type as its operands.</p>
3211 </div>
3213 <!-- _______________________________________________________________________ -->
3216 </div>
3221 <pre>
3226 </pre>
3234 integer values. Both arguments must have identical types.</p>
3239 <p>If the sum has unsigned overflow, the result returned is the mathematical
3242 <p>Because LLVM integers use a two's complement representation, this instruction
3243 is appropriate for both signed and unsigned integers.</p>
3249 respectively, occurs.</p>
3252 <pre>
3254 </pre>
3256 </div>
3258 <!-- _______________________________________________________________________ -->
3261 </div>
3266 <pre>
3268 </pre>
3276 floating point values. Both arguments must have identical types.</p>
3282 <pre>
3284 </pre>
3286 </div>
3288 <!-- _______________________________________________________________________ -->
3291 </div>
3296 <pre>
3301 </pre>
3305 operands.</p>
3309 representations.</p>
3314 integer values. Both arguments must have identical types.</p>
3319 <p>If the difference has unsigned overflow, the result returned is the
3321 result.</p>
3323 <p>Because LLVM integers use a two's complement representation, this instruction
3324 is appropriate for both signed and unsigned integers.</p>
3330 respectively, occurs.</p>
3333 <pre>
3336 </pre>
3338 </div>
3340 <!-- _______________________________________________________________________ -->
3343 </div>
3348 <pre>
3350 </pre>
3354 operands.</p>
3358 representations.</p>
3363 floating point values. Both arguments must have identical types.</p>
3369 <pre>
3372 </pre>
3374 </div>
3376 <!-- _______________________________________________________________________ -->
3379 </div>
3384 <pre>
3389 </pre>
3397 integer values. Both arguments must have identical types.</p>
3402 <p>If the result of the multiplication has unsigned overflow, the result
3404 width of the result.</p>
3406 <p>Because LLVM integers use a two's complement representation, and the result
3407 is the same width as the operands, this instruction returns the correct
3408 result for both signed and unsigned integers. If a full product
3410 be sign-extended or zero-extended as appropriate to the width of the full
3411 product.</p>
3417 respectively, occurs.</p>
3420 <pre>
3422 </pre>
3424 </div>
3426 <!-- _______________________________________________________________________ -->
3429 </div>
3434 <pre>
3436 </pre>
3444 floating point values. Both arguments must have identical types.</p>
3450 <pre>
3452 </pre>
3454 </div>
3456 <!-- _______________________________________________________________________ -->
3458 </a></div>
3463 <pre>
3465 </pre>
3473 values. Both arguments must have identical types.</p>
3478 <p>Note that unsigned integer division and signed integer division are distinct
3484 <pre>
3486 </pre>
3488 </div>
3490 <!-- _______________________________________________________________________ -->
3492 </a> </div>
3497 <pre>
3500 </pre>
3508 values. Both arguments must have identical types.</p>
3511 <p>The value produced is the signed integer quotient of the two operands rounded
3512 towards zero.</p>
3514 <p>Note that signed integer division and unsigned integer division are distinct
3517 <p>Division by zero leads to undefined behavior. Overflow also leads to
3518 undefined behavior; this is a rare case, but can occur, for example, by doing
3523 be rounded or if overflow would occur.</p>
3526 <pre>
3528 </pre>
3530 </div>
3532 <!-- _______________________________________________________________________ -->
3534 Instruction</a> </div>
3539 <pre>
3541 </pre>
3549 floating point values. Both arguments must have identical types.</p>
3555 <pre>
3557 </pre>
3559 </div>
3561 <!-- _______________________________________________________________________ -->
3563 </div>
3568 <pre>
3570 </pre>
3574 division of its two arguments.</p>
3579 values. Both arguments must have identical types.</p>
3583 This instruction always performs an unsigned division to get the
3584 remainder.</p>
3586 <p>Note that unsigned integer remainder and signed integer remainder are
3592 <pre>
3594 </pre>
3596 </div>
3598 <!-- _______________________________________________________________________ -->
3601 </div>
3606 <pre>
3608 </pre>
3612 division of its two operands. This instruction can also take
3614 elements must be integers.</p>
3619 values. Both arguments must have identical types.</p>
3625 a value. For more information about the difference,
3627 Math Forum</a>. For a table of how this is implemented in various languages,
3631 <p>Note that signed integer remainder and unsigned integer remainder are
3634 <p>Taking the remainder of a division by zero leads to undefined behavior.
3635 Overflow also leads to undefined behavior; this is a rare case, but can
3636 occur, for example, by taking the remainder of a 32-bit division of
3638 lets srem be implemented using instructions that return both the result of
3639 the division and the remainder.)</p>
3642 <pre>
3644 </pre>
3646 </div>
3648 <!-- _______________________________________________________________________ -->
3655 <pre>
3657 </pre>
3661 its two operands.</p>
3666 floating point values. Both arguments must have identical types.</p>
3670 has the same sign as the dividend.</p>
3673 <pre>
3675 </pre>
3677 </div>
3679 <!-- ======================================================================= -->
3681 Operations</a> </div>
3685 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3686 program. They are generally very efficient instructions and can commonly be
3687 strength reduced from other instructions. They require two operands of the
3688 same type, execute an operation on them, and produce a single value. The
3689 resulting value is the same type as its operands.</p>
3691 </div>
3693 <!-- _______________________________________________________________________ -->
3695 Instruction</a> </div>
3700 <pre>
3702 </pre>
3706 a specified number of bits.</p>
3716 is (statically or dynamically) negative or equal to or larger than the number
3722 <pre>
3727 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3728 </pre>
3730 </div>
3732 <!-- _______________________________________________________________________ -->
3734 Instruction</a> </div>
3739 <pre>
3741 </pre>
3745 operand shifted to the right a specified number of bits with zero fill.</p>
3753 <p>This instruction always performs a logical shift right operation. The most
3754 significant bits of the result will be filled with zero bits after the shift.
3761 <pre>
3767 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3768 </pre>
3770 </div>
3772 <!-- _______________________________________________________________________ -->
3774 Instruction</a> </div>
3778 <pre>
3780 </pre>
3784 operand shifted to the right a specified number of bits with sign
3785 extension.</p>
3793 <p>This instruction always performs an arithmetic shift right operation, The
3794 most significant bits of the result will be filled with the sign bit
3801 <pre>
3807 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3808 </pre>
3810 </div>
3812 <!-- _______________________________________________________________________ -->
3814 Instruction</a> </div>
3819 <pre>
3821 </pre>
3825 operands.</p>
3830 values. Both arguments must have identical types.</p>
3836 <tbody>
3837 <tr>
3841 </tr>
3842 <tr>
3846 </tr>
3847 <tr>
3851 </tr>
3852 <tr>
3856 </tr>
3857 <tr>
3861 </tr>
3862 </tbody>
3863 </table>
3866 <pre>
3870 </pre>
3871 </div>
3872 <!-- _______________________________________________________________________ -->
3878 <pre>
3880 </pre>
3884 two operands.</p>
3889 values. Both arguments must have identical types.</p>
3895 <tbody>
3896 <tr>
3900 </tr>
3901 <tr>
3905 </tr>
3906 <tr>
3910 </tr>
3911 <tr>
3915 </tr>
3916 <tr>
3920 </tr>
3921 </tbody>
3922 </table>
3925 <pre>
3929 </pre>
3931 </div>
3933 <!-- _______________________________________________________________________ -->
3935 Instruction</a> </div>
3940 <pre>
3942 </pre>
3952 values. Both arguments must have identical types.</p>
3958 <tbody>
3959 <tr>
3963 </tr>
3964 <tr>
3968 </tr>
3969 <tr>
3973 </tr>
3974 <tr>
3978 </tr>
3979 <tr>
3983 </tr>
3984 </tbody>
3985 </table>
3988 <pre>
3993 </pre>
3995 </div>
3997 <!-- ======================================================================= -->
4000 </div>
4004 <p>LLVM supports several instructions to represent vector operations in a
4005 target-independent manner. These instructions cover the element-access and
4006 vector-specific operations needed to process vectors effectively. While LLVM
4007 does directly support these vector operations, many sophisticated algorithms
4008 will want to use target-specific intrinsics to take full advantage of a
4009 specific target.</p>
4011 </div>
4013 <!-- _______________________________________________________________________ -->
4016 </div>
4021 <pre>
4022 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4023 </pre>
4027 from a vector at a specified index.</p>
4033 indicating the position from which to extract the element. The index may be
4034 a variable.</p>
4037 <p>The result is a scalar of the same type as the element type of
4040 results are undefined.</p>
4043 <pre>
4045 </pre>
4047 </div>
4049 <!-- _______________________________________________________________________ -->
4052 </div>
4057 <pre>
4058 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4059 </pre>
4063 vector at a specified index.</p>
4068 whose type must equal the element type of the first operand. The third
4069 operand is an index indicating the position at which to insert the value.
4070 The index may be a variable.</p>
4076 results are undefined.</p>
4079 <pre>
4080 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4081 </pre>
4083 </div>
4085 <!-- _______________________________________________________________________ -->
4088 </div>
4093 <pre>
4094 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4095 </pre>
4099 from two input vectors, returning a vector with the same element type as the
4100 input and length that is the same as the shuffle mask.</p>
4104 with types that match each other. The third argument is a shuffle mask whose
4105 element type is always 'i32'. The result of the instruction is a vector
4106 whose length is the same as the shuffle mask and whose element type is the
4107 same as the element type of the first two operands.</p>
4109 <p>The shuffle mask operand is required to be a constant vector with either
4110 constant integer or undef values.</p>
4113 <p>The elements of the two input vectors are numbered from left to right across
4114 both of the vectors. The shuffle mask operand specifies, for each element of
4115 the result vector, which element of the two input vectors the result element
4116 gets. The element selector may be undef (meaning "don't care") and the
4117 second operand may be undef if performing a shuffle from only one vector.</p>
4120 <pre>
4124 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4128 <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>
4129 </pre>
4131 </div>
4133 <!-- ======================================================================= -->
4136 </div>
4140 <p>LLVM supports several instructions for working with
4143 </div>
4145 <!-- _______________________________________________________________________ -->
4148 </div>
4153 <pre>
4155 </pre>
4165 specify which value to extract in a similar manner as indices in a
4169 <p>The result is the value at the position in the aggregate specified by the
4170 index operands.</p>
4173 <pre>
4175 </pre>
4177 </div>
4179 <!-- _______________________________________________________________________ -->
4182 </div>
4187 <pre>
4188 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx> <i>; yields <aggregate type></i>
4189 </pre>
4199 value to insert. The following operands are constant indices indicating
4200 the position at which to insert the value in a similar manner as indices in a
4202 value to insert must have the same type as the value identified by the
4203 indices.</p>
4211 <pre>
4214 </pre>
4216 </div>
4219 <!-- ======================================================================= -->
4222 </div>
4226 <p>A key design point of an SSA-based representation is how it represents
4227 memory. In LLVM, no memory locations are in SSA form, which makes things
4228 very simple. This section describes how to read, write, and allocate
4229 memory in LLVM.</p>
4231 </div>
4233 <!-- _______________________________________________________________________ -->
4236 </div>
4241 <pre>
4242 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4243 </pre>
4247 currently executing function, to be automatically released when this function
4248 returns to its caller. The object is always allocated in the generic address
4249 space (address space zero).</p>
4254 runtime stack, returning a pointer of the appropriate type to the program.
4255 If "NumElements" is specified, it is the number of elements allocated,
4256 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4257 specified, the value result of the allocation is guaranteed to be aligned to
4258 at least that boundary. If not specified, or if zero, the target can choose
4259 to align the allocation on any convenient boundary compatible with the
4260 type.</p>
4265 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4267 memory is automatically released when the function returns. The
4269 variables that must have an address available. When the function returns
4272 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4275 <pre>
4280 </pre>
4282 </div>
4284 <!-- _______________________________________________________________________ -->
4286 Instruction</a> </div>
4291 <pre>
4292 <result> = load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4293 <result> = volatile load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4295 </pre>
4302 from which to load. The pointer must point to
4309 operation (that is, the alignment of the memory address). A value of 0 or an
4311 alignment for the target. It is the responsibility of the code emitter to
4312 ensure that the alignment information is correct. Overestimating the
4313 alignment results in undefined behavior. Underestimating the alignment may
4320 and code generator that this load is not expected to be reused in the cache.
4321 The code generator may select special instructions to save cache bandwidth,
4325 <p>The location of memory pointed to is loaded. If the value being loaded is of
4326 scalar type then the number of bytes read does not exceed the minimum number
4327 of bytes needed to hold all bits of the type. For example, loading an
4330 is undefined if the value was not originally written using a store of the
4331 same type.</p>
4334 <pre>
4338 </pre>
4340 </div>
4342 <!-- _______________________________________________________________________ -->
4344 Instruction</a> </div>
4349 <pre>
4350 store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4351 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4352 </pre>
4359 and an address at which to store it. The type of the
4368 operation (that is, the alignment of the memory address). A value of 0 or an
4369 omitted "align" argument means that the operation has the preferential
4370 alignment for the target. It is the responsibility of the code emitter to
4371 ensure that the alignment information is correct. Overestimating the
4372 alignment results in an undefined behavior. Underestimating the alignment may
4375 <p>The optional !nontemporal metadata must reference a single metatadata
4376 name <index> corresponding to a metadata node with one i32 entry of
4377 value 1. The existence of the !nontemporal metatadata on the
4378 instruction tells the optimizer and code generator that this load is
4379 not expected to be reused in the cache. The code generator may
4380 select special instructions to save cache bandwidth, such as the
4381 MOVNT instruction on x86.</p>
4388 does not exceed the minimum number of bytes needed to hold all bits of the
4391 integral number of bytes, it is unspecified what happens to the extra bits
4392 that do not belong to the type, but they will typically be overwritten.</p>
4395 <pre>
4399 </pre>
4401 </div>
4403 <!-- _______________________________________________________________________ -->
4406 </div>
4411 <pre>
4414 </pre>
4419 It performs address calculation only and does not access memory.</p>
4422 <p>The first argument is always a pointer, and forms the basis of the
4423 calculation. The remaining arguments are indices that indicate which of the
4424 elements of the aggregate object are indexed. The interpretation of each
4425 index is dependent on the type being indexed into. The first index always
4426 indexes the pointer value given as the first argument, the second index
4427 indexes a value of the type pointed to (not necessarily the value directly
4428 pointed to, since the first index can be non-zero), etc. The first type
4429 indexed into must be a pointer value, subsequent types can be arrays,
4430 vectors, structs and unions. Note that subsequent types being indexed into
4431 can never be pointers, since that would require loading the pointer before
4432 continuing calculation.</p>
4434 <p>The type of each index argument depends on the type it is indexing into.
4437 or vector, integers of any width are allowed, and they are not required to be
4438 constant.</p>
4440 <p>For example, let's consider a C code fragment and how it gets compiled to
4441 LLVM:</p>
4444 <pre>
4445 struct RT {
4446 char A;
4448 char C;
4449 };
4450 struct ST {
4451 int X;
4452 double Y;
4453 struct RT Z;
4454 };
4456 int *foo(struct ST *s) {
4458 }
4459 </pre>
4460 </div>
4465 <pre>
4469 define i32* @foo(%ST* %s) {
4470 entry:
4472 ret i32* %reg
4473 }
4474 </pre>
4475 </div>
4480 }</tt>' type, a structure. The second index indexes into the third element
4482 i8 }</tt>' type, another structure. The third index indexes into the second
4484 array. The two dimensions of the array are subscripted into, yielding an
4488 <p>Note that it is perfectly legal to index partially through a structure,
4489 returning a pointer to an inner element. Because of this, the LLVM code for
4490 the given testcase is equivalent to:</p>
4492 <pre>
4493 define i32* @foo(%ST* %s) {
4499 ret i32* %t5
4500 }
4501 </pre>
4506 or if any of the addresses that would be formed by successive addition of
4507 the offsets implied by the indices to the base address with infinitely
4510 the addresses that point into the object, plus the address one byte past
4511 the end.</p>
4514 the base address with silently-wrapping two's complement arithmetic, and
4516 pointed to by the base pointer. The result value may not necessarily be
4517 used to access memory though, even if it happens to point into allocated
4519 section for more information.</p>
4521 <p>The getelementptr instruction is often confusing. For some more insight into
4525 <pre>
4534 </pre>
4536 </div>
4538 <!-- ======================================================================= -->
4540 </div>
4544 <p>The instructions in this category are the conversion instructions (casting)
4545 which all take a single operand and a type. They perform various bit
4546 conversions on the operand.</p>
4548 </div>
4550 <!-- _______________________________________________________________________ -->
4553 </div>
4557 <pre>
4559 </pre>
4568 size and type of the result, which must be
4571 allowed.</p>
4580 <pre>
4584 </pre>
4586 </div>
4588 <!-- _______________________________________________________________________ -->
4591 </div>
4595 <pre>
4597 </pre>
4618 <pre>
4621 </pre>
4623 </div>
4625 <!-- _______________________________________________________________________ -->
4628 </div>
4632 <pre>
4634 </pre>
4654 <pre>
4657 </pre>
4659 </div>
4661 <!-- _______________________________________________________________________ -->
4664 </div>
4669 <pre>
4671 </pre>
4689 undefined.</p>
4692 <pre>
4695 </pre>
4697 </div>
4699 <!-- _______________________________________________________________________ -->
4702 </div>
4706 <pre>
4708 </pre>
4712 floating point value.</p>
4718 type must be smaller than the destination type.</p>
4728 <pre>
4731 </pre>
4733 </div>
4735 <!-- _______________________________________________________________________ -->
4738 </div>
4742 <pre>
4744 </pre>
4760 towards zero) unsigned integer value. If the value cannot fit
4764 <pre>
4768 </pre>
4770 </div>
4772 <!-- _______________________________________________________________________ -->
4775 </div>
4779 <pre>
4781 </pre>
4799 the results are undefined.</p>
4802 <pre>
4806 </pre>
4808 </div>
4810 <!-- _______________________________________________________________________ -->
4813 </div>
4817 <pre>
4819 </pre>
4834 integer quantity and converts it to the corresponding floating point
4835 value. If the value cannot fit in the floating point value, the results are
4836 undefined.</p>
4839 <pre>
4842 </pre>
4844 </div>
4846 <!-- _______________________________________________________________________ -->
4849 </div>
4853 <pre>
4855 </pre>
4870 quantity and converts it to the corresponding floating point value. If the
4871 value cannot fit in the floating point value, the results are undefined.</p>
4874 <pre>
4877 </pre>
4879 </div>
4881 <!-- _______________________________________________________________________ -->
4884 </div>
4888 <pre>
4890 </pre>
4904 truncating or zero extending that value to the size of the integer type. If
4908 change.</p>
4911 <pre>
4914 </pre>
4916 </div>
4918 <!-- _______________________________________________________________________ -->
4921 </div>
4925 <pre>
4927 </pre>
4935 value to cast, and a type to cast it to, which must be a
4943 than the size of a pointer then a zero extension is done. If they are the
4947 <pre>
4951 </pre>
4953 </div>
4955 <!-- _______________________________________________________________________ -->
4958 </div>
4962 <pre>
4964 </pre>
4972 non-aggregate first class value, and a type to cast it to, which must also be
4975 identical. If the source type is a pointer, the destination type must also be
4976 a pointer. This instruction supports bitwise conversion of vectors to
4977 integers and to vectors of other types (as long as they have the same
4978 size).</p>
4985 be converted to other pointer types with this instruction. To convert
4990 <pre>
4994 </pre>
4996 </div>
4998 <!-- ======================================================================= -->
5004 defy better classification.</p>
5006 </div>
5008 <!-- _______________________________________________________________________ -->
5010 </div>
5015 <pre>
5016 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5017 </pre>
5021 boolean values based on comparison of its two integer, integer vector, or
5022 pointer operands.</p>
5026 the condition code indicating the kind of comparison to perform. It is not a
5027 value, just a keyword. The possible condition code are:</p>
5029 <ol>
5040 </ol>
5044 typed. They must also be identical types.</p>
5050 result, as follows:</p>
5052 <ol>
5055 performed.</li>
5059 performed.</li>
5086 </ol>
5089 values are compared as if they were integers.</p>
5091 <p>If the operands are integer vectors, then they are compared element by
5096 <pre>
5103 </pre>
5105 <p>Note that the code generator does not yet support vector types with
5108 </div>
5110 <!-- _______________________________________________________________________ -->
5112 </div>
5117 <pre>
5118 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5119 </pre>
5123 values based on comparison of its operands.</p>
5125 <p>If the operands are floating point scalars, then the result type is a boolean
5128 <p>If the operands are floating point vectors, then the result type is a vector
5129 of boolean with the same number of elements as the operands being
5130 compared.</p>
5134 the condition code indicating the kind of comparison to perform. It is not a
5135 value, just a keyword. The possible condition code are:</p>
5137 <ol>
5154 </ol>
5162 identical types.</p>
5167 vectors, then the vectors are compared element by element. Each comparison
5169 follows:</p>
5171 <ol>
5215 </ol>
5218 <pre>
5223 </pre>
5225 <p>Note that the code generator does not yet support vector types with
5228 </div>
5230 <!-- _______________________________________________________________________ -->
5233 </div>
5238 <pre>
5240 </pre>
5244 SSA graph representing the function.</p>
5247 <p>The type of the incoming values is specified with the first type field. After
5249 one pair for each predecessor basic block of the current block. Only values
5251 arguments to the PHI node. Only labels may be used as the label
5252 arguments.</p>
5254 <p>There must be no non-phi instructions between the start of a basic block and
5255 the PHI instructions: i.e. PHI instructions must be first in a basic
5256 block.</p>
5258 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5259 occur on the edge from the corresponding predecessor block to the current
5261 value on the same edge).</p>
5265 specified by the pair corresponding to the predecessor basic block that
5266 executed just prior to the current block.</p>
5269 <pre>
5270 Loop: ; Infinite loop that counts from 0 on up...
5271 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5272 %nextindvar = add i32 %indvar, 1
5273 br label %Loop
5274 </pre>
5276 </div>
5278 <!-- _______________________________________________________________________ -->
5281 </div>
5286 <pre>
5287 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5290 </pre>
5294 condition, without branching.</p>
5299 values indicating the condition, and two values of the
5301 vectors and the condition is a scalar, then entire vectors are selected, not
5302 individual elements.</p>
5306 first value argument; otherwise, it returns the second value argument.</p>
5308 <p>If the condition is a vector of i1, then the value arguments must be vectors
5309 of the same size, and the selection is done element by element.</p>
5312 <pre>
5314 </pre>
5316 <p>Note that the code generator does not yet support conditions
5317 with vector type.</p>
5319 </div>
5321 <!-- _______________________________________________________________________ -->
5324 </div>
5329 <pre>
5330 <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>]
5331 </pre>
5339 <ol>
5341 access any allocas or varargs in the caller. Note that calls may be
5342 marked "tail" even if they do not occur before
5344 present, the function call is eligible for tail call optimization,
5346 optimized into a jump</a>. The code generator may optimize calls marked
5348 sibling call optimization</a> when the caller and callee have
5349 matching signatures, or 2) forced tail call optimization when the
5350 following extra requirements are met:
5351 <ul>
5352 <li>Caller and callee both have the calling
5354 <li>The call is in tail position (ret immediately follows call and ret
5355 uses value of call or is void).</li>
5359 constraints are met.</a></li>
5360 </ul>
5361 </li>
5364 convention</a> the call should use. If none is specified, the call
5365 defaults to using C calling conventions. The calling convention of the
5366 call must match the calling convention of the target function, or else the
5367 behavior is undefined.</li>
5374 type of the return value. Functions that return no value are marked
5378 being invoked. The argument types must match the types implied by this
5379 signature. This type can be omitted if the function is not varargs and if
5380 the function type does not return a pointer to a function.</li>
5383 be invoked. In most cases, this is a direct function invocation, but
5385 to function value.</li>
5388 signature argument types and parameter attributes. All arguments must be
5390 signature indicates the function accepts a variable number of arguments,
5391 the extra arguments can be specified.</li>
5396 </ol>
5400 a specified function, with its incoming arguments bound to the specified
5402 function, control flow continues with the instruction after the function
5403 call, and the return value of the function is bound to the result
5404 argument.</p>
5407 <pre>
5408 %retval = call i32 @test(i32 %argc)
5412 call void %foo(i8 97 signext)
5414 %struct.A = type { i32, i8 }
5420 </pre>
5422 <p>llvm treats calls to some functions with names and arguments that match the
5423 standard C99 library as being the C99 library functions, and may perform
5424 optimizations or generate code for them under that assumption. This is
5425 something we'd like to change in the future to provide better support for
5426 freestanding environments and non-C-based languages.</p>
5428 </div>
5430 <!-- _______________________________________________________________________ -->
5433 </div>
5438 <pre>
5440 </pre>
5444 the "variable argument" area of a function call. It is used to implement the
5449 argument. It returns a value of the specified argument type and increments
5456 to the next argument. For more information, see the variable argument
5459 <p>It is legal for this instruction to be called in a function which does not
5461 function.</p>
5465 argument.</p>
5470 <p>Note that the code generator does not yet fully support va_arg on many
5471 targets. Also, it does not currently support va_arg with aggregate types on
5472 any target.</p>
5474 </div>
5476 <!-- *********************************************************************** -->
5478 <!-- *********************************************************************** -->
5483 well known names and semantics and are required to follow certain
5484 restrictions. Overall, these intrinsics represent an extension mechanism for
5485 the LLVM language that does not require changing all of the transformations
5486 in LLVM when adding to the language (or the bitcode reader/writer, the
5487 parser, etc...).</p>
5490 prefix is reserved in LLVM for intrinsic names; thus, function names may not
5491 begin with this prefix. Intrinsic functions must always be external
5492 functions: you cannot define the body of intrinsic functions. Intrinsic
5493 functions may only be used in call or invoke instructions: it is illegal to
5494 take the address of an intrinsic function. Additionally, because intrinsic
5495 functions are part of the LLVM language, it is required if any are added that
5496 they be documented here.</p>
5498 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
5499 family of functions that perform the same operation but on different data
5500 types. Because LLVM can represent over 8 million different integer types,
5501 overloading is used commonly to allow an intrinsic function to operate on any
5502 integer type. One or more of the argument types or the result type can be
5503 overloaded to accept any integer type. Argument types may also be defined as
5504 exactly matching a previous argument's type or the result type. This allows
5505 an intrinsic function which accepts multiple arguments, but needs all of them
5506 to be of the same type, to only be overloaded with respect to a single
5507 argument or the result.</p>
5509 <p>Overloaded intrinsics will have the names of its overloaded argument types
5510 encoded into its function name, each preceded by a period. Only those types
5511 which are overloaded result in a name suffix. Arguments whose type is matched
5513 can take an integer of any width and returns an integer of exactly the same
5514 integer width. This leads to a family of functions such as
5516 %val)</tt>. Only one type, the return type, is overloaded, and only one type
5517 suffix is required. Because the argument's type is matched against the return
5518 type, it does not require its own name suffix.</p>
5520 <p>To learn how to add an intrinsic function, please see the
5523 </div>
5525 <!-- ======================================================================= -->
5528 </div>
5532 <p>Variable argument support is defined in LLVM with
5534 intrinsic functions. These functions are related to the similarly named
5537 <p>All of these functions operate on arguments that use a target-specific value
5538 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
5539 not define what this type is, so all transformations should be prepared to
5540 handle these functions regardless of the type used.</p>
5543 instruction and the variable argument handling intrinsic functions are
5544 used.</p>
5547 <pre>
5548 define i32 @test(i32 %X, ...) {
5549 ; Initialize variable argument processing
5550 %ap = alloca i8*
5551 %ap2 = bitcast i8** %ap to i8*
5552 call void @llvm.va_start(i8* %ap2)
5554 ; Read a single integer argument
5555 %tmp = va_arg i8** %ap, i32
5557 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5558 %aq = alloca i8*
5559 %aq2 = bitcast i8** %aq to i8*
5560 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5561 call void @llvm.va_end(i8* %aq2)
5563 ; Stop processing of arguments.
5564 call void @llvm.va_end(i8* %ap2)
5565 ret i32 %tmp
5566 }
5568 declare void @llvm.va_start(i8*)
5569 declare void @llvm.va_copy(i8*, i8*)
5570 declare void @llvm.va_end(i8*)
5571 </pre>
5572 </div>
5574 </div>
5576 <!-- _______________________________________________________________________ -->
5579 </div>
5585 <pre>
5587 </pre>
5598 macro available in C. In a target-dependent way, it initializes
5602 need to know the last argument of the function as the compiler can figure
5603 that out.</p>
5605 </div>
5607 <!-- _______________________________________________________________________ -->
5610 </div>
5615 <pre>
5617 </pre>
5621 which has been initialized previously
5630 macro available in C. In a target-dependent way, it destroys
5636 </div>
5638 <!-- _______________________________________________________________________ -->
5641 </div>
5646 <pre>
5648 </pre>
5652 from the source argument list to the destination argument list.</p>
5657 from.</p>
5661 macro available in C. In a target-dependent way, it copies the
5663 element. This intrinsic is necessary because
5665 arbitrarily complex and require, for example, memory allocation.</p>
5667 </div>
5669 <!-- ======================================================================= -->
5672 </div>
5677 Collection</a> (GC) requires the implementation and generation of these
5679 roots on the stack</a>, as well as garbage collector implementations that
5681 barriers. Front-ends for type-safe garbage collected languages should generate
5682 these intrinsics to make use of the LLVM garbage collectors. For more details,
5686 <p>The garbage collection intrinsics only operate on objects in the generic
5687 address space (address space zero).</p>
5689 </div>
5691 <!-- _______________________________________________________________________ -->
5694 </div>
5699 <pre>
5700 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5701 </pre>
5705 the code generator, and allows some metadata to be associated with it.</p>
5708 <p>The first argument specifies the address of a stack object that contains the
5709 root pointer. The second pointer (which must be either a constant or a
5710 global value address) contains the meta-data to be associated with the
5711 root.</p>
5715 location. At compile-time, the code generator generates information to allow
5720 </div>
5722 <!-- _______________________________________________________________________ -->
5725 </div>
5730 <pre>
5731 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5732 </pre>
5736 locations, allowing garbage collector implementations that require read
5737 barriers.</p>
5740 <p>The second argument is the address to read from, which should be an address
5741 allocated from the garbage collector. The first object is a pointer to the
5742 start of the referenced object, if needed by the language runtime (otherwise
5743 null).</p>
5747 instruction, but may be replaced with substantially more complex code by the
5752 </div>
5754 <!-- _______________________________________________________________________ -->
5757 </div>
5762 <pre>
5763 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5764 </pre>
5768 locations, allowing garbage collector implementations that require write
5769 barriers (such as generational or reference counting collectors).</p>
5772 <p>The first argument is the reference to store, the second is the start of the
5773 object to store it to, and the third is the address of the field of Obj to
5774 store to. If the runtime does not require a pointer to the object, Obj may
5775 be null.</p>
5779 instruction, but may be replaced with substantially more complex code by the
5784 </div>
5786 <!-- ======================================================================= -->
5789 </div>
5793 <p>These intrinsics are provided by LLVM to expose special features that may
5794 only be implemented with code generator support.</p>
5796 </div>
5798 <!-- _______________________________________________________________________ -->
5801 </div>
5806 <pre>
5808 </pre>
5812 target-specific value indicating the return address of the current function
5813 or one of its callers.</p>
5816 <p>The argument to this intrinsic indicates which function to return the address
5817 for. Zero indicates the calling function, one indicates its caller, etc.
5822 indicating the return address of the specified call frame, or zero if it
5823 cannot be identified. The value returned by this intrinsic is likely to be
5824 incorrect or 0 for arguments other than zero, so it should only be used for
5825 debugging purposes.</p>
5827 <p>Note that calling this intrinsic does not prevent function inlining or other
5828 aggressive transformations, so the value returned may not be that of the
5829 obvious source-language caller.</p>
5831 </div>
5833 <!-- _______________________________________________________________________ -->
5836 </div>
5841 <pre>
5843 </pre>
5847 target-specific frame pointer value for the specified stack frame.</p>
5850 <p>The argument to this intrinsic indicates which function to return the frame
5851 pointer for. Zero indicates the calling function, one indicates its caller,
5856 indicating the frame address of the specified call frame, or zero if it
5857 cannot be identified. The value returned by this intrinsic is likely to be
5858 incorrect or 0 for arguments other than zero, so it should only be used for
5859 debugging purposes.</p>
5861 <p>Note that calling this intrinsic does not prevent function inlining or other
5862 aggressive transformations, so the value returned may not be that of the
5863 obvious source-language caller.</p>
5865 </div>
5867 <!-- _______________________________________________________________________ -->
5870 </div>
5875 <pre>
5876 declare i8 *@llvm.stacksave()
5877 </pre>
5881 of the function stack, for use
5883 useful for implementing language features like scoped automatic variable
5884 sized arrays in C99.</p>
5887 <p>This intrinsic returns a opaque pointer value that can be passed
5895 </div>
5897 <!-- _______________________________________________________________________ -->
5900 </div>
5905 <pre>
5906 declare void @llvm.stackrestore(i8 * %ptr)
5907 </pre>
5911 the function stack to the state it was in when the
5913 executed. This is useful for implementing language features like scoped
5914 automatic variable sized arrays in C99.</p>
5917 <p>See the description
5920 </div>
5922 <!-- _______________________________________________________________________ -->
5925 </div>
5930 <pre>
5932 </pre>
5936 insert a prefetch instruction if supported; otherwise, it is a noop.
5937 Prefetches have no effect on the behavior of the program but can change its
5938 performance characteristics.</p>
5948 <p>This intrinsic does not modify the behavior of the program. In particular,
5949 prefetches cannot trap and do not produce a value. On targets that support
5950 this intrinsic, the prefetch can provide hints to the processor cache for
5951 better performance.</p>
5953 </div>
5955 <!-- _______________________________________________________________________ -->
5958 </div>
5963 <pre>
5965 </pre>
5969 Counter (PC) in a region of code to simulators and other tools. The method
5970 is target specific, but it is expected that the marker will use exported
5971 symbols to transmit the PC of the marker. The marker makes no guarantees
5972 that it will remain with any specific instruction after optimizations. It is
5973 possible that the presence of a marker will inhibit optimizations. The
5974 intended use is to be inserted after optimizations to allow correlations of
5975 simulation runs.</p>
5981 <p>This intrinsic does not modify the behavior of the program. Backends that do
5982 not support this intrinsic may ignore it.</p>
5984 </div>
5986 <!-- _______________________________________________________________________ -->
5989 </div>
5994 <pre>
5995 declare i64 @llvm.readcyclecounter( )
5996 </pre>
6000 counter register (or similar low latency, high accuracy clocks) on those
6001 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6002 should map to RPCC. As the backing counters overflow quickly (on the order
6006 <p>When directly supported, reading the cycle counter should not modify any
6007 memory. Implementations are allowed to either return a application specific
6008 value or a system wide value. On backends without support, this is lowered
6011 </div>
6013 <!-- ======================================================================= -->
6016 </div>
6020 <p>LLVM provides intrinsics for a few important standard C library functions.
6021 These intrinsics allow source-language front-ends to pass information about
6022 the alignment of the pointer arguments to the code generator, providing
6023 opportunity for more efficient code generation.</p>
6025 </div>
6027 <!-- _______________________________________________________________________ -->
6030 </div>
6036 integer bit width and for different address spaces. Not all targets support
6037 all bit widths however.</p>
6039 <pre>
6044 </pre>
6048 source location to the destination location.</p>
6051 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6052 and the pointers can be in specified address spaces.</p>
6056 <p>The first argument is a pointer to the destination, the second is a pointer
6057 to the source. The third argument is an integer argument specifying the
6058 number of bytes to copy, the fourth argument is the alignment of the
6059 source and destination locations, and the fifth is a boolean indicating a
6060 volatile access.</p>
6063 then the caller guarantees that both the source and destination pointers are
6064 aligned to that boundary.</p>
6068 The detailed access behavior is not very cleanly specified and it is unwise
6069 to depend on it.</p>
6074 source location to the destination location, which are not allowed to
6075 overlap. It copies "len" bytes of memory over. If the argument is known to
6076 be aligned to some boundary, this can be specified as the fourth argument,
6079 </div>
6081 <!-- _______________________________________________________________________ -->
6084 </div>
6089 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
6090 width and for different address space. Not all targets support all bit
6091 widths however.</p>
6093 <pre>
6098 </pre>
6102 source location to the destination location. It is similar to the
6104 overlap.</p>
6107 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6108 and the pointers can be in specified address spaces.</p>
6112 <p>The first argument is a pointer to the destination, the second is a pointer
6113 to the source. The third argument is an integer argument specifying the
6114 number of bytes to copy, the fourth argument is the alignment of the
6115 source and destination locations, and the fifth is a boolean indicating a
6116 volatile access.</p>
6119 then the caller guarantees that the source and destination pointers are
6120 aligned to that boundary.</p>
6124 The detailed access behavior is not very cleanly specified and it is unwise
6125 to depend on it.</p>
6130 source location to the destination location, which may overlap. It copies
6131 "len" bytes of memory over. If the argument is known to be aligned to some
6132 boundary, this can be specified as the fourth argument, otherwise it should
6135 </div>
6137 <!-- _______________________________________________________________________ -->
6140 </div>
6145 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
6146 width and for different address spaces. Not all targets support all bit
6147 widths however.</p>
6149 <pre>
6154 </pre>
6158 particular byte value.</p>
6161 intrinsic does not return a value, takes extra alignment/volatile arguments,
6162 and the destination can be in an arbitrary address space.</p>
6165 <p>The first argument is a pointer to the destination to fill, the second is the
6166 byte value to fill it with, the third argument is an integer argument
6167 specifying the number of bytes to fill, and the fourth argument is the known
6168 alignment of destination location.</p>
6171 then the caller guarantees that the destination pointer is aligned to that
6172 boundary.</p>
6176 The detailed access behavior is not very cleanly specified and it is unwise
6177 to depend on it.</p>
6181 at the destination location. If the argument is known to be aligned to some
6182 boundary, this can be specified as the fourth argument, otherwise it should
6185 </div>
6187 <!-- _______________________________________________________________________ -->
6190 </div>
6196 floating point or vector of floating point type. Not all targets support all
6197 types however.</p>
6199 <pre>
6200 declare float @llvm.sqrt.f32(float %Val)
6201 declare double @llvm.sqrt.f64(double %Val)
6202 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6203 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6204 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6205 </pre>
6211 behavior for negative numbers other than -0.0 (which allows for better
6212 optimization, because there is no need to worry about errno being
6216 <p>The argument and return value are floating point numbers of the same
6217 type.</p>
6220 <p>This function returns the sqrt of the specified operand if it is a
6221 nonnegative floating point number.</p>
6223 </div>
6225 <!-- _______________________________________________________________________ -->
6228 </div>
6234 floating point or vector of floating point type. Not all targets support all
6235 types however.</p>
6237 <pre>
6238 declare float @llvm.powi.f32(float %Val, i32 %power)
6239 declare double @llvm.powi.f64(double %Val, i32 %power)
6240 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6241 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6242 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6243 </pre>
6247 specified (positive or negative) power. The order of evaluation of
6248 multiplications is not defined. When a vector of floating point type is
6249 used, the second argument remains a scalar integer value.</p>
6252 <p>The second argument is an integer power, and the first is a value to raise to
6253 that power.</p>
6256 <p>This function returns the first value raised to the second power with an
6257 unspecified sequence of rounding operations.</p>
6259 </div>
6261 <!-- _______________________________________________________________________ -->
6264 </div>
6270 floating point or vector of floating point type. Not all targets support all
6271 types however.</p>
6273 <pre>
6274 declare float @llvm.sin.f32(float %Val)
6275 declare double @llvm.sin.f64(double %Val)
6276 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6277 declare fp128 @llvm.sin.f128(fp128 %Val)
6278 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6279 </pre>
6285 <p>The argument and return value are floating point numbers of the same
6286 type.</p>
6289 <p>This function returns the sine of the specified operand, returning the same
6291 in the same way.</p>
6293 </div>
6295 <!-- _______________________________________________________________________ -->
6298 </div>
6304 floating point or vector of floating point type. Not all targets support all
6305 types however.</p>
6307 <pre>
6308 declare float @llvm.cos.f32(float %Val)
6309 declare double @llvm.cos.f64(double %Val)
6310 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6311 declare fp128 @llvm.cos.f128(fp128 %Val)
6312 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6313 </pre>
6319 <p>The argument and return value are floating point numbers of the same
6320 type.</p>
6323 <p>This function returns the cosine of the specified operand, returning the same
6325 in the same way.</p>
6327 </div>
6329 <!-- _______________________________________________________________________ -->
6332 </div>
6338 floating point or vector of floating point type. Not all targets support all
6339 types however.</p>
6341 <pre>
6342 declare float @llvm.pow.f32(float %Val, float %Power)
6343 declare double @llvm.pow.f64(double %Val, double %Power)
6344 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6345 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6346 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6347 </pre>
6351 specified (positive or negative) power.</p>
6354 <p>The second argument is a floating point power, and the first is a value to
6355 raise to that power.</p>
6358 <p>This function returns the first value raised to the second power, returning
6360 conditions in the same way.</p>
6362 </div>
6364 <!-- ======================================================================= -->
6367 </div>
6371 <p>LLVM provides intrinsics for a few important bit manipulation operations.
6372 These allow efficient code generation for some algorithms.</p>
6374 </div>
6376 <!-- _______________________________________________________________________ -->
6379 </div>
6384 <p>This is an overloaded intrinsic function. You can use bswap on any integer
6387 <pre>
6391 </pre>
6395 values with an even number of bytes (positive multiple of 16 bits). These
6396 are useful for performing operations on data that is not in the target's
6397 native byte order.</p>
6401 and low byte of the input i16 swapped. Similarly,
6407 more, respectively).</p>
6409 </div>
6411 <!-- _______________________________________________________________________ -->
6414 </div>
6419 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
6420 width. Not all targets support all bit widths however.</p>
6422 <pre>
6428 </pre>
6432 in a value.</p>
6435 <p>The only argument is the value to be counted. The argument may be of any
6436 integer type. The return type must match the argument type.</p>
6441 </div>
6443 <!-- _______________________________________________________________________ -->
6446 </div>
6452 integer bit width. Not all targets support all bit widths however.</p>
6454 <pre>
6460 </pre>
6464 leading zeros in a variable.</p>
6467 <p>The only argument is the value to be counted. The argument may be of any
6468 integer type. The return type must match the argument type.</p>
6472 zeros in a variable. If the src == 0 then the result is the size in bits of
6475 </div>
6477 <!-- _______________________________________________________________________ -->
6480 </div>
6486 integer bit width. Not all targets support all bit widths however.</p>
6488 <pre>
6494 </pre>
6498 trailing zeros.</p>
6501 <p>The only argument is the value to be counted. The argument may be of any
6502 integer type. The return type must match the argument type.</p>
6506 zeros in a variable. If the src == 0 then the result is the size in bits of
6509 </div>
6511 <!-- ======================================================================= -->
6514 </div>
6520 </div>
6522 <!-- _______________________________________________________________________ -->
6525 </div>
6531 on any integer bit width.</p>
6533 <pre>
6534 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6535 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6536 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6537 </pre>
6541 a signed addition of the two arguments, and indicate whether an overflow
6542 occurred during the signed summation.</p>
6545 <p>The arguments (%a and %b) and the first element of the result structure may
6546 be of integer types of any bit width, but they must have the same bit
6547 width. The second element of the result structure must be of
6549 undergo signed addition.</p>
6553 a signed addition of the two variables. They return a structure — the
6554 first element of which is the signed summation, and the second element of
6555 which is a bit specifying if the signed summation resulted in an
6556 overflow.</p>
6559 <pre>
6560 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6561 %sum = extractvalue {i32, i1} %res, 0
6562 %obit = extractvalue {i32, i1} %res, 1
6563 br i1 %obit, label %overflow, label %normal
6564 </pre>
6566 </div>
6568 <!-- _______________________________________________________________________ -->
6571 </div>
6577 on any integer bit width.</p>
6579 <pre>
6580 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6581 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6582 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6583 </pre>
6587 an unsigned addition of the two arguments, and indicate whether a carry
6588 occurred during the unsigned summation.</p>
6591 <p>The arguments (%a and %b) and the first element of the result structure may
6592 be of integer types of any bit width, but they must have the same bit
6593 width. The second element of the result structure must be of
6595 undergo unsigned addition.</p>
6599 an unsigned addition of the two arguments. They return a structure —
6600 the first element of which is the sum, and the second element of which is a
6601 bit specifying if the unsigned summation resulted in a carry.</p>
6604 <pre>
6605 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6606 %sum = extractvalue {i32, i1} %res, 0
6607 %obit = extractvalue {i32, i1} %res, 1
6608 br i1 %obit, label %carry, label %normal
6609 </pre>
6611 </div>
6613 <!-- _______________________________________________________________________ -->
6616 </div>
6622 on any integer bit width.</p>
6624 <pre>
6625 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6626 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6627 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6628 </pre>
6632 a signed subtraction of the two arguments, and indicate whether an overflow
6633 occurred during the signed subtraction.</p>
6636 <p>The arguments (%a and %b) and the first element of the result structure may
6637 be of integer types of any bit width, but they must have the same bit
6638 width. The second element of the result structure must be of
6640 undergo signed subtraction.</p>
6644 a signed subtraction of the two arguments. They return a structure —
6645 the first element of which is the subtraction, and the second element of
6646 which is a bit specifying if the signed subtraction resulted in an
6647 overflow.</p>
6650 <pre>
6651 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6652 %sum = extractvalue {i32, i1} %res, 0
6653 %obit = extractvalue {i32, i1} %res, 1
6654 br i1 %obit, label %overflow, label %normal
6655 </pre>
6657 </div>
6659 <!-- _______________________________________________________________________ -->
6662 </div>
6668 on any integer bit width.</p>
6670 <pre>
6671 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6672 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6673 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6674 </pre>
6678 an unsigned subtraction of the two arguments, and indicate whether an
6679 overflow occurred during the unsigned subtraction.</p>
6682 <p>The arguments (%a and %b) and the first element of the result structure may
6683 be of integer types of any bit width, but they must have the same bit
6684 width. The second element of the result structure must be of
6686 undergo unsigned subtraction.</p>
6690 an unsigned subtraction of the two arguments. They return a structure —
6691 the first element of which is the subtraction, and the second element of
6692 which is a bit specifying if the unsigned subtraction resulted in an
6693 overflow.</p>
6696 <pre>
6697 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6698 %sum = extractvalue {i32, i1} %res, 0
6699 %obit = extractvalue {i32, i1} %res, 1
6700 br i1 %obit, label %overflow, label %normal
6701 </pre>
6703 </div>
6705 <!-- _______________________________________________________________________ -->
6708 </div>
6714 on any integer bit width.</p>
6716 <pre>
6717 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6718 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6719 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6720 </pre>
6725 a signed multiplication of the two arguments, and indicate whether an
6726 overflow occurred during the signed multiplication.</p>
6729 <p>The arguments (%a and %b) and the first element of the result structure may
6730 be of integer types of any bit width, but they must have the same bit
6731 width. The second element of the result structure must be of
6733 undergo signed multiplication.</p>
6737 a signed multiplication of the two arguments. They return a structure —
6738 the first element of which is the multiplication, and the second element of
6739 which is a bit specifying if the signed multiplication resulted in an
6740 overflow.</p>
6743 <pre>
6744 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6745 %sum = extractvalue {i32, i1} %res, 0
6746 %obit = extractvalue {i32, i1} %res, 1
6747 br i1 %obit, label %overflow, label %normal
6748 </pre>
6750 </div>
6752 <!-- _______________________________________________________________________ -->
6755 </div>
6761 on any integer bit width.</p>
6763 <pre>
6764 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6765 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6766 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6767 </pre>
6771 a unsigned multiplication of the two arguments, and indicate whether an
6772 overflow occurred during the unsigned multiplication.</p>
6775 <p>The arguments (%a and %b) and the first element of the result structure may
6776 be of integer types of any bit width, but they must have the same bit
6777 width. The second element of the result structure must be of
6779 undergo unsigned multiplication.</p>
6783 an unsigned multiplication of the two arguments. They return a structure
6784 — the first element of which is the multiplication, and the second
6785 element of which is a bit specifying if the unsigned multiplication resulted
6786 in an overflow.</p>
6789 <pre>
6790 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6791 %sum = extractvalue {i32, i1} %res, 0
6792 %obit = extractvalue {i32, i1} %res, 1
6793 br i1 %obit, label %overflow, label %normal
6794 </pre>
6796 </div>
6798 <!-- ======================================================================= -->
6801 </div>
6805 <p>Half precision floating point is a storage-only format. This means that it is
6806 a dense encoding (in memory) but does not support computation in the
6807 format.</p>
6809 <p>This means that code must first load the half-precision floating point
6810 value as an i16, then convert it to float with <a
6812 Computation can then be performed on the float value (including extending to
6813 double etc). To store the value back to memory, it is first converted to
6814 float if needed, then converted to i16 with
6816 storing as an i16 value.</p>
6817 </div>
6819 <!-- _______________________________________________________________________ -->
6822 </div>
6827 <pre>
6828 declare i16 @llvm.convert.to.fp16(f32 %a)
6829 </pre>
6833 a conversion from single precision floating point format to half precision
6834 floating point format.</p>
6837 <p>The intrinsic function contains single argument - the value to be
6838 converted.</p>
6842 a conversion from single precision floating point format to half precision
6844 contains the converted number.</p>
6847 <pre>
6848 %res = call i16 @llvm.convert.to.fp16(f32 %a)
6849 store i16 %res, i16* @x, align 2
6850 </pre>
6852 </div>
6854 <!-- _______________________________________________________________________ -->
6857 </div>
6862 <pre>
6863 declare f32 @llvm.convert.from.fp16(i16 %a)
6864 </pre>
6868 a conversion from half precision floating point format to single precision
6869 floating point format.</p>
6872 <p>The intrinsic function contains single argument - the value to be
6873 converted.</p>
6877 conversion from half single precision floating point format to single
6878 precision floating point format. The input half-float value is represented by
6882 <pre>
6883 %a = load i16* @x, align 2
6884 %res = call f32 @llvm.convert.from.fp16(i16 %a)
6885 </pre>
6887 </div>
6889 <!-- ======================================================================= -->
6892 </div>
6897 prefix), are described in
6901 </div>
6903 <!-- ======================================================================= -->
6906 </div>
6910 <p>The LLVM exception handling intrinsics (which all start with
6915 </div>
6917 <!-- ======================================================================= -->
6920 </div>
6924 <p>This intrinsic makes it possible to excise one parameter, marked with
6926 function pointer lacking the nest parameter - the caller does not need to
6927 provide a value for it. Instead, the value to use is stored in advance in a
6928 "trampoline", a block of memory usually allocated on the stack, which also
6929 contains code to splice the nest value into the argument list. This is used
6930 to implement the GCC nested function address extension.</p>
6932 <p>For example, if the function is
6935 follows:</p>
6938 <pre>
6941 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6942 %fp = bitcast i8* %p to i32 (i32, i32)*
6943 </pre>
6944 </div>
6949 </div>
6951 <!-- _______________________________________________________________________ -->
6954 </div>
6959 <pre>
6961 </pre>
6965 function pointer suitable for executing it.</p>
6970 sufficiently aligned block of memory; this memory is written to by the
6971 intrinsic. Note that the size and the alignment are target-specific - LLVM
6972 currently provides no portable way of determining them, so a front-end that
6973 generates this intrinsic needs to have some target-specific knowledge.
6979 dependent code, turning it into a function. A pointer to this function is
6981 function pointer type</a> before being called. The new function's signature
6984 is allowed, and it must be of pointer type. Calling the new function is
6989 returned function pointer is undefined.</p>
6991 </div>
6993 <!-- ======================================================================= -->
6996 </div>
7001 hardware constructs for atomic operations and memory synchronization. This
7002 provides an interface to the hardware, not an interface to the programmer. It
7003 is aimed at a low enough level to allow any programming models or APIs
7004 (Application Programming Interfaces) which need atomic behaviors to map
7005 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
7006 hardware provides a "universal IR" for source languages, it also provides a
7007 starting point for developing a "universal" atomic operation and
7008 synchronization IR.</p>
7011 software transaction memory systems, atomic primitives, and intrinsic
7012 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
7013 application libraries. The hardware interface provided by LLVM should allow
7014 a clean implementation of all of these APIs and parallel programming models.
7015 No one model or paradigm should be selected above others unless the hardware
7016 itself ubiquitously does so.</p>
7018 </div>
7020 <!-- _______________________________________________________________________ -->
7023 </div>
7026 <pre>
7027 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device> )
7028 </pre>
7032 specific pairs of memory access types.</p>
7036 The first four arguments enables a specific barrier as listed below. The
7037 fifth argument specifies that the barrier applies to io or device or uncached
7038 memory.</p>
7040 <ul>
7046 </ul>
7049 <p>This intrinsic causes the system to enforce some ordering constraints upon
7050 the loads and stores of the program. This barrier does not
7053 and store operations (f.ex. load-load, or store-load), all of the first
7054 operations preceding the barrier will complete before any of the second
7055 operations succeeding the barrier begin. Specifically the semantics for each
7056 pairing is as follows:</p>
7058 <ul>
7060 after the barrier begins.</li>
7062 store after the barrier begins.</li>
7064 store after the barrier begins.</li>
7066 load after the barrier begins.</li>
7067 </ul>
7070 is enabled in a single memory barrier intrinsic.</p>
7072 <p>Backends may implement stronger barriers than those requested when they do
7073 not support as fine grained a barrier as requested. Some architectures do
7074 not need all types of barriers and on such architectures, these become
7075 noops.</p>
7078 <pre>
7079 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7080 %ptr = bitcast i8* %mallocP to i32*
7081 store i32 4, %ptr
7084 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
7087 </pre>
7089 </div>
7091 <!-- _______________________________________________________________________ -->
7094 </div>
7100 any integer bit width and for different address spaces. Not all targets
7101 support all bit widths however.</p>
7103 <pre>
7105 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
7106 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
7107 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
7108 </pre>
7111 <p>This loads a value in memory and compares it to a given value. If they are
7112 equal, it stores a new value into the memory.</p>
7118 this integer type. While any bit width integer may be used, targets may only
7119 lower representations they support in hardware.</p>
7122 <p>This entire intrinsic must be executed atomically. It first loads the value
7125 memory. The loaded value is yielded in all cases. This provides the
7126 equivalent of an atomic compare-and-swap operation within the SSA
7127 framework.</p>
7130 <pre>
7131 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7132 %ptr = bitcast i8* %mallocP to i32*
7133 store i32 4, %ptr
7136 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
7142 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
7147 </pre>
7149 </div>
7151 <!-- _______________________________________________________________________ -->
7154 </div>
7159 integer bit width. Not all targets support all bit widths however.</p>
7161 <pre>
7166 </pre>
7177 integer type. The targets may only lower integer representations they
7178 support.</p>
7183 equivalent of an atomic swap operation within the SSA framework.</p>
7186 <pre>
7187 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7188 %ptr = bitcast i8* %mallocP to i32*
7189 store i32 4, %ptr
7192 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
7198 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
7203 </pre>
7205 </div>
7207 <!-- _______________________________________________________________________ -->
7211 </div>
7217 any integer bit width. Not all targets support all bit widths however.</p>
7219 <pre>
7224 </pre>
7231 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7232 and the second an integer value. The result is also an integer value. These
7233 integer types can have any bit width, but they must all have the same bit
7234 width. The targets may only lower integer representations they support.</p>
7237 <p>This intrinsic does a series of operations atomically. It first loads the
7242 <pre>
7243 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7244 %ptr = bitcast i8* %mallocP to i32*
7245 store i32 4, %ptr
7246 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
7248 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
7250 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
7253 </pre>
7255 </div>
7257 <!-- _______________________________________________________________________ -->
7261 </div>
7267 any integer bit width and for different address spaces. Not all targets
7268 support all bit widths however.</p>
7270 <pre>
7275 </pre>
7282 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7283 and the second an integer value. The result is also an integer value. These
7284 integer types can have any bit width, but they must all have the same bit
7285 width. The targets may only lower integer representations they support.</p>
7288 <p>This intrinsic does a series of operations atomically. It first loads the
7294 <pre>
7295 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7296 %ptr = bitcast i8* %mallocP to i32*
7297 store i32 8, %ptr
7298 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
7300 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
7302 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
7305 </pre>
7307 </div>
7309 <!-- _______________________________________________________________________ -->
7315 </div>
7320 <p>These are overloaded intrinsics. You can
7323 bit width and for different address spaces. Not all targets support all bit
7324 widths however.</p>
7326 <pre>
7331 </pre>
7333 <pre>
7338 </pre>
7340 <pre>
7345 </pre>
7347 <pre>
7352 </pre>
7360 <p>These intrinsics take two arguments, the first a pointer to an integer value
7361 and the second an integer value. The result is also an integer value. These
7362 integer types can have any bit width, but they must all have the same bit
7363 width. The targets may only lower integer representations they support.</p>
7366 <p>These intrinsics does a series of operations atomically. They first load the
7372 <pre>
7373 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7374 %ptr = bitcast i8* %mallocP to i32*
7375 store i32 0x0F0F, %ptr
7376 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
7378 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
7380 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
7382 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
7385 </pre>
7387 </div>
7389 <!-- _______________________________________________________________________ -->
7395 </div>
7403 address spaces. Not all targets support all bit widths however.</p>
7405 <pre>
7410 </pre>
7412 <pre>
7417 </pre>
7419 <pre>
7424 </pre>
7426 <pre>
7431 </pre>
7434 <p>These intrinsics takes the signed or unsigned minimum or maximum of
7439 <p>These intrinsics take two arguments, the first a pointer to an integer value
7440 and the second an integer value. The result is also an integer value. These
7441 integer types can have any bit width, but they must all have the same bit
7442 width. The targets may only lower integer representations they support.</p>
7445 <p>These intrinsics does a series of operations atomically. They first load the
7451 <pre>
7452 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7453 %ptr = bitcast i8* %mallocP to i32*
7454 store i32 7, %ptr
7455 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
7457 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
7459 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
7461 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
7464 </pre>
7466 </div>
7469 <!-- ======================================================================= -->
7472 </div>
7476 <p>This class of intrinsics exists to information about the lifetime of memory
7477 objects and ranges where variables are immutable.</p>
7479 </div>
7481 <!-- _______________________________________________________________________ -->
7484 </div>
7489 <pre>
7491 </pre>
7495 object's lifetime.</p>
7498 <p>The first argument is a constant integer representing the size of the
7499 object, or -1 if it is variable sized. The second argument is a pointer to
7500 the object.</p>
7503 <p>This intrinsic indicates that before this point in the code, the value of the
7505 never be used and has an undefined value. A load from the pointer that
7506 precedes this intrinsic can be replaced with
7509 </div>
7511 <!-- _______________________________________________________________________ -->
7514 </div>
7519 <pre>
7521 </pre>
7525 object's lifetime.</p>
7528 <p>The first argument is a constant integer representing the size of the
7529 object, or -1 if it is variable sized. The second argument is a pointer to
7530 the object.</p>
7533 <p>This intrinsic indicates that after this point in the code, the value of the
7535 never be used and has an undefined value. Any stores into the memory object
7536 following this intrinsic may be removed as dead.
7538 </div>
7540 <!-- _______________________________________________________________________ -->
7543 </div>
7548 <pre>
7550 </pre>
7554 a memory object will not change.</p>
7557 <p>The first argument is a constant integer representing the size of the
7558 object, or -1 if it is variable sized. The second argument is a pointer to
7559 the object.</p>
7563 the return value, the referenced memory location is constant and
7564 unchanging.</p>
7566 </div>
7568 <!-- _______________________________________________________________________ -->
7571 </div>
7576 <pre>
7577 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
7578 </pre>
7582 a memory object are mutable.</p>
7586 The second argument is a constant integer representing the size of the
7587 object, or -1 if it is variable sized and the third argument is a pointer
7588 to the object.</p>
7593 </div>
7595 <!-- ======================================================================= -->
7598 </div>
7602 <p>This class of intrinsics is designed to be generic and has no specific
7603 purpose.</p>
7605 </div>
7607 <!-- _______________________________________________________________________ -->
7610 </div>
7615 <pre>
7616 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
7617 </pre>
7623 <p>The first argument is a pointer to a value, the second is a pointer to a
7624 global string, the third is a pointer to a global string which is the source
7625 file name, and the last argument is the line number.</p>
7628 <p>This intrinsic allows annotation of local variables with arbitrary strings.
7629 This can be useful for special purpose optimizations that want to look for
7630 these annotations. These have no other defined use, they are ignored by code
7631 generation and optimization.</p>
7633 </div>
7635 <!-- _______________________________________________________________________ -->
7638 </div>
7644 any integer bit width.</p>
7646 <pre>
7647 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
7648 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
7649 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
7650 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
7651 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
7652 </pre>
7658 <p>The first argument is an integer value (result of some expression), the
7659 second is a pointer to a global string, the third is a pointer to a global
7660 string which is the source file name, and the last argument is the line
7661 number. It returns the value of the first argument.</p>
7664 <p>This intrinsic allows annotations to be put on arbitrary expressions with
7665 arbitrary strings. This can be useful for special purpose optimizations that
7666 want to look for these annotations. These have no other defined use, they
7667 are ignored by code generation and optimization.</p>
7669 </div>
7671 <!-- _______________________________________________________________________ -->
7674 </div>
7679 <pre>
7680 declare void @llvm.trap()
7681 </pre>
7690 <p>This intrinsics is lowered to the target dependent trap instruction. If the
7691 target does not have a trap instruction, this intrinsic will be lowered to
7694 </div>
7696 <!-- _______________________________________________________________________ -->
7699 </div>
7704 <pre>
7706 </pre>
7711 ensure that it is placed on the stack before local variables.</p>
7715 arguments. The first argument is the value loaded from the stack
7717 that has enough space to hold the value of the guard.</p>
7720 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
7722 stack. This is to ensure that if a local variable on the stack is
7723 overwritten, it will destroy the value of the guard. When the function exits,
7724 the guard on the stack is checked against the original guard. If they're
7726 function.</p>
7728 </div>
7730 <!-- _______________________________________________________________________ -->
7733 </div>
7738 <pre>
7741 </pre>
7745 to the optimizers to discover at compile time either a) when an
7746 operation like memcpy will either overflow a buffer that corresponds to
7747 an object, or b) to determine that a runtime check for overflow isn't
7748 necessary. An object in this context means an allocation of a
7749 specific class, structure, array, or other object.</p>
7762 at compile time.</p>
7764 </div>
7766 <!-- *********************************************************************** -->
7767 <hr>
7768 <address>
7776 Last modified: $Date$
7777 </address>
7779 </body>
7780 </html>