| blob | 45f6f38f598fc9c34bc7b9ce70feb8905d7b2f40 |
3 <html>
4 <head>
11 </head>
13 <body>
16 <ol>
21 <ol>
24 <ol>
39 </ol>
40 </li>
52 </ol>
53 </li>
55 <ol>
58 <ol>
64 </ol>
65 </li>
67 <ol>
75 </ol>
76 </li>
78 </ol>
79 </li>
81 <ol>
89 </ol>
90 </li>
92 <ol>
94 </ol>
95 </li>
97 <ol>
100 Global Variable</a></li>
102 Global Variable</a></li>
104 Global Variable</a></li>
105 </ol>
106 </li>
108 <ol>
110 <ol>
118 </ol>
119 </li>
121 <ol>
134 </ol>
135 </li>
137 <ol>
144 </ol>
145 </li>
147 <ol>
151 </ol>
152 </li>
154 <ol>
157 </ol>
158 </li>
160 <ol>
165 </ol>
166 </li>
168 <ol>
181 </ol>
182 </li>
184 <ol>
191 </ol>
192 </li>
193 </ol>
194 </li>
196 <ol>
198 <ol>
202 </ol>
203 </li>
205 <ol>
209 </ol>
210 </li>
212 <ol>
220 </ol>
221 </li>
223 <ol>
232 </ol>
233 </li>
235 <ol>
240 </ol>
241 </li>
243 <ol>
250 </ol>
251 </li>
255 <ol>
257 </ol>
258 </li>
260 <ol>
274 </ol>
275 </li>
277 <ol>
282 </ol>
283 </li>
285 <ol>
296 </ol>
297 </li>
298 </ol>
299 </li>
300 </ol>
305 </div>
307 <!-- *********************************************************************** -->
309 <!-- *********************************************************************** -->
313 <p>This document is a reference manual for the LLVM assembly language. LLVM is
314 a Static Single Assignment (SSA) based representation that provides type
315 safety, low-level operations, flexibility, and the capability of representing
316 'all' high-level languages cleanly. It is the common code representation
317 used throughout all phases of the LLVM compilation strategy.</p>
319 </div>
321 <!-- *********************************************************************** -->
323 <!-- *********************************************************************** -->
327 <p>The LLVM code representation is designed to be used in three different forms:
328 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
329 for fast loading by a Just-In-Time compiler), and as a human readable
330 assembly language representation. This allows LLVM to provide a powerful
331 intermediate representation for efficient compiler transformations and
332 analysis, while providing a natural means to debug and visualize the
333 transformations. The three different forms of LLVM are all equivalent. This
334 document describes the human readable representation and notation.</p>
336 <p>The LLVM representation aims to be light-weight and low-level while being
337 expressive, typed, and extensible at the same time. It aims to be a
338 "universal IR" of sorts, by being at a low enough level that high-level ideas
339 may be cleanly mapped to it (similar to how microprocessors are "universal
340 IR's", allowing many source languages to be mapped to them). By providing
341 type information, LLVM can be used as the target of optimizations: for
342 example, through pointer analysis, it can be proven that a C automatic
343 variable is never accessed outside of the current function, allowing it to
344 be promoted to a simple SSA value instead of a memory location.</p>
346 </div>
348 <!-- _______________________________________________________________________ -->
353 <p>It is important to note that this document describes 'well formed' LLVM
354 assembly language. There is a difference between what the parser accepts and
355 what is considered 'well formed'. For example, the following instruction is
356 syntactically okay, but not well formed:</p>
359 <pre>
361 </pre>
362 </div>
365 LLVM infrastructure provides a verification pass that may be used to verify
366 that an LLVM module is well formed. This pass is automatically run by the
367 parser after parsing input assembly and by the optimizer before it outputs
368 bitcode. The violations pointed out by the verifier pass indicate bugs in
369 transformation passes or input to the parser.</p>
371 </div>
373 <!-- Describe the typesetting conventions here. -->
375 <!-- *********************************************************************** -->
377 <!-- *********************************************************************** -->
381 <p>LLVM identifiers come in two basic types: global and local. Global
383 character. Local identifiers (register names, types) begin with
385 for identifiers, for different purposes:</p>
387 <ol>
388 <li>Named values are represented as a string of characters with their prefix.
392 other characters in their names can be surrounded with quotes. Special
394 ASCII code for the character in hexadecimal. In this way, any character
395 can be used in a name value, even quotes themselves.</li>
397 <li>Unnamed values are represented as an unsigned numeric value with their
402 </ol>
404 <p>LLVM requires that values start with a prefix for two reasons: Compilers
405 don't need to worry about name clashes with reserved words, and the set of
406 reserved words may be expanded in the future without penalty. Additionally,
407 unnamed identifiers allow a compiler to quickly come up with a temporary
408 variable without having to avoid symbol table conflicts.</p>
410 <p>Reserved words in LLVM are very similar to reserved words in other
411 languages. There are keywords for different opcodes
417 reserved words cannot conflict with variable names, because none of them
420 <p>Here is an example of LLVM code to multiply the integer variable
426 <pre>
428 </pre>
429 </div>
434 <pre>
436 </pre>
437 </div>
442 <pre>
446 </pre>
447 </div>
450 lexical features of LLVM:</p>
452 <ol>
454 line.</li>
456 <li>Unnamed temporaries are created when the result of a computation is not
457 assigned to a named value.</li>
460 </ol>
462 <p>It also shows a convention that we follow in this document. When
463 demonstrating instructions, we will follow an instruction with a comment that
464 defines the type and name of value produced. Comments are shown in italic
465 text.</p>
467 </div>
469 <!-- *********************************************************************** -->
471 <!-- *********************************************************************** -->
473 <!-- ======================================================================= -->
475 </div>
480 of the input programs. Each module consists of functions, global variables,
481 and symbol table entries. Modules may be combined together with the LLVM
482 linker, which merges function (and global variable) definitions, resolves
483 forward declarations, and merges symbol table entries. Here is an example of
487 <pre>
489 <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>
497 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
502 </pre>
503 </div>
510 <p>In general, a module is made up of a list of global values, where both
511 functions and global variables are global values. Global values are
512 represented by a pointer to a memory location (in this case, a pointer to an
513 array of char, and a pointer to a function), and have one of the
516 </div>
518 <!-- ======================================================================= -->
521 </div>
525 <p>All Global Variables and Functions have one of the following types of
526 linkage:</p>
528 <dl>
530 <dd>Global values with private linkage are only directly accessible by objects
531 in the current module. In particular, linking code into a module with an
532 private global value may cause the private to be renamed as necessary to
533 avoid collisions. Because the symbol is private to the module, all
534 references can be updated. This doesn't show up in any symbol table in the
535 object file.</dd>
538 <dd>Similar to private, but the symbol is passed through the assembler and
539 removed by the linker after evaluation. Note that (unlike private
540 symbols) linker_private symbols are subject to coalescing by the linker:
541 weak symbols get merged and redefinitions are rejected. However, unlike
542 normal strong symbols, they are removed by the linker from the final
543 linked image (executable or dynamic library).</dd>
546 <dd>Similar to private, but the value shows as a local symbol
552 into the object file corresponding to the LLVM module. They exist to
553 allow inlining and other optimizations to take place given knowledge of
554 the definition of the global, which is known to be somewhere outside the
557 This linkage type is only allowed on definitions, not declarations.</dd>
561 the same name when linkage occurs. This is typically used to implement
562 inline functions, templates, or other code which must be generated in each
564 allowed to be discarded.</dd>
574 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
575 global scope.
576 Symbols with "<tt>common</tt>" linkage are merged in the same way as
579 must have a zero initializer, and may not be marked '<a
581 have common linkage.</dd>
586 pointer to array type. When two global variables with appending linkage
587 are linked together, the two global arrays are appended together. This is
588 the LLVM, typesafe, equivalent of having the system linker append together
592 <dd>The semantics of this linkage follow the ELF object file model: the symbol
593 is weak until linked, if not linked, the symbol becomes null instead of
594 being an undefined reference.</dd>
598 <dd>Some languages allow differing globals to be merged, such as two functions
600 that only equivalent globals are ever merged (the "one definition rule" -
603 be merged with equivalent globals. These linkage types are otherwise the
607 <dd>If none of the above identifiers are used, the global is externally
608 visible, meaning that it participates in linkage and can be used to
609 resolve external symbol references.</dd>
610 </dl>
612 <p>The next two types of linkage are targeted for Microsoft Windows platform
613 only. They are designed to support importing (exporting) symbols from (to)
614 DLLs (Dynamic Link Libraries).</p>
616 <dl>
619 or variable via a global pointer to a pointer that is set up by the DLL
620 exporting the symbol. On Microsoft Windows targets, the pointer name is
622 name.</dd>
626 pointer to a pointer in a DLL, so that it can be referenced with the
629 variable name.</dd>
630 </dl>
633 another module defined a "<tt>.LC0</tt>" variable and was linked with this
634 one, one of the two would be renamed, preventing a collision. Since
636 declarations), they are accessible outside of the current module.</p>
645 </div>
647 <!-- ======================================================================= -->
650 </div>
656 convention specified for the call. The calling convention of any pair of
657 dynamic caller/callee must match, or the behavior of the program is
658 undefined. The following calling conventions are supported by LLVM, and more
659 may be added in the future:</p>
661 <dl>
663 <dd>This calling convention (the default if no other calling convention is
664 specified) matches the target C calling conventions. This calling
665 convention supports varargs function calls and tolerates some mismatch in
666 the declared prototype and implemented declaration of the function (as
667 does normal C).</dd>
670 <dd>This calling convention attempts to make calls as fast as possible
671 (e.g. by passing things in registers). This calling convention allows the
672 target to use whatever tricks it wants to produce fast code for the
673 target, without having to conform to an externally specified ABI
674 (Application Binary Interface). Implementations of this convention should
676 optimization</a> to be supported. This calling convention does not
677 support varargs and requires the prototype of all callees to exactly match
678 the prototype of the function definition.</dd>
681 <dd>This calling convention attempts to make code in the caller as efficient
682 as possible under the assumption that the call is not commonly executed.
683 As such, these calls often preserve all registers so that the call does
684 not break any live ranges in the caller side. This calling convention
685 does not support varargs and requires the prototype of all callees to
686 exactly match the prototype of the function definition.</dd>
689 <dd>Any calling convention may be specified by number, allowing
690 target-specific calling conventions to be used. Target specific calling
692 </dl>
694 <p>More calling conventions can be added/defined on an as-needed basis, to
695 support Pascal conventions or any other well-known target-independent
696 convention.</p>
698 </div>
700 <!-- ======================================================================= -->
703 </div>
707 <p>All Global Variables and Functions have one of the following visibility
708 styles:</p>
710 <dl>
712 <dd>On targets that use the ELF object file format, default visibility means
713 that the declaration is visible to other modules and, in shared libraries,
714 means that the declared entity may be overridden. On Darwin, default
715 visibility means that the declaration is visible to other modules. Default
719 <dd>Two declarations of an object with hidden visibility refer to the same
720 object if they are in the same shared object. Usually, hidden visibility
721 indicates that the symbol will not be placed into the dynamic symbol
722 table, so no other module (executable or shared library) can reference it
723 directly.</dd>
726 <dd>On ELF, protected visibility indicates that the symbol will be placed in
727 the dynamic symbol table, but that references within the defining module
728 will bind to the local symbol. That is, the symbol cannot be overridden by
729 another module.</dd>
730 </dl>
732 </div>
734 <!-- ======================================================================= -->
737 </div>
741 <p>LLVM IR allows you to specify name aliases for certain types. This can make
742 it easier to read the IR and make the IR more condensed (particularly when
743 recursive types are involved). An example of a name specification is:</p>
746 <pre>
747 %mytype = type { %mytype*, i32 }
748 </pre>
749 </div>
755 <p>Note that type names are aliases for the structural type that they indicate,
756 and that you can therefore specify multiple names for the same type. This
757 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
758 uses structural typing, the name is not part of the type. When printing out
760 particular shape. This means that if you have code where two different
761 source types end up having the same LLVM type, that the dumper will sometimes
762 print the "wrong" or unexpected type. This is an important design point and
763 isn't going to change.</p>
765 </div>
767 <!-- ======================================================================= -->
770 </div>
774 <p>Global variables define regions of memory allocated at compilation time
775 instead of run-time. Global variables may optionally be initialized, may
776 have an explicit section to be placed in, and may have an optional explicit
777 alignment specified. A variable may be defined as "thread_local", which
778 means that it will not be shared by threads (each thread will have a
779 separated copy of the variable). A variable may be defined as a global
780 "constant," which indicates that the contents of the variable
782 global data to be placed in the read-only section of an executable, etc).
783 Note that variables that need runtime initialization cannot be marked
787 constant, even if the final definition of the global is not. This capability
788 can be used to enable slightly better optimization of the program, but
789 requires the language definition to guarantee that optimizations based on the
790 'constantness' are valid for the translation units that do not include the
791 definition.</p>
793 <p>As SSA values, global variables define pointer values that are in scope
794 (i.e. they dominate) all basic blocks in the program. Global variables
795 always define a pointer to their "content" type because they describe a
796 region of memory, and all memory objects in LLVM are accessed through
797 pointers.</p>
799 <p>A global variable may be declared to reside in a target-specific numbered
800 address space. For targets that support them, address spaces may affect how
801 optimizations are performed and/or what target instructions are used to
802 access the variable. The default address space is zero. The address space
803 qualifier must precede any other attributes.</p>
805 <p>LLVM allows an explicit section to be specified for globals. If the target
806 supports it, it will emit globals to the section specified.</p>
808 <p>An explicit alignment may be specified for a global. If not present, or if
809 the alignment is set to zero, the alignment of the global is set by the
810 target to whatever it feels convenient. If an explicit alignment is
811 specified, the global is forced to have at least that much alignment. All
814 <p>For example, the following defines a global in a numbered address space with
815 an initializer, section, and alignment:</p>
818 <pre>
820 </pre>
821 </div>
823 </div>
826 <!-- ======================================================================= -->
829 </div>
838 name, a (possibly empty) argument list (each with optional
842 curly brace, a list of basic blocks, and a closing curly brace.</p>
849 name, a possibly empty list of arguments, an optional alignment, and an
852 <p>A function definition contains a list of basic blocks, forming the CFG
853 (Control Flow Graph) for the function. Each basic block may optionally start
854 with a label (giving the basic block a symbol table entry), contains a list
856 instruction (such as a branch or function return).</p>
858 <p>The first basic block in a function is special in two ways: it is immediately
859 executed on entrance to the function, and it is not allowed to have
860 predecessor basic blocks (i.e. there can not be any branches to the entry
861 block of a function). Because the block can have no predecessors, it also
864 <p>LLVM allows an explicit section to be specified for functions. If the target
865 supports it, it will emit functions to the section specified.</p>
867 <p>An explicit alignment may be specified for a function. If not present, or if
868 the alignment is set to zero, the alignment of the function is set by the
869 target to whatever it feels convenient. If an explicit alignment is
870 specified, the function is forced to have at least that much alignment. All
875 <pre>
881 </pre>
882 </div>
884 </div>
886 <!-- ======================================================================= -->
889 </div>
894 function, global variable, another alias or bitcast of global value). Aliases
900 <pre>
902 </pre>
903 </div>
905 </div>
907 <!-- ======================================================================= -->
912 <p>The return type and each parameter of a function type may have a set of
914 used to communicate additional information about the result or parameters of
915 a function. Parameter attributes are considered to be part of the function,
916 not of the function type, so functions with different parameter attributes
917 can have the same function type.</p>
919 <p>Parameter attributes are simple keywords that follow the type specified. If
920 multiple parameter attributes are needed, they are space separated. For
921 example:</p>
924 <pre>
925 declare i32 @printf(i8* noalias nocapture, ...)
926 declare i32 @atoi(i8 zeroext)
927 declare signext i8 @returns_signed_char()
928 </pre>
929 </div>
936 <dl>
938 <dd>This indicates to the code generator that the parameter or return value
939 should be zero-extended to a 32-bit value by the caller (for a parameter)
940 or the callee (for a return value).</dd>
943 <dd>This indicates to the code generator that the parameter or return value
944 should be sign-extended to a 32-bit value by the caller (for a parameter)
945 or the callee (for a return value).</dd>
948 <dd>This indicates that this parameter or return value should be treated in a
949 special target-dependent fashion during while emitting code for a function
950 call or return (usually, by putting it in a register as opposed to memory,
951 though some targets use it to distinguish between two different kinds of
952 registers). Use of this attribute is target-specific.</dd>
955 <dd>This indicates that the pointer parameter should really be passed by value
956 to the function. The attribute implies that a hidden copy of the pointee
957 is made between the caller and the callee, so the callee is unable to
958 modify the value in the callee. This attribute is only valid on LLVM
959 pointer arguments. It is generally used to pass structs and arrays by
960 value, but is also valid on pointers to scalars. The copy is considered
961 to belong to the caller not the callee (for example,
964 values. The byval attribute also supports specifying an alignment with
965 the align attribute. This has a target-specific effect on the code
966 generator that usually indicates a desired alignment for the synthesized
967 stack slot.</dd>
970 <dd>This indicates that the pointer parameter specifies the address of a
971 structure that is the return value of the function in the source program.
972 This pointer must be guaranteed by the caller to be valid: loads and
973 stores to the structure may be assumed by the callee to not to trap. This
974 may only be applied to the first parameter. This is not a valid attribute
975 for return values. </dd>
978 <dd>This indicates that the pointer does not alias any global or any other
979 parameter. The caller is responsible for ensuring that this is the
981 that the pointer does not alias any other pointers visible to the
982 caller. For further details, please see the discussion of the NoAlias
983 response in
988 <dd>This indicates that the callee does not make any copies of the pointer
989 that outlive the callee itself. This is not a valid attribute for return
990 values.</dd>
993 <dd>This indicates that the pointer parameter can be excised using the
995 attribute for return values.</dd>
996 </dl>
998 </div>
1000 <!-- ======================================================================= -->
1003 </div>
1007 <p>Each function may specify a garbage collector name, which is simply a
1008 string:</p>
1011 <pre>
1012 define void @f() gc "name" { ... }
1013 </pre>
1014 </div>
1017 collector which will cause the compiler to alter its output in order to
1018 support the named garbage collection algorithm.</p>
1020 </div>
1022 <!-- ======================================================================= -->
1025 </div>
1029 <p>Function attributes are set to communicate additional information about a
1030 function. Function attributes are considered to be part of the function, not
1031 of the function type, so functions with different parameter attributes can
1032 have the same function type.</p>
1034 <p>Function attributes are simple keywords that follow the type specified. If
1035 multiple attributes are needed, they are space separated. For example:</p>
1038 <pre>
1039 define void @f() noinline { ... }
1040 define void @f() alwaysinline { ... }
1041 define void @f() alwaysinline optsize { ... }
1042 define void @f() optsize { ... }
1043 </pre>
1044 </div>
1046 <dl>
1048 <dd>This attribute indicates that the inliner should attempt to inline this
1049 function into callers whenever possible, ignoring any active inlining size
1050 threshold for this caller.</dd>
1053 <dd>This attribute indicates that the source code contained a hint that inlining
1054 this function is desirable (such as the "inline" keyword in C/C++). It
1055 is just a hint; it imposes no requirements on the inliner.</dd>
1058 <dd>This attribute indicates that the inliner should never inline this
1059 function in any situation. This attribute may not be used together with
1063 <dd>This attribute suggests that optimization passes and code generator passes
1064 make choices that keep the code size of this function low, and otherwise
1065 do optimizations specifically to reduce code size.</dd>
1068 <dd>This function attribute indicates that the function never returns
1069 normally. This produces undefined behavior at runtime if the function
1070 ever does dynamically return.</dd>
1073 <dd>This function attribute indicates that the function never returns with an
1074 unwind or exceptional control flow. If the function does unwind, its
1075 runtime behavior is undefined.</dd>
1078 <dd>This attribute indicates that the function computes its result (or decides
1079 to unwind an exception) based strictly on its arguments, without
1080 dereferencing any pointer arguments or otherwise accessing any mutable
1081 state (e.g. memory, control registers, etc) visible to caller functions.
1082 It does not write through any pointer arguments
1084 changes any state visible to callers. This means that it cannot unwind
1089 <dd>This attribute indicates that the function does not write through any
1091 arguments) or otherwise modify any state (e.g. memory, control registers,
1092 etc) visible to caller functions. It may dereference pointer arguments
1093 and read state that may be set in the caller. A readonly function always
1094 returns the same value (or unwinds an exception identically) when called
1095 with the same set of arguments and global state. It cannot unwind an
1100 <dd>This attribute indicates that the function should emit a stack smashing
1102 the stack before the local variables that's checked upon return from the
1103 function to see if it has been overwritten. A heuristic is used to
1104 determine if a function needs stack protectors or not.<br>
1105 <br>
1112 stack smashing protector. This overrides
1114 <br>
1121 <dd>This attribute indicates that the code generator should not use a red
1122 zone, even if the target-specific ABI normally permits it.</dd>
1128 <dd>This attribute disables prologue / epilogue emission for the function.
1129 This can have very system-specific consequences.</dd>
1130 </dl>
1132 </div>
1134 <!-- ======================================================================= -->
1137 </div>
1142 the GCC "file scope inline asm" blocks. These blocks are internally
1143 concatenated by LLVM and treated as a single unit, but may be separated in
1147 <pre>
1148 module asm "inline asm code goes here"
1149 module asm "more can go here"
1150 </pre>
1151 </div>
1153 <p>The strings can contain any character by escaping non-printable characters.
1155 for the number.</p>
1157 <p>The inline asm code is simply printed to the machine code .s file when
1158 assembly code is generated.</p>
1160 </div>
1162 <!-- ======================================================================= -->
1165 </div>
1169 <p>A module may specify a target specific data layout string that specifies how
1170 data is to be laid out in memory. The syntax for the data layout is
1171 simply:</p>
1174 <pre>
1175 target datalayout = "<i>layout specification</i>"
1176 </pre>
1177 </div>
1180 separated by the minus sign character ('-'). Each specification starts with
1181 a letter and may include other information after the letter to define some
1182 aspect of the data layout. The specifications accepted are as follows:</p>
1184 <dl>
1186 <dd>Specifies that the target lays out data in big-endian form. That is, the
1187 bits with the most significance have the lowest address location.</dd>
1190 <dd>Specifies that the target lays out data in little-endian form. That is,
1191 the bits with the least significance have the lowest address
1192 location.</dd>
1201 <dd>This specifies the alignment for an integer type of a given bit
1205 <dd>This specifies the alignment for a vector type of a given bit
1209 <dd>This specifies the alignment for a floating point type of a given bit
1211 (double).</dd>
1214 <dd>This specifies the alignment for an aggregate type of a given bit
1218 <dd>This specifies the alignment for a stack object of a given bit
1222 <dd>This specifies a set of native integer widths for the target CPU
1225 this set are considered to support most general arithmetic
1226 operations efficiently.</dd>
1227 </dl>
1229 <p>When constructing the data layout for a given target, LLVM starts with a
1230 default set of specifications which are then (possibly) overriden by the
1232 are given in this list:</p>
1234 <ul>
1249 </ul>
1251 <p>When LLVM is determining the alignment for a given type, it uses the
1252 following rules:</p>
1254 <ol>
1255 <li>If the type sought is an exact match for one of the specifications, that
1256 specification is used.</li>
1258 <li>If no match is found, and the type sought is an integer type, then the
1259 smallest integer type that is larger than the bitwidth of the sought type
1260 is used. If none of the specifications are larger than the bitwidth then
1261 the the largest integer type is used. For example, given the default
1262 specifications above, the i7 type will use the alignment of i8 (next
1263 largest) while both i65 and i256 will use the alignment of i64 (largest
1264 specified).</li>
1266 <li>If no match is found, and the type sought is a vector type, then the
1267 largest vector type that is smaller than the sought vector type will be
1270 </ol>
1272 </div>
1274 <!-- ======================================================================= -->
1277 </div>
1281 <p>Any memory access must be done through a pointer value associated
1282 with an address range of the memory access, otherwise the behavior
1283 is undefined. Pointer values are associated with address ranges
1284 according to the following rules:</p>
1286 <ul>
1287 <li>A pointer value formed from a
1289 is associated with the addresses associated with the first operand
1291 <li>An address of a global variable is associated with the address
1292 range of the variable's storage.</li>
1293 <li>The result value of an allocation instruction is associated with
1294 the address range of the allocated storage.</li>
1295 <li>A null pointer in the default address-space is associated with
1296 no address.</li>
1297 <li>A pointer value formed by an
1299 address ranges of all pointer values that contribute (directly or
1300 indirectly) to the computation of the pointer's value.</li>
1301 <li>The result value of a
1304 <li>An integer constant other than zero or a pointer value returned
1305 from a function not defined within LLVM may be associated with address
1306 ranges allocated through mechanisms other than those provided by
1307 LLVM. Such ranges shall not overlap with any ranges of addresses
1308 allocated by mechanisms provided by LLVM.</li>
1309 </ul>
1311 <p>LLVM IR does not associate types with memory. The result type of a
1313 alignment of the memory from which to load, as well as the
1314 interpretation of the value. The first operand of a
1316 and alignment of the store.</p>
1318 <p>Consequently, type-based alias analysis, aka TBAA, aka
1321 additional information which specialized optimization passes may use
1322 to implement type-based alias analysis.</p>
1324 </div>
1326 <!-- *********************************************************************** -->
1328 <!-- *********************************************************************** -->
1332 <p>The LLVM type system is one of the most important features of the
1333 intermediate representation. Being typed enables a number of optimizations
1334 to be performed on the intermediate representation directly, without having
1335 to do extra analyses on the side before the transformation. A strong type
1336 system makes it easier to read the generated code and enables novel analyses
1337 and transformations that are not feasible to perform on normal three address
1338 code representations.</p>
1340 </div>
1342 <!-- ======================================================================= -->
1344 Classifications</a> </div>
1351 <tbody>
1353 <tr>
1356 </tr>
1357 <tr>
1360 </tr>
1361 <tr>
1371 </td>
1372 </tr>
1373 <tr>
1379 </tr>
1380 <tr>
1390 </td>
1391 </tr>
1392 </tbody>
1393 </table>
1396 important. Values of these types are the only ones which can be produced by
1397 instructions.</p>
1399 </div>
1401 <!-- ======================================================================= -->
1406 <p>The primitive types are the fundamental building blocks of the LLVM
1407 system.</p>
1409 </div>
1411 <!-- _______________________________________________________________________ -->
1417 <p>The integer type is a very simple type that simply specifies an arbitrary
1418 bit width for the integer type desired. Any bit width from 1 bit to
1422 <pre>
1423 iN
1424 </pre>
1427 value.</p>
1434 </tr>
1438 </tr>
1442 </tr>
1443 </table>
1445 </div>
1447 <!-- _______________________________________________________________________ -->
1452 <table>
1453 <tbody>
1460 </tbody>
1461 </table>
1463 </div>
1465 <!-- _______________________________________________________________________ -->
1474 <pre>
1475 void
1476 </pre>
1478 </div>
1480 <!-- _______________________________________________________________________ -->
1489 <pre>
1490 label
1491 </pre>
1493 </div>
1495 <!-- _______________________________________________________________________ -->
1501 <p>The metadata type represents embedded metadata. No derived types may be
1503 arguments.
1506 <pre>
1507 metadata
1508 </pre>
1510 </div>
1513 <!-- ======================================================================= -->
1518 <p>The real power in LLVM comes from the derived types in the system. This is
1519 what allows a programmer to represent arrays, functions, pointers, and other
1520 useful types. Each of these types contain one or more element types which
1521 may be a primitive type, or another derived type. For example, it is
1522 possible to have a two dimensional array, using an array as the element type
1523 of another array.</p>
1525 </div>
1527 <!-- _______________________________________________________________________ -->
1533 <p>The array type is a very simple derived type that arranges elements
1534 sequentially in memory. The array type requires a size (number of elements)
1535 and an underlying data type.</p>
1538 <pre>
1540 </pre>
1543 be any type with a size.</p>
1550 </tr>
1554 </tr>
1558 </tr>
1559 </table>
1565 </tr>
1569 </tr>
1573 </tr>
1574 </table>
1576 <p>There is no restriction on indexing beyond the end of the array implied by
1577 a static type (though there are restrictions on indexing beyond the bounds
1578 of an allocated object in some cases). This means that single-dimension
1579 'variable sized array' addressing can be implemented in LLVM with a zero
1580 length array type. An implementation of 'pascal style arrays' in LLVM could
1583 </div>
1585 <!-- _______________________________________________________________________ -->
1591 <p>The function type can be thought of as a function signature. It consists of
1592 a return type and a list of formal parameter types. The return type of a
1593 function type is a scalar type, a void type, or a struct type. If the return
1594 type is a struct type then all struct elements must be of first class types,
1595 and the struct must have at least one element.</p>
1598 <pre>
1600 </pre>
1604 which indicates that the function takes a variable number of arguments.
1605 Variable argument functions can access their arguments with
1615 </td>
1618 </tt></td>
1623 </td>
1629 LLVM.
1630 </td>
1635 </td>
1636 </tr>
1637 </table>
1639 </div>
1641 <!-- _______________________________________________________________________ -->
1647 <p>The structure type is used to represent a collection of data members together
1648 in memory. The packing of the field types is defined to match the ABI of the
1649 underlying processor. The elements of a structure may be any type that has a
1650 size.</p>
1657 <pre>
1659 </pre>
1672 </tr>
1673 </table>
1675 </div>
1677 <!-- _______________________________________________________________________ -->
1679 </div>
1684 <p>The packed structure type is used to represent a collection of data members
1685 together in memory. There is no padding between fields. Further, the
1686 alignment of a packed structure is 1 byte. The elements of a packed
1687 structure may be any type that has a size.</p>
1694 <pre>
1696 </pre>
1710 </tr>
1711 </table>
1713 </div>
1715 <!-- _______________________________________________________________________ -->
1721 <p>As in many languages, the pointer type represents a pointer or reference to
1722 another object, which must live in memory. Pointer types may have an optional
1723 address space attribute defining the target-specific numbered address space
1724 where the pointed-to object resides. The default address space is zero.</p>
1730 <pre>
1732 </pre>
1740 </tr>
1746 </tr>
1751 </tr>
1752 </table>
1754 </div>
1756 <!-- _______________________________________________________________________ -->
1762 <p>A vector type is a simple derived type that represents a vector of elements.
1763 Vector types are used when multiple primitive data are operated in parallel
1764 using a single instruction (SIMD). A vector type requires a size (number of
1765 elements) and an underlying primitive data type. Vector types are considered
1769 <pre>
1771 </pre>
1773 <p>The number of elements is a constant integer value; elementtype may be any
1774 integer or floating point type.</p>
1781 </tr>
1785 </tr>
1789 </tr>
1790 </table>
1792 </div>
1794 <!-- _______________________________________________________________________ -->
1799 <p>Opaque types are used to represent unknown types in the system. This
1800 corresponds (for example) to the C notion of a forward declared structure
1801 type. In LLVM, opaque types can eventually be resolved to any type (not just
1802 a structure type).</p>
1805 <pre>
1806 opaque
1807 </pre>
1814 </tr>
1815 </table>
1817 </div>
1819 <!-- ======================================================================= -->
1822 </div>
1828 requiring it to have a name. For instance, a structure declaration may
1829 contain a pointer to any of the types it is lexically a member of. Example
1830 of up references (with their equivalent as named type declarations)
1831 include:</p>
1833 <pre>
1834 { \2 * } %x = type { %x* }
1835 { \2 }* %y = type { %y }*
1836 \1* %z = type %z*
1837 </pre>
1839 <p>An up reference is needed by the asmprinter for printing out cyclic types
1840 when there is no declared name for a type in the cycle. Because the
1841 asmprinter does not want to print out an infinite type string, it needs a
1842 syntax to handle recursive types that have no names (all names are optional
1843 in llvm IR).</p>
1846 <pre>
1848 </pre>
1857 </tr>
1861 structure.</td>
1862 </tr>
1863 </table>
1865 </div>
1867 <!-- *********************************************************************** -->
1869 <!-- *********************************************************************** -->
1873 <p>LLVM has several different basic types of constants. This section describes
1874 them all and their syntax.</p>
1876 </div>
1878 <!-- ======================================================================= -->
1883 <dl>
1891 with integer types.</dd>
1895 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1896 notation (see below). The assembler requires the exact decimal value of a
1897 floating-point constant. For example, the assembler accepts 1.25 but
1904 </dl>
1906 <p>The one non-intuitive notation for constants is the hexadecimal form of
1907 floating point constants. For example, the form '<tt>double
1910 constants are required (and the only time that they are generated by the
1911 disassembler) is when a floating point constant must be emitted but it cannot
1912 be represented as a decimal floating point number in a reasonable number of
1913 digits. For example, NaN's, infinities, and other special values are
1914 represented in their IEEE hexadecimal format so that assembly and disassembly
1915 do not cause any bits to change in the constants.</p>
1917 <p>When using the hexadecimal form, constants of types float and double are
1918 represented using the 16-digit form shown above (which matches the IEEE754
1919 representation for double); float values must, however, be exactly
1920 representable as IEE754 single precision. Hexadecimal format is always used
1921 for long double, and there are three forms of long double. The 80-bit format
1923 The 128-bit format used by PowerPC (two adjacent doubles) is represented
1926 currently supported target uses this format. Long doubles will only work if
1927 they match the long double format on your target. All hexadecimal formats
1928 are big-endian (sign bit at the left).</p>
1930 </div>
1932 <!-- ======================================================================= -->
1936 </div>
1940 <p>Complex constants are a (potentially recursive) combination of simple
1941 constants and smaller complex constants.</p>
1943 <dl>
1945 <dd>Structure constants are represented with notation similar to structure
1946 type definitions (a comma separated list of elements, surrounded by braces
1950 the number and types of elements must match those specified by the
1951 type.</dd>
1954 <dd>Array constants are represented with notation similar to array type
1955 definitions (a comma separated list of elements, surrounded by square
1958 the number and types of elements must match those specified by the
1959 type.</dd>
1962 <dd>Vector constants are represented with notation similar to vector type
1963 definitions (a comma separated list of elements, surrounded by
1965 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
1967 elements must match those specified by the type.</dd>
1972 This is often used to avoid having to print large zero initializers
1973 (e.g. for large arrays) and is always exactly equivalent to using explicit
1974 zero initializers.</dd>
1977 <dd>A metadata node is a structure-like constant with
1980 be interpreted as part of the instruction stream, metadata is a place to
1981 attach additional information such as debug info.</dd>
1982 </dl>
1984 </div>
1986 <!-- ======================================================================= -->
1989 </div>
1995 (link-time) constants. These constants are explicitly referenced when
1998 legal LLVM file:</p>
2001 <pre>
2002 @X = global i32 17
2003 @Y = global i32 42
2004 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2005 </pre>
2006 </div>
2008 </div>
2010 <!-- ======================================================================= -->
2015 indicates that the user of the value may receive an unspecified bit-pattern.
2016 Undefined values may be of any type (other than label or void) and be used
2017 anywhere a constant is permitted.</p>
2019 <p>Undefined values are useful because they indicate to the compiler that the
2020 program is well defined no matter what value is used. This gives the
2021 compiler more freedom to optimize. Here are some examples of (potentially
2022 surprising) transformations that are valid (in pseudo IR):</p>
2026 <pre>
2027 %A = add %X, undef
2028 %B = sub %X, undef
2029 %C = xor %X, undef
2030 Safe:
2031 %A = undef
2032 %B = undef
2033 %C = undef
2034 </pre>
2035 </div>
2037 <p>This is safe because all of the output bits are affected by the undef bits.
2038 Any output bit can have a zero or one depending on the input bits.</p>
2041 <pre>
2042 %A = or %X, undef
2043 %B = and %X, undef
2044 Safe:
2045 %A = -1
2046 %B = 0
2047 Unsafe:
2048 %A = undef
2049 %B = undef
2050 </pre>
2051 </div>
2053 <p>These logical operations have bits that are not always affected by the input.
2054 For example, if "%X" has a zero bit, then the output of the 'and' operation will
2055 always be a zero, no matter what the corresponding bit from the undef is. As
2056 such, it is unsafe to optimize or assume that the result of the and is undef.
2057 However, it is safe to assume that all bits of the undef could be 0, and
2058 optimize the and to 0. Likewise, it is safe to assume that all the bits of
2059 the undef operand to the or could be set, allowing the or to be folded to
2063 <pre>
2064 %A = select undef, %X, %Y
2065 %B = select undef, 42, %Y
2066 %C = select %X, %Y, undef
2067 Safe:
2068 %A = %X (or %Y)
2069 %B = 42 (or %Y)
2070 %C = %Y
2071 Unsafe:
2072 %A = undef
2073 %B = undef
2074 %C = undef
2075 </pre>
2076 </div>
2078 <p>This set of examples show that undefined select (and conditional branch)
2079 conditions can go "either way" but they have to come from one of the two
2080 operands. In the %A example, if %X and %Y were both known to have a clear low
2081 bit, then %A would have to have a cleared low bit. However, in the %C example,
2082 the optimizer is allowed to assume that the undef operand could be the same as
2083 %Y, allowing the whole select to be eliminated.</p>
2087 <pre>
2088 %A = xor undef, undef
2090 %B = undef
2091 %C = xor %B, %B
2093 %D = undef
2094 %E = icmp lt %D, 4
2095 %F = icmp gte %D, 4
2097 Safe:
2098 %A = undef
2099 %B = undef
2100 %C = undef
2101 %D = undef
2102 %E = undef
2103 %F = undef
2104 </pre>
2105 </div>
2107 <p>This example points out that two undef operands are not necessarily the same.
2108 This can be surprising to people (and also matches C semantics) where they
2109 assume that "X^X" is always zero, even if X is undef. This isn't true for a
2110 number of reasons, but the short answer is that an undef "variable" can
2111 arbitrarily change its value over its "live range". This is true because the
2113 logically read from arbitrary registers that happen to be around when needed,
2114 so the value is not necessarily consistent over time. In fact, %A and %C need
2115 to have the same semantics or the core LLVM "replace all uses with" concept
2116 would not hold.</p>
2119 <pre>
2120 %A = fdiv undef, %X
2121 %B = fdiv %X, undef
2122 Safe:
2123 %A = undef
2124 b: unreachable
2125 </pre>
2126 </div>
2130 allowed to have an arbitrary bit-pattern. This means that the %A operation
2131 can be constant folded to undef because the undef could be an SNaN, and fdiv is
2132 not (currently) defined on SNaN's. However, in the second example, we can make
2133 a more aggressive assumption: because the undef is allowed to be an arbitrary
2134 value, we are allowed to assume that it could be zero. Since a divide by zero
2136 does not execute at all. This allows us to delete the divide and all code after
2137 it: since the undefined operation "can't happen", the optimizer can assume that
2138 it occurs in dead code.
2139 </p>
2142 <pre>
2143 a: store undef -> %X
2144 b: store %X -> undef
2145 Safe:
2147 b: unreachable
2148 </pre>
2149 </div>
2152 can be assumed to not have any effect: we can assume that the value is
2153 overwritten with bits that happen to match what was already there. However, a
2154 store "to" an undefined location could clobber arbitrary memory, therefore, it
2155 has undefined behavior.</p>
2157 </div>
2159 <!-- ======================================================================= -->
2161 Blocks</a></div>
2167 basic block in the specified function, and always has an i8* type. Taking
2168 the address of the entry block is illegal.</p>
2170 <p>This value only has defined behavior when used as an operand to the
2172 against null. Pointer equality tests between labels addresses is undefined
2173 behavior - though, again, comparison against null is ok, and no label is
2174 equal to the null pointer. This may also be passed around as an opaque
2175 pointer sized value as long as the bits are not inspected. This allows
2179 <p>Finally, some targets may provide defined semantics when
2180 using the value as the operand to an inline assembly, but that is target
2181 specific.
2182 </p>
2184 </div>
2187 <!-- ======================================================================= -->
2189 </div>
2193 <p>Constant expressions are used to allow expressions involving other constants
2194 to be used as constants. Constant expressions may be of
2196 operation that does not have side effects (e.g. load and call are not
2197 supported). The following is the syntax for constant expressions:</p>
2199 <dl>
2201 <dd>Truncate a constant to another type. The bit size of CST must be larger
2202 than the bit size of TYPE. Both types must be integers.</dd>
2205 <dd>Zero extend a constant to another type. The bit size of CST must be
2206 smaller or equal to the bit size of TYPE. Both types must be
2207 integers.</dd>
2210 <dd>Sign extend a constant to another type. The bit size of CST must be
2211 smaller or equal to the bit size of TYPE. Both types must be
2212 integers.</dd>
2215 <dd>Truncate a floating point constant to another floating point type. The
2216 size of CST must be larger than the size of TYPE. Both types must be
2217 floating point.</dd>
2220 <dd>Floating point extend a constant to another type. The size of CST must be
2221 smaller or equal to the size of TYPE. Both types must be floating
2222 point.</dd>
2225 <dd>Convert a floating point constant to the corresponding unsigned integer
2226 constant. TYPE must be a scalar or vector integer type. CST must be of
2227 scalar or vector floating point type. Both CST and TYPE must be scalars,
2228 or vectors of the same number of elements. If the value won't fit in the
2229 integer type, the results are undefined.</dd>
2232 <dd>Convert a floating point constant to the corresponding signed integer
2233 constant. TYPE must be a scalar or vector integer type. CST must be of
2234 scalar or vector floating point type. Both CST and TYPE must be scalars,
2235 or vectors of the same number of elements. If the value won't fit in the
2236 integer type, the results are undefined.</dd>
2239 <dd>Convert an unsigned integer constant to the corresponding floating point
2240 constant. TYPE must be a scalar or vector floating point type. CST must be
2241 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2242 vectors of the same number of elements. If the value won't fit in the
2243 floating point type, the results are undefined.</dd>
2246 <dd>Convert a signed integer constant to the corresponding floating point
2247 constant. TYPE must be a scalar or vector floating point type. CST must be
2248 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2249 vectors of the same number of elements. If the value won't fit in the
2250 floating point type, the results are undefined.</dd>
2253 <dd>Convert a pointer typed constant to the corresponding integer constant
2259 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2260 type. CST must be of integer type. The CST value is zero extended,
2261 truncated, or unchanged to make it fit in a pointer size. This one is
2265 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2273 instruction, the index list may have zero or more indexes, which are
2287 constants.</dd>
2291 constants.</dd>
2295 constants.</dd>
2298 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2301 on operands are the same as those for the corresponding instruction
2302 (e.g. no bitwise operations on floating point values are allowed).</dd>
2303 </dl>
2305 </div>
2307 <!-- ======================================================================= -->
2309 </div>
2313 <p>Embedded metadata provides a way to attach arbitrary data to the instruction
2314 stream without affecting the behaviour of the program. There are two
2315 metadata primitives, strings and nodes. All metadata has the
2319 <p>A metadata string is a string surrounded by double quotes. It can contain
2323 <p>Metadata nodes are represented with notation similar to structure constants
2324 (a comma separated list of elements, surrounded by braces and preceded by an
2328 <p>A metadata node will attempt to track changes to the values it holds. In the
2329 event that a value is deleted, it will be replaced with a typeless
2332 <p>Optimizations may rely on metadata to provide additional information about
2333 the program that isn't available in the instructions, or that isn't easily
2334 computable. Similarly, the code generator may expect a certain metadata
2335 format to be used to express debugging information.</p>
2337 </div>
2339 <!-- *********************************************************************** -->
2341 <!-- *********************************************************************** -->
2343 <!-- ======================================================================= -->
2346 </div>
2350 <p>LLVM supports inline assembler expressions (as opposed
2352 a special value. This value represents the inline assembler as a string
2353 (containing the instructions to emit), a list of operand constraints (stored
2354 as a string), a flag that indicates whether or not the inline asm
2355 expression has side effects, and a flag indicating whether the function
2356 containing the asm needs to align its stack conservatively. An example
2357 inline assembler expression is:</p>
2360 <pre>
2362 </pre>
2363 </div>
2367 have:</p>
2370 <pre>
2372 </pre>
2373 </div>
2375 <p>Inline asms with side effects not visible in the constraint list must be
2376 marked as having side effects. This is done through the use of the
2380 <pre>
2382 </pre>
2383 </div>
2385 <p>In some cases inline asms will contain code that will not work unless the
2386 stack is aligned in some way, such as calls or SSE instructions on x86,
2387 yet will not contain code that does that alignment within the asm.
2388 The compiler should make conservative assumptions about what the asm might
2389 contain and should generate its usual stack alignment code in the prologue
2393 <pre>
2395 </pre>
2396 </div>
2399 first.</p>
2401 <p>TODO: The format of the asm and constraints string still need to be
2402 documented here. Constraints on what can be done (e.g. duplication, moving,
2403 etc need to be documented). This is probably best done by reference to
2404 another document that covers inline asm from a holistic perspective.</p>
2406 </div>
2409 <!-- *********************************************************************** -->
2412 </div>
2413 <!-- *********************************************************************** -->
2416 code generation or other IR semantics. These are documented here. All globals
2417 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2418 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2419 by LLVM.</p>
2421 <!-- ======================================================================= -->
2424 </div>
2430 pointers to global variables and functions which may optionally have a pointer
2431 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2433 <pre>
2434 @X = global i8 4
2435 @Y = global i32 123
2437 @llvm.used = appending global [2 x i8*] [
2438 i8* @X,
2439 i8* bitcast (i32* @Y to i8*)
2440 ], section "llvm.metadata"
2441 </pre>
2444 compiler, assembler, and linker are required to treat the symbol as if there is
2445 a reference to the global that it cannot see. For example, if a variable has
2447 list, it cannot be deleted. This is commonly used to represent references from
2448 inline asms and other things the compiler cannot "see", and corresponds to
2451 <p>On some targets, the code generator must emit a directive to the assembler or
2452 object file to prevent the assembler and linker from molesting the symbol.</p>
2454 </div>
2456 <!-- ======================================================================= -->
2459 </div>
2465 touching the symbol. On targets that support it, this allows an intelligent
2466 linker to optimize references to the symbol without being impeded as it would be
2469 <p>This is a rare construct that should only be used in rare circumstances, and
2470 should not be exposed to source languages.</p>
2472 </div>
2474 <!-- ======================================================================= -->
2477 </div>
2483 </div>
2485 <!-- ======================================================================= -->
2488 </div>
2494 </div>
2497 <!-- *********************************************************************** -->
2499 <!-- *********************************************************************** -->
2503 <p>The LLVM instruction set consists of several different classifications of
2510 </div>
2512 <!-- ======================================================================= -->
2514 Instructions</a> </div>
2519 in a program ends with a "Terminator" instruction, which indicates which
2520 block should be executed after the current block is finished. These
2522 control flow, not values (the one exception being the
2525 <p>There are six different terminator instructions: the
2534 </div>
2536 <!-- _______________________________________________________________________ -->
2538 Instruction</a> </div>
2543 <pre>
2546 </pre>
2550 a value) from a function back to the caller.</p>
2553 value and then causes control flow, and one that just causes control flow to
2554 occur.</p>
2558 return value. The type of the return value must be a
2563 value or a return value with a type that does not match its type, or if it
2565 return value.</p>
2569 the calling function's context. If the caller is a
2571 instruction after the call. If the caller was an
2573 the beginning of the "normal" destination block. If the instruction returns
2574 a value, that value shall set the call or invoke instruction's return
2575 value.</p>
2578 <pre>
2582 </pre>
2584 </div>
2585 <!-- _______________________________________________________________________ -->
2591 <pre>
2592 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2593 </pre>
2597 different basic block in the current function. There are two forms of this
2598 instruction, corresponding to a conditional branch and an unconditional
2599 branch.</p>
2605 target.</p>
2614 <pre>
2615 Test:
2617 br i1 %cond, label %IfEqual, label %IfUnequal
2618 IfEqual:
2620 IfUnequal:
2622 </pre>
2624 </div>
2626 <!-- _______________________________________________________________________ -->
2629 </div>
2634 <pre>
2635 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2636 </pre>
2641 instruction, allowing a branch to occur to one of many possible
2642 destinations.</p>
2648 The table is not allowed to contain duplicate constant entries.</p>
2653 is searched for the given value. If the value is found, control flow is
2654 transferred to the corresponding destination; otherwise, control flow is
2655 transferred to the default destination.</p>
2658 <p>Depending on properties of the target machine and the particular
2660 different ways. For example, it could be generated as a series of chained
2661 conditional branches or with a lookup table.</p>
2664 <pre>
2667 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2670 switch i32 0, label %dest [ ]
2673 switch i32 %val, label %otherwise [ i32 0, label %onzero
2674 i32 1, label %onone
2675 i32 2, label %ontwo ]
2676 </pre>
2678 </div>
2681 <!-- _______________________________________________________________________ -->
2684 </div>
2689 <pre>
2691 </pre>
2696 within the current function, whose address is specified by
2703 rest of the arguments indicate the full set of possible destinations that the
2704 address may point to. Blocks are allowed to occur multiple times in the
2705 destination list, though this isn't particularly useful.</p>
2707 <p>This destination list is required so that dataflow analysis has an accurate
2708 understanding of the CFG.</p>
2712 <p>Control transfers to the block specified in the address argument. All
2713 possible destination blocks must be listed in the label list, otherwise this
2714 instruction has undefined behavior. This implies that jumps to labels
2715 defined in other functions have undefined behavior as well.</p>
2722 <pre>
2723 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
2724 </pre>
2726 </div>
2729 <!-- _______________________________________________________________________ -->
2732 </div>
2737 <pre>
2738 <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>]
2740 </pre>
2744 function, with the possibility of control flow transfer to either the
2747 control flow will return to the "normal" label. If the callee (or any
2749 instruction, control is interrupted and continued at the dynamically nearest
2755 <ol>
2757 convention</a> the call should use. If none is specified, the call
2758 defaults to using C calling conventions.</li>
2765 function value being invoked. In most cases, this is a direct function
2767 off an arbitrary pointer to function value.</li>
2770 function to be invoked. </li>
2773 signature argument types. If the function signature indicates the
2774 function accepts a variable number of arguments, the extra arguments can
2775 be specified.</li>
2786 </ol>
2789 <p>This instruction is designed to operate as a standard
2791 primary difference is that it establishes an association with a label, which
2792 is used by the runtime library to unwind the stack.</p>
2794 <p>This instruction is used in languages with destructors to ensure that proper
2796 exception. Additionally, this is important for implementation of
2799 <p>For the purposes of the SSA form, the definition of the value returned by the
2801 block to the "normal" label. If the callee unwinds then no return value is
2802 available.</p>
2805 <pre>
2806 %retval = invoke i32 @Test(i32 15) to label %Continue
2810 </pre>
2812 </div>
2814 <!-- _______________________________________________________________________ -->
2817 Instruction</a> </div>
2822 <pre>
2823 unwind
2824 </pre>
2828 at the first callee in the dynamic call stack which used
2830 This is primarily used to implement exception handling.</p>
2834 immediately halt. The dynamic call stack is then searched for the
2836 Once found, execution continues at the "exceptional" destination block
2838 instruction in the dynamic call chain, undefined behavior results.</p>
2840 </div>
2842 <!-- _______________________________________________________________________ -->
2845 Instruction</a> </div>
2850 <pre>
2851 unreachable
2852 </pre>
2856 instruction is used to inform the optimizer that a particular portion of the
2857 code is not reachable. This can be used to indicate that the code after a
2858 no-return function cannot be reached, and other facts.</p>
2863 </div>
2865 <!-- ======================================================================= -->
2870 <p>Binary operators are used to do most of the computation in a program. They
2871 require two operands of the same type, execute an operation on them, and
2872 produce a single value. The operands might represent multiple data, as is
2874 has the same type as its operands.</p>
2878 </div>
2880 <!-- _______________________________________________________________________ -->
2883 </div>
2888 <pre>
2893 </pre>
2901 integer values. Both arguments must have identical types.</p>
2906 <p>If the sum has unsigned overflow, the result returned is the mathematical
2909 <p>Because LLVM integers use a two's complement representation, this instruction
2910 is appropriate for both signed and unsigned integers.</p>
2915 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
2918 <pre>
2920 </pre>
2922 </div>
2924 <!-- _______________________________________________________________________ -->
2927 </div>
2932 <pre>
2934 </pre>
2942 floating point values. Both arguments must have identical types.</p>
2948 <pre>
2950 </pre>
2952 </div>
2954 <!-- _______________________________________________________________________ -->
2957 </div>
2962 <pre>
2967 </pre>
2971 operands.</p>
2975 representations.</p>
2980 integer values. Both arguments must have identical types.</p>
2985 <p>If the difference has unsigned overflow, the result returned is the
2987 result.</p>
2989 <p>Because LLVM integers use a two's complement representation, this instruction
2990 is appropriate for both signed and unsigned integers.</p>
2995 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
2998 <pre>
3001 </pre>
3003 </div>
3005 <!-- _______________________________________________________________________ -->
3008 </div>
3013 <pre>
3015 </pre>
3019 operands.</p>
3023 representations.</p>
3028 floating point values. Both arguments must have identical types.</p>
3034 <pre>
3037 </pre>
3039 </div>
3041 <!-- _______________________________________________________________________ -->
3044 </div>
3049 <pre>
3054 </pre>
3062 integer values. Both arguments must have identical types.</p>
3067 <p>If the result of the multiplication has unsigned overflow, the result
3069 width of the result.</p>
3071 <p>Because LLVM integers use a two's complement representation, and the result
3072 is the same width as the operands, this instruction returns the correct
3073 result for both signed and unsigned integers. If a full product
3075 be sign-extended or zero-extended as appropriate to the width of the full
3076 product.</p>
3081 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
3084 <pre>
3086 </pre>
3088 </div>
3090 <!-- _______________________________________________________________________ -->
3093 </div>
3098 <pre>
3100 </pre>
3108 floating point values. Both arguments must have identical types.</p>
3114 <pre>
3116 </pre>
3118 </div>
3120 <!-- _______________________________________________________________________ -->
3122 </a></div>
3127 <pre>
3129 </pre>
3137 values. Both arguments must have identical types.</p>
3142 <p>Note that unsigned integer division and signed integer division are distinct
3148 <pre>
3150 </pre>
3152 </div>
3154 <!-- _______________________________________________________________________ -->
3156 </a> </div>
3161 <pre>
3164 </pre>
3172 values. Both arguments must have identical types.</p>
3175 <p>The value produced is the signed integer quotient of the two operands rounded
3176 towards zero.</p>
3178 <p>Note that signed integer division and unsigned integer division are distinct
3181 <p>Division by zero leads to undefined behavior. Overflow also leads to
3182 undefined behavior; this is a rare case, but can occur, for example, by doing
3187 would occur.</p>
3190 <pre>
3192 </pre>
3194 </div>
3196 <!-- _______________________________________________________________________ -->
3198 Instruction</a> </div>
3203 <pre>
3205 </pre>
3213 floating point values. Both arguments must have identical types.</p>
3219 <pre>
3221 </pre>
3223 </div>
3225 <!-- _______________________________________________________________________ -->
3227 </div>
3232 <pre>
3234 </pre>
3238 division of its two arguments.</p>
3243 values. Both arguments must have identical types.</p>
3247 This instruction always performs an unsigned division to get the
3248 remainder.</p>
3250 <p>Note that unsigned integer remainder and signed integer remainder are
3256 <pre>
3258 </pre>
3260 </div>
3262 <!-- _______________________________________________________________________ -->
3265 </div>
3270 <pre>
3272 </pre>
3276 division of its two operands. This instruction can also take
3278 elements must be integers.</p>
3283 values. Both arguments must have identical types.</p>
3289 a value. For more information about the difference,
3291 Math Forum</a>. For a table of how this is implemented in various languages,
3295 <p>Note that signed integer remainder and unsigned integer remainder are
3298 <p>Taking the remainder of a division by zero leads to undefined behavior.
3299 Overflow also leads to undefined behavior; this is a rare case, but can
3300 occur, for example, by taking the remainder of a 32-bit division of
3302 lets srem be implemented using instructions that return both the result of
3303 the division and the remainder.)</p>
3306 <pre>
3308 </pre>
3310 </div>
3312 <!-- _______________________________________________________________________ -->
3319 <pre>
3321 </pre>
3325 its two operands.</p>
3330 floating point values. Both arguments must have identical types.</p>
3334 has the same sign as the dividend.</p>
3337 <pre>
3339 </pre>
3341 </div>
3343 <!-- ======================================================================= -->
3345 Operations</a> </div>
3349 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3350 program. They are generally very efficient instructions and can commonly be
3351 strength reduced from other instructions. They require two operands of the
3352 same type, execute an operation on them, and produce a single value. The
3353 resulting value is the same type as its operands.</p>
3355 </div>
3357 <!-- _______________________________________________________________________ -->
3359 Instruction</a> </div>
3364 <pre>
3366 </pre>
3370 a specified number of bits.</p>
3380 is (statically or dynamically) negative or equal to or larger than the number
3386 <pre>
3391 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3392 </pre>
3394 </div>
3396 <!-- _______________________________________________________________________ -->
3398 Instruction</a> </div>
3403 <pre>
3405 </pre>
3409 operand shifted to the right a specified number of bits with zero fill.</p>
3417 <p>This instruction always performs a logical shift right operation. The most
3418 significant bits of the result will be filled with zero bits after the shift.
3425 <pre>
3431 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3432 </pre>
3434 </div>
3436 <!-- _______________________________________________________________________ -->
3438 Instruction</a> </div>
3442 <pre>
3444 </pre>
3448 operand shifted to the right a specified number of bits with sign
3449 extension.</p>
3457 <p>This instruction always performs an arithmetic shift right operation, The
3458 most significant bits of the result will be filled with the sign bit
3465 <pre>
3471 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3472 </pre>
3474 </div>
3476 <!-- _______________________________________________________________________ -->
3478 Instruction</a> </div>
3483 <pre>
3485 </pre>
3489 operands.</p>
3494 values. Both arguments must have identical types.</p>
3500 <tbody>
3501 <tr>
3505 </tr>
3506 <tr>
3510 </tr>
3511 <tr>
3515 </tr>
3516 <tr>
3520 </tr>
3521 <tr>
3525 </tr>
3526 </tbody>
3527 </table>
3530 <pre>
3534 </pre>
3535 </div>
3536 <!-- _______________________________________________________________________ -->
3542 <pre>
3544 </pre>
3548 two operands.</p>
3553 values. Both arguments must have identical types.</p>
3559 <tbody>
3560 <tr>
3564 </tr>
3565 <tr>
3569 </tr>
3570 <tr>
3574 </tr>
3575 <tr>
3579 </tr>
3580 <tr>
3584 </tr>
3585 </tbody>
3586 </table>
3589 <pre>
3593 </pre>
3595 </div>
3597 <!-- _______________________________________________________________________ -->
3599 Instruction</a> </div>
3604 <pre>
3606 </pre>
3616 values. Both arguments must have identical types.</p>
3622 <tbody>
3623 <tr>
3627 </tr>
3628 <tr>
3632 </tr>
3633 <tr>
3637 </tr>
3638 <tr>
3642 </tr>
3643 <tr>
3647 </tr>
3648 </tbody>
3649 </table>
3652 <pre>
3657 </pre>
3659 </div>
3661 <!-- ======================================================================= -->
3664 </div>
3668 <p>LLVM supports several instructions to represent vector operations in a
3669 target-independent manner. These instructions cover the element-access and
3670 vector-specific operations needed to process vectors effectively. While LLVM
3671 does directly support these vector operations, many sophisticated algorithms
3672 will want to use target-specific intrinsics to take full advantage of a
3673 specific target.</p>
3675 </div>
3677 <!-- _______________________________________________________________________ -->
3680 </div>
3685 <pre>
3686 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3687 </pre>
3691 from a vector at a specified index.</p>
3697 indicating the position from which to extract the element. The index may be
3698 a variable.</p>
3701 <p>The result is a scalar of the same type as the element type of
3704 results are undefined.</p>
3707 <pre>
3709 </pre>
3711 </div>
3713 <!-- _______________________________________________________________________ -->
3716 </div>
3721 <pre>
3722 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3723 </pre>
3727 vector at a specified index.</p>
3732 whose type must equal the element type of the first operand. The third
3733 operand is an index indicating the position at which to insert the value.
3734 The index may be a variable.</p>
3740 results are undefined.</p>
3743 <pre>
3744 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
3745 </pre>
3747 </div>
3749 <!-- _______________________________________________________________________ -->
3752 </div>
3757 <pre>
3758 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3759 </pre>
3763 from two input vectors, returning a vector with the same element type as the
3764 input and length that is the same as the shuffle mask.</p>
3768 with types that match each other. The third argument is a shuffle mask whose
3769 element type is always 'i32'. The result of the instruction is a vector
3770 whose length is the same as the shuffle mask and whose element type is the
3771 same as the element type of the first two operands.</p>
3773 <p>The shuffle mask operand is required to be a constant vector with either
3774 constant integer or undef values.</p>
3777 <p>The elements of the two input vectors are numbered from left to right across
3778 both of the vectors. The shuffle mask operand specifies, for each element of
3779 the result vector, which element of the two input vectors the result element
3780 gets. The element selector may be undef (meaning "don't care") and the
3781 second operand may be undef if performing a shuffle from only one vector.</p>
3784 <pre>
3788 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3792 <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>
3793 </pre>
3795 </div>
3797 <!-- ======================================================================= -->
3800 </div>
3806 </div>
3808 <!-- _______________________________________________________________________ -->
3811 </div>
3816 <pre>
3818 </pre>
3822 or array element from an aggregate value.</p>
3827 operands are constant indices to specify which value to extract in a similar
3828 manner as indices in a
3832 <p>The result is the value at the position in the aggregate specified by the
3833 index operands.</p>
3836 <pre>
3838 </pre>
3840 </div>
3842 <!-- _______________________________________________________________________ -->
3845 </div>
3850 <pre>
3851 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3852 </pre>
3856 array element in an aggregate.</p>
3862 second operand is a first-class value to insert. The following operands are
3863 constant indices indicating the position at which to insert the value in a
3864 similar manner as indices in a
3866 value to insert must have the same type as the value identified by the
3867 indices.</p>
3875 <pre>
3877 </pre>
3879 </div>
3882 <!-- ======================================================================= -->
3885 </div>
3889 <p>A key design point of an SSA-based representation is how it represents
3890 memory. In LLVM, no memory locations are in SSA form, which makes things
3891 very simple. This section describes how to read, write, and allocate
3892 memory in LLVM.</p>
3894 </div>
3896 <!-- _______________________________________________________________________ -->
3899 </div>
3904 <pre>
3905 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3906 </pre>
3910 currently executing function, to be automatically released when this function
3911 returns to its caller. The object is always allocated in the generic address
3912 space (address space zero).</p>
3917 runtime stack, returning a pointer of the appropriate type to the program.
3918 If "NumElements" is specified, it is the number of elements allocated,
3919 otherwise "NumElements" is defaulted to be one. If a constant alignment is
3920 specified, the value result of the allocation is guaranteed to be aligned to
3921 at least that boundary. If not specified, or if zero, the target can choose
3922 to align the allocation on any convenient boundary compatible with the
3923 type.</p>
3928 <p>Memory is allocated; a pointer is returned. The operation is undefined if
3930 memory is automatically released when the function returns. The
3932 variables that must have an address available. When the function returns
3935 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
3938 <pre>
3943 </pre>
3945 </div>
3947 <!-- _______________________________________________________________________ -->
3949 Instruction</a> </div>
3954 <pre>
3957 </pre>
3964 from which to load. The pointer must point to
3969 instructions. </p>
3972 operation (that is, the alignment of the memory address). A value of 0 or an
3973 omitted "align" argument means that the operation has the preferential
3974 alignment for the target. It is the responsibility of the code emitter to
3975 ensure that the alignment information is correct. Overestimating the
3976 alignment results in an undefined behavior. Underestimating the alignment may
3980 <p>The location of memory pointed to is loaded. If the value being loaded is of
3981 scalar type then the number of bytes read does not exceed the minimum number
3982 of bytes needed to hold all bits of the type. For example, loading an
3985 is undefined if the value was not originally written using a store of the
3986 same type.</p>
3989 <pre>
3993 </pre>
3995 </div>
3997 <!-- _______________________________________________________________________ -->
3999 Instruction</a> </div>
4004 <pre>
4005 store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
4006 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
4007 </pre>
4014 and an address at which to store it. The type of the
4021 instructions.</p>
4024 operation (that is, the alignment of the memory address). A value of 0 or an
4025 omitted "align" argument means that the operation has the preferential
4026 alignment for the target. It is the responsibility of the code emitter to
4027 ensure that the alignment information is correct. Overestimating the
4028 alignment results in an undefined behavior. Underestimating the alignment may
4035 does not exceed the minimum number of bytes needed to hold all bits of the
4038 integral number of bytes, it is unspecified what happens to the extra bits
4039 that do not belong to the type, but they will typically be overwritten.</p>
4042 <pre>
4046 </pre>
4048 </div>
4050 <!-- _______________________________________________________________________ -->
4053 </div>
4058 <pre>
4061 </pre>
4065 subelement of an aggregate data structure. It performs address calculation
4066 only and does not access memory.</p>
4069 <p>The first argument is always a pointer, and forms the basis of the
4070 calculation. The remaining arguments are indices that indicate which of the
4071 elements of the aggregate object are indexed. The interpretation of each
4072 index is dependent on the type being indexed into. The first index always
4073 indexes the pointer value given as the first argument, the second index
4074 indexes a value of the type pointed to (not necessarily the value directly
4075 pointed to, since the first index can be non-zero), etc. The first type
4076 indexed into must be a pointer value, subsequent types can be arrays, vectors
4077 and structs. Note that subsequent types being indexed into can never be
4078 pointers, since that would require loading the pointer before continuing
4079 calculation.</p>
4081 <p>The type of each index argument depends on the type it is indexing into.
4084 vector, integers of any width are allowed, and they are not required to be
4085 constant.</p>
4087 <p>For example, let's consider a C code fragment and how it gets compiled to
4088 LLVM:</p>
4091 <pre>
4092 struct RT {
4093 char A;
4095 char C;
4096 };
4097 struct ST {
4098 int X;
4099 double Y;
4100 struct RT Z;
4101 };
4103 int *foo(struct ST *s) {
4105 }
4106 </pre>
4107 </div>
4112 <pre>
4116 define i32* @foo(%ST* %s) {
4117 entry:
4119 ret i32* %reg
4120 }
4121 </pre>
4122 </div>
4127 }</tt>' type, a structure. The second index indexes into the third element
4129 i8 }</tt>' type, another structure. The third index indexes into the second
4131 array. The two dimensions of the array are subscripted into, yielding an
4135 <p>Note that it is perfectly legal to index partially through a structure,
4136 returning a pointer to an inner element. Because of this, the LLVM code for
4137 the given testcase is equivalent to:</p>
4139 <pre>
4140 define i32* @foo(%ST* %s) {
4146 ret i32* %t5
4147 }
4148 </pre>
4153 that would be formed by successive addition of the offsets implied by the
4154 indices to the base address with infinitely precise arithmetic are not an
4157 that point into the object, plus the address one byte past the end.</p>
4160 the base address with silently-wrapping two's complement arithmetic, and
4162 pointed to by the base pointer. The result value may not necessarily be
4163 used to access memory though, even if it happens to point into allocated
4165 section for more information.</p>
4167 <p>The getelementptr instruction is often confusing. For some more insight into
4171 <pre>
4180 </pre>
4182 </div>
4184 <!-- ======================================================================= -->
4186 </div>
4190 <p>The instructions in this category are the conversion instructions (casting)
4191 which all take a single operand and a type. They perform various bit
4192 conversions on the operand.</p>
4194 </div>
4196 <!-- _______________________________________________________________________ -->
4199 </div>
4203 <pre>
4205 </pre>
4214 size and type of the result, which must be
4217 allowed.</p>
4226 <pre>
4230 </pre>
4232 </div>
4234 <!-- _______________________________________________________________________ -->
4237 </div>
4241 <pre>
4243 </pre>
4264 <pre>
4267 </pre>
4269 </div>
4271 <!-- _______________________________________________________________________ -->
4274 </div>
4278 <pre>
4280 </pre>
4300 <pre>
4303 </pre>
4305 </div>
4307 <!-- _______________________________________________________________________ -->
4310 </div>
4315 <pre>
4317 </pre>
4335 undefined.</p>
4338 <pre>
4341 </pre>
4343 </div>
4345 <!-- _______________________________________________________________________ -->
4348 </div>
4352 <pre>
4354 </pre>
4358 floating point value.</p>
4364 type must be smaller than the destination type.</p>
4374 <pre>
4377 </pre>
4379 </div>
4381 <!-- _______________________________________________________________________ -->
4384 </div>
4388 <pre>
4390 </pre>
4406 towards zero) unsigned integer value. If the value cannot fit
4410 <pre>
4414 </pre>
4416 </div>
4418 <!-- _______________________________________________________________________ -->
4421 </div>
4425 <pre>
4427 </pre>
4445 the results are undefined.</p>
4448 <pre>
4452 </pre>
4454 </div>
4456 <!-- _______________________________________________________________________ -->
4459 </div>
4463 <pre>
4465 </pre>
4480 integer quantity and converts it to the corresponding floating point
4481 value. If the value cannot fit in the floating point value, the results are
4482 undefined.</p>
4485 <pre>
4488 </pre>
4490 </div>
4492 <!-- _______________________________________________________________________ -->
4495 </div>
4499 <pre>
4501 </pre>
4516 quantity and converts it to the corresponding floating point value. If the
4517 value cannot fit in the floating point value, the results are undefined.</p>
4520 <pre>
4523 </pre>
4525 </div>
4527 <!-- _______________________________________________________________________ -->
4530 </div>
4534 <pre>
4536 </pre>
4550 truncating or zero extending that value to the size of the integer type. If
4554 change.</p>
4557 <pre>
4560 </pre>
4562 </div>
4564 <!-- _______________________________________________________________________ -->
4567 </div>
4571 <pre>
4573 </pre>
4581 value to cast, and a type to cast it to, which must be a
4589 than the size of a pointer then a zero extension is done. If they are the
4593 <pre>
4597 </pre>
4599 </div>
4601 <!-- _______________________________________________________________________ -->
4604 </div>
4608 <pre>
4610 </pre>
4618 non-aggregate first class value, and a type to cast it to, which must also be
4621 identical. If the source type is a pointer, the destination type must also be
4622 a pointer. This instruction supports bitwise conversion of vectors to
4623 integers and to vectors of other types (as long as they have the same
4624 size).</p>
4631 be converted to other pointer types with this instruction. To convert
4636 <pre>
4640 </pre>
4642 </div>
4644 <!-- ======================================================================= -->
4650 defy better classification.</p>
4652 </div>
4654 <!-- _______________________________________________________________________ -->
4656 </div>
4661 <pre>
4662 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4663 </pre>
4667 boolean values based on comparison of its two integer, integer vector, or
4668 pointer operands.</p>
4672 the condition code indicating the kind of comparison to perform. It is not a
4673 value, just a keyword. The possible condition code are:</p>
4675 <ol>
4686 </ol>
4690 typed. They must also be identical types.</p>
4696 result, as follows:</p>
4698 <ol>
4701 performed.</li>
4705 performed.</li>
4732 </ol>
4735 values are compared as if they were integers.</p>
4737 <p>If the operands are integer vectors, then they are compared element by
4742 <pre>
4749 </pre>
4751 <p>Note that the code generator does not yet support vector types with
4754 </div>
4756 <!-- _______________________________________________________________________ -->
4758 </div>
4763 <pre>
4764 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4765 </pre>
4769 values based on comparison of its operands.</p>
4771 <p>If the operands are floating point scalars, then the result type is a boolean
4774 <p>If the operands are floating point vectors, then the result type is a vector
4775 of boolean with the same number of elements as the operands being
4776 compared.</p>
4780 the condition code indicating the kind of comparison to perform. It is not a
4781 value, just a keyword. The possible condition code are:</p>
4783 <ol>
4800 </ol>
4808 identical types.</p>
4813 vectors, then the vectors are compared element by element. Each comparison
4815 follows:</p>
4817 <ol>
4861 </ol>
4864 <pre>
4869 </pre>
4871 <p>Note that the code generator does not yet support vector types with
4874 </div>
4876 <!-- _______________________________________________________________________ -->
4879 </div>
4884 <pre>
4886 </pre>
4890 SSA graph representing the function.</p>
4893 <p>The type of the incoming values is specified with the first type field. After
4895 one pair for each predecessor basic block of the current block. Only values
4897 arguments to the PHI node. Only labels may be used as the label
4898 arguments.</p>
4900 <p>There must be no non-phi instructions between the start of a basic block and
4901 the PHI instructions: i.e. PHI instructions must be first in a basic
4902 block.</p>
4904 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
4905 occur on the edge from the corresponding predecessor block to the current
4907 value on the same edge).</p>
4911 specified by the pair corresponding to the predecessor basic block that
4912 executed just prior to the current block.</p>
4915 <pre>
4916 Loop: ; Infinite loop that counts from 0 on up...
4917 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4918 %nextindvar = add i32 %indvar, 1
4919 br label %Loop
4920 </pre>
4922 </div>
4924 <!-- _______________________________________________________________________ -->
4927 </div>
4932 <pre>
4933 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4936 </pre>
4940 condition, without branching.</p>
4945 values indicating the condition, and two values of the
4947 vectors and the condition is a scalar, then entire vectors are selected, not
4948 individual elements.</p>
4952 first value argument; otherwise, it returns the second value argument.</p>
4954 <p>If the condition is a vector of i1, then the value arguments must be vectors
4955 of the same size, and the selection is done element by element.</p>
4958 <pre>
4960 </pre>
4962 <p>Note that the code generator does not yet support conditions
4963 with vector type.</p>
4965 </div>
4967 <!-- _______________________________________________________________________ -->
4970 </div>
4975 <pre>
4976 <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>]
4977 </pre>
4985 <ol>
4987 any allocas or varargs in the caller. If the "tail" marker is present,
4988 the function call is eligible for tail call optimization. Note that calls
4989 may be marked "tail" even if they do not occur before
4993 convention</a> the call should use. If none is specified, the call
4994 defaults to using C calling conventions.</li>
5001 type of the return value. Functions that return no value are marked
5005 being invoked. The argument types must match the types implied by this
5006 signature. This type can be omitted if the function is not varargs and if
5007 the function type does not return a pointer to a function.</li>
5010 be invoked. In most cases, this is a direct function invocation, but
5012 to function value.</li>
5015 signature argument types. All arguments must be of
5017 indicates the function accepts a variable number of arguments, the extra
5018 arguments can be specified.</li>
5023 </ol>
5027 a specified function, with its incoming arguments bound to the specified
5029 function, control flow continues with the instruction after the function
5030 call, and the return value of the function is bound to the result
5031 argument.</p>
5034 <pre>
5035 %retval = call i32 @test(i32 %argc)
5039 call void %foo(i8 97 signext)
5041 %struct.A = type { i32, i8 }
5047 </pre>
5049 <p>llvm treats calls to some functions with names and arguments that match the
5050 standard C99 library as being the C99 library functions, and may perform
5051 optimizations or generate code for them under that assumption. This is
5052 something we'd like to change in the future to provide better support for
5053 freestanding environments and non-C-based langauges.</p>
5055 </div>
5057 <!-- _______________________________________________________________________ -->
5060 </div>
5065 <pre>
5067 </pre>
5071 the "variable argument" area of a function call. It is used to implement the
5076 argument. It returns a value of the specified argument type and increments
5083 to the next argument. For more information, see the variable argument
5086 <p>It is legal for this instruction to be called in a function which does not
5088 function.</p>
5092 argument.</p>
5097 <p>Note that the code generator does not yet fully support va_arg on many
5098 targets. Also, it does not currently support va_arg with aggregate types on
5099 any target.</p>
5101 </div>
5103 <!-- *********************************************************************** -->
5105 <!-- *********************************************************************** -->
5110 well known names and semantics and are required to follow certain
5111 restrictions. Overall, these intrinsics represent an extension mechanism for
5112 the LLVM language that does not require changing all of the transformations
5113 in LLVM when adding to the language (or the bitcode reader/writer, the
5114 parser, etc...).</p>
5117 prefix is reserved in LLVM for intrinsic names; thus, function names may not
5118 begin with this prefix. Intrinsic functions must always be external
5119 functions: you cannot define the body of intrinsic functions. Intrinsic
5120 functions may only be used in call or invoke instructions: it is illegal to
5121 take the address of an intrinsic function. Additionally, because intrinsic
5122 functions are part of the LLVM language, it is required if any are added that
5123 they be documented here.</p>
5125 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
5126 family of functions that perform the same operation but on different data
5127 types. Because LLVM can represent over 8 million different integer types,
5128 overloading is used commonly to allow an intrinsic function to operate on any
5129 integer type. One or more of the argument types or the result type can be
5130 overloaded to accept any integer type. Argument types may also be defined as
5131 exactly matching a previous argument's type or the result type. This allows
5132 an intrinsic function which accepts multiple arguments, but needs all of them
5133 to be of the same type, to only be overloaded with respect to a single
5134 argument or the result.</p>
5136 <p>Overloaded intrinsics will have the names of its overloaded argument types
5137 encoded into its function name, each preceded by a period. Only those types
5138 which are overloaded result in a name suffix. Arguments whose type is matched
5140 can take an integer of any width and returns an integer of exactly the same
5141 integer width. This leads to a family of functions such as
5143 %val)</tt>. Only one type, the return type, is overloaded, and only one type
5144 suffix is required. Because the argument's type is matched against the return
5145 type, it does not require its own name suffix.</p>
5147 <p>To learn how to add an intrinsic function, please see the
5150 </div>
5152 <!-- ======================================================================= -->
5155 </div>
5159 <p>Variable argument support is defined in LLVM with
5161 intrinsic functions. These functions are related to the similarly named
5164 <p>All of these functions operate on arguments that use a target-specific value
5165 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
5166 not define what this type is, so all transformations should be prepared to
5167 handle these functions regardless of the type used.</p>
5170 instruction and the variable argument handling intrinsic functions are
5171 used.</p>
5174 <pre>
5175 define i32 @test(i32 %X, ...) {
5176 ; Initialize variable argument processing
5177 %ap = alloca i8*
5178 %ap2 = bitcast i8** %ap to i8*
5179 call void @llvm.va_start(i8* %ap2)
5181 ; Read a single integer argument
5182 %tmp = va_arg i8** %ap, i32
5184 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5185 %aq = alloca i8*
5186 %aq2 = bitcast i8** %aq to i8*
5187 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5188 call void @llvm.va_end(i8* %aq2)
5190 ; Stop processing of arguments.
5191 call void @llvm.va_end(i8* %ap2)
5192 ret i32 %tmp
5193 }
5195 declare void @llvm.va_start(i8*)
5196 declare void @llvm.va_copy(i8*, i8*)
5197 declare void @llvm.va_end(i8*)
5198 </pre>
5199 </div>
5201 </div>
5203 <!-- _______________________________________________________________________ -->
5206 </div>
5212 <pre>
5214 </pre>
5225 macro available in C. In a target-dependent way, it initializes
5229 need to know the last argument of the function as the compiler can figure
5230 that out.</p>
5232 </div>
5234 <!-- _______________________________________________________________________ -->
5237 </div>
5242 <pre>
5244 </pre>
5248 which has been initialized previously
5257 macro available in C. In a target-dependent way, it destroys
5263 </div>
5265 <!-- _______________________________________________________________________ -->
5268 </div>
5273 <pre>
5275 </pre>
5279 from the source argument list to the destination argument list.</p>
5284 from.</p>
5288 macro available in C. In a target-dependent way, it copies the
5290 element. This intrinsic is necessary because
5292 arbitrarily complex and require, for example, memory allocation.</p>
5294 </div>
5296 <!-- ======================================================================= -->
5299 </div>
5304 Collection</a> (GC) requires the implementation and generation of these
5306 roots on the stack</a>, as well as garbage collector implementations that
5308 barriers. Front-ends for type-safe garbage collected languages should generate
5309 these intrinsics to make use of the LLVM garbage collectors. For more details,
5313 <p>The garbage collection intrinsics only operate on objects in the generic
5314 address space (address space zero).</p>
5316 </div>
5318 <!-- _______________________________________________________________________ -->
5321 </div>
5326 <pre>
5327 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5328 </pre>
5332 the code generator, and allows some metadata to be associated with it.</p>
5335 <p>The first argument specifies the address of a stack object that contains the
5336 root pointer. The second pointer (which must be either a constant or a
5337 global value address) contains the meta-data to be associated with the
5338 root.</p>
5342 location. At compile-time, the code generator generates information to allow
5347 </div>
5349 <!-- _______________________________________________________________________ -->
5352 </div>
5357 <pre>
5358 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5359 </pre>
5363 locations, allowing garbage collector implementations that require read
5364 barriers.</p>
5367 <p>The second argument is the address to read from, which should be an address
5368 allocated from the garbage collector. The first object is a pointer to the
5369 start of the referenced object, if needed by the language runtime (otherwise
5370 null).</p>
5374 instruction, but may be replaced with substantially more complex code by the
5379 </div>
5381 <!-- _______________________________________________________________________ -->
5384 </div>
5389 <pre>
5390 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5391 </pre>
5395 locations, allowing garbage collector implementations that require write
5396 barriers (such as generational or reference counting collectors).</p>
5399 <p>The first argument is the reference to store, the second is the start of the
5400 object to store it to, and the third is the address of the field of Obj to
5401 store to. If the runtime does not require a pointer to the object, Obj may
5402 be null.</p>
5406 instruction, but may be replaced with substantially more complex code by the
5411 </div>
5413 <!-- ======================================================================= -->
5416 </div>
5420 <p>These intrinsics are provided by LLVM to expose special features that may
5421 only be implemented with code generator support.</p>
5423 </div>
5425 <!-- _______________________________________________________________________ -->
5428 </div>
5433 <pre>
5435 </pre>
5439 target-specific value indicating the return address of the current function
5440 or one of its callers.</p>
5443 <p>The argument to this intrinsic indicates which function to return the address
5444 for. Zero indicates the calling function, one indicates its caller, etc.
5449 indicating the return address of the specified call frame, or zero if it
5450 cannot be identified. The value returned by this intrinsic is likely to be
5451 incorrect or 0 for arguments other than zero, so it should only be used for
5452 debugging purposes.</p>
5454 <p>Note that calling this intrinsic does not prevent function inlining or other
5455 aggressive transformations, so the value returned may not be that of the
5456 obvious source-language caller.</p>
5458 </div>
5460 <!-- _______________________________________________________________________ -->
5463 </div>
5468 <pre>
5470 </pre>
5474 target-specific frame pointer value for the specified stack frame.</p>
5477 <p>The argument to this intrinsic indicates which function to return the frame
5478 pointer for. Zero indicates the calling function, one indicates its caller,
5483 indicating the frame address of the specified call frame, or zero if it
5484 cannot be identified. The value returned by this intrinsic is likely to be
5485 incorrect or 0 for arguments other than zero, so it should only be used for
5486 debugging purposes.</p>
5488 <p>Note that calling this intrinsic does not prevent function inlining or other
5489 aggressive transformations, so the value returned may not be that of the
5490 obvious source-language caller.</p>
5492 </div>
5494 <!-- _______________________________________________________________________ -->
5497 </div>
5502 <pre>
5503 declare i8 *@llvm.stacksave()
5504 </pre>
5508 of the function stack, for use
5510 useful for implementing language features like scoped automatic variable
5511 sized arrays in C99.</p>
5514 <p>This intrinsic returns a opaque pointer value that can be passed
5522 </div>
5524 <!-- _______________________________________________________________________ -->
5527 </div>
5532 <pre>
5533 declare void @llvm.stackrestore(i8 * %ptr)
5534 </pre>
5538 the function stack to the state it was in when the
5540 executed. This is useful for implementing language features like scoped
5541 automatic variable sized arrays in C99.</p>
5544 <p>See the description
5547 </div>
5549 <!-- _______________________________________________________________________ -->
5552 </div>
5557 <pre>
5559 </pre>
5563 insert a prefetch instruction if supported; otherwise, it is a noop.
5564 Prefetches have no effect on the behavior of the program but can change its
5565 performance characteristics.</p>
5575 <p>This intrinsic does not modify the behavior of the program. In particular,
5576 prefetches cannot trap and do not produce a value. On targets that support
5577 this intrinsic, the prefetch can provide hints to the processor cache for
5578 better performance.</p>
5580 </div>
5582 <!-- _______________________________________________________________________ -->
5585 </div>
5590 <pre>
5592 </pre>
5596 Counter (PC) in a region of code to simulators and other tools. The method
5597 is target specific, but it is expected that the marker will use exported
5598 symbols to transmit the PC of the marker. The marker makes no guarantees
5599 that it will remain with any specific instruction after optimizations. It is
5600 possible that the presence of a marker will inhibit optimizations. The
5601 intended use is to be inserted after optimizations to allow correlations of
5602 simulation runs.</p>
5608 <p>This intrinsic does not modify the behavior of the program. Backends that do
5609 not support this intrinisic may ignore it.</p>
5611 </div>
5613 <!-- _______________________________________________________________________ -->
5616 </div>
5621 <pre>
5622 declare i64 @llvm.readcyclecounter( )
5623 </pre>
5627 counter register (or similar low latency, high accuracy clocks) on those
5628 targets that support it. On X86, it should map to RDTSC. On Alpha, it
5629 should map to RPCC. As the backing counters overflow quickly (on the order
5633 <p>When directly supported, reading the cycle counter should not modify any
5634 memory. Implementations are allowed to either return a application specific
5635 value or a system wide value. On backends without support, this is lowered
5638 </div>
5640 <!-- ======================================================================= -->
5643 </div>
5647 <p>LLVM provides intrinsics for a few important standard C library functions.
5648 These intrinsics allow source-language front-ends to pass information about
5649 the alignment of the pointer arguments to the code generator, providing
5650 opportunity for more efficient code generation.</p>
5652 </div>
5654 <!-- _______________________________________________________________________ -->
5657 </div>
5663 integer bit width. Not all targets support all bit widths however.</p>
5665 <pre>
5674 </pre>
5678 source location to the destination location.</p>
5681 intrinsics do not return a value, and takes an extra alignment argument.</p>
5684 <p>The first argument is a pointer to the destination, the second is a pointer
5685 to the source. The third argument is an integer argument specifying the
5686 number of bytes to copy, and the fourth argument is the alignment of the
5687 source and destination locations.</p>
5690 then the caller guarantees that both the source and destination pointers are
5691 aligned to that boundary.</p>
5695 source location to the destination location, which are not allowed to
5696 overlap. It copies "len" bytes of memory over. If the argument is known to
5697 be aligned to some boundary, this can be specified as the fourth argument,
5700 </div>
5702 <!-- _______________________________________________________________________ -->
5705 </div>
5710 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5711 width. Not all targets support all bit widths however.</p>
5713 <pre>
5722 </pre>
5726 source location to the destination location. It is similar to the
5728 overlap.</p>
5731 intrinsics do not return a value, and takes an extra alignment argument.</p>
5734 <p>The first argument is a pointer to the destination, the second is a pointer
5735 to the source. The third argument is an integer argument specifying the
5736 number of bytes to copy, and the fourth argument is the alignment of the
5737 source and destination locations.</p>
5740 then the caller guarantees that the source and destination pointers are
5741 aligned to that boundary.</p>
5745 source location to the destination location, which may overlap. It copies
5746 "len" bytes of memory over. If the argument is known to be aligned to some
5747 boundary, this can be specified as the fourth argument, otherwise it should
5750 </div>
5752 <!-- _______________________________________________________________________ -->
5755 </div>
5760 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5761 width. Not all targets support all bit widths however.</p>
5763 <pre>
5772 </pre>
5776 particular byte value.</p>
5779 intrinsic does not return a value, and takes an extra alignment argument.</p>
5782 <p>The first argument is a pointer to the destination to fill, the second is the
5783 byte value to fill it with, the third argument is an integer argument
5784 specifying the number of bytes to fill, and the fourth argument is the known
5785 alignment of destination location.</p>
5788 then the caller guarantees that the destination pointer is aligned to that
5789 boundary.</p>
5793 at the destination location. If the argument is known to be aligned to some
5794 boundary, this can be specified as the fourth argument, otherwise it should
5797 </div>
5799 <!-- _______________________________________________________________________ -->
5802 </div>
5808 floating point or vector of floating point type. Not all targets support all
5809 types however.</p>
5811 <pre>
5812 declare float @llvm.sqrt.f32(float %Val)
5813 declare double @llvm.sqrt.f64(double %Val)
5814 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5815 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5816 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5817 </pre>
5823 behavior for negative numbers other than -0.0 (which allows for better
5824 optimization, because there is no need to worry about errno being
5828 <p>The argument and return value are floating point numbers of the same
5829 type.</p>
5832 <p>This function returns the sqrt of the specified operand if it is a
5833 nonnegative floating point number.</p>
5835 </div>
5837 <!-- _______________________________________________________________________ -->
5840 </div>
5846 floating point or vector of floating point type. Not all targets support all
5847 types however.</p>
5849 <pre>
5850 declare float @llvm.powi.f32(float %Val, i32 %power)
5851 declare double @llvm.powi.f64(double %Val, i32 %power)
5852 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5853 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5854 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5855 </pre>
5859 specified (positive or negative) power. The order of evaluation of
5860 multiplications is not defined. When a vector of floating point type is
5861 used, the second argument remains a scalar integer value.</p>
5864 <p>The second argument is an integer power, and the first is a value to raise to
5865 that power.</p>
5868 <p>This function returns the first value raised to the second power with an
5869 unspecified sequence of rounding operations.</p>
5871 </div>
5873 <!-- _______________________________________________________________________ -->
5876 </div>
5882 floating point or vector of floating point type. Not all targets support all
5883 types however.</p>
5885 <pre>
5886 declare float @llvm.sin.f32(float %Val)
5887 declare double @llvm.sin.f64(double %Val)
5888 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5889 declare fp128 @llvm.sin.f128(fp128 %Val)
5890 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5891 </pre>
5897 <p>The argument and return value are floating point numbers of the same
5898 type.</p>
5901 <p>This function returns the sine of the specified operand, returning the same
5903 in the same way.</p>
5905 </div>
5907 <!-- _______________________________________________________________________ -->
5910 </div>
5916 floating point or vector of floating point type. Not all targets support all
5917 types however.</p>
5919 <pre>
5920 declare float @llvm.cos.f32(float %Val)
5921 declare double @llvm.cos.f64(double %Val)
5922 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5923 declare fp128 @llvm.cos.f128(fp128 %Val)
5924 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5925 </pre>
5931 <p>The argument and return value are floating point numbers of the same
5932 type.</p>
5935 <p>This function returns the cosine of the specified operand, returning the same
5937 in the same way.</p>
5939 </div>
5941 <!-- _______________________________________________________________________ -->
5944 </div>
5950 floating point or vector of floating point type. Not all targets support all
5951 types however.</p>
5953 <pre>
5954 declare float @llvm.pow.f32(float %Val, float %Power)
5955 declare double @llvm.pow.f64(double %Val, double %Power)
5956 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5957 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5958 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5959 </pre>
5963 specified (positive or negative) power.</p>
5966 <p>The second argument is a floating point power, and the first is a value to
5967 raise to that power.</p>
5970 <p>This function returns the first value raised to the second power, returning
5972 conditions in the same way.</p>
5974 </div>
5976 <!-- ======================================================================= -->
5979 </div>
5983 <p>LLVM provides intrinsics for a few important bit manipulation operations.
5984 These allow efficient code generation for some algorithms.</p>
5986 </div>
5988 <!-- _______________________________________________________________________ -->
5991 </div>
5996 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5999 <pre>
6003 </pre>
6007 values with an even number of bytes (positive multiple of 16 bits). These
6008 are useful for performing operations on data that is not in the target's
6009 native byte order.</p>
6013 and low byte of the input i16 swapped. Similarly,
6019 more, respectively).</p>
6021 </div>
6023 <!-- _______________________________________________________________________ -->
6026 </div>
6031 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
6032 width. Not all targets support all bit widths however.</p>
6034 <pre>
6040 </pre>
6044 in a value.</p>
6047 <p>The only argument is the value to be counted. The argument may be of any
6048 integer type. The return type must match the argument type.</p>
6053 </div>
6055 <!-- _______________________________________________________________________ -->
6058 </div>
6064 integer bit width. Not all targets support all bit widths however.</p>
6066 <pre>
6072 </pre>
6076 leading zeros in a variable.</p>
6079 <p>The only argument is the value to be counted. The argument may be of any
6080 integer type. The return type must match the argument type.</p>
6084 zeros in a variable. If the src == 0 then the result is the size in bits of
6087 </div>
6089 <!-- _______________________________________________________________________ -->
6092 </div>
6098 integer bit width. Not all targets support all bit widths however.</p>
6100 <pre>
6106 </pre>
6110 trailing zeros.</p>
6113 <p>The only argument is the value to be counted. The argument may be of any
6114 integer type. The return type must match the argument type.</p>
6118 zeros in a variable. If the src == 0 then the result is the size in bits of
6121 </div>
6123 <!-- ======================================================================= -->
6126 </div>
6132 </div>
6134 <!-- _______________________________________________________________________ -->
6137 </div>
6143 on any integer bit width.</p>
6145 <pre>
6146 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6147 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6148 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6149 </pre>
6153 a signed addition of the two arguments, and indicate whether an overflow
6154 occurred during the signed summation.</p>
6157 <p>The arguments (%a and %b) and the first element of the result structure may
6158 be of integer types of any bit width, but they must have the same bit
6159 width. The second element of the result structure must be of
6161 undergo signed addition.</p>
6165 a signed addition of the two variables. They return a structure — the
6166 first element of which is the signed summation, and the second element of
6167 which is a bit specifying if the signed summation resulted in an
6168 overflow.</p>
6171 <pre>
6172 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6173 %sum = extractvalue {i32, i1} %res, 0
6174 %obit = extractvalue {i32, i1} %res, 1
6175 br i1 %obit, label %overflow, label %normal
6176 </pre>
6178 </div>
6180 <!-- _______________________________________________________________________ -->
6183 </div>
6189 on any integer bit width.</p>
6191 <pre>
6192 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6193 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6194 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6195 </pre>
6199 an unsigned addition of the two arguments, and indicate whether a carry
6200 occurred during the unsigned summation.</p>
6203 <p>The arguments (%a and %b) and the first element of the result structure may
6204 be of integer types of any bit width, but they must have the same bit
6205 width. The second element of the result structure must be of
6207 undergo unsigned addition.</p>
6211 an unsigned addition of the two arguments. They return a structure —
6212 the first element of which is the sum, and the second element of which is a
6213 bit specifying if the unsigned summation resulted in a carry.</p>
6216 <pre>
6217 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6218 %sum = extractvalue {i32, i1} %res, 0
6219 %obit = extractvalue {i32, i1} %res, 1
6220 br i1 %obit, label %carry, label %normal
6221 </pre>
6223 </div>
6225 <!-- _______________________________________________________________________ -->
6228 </div>
6234 on any integer bit width.</p>
6236 <pre>
6237 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6238 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6239 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6240 </pre>
6244 a signed subtraction of the two arguments, and indicate whether an overflow
6245 occurred during the signed subtraction.</p>
6248 <p>The arguments (%a and %b) and the first element of the result structure may
6249 be of integer types of any bit width, but they must have the same bit
6250 width. The second element of the result structure must be of
6252 undergo signed subtraction.</p>
6256 a signed subtraction of the two arguments. They return a structure —
6257 the first element of which is the subtraction, and the second element of
6258 which is a bit specifying if the signed subtraction resulted in an
6259 overflow.</p>
6262 <pre>
6263 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6264 %sum = extractvalue {i32, i1} %res, 0
6265 %obit = extractvalue {i32, i1} %res, 1
6266 br i1 %obit, label %overflow, label %normal
6267 </pre>
6269 </div>
6271 <!-- _______________________________________________________________________ -->
6274 </div>
6280 on any integer bit width.</p>
6282 <pre>
6283 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6284 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6285 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6286 </pre>
6290 an unsigned subtraction of the two arguments, and indicate whether an
6291 overflow occurred during the unsigned subtraction.</p>
6294 <p>The arguments (%a and %b) and the first element of the result structure may
6295 be of integer types of any bit width, but they must have the same bit
6296 width. The second element of the result structure must be of
6298 undergo unsigned subtraction.</p>
6302 an unsigned subtraction of the two arguments. They return a structure —
6303 the first element of which is the subtraction, and the second element of
6304 which is a bit specifying if the unsigned subtraction resulted in an
6305 overflow.</p>
6308 <pre>
6309 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6310 %sum = extractvalue {i32, i1} %res, 0
6311 %obit = extractvalue {i32, i1} %res, 1
6312 br i1 %obit, label %overflow, label %normal
6313 </pre>
6315 </div>
6317 <!-- _______________________________________________________________________ -->
6320 </div>
6326 on any integer bit width.</p>
6328 <pre>
6329 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6330 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6331 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6332 </pre>
6337 a signed multiplication of the two arguments, and indicate whether an
6338 overflow occurred during the signed multiplication.</p>
6341 <p>The arguments (%a and %b) and the first element of the result structure may
6342 be of integer types of any bit width, but they must have the same bit
6343 width. The second element of the result structure must be of
6345 undergo signed multiplication.</p>
6349 a signed multiplication of the two arguments. They return a structure —
6350 the first element of which is the multiplication, and the second element of
6351 which is a bit specifying if the signed multiplication resulted in an
6352 overflow.</p>
6355 <pre>
6356 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6357 %sum = extractvalue {i32, i1} %res, 0
6358 %obit = extractvalue {i32, i1} %res, 1
6359 br i1 %obit, label %overflow, label %normal
6360 </pre>
6362 </div>
6364 <!-- _______________________________________________________________________ -->
6367 </div>
6373 on any integer bit width.</p>
6375 <pre>
6376 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6377 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6378 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6379 </pre>
6383 a unsigned multiplication of the two arguments, and indicate whether an
6384 overflow occurred during the unsigned multiplication.</p>
6387 <p>The arguments (%a and %b) and the first element of the result structure may
6388 be of integer types of any bit width, but they must have the same bit
6389 width. The second element of the result structure must be of
6391 undergo unsigned multiplication.</p>
6395 an unsigned multiplication of the two arguments. They return a structure
6396 — the first element of which is the multiplication, and the second
6397 element of which is a bit specifying if the unsigned multiplication resulted
6398 in an overflow.</p>
6401 <pre>
6402 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6403 %sum = extractvalue {i32, i1} %res, 0
6404 %obit = extractvalue {i32, i1} %res, 1
6405 br i1 %obit, label %overflow, label %normal
6406 </pre>
6408 </div>
6410 <!-- ======================================================================= -->
6413 </div>
6418 prefix), are described in
6422 </div>
6424 <!-- ======================================================================= -->
6427 </div>
6431 <p>The LLVM exception handling intrinsics (which all start with
6436 </div>
6438 <!-- ======================================================================= -->
6441 </div>
6445 <p>This intrinsic makes it possible to excise one parameter, marked with
6447 function pointer lacking the nest parameter - the caller does not need to
6448 provide a value for it. Instead, the value to use is stored in advance in a
6449 "trampoline", a block of memory usually allocated on the stack, which also
6450 contains code to splice the nest value into the argument list. This is used
6451 to implement the GCC nested function address extension.</p>
6453 <p>For example, if the function is
6456 follows:</p>
6459 <pre>
6462 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6463 %fp = bitcast i8* %p to i32 (i32, i32)*
6464 </pre>
6465 </div>
6470 </div>
6472 <!-- _______________________________________________________________________ -->
6475 </div>
6480 <pre>
6482 </pre>
6486 function pointer suitable for executing it.</p>
6491 sufficiently aligned block of memory; this memory is written to by the
6492 intrinsic. Note that the size and the alignment are target-specific - LLVM
6493 currently provides no portable way of determining them, so a front-end that
6494 generates this intrinsic needs to have some target-specific knowledge.
6500 dependent code, turning it into a function. A pointer to this function is
6502 function pointer type</a> before being called. The new function's signature
6505 is allowed, and it must be of pointer type. Calling the new function is
6510 returned function pointer is undefined.</p>
6512 </div>
6514 <!-- ======================================================================= -->
6517 </div>
6522 hardware constructs for atomic operations and memory synchronization. This
6523 provides an interface to the hardware, not an interface to the programmer. It
6524 is aimed at a low enough level to allow any programming models or APIs
6525 (Application Programming Interfaces) which need atomic behaviors to map
6526 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
6527 hardware provides a "universal IR" for source languages, it also provides a
6528 starting point for developing a "universal" atomic operation and
6529 synchronization IR.</p>
6532 software transaction memory systems, atomic primitives, and intrinsic
6533 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6534 application libraries. The hardware interface provided by LLVM should allow
6535 a clean implementation of all of these APIs and parallel programming models.
6536 No one model or paradigm should be selected above others unless the hardware
6537 itself ubiquitously does so.</p>
6539 </div>
6541 <!-- _______________________________________________________________________ -->
6544 </div>
6547 <pre>
6548 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device> )
6549 </pre>
6553 specific pairs of memory access types.</p>
6557 The first four arguments enables a specific barrier as listed below. The
6558 fith argument specifies that the barrier applies to io or device or uncached
6559 memory.</p>
6561 <ul>
6567 </ul>
6570 <p>This intrinsic causes the system to enforce some ordering constraints upon
6571 the loads and stores of the program. This barrier does not
6574 and store operations (f.ex. load-load, or store-load), all of the first
6575 operations preceding the barrier will complete before any of the second
6576 operations succeeding the barrier begin. Specifically the semantics for each
6577 pairing is as follows:</p>
6579 <ul>
6581 after the barrier begins.</li>
6583 store after the barrier begins.</li>
6585 store after the barrier begins.</li>
6587 load after the barrier begins.</li>
6588 </ul>
6591 is enabled in a single memory barrier intrinsic.</p>
6593 <p>Backends may implement stronger barriers than those requested when they do
6594 not support as fine grained a barrier as requested. Some architectures do
6595 not need all types of barriers and on such architectures, these become
6596 noops.</p>
6599 <pre>
6600 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6601 %ptr = bitcast i8* %mallocP to i32*
6602 store i32 4, %ptr
6605 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6608 </pre>
6610 </div>
6612 <!-- _______________________________________________________________________ -->
6615 </div>
6621 any integer bit width and for different address spaces. Not all targets
6622 support all bit widths however.</p>
6624 <pre>
6626 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6627 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6628 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6629 </pre>
6632 <p>This loads a value in memory and compares it to a given value. If they are
6633 equal, it stores a new value into the memory.</p>
6639 this integer type. While any bit width integer may be used, targets may only
6640 lower representations they support in hardware.</p>
6643 <p>This entire intrinsic must be executed atomically. It first loads the value
6646 memory. The loaded value is yielded in all cases. This provides the
6647 equivalent of an atomic compare-and-swap operation within the SSA
6648 framework.</p>
6651 <pre>
6652 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6653 %ptr = bitcast i8* %mallocP to i32*
6654 store i32 4, %ptr
6657 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6663 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6668 </pre>
6670 </div>
6672 <!-- _______________________________________________________________________ -->
6675 </div>
6680 integer bit width. Not all targets support all bit widths however.</p>
6682 <pre>
6687 </pre>
6698 integer type. The targets may only lower integer representations they
6699 support.</p>
6704 equivalent of an atomic swap operation within the SSA framework.</p>
6707 <pre>
6708 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6709 %ptr = bitcast i8* %mallocP to i32*
6710 store i32 4, %ptr
6713 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6719 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6724 </pre>
6726 </div>
6728 <!-- _______________________________________________________________________ -->
6732 </div>
6738 any integer bit width. Not all targets support all bit widths however.</p>
6740 <pre>
6745 </pre>
6752 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6753 and the second an integer value. The result is also an integer value. These
6754 integer types can have any bit width, but they must all have the same bit
6755 width. The targets may only lower integer representations they support.</p>
6758 <p>This intrinsic does a series of operations atomically. It first loads the
6763 <pre>
6764 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6765 %ptr = bitcast i8* %mallocP to i32*
6766 store i32 4, %ptr
6767 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6769 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6771 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6774 </pre>
6776 </div>
6778 <!-- _______________________________________________________________________ -->
6782 </div>
6788 any integer bit width and for different address spaces. Not all targets
6789 support all bit widths however.</p>
6791 <pre>
6796 </pre>
6803 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6804 and the second an integer value. The result is also an integer value. These
6805 integer types can have any bit width, but they must all have the same bit
6806 width. The targets may only lower integer representations they support.</p>
6809 <p>This intrinsic does a series of operations atomically. It first loads the
6815 <pre>
6816 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6817 %ptr = bitcast i8* %mallocP to i32*
6818 store i32 8, %ptr
6819 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6821 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6823 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6826 </pre>
6828 </div>
6830 <!-- _______________________________________________________________________ -->
6836 </div>
6841 <p>These are overloaded intrinsics. You can
6844 bit width and for different address spaces. Not all targets support all bit
6845 widths however.</p>
6847 <pre>
6852 </pre>
6854 <pre>
6859 </pre>
6861 <pre>
6866 </pre>
6868 <pre>
6873 </pre>
6881 <p>These intrinsics take two arguments, the first a pointer to an integer value
6882 and the second an integer value. The result is also an integer value. These
6883 integer types can have any bit width, but they must all have the same bit
6884 width. The targets may only lower integer representations they support.</p>
6887 <p>These intrinsics does a series of operations atomically. They first load the
6893 <pre>
6894 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6895 %ptr = bitcast i8* %mallocP to i32*
6896 store i32 0x0F0F, %ptr
6897 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6899 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6901 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6903 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6906 </pre>
6908 </div>
6910 <!-- _______________________________________________________________________ -->
6916 </div>
6924 address spaces. Not all targets support all bit widths however.</p>
6926 <pre>
6931 </pre>
6933 <pre>
6938 </pre>
6940 <pre>
6945 </pre>
6947 <pre>
6952 </pre>
6955 <p>These intrinsics takes the signed or unsigned minimum or maximum of
6960 <p>These intrinsics take two arguments, the first a pointer to an integer value
6961 and the second an integer value. The result is also an integer value. These
6962 integer types can have any bit width, but they must all have the same bit
6963 width. The targets may only lower integer representations they support.</p>
6966 <p>These intrinsics does a series of operations atomically. They first load the
6972 <pre>
6973 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6974 %ptr = bitcast i8* %mallocP to i32*
6975 store i32 7, %ptr
6976 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6978 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6980 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6982 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6985 </pre>
6987 </div>
6990 <!-- ======================================================================= -->
6993 </div>
6997 <p>This class of intrinsics exists to information about the lifetime of memory
6998 objects and ranges where variables are immutable.</p>
7000 </div>
7002 <!-- _______________________________________________________________________ -->
7005 </div>
7010 <pre>
7012 </pre>
7016 object's lifetime.</p>
7019 <p>The first argument is a constant integer representing the size of the
7020 object, or -1 if it is variable sized. The second argument is a pointer to
7021 the object.</p>
7024 <p>This intrinsic indicates that before this point in the code, the value of the
7026 never be used and has an undefined value. A load from the pointer that
7027 precedes this intrinsic can be replaced with
7030 </div>
7032 <!-- _______________________________________________________________________ -->
7035 </div>
7040 <pre>
7042 </pre>
7046 object's lifetime.</p>
7049 <p>The first argument is a constant integer representing the size of the
7050 object, or -1 if it is variable sized. The second argument is a pointer to
7051 the object.</p>
7054 <p>This intrinsic indicates that after this point in the code, the value of the
7056 never be used and has an undefined value. Any stores into the memory object
7057 following this intrinsic may be removed as dead.
7059 </div>
7061 <!-- _______________________________________________________________________ -->
7064 </div>
7069 <pre>
7071 </pre>
7075 a memory object will not change.</p>
7078 <p>The first argument is a constant integer representing the size of the
7079 object, or -1 if it is variable sized. The second argument is a pointer to
7080 the object.</p>
7084 the return value, the referenced memory location is constant and
7085 unchanging.</p>
7087 </div>
7089 <!-- _______________________________________________________________________ -->
7092 </div>
7097 <pre>
7098 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
7099 </pre>
7103 a memory object are mutable.</p>
7107 The second argument is a constant integer representing the size of the
7108 object, or -1 if it is variable sized and the third argument is a pointer
7109 to the object.</p>
7114 </div>
7116 <!-- ======================================================================= -->
7119 </div>
7123 <p>This class of intrinsics is designed to be generic and has no specific
7124 purpose.</p>
7126 </div>
7128 <!-- _______________________________________________________________________ -->
7131 </div>
7136 <pre>
7137 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
7138 </pre>
7144 <p>The first argument is a pointer to a value, the second is a pointer to a
7145 global string, the third is a pointer to a global string which is the source
7146 file name, and the last argument is the line number.</p>
7149 <p>This intrinsic allows annotation of local variables with arbitrary strings.
7150 This can be useful for special purpose optimizations that want to look for
7151 these annotations. These have no other defined use, they are ignored by code
7152 generation and optimization.</p>
7154 </div>
7156 <!-- _______________________________________________________________________ -->
7159 </div>
7165 any integer bit width.</p>
7167 <pre>
7168 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
7169 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
7170 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
7171 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
7172 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
7173 </pre>
7179 <p>The first argument is an integer value (result of some expression), the
7180 second is a pointer to a global string, the third is a pointer to a global
7181 string which is the source file name, and the last argument is the line
7182 number. It returns the value of the first argument.</p>
7185 <p>This intrinsic allows annotations to be put on arbitrary expressions with
7186 arbitrary strings. This can be useful for special purpose optimizations that
7187 want to look for these annotations. These have no other defined use, they
7188 are ignored by code generation and optimization.</p>
7190 </div>
7192 <!-- _______________________________________________________________________ -->
7195 </div>
7200 <pre>
7201 declare void @llvm.trap()
7202 </pre>
7211 <p>This intrinsics is lowered to the target dependent trap instruction. If the
7212 target does not have a trap instruction, this intrinsic will be lowered to
7215 </div>
7217 <!-- _______________________________________________________________________ -->
7220 </div>
7225 <pre>
7227 </pre>
7232 ensure that it is placed on the stack before local variables.</p>
7236 arguments. The first argument is the value loaded from the stack
7238 that has enough space to hold the value of the guard.</p>
7241 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
7243 stack. This is to ensure that if a local variable on the stack is
7244 overwritten, it will destroy the value of the guard. When the function exits,
7245 the guard on the stack is checked against the original guard. If they're
7247 function.</p>
7249 </div>
7251 <!-- _______________________________________________________________________ -->
7254 </div>
7259 <pre>
7262 </pre>
7266 to the optimizers to either discover at compile time either a) when an
7267 operation like memcpy will either overflow a buffer that corresponds to
7268 an object, or b) to determine that a runtime check for overflow isn't
7269 necessary. An object in this context means an allocation of a
7276 the type corresponds to a return value based on whole objects,
7277 and the second bit whether or not we return the maximum or minimum
7278 remaining bytes computed.</p>
7283 </tr>
7287 </tr>
7291 </tr>
7295 </tr>
7296 </table>
7302 at compile time.</p>
7304 </div>
7306 <!-- *********************************************************************** -->
7307 <hr>
7308 <address>
7316 Last modified: $Date$
7317 </address>
7319 </body>
7320 </html>