| blob | 44ef03b4efee24a1da4d066b2799c4f5ad91e7ab |
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>
63 </ol>
64 </li>
66 <ol>
75 </ol>
76 </li>
78 </ol>
79 </li>
81 <ol>
88 </ol>
89 </li>
91 <ol>
93 </ol>
94 </li>
96 <ol>
99 Global Variable</a></li>
101 Global Variable</a></li>
103 Global Variable</a></li>
104 </ol>
105 </li>
107 <ol>
109 <ol>
116 </ol>
117 </li>
119 <ol>
132 </ol>
133 </li>
135 <ol>
142 </ol>
143 </li>
145 <ol>
149 </ol>
150 </li>
152 <ol>
155 </ol>
156 </li>
158 <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>
286 </ol>
287 </li>
288 </ol>
289 </li>
290 </ol>
295 </div>
297 <!-- *********************************************************************** -->
299 <!-- *********************************************************************** -->
303 <p>This document is a reference manual for the LLVM assembly language. LLVM is
304 a Static Single Assignment (SSA) based representation that provides type
305 safety, low-level operations, flexibility, and the capability of representing
306 'all' high-level languages cleanly. It is the common code representation
307 used throughout all phases of the LLVM compilation strategy.</p>
309 </div>
311 <!-- *********************************************************************** -->
313 <!-- *********************************************************************** -->
317 <p>The LLVM code representation is designed to be used in three different forms:
318 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
319 for fast loading by a Just-In-Time compiler), and as a human readable
320 assembly language representation. This allows LLVM to provide a powerful
321 intermediate representation for efficient compiler transformations and
322 analysis, while providing a natural means to debug and visualize the
323 transformations. The three different forms of LLVM are all equivalent. This
324 document describes the human readable representation and notation.</p>
326 <p>The LLVM representation aims to be light-weight and low-level while being
327 expressive, typed, and extensible at the same time. It aims to be a
328 "universal IR" of sorts, by being at a low enough level that high-level ideas
329 may be cleanly mapped to it (similar to how microprocessors are "universal
330 IR's", allowing many source languages to be mapped to them). By providing
331 type information, LLVM can be used as the target of optimizations: for
332 example, through pointer analysis, it can be proven that a C automatic
333 variable is never accessed outside of the current function... allowing it to
334 be promoted to a simple SSA value instead of a memory location.</p>
336 </div>
338 <!-- _______________________________________________________________________ -->
343 <p>It is important to note that this document describes 'well formed' LLVM
344 assembly language. There is a difference between what the parser accepts and
345 what is considered 'well formed'. For example, the following instruction is
346 syntactically okay, but not well formed:</p>
349 <pre>
351 </pre>
352 </div>
355 uses. The LLVM infrastructure provides a verification pass that may be used
356 to verify that an LLVM module is well formed. This pass is automatically run
357 by the parser after parsing input assembly and by the optimizer before it
358 outputs bitcode. The violations pointed out by the verifier pass indicate
359 bugs in transformation passes or input to the parser.</p>
361 </div>
363 <!-- Describe the typesetting conventions here. -->
365 <!-- *********************************************************************** -->
367 <!-- *********************************************************************** -->
371 <p>LLVM identifiers come in two basic types: global and local. Global
373 character. Local identifiers (register names, types) begin with
375 for identifiers, for different purposes:</p>
377 <ol>
378 <li>Named values are represented as a string of characters with their prefix.
382 other characters in their names can be surrounded with quotes. Special
384 ASCII code for the character in hexadecimal. In this way, any character
385 can be used in a name value, even quotes themselves.</li>
387 <li>Unnamed values are represented as an unsigned numeric value with their
392 </ol>
394 <p>LLVM requires that values start with a prefix for two reasons: Compilers
395 don't need to worry about name clashes with reserved words, and the set of
396 reserved words may be expanded in the future without penalty. Additionally,
397 unnamed identifiers allow a compiler to quickly come up with a temporary
398 variable without having to avoid symbol table conflicts.</p>
400 <p>Reserved words in LLVM are very similar to reserved words in other
401 languages. There are keywords for different opcodes
407 reserved words cannot conflict with variable names, because none of them
410 <p>Here is an example of LLVM code to multiply the integer variable
416 <pre>
418 </pre>
419 </div>
424 <pre>
426 </pre>
427 </div>
432 <pre>
436 </pre>
437 </div>
440 lexical features of LLVM:</p>
442 <ol>
444 line.</li>
446 <li>Unnamed temporaries are created when the result of a computation is not
447 assigned to a named value.</li>
450 </ol>
452 <p>...and it also shows a convention that we follow in this document. When
453 demonstrating instructions, we will follow an instruction with a comment that
454 defines the type and name of value produced. Comments are shown in italic
455 text.</p>
457 </div>
459 <!-- *********************************************************************** -->
461 <!-- *********************************************************************** -->
463 <!-- ======================================================================= -->
465 </div>
470 of the input programs. Each module consists of functions, global variables,
471 and symbol table entries. Modules may be combined together with the LLVM
472 linker, which merges function (and global variable) definitions, resolves
473 forward declarations, and merges symbol table entries. Here is an example of
479 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
487 %cast210 = <a
491 <a
493 <a
495 </pre>
496 </div>
503 <p>In general, a module is made up of a list of global values, where both
504 functions and global variables are global values. Global values are
505 represented by a pointer to a memory location (in this case, a pointer to an
506 array of char, and a pointer to a function), and have one of the
509 </div>
511 <!-- ======================================================================= -->
514 </div>
518 <p>All Global Variables and Functions have one of the following types of
519 linkage:</p>
521 <dl>
523 <dd>Global values with private linkage are only directly accessible by objects
524 in the current module. In particular, linking code into a module with an
525 private global value may cause the private to be renamed as necessary to
526 avoid collisions. Because the symbol is private to the module, all
527 references can be updated. This doesn't show up in any symbol table in the
528 object file.</dd>
531 <dd>Similar to private, but the symbol is passed through the assembler and
532 removed by the linker after evaluation. Note that (unlike private
533 symbols) linker_private symbols are subject to coalescing by the linker:
534 weak symbols get merged and redefinitions are rejected. However, unlike
535 normal strong symbols, they are removed by the linker from the final
536 linked image (executable or dynamic library).</dd>
539 <dd>Similar to private, but the value shows as a local symbol
545 into the object file corresponding to the LLVM module. They exist to
546 allow inlining and other optimizations to take place given knowledge of
547 the definition of the global, which is known to be somewhere outside the
550 This linkage type is only allowed on definitions, not declarations.</dd>
554 the same name when linkage occurs. This is typically used to implement
555 inline functions, templates, or other code which must be generated in each
557 allowed to be discarded.</dd>
567 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
568 global scope.
569 Symbols with "<tt>common</tt>" linkage are merged in the same way as
572 must have a zero initializer, and may not be marked '<a
574 have common linkage.</dd>
579 pointer to array type. When two global variables with appending linkage
580 are linked together, the two global arrays are appended together. This is
581 the LLVM, typesafe, equivalent of having the system linker append together
585 <dd>The semantics of this linkage follow the ELF object file model: the symbol
586 is weak until linked, if not linked, the symbol becomes null instead of
587 being an undefined reference.</dd>
591 <dd>Some languages allow differing globals to be merged, such as two functions
593 that only equivalent globals are ever merged (the "one definition rule" -
596 be merged with equivalent globals. These linkage types are otherwise the
600 <dd>If none of the above identifiers are used, the global is externally
601 visible, meaning that it participates in linkage and can be used to
602 resolve external symbol references.</dd>
603 </dl>
605 <p>The next two types of linkage are targeted for Microsoft Windows platform
606 only. They are designed to support importing (exporting) symbols from (to)
607 DLLs (Dynamic Link Libraries).</p>
609 <dl>
612 or variable via a global pointer to a pointer that is set up by the DLL
613 exporting the symbol. On Microsoft Windows targets, the pointer name is
615 name.</dd>
619 pointer to a pointer in a DLL, so that it can be referenced with the
622 variable name.</dd>
623 </dl>
626 another module defined a "<tt>.LC0</tt>" variable and was linked with this
627 one, one of the two would be renamed, preventing a collision. Since
629 declarations), they are accessible outside of the current module.</p>
638 </div>
640 <!-- ======================================================================= -->
643 </div>
649 convention specified for the call. The calling convention of any pair of
650 dynamic caller/callee must match, or the behavior of the program is
651 undefined. The following calling conventions are supported by LLVM, and more
652 may be added in the future:</p>
654 <dl>
656 <dd>This calling convention (the default if no other calling convention is
657 specified) matches the target C calling conventions. This calling
658 convention supports varargs function calls and tolerates some mismatch in
659 the declared prototype and implemented declaration of the function (as
660 does normal C).</dd>
663 <dd>This calling convention attempts to make calls as fast as possible
664 (e.g. by passing things in registers). This calling convention allows the
665 target to use whatever tricks it wants to produce fast code for the
666 target, without having to conform to an externally specified ABI
667 (Application Binary Interface). Implementations of this convention should
669 optimization</a> to be supported. This calling convention does not
670 support varargs and requires the prototype of all callees to exactly match
671 the prototype of the function definition.</dd>
674 <dd>This calling convention attempts to make code in the caller as efficient
675 as possible under the assumption that the call is not commonly executed.
676 As such, these calls often preserve all registers so that the call does
677 not break any live ranges in the caller side. This calling convention
678 does not support varargs and requires the prototype of all callees to
679 exactly match the prototype of the function definition.</dd>
682 <dd>Any calling convention may be specified by number, allowing
683 target-specific calling conventions to be used. Target specific calling
685 </dl>
687 <p>More calling conventions can be added/defined on an as-needed basis, to
688 support Pascal conventions or any other well-known target-independent
689 convention.</p>
691 </div>
693 <!-- ======================================================================= -->
696 </div>
700 <p>All Global Variables and Functions have one of the following visibility
701 styles:</p>
703 <dl>
705 <dd>On targets that use the ELF object file format, default visibility means
706 that the declaration is visible to other modules and, in shared libraries,
707 means that the declared entity may be overridden. On Darwin, default
708 visibility means that the declaration is visible to other modules. Default
712 <dd>Two declarations of an object with hidden visibility refer to the same
713 object if they are in the same shared object. Usually, hidden visibility
714 indicates that the symbol will not be placed into the dynamic symbol
715 table, so no other module (executable or shared library) can reference it
716 directly.</dd>
719 <dd>On ELF, protected visibility indicates that the symbol will be placed in
720 the dynamic symbol table, but that references within the defining module
721 will bind to the local symbol. That is, the symbol cannot be overridden by
722 another module.</dd>
723 </dl>
725 </div>
727 <!-- ======================================================================= -->
730 </div>
734 <p>LLVM IR allows you to specify name aliases for certain types. This can make
735 it easier to read the IR and make the IR more condensed (particularly when
736 recursive types are involved). An example of a name specification is:</p>
739 <pre>
740 %mytype = type { %mytype*, i32 }
741 </pre>
742 </div>
748 <p>Note that type names are aliases for the structural type that they indicate,
749 and that you can therefore specify multiple names for the same type. This
750 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
751 uses structural typing, the name is not part of the type. When printing out
753 particular shape. This means that if you have code where two different
754 source types end up having the same LLVM type, that the dumper will sometimes
755 print the "wrong" or unexpected type. This is an important design point and
756 isn't going to change.</p>
758 </div>
760 <!-- ======================================================================= -->
763 </div>
767 <p>Global variables define regions of memory allocated at compilation time
768 instead of run-time. Global variables may optionally be initialized, may
769 have an explicit section to be placed in, and may have an optional explicit
770 alignment specified. A variable may be defined as "thread_local", which
771 means that it will not be shared by threads (each thread will have a
772 separated copy of the variable). A variable may be defined as a global
773 "constant," which indicates that the contents of the variable
775 global data to be placed in the read-only section of an executable, etc).
776 Note that variables that need runtime initialization cannot be marked
780 constant, even if the final definition of the global is not. This capability
781 can be used to enable slightly better optimization of the program, but
782 requires the language definition to guarantee that optimizations based on the
783 'constantness' are valid for the translation units that do not include the
784 definition.</p>
786 <p>As SSA values, global variables define pointer values that are in scope
787 (i.e. they dominate) all basic blocks in the program. Global variables
788 always define a pointer to their "content" type because they describe a
789 region of memory, and all memory objects in LLVM are accessed through
790 pointers.</p>
792 <p>A global variable may be declared to reside in a target-specific numbered
793 address space. For targets that support them, address spaces may affect how
794 optimizations are performed and/or what target instructions are used to
795 access the variable. The default address space is zero. The address space
796 qualifier must precede any other attributes.</p>
798 <p>LLVM allows an explicit section to be specified for globals. If the target
799 supports it, it will emit globals to the section specified.</p>
801 <p>An explicit alignment may be specified for a global. If not present, or if
802 the alignment is set to zero, the alignment of the global is set by the
803 target to whatever it feels convenient. If an explicit alignment is
804 specified, the global is forced to have at least that much alignment. All
807 <p>For example, the following defines a global in a numbered address space with
808 an initializer, section, and alignment:</p>
811 <pre>
813 </pre>
814 </div>
816 </div>
819 <!-- ======================================================================= -->
822 </div>
831 name, a (possibly empty) argument list (each with optional
835 curly brace, a list of basic blocks, and a closing curly brace.</p>
842 name, a possibly empty list of arguments, an optional alignment, and an
845 <p>A function definition contains a list of basic blocks, forming the CFG
846 (Control Flow Graph) for the function. Each basic block may optionally start
847 with a label (giving the basic block a symbol table entry), contains a list
849 instruction (such as a branch or function return).</p>
851 <p>The first basic block in a function is special in two ways: it is immediately
852 executed on entrance to the function, and it is not allowed to have
853 predecessor basic blocks (i.e. there can not be any branches to the entry
854 block of a function). Because the block can have no predecessors, it also
857 <p>LLVM allows an explicit section to be specified for functions. If the target
858 supports it, it will emit functions to the section specified.</p>
860 <p>An explicit alignment may be specified for a function. If not present, or if
861 the alignment is set to zero, the alignment of the function is set by the
862 target to whatever it feels convenient. If an explicit alignment is
863 specified, the function is forced to have at least that much alignment. All
868 <pre>
874 </pre>
875 </div>
877 </div>
879 <!-- ======================================================================= -->
882 </div>
887 function, global variable, another alias or bitcast of global value). Aliases
893 <pre>
895 </pre>
896 </div>
898 </div>
900 <!-- ======================================================================= -->
905 <p>The return type and each parameter of a function type may have a set of
907 used to communicate additional information about the result or parameters of
908 a function. Parameter attributes are considered to be part of the function,
909 not of the function type, so functions with different parameter attributes
910 can have the same function type.</p>
912 <p>Parameter attributes are simple keywords that follow the type specified. If
913 multiple parameter attributes are needed, they are space separated. For
914 example:</p>
917 <pre>
918 declare i32 @printf(i8* noalias nocapture, ...)
919 declare i32 @atoi(i8 zeroext)
920 declare signext i8 @returns_signed_char()
921 </pre>
922 </div>
929 <dl>
931 <dd>This indicates to the code generator that the parameter or return value
932 should be zero-extended to a 32-bit value by the caller (for a parameter)
933 or the callee (for a return value).</dd>
936 <dd>This indicates to the code generator that the parameter or return value
937 should be sign-extended to a 32-bit value by the caller (for a parameter)
938 or the callee (for a return value).</dd>
941 <dd>This indicates that this parameter or return value should be treated in a
942 special target-dependent fashion during while emitting code for a function
943 call or return (usually, by putting it in a register as opposed to memory,
944 though some targets use it to distinguish between two different kinds of
945 registers). Use of this attribute is target-specific.</dd>
948 <dd>This indicates that the pointer parameter should really be passed by value
949 to the function. The attribute implies that a hidden copy of the pointee
950 is made between the caller and the callee, so the callee is unable to
951 modify the value in the callee. This attribute is only valid on LLVM
952 pointer arguments. It is generally used to pass structs and arrays by
953 value, but is also valid on pointers to scalars. The copy is considered
954 to belong to the caller not the callee (for example,
957 values. The byval attribute also supports specifying an alignment with
958 the align attribute. This has a target-specific effect on the code
959 generator that usually indicates a desired alignment for the synthesized
960 stack slot.</dd>
963 <dd>This indicates that the pointer parameter specifies the address of a
964 structure that is the return value of the function in the source program.
965 This pointer must be guaranteed by the caller to be valid: loads and
966 stores to the structure may be assumed by the callee to not to trap. This
967 may only be applied to the first parameter. This is not a valid attribute
968 for return values. </dd>
971 <dd>This indicates that the pointer does not alias any global or any other
972 parameter. The caller is responsible for ensuring that this is the
974 that the pointer does not alias any other pointers visible to the
975 caller. For further details, please see the discussion of the NoAlias
976 response in
981 <dd>This indicates that the callee does not make any copies of the pointer
982 that outlive the callee itself. This is not a valid attribute for return
983 values.</dd>
986 <dd>This indicates that the pointer parameter can be excised using the
988 attribute for return values.</dd>
989 </dl>
991 </div>
993 <!-- ======================================================================= -->
996 </div>
1000 <p>Each function may specify a garbage collector name, which is simply a
1001 string:</p>
1004 <pre>
1005 define void @f() gc "name" { ...
1006 </pre>
1007 </div>
1010 collector which will cause the compiler to alter its output in order to
1011 support the named garbage collection algorithm.</p>
1013 </div>
1015 <!-- ======================================================================= -->
1018 </div>
1022 <p>Function attributes are set to communicate additional information about a
1023 function. Function attributes are considered to be part of the function, not
1024 of the function type, so functions with different parameter attributes can
1025 have the same function type.</p>
1027 <p>Function attributes are simple keywords that follow the type specified. If
1028 multiple attributes are needed, they are space separated. For example:</p>
1031 <pre>
1032 define void @f() noinline { ... }
1033 define void @f() alwaysinline { ... }
1034 define void @f() alwaysinline optsize { ... }
1035 define void @f() optsize
1036 </pre>
1037 </div>
1039 <dl>
1041 <dd>This attribute indicates that the inliner should attempt to inline this
1042 function into callers whenever possible, ignoring any active inlining size
1043 threshold for this caller.</dd>
1046 <dd>This attribute indicates that the source code contained a hint that inlining
1047 this function is desirable (such as the "inline" keyword in C/C++). It
1048 is just a hint; it imposes no requirements on the inliner.</dd>
1051 <dd>This attribute indicates that the inliner should never inline this
1052 function in any situation. This attribute may not be used together with
1056 <dd>This attribute suggests that optimization passes and code generator passes
1057 make choices that keep the code size of this function low, and otherwise
1058 do optimizations specifically to reduce code size.</dd>
1061 <dd>This function attribute indicates that the function never returns
1062 normally. This produces undefined behavior at runtime if the function
1063 ever does dynamically return.</dd>
1066 <dd>This function attribute indicates that the function never returns with an
1067 unwind or exceptional control flow. If the function does unwind, its
1068 runtime behavior is undefined.</dd>
1071 <dd>This attribute indicates that the function computes its result (or decides
1072 to unwind an exception) based strictly on its arguments, without
1073 dereferencing any pointer arguments or otherwise accessing any mutable
1074 state (e.g. memory, control registers, etc) visible to caller functions.
1075 It does not write through any pointer arguments
1077 changes any state visible to callers. This means that it cannot unwind
1082 <dd>This attribute indicates that the function does not write through any
1084 arguments) or otherwise modify any state (e.g. memory, control registers,
1085 etc) visible to caller functions. It may dereference pointer arguments
1086 and read state that may be set in the caller. A readonly function always
1087 returns the same value (or unwinds an exception identically) when called
1088 with the same set of arguments and global state. It cannot unwind an
1093 <dd>This attribute indicates that the function should emit a stack smashing
1095 the stack before the local variables that's checked upon return from the
1096 function to see if it has been overwritten. A heuristic is used to
1097 determine if a function needs stack protectors or not.<br>
1098 <br>
1105 stack smashing protector. This overrides
1107 <br>
1114 <dd>This attribute indicates that the code generator should not use a red
1115 zone, even if the target-specific ABI normally permits it.</dd>
1121 <dd>This attribute disables prologue / epilogue emission for the function.
1122 This can have very system-specific consequences.</dd>
1123 </dl>
1125 </div>
1127 <!-- ======================================================================= -->
1130 </div>
1135 the GCC "file scope inline asm" blocks. These blocks are internally
1136 concatenated by LLVM and treated as a single unit, but may be separated in
1140 <pre>
1141 module asm "inline asm code goes here"
1142 module asm "more can go here"
1143 </pre>
1144 </div>
1146 <p>The strings can contain any character by escaping non-printable characters.
1148 for the number.</p>
1150 <p>The inline asm code is simply printed to the machine code .s file when
1151 assembly code is generated.</p>
1153 </div>
1155 <!-- ======================================================================= -->
1158 </div>
1162 <p>A module may specify a target specific data layout string that specifies how
1163 data is to be laid out in memory. The syntax for the data layout is
1164 simply:</p>
1167 <pre>
1168 target datalayout = "<i>layout specification</i>"
1169 </pre>
1170 </div>
1173 separated by the minus sign character ('-'). Each specification starts with
1174 a letter and may include other information after the letter to define some
1175 aspect of the data layout. The specifications accepted are as follows:</p>
1177 <dl>
1179 <dd>Specifies that the target lays out data in big-endian form. That is, the
1180 bits with the most significance have the lowest address location.</dd>
1183 <dd>Specifies that the target lays out data in little-endian form. That is,
1184 the bits with the least significance have the lowest address
1185 location.</dd>
1194 <dd>This specifies the alignment for an integer type of a given bit
1198 <dd>This specifies the alignment for a vector type of a given bit
1202 <dd>This specifies the alignment for a floating point type of a given bit
1204 (double).</dd>
1207 <dd>This specifies the alignment for an aggregate type of a given bit
1211 <dd>This specifies the alignment for a stack object of a given bit
1213 </dl>
1215 <p>When constructing the data layout for a given target, LLVM starts with a
1216 default set of specifications which are then (possibly) overriden by the
1218 are given in this list:</p>
1220 <ul>
1235 </ul>
1237 <p>When LLVM is determining the alignment for a given type, it uses the
1238 following rules:</p>
1240 <ol>
1241 <li>If the type sought is an exact match for one of the specifications, that
1242 specification is used.</li>
1244 <li>If no match is found, and the type sought is an integer type, then the
1245 smallest integer type that is larger than the bitwidth of the sought type
1246 is used. If none of the specifications are larger than the bitwidth then
1247 the the largest integer type is used. For example, given the default
1248 specifications above, the i7 type will use the alignment of i8 (next
1249 largest) while both i65 and i256 will use the alignment of i64 (largest
1250 specified).</li>
1252 <li>If no match is found, and the type sought is a vector type, then the
1253 largest vector type that is smaller than the sought vector type will be
1256 </ol>
1258 </div>
1260 <!-- ======================================================================= -->
1263 </div>
1267 <p>Any memory access must be done through a pointer value associated
1268 with an address range of the memory access, otherwise the behavior
1269 is undefined. Pointer values are associated with address ranges
1270 according to the following rules:</p>
1272 <ul>
1273 <li>A pointer value formed from a
1275 is associated with the addresses associated with the first operand
1277 <li>An address of a global variable is associated with the address
1278 range of the variable's storage.</li>
1279 <li>The result value of an allocation instruction is associated with
1280 the address range of the allocated storage.</li>
1281 <li>A null pointer in the default address-space is associated with
1282 no address.</li>
1283 <li>A pointer value formed by an
1285 address ranges of all pointer values that contribute (directly or
1286 indirectly) to the computation of the pointer's value.</li>
1287 <li>The result value of a
1290 <li>An integer constant other than zero or a pointer value returned
1291 from a function not defined within LLVM may be associated with address
1292 ranges allocated through mechanisms other than those provided by
1293 LLVM. Such ranges shall not overlap with any ranges of addresses
1294 allocated by mechanisms provided by LLVM.</li>
1295 </ul>
1297 <p>LLVM IR does not associate types with memory. The result type of a
1299 alignment of the memory from which to load, as well as the
1300 interpretation of the value. The first operand of a
1302 and alignment of the store.</p>
1304 <p>Consequently, type-based alias analysis, aka TBAA, aka
1307 additional information which specialized optimization passes may use
1308 to implement type-based alias analysis.</p>
1310 </div>
1312 <!-- *********************************************************************** -->
1314 <!-- *********************************************************************** -->
1318 <p>The LLVM type system is one of the most important features of the
1319 intermediate representation. Being typed enables a number of optimizations
1320 to be performed on the intermediate representation directly, without having
1321 to do extra analyses on the side before the transformation. A strong type
1322 system makes it easier to read the generated code and enables novel analyses
1323 and transformations that are not feasible to perform on normal three address
1324 code representations.</p>
1326 </div>
1328 <!-- ======================================================================= -->
1330 Classifications</a> </div>
1337 <tbody>
1339 <tr>
1342 </tr>
1343 <tr>
1346 </tr>
1347 <tr>
1357 </td>
1358 </tr>
1359 <tr>
1365 </tr>
1366 <tr>
1376 </td>
1377 </tr>
1378 </tbody>
1379 </table>
1382 important. Values of these types are the only ones which can be produced by
1383 instructions, passed as arguments, or used as operands to instructions.</p>
1385 </div>
1387 <!-- ======================================================================= -->
1392 <p>The primitive types are the fundamental building blocks of the LLVM
1393 system.</p>
1395 </div>
1397 <!-- _______________________________________________________________________ -->
1402 <table>
1403 <tbody>
1410 </tbody>
1411 </table>
1413 </div>
1415 <!-- _______________________________________________________________________ -->
1424 <pre>
1425 void
1426 </pre>
1428 </div>
1430 <!-- _______________________________________________________________________ -->
1439 <pre>
1440 label
1441 </pre>
1443 </div>
1445 <!-- _______________________________________________________________________ -->
1451 <p>The metadata type represents embedded metadata. The only derived type that
1453 takes metadata typed parameters, but not pointer to metadata types.</p>
1456 <pre>
1457 metadata
1458 </pre>
1460 </div>
1463 <!-- ======================================================================= -->
1468 <p>The real power in LLVM comes from the derived types in the system. This is
1469 what allows a programmer to represent arrays, functions, pointers, and other
1470 useful types. Note that these derived types may be recursive: For example,
1471 it is possible to have a two dimensional array.</p>
1473 </div>
1475 <!-- _______________________________________________________________________ -->
1481 <p>The integer type is a very simple derived type that simply specifies an
1482 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1486 <pre>
1487 iN
1488 </pre>
1491 value.</p>
1498 </tr>
1502 </tr>
1506 </tr>
1507 </table>
1509 <p>Note that the code generator does not yet support large integer types to be
1510 used as function return types. The specific limit on how large a return type
1511 the code generator can currently handle is target-dependent; currently it's
1514 </div>
1516 <!-- _______________________________________________________________________ -->
1522 <p>The array type is a very simple derived type that arranges elements
1523 sequentially in memory. The array type requires a size (number of elements)
1524 and an underlying data type.</p>
1527 <pre>
1529 </pre>
1532 be any type with a size.</p>
1539 </tr>
1543 </tr>
1547 </tr>
1548 </table>
1554 </tr>
1558 </tr>
1562 </tr>
1563 </table>
1565 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1566 length array. Normally, accesses past the end of an array are undefined in
1568 a special case, however, zero length arrays are recognized to be variable
1569 length. This allows implementation of 'pascal style arrays' with the LLVM
1572 <p>Note that the code generator does not yet support large aggregate types to be
1573 used as function return types. The specific limit on how large an aggregate
1574 return type the code generator can currently handle is target-dependent, and
1575 also dependent on the aggregate element types.</p>
1577 </div>
1579 <!-- _______________________________________________________________________ -->
1585 <p>The function type can be thought of as a function signature. It consists of
1586 a return type and a list of formal parameter types. The return type of a
1587 function type is a scalar type, a void type, or a struct type. If the return
1588 type is a struct type then all struct elements must be of first class types,
1589 and the struct must have at least one element.</p>
1592 <pre>
1594 </pre>
1598 which indicates that the function takes a variable number of arguments.
1599 Variable argument functions can access their arguments with
1609 </td>
1612 </tt></td>
1617 </td>
1623 LLVM.
1624 </td>
1629 </td>
1630 </tr>
1631 </table>
1633 </div>
1635 <!-- _______________________________________________________________________ -->
1641 <p>The structure type is used to represent a collection of data members together
1642 in memory. The packing of the field types is defined to match the ABI of the
1643 underlying processor. The elements of a structure may be any type that has a
1644 size.</p>
1651 <pre>
1653 </pre>
1666 </tr>
1667 </table>
1669 <p>Note that the code generator does not yet support large aggregate types to be
1670 used as function return types. The specific limit on how large an aggregate
1671 return type the code generator can currently handle is target-dependent, and
1672 also dependent on the aggregate element types.</p>
1674 </div>
1676 <!-- _______________________________________________________________________ -->
1678 </div>
1683 <p>The packed structure type is used to represent a collection of data members
1684 together in memory. There is no padding between fields. Further, the
1685 alignment of a packed structure is 1 byte. The elements of a packed
1686 structure may be any type that has a size.</p>
1693 <pre>
1695 </pre>
1709 </tr>
1710 </table>
1712 </div>
1714 <!-- _______________________________________________________________________ -->
1720 <p>As in many languages, the pointer type represents a pointer or reference to
1721 another object, which must live in memory. Pointer types may have an optional
1722 address space attribute defining the target-specific numbered address space
1723 where the pointed-to object resides. The default address space is zero.</p>
1729 <pre>
1731 </pre>
1739 </tr>
1745 </tr>
1750 </tr>
1751 </table>
1753 </div>
1755 <!-- _______________________________________________________________________ -->
1761 <p>A vector type is a simple derived type that represents a vector of elements.
1762 Vector types are used when multiple primitive data are operated in parallel
1763 using a single instruction (SIMD). A vector type requires a size (number of
1764 elements) and an underlying primitive data type. Vectors must have a power
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 <p>Note that the code generator does not yet support large vector types to be
1793 used as function return types. The specific limit on how large a vector
1794 return type codegen can currently handle is target-dependent; currently it's
1795 often a few times longer than a hardware vector register.</p>
1797 </div>
1799 <!-- _______________________________________________________________________ -->
1804 <p>Opaque types are used to represent unknown types in the system. This
1805 corresponds (for example) to the C notion of a forward declared structure
1806 type. In LLVM, opaque types can eventually be resolved to any type (not just
1807 a structure type).</p>
1810 <pre>
1811 opaque
1812 </pre>
1819 </tr>
1820 </table>
1822 </div>
1824 <!-- ======================================================================= -->
1827 </div>
1833 requiring it to have a name. For instance, a structure declaration may
1834 contain a pointer to any of the types it is lexically a member of. Example
1835 of up references (with their equivalent as named type declarations)
1836 include:</p>
1838 <pre>
1839 { \2 * } %x = type { %x* }
1840 { \2 }* %y = type { %y }*
1841 \1* %z = type %z*
1842 </pre>
1844 <p>An up reference is needed by the asmprinter for printing out cyclic types
1845 when there is no declared name for a type in the cycle. Because the
1846 asmprinter does not want to print out an infinite type string, it needs a
1847 syntax to handle recursive types that have no names (all names are optional
1848 in llvm IR).</p>
1851 <pre>
1853 </pre>
1862 </tr>
1866 structure.</td>
1867 </tr>
1868 </table>
1870 </div>
1872 <!-- *********************************************************************** -->
1874 <!-- *********************************************************************** -->
1878 <p>LLVM has several different basic types of constants. This section describes
1879 them all and their syntax.</p>
1881 </div>
1883 <!-- ======================================================================= -->
1888 <dl>
1896 with integer types.</dd>
1900 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1901 notation (see below). The assembler requires the exact decimal value of a
1902 floating-point constant. For example, the assembler accepts 1.25 but
1909 </dl>
1911 <p>The one non-intuitive notation for constants is the hexadecimal form of
1912 floating point constants. For example, the form '<tt>double
1915 constants are required (and the only time that they are generated by the
1916 disassembler) is when a floating point constant must be emitted but it cannot
1917 be represented as a decimal floating point number in a reasonable number of
1918 digits. For example, NaN's, infinities, and other special values are
1919 represented in their IEEE hexadecimal format so that assembly and disassembly
1920 do not cause any bits to change in the constants.</p>
1922 <p>When using the hexadecimal form, constants of types float and double are
1923 represented using the 16-digit form shown above (which matches the IEEE754
1924 representation for double); float values must, however, be exactly
1925 representable as IEE754 single precision. Hexadecimal format is always used
1926 for long double, and there are three forms of long double. The 80-bit format
1928 The 128-bit format used by PowerPC (two adjacent doubles) is represented
1931 currently supported target uses this format. Long doubles will only work if
1932 they match the long double format on your target. All hexadecimal formats
1933 are big-endian (sign bit at the left).</p>
1935 </div>
1937 <!-- ======================================================================= -->
1941 </div>
1945 <p>Complex constants are a (potentially recursive) combination of simple
1946 constants and smaller complex constants.</p>
1948 <dl>
1950 <dd>Structure constants are represented with notation similar to structure
1951 type definitions (a comma separated list of elements, surrounded by braces
1955 the number and types of elements must match those specified by the
1956 type.</dd>
1959 <dd>Array constants are represented with notation similar to array type
1960 definitions (a comma separated list of elements, surrounded by square
1963 the number and types of elements must match those specified by the
1964 type.</dd>
1967 <dd>Vector constants are represented with notation similar to vector type
1968 definitions (a comma separated list of elements, surrounded by
1970 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
1972 elements must match those specified by the type.</dd>
1977 This is often used to avoid having to print large zero initializers
1978 (e.g. for large arrays) and is always exactly equivalent to using explicit
1979 zero initializers.</dd>
1982 <dd>A metadata node is a structure-like constant with
1985 be interpreted as part of the instruction stream, metadata is a place to
1986 attach additional information such as debug info.</dd>
1987 </dl>
1989 </div>
1991 <!-- ======================================================================= -->
1994 </div>
2000 (link-time) constants. These constants are explicitly referenced when
2003 legal LLVM file:</p>
2006 <pre>
2007 @X = global i32 17
2008 @Y = global i32 42
2009 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2010 </pre>
2011 </div>
2013 </div>
2015 <!-- ======================================================================= -->
2020 specific value. Undefined values may be of any type and be used anywhere a
2021 constant is permitted.</p>
2023 <p>Undefined values indicate to the compiler that the program is well defined no
2024 matter what value is used, giving the compiler more freedom to optimize.</p>
2026 </div>
2028 <!-- ======================================================================= -->
2030 </div>
2034 <p>Constant expressions are used to allow expressions involving other constants
2035 to be used as constants. Constant expressions may be of
2037 operation that does not have side effects (e.g. load and call are not
2038 supported). The following is the syntax for constant expressions:</p>
2040 <dl>
2042 <dd>Truncate a constant to another type. The bit size of CST must be larger
2043 than the bit size of TYPE. Both types must be integers.</dd>
2046 <dd>Zero extend a constant to another type. The bit size of CST must be
2047 smaller or equal to the bit size of TYPE. Both types must be
2048 integers.</dd>
2051 <dd>Sign extend a constant to another type. The bit size of CST must be
2052 smaller or equal to the bit size of TYPE. Both types must be
2053 integers.</dd>
2056 <dd>Truncate a floating point constant to another floating point type. The
2057 size of CST must be larger than the size of TYPE. Both types must be
2058 floating point.</dd>
2061 <dd>Floating point extend a constant to another type. The size of CST must be
2062 smaller or equal to the size of TYPE. Both types must be floating
2063 point.</dd>
2066 <dd>Convert a floating point constant to the corresponding unsigned integer
2067 constant. TYPE must be a scalar or vector integer type. CST must be of
2068 scalar or vector floating point type. Both CST and TYPE must be scalars,
2069 or vectors of the same number of elements. If the value won't fit in the
2070 integer type, the results are undefined.</dd>
2073 <dd>Convert a floating point constant to the corresponding signed integer
2074 constant. TYPE must be a scalar or vector integer type. CST must be of
2075 scalar or vector floating point type. Both CST and TYPE must be scalars,
2076 or vectors of the same number of elements. If the value won't fit in the
2077 integer type, the results are undefined.</dd>
2080 <dd>Convert an unsigned integer constant to the corresponding floating point
2081 constant. TYPE must be a scalar or vector floating point type. CST must be
2082 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2083 vectors of the same number of elements. If the value won't fit in the
2084 floating point type, the results are undefined.</dd>
2087 <dd>Convert a signed integer constant to the corresponding floating point
2088 constant. TYPE must be a scalar or vector floating point type. CST must be
2089 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2090 vectors of the same number of elements. If the value won't fit in the
2091 floating point type, the results are undefined.</dd>
2094 <dd>Convert a pointer typed constant to the corresponding integer constant
2100 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2101 type. CST must be of integer type. The CST value is zero extended,
2102 truncated, or unchanged to make it fit in a pointer size. This one is
2106 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2114 instruction, the index list may have zero or more indexes, which are
2128 constants.</dd>
2132 constants.</dd>
2136 constants.</dd>
2139 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2142 on operands are the same as those for the corresponding instruction
2143 (e.g. no bitwise operations on floating point values are allowed).</dd>
2144 </dl>
2146 </div>
2148 <!-- ======================================================================= -->
2150 </div>
2154 <p>Embedded metadata provides a way to attach arbitrary data to the instruction
2155 stream without affecting the behaviour of the program. There are two
2156 metadata primitives, strings and nodes. All metadata has the
2160 <p>A metadata string is a string surrounded by double quotes. It can contain
2164 <p>Metadata nodes are represented with notation similar to structure constants
2165 (a comma separated list of elements, surrounded by braces and preceeded by an
2169 <p>A metadata node will attempt to track changes to the values it holds. In the
2170 event that a value is deleted, it will be replaced with a typeless
2173 <p>Optimizations may rely on metadata to provide additional information about
2174 the program that isn't available in the instructions, or that isn't easily
2175 computable. Similarly, the code generator may expect a certain metadata
2176 format to be used to express debugging information.</p>
2178 </div>
2180 <!-- *********************************************************************** -->
2182 <!-- *********************************************************************** -->
2184 <!-- ======================================================================= -->
2187 </div>
2191 <p>LLVM supports inline assembler expressions (as opposed
2193 a special value. This value represents the inline assembler as a string
2194 (containing the instructions to emit), a list of operand constraints (stored
2195 as a string), and a flag that indicates whether or not the inline asm
2196 expression has side effects. An example inline assembler expression is:</p>
2199 <pre>
2201 </pre>
2202 </div>
2206 have:</p>
2209 <pre>
2211 </pre>
2212 </div>
2214 <p>Inline asms with side effects not visible in the constraint list must be
2215 marked as having side effects. This is done through the use of the
2219 <pre>
2221 </pre>
2222 </div>
2224 <p>TODO: The format of the asm and constraints string still need to be
2225 documented here. Constraints on what can be done (e.g. duplication, moving,
2226 etc need to be documented). This is probably best done by reference to
2227 another document that covers inline asm from a holistic perspective.</p>
2229 </div>
2232 <!-- *********************************************************************** -->
2235 </div>
2236 <!-- *********************************************************************** -->
2239 code generation or other IR semantics. These are documented here. All globals
2240 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2241 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2242 by LLVM.</p>
2244 <!-- ======================================================================= -->
2247 </div>
2253 pointers to global variables and functions which may optionally have a pointer
2254 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2256 <pre>
2257 @X = global i8 4
2258 @Y = global i32 123
2260 @llvm.used = appending global [2 x i8*] [
2261 i8* @X,
2262 i8* bitcast (i32* @Y to i8*)
2263 ], section "llvm.metadata"
2264 </pre>
2267 compiler, assembler, and linker are required to treat the symbol as if there is
2268 a reference to the global that it cannot see. For example, if a variable has
2270 list, it cannot be deleted. This is commonly used to represent references from
2271 inline asms and other things the compiler cannot "see", and corresponds to
2274 <p>On some targets, the code generator must emit a directive to the assembler or
2275 object file to prevent the assembler and linker from molesting the symbol.</p>
2277 </div>
2279 <!-- ======================================================================= -->
2282 </div>
2288 touching the symbol. On targets that support it, this allows an intelligent
2289 linker to optimize references to the symbol without being impeded as it would be
2292 <p>This is a rare construct that should only be used in rare circumstances, and
2293 should not be exposed to source languages.</p>
2295 </div>
2297 <!-- ======================================================================= -->
2300 </div>
2306 </div>
2308 <!-- ======================================================================= -->
2311 </div>
2317 </div>
2320 <!-- *********************************************************************** -->
2322 <!-- *********************************************************************** -->
2326 <p>The LLVM instruction set consists of several different classifications of
2333 </div>
2335 <!-- ======================================================================= -->
2337 Instructions</a> </div>
2342 in a program ends with a "Terminator" instruction, which indicates which
2343 block should be executed after the current block is finished. These
2345 control flow, not values (the one exception being the
2348 <p>There are six different terminator instructions: the
2356 </div>
2358 <!-- _______________________________________________________________________ -->
2360 Instruction</a> </div>
2365 <pre>
2368 </pre>
2372 a value) from a function back to the caller.</p>
2375 value and then causes control flow, and one that just causes control flow to
2376 occur.</p>
2380 return value. The type of the return value must be a
2385 value or a return value with a type that does not match its type, or if it
2387 return value.</p>
2391 the calling function's context. If the caller is a
2393 instruction after the call. If the caller was an
2395 the beginning of the "normal" destination block. If the instruction returns
2396 a value, that value shall set the call or invoke instruction's return
2397 value.</p>
2400 <pre>
2404 </pre>
2406 <p>Note that the code generator does not yet fully support large
2407 return values. The specific sizes that are currently supported are
2408 dependent on the target. For integers, on 32-bit targets the limit
2410 For aggregate types, the current limits are dependent on the element
2411 types; for example targets are often limited to 2 total integer
2414 </div>
2415 <!-- _______________________________________________________________________ -->
2421 <pre>
2422 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2423 </pre>
2427 different basic block in the current function. There are two forms of this
2428 instruction, corresponding to a conditional branch and an unconditional
2429 branch.</p>
2435 target.</p>
2444 <pre>
2445 Test:
2447 br i1 %cond, label %IfEqual, label %IfUnequal
2448 IfEqual:
2450 IfUnequal:
2452 </pre>
2454 </div>
2456 <!-- _______________________________________________________________________ -->
2459 </div>
2464 <pre>
2465 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2466 </pre>
2471 instruction, allowing a branch to occur to one of many possible
2472 destinations.</p>
2478 The table is not allowed to contain duplicate constant entries.</p>
2483 is searched for the given value. If the value is found, control flow is
2484 transfered to the corresponding destination; otherwise, control flow is
2485 transfered to the default destination.</p>
2488 <p>Depending on properties of the target machine and the particular
2490 different ways. For example, it could be generated as a series of chained
2491 conditional branches or with a lookup table.</p>
2494 <pre>
2497 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2500 switch i32 0, label %dest [ ]
2503 switch i32 %val, label %otherwise [ i32 0, label %onzero
2504 i32 1, label %onone
2505 i32 2, label %ontwo ]
2506 </pre>
2508 </div>
2510 <!-- _______________________________________________________________________ -->
2513 </div>
2518 <pre>
2519 <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>]
2521 </pre>
2525 function, with the possibility of control flow transfer to either the
2528 control flow will return to the "normal" label. If the callee (or any
2530 instruction, control is interrupted and continued at the dynamically nearest
2536 <ol>
2538 convention</a> the call should use. If none is specified, the call
2539 defaults to using C calling conventions.</li>
2546 function value being invoked. In most cases, this is a direct function
2548 off an arbitrary pointer to function value.</li>
2551 function to be invoked. </li>
2554 signature argument types. If the function signature indicates the
2555 function accepts a variable number of arguments, the extra arguments can
2556 be specified.</li>
2567 </ol>
2570 <p>This instruction is designed to operate as a standard
2572 primary difference is that it establishes an association with a label, which
2573 is used by the runtime library to unwind the stack.</p>
2575 <p>This instruction is used in languages with destructors to ensure that proper
2577 exception. Additionally, this is important for implementation of
2580 <p>For the purposes of the SSA form, the definition of the value returned by the
2582 block to the "normal" label. If the callee unwinds then no return value is
2583 available.</p>
2586 <pre>
2587 %retval = invoke i32 @Test(i32 15) to label %Continue
2591 </pre>
2593 </div>
2595 <!-- _______________________________________________________________________ -->
2598 Instruction</a> </div>
2603 <pre>
2604 unwind
2605 </pre>
2609 at the first callee in the dynamic call stack which used
2611 This is primarily used to implement exception handling.</p>
2615 immediately halt. The dynamic call stack is then searched for the
2617 Once found, execution continues at the "exceptional" destination block
2619 instruction in the dynamic call chain, undefined behavior results.</p>
2621 </div>
2623 <!-- _______________________________________________________________________ -->
2626 Instruction</a> </div>
2631 <pre>
2632 unreachable
2633 </pre>
2637 instruction is used to inform the optimizer that a particular portion of the
2638 code is not reachable. This can be used to indicate that the code after a
2639 no-return function cannot be reached, and other facts.</p>
2644 </div>
2646 <!-- ======================================================================= -->
2651 <p>Binary operators are used to do most of the computation in a program. They
2652 require two operands of the same type, execute an operation on them, and
2653 produce a single value. The operands might represent multiple data, as is
2655 has the same type as its operands.</p>
2659 </div>
2661 <!-- _______________________________________________________________________ -->
2664 </div>
2669 <pre>
2674 </pre>
2682 integer values. Both arguments must have identical types.</p>
2687 <p>If the sum has unsigned overflow, the result returned is the mathematical
2690 <p>Because LLVM integers use a two's complement representation, this instruction
2691 is appropriate for both signed and unsigned integers.</p>
2696 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
2699 <pre>
2701 </pre>
2703 </div>
2705 <!-- _______________________________________________________________________ -->
2708 </div>
2713 <pre>
2715 </pre>
2723 floating point values. Both arguments must have identical types.</p>
2729 <pre>
2731 </pre>
2733 </div>
2735 <!-- _______________________________________________________________________ -->
2738 </div>
2743 <pre>
2748 </pre>
2752 operands.</p>
2756 representations.</p>
2761 integer values. Both arguments must have identical types.</p>
2766 <p>If the difference has unsigned overflow, the result returned is the
2768 result.</p>
2770 <p>Because LLVM integers use a two's complement representation, this instruction
2771 is appropriate for both signed and unsigned integers.</p>
2776 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
2779 <pre>
2782 </pre>
2784 </div>
2786 <!-- _______________________________________________________________________ -->
2789 </div>
2794 <pre>
2796 </pre>
2800 operands.</p>
2804 representations.</p>
2809 floating point values. Both arguments must have identical types.</p>
2815 <pre>
2818 </pre>
2820 </div>
2822 <!-- _______________________________________________________________________ -->
2825 </div>
2830 <pre>
2835 </pre>
2843 integer values. Both arguments must have identical types.</p>
2848 <p>If the result of the multiplication has unsigned overflow, the result
2850 width of the result.</p>
2852 <p>Because LLVM integers use a two's complement representation, and the result
2853 is the same width as the operands, this instruction returns the correct
2854 result for both signed and unsigned integers. If a full product
2856 be sign-extended or zero-extended as appropriate to the width of the full
2857 product.</p>
2862 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
2865 <pre>
2867 </pre>
2869 </div>
2871 <!-- _______________________________________________________________________ -->
2874 </div>
2879 <pre>
2881 </pre>
2889 floating point values. Both arguments must have identical types.</p>
2895 <pre>
2897 </pre>
2899 </div>
2901 <!-- _______________________________________________________________________ -->
2903 </a></div>
2908 <pre>
2910 </pre>
2918 values. Both arguments must have identical types.</p>
2923 <p>Note that unsigned integer division and signed integer division are distinct
2929 <pre>
2931 </pre>
2933 </div>
2935 <!-- _______________________________________________________________________ -->
2937 </a> </div>
2942 <pre>
2945 </pre>
2953 values. Both arguments must have identical types.</p>
2956 <p>The value produced is the signed integer quotient of the two operands rounded
2957 towards zero.</p>
2959 <p>Note that signed integer division and unsigned integer division are distinct
2962 <p>Division by zero leads to undefined behavior. Overflow also leads to
2963 undefined behavior; this is a rare case, but can occur, for example, by doing
2968 would occur.</p>
2971 <pre>
2973 </pre>
2975 </div>
2977 <!-- _______________________________________________________________________ -->
2979 Instruction</a> </div>
2984 <pre>
2986 </pre>
2994 floating point values. Both arguments must have identical types.</p>
3000 <pre>
3002 </pre>
3004 </div>
3006 <!-- _______________________________________________________________________ -->
3008 </div>
3013 <pre>
3015 </pre>
3019 division of its two arguments.</p>
3024 values. Both arguments must have identical types.</p>
3028 This instruction always performs an unsigned division to get the
3029 remainder.</p>
3031 <p>Note that unsigned integer remainder and signed integer remainder are
3037 <pre>
3039 </pre>
3041 </div>
3043 <!-- _______________________________________________________________________ -->
3046 </div>
3051 <pre>
3053 </pre>
3057 division of its two operands. This instruction can also take
3059 elements must be integers.</p>
3064 values. Both arguments must have identical types.</p>
3070 a value. For more information about the difference,
3072 Math Forum</a>. For a table of how this is implemented in various languages,
3076 <p>Note that signed integer remainder and unsigned integer remainder are
3079 <p>Taking the remainder of a division by zero leads to undefined behavior.
3080 Overflow also leads to undefined behavior; this is a rare case, but can
3081 occur, for example, by taking the remainder of a 32-bit division of
3083 lets srem be implemented using instructions that return both the result of
3084 the division and the remainder.)</p>
3087 <pre>
3089 </pre>
3091 </div>
3093 <!-- _______________________________________________________________________ -->
3100 <pre>
3102 </pre>
3106 its two operands.</p>
3111 floating point values. Both arguments must have identical types.</p>
3115 has the same sign as the dividend.</p>
3118 <pre>
3120 </pre>
3122 </div>
3124 <!-- ======================================================================= -->
3126 Operations</a> </div>
3130 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3131 program. They are generally very efficient instructions and can commonly be
3132 strength reduced from other instructions. They require two operands of the
3133 same type, execute an operation on them, and produce a single value. The
3134 resulting value is the same type as its operands.</p>
3136 </div>
3138 <!-- _______________________________________________________________________ -->
3140 Instruction</a> </div>
3145 <pre>
3147 </pre>
3151 a specified number of bits.</p>
3161 is (statically or dynamically) negative or equal to or larger than the number
3167 <pre>
3172 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3173 </pre>
3175 </div>
3177 <!-- _______________________________________________________________________ -->
3179 Instruction</a> </div>
3184 <pre>
3186 </pre>
3190 operand shifted to the right a specified number of bits with zero fill.</p>
3198 <p>This instruction always performs a logical shift right operation. The most
3199 significant bits of the result will be filled with zero bits after the shift.
3206 <pre>
3212 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3213 </pre>
3215 </div>
3217 <!-- _______________________________________________________________________ -->
3219 Instruction</a> </div>
3223 <pre>
3225 </pre>
3229 operand shifted to the right a specified number of bits with sign
3230 extension.</p>
3238 <p>This instruction always performs an arithmetic shift right operation, The
3239 most significant bits of the result will be filled with the sign bit
3246 <pre>
3252 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3253 </pre>
3255 </div>
3257 <!-- _______________________________________________________________________ -->
3259 Instruction</a> </div>
3264 <pre>
3266 </pre>
3270 operands.</p>
3275 values. Both arguments must have identical types.</p>
3281 <tbody>
3282 <tr>
3286 </tr>
3287 <tr>
3291 </tr>
3292 <tr>
3296 </tr>
3297 <tr>
3301 </tr>
3302 <tr>
3306 </tr>
3307 </tbody>
3308 </table>
3311 <pre>
3315 </pre>
3316 </div>
3317 <!-- _______________________________________________________________________ -->
3323 <pre>
3325 </pre>
3329 two operands.</p>
3334 values. Both arguments must have identical types.</p>
3340 <tbody>
3341 <tr>
3345 </tr>
3346 <tr>
3350 </tr>
3351 <tr>
3355 </tr>
3356 <tr>
3360 </tr>
3361 <tr>
3365 </tr>
3366 </tbody>
3367 </table>
3370 <pre>
3374 </pre>
3376 </div>
3378 <!-- _______________________________________________________________________ -->
3380 Instruction</a> </div>
3385 <pre>
3387 </pre>
3397 values. Both arguments must have identical types.</p>
3403 <tbody>
3404 <tr>
3408 </tr>
3409 <tr>
3413 </tr>
3414 <tr>
3418 </tr>
3419 <tr>
3423 </tr>
3424 <tr>
3428 </tr>
3429 </tbody>
3430 </table>
3433 <pre>
3438 </pre>
3440 </div>
3442 <!-- ======================================================================= -->
3445 </div>
3449 <p>LLVM supports several instructions to represent vector operations in a
3450 target-independent manner. These instructions cover the element-access and
3451 vector-specific operations needed to process vectors effectively. While LLVM
3452 does directly support these vector operations, many sophisticated algorithms
3453 will want to use target-specific intrinsics to take full advantage of a
3454 specific target.</p>
3456 </div>
3458 <!-- _______________________________________________________________________ -->
3461 </div>
3466 <pre>
3467 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3468 </pre>
3472 from a vector at a specified index.</p>
3478 indicating the position from which to extract the element. The index may be
3479 a variable.</p>
3482 <p>The result is a scalar of the same type as the element type of
3485 results are undefined.</p>
3488 <pre>
3490 </pre>
3492 </div>
3494 <!-- _______________________________________________________________________ -->
3497 </div>
3502 <pre>
3503 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3504 </pre>
3508 vector at a specified index.</p>
3513 whose type must equal the element type of the first operand. The third
3514 operand is an index indicating the position at which to insert the value.
3515 The index may be a variable.</p>
3521 results are undefined.</p>
3524 <pre>
3526 </pre>
3528 </div>
3530 <!-- _______________________________________________________________________ -->
3533 </div>
3538 <pre>
3539 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3540 </pre>
3544 from two input vectors, returning a vector with the same element type as the
3545 input and length that is the same as the shuffle mask.</p>
3549 with types that match each other. The third argument is a shuffle mask whose
3550 element type is always 'i32'. The result of the instruction is a vector
3551 whose length is the same as the shuffle mask and whose element type is the
3552 same as the element type of the first two operands.</p>
3554 <p>The shuffle mask operand is required to be a constant vector with either
3555 constant integer or undef values.</p>
3558 <p>The elements of the two input vectors are numbered from left to right across
3559 both of the vectors. The shuffle mask operand specifies, for each element of
3560 the result vector, which element of the two input vectors the result element
3561 gets. The element selector may be undef (meaning "don't care") and the
3562 second operand may be undef if performing a shuffle from only one vector.</p>
3565 <pre>
3569 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3573 <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>
3574 </pre>
3576 </div>
3578 <!-- ======================================================================= -->
3581 </div>
3587 </div>
3589 <!-- _______________________________________________________________________ -->
3592 </div>
3597 <pre>
3599 </pre>
3603 or array element from an aggregate value.</p>
3608 operands are constant indices to specify which value to extract in a similar
3609 manner as indices in a
3613 <p>The result is the value at the position in the aggregate specified by the
3614 index operands.</p>
3617 <pre>
3619 </pre>
3621 </div>
3623 <!-- _______________________________________________________________________ -->
3626 </div>
3631 <pre>
3632 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3633 </pre>
3637 array element in an aggregate.</p>
3643 second operand is a first-class value to insert. The following operands are
3644 constant indices indicating the position at which to insert the value in a
3645 similar manner as indices in a
3647 value to insert must have the same type as the value identified by the
3648 indices.</p>
3656 <pre>
3658 </pre>
3660 </div>
3663 <!-- ======================================================================= -->
3666 </div>
3670 <p>A key design point of an SSA-based representation is how it represents
3671 memory. In LLVM, no memory locations are in SSA form, which makes things
3672 very simple. This section describes how to read, write, allocate, and free
3673 memory in LLVM.</p>
3675 </div>
3677 <!-- _______________________________________________________________________ -->
3680 </div>
3685 <pre>
3686 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3687 </pre>
3691 returns a pointer to it. The object is always allocated in the generic
3692 address space (address space zero).</p>
3697 system and returns a pointer of the appropriate type to the program. If
3698 "NumElements" is specified, it is the number of elements allocated, otherwise
3699 "NumElements" is defaulted to be one. If a constant alignment is specified,
3700 the value result of the allocation is guaranteed to be aligned to at least
3701 that boundary. If not specified, or if zero, the target can choose to align
3702 the allocation on any convenient boundary compatible with the type.</p>
3708 pointer is returned. The result of a zero byte allocation is undefined. The
3709 result is null if there is insufficient memory available.</p>
3712 <pre>
3720 </pre>
3724 </div>
3726 <!-- _______________________________________________________________________ -->
3729 </div>
3734 <pre>
3736 </pre>
3740 to be reallocated in the future.</p>
3747 <p>Access to the memory pointed to by the pointer is no longer defined after
3748 this instruction executes. If the pointer is null, the operation is a
3749 noop.</p>
3752 <pre>
3754 free [4 x i8]* %array
3755 </pre>
3757 </div>
3759 <!-- _______________________________________________________________________ -->
3762 </div>
3767 <pre>
3768 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3769 </pre>
3773 currently executing function, to be automatically released when this function
3774 returns to its caller. The object is always allocated in the generic address
3775 space (address space zero).</p>
3780 runtime stack, returning a pointer of the appropriate type to the program.
3781 If "NumElements" is specified, it is the number of elements allocated,
3782 otherwise "NumElements" is defaulted to be one. If a constant alignment is
3783 specified, the value result of the allocation is guaranteed to be aligned to
3784 at least that boundary. If not specified, or if zero, the target can choose
3785 to align the allocation on any convenient boundary compatible with the
3786 type.</p>
3791 <p>Memory is allocated; a pointer is returned. The operation is undefined if
3793 memory is automatically released when the function returns. The
3795 variables that must have an address available. When the function returns
3798 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
3801 <pre>
3806 </pre>
3808 </div>
3810 <!-- _______________________________________________________________________ -->
3812 Instruction</a> </div>
3817 <pre>
3820 </pre>
3827 from which to load. The pointer must point to
3832 instructions. </p>
3835 operation (that is, the alignment of the memory address). A value of 0 or an
3836 omitted "align" argument means that the operation has the preferential
3837 alignment for the target. It is the responsibility of the code emitter to
3838 ensure that the alignment information is correct. Overestimating the
3839 alignment results in an undefined behavior. Underestimating the alignment may
3843 <p>The location of memory pointed to is loaded. If the value being loaded is of
3844 scalar type then the number of bytes read does not exceed the minimum number
3845 of bytes needed to hold all bits of the type. For example, loading an
3848 is undefined if the value was not originally written using a store of the
3849 same type.</p>
3852 <pre>
3856 </pre>
3858 </div>
3860 <!-- _______________________________________________________________________ -->
3862 Instruction</a> </div>
3867 <pre>
3868 store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3869 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3870 </pre>
3877 and an address at which to store it. The type of the
3884 instructions.</p>
3887 operation (that is, the alignment of the memory address). A value of 0 or an
3888 omitted "align" argument means that the operation has the preferential
3889 alignment for the target. It is the responsibility of the code emitter to
3890 ensure that the alignment information is correct. Overestimating the
3891 alignment results in an undefined behavior. Underestimating the alignment may
3898 does not exceed the minimum number of bytes needed to hold all bits of the
3901 integral number of bytes, it is unspecified what happens to the extra bits
3902 that do not belong to the type, but they will typically be overwritten.</p>
3905 <pre>
3909 </pre>
3911 </div>
3913 <!-- _______________________________________________________________________ -->
3916 </div>
3921 <pre>
3924 </pre>
3928 subelement of an aggregate data structure. It performs address calculation
3929 only and does not access memory.</p>
3932 <p>The first argument is always a pointer, and forms the basis of the
3933 calculation. The remaining arguments are indices that indicate which of the
3934 elements of the aggregate object are indexed. The interpretation of each
3935 index is dependent on the type being indexed into. The first index always
3936 indexes the pointer value given as the first argument, the second index
3937 indexes a value of the type pointed to (not necessarily the value directly
3938 pointed to, since the first index can be non-zero), etc. The first type
3939 indexed into must be a pointer value, subsequent types can be arrays, vectors
3940 and structs. Note that subsequent types being indexed into can never be
3941 pointers, since that would require loading the pointer before continuing
3942 calculation.</p>
3944 <p>The type of each index argument depends on the type it is indexing into.
3947 vector, integers of any width are allowed, and they are not required to be
3948 constant.</p>
3950 <p>For example, let's consider a C code fragment and how it gets compiled to
3951 LLVM:</p>
3954 <pre>
3955 struct RT {
3956 char A;
3958 char C;
3959 };
3960 struct ST {
3961 int X;
3962 double Y;
3963 struct RT Z;
3964 };
3966 int *foo(struct ST *s) {
3968 }
3969 </pre>
3970 </div>
3975 <pre>
3979 define i32* @foo(%ST* %s) {
3980 entry:
3982 ret i32* %reg
3983 }
3984 </pre>
3985 </div>
3990 }</tt>' type, a structure. The second index indexes into the third element
3992 i8 }</tt>' type, another structure. The third index indexes into the second
3994 array. The two dimensions of the array are subscripted into, yielding an
3998 <p>Note that it is perfectly legal to index partially through a structure,
3999 returning a pointer to an inner element. Because of this, the LLVM code for
4000 the given testcase is equivalent to:</p>
4002 <pre>
4003 define i32* @foo(%ST* %s) {
4009 ret i32* %t5
4010 }
4011 </pre>
4016 that would be formed by successive addition of the offsets implied by the
4017 indices to the base address with infinitely precise arithmetic are not an
4020 that point into the object, plus the address one byte past the end.</p>
4023 the base address with silently-wrapping two's complement arithmetic, and
4025 pointed to by the base pointer. The result value may not necessarily be
4026 used to access memory though, even if it happens to point into allocated
4028 section for more information.</p>
4030 <p>The getelementptr instruction is often confusing. For some more insight into
4034 <pre>
4043 </pre>
4045 </div>
4047 <!-- ======================================================================= -->
4049 </div>
4053 <p>The instructions in this category are the conversion instructions (casting)
4054 which all take a single operand and a type. They perform various bit
4055 conversions on the operand.</p>
4057 </div>
4059 <!-- _______________________________________________________________________ -->
4062 </div>
4066 <pre>
4068 </pre>
4077 size and type of the result, which must be
4080 allowed.</p>
4089 <pre>
4093 </pre>
4095 </div>
4097 <!-- _______________________________________________________________________ -->
4100 </div>
4104 <pre>
4106 </pre>
4127 <pre>
4130 </pre>
4132 </div>
4134 <!-- _______________________________________________________________________ -->
4137 </div>
4141 <pre>
4143 </pre>
4163 <pre>
4166 </pre>
4168 </div>
4170 <!-- _______________________________________________________________________ -->
4173 </div>
4178 <pre>
4180 </pre>
4198 undefined.</p>
4201 <pre>
4204 </pre>
4206 </div>
4208 <!-- _______________________________________________________________________ -->
4211 </div>
4215 <pre>
4217 </pre>
4221 floating point value.</p>
4227 type must be smaller than the destination type.</p>
4237 <pre>
4240 </pre>
4242 </div>
4244 <!-- _______________________________________________________________________ -->
4247 </div>
4251 <pre>
4253 </pre>
4269 towards zero) unsigned integer value. If the value cannot fit
4273 <pre>
4277 </pre>
4279 </div>
4281 <!-- _______________________________________________________________________ -->
4284 </div>
4288 <pre>
4290 </pre>
4308 the results are undefined.</p>
4311 <pre>
4315 </pre>
4317 </div>
4319 <!-- _______________________________________________________________________ -->
4322 </div>
4326 <pre>
4328 </pre>
4343 integer quantity and converts it to the corresponding floating point
4344 value. If the value cannot fit in the floating point value, the results are
4345 undefined.</p>
4348 <pre>
4351 </pre>
4353 </div>
4355 <!-- _______________________________________________________________________ -->
4358 </div>
4362 <pre>
4364 </pre>
4379 quantity and converts it to the corresponding floating point value. If the
4380 value cannot fit in the floating point value, the results are undefined.</p>
4383 <pre>
4386 </pre>
4388 </div>
4390 <!-- _______________________________________________________________________ -->
4393 </div>
4397 <pre>
4399 </pre>
4413 truncating or zero extending that value to the size of the integer type. If
4417 change.</p>
4420 <pre>
4423 </pre>
4425 </div>
4427 <!-- _______________________________________________________________________ -->
4430 </div>
4434 <pre>
4436 </pre>
4444 value to cast, and a type to cast it to, which must be a
4452 than the size of a pointer then a zero extension is done. If they are the
4456 <pre>
4460 </pre>
4462 </div>
4464 <!-- _______________________________________________________________________ -->
4467 </div>
4471 <pre>
4473 </pre>
4481 non-aggregate first class value, and a type to cast it to, which must also be
4484 identical. If the source type is a pointer, the destination type must also be
4485 a pointer. This instruction supports bitwise conversion of vectors to
4486 integers and to vectors of other types (as long as they have the same
4487 size).</p>
4494 be converted to other pointer types with this instruction. To convert
4499 <pre>
4503 </pre>
4505 </div>
4507 <!-- ======================================================================= -->
4513 defy better classification.</p>
4515 </div>
4517 <!-- _______________________________________________________________________ -->
4519 </div>
4524 <pre>
4525 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4526 </pre>
4530 boolean values based on comparison of its two integer, integer vector, or
4531 pointer operands.</p>
4535 the condition code indicating the kind of comparison to perform. It is not a
4536 value, just a keyword. The possible condition code are:</p>
4538 <ol>
4549 </ol>
4553 typed. They must also be identical types.</p>
4559 result, as follows:</p>
4561 <ol>
4564 performed.</li>
4568 performed.</li>
4595 </ol>
4598 values are compared as if they were integers.</p>
4600 <p>If the operands are integer vectors, then they are compared element by
4605 <pre>
4612 </pre>
4614 <p>Note that the code generator does not yet support vector types with
4617 </div>
4619 <!-- _______________________________________________________________________ -->
4621 </div>
4626 <pre>
4627 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4628 </pre>
4632 values based on comparison of its operands.</p>
4634 <p>If the operands are floating point scalars, then the result type is a boolean
4637 <p>If the operands are floating point vectors, then the result type is a vector
4638 of boolean with the same number of elements as the operands being
4639 compared.</p>
4643 the condition code indicating the kind of comparison to perform. It is not a
4644 value, just a keyword. The possible condition code are:</p>
4646 <ol>
4663 </ol>
4671 identical types.</p>
4676 vectors, then the vectors are compared element by element. Each comparison
4678 follows:</p>
4680 <ol>
4724 </ol>
4727 <pre>
4732 </pre>
4734 <p>Note that the code generator does not yet support vector types with
4737 </div>
4739 <!-- _______________________________________________________________________ -->
4742 </div>
4747 <pre>
4749 </pre>
4753 SSA graph representing the function.</p>
4756 <p>The type of the incoming values is specified with the first type field. After
4758 one pair for each predecessor basic block of the current block. Only values
4760 arguments to the PHI node. Only labels may be used as the label
4761 arguments.</p>
4763 <p>There must be no non-phi instructions between the start of a basic block and
4764 the PHI instructions: i.e. PHI instructions must be first in a basic
4765 block.</p>
4767 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
4768 occur on the edge from the corresponding predecessor block to the current
4770 value on the same edge).</p>
4774 specified by the pair corresponding to the predecessor basic block that
4775 executed just prior to the current block.</p>
4778 <pre>
4779 Loop: ; Infinite loop that counts from 0 on up...
4780 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4781 %nextindvar = add i32 %indvar, 1
4782 br label %Loop
4783 </pre>
4785 </div>
4787 <!-- _______________________________________________________________________ -->
4790 </div>
4795 <pre>
4796 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4799 </pre>
4803 condition, without branching.</p>
4808 values indicating the condition, and two values of the
4810 vectors and the condition is a scalar, then entire vectors are selected, not
4811 individual elements.</p>
4815 first value argument; otherwise, it returns the second value argument.</p>
4817 <p>If the condition is a vector of i1, then the value arguments must be vectors
4818 of the same size, and the selection is done element by element.</p>
4821 <pre>
4823 </pre>
4825 <p>Note that the code generator does not yet support conditions
4826 with vector type.</p>
4828 </div>
4830 <!-- _______________________________________________________________________ -->
4833 </div>
4838 <pre>
4839 <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>]
4840 </pre>
4848 <ol>
4850 any allocas or varargs in the caller. If the "tail" marker is present,
4851 the function call is eligible for tail call optimization. Note that calls
4852 may be marked "tail" even if they do not occur before
4856 convention</a> the call should use. If none is specified, the call
4857 defaults to using C calling conventions.</li>
4864 type of the return value. Functions that return no value are marked
4868 being invoked. The argument types must match the types implied by this
4869 signature. This type can be omitted if the function is not varargs and if
4870 the function type does not return a pointer to a function.</li>
4873 be invoked. In most cases, this is a direct function invocation, but
4875 to function value.</li>
4878 signature argument types. All arguments must be of
4880 indicates the function accepts a variable number of arguments, the extra
4881 arguments can be specified.</li>
4886 </ol>
4890 a specified function, with its incoming arguments bound to the specified
4892 function, control flow continues with the instruction after the function
4893 call, and the return value of the function is bound to the result
4894 argument.</p>
4897 <pre>
4898 %retval = call i32 @test(i32 %argc)
4902 call void %foo(i8 97 signext)
4904 %struct.A = type { i32, i8 }
4910 </pre>
4912 </div>
4914 <!-- _______________________________________________________________________ -->
4917 </div>
4922 <pre>
4924 </pre>
4928 the "variable argument" area of a function call. It is used to implement the
4933 argument. It returns a value of the specified argument type and increments
4940 to the next argument. For more information, see the variable argument
4943 <p>It is legal for this instruction to be called in a function which does not
4945 function.</p>
4949 argument.</p>
4954 <p>Note that the code generator does not yet fully support va_arg on many
4955 targets. Also, it does not currently support va_arg with aggregate types on
4956 any target.</p>
4958 </div>
4960 <!-- *********************************************************************** -->
4962 <!-- *********************************************************************** -->
4967 well known names and semantics and are required to follow certain
4968 restrictions. Overall, these intrinsics represent an extension mechanism for
4969 the LLVM language that does not require changing all of the transformations
4970 in LLVM when adding to the language (or the bitcode reader/writer, the
4971 parser, etc...).</p>
4974 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4975 begin with this prefix. Intrinsic functions must always be external
4976 functions: you cannot define the body of intrinsic functions. Intrinsic
4977 functions may only be used in call or invoke instructions: it is illegal to
4978 take the address of an intrinsic function. Additionally, because intrinsic
4979 functions are part of the LLVM language, it is required if any are added that
4980 they be documented here.</p>
4982 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
4983 family of functions that perform the same operation but on different data
4984 types. Because LLVM can represent over 8 million different integer types,
4985 overloading is used commonly to allow an intrinsic function to operate on any
4986 integer type. One or more of the argument types or the result type can be
4987 overloaded to accept any integer type. Argument types may also be defined as
4988 exactly matching a previous argument's type or the result type. This allows
4989 an intrinsic function which accepts multiple arguments, but needs all of them
4990 to be of the same type, to only be overloaded with respect to a single
4991 argument or the result.</p>
4993 <p>Overloaded intrinsics will have the names of its overloaded argument types
4994 encoded into its function name, each preceded by a period. Only those types
4995 which are overloaded result in a name suffix. Arguments whose type is matched
4997 can take an integer of any width and returns an integer of exactly the same
4998 integer width. This leads to a family of functions such as
5000 %val)</tt>. Only one type, the return type, is overloaded, and only one type
5001 suffix is required. Because the argument's type is matched against the return
5002 type, it does not require its own name suffix.</p>
5004 <p>To learn how to add an intrinsic function, please see the
5007 </div>
5009 <!-- ======================================================================= -->
5012 </div>
5016 <p>Variable argument support is defined in LLVM with
5018 intrinsic functions. These functions are related to the similarly named
5021 <p>All of these functions operate on arguments that use a target-specific value
5022 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
5023 not define what this type is, so all transformations should be prepared to
5024 handle these functions regardless of the type used.</p>
5027 instruction and the variable argument handling intrinsic functions are
5028 used.</p>
5031 <pre>
5032 define i32 @test(i32 %X, ...) {
5033 ; Initialize variable argument processing
5034 %ap = alloca i8*
5035 %ap2 = bitcast i8** %ap to i8*
5036 call void @llvm.va_start(i8* %ap2)
5038 ; Read a single integer argument
5039 %tmp = va_arg i8** %ap, i32
5041 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5042 %aq = alloca i8*
5043 %aq2 = bitcast i8** %aq to i8*
5044 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5045 call void @llvm.va_end(i8* %aq2)
5047 ; Stop processing of arguments.
5048 call void @llvm.va_end(i8* %ap2)
5049 ret i32 %tmp
5050 }
5052 declare void @llvm.va_start(i8*)
5053 declare void @llvm.va_copy(i8*, i8*)
5054 declare void @llvm.va_end(i8*)
5055 </pre>
5056 </div>
5058 </div>
5060 <!-- _______________________________________________________________________ -->
5063 </div>
5069 <pre>
5071 </pre>
5082 macro available in C. In a target-dependent way, it initializes
5086 need to know the last argument of the function as the compiler can figure
5087 that out.</p>
5089 </div>
5091 <!-- _______________________________________________________________________ -->
5094 </div>
5099 <pre>
5101 </pre>
5105 which has been initialized previously
5114 macro available in C. In a target-dependent way, it destroys
5120 </div>
5122 <!-- _______________________________________________________________________ -->
5125 </div>
5130 <pre>
5132 </pre>
5136 from the source argument list to the destination argument list.</p>
5141 from.</p>
5145 macro available in C. In a target-dependent way, it copies the
5147 element. This intrinsic is necessary because
5149 arbitrarily complex and require, for example, memory allocation.</p>
5151 </div>
5153 <!-- ======================================================================= -->
5156 </div>
5161 Collection</a> (GC) requires the implementation and generation of these
5163 roots on the stack</a>, as well as garbage collector implementations that
5165 barriers. Front-ends for type-safe garbage collected languages should generate
5166 these intrinsics to make use of the LLVM garbage collectors. For more details,
5170 <p>The garbage collection intrinsics only operate on objects in the generic
5171 address space (address space zero).</p>
5173 </div>
5175 <!-- _______________________________________________________________________ -->
5178 </div>
5183 <pre>
5184 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5185 </pre>
5189 the code generator, and allows some metadata to be associated with it.</p>
5192 <p>The first argument specifies the address of a stack object that contains the
5193 root pointer. The second pointer (which must be either a constant or a
5194 global value address) contains the meta-data to be associated with the
5195 root.</p>
5199 location. At compile-time, the code generator generates information to allow
5204 </div>
5206 <!-- _______________________________________________________________________ -->
5209 </div>
5214 <pre>
5215 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5216 </pre>
5220 locations, allowing garbage collector implementations that require read
5221 barriers.</p>
5224 <p>The second argument is the address to read from, which should be an address
5225 allocated from the garbage collector. The first object is a pointer to the
5226 start of the referenced object, if needed by the language runtime (otherwise
5227 null).</p>
5231 instruction, but may be replaced with substantially more complex code by the
5236 </div>
5238 <!-- _______________________________________________________________________ -->
5241 </div>
5246 <pre>
5247 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5248 </pre>
5252 locations, allowing garbage collector implementations that require write
5253 barriers (such as generational or reference counting collectors).</p>
5256 <p>The first argument is the reference to store, the second is the start of the
5257 object to store it to, and the third is the address of the field of Obj to
5258 store to. If the runtime does not require a pointer to the object, Obj may
5259 be null.</p>
5263 instruction, but may be replaced with substantially more complex code by the
5268 </div>
5270 <!-- ======================================================================= -->
5273 </div>
5277 <p>These intrinsics are provided by LLVM to expose special features that may
5278 only be implemented with code generator support.</p>
5280 </div>
5282 <!-- _______________________________________________________________________ -->
5285 </div>
5290 <pre>
5292 </pre>
5296 target-specific value indicating the return address of the current function
5297 or one of its callers.</p>
5300 <p>The argument to this intrinsic indicates which function to return the address
5301 for. Zero indicates the calling function, one indicates its caller, etc.
5306 indicating the return address of the specified call frame, or zero if it
5307 cannot be identified. The value returned by this intrinsic is likely to be
5308 incorrect or 0 for arguments other than zero, so it should only be used for
5309 debugging purposes.</p>
5311 <p>Note that calling this intrinsic does not prevent function inlining or other
5312 aggressive transformations, so the value returned may not be that of the
5313 obvious source-language caller.</p>
5315 </div>
5317 <!-- _______________________________________________________________________ -->
5320 </div>
5325 <pre>
5327 </pre>
5331 target-specific frame pointer value for the specified stack frame.</p>
5334 <p>The argument to this intrinsic indicates which function to return the frame
5335 pointer for. Zero indicates the calling function, one indicates its caller,
5340 indicating the frame address of the specified call frame, or zero if it
5341 cannot be identified. The value returned by this intrinsic is likely to be
5342 incorrect or 0 for arguments other than zero, so it should only be used for
5343 debugging purposes.</p>
5345 <p>Note that calling this intrinsic does not prevent function inlining or other
5346 aggressive transformations, so the value returned may not be that of the
5347 obvious source-language caller.</p>
5349 </div>
5351 <!-- _______________________________________________________________________ -->
5354 </div>
5359 <pre>
5360 declare i8 *@llvm.stacksave()
5361 </pre>
5365 of the function stack, for use
5367 useful for implementing language features like scoped automatic variable
5368 sized arrays in C99.</p>
5371 <p>This intrinsic returns a opaque pointer value that can be passed
5379 </div>
5381 <!-- _______________________________________________________________________ -->
5384 </div>
5389 <pre>
5390 declare void @llvm.stackrestore(i8 * %ptr)
5391 </pre>
5395 the function stack to the state it was in when the
5397 executed. This is useful for implementing language features like scoped
5398 automatic variable sized arrays in C99.</p>
5401 <p>See the description
5404 </div>
5406 <!-- _______________________________________________________________________ -->
5409 </div>
5414 <pre>
5416 </pre>
5420 insert a prefetch instruction if supported; otherwise, it is a noop.
5421 Prefetches have no effect on the behavior of the program but can change its
5422 performance characteristics.</p>
5432 <p>This intrinsic does not modify the behavior of the program. In particular,
5433 prefetches cannot trap and do not produce a value. On targets that support
5434 this intrinsic, the prefetch can provide hints to the processor cache for
5435 better performance.</p>
5437 </div>
5439 <!-- _______________________________________________________________________ -->
5442 </div>
5447 <pre>
5449 </pre>
5453 Counter (PC) in a region of code to simulators and other tools. The method
5454 is target specific, but it is expected that the marker will use exported
5455 symbols to transmit the PC of the marker. The marker makes no guarantees
5456 that it will remain with any specific instruction after optimizations. It is
5457 possible that the presence of a marker will inhibit optimizations. The
5458 intended use is to be inserted after optimizations to allow correlations of
5459 simulation runs.</p>
5465 <p>This intrinsic does not modify the behavior of the program. Backends that do
5466 not support this intrinisic may ignore it.</p>
5468 </div>
5470 <!-- _______________________________________________________________________ -->
5473 </div>
5478 <pre>
5479 declare i64 @llvm.readcyclecounter( )
5480 </pre>
5484 counter register (or similar low latency, high accuracy clocks) on those
5485 targets that support it. On X86, it should map to RDTSC. On Alpha, it
5486 should map to RPCC. As the backing counters overflow quickly (on the order
5490 <p>When directly supported, reading the cycle counter should not modify any
5491 memory. Implementations are allowed to either return a application specific
5492 value or a system wide value. On backends without support, this is lowered
5495 </div>
5497 <!-- ======================================================================= -->
5500 </div>
5504 <p>LLVM provides intrinsics for a few important standard C library functions.
5505 These intrinsics allow source-language front-ends to pass information about
5506 the alignment of the pointer arguments to the code generator, providing
5507 opportunity for more efficient code generation.</p>
5509 </div>
5511 <!-- _______________________________________________________________________ -->
5514 </div>
5520 integer bit width. Not all targets support all bit widths however.</p>
5522 <pre>
5531 </pre>
5535 source location to the destination location.</p>
5538 intrinsics do not return a value, and takes an extra alignment argument.</p>
5541 <p>The first argument is a pointer to the destination, the second is a pointer
5542 to the source. The third argument is an integer argument specifying the
5543 number of bytes to copy, and the fourth argument is the alignment of the
5544 source and destination locations.</p>
5547 then the caller guarantees that both the source and destination pointers are
5548 aligned to that boundary.</p>
5552 source location to the destination location, which are not allowed to
5553 overlap. It copies "len" bytes of memory over. If the argument is known to
5554 be aligned to some boundary, this can be specified as the fourth argument,
5557 </div>
5559 <!-- _______________________________________________________________________ -->
5562 </div>
5567 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5568 width. Not all targets support all bit widths however.</p>
5570 <pre>
5579 </pre>
5583 source location to the destination location. It is similar to the
5585 overlap.</p>
5588 intrinsics do not return a value, and takes an extra alignment argument.</p>
5591 <p>The first argument is a pointer to the destination, the second is a pointer
5592 to the source. The third argument is an integer argument specifying the
5593 number of bytes to copy, and the fourth argument is the alignment of the
5594 source and destination locations.</p>
5597 then the caller guarantees that the source and destination pointers are
5598 aligned to that boundary.</p>
5602 source location to the destination location, which may overlap. It copies
5603 "len" bytes of memory over. If the argument is known to be aligned to some
5604 boundary, this can be specified as the fourth argument, otherwise it should
5607 </div>
5609 <!-- _______________________________________________________________________ -->
5612 </div>
5617 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5618 width. Not all targets support all bit widths however.</p>
5620 <pre>
5629 </pre>
5633 particular byte value.</p>
5636 intrinsic does not return a value, and takes an extra alignment argument.</p>
5639 <p>The first argument is a pointer to the destination to fill, the second is the
5640 byte value to fill it with, the third argument is an integer argument
5641 specifying the number of bytes to fill, and the fourth argument is the known
5642 alignment of destination location.</p>
5645 then the caller guarantees that the destination pointer is aligned to that
5646 boundary.</p>
5650 at the destination location. If the argument is known to be aligned to some
5651 boundary, this can be specified as the fourth argument, otherwise it should
5654 </div>
5656 <!-- _______________________________________________________________________ -->
5659 </div>
5665 floating point or vector of floating point type. Not all targets support all
5666 types however.</p>
5668 <pre>
5669 declare float @llvm.sqrt.f32(float %Val)
5670 declare double @llvm.sqrt.f64(double %Val)
5671 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5672 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5673 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5674 </pre>
5680 behavior for negative numbers other than -0.0 (which allows for better
5681 optimization, because there is no need to worry about errno being
5685 <p>The argument and return value are floating point numbers of the same
5686 type.</p>
5689 <p>This function returns the sqrt of the specified operand if it is a
5690 nonnegative floating point number.</p>
5692 </div>
5694 <!-- _______________________________________________________________________ -->
5697 </div>
5703 floating point or vector of floating point type. Not all targets support all
5704 types however.</p>
5706 <pre>
5707 declare float @llvm.powi.f32(float %Val, i32 %power)
5708 declare double @llvm.powi.f64(double %Val, i32 %power)
5709 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5710 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5711 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5712 </pre>
5716 specified (positive or negative) power. The order of evaluation of
5717 multiplications is not defined. When a vector of floating point type is
5718 used, the second argument remains a scalar integer value.</p>
5721 <p>The second argument is an integer power, and the first is a value to raise to
5722 that power.</p>
5725 <p>This function returns the first value raised to the second power with an
5726 unspecified sequence of rounding operations.</p>
5728 </div>
5730 <!-- _______________________________________________________________________ -->
5733 </div>
5739 floating point or vector of floating point type. Not all targets support all
5740 types however.</p>
5742 <pre>
5743 declare float @llvm.sin.f32(float %Val)
5744 declare double @llvm.sin.f64(double %Val)
5745 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5746 declare fp128 @llvm.sin.f128(fp128 %Val)
5747 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5748 </pre>
5754 <p>The argument and return value are floating point numbers of the same
5755 type.</p>
5758 <p>This function returns the sine of the specified operand, returning the same
5760 in the same way.</p>
5762 </div>
5764 <!-- _______________________________________________________________________ -->
5767 </div>
5773 floating point or vector of floating point type. Not all targets support all
5774 types however.</p>
5776 <pre>
5777 declare float @llvm.cos.f32(float %Val)
5778 declare double @llvm.cos.f64(double %Val)
5779 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5780 declare fp128 @llvm.cos.f128(fp128 %Val)
5781 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5782 </pre>
5788 <p>The argument and return value are floating point numbers of the same
5789 type.</p>
5792 <p>This function returns the cosine of the specified operand, returning the same
5794 in the same way.</p>
5796 </div>
5798 <!-- _______________________________________________________________________ -->
5801 </div>
5807 floating point or vector of floating point type. Not all targets support all
5808 types however.</p>
5810 <pre>
5811 declare float @llvm.pow.f32(float %Val, float %Power)
5812 declare double @llvm.pow.f64(double %Val, double %Power)
5813 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5814 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5815 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5816 </pre>
5820 specified (positive or negative) power.</p>
5823 <p>The second argument is a floating point power, and the first is a value to
5824 raise to that power.</p>
5827 <p>This function returns the first value raised to the second power, returning
5829 conditions in the same way.</p>
5831 </div>
5833 <!-- ======================================================================= -->
5836 </div>
5840 <p>LLVM provides intrinsics for a few important bit manipulation operations.
5841 These allow efficient code generation for some algorithms.</p>
5843 </div>
5845 <!-- _______________________________________________________________________ -->
5848 </div>
5853 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5856 <pre>
5860 </pre>
5864 values with an even number of bytes (positive multiple of 16 bits). These
5865 are useful for performing operations on data that is not in the target's
5866 native byte order.</p>
5870 and low byte of the input i16 swapped. Similarly,
5876 more, respectively).</p>
5878 </div>
5880 <!-- _______________________________________________________________________ -->
5883 </div>
5888 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5889 width. Not all targets support all bit widths however.</p>
5891 <pre>
5897 </pre>
5901 in a value.</p>
5904 <p>The only argument is the value to be counted. The argument may be of any
5905 integer type. The return type must match the argument type.</p>
5910 </div>
5912 <!-- _______________________________________________________________________ -->
5915 </div>
5921 integer bit width. Not all targets support all bit widths however.</p>
5923 <pre>
5929 </pre>
5933 leading zeros in a variable.</p>
5936 <p>The only argument is the value to be counted. The argument may be of any
5937 integer type. The return type must match the argument type.</p>
5941 zeros in a variable. If the src == 0 then the result is the size in bits of
5944 </div>
5946 <!-- _______________________________________________________________________ -->
5949 </div>
5955 integer bit width. Not all targets support all bit widths however.</p>
5957 <pre>
5963 </pre>
5967 trailing zeros.</p>
5970 <p>The only argument is the value to be counted. The argument may be of any
5971 integer type. The return type must match the argument type.</p>
5975 zeros in a variable. If the src == 0 then the result is the size in bits of
5978 </div>
5980 <!-- ======================================================================= -->
5983 </div>
5989 </div>
5991 <!-- _______________________________________________________________________ -->
5994 </div>
6000 on any integer bit width.</p>
6002 <pre>
6003 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6004 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6005 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6006 </pre>
6010 a signed addition of the two arguments, and indicate whether an overflow
6011 occurred during the signed summation.</p>
6014 <p>The arguments (%a and %b) and the first element of the result structure may
6015 be of integer types of any bit width, but they must have the same bit
6016 width. The second element of the result structure must be of
6018 undergo signed addition.</p>
6022 a signed addition of the two variables. They return a structure — the
6023 first element of which is the signed summation, and the second element of
6024 which is a bit specifying if the signed summation resulted in an
6025 overflow.</p>
6028 <pre>
6029 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6030 %sum = extractvalue {i32, i1} %res, 0
6031 %obit = extractvalue {i32, i1} %res, 1
6032 br i1 %obit, label %overflow, label %normal
6033 </pre>
6035 </div>
6037 <!-- _______________________________________________________________________ -->
6040 </div>
6046 on any integer bit width.</p>
6048 <pre>
6049 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6050 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6051 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6052 </pre>
6056 an unsigned addition of the two arguments, and indicate whether a carry
6057 occurred during the unsigned summation.</p>
6060 <p>The arguments (%a and %b) and the first element of the result structure may
6061 be of integer types of any bit width, but they must have the same bit
6062 width. The second element of the result structure must be of
6064 undergo unsigned addition.</p>
6068 an unsigned addition of the two arguments. They return a structure —
6069 the first element of which is the sum, and the second element of which is a
6070 bit specifying if the unsigned summation resulted in a carry.</p>
6073 <pre>
6074 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6075 %sum = extractvalue {i32, i1} %res, 0
6076 %obit = extractvalue {i32, i1} %res, 1
6077 br i1 %obit, label %carry, label %normal
6078 </pre>
6080 </div>
6082 <!-- _______________________________________________________________________ -->
6085 </div>
6091 on any integer bit width.</p>
6093 <pre>
6094 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6095 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6096 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6097 </pre>
6101 a signed subtraction of the two arguments, and indicate whether an overflow
6102 occurred during the signed subtraction.</p>
6105 <p>The arguments (%a and %b) and the first element of the result structure may
6106 be of integer types of any bit width, but they must have the same bit
6107 width. The second element of the result structure must be of
6109 undergo signed subtraction.</p>
6113 a signed subtraction of the two arguments. They return a structure —
6114 the first element of which is the subtraction, and the second element of
6115 which is a bit specifying if the signed subtraction resulted in an
6116 overflow.</p>
6119 <pre>
6120 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6121 %sum = extractvalue {i32, i1} %res, 0
6122 %obit = extractvalue {i32, i1} %res, 1
6123 br i1 %obit, label %overflow, label %normal
6124 </pre>
6126 </div>
6128 <!-- _______________________________________________________________________ -->
6131 </div>
6137 on any integer bit width.</p>
6139 <pre>
6140 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6141 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6142 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6143 </pre>
6147 an unsigned subtraction of the two arguments, and indicate whether an
6148 overflow occurred during the unsigned subtraction.</p>
6151 <p>The arguments (%a and %b) and the first element of the result structure may
6152 be of integer types of any bit width, but they must have the same bit
6153 width. The second element of the result structure must be of
6155 undergo unsigned subtraction.</p>
6159 an unsigned subtraction of the two arguments. They return a structure —
6160 the first element of which is the subtraction, and the second element of
6161 which is a bit specifying if the unsigned subtraction resulted in an
6162 overflow.</p>
6165 <pre>
6166 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6167 %sum = extractvalue {i32, i1} %res, 0
6168 %obit = extractvalue {i32, i1} %res, 1
6169 br i1 %obit, label %overflow, label %normal
6170 </pre>
6172 </div>
6174 <!-- _______________________________________________________________________ -->
6177 </div>
6183 on any integer bit width.</p>
6185 <pre>
6186 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6187 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6188 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6189 </pre>
6194 a signed multiplication of the two arguments, and indicate whether an
6195 overflow occurred during the signed multiplication.</p>
6198 <p>The arguments (%a and %b) and the first element of the result structure may
6199 be of integer types of any bit width, but they must have the same bit
6200 width. The second element of the result structure must be of
6202 undergo signed multiplication.</p>
6206 a signed multiplication of the two arguments. They return a structure —
6207 the first element of which is the multiplication, and the second element of
6208 which is a bit specifying if the signed multiplication resulted in an
6209 overflow.</p>
6212 <pre>
6213 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6214 %sum = extractvalue {i32, i1} %res, 0
6215 %obit = extractvalue {i32, i1} %res, 1
6216 br i1 %obit, label %overflow, label %normal
6217 </pre>
6219 </div>
6221 <!-- _______________________________________________________________________ -->
6224 </div>
6230 on any integer bit width.</p>
6232 <pre>
6233 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6234 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6235 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6236 </pre>
6240 a unsigned multiplication of the two arguments, and indicate whether an
6241 overflow occurred during the unsigned multiplication.</p>
6244 <p>The arguments (%a and %b) and the first element of the result structure may
6245 be of integer types of any bit width, but they must have the same bit
6246 width. The second element of the result structure must be of
6248 undergo unsigned multiplication.</p>
6252 an unsigned multiplication of the two arguments. They return a structure
6253 — the first element of which is the multiplication, and the second
6254 element of which is a bit specifying if the unsigned multiplication resulted
6255 in an overflow.</p>
6258 <pre>
6259 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6260 %sum = extractvalue {i32, i1} %res, 0
6261 %obit = extractvalue {i32, i1} %res, 1
6262 br i1 %obit, label %overflow, label %normal
6263 </pre>
6265 </div>
6267 <!-- ======================================================================= -->
6270 </div>
6275 prefix), are described in
6279 </div>
6281 <!-- ======================================================================= -->
6284 </div>
6288 <p>The LLVM exception handling intrinsics (which all start with
6293 </div>
6295 <!-- ======================================================================= -->
6298 </div>
6302 <p>This intrinsic makes it possible to excise one parameter, marked with
6304 function pointer lacking the nest parameter - the caller does not need to
6305 provide a value for it. Instead, the value to use is stored in advance in a
6306 "trampoline", a block of memory usually allocated on the stack, which also
6307 contains code to splice the nest value into the argument list. This is used
6308 to implement the GCC nested function address extension.</p>
6310 <p>For example, if the function is
6313 follows:</p>
6316 <pre>
6319 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6320 %fp = bitcast i8* %p to i32 (i32, i32)*
6321 </pre>
6322 </div>
6327 </div>
6329 <!-- _______________________________________________________________________ -->
6332 </div>
6337 <pre>
6339 </pre>
6343 function pointer suitable for executing it.</p>
6348 sufficiently aligned block of memory; this memory is written to by the
6349 intrinsic. Note that the size and the alignment are target-specific - LLVM
6350 currently provides no portable way of determining them, so a front-end that
6351 generates this intrinsic needs to have some target-specific knowledge.
6357 dependent code, turning it into a function. A pointer to this function is
6359 function pointer type</a> before being called. The new function's signature
6362 is allowed, and it must be of pointer type. Calling the new function is
6367 returned function pointer is undefined.</p>
6369 </div>
6371 <!-- ======================================================================= -->
6374 </div>
6379 hardware constructs for atomic operations and memory synchronization. This
6380 provides an interface to the hardware, not an interface to the programmer. It
6381 is aimed at a low enough level to allow any programming models or APIs
6382 (Application Programming Interfaces) which need atomic behaviors to map
6383 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
6384 hardware provides a "universal IR" for source languages, it also provides a
6385 starting point for developing a "universal" atomic operation and
6386 synchronization IR.</p>
6389 software transaction memory systems, atomic primitives, and intrinsic
6390 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6391 application libraries. The hardware interface provided by LLVM should allow
6392 a clean implementation of all of these APIs and parallel programming models.
6393 No one model or paradigm should be selected above others unless the hardware
6394 itself ubiquitously does so.</p>
6396 </div>
6398 <!-- _______________________________________________________________________ -->
6401 </div>
6404 <pre>
6405 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device> )
6406 </pre>
6410 specific pairs of memory access types.</p>
6414 The first four arguments enables a specific barrier as listed below. The
6415 fith argument specifies that the barrier applies to io or device or uncached
6416 memory.</p>
6418 <ul>
6424 </ul>
6427 <p>This intrinsic causes the system to enforce some ordering constraints upon
6428 the loads and stores of the program. This barrier does not
6431 and store operations (f.ex. load-load, or store-load), all of the first
6432 operations preceding the barrier will complete before any of the second
6433 operations succeeding the barrier begin. Specifically the semantics for each
6434 pairing is as follows:</p>
6436 <ul>
6438 after the barrier begins.</li>
6440 store after the barrier begins.</li>
6442 store after the barrier begins.</li>
6444 load after the barrier begins.</li>
6445 </ul>
6448 is enabled in a single memory barrier intrinsic.</p>
6450 <p>Backends may implement stronger barriers than those requested when they do
6451 not support as fine grained a barrier as requested. Some architectures do
6452 not need all types of barriers and on such architectures, these become
6453 noops.</p>
6456 <pre>
6457 %ptr = malloc i32
6458 store i32 4, %ptr
6461 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6464 </pre>
6466 </div>
6468 <!-- _______________________________________________________________________ -->
6471 </div>
6477 any integer bit width and for different address spaces. Not all targets
6478 support all bit widths however.</p>
6480 <pre>
6482 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6483 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6484 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6485 </pre>
6488 <p>This loads a value in memory and compares it to a given value. If they are
6489 equal, it stores a new value into the memory.</p>
6495 this integer type. While any bit width integer may be used, targets may only
6496 lower representations they support in hardware.</p>
6499 <p>This entire intrinsic must be executed atomically. It first loads the value
6502 memory. The loaded value is yielded in all cases. This provides the
6503 equivalent of an atomic compare-and-swap operation within the SSA
6504 framework.</p>
6507 <pre>
6508 %ptr = malloc i32
6509 store i32 4, %ptr
6512 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6518 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6523 </pre>
6525 </div>
6527 <!-- _______________________________________________________________________ -->
6530 </div>
6535 integer bit width. Not all targets support all bit widths however.</p>
6537 <pre>
6542 </pre>
6553 integer type. The targets may only lower integer representations they
6554 support.</p>
6559 equivalent of an atomic swap operation within the SSA framework.</p>
6562 <pre>
6563 %ptr = malloc i32
6564 store i32 4, %ptr
6567 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6573 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6578 </pre>
6580 </div>
6582 <!-- _______________________________________________________________________ -->
6586 </div>
6592 any integer bit width. Not all targets support all bit widths however.</p>
6594 <pre>
6599 </pre>
6606 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6607 and the second an integer value. The result is also an integer value. These
6608 integer types can have any bit width, but they must all have the same bit
6609 width. The targets may only lower integer representations they support.</p>
6612 <p>This intrinsic does a series of operations atomically. It first loads the
6617 <pre>
6618 %ptr = malloc i32
6619 store i32 4, %ptr
6620 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6622 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6624 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6627 </pre>
6629 </div>
6631 <!-- _______________________________________________________________________ -->
6635 </div>
6641 any integer bit width and for different address spaces. Not all targets
6642 support all bit widths however.</p>
6644 <pre>
6649 </pre>
6656 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6657 and the second an integer value. The result is also an integer value. These
6658 integer types can have any bit width, but they must all have the same bit
6659 width. The targets may only lower integer representations they support.</p>
6662 <p>This intrinsic does a series of operations atomically. It first loads the
6668 <pre>
6669 %ptr = malloc i32
6670 store i32 8, %ptr
6671 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6673 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6675 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6678 </pre>
6680 </div>
6682 <!-- _______________________________________________________________________ -->
6688 </div>
6693 <p>These are overloaded intrinsics. You can
6696 bit width and for different address spaces. Not all targets support all bit
6697 widths however.</p>
6699 <pre>
6704 </pre>
6706 <pre>
6711 </pre>
6713 <pre>
6718 </pre>
6720 <pre>
6725 </pre>
6733 <p>These intrinsics take two arguments, the first a pointer to an integer value
6734 and the second an integer value. The result is also an integer value. These
6735 integer types can have any bit width, but they must all have the same bit
6736 width. The targets may only lower integer representations they support.</p>
6739 <p>These intrinsics does a series of operations atomically. They first load the
6745 <pre>
6746 %ptr = malloc i32
6747 store i32 0x0F0F, %ptr
6748 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6750 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6752 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6754 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6757 </pre>
6759 </div>
6761 <!-- _______________________________________________________________________ -->
6767 </div>
6775 address spaces. Not all targets support all bit widths however.</p>
6777 <pre>
6782 </pre>
6784 <pre>
6789 </pre>
6791 <pre>
6796 </pre>
6798 <pre>
6803 </pre>
6806 <p>These intrinsics takes the signed or unsigned minimum or maximum of
6811 <p>These intrinsics take two arguments, the first a pointer to an integer value
6812 and the second an integer value. The result is also an integer value. These
6813 integer types can have any bit width, but they must all have the same bit
6814 width. The targets may only lower integer representations they support.</p>
6817 <p>These intrinsics does a series of operations atomically. They first load the
6823 <pre>
6824 %ptr = malloc i32
6825 store i32 7, %ptr
6826 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6828 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6830 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6832 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6835 </pre>
6837 </div>
6839 <!-- ======================================================================= -->
6842 </div>
6846 <p>This class of intrinsics is designed to be generic and has no specific
6847 purpose.</p>
6849 </div>
6851 <!-- _______________________________________________________________________ -->
6854 </div>
6859 <pre>
6860 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6861 </pre>
6867 <p>The first argument is a pointer to a value, the second is a pointer to a
6868 global string, the third is a pointer to a global string which is the source
6869 file name, and the last argument is the line number.</p>
6872 <p>This intrinsic allows annotation of local variables with arbitrary strings.
6873 This can be useful for special purpose optimizations that want to look for
6874 these annotations. These have no other defined use, they are ignored by code
6875 generation and optimization.</p>
6877 </div>
6879 <!-- _______________________________________________________________________ -->
6882 </div>
6888 any integer bit width.</p>
6890 <pre>
6891 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6892 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6893 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6894 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6895 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6896 </pre>
6902 <p>The first argument is an integer value (result of some expression), the
6903 second is a pointer to a global string, the third is a pointer to a global
6904 string which is the source file name, and the last argument is the line
6905 number. It returns the value of the first argument.</p>
6908 <p>This intrinsic allows annotations to be put on arbitrary expressions with
6909 arbitrary strings. This can be useful for special purpose optimizations that
6910 want to look for these annotations. These have no other defined use, they
6911 are ignored by code generation and optimization.</p>
6913 </div>
6915 <!-- _______________________________________________________________________ -->
6918 </div>
6923 <pre>
6924 declare void @llvm.trap()
6925 </pre>
6934 <p>This intrinsics is lowered to the target dependent trap instruction. If the
6935 target does not have a trap instruction, this intrinsic will be lowered to
6938 </div>
6940 <!-- _______________________________________________________________________ -->
6943 </div>
6948 <pre>
6950 </pre>
6955 ensure that it is placed on the stack before local variables.</p>
6959 arguments. The first argument is the value loaded from the stack
6961 that has enough space to hold the value of the guard.</p>
6964 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
6966 stack. This is to ensure that if a local variable on the stack is
6967 overwritten, it will destroy the value of the guard. When the function exits,
6968 the guard on the stack is checked against the original guard. If they're
6970 function.</p>
6972 </div>
6974 <!-- *********************************************************************** -->
6975 <hr>
6976 <address>
6984 Last modified: $Date$
6985 </address>
6987 </body>
6988 </html>