| blob | be8ff73331801eb5e13f89d268a1f6592c9034f7 |
3 <html>
4 <head>
11 </head>
13 <body>
16 <ol>
21 <ol>
34 </ol>
35 </li>
37 <ol>
40 <ol>
44 </ol>
45 </li>
47 <ol>
56 </ol>
57 </li>
59 </ol>
60 </li>
62 <ol>
69 </ol>
70 </li>
72 <ol>
74 </ol>
75 </li>
77 <ol>
79 <ol>
86 </ol>
87 </li>
89 <ol>
99 </ol>
100 </li>
102 <ol>
109 </ol>
110 </li>
112 <ol>
116 </ol>
117 </li>
119 <ol>
122 </ol>
123 </li>
125 <ol>
132 </ol>
133 </li>
135 <ol>
148 </ol>
149 </li>
151 <ol>
160 </ol>
161 </li>
162 </ol>
163 </li>
165 <ol>
167 <ol>
171 </ol>
172 </li>
174 <ol>
178 </ol>
179 </li>
181 <ol>
189 </ol>
190 </li>
192 <ol>
201 </ol>
202 </li>
204 <ol>
211 </ol>
212 </li>
214 <ol>
221 </ol>
222 </li>
226 <ol>
228 </ol>
229 </li>
231 <ol>
245 </ol>
246 </li>
248 <ol>
257 </ol>
258 </li>
259 </ol>
260 </li>
261 </ol>
266 </div>
268 <!-- *********************************************************************** -->
270 <!-- *********************************************************************** -->
273 <p>This document is a reference manual for the LLVM assembly language.
274 LLVM is a Static Single Assignment (SSA) based representation that provides
275 type safety, low-level operations, flexibility, and the capability of
276 representing 'all' high-level languages cleanly. It is the common code
277 representation used throughout all phases of the LLVM compilation
278 strategy.</p>
279 </div>
281 <!-- *********************************************************************** -->
283 <!-- *********************************************************************** -->
287 <p>The LLVM code representation is designed to be used in three
288 different forms: as an in-memory compiler IR, as an on-disk bitcode
289 representation (suitable for fast loading by a Just-In-Time compiler),
290 and as a human readable assembly language representation. This allows
291 LLVM to provide a powerful intermediate representation for efficient
292 compiler transformations and analysis, while providing a natural means
293 to debug and visualize the transformations. The three different forms
294 of LLVM are all equivalent. This document describes the human readable
295 representation and notation.</p>
297 <p>The LLVM representation aims to be light-weight and low-level
298 while being expressive, typed, and extensible at the same time. It
299 aims to be a "universal IR" of sorts, by being at a low enough level
300 that high-level ideas may be cleanly mapped to it (similar to how
301 microprocessors are "universal IR's", allowing many source languages to
302 be mapped to them). By providing type information, LLVM can be used as
303 the target of optimizations: for example, through pointer analysis, it
304 can be proven that a C automatic variable is never accessed outside of
305 the current function... allowing it to be promoted to a simple SSA
306 value instead of a memory location.</p>
308 </div>
310 <!-- _______________________________________________________________________ -->
315 <p>It is important to note that this document describes 'well formed'
316 LLVM assembly language. There is a difference between what the parser
317 accepts and what is considered 'well formed'. For example, the
318 following instruction is syntactically okay, but not well formed:</p>
321 <pre>
323 </pre>
324 </div>
327 its uses. The LLVM infrastructure provides a verification pass that may
328 be used to verify that an LLVM module is well formed. This pass is
329 automatically run by the parser after parsing input assembly and by
330 the optimizer before it outputs bitcode. The violations pointed out
331 by the verifier pass indicate bugs in transformation passes or input to
332 the parser.</p>
333 </div>
335 <!-- Describe the typesetting conventions here. -->
337 <!-- *********************************************************************** -->
339 <!-- *********************************************************************** -->
343 <p>LLVM identifiers come in two basic types: global and local. Global
344 identifiers (functions, global variables) begin with the @ character. Local
345 identifiers (register names, types) begin with the % character. Additionally,
346 there are three different formats for identifiers, for different purposes:</p>
348 <ol>
349 <li>Named values are represented as a string of characters with their prefix.
350 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
352 Identifiers which require other characters in their names can be surrounded
353 with quotes. Special characters may be escaped using "\xx" where xx is the
354 ASCII code for the character in hexadecimal. In this way, any character can
355 be used in a name value, even quotes themselves.
357 <li>Unnamed values are represented as an unsigned numeric value with their
362 </ol>
364 <p>LLVM requires that values start with a prefix for two reasons: Compilers
365 don't need to worry about name clashes with reserved words, and the set of
366 reserved words may be expanded in the future without penalty. Additionally,
367 unnamed identifiers allow a compiler to quickly come up with a temporary
368 variable without having to avoid symbol table conflicts.</p>
370 <p>Reserved words in LLVM are very similar to reserved words in other
371 languages. There are keywords for different opcodes
376 and others. These reserved words cannot conflict with variable names, because
377 none of them start with a prefix character ('%' or '@').</p>
379 <p>Here is an example of LLVM code to multiply the integer variable
385 <pre>
387 </pre>
388 </div>
393 <pre>
395 </pre>
396 </div>
401 <pre>
405 </pre>
406 </div>
409 important lexical features of LLVM:</p>
411 <ol>
414 line.</li>
416 <li>Unnamed temporaries are created when the result of a computation is not
417 assigned to a named value.</li>
421 </ol>
423 <p>...and it also shows a convention that we follow in this document. When
424 demonstrating instructions, we will follow an instruction with a comment that
425 defines the type and name of value produced. Comments are shown in italic
426 text.</p>
428 </div>
430 <!-- *********************************************************************** -->
432 <!-- *********************************************************************** -->
434 <!-- ======================================================================= -->
436 </div>
441 translation unit of the input programs. Each module consists of
442 functions, global variables, and symbol table entries. Modules may be
443 combined together with the LLVM linker, which merges function (and
444 global variable) definitions, resolves forward declarations, and merges
450 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
458 %cast210 = <a
462 <a
464 <a
466 </pre>
467 </div>
474 <p>In general, a module is made up of a list of global values,
475 where both functions and global variables are global values. Global values are
476 represented by a pointer to a memory location (in this case, a pointer to an
477 array of char, and a pointer to a function), and have one of the following <a
480 </div>
482 <!-- ======================================================================= -->
485 </div>
489 <p>
490 All Global Variables and Functions have one of the following types of linkage:
491 </p>
493 <dl>
497 <dd>Global values with private linkage are only directly accessible by
498 objects in the current module. In particular, linking code into a module with
499 an private global value may cause the private to be renamed as necessary to
500 avoid collisions. Because the symbol is private to the module, all
501 references can be updated. This doesn't show up in any symbol table in the
502 object file.
503 </dd>
507 <dd> Similar to private, but the value shows as a local symbol (STB_LOCAL in
508 the case of ELF) in the object file. This corresponds to the notion of the
510 </dd>
513 </dt>
516 into the object file corresponding to the LLVM module. They exist to
517 allow inlining and other optimizations to take place given knowledge of the
518 definition of the global, which is known to be somewhere outside the module.
521 type is only allowed on definitions, not declarations.</dd>
526 the same name when linkage occurs. This is typically used to implement
527 inline functions, templates, or other code which must be generated in each
529 allowed to be discarded.
530 </dd>
536 discarded. This is used for globals that may be emitted in multiple
537 translation units, but that are not guaranteed to be emitted into every
538 translation unit that uses them. One example of this is tentative
539 definitions in C, such as "<tt>int X;</tt>" at global scope.
540 </dd>
545 that some targets may choose to emit different assembly sequences for them
546 for target-dependent reasons. This is used for globals that are declared
547 "weak" in C source code.
548 </dd>
553 pointer to array type. When two global variables with appending linkage are
554 linked together, the two global arrays are appended together. This is the
555 LLVM, typesafe, equivalent of having the system linker append together
556 "sections" with identical names when .o files are linked.
557 </dd>
561 <dd>The semantics of this linkage follow the ELF object file model: the
562 symbol is weak until linked, if not linked, the symbol becomes null instead
563 of being an undefined reference.
564 </dd>
568 <dd>Some languages allow differing globals to be merged, such as two
570 ensure that only equivalent globals are ever merged (the "one definition
573 be merged with equivalent globals. These linkage types are otherwise the
575 </dd>
579 <dd>If none of the above identifiers are used, the global is externally
580 visible, meaning that it participates in linkage and can be used to resolve
581 external symbol references.
582 </dd>
583 </dl>
585 <p>
586 The next two types of linkage are targeted for Microsoft Windows platform
587 only. They are designed to support importing (exporting) symbols from (to)
588 DLLs (Dynamic Link Libraries).
589 </p>
591 <dl>
595 or variable via a global pointer to a pointer that is set up by the DLL
596 exporting the symbol. On Microsoft Windows targets, the pointer name is
598 </dd>
603 pointer to a pointer in a DLL, so that it can be referenced with the
606 name.
607 </dd>
609 </dl>
612 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
613 variable and was linked with this one, one of the two would be renamed,
615 external (i.e., lacking any linkage declarations), they are accessible
616 outside of the current module.</p>
622 </div>
624 <!-- ======================================================================= -->
627 </div>
633 specified for the call. The calling convention of any pair of dynamic
634 caller/callee must match, or the behavior of the program is undefined. The
635 following calling conventions are supported by LLVM, and more may be added in
636 the future:</p>
638 <dl>
641 <dd>This calling convention (the default if no other calling convention is
642 specified) matches the target C calling conventions. This calling convention
643 supports varargs function calls and tolerates some mismatch in the declared
644 prototype and implemented declaration of the function (as does normal C).
645 </dd>
649 <dd>This calling convention attempts to make calls as fast as possible
650 (e.g. by passing things in registers). This calling convention allows the
651 target to use whatever tricks it wants to produce fast code for the target,
652 without having to conform to an externally specified ABI (Application Binary
653 Interface). Implementations of this convention should allow arbitrary
655 supported. This calling convention does not support varargs and requires the
656 prototype of all callees to exactly match the prototype of the function
657 definition.
658 </dd>
662 <dd>This calling convention attempts to make code in the caller as efficient
663 as possible under the assumption that the call is not commonly executed. As
664 such, these calls often preserve all registers so that the call does not break
665 any live ranges in the caller side. This calling convention does not support
666 varargs and requires the prototype of all callees to exactly match the
667 prototype of the function definition.
668 </dd>
672 <dd>Any calling convention may be specified by number, allowing
673 target-specific calling conventions to be used. Target specific calling
674 conventions start at 64.
675 </dd>
676 </dl>
678 <p>More calling conventions can be added/defined on an as-needed basis, to
679 support pascal conventions or any other well-known target-independent
680 convention.</p>
682 </div>
684 <!-- ======================================================================= -->
687 </div>
691 <p>
692 All Global Variables and Functions have one of the following visibility styles:
693 </p>
695 <dl>
698 <dd>On targets that use the ELF object file format, default visibility means
699 that the declaration is visible to other
700 modules and, in shared libraries, means that the declared entity may be
701 overridden. On Darwin, default visibility means that the declaration is
702 visible to other modules. Default visibility corresponds to "external
703 linkage" in the language.
704 </dd>
708 <dd>Two declarations of an object with hidden visibility refer to the same
709 object if they are in the same shared object. Usually, hidden visibility
710 indicates that the symbol will not be placed into the dynamic symbol table,
711 so no other module (executable or shared library) can reference it
712 directly.
713 </dd>
717 <dd>On ELF, protected visibility indicates that the symbol will be placed in
718 the dynamic symbol table, but that references within the defining module will
719 bind to the local symbol. That is, the symbol cannot be overridden by another
720 module.
721 </dd>
722 </dl>
724 </div>
726 <!-- ======================================================================= -->
729 </div>
733 <p>LLVM IR allows you to specify name aliases for certain types. This can make
734 it easier to read the IR and make the IR more condensed (particularly when
735 recursive types are involved). An example of a name specification is:
736 </p>
739 <pre>
740 %mytype = type { %mytype*, i32 }
741 </pre>
742 </div>
748 <p>Note that type names are aliases for the structural type that they indicate,
749 and that you can therefore specify multiple names for the same type. This often
750 leads to confusing behavior when dumping out a .ll file. Since LLVM IR uses
751 structural typing, the name is not part of the type. When printing out LLVM IR,
753 shape. This means that if you have code where two different source types end up
754 having the same LLVM type, that the dumper will sometimes print the "wrong" or
755 unexpected type. This is an important design point and isn't going to
756 change.</p>
758 </div>
760 <!-- ======================================================================= -->
763 </div>
767 <p>Global variables define regions of memory allocated at compilation time
768 instead of run-time. Global variables may optionally be initialized, may have
769 an explicit section to be placed in, and may have an optional explicit alignment
770 specified. A variable may be defined as "thread_local", which means that it
771 will not be shared by threads (each thread will have a separated copy of the
772 variable). A variable may be defined as a global "constant," which indicates
774 optimization, allowing the global data to be placed in the read-only section of
775 an executable, etc). Note that variables that need runtime initialization
778 <p>
780 constant, even if the final definition of the global is not. This capability
781 can be used to enable slightly better optimization of the program, but requires
782 the language definition to guarantee that optimizations based on the
783 'constantness' are valid for the translation units that do not include the
784 definition.
785 </p>
787 <p>As SSA values, global variables define pointer values that are in
788 scope (i.e. they dominate) all basic blocks in the program. Global
789 variables always define a pointer to their "content" type because they
790 describe a region of memory, and all memory objects in LLVM are
791 accessed through pointers.</p>
793 <p>A global variable may be declared to reside in a target-specifc numbered
794 address space. For targets that support them, address spaces may affect how
795 optimizations are performed and/or what target instructions are used to access
796 the variable. The default address space is zero. The address space qualifier
797 must precede any other attributes.</p>
799 <p>LLVM allows an explicit section to be specified for globals. If the target
800 supports it, it will emit globals to the section specified.</p>
802 <p>An explicit alignment may be specified for a global. If not present, or if
803 the alignment is set to zero, the alignment of the global is set by the target
804 to whatever it feels convenient. If an explicit alignment is specified, the
805 global is forced to have at least that much alignment. All alignments must be
808 <p>For example, the following defines a global in a numbered address space with
809 an initializer, section, and alignment:</p>
812 <pre>
814 </pre>
815 </div>
817 </div>
820 <!-- ======================================================================= -->
823 </div>
832 name, a (possibly empty) argument list (each with optional
836 an opening curly brace, a list of basic blocks, and a closing curly brace.
838 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
843 name, a possibly empty list of arguments, an optional alignment, and an optional
846 <p>A function definition contains a list of basic blocks, forming the CFG
847 (Control Flow Graph) for
848 the function. Each basic block may optionally start with a label (giving the
849 basic block a symbol table entry), contains a list of instructions, and ends
851 function return).</p>
853 <p>The first basic block in a function is special in two ways: it is immediately
854 executed on entrance to the function, and it is not allowed to have predecessor
855 basic blocks (i.e. there can not be any branches to the entry block of a
856 function). Because the block can have no predecessors, it also cannot have any
859 <p>LLVM allows an explicit section to be specified for functions. If the target
860 supports it, it will emit functions to the section specified.</p>
862 <p>An explicit alignment may be specified for a function. If not present, or if
863 the alignment is set to zero, the alignment of the function is set by the target
864 to whatever it feels convenient. If an explicit alignment is specified, the
865 function is forced to have at least that much alignment. All alignments must be
871 <tt>
877 </tt>
878 </div>
880 </div>
883 <!-- ======================================================================= -->
886 </div>
889 function, global variable, another alias or bitcast of global value). Aliases
896 <pre>
898 </pre>
899 </div>
901 </div>
905 <!-- ======================================================================= -->
908 <p>The return type and each parameter of a function type may have a set of
910 used to communicate additional information about the result or parameters of
911 a function. Parameter attributes are considered to be part of the function,
912 not of the function type, so functions with different parameter attributes
913 can have the same function type.</p>
915 <p>Parameter attributes are simple keywords that follow the type specified. If
916 multiple parameter attributes are needed, they are space separated. For
917 example:</p>
920 <pre>
921 declare i32 @printf(i8* noalias nocapture, ...)
922 declare i32 @atoi(i8 zeroext)
923 declare signext i8 @returns_signed_char()
924 </pre>
925 </div>
931 <dl>
933 <dd>This indicates to the code generator that the parameter or return value
934 should be zero-extended to a 32-bit value by the caller (for a parameter)
935 or the callee (for a return value).</dd>
938 <dd>This indicates to the code generator that the parameter or return value
939 should be sign-extended to a 32-bit value by the caller (for a parameter)
940 or the callee (for a return value).</dd>
943 <dd>This indicates that this parameter or return value should be treated
944 in a special target-dependent fashion during while emitting code for a
945 function call or return (usually, by putting it in a register as opposed
946 to memory, though some targets use it to distinguish between two different
947 kinds of registers). Use of this attribute is target-specific.</dd>
950 <dd>This indicates that the pointer parameter should really be passed by
951 value to the function. The attribute implies that a hidden copy of the
952 pointee is made between the caller and the callee, so the callee is unable
953 to modify the value in the callee. This attribute is only valid on LLVM
954 pointer arguments. It is generally used to pass structs and arrays by
955 value, but is also valid on pointers to scalars. The copy is considered to
956 belong to the caller not the callee (for example,
959 values. The byval attribute also supports specifying an alignment with the
960 align attribute. This has a target-specific effect on the code generator
961 that usually indicates a desired alignment for the synthesized stack
962 slot.</dd>
965 <dd>This indicates that the pointer parameter specifies the address of a
966 structure that is the return value of the function in the source program.
967 This pointer must be guaranteed by the caller to be valid: loads and stores
968 to the structure may be assumed by the callee to not to trap. This may only
969 be applied to the first parameter. This is not a valid attribute for
970 return values. </dd>
973 <dd>This indicates that the pointer does not alias any global or any other
974 parameter. The caller is responsible for ensuring that this is the
976 that the pointer does not alias any other pointers visible to the
977 caller. For further details, please see the discussion of the NoAlias
978 response in
983 <dd>This indicates that the callee does not make any copies of the pointer
984 that outlive the callee itself. This is not a valid attribute for return
985 values.</dd>
988 <dd>This indicates that the pointer parameter can be excised using the
990 attribute for return values.</dd>
991 </dl>
993 </div>
995 <!-- ======================================================================= -->
998 </div>
1001 <p>Each function may specify a garbage collector name, which is simply a
1002 string.</p>
1008 collector which will cause the compiler to alter its output in order to support
1009 the named garbage collection algorithm.</p>
1010 </div>
1012 <!-- ======================================================================= -->
1015 </div>
1019 <p>Function attributes are set to communicate additional information about
1020 a function. Function attributes are considered to be part of the function,
1021 not of the function type, so functions with different parameter attributes
1022 can have the same function type.</p>
1024 <p>Function attributes are simple keywords that follow the type specified. If
1025 multiple attributes are needed, they are space separated. For
1026 example:</p>
1029 <pre>
1030 define void @f() noinline { ... }
1031 define void @f() alwaysinline { ... }
1032 define void @f() alwaysinline optsize { ... }
1033 define void @f() optsize
1034 </pre>
1035 </div>
1037 <dl>
1039 <dd>This attribute indicates that the inliner should attempt to inline this
1040 function into callers whenever possible, ignoring any active inlining size
1041 threshold for this caller.</dd>
1044 <dd>This attribute indicates that the inliner should never inline this function
1045 in any situation. This attribute may not be used together with the
1049 <dd>This attribute suggests that optimization passes and code generator passes
1050 make choices that keep the code size of this function low, and otherwise do
1051 optimizations specifically to reduce code size.</dd>
1054 <dd>This function attribute indicates that the function never returns normally.
1055 This produces undefined behavior at runtime if the function ever does
1056 dynamically return.</dd>
1059 <dd>This function attribute indicates that the function never returns with an
1060 unwind or exceptional control flow. If the function does unwind, its runtime
1061 behavior is undefined.</dd>
1064 <dd>This attribute indicates that the function computes its result (or the
1065 exception it throws) based strictly on its arguments, without dereferencing any
1066 pointer arguments or otherwise accessing any mutable state (e.g. memory, control
1067 registers, etc) visible to caller functions. It does not write through any
1069 never changes any state visible to callers.</dd>
1072 <dd>This attribute indicates that the function does not write through any
1074 or otherwise modify any state (e.g. memory, control registers, etc) visible to
1075 caller functions. It may dereference pointer arguments and read state that may
1076 be set in the caller. A readonly function always returns the same value (or
1077 throws the same exception) when called with the same set of arguments and global
1078 state.</dd>
1081 <dd>This attribute indicates that the function should emit a stack smashing
1083 stack before the local variables that's checked upon return from the function to
1084 see if it has been overwritten. A heuristic is used to determine if a function
1085 needs stack protectors or not.
1094 function attribute.
1100 </dl>
1102 </div>
1104 <!-- ======================================================================= -->
1107 </div>
1110 <p>
1111 Modules may contain "module-level inline asm" blocks, which corresponds to the
1112 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
1113 LLVM and treated as a single unit, but may be separated in the .ll file if
1114 desired. The syntax is very simple:
1115 </p>
1118 <pre>
1119 module asm "inline asm code goes here"
1120 module asm "more can go here"
1121 </pre>
1122 </div>
1124 <p>The strings can contain any character by escaping non-printable characters.
1126 for the number.
1127 </p>
1129 <p>
1130 The inline asm code is simply printed to the machine code .s file when
1131 assembly code is generated.
1132 </p>
1133 </div>
1135 <!-- ======================================================================= -->
1138 </div>
1141 <p>A module may specify a target specific data layout string that specifies how
1142 data is to be laid out in memory. The syntax for the data layout is simply:</p>
1145 separated by the minus sign character ('-'). Each specification starts with a
1146 letter and may include other information after the letter to define some
1147 aspect of the data layout. The specifications accepted are as follows: </p>
1148 <dl>
1150 <dd>Specifies that the target lays out data in big-endian form. That is, the
1151 bits with the most significance have the lowest address location.</dd>
1153 <dd>Specifies that the target lays out data in little-endian form. That is,
1154 the bits with the least significance have the lowest address location.</dd>
1159 too.</dd>
1161 <dd>This specifies the alignment for an integer type of a given bit
1164 <dd>This specifies the alignment for a vector type of a given bit
1167 <dd>This specifies the alignment for a floating point type of a given bit
1169 (double).</dd>
1171 <dd>This specifies the alignment for an aggregate type of a given bit
1173 </dl>
1174 <p>When constructing the data layout for a given target, LLVM starts with a
1175 default set of specifications which are then (possibly) overriden by the
1177 are given in this list:</p>
1178 <ul>
1192 </ul>
1193 <p>When LLVM is determining the alignment for a given type, it uses the
1194 following rules:</p>
1195 <ol>
1196 <li>If the type sought is an exact match for one of the specifications, that
1197 specification is used.</li>
1198 <li>If no match is found, and the type sought is an integer type, then the
1199 smallest integer type that is larger than the bitwidth of the sought type is
1200 used. If none of the specifications are larger than the bitwidth then the the
1201 largest integer type is used. For example, given the default specifications
1202 above, the i7 type will use the alignment of i8 (next largest) while both
1203 i65 and i256 will use the alignment of i64 (largest specified).</li>
1204 <li>If no match is found, and the type sought is a vector type, then the
1205 largest vector type that is smaller than the sought vector type will be used
1208 </ol>
1209 </div>
1211 <!-- *********************************************************************** -->
1213 <!-- *********************************************************************** -->
1217 <p>The LLVM type system is one of the most important features of the
1218 intermediate representation. Being typed enables a number of
1219 optimizations to be performed on the intermediate representation directly,
1220 without having to do
1221 extra analyses on the side before the transformation. A strong type
1222 system makes it easier to read the generated code and enables novel
1223 analyses and transformations that are not feasible to perform on normal
1224 three address code representations.</p>
1226 </div>
1228 <!-- ======================================================================= -->
1230 Classifications</a> </div>
1232 <p>The types fall into a few useful
1233 classifications:</p>
1236 <tbody>
1238 <tr>
1241 </tr>
1242 <tr>
1245 </tr>
1246 <tr>
1255 </td>
1256 </tr>
1257 <tr>
1262 </tr>
1263 <tr>
1273 </td>
1274 </tr>
1275 </tbody>
1276 </table>
1279 most important. Values of these types are the only ones which can be
1280 produced by instructions, passed as arguments, or used as operands to
1281 instructions.</p>
1282 </div>
1284 <!-- ======================================================================= -->
1288 <p>The primitive types are the fundamental building blocks of the LLVM
1289 system.</p>
1291 </div>
1293 <!-- _______________________________________________________________________ -->
1297 <table>
1298 <tbody>
1305 </tbody>
1306 </table>
1307 </div>
1309 <!-- _______________________________________________________________________ -->
1318 <pre>
1319 void
1320 </pre>
1321 </div>
1323 <!-- _______________________________________________________________________ -->
1332 <pre>
1333 label
1334 </pre>
1335 </div>
1338 <!-- ======================================================================= -->
1343 <p>The real power in LLVM comes from the derived types in the system.
1344 This is what allows a programmer to represent arrays, functions,
1345 pointers, and other useful types. Note that these derived types may be
1346 recursive: For example, it is possible to have a two dimensional array.</p>
1348 </div>
1350 <!-- _______________________________________________________________________ -->
1356 <p>The integer type is a very simple derived type that simply specifies an
1357 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1362 <pre>
1363 iN
1364 </pre>
1367 value.</p>
1371 <tbody>
1372 <tr>
1375 </tr><tr>
1378 </tr><tr>
1381 </tr>
1382 </tbody>
1383 </table>
1385 <p>Note that the code generator does not yet support large integer types
1386 to be used as function return types. The specific limit on how large a
1387 return type the code generator can currently handle is target-dependent;
1389 targets.</p>
1391 </div>
1393 <!-- _______________________________________________________________________ -->
1400 <p>The array type is a very simple derived type that arranges elements
1401 sequentially in memory. The array type requires a size (number of
1402 elements) and an underlying data type.</p>
1406 <pre>
1408 </pre>
1410 <p>The number of elements is a constant integer value; elementtype may
1411 be any type with a size.</p>
1418 </tr>
1422 </tr>
1426 </tr>
1427 </table>
1433 </tr>
1437 </tr>
1441 </tr>
1442 </table>
1444 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1445 length array. Normally, accesses past the end of an array are undefined in
1447 As a special case, however, zero length arrays are recognized to be variable
1448 length. This allows implementation of 'pascal style arrays' with the LLVM
1451 <p>Note that the code generator does not yet support large aggregate types
1452 to be used as function return types. The specific limit on how large an
1453 aggregate return type the code generator can currently handle is
1454 target-dependent, and also dependent on the aggregate element types.</p>
1456 </div>
1458 <!-- _______________________________________________________________________ -->
1464 <p>The function type can be thought of as a function signature. It
1465 consists of a return type and a list of formal parameter types. The
1466 return type of a function type is a scalar type, a void type, or a struct type.
1467 If the return type is a struct type then all struct elements must be of first
1468 class types, and the struct must have at least one element.</p>
1472 <pre>
1474 </pre>
1478 which indicates that the function takes a variable number of arguments.
1479 Variable argument functions can access their arguments with the <a
1489 </td>
1492 </tt></td>
1497 </td>
1503 LLVM.
1504 </td>
1509 </td>
1510 </tr>
1511 </table>
1513 </div>
1514 <!-- _______________________________________________________________________ -->
1518 <p>The structure type is used to represent a collection of data members
1519 together in memory. The packing of the field types is defined to match
1520 the ABI of the underlying processor. The elements of a structure may
1521 be any type that has a size.</p>
1525 instruction.</p>
1539 </tr>
1540 </table>
1542 <p>Note that the code generator does not yet support large aggregate types
1543 to be used as function return types. The specific limit on how large an
1544 aggregate return type the code generator can currently handle is
1545 target-dependent, and also dependent on the aggregate element types.</p>
1547 </div>
1549 <!-- _______________________________________________________________________ -->
1551 </div>
1554 <p>The packed structure type is used to represent a collection of data members
1555 together in memory. There is no padding between fields. Further, the alignment
1556 of a packed structure is 1 byte. The elements of a packed structure may
1557 be any type that has a size.</p>
1561 instruction.</p>
1576 </tr>
1577 </table>
1578 </div>
1580 <!-- _______________________________________________________________________ -->
1584 <p>As in many languages, the pointer type represents a pointer or
1585 reference to another object, which must live in memory. Pointer types may have
1586 an optional address space attribute defining the target-specific numbered
1587 address space where the pointed-to object resides. The default address space is
1588 zero.</p>
1601 </tr>
1607 </tr>
1612 </tr>
1613 </table>
1614 </div>
1616 <!-- _______________________________________________________________________ -->
1622 <p>A vector type is a simple derived type that represents a vector
1623 of elements. Vector types are used when multiple primitive data
1624 are operated in parallel using a single instruction (SIMD).
1625 A vector type requires a size (number of
1626 elements) and an underlying primitive data type. Vectors must have a power
1632 <pre>
1634 </pre>
1636 <p>The number of elements is a constant integer value; elementtype may
1637 be any integer or floating point type.</p>
1645 </tr>
1649 </tr>
1653 </tr>
1654 </table>
1656 <p>Note that the code generator does not yet support large vector types
1657 to be used as function return types. The specific limit on how large a
1658 vector return type codegen can currently handle is target-dependent;
1659 currently it's often a few times longer than a hardware vector register.</p>
1661 </div>
1663 <!-- _______________________________________________________________________ -->
1669 <p>Opaque types are used to represent unknown types in the system. This
1670 corresponds (for example) to the C notion of a forward declared structure type.
1671 In LLVM, opaque types can eventually be resolved to any type (not just a
1672 structure type).</p>
1676 <pre>
1677 opaque
1678 </pre>
1686 </tr>
1687 </table>
1688 </div>
1690 <!-- ======================================================================= -->
1693 </div>
1697 <p>
1698 An "up reference" allows you to refer to a lexically enclosing type without
1699 requiring it to have a name. For instance, a structure declaration may contain a
1700 pointer to any of the types it is lexically a member of. Example of up
1701 references (with their equivalent as named type declarations) include:</p>
1703 <pre>
1704 { \2 * } %x = type { %x* }
1705 { \2 }* %y = type { %y }*
1706 \1* %z = type %z*
1707 </pre>
1709 <p>
1710 An up reference is needed by the asmprinter for printing out cyclic types when
1711 there is no declared name for a type in the cycle. Because the asmprinter does
1712 not want to print out an infinite type string, it needs a syntax to handle
1713 recursive types that have no names (all names are optional in llvm IR).
1714 </p>
1717 <pre>
1719 </pre>
1721 <p>
1722 The level is the count of the lexical type that is being referred to.
1723 </p>
1731 </tr>
1735 structure.</td>
1736 </tr>
1737 </table>
1738 </div>
1741 <!-- *********************************************************************** -->
1743 <!-- *********************************************************************** -->
1747 <p>LLVM has several different basic types of constants. This section describes
1748 them all and their syntax.</p>
1750 </div>
1752 <!-- ======================================================================= -->
1757 <dl>
1762 </dd>
1768 integer types.
1769 </dd>
1774 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1775 notation (see below). The assembler requires the exact decimal value of
1776 a floating-point constant. For example, the assembler accepts 1.25 but
1785 </dl>
1787 <p>The one non-intuitive notation for constants is the hexadecimal form
1788 of floating point constants. For example, the form '<tt>double
1791 (and the only time that they are generated by the disassembler) is when a
1792 floating point constant must be emitted but it cannot be represented as a
1793 decimal floating point number in a reasonable number of digits. For example,
1794 NaN's, infinities, and other
1795 special values are represented in their IEEE hexadecimal format so that
1796 assembly and disassembly do not cause any bits to change in the constants.</p>
1797 <p>When using the hexadecimal form, constants of types float and double are
1798 represented using the 16-digit form shown above (which matches the IEEE754
1799 representation for double); float values must, however, be exactly representable
1800 as IEE754 single precision.
1801 Hexadecimal format is always used for long
1802 double, and there are three forms of long double. The 80-bit
1804 followed by 20 hexadecimal digits.
1805 The 128-bit format used by PowerPC (two adjacent doubles) is represented
1807 format is represented
1809 target uses this format. Long doubles will only work if they match
1810 the long double format on your target. All hexadecimal formats are big-endian
1811 (sign bit at the left).</p>
1812 </div>
1814 <!-- ======================================================================= -->
1818 </div>
1821 <p>Complex constants are a (potentially recursive) combination of simple
1822 constants and smaller complex constants.</p>
1824 <dl>
1827 <dd>Structure constants are represented with notation similar to structure
1828 type definitions (a comma separated list of elements, surrounded by braces
1832 types of elements must match those specified by the type.
1833 </dd>
1837 <dd>Array constants are represented with notation similar to array type
1838 definitions (a comma separated list of elements, surrounded by square brackets
1841 types of elements must match those specified by the type.
1842 </dd>
1846 <dd>Vector constants are represented with notation similar to vector type
1847 definitions (a comma separated list of elements, surrounded by
1851 match those specified by the type.
1852 </dd>
1858 This is often used to avoid having to print large zero initializers (e.g. for
1859 large arrays) and is always exactly equivalent to using explicit zero
1860 initializers.
1861 </dd>
1865 <dd>A metadata node is a structure-like constant with the type of an empty
1867 constants that are meant to be interpreted as part of the instruction stream,
1868 metadata is a place to attach additional information such as debug info.
1869 </dd>
1870 </dl>
1872 </div>
1874 <!-- ======================================================================= -->
1877 </div>
1883 constants. These constants are explicitly referenced when the <a
1886 file:</p>
1889 <pre>
1890 @X = global i32 17
1891 @Y = global i32 42
1892 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1893 </pre>
1894 </div>
1896 </div>
1898 <!-- ======================================================================= -->
1902 no specific value. Undefined values may be of any type and be used anywhere
1903 a constant is permitted.</p>
1905 <p>Undefined values indicate to the compiler that the program is well defined
1906 no matter what value is used, giving the compiler more freedom to optimize.
1907 </p>
1908 </div>
1910 <!-- ======================================================================= -->
1912 </div>
1916 <p>Constant expressions are used to allow expressions involving other constants
1917 to be used as constants. Constant expressions may be of any <a
1919 that does not have side effects (e.g. load and call are not supported). The
1920 following is the syntax for constant expressions:</p>
1922 <dl>
1924 <dd>Truncate a constant to another type. The bit size of CST must be larger
1925 than the bit size of TYPE. Both types must be integers.</dd>
1928 <dd>Zero extend a constant to another type. The bit size of CST must be
1929 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1932 <dd>Sign extend a constant to another type. The bit size of CST must be
1933 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1936 <dd>Truncate a floating point constant to another floating point type. The
1937 size of CST must be larger than the size of TYPE. Both types must be
1938 floating point.</dd>
1941 <dd>Floating point extend a constant to another type. The size of CST must be
1942 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1945 <dd>Convert a floating point constant to the corresponding unsigned integer
1946 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1947 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1948 of the same number of elements. If the value won't fit in the integer type,
1949 the results are undefined.</dd>
1952 <dd>Convert a floating point constant to the corresponding signed integer
1953 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1954 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1955 of the same number of elements. If the value won't fit in the integer type,
1956 the results are undefined.</dd>
1959 <dd>Convert an unsigned integer constant to the corresponding floating point
1960 constant. TYPE must be a scalar or vector floating point type. CST must be of
1961 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1962 of the same number of elements. If the value won't fit in the floating point
1963 type, the results are undefined.</dd>
1966 <dd>Convert a signed integer constant to the corresponding floating point
1967 constant. TYPE must be a scalar or vector floating point type. CST must be of
1968 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1969 of the same number of elements. If the value won't fit in the floating point
1970 type, the results are undefined.</dd>
1973 <dd>Convert a pointer typed constant to the corresponding integer constant
1974 TYPE must be an integer type. CST must be of pointer type. The CST value is
1975 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1978 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1979 pointer type. CST must be of integer type. The CST value is zero extended,
1980 truncated, or unchanged to make it fit in a pointer size. This one is
1984 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
1992 instruction, the index list may have zero or more indexes, which are required
1998 constants.</dd>
2030 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2032 binary</a> operations. The constraints on operands are the same as those for
2033 the corresponding instruction (e.g. no bitwise operations on floating point
2034 values are allowed).</dd>
2035 </dl>
2036 </div>
2038 <!-- ======================================================================= -->
2040 </div>
2044 <p>Embedded metadata provides a way to attach arbitrary data to the
2045 instruction stream without affecting the behaviour of the program. There are
2046 two metadata primitives, strings and nodes. All metadata has the type of an
2047 empty struct and is identified in syntax by a preceding exclamation point
2049 </p>
2051 <p>A metadata string is a string surrounded by double quotes. It can contain
2054 </p>
2056 <p>Metadata nodes are represented with notation similar to structure constants
2057 (a comma separated list of elements, surrounded by braces and preceeded by an
2059 </p>
2061 <p>Optimizations may rely on metadata to provide additional information about
2062 the program that isn't available in the instructions, or that isn't easily
2063 computable. Similarly, the code generator may expect a certain metadata format
2064 to be used to express debugging information.</p>
2065 </div>
2067 <!-- *********************************************************************** -->
2069 <!-- *********************************************************************** -->
2071 <!-- ======================================================================= -->
2074 </div>
2078 <p>
2080 Module-Level Inline Assembly</a>) through the use of a special value. This
2081 value represents the inline assembler as a string (containing the instructions
2082 to emit), a list of operand constraints (stored as a string), and a flag that
2083 indicates whether or not the inline asm expression has side effects. An example
2084 inline assembler expression is:
2085 </p>
2088 <pre>
2090 </pre>
2091 </div>
2093 <p>
2096 </p>
2099 <pre>
2101 </pre>
2102 </div>
2104 <p>
2105 Inline asms with side effects not visible in the constraint list must be marked
2106 as having side effects. This is done through the use of the
2108 </p>
2111 <pre>
2113 </pre>
2114 </div>
2116 <p>TODO: The format of the asm and constraints string still need to be
2117 documented here. Constraints on what can be done (e.g. duplication, moving, etc
2118 need to be documented). This is probably best done by reference to another
2119 document that covers inline asm from a holistic perspective.
2120 </p>
2122 </div>
2124 <!-- *********************************************************************** -->
2126 <!-- *********************************************************************** -->
2130 <p>The LLVM instruction set consists of several different
2137 </div>
2139 <!-- ======================================================================= -->
2141 Instructions</a> </div>
2146 basic block in a program ends with a "Terminator" instruction, which
2147 indicates which block should be executed after the current block is
2149 value: they produce control flow, not values (the one exception being
2158 </div>
2160 <!-- _______________________________________________________________________ -->
2162 Instruction</a> </div>
2165 <pre>
2168 </pre>
2173 optionally a value) from a function back to the caller.</p>
2175 returns a value and then causes control flow, and one that just causes
2176 control flow to occur.</p>
2181 the return value. The type of the return value must be a
2186 instruction with no return value or a return value with a type that
2187 does not match its type, or if it has a void return type and contains
2193 returns back to the calling function's context. If the caller is a "<a
2195 the instruction after the call. If the caller was an "<a
2197 at the beginning of the "normal" destination block. If the instruction
2198 returns a value, that value shall set the call or invoke instruction's
2199 return value.</p>
2203 <pre>
2207 </pre>
2209 <p>Note that the code generator does not yet fully support large
2210 return values. The specific sizes that are currently supported are
2211 dependent on the target. For integers, on 32-bit targets the limit
2213 For aggregate types, the current limits are dependent on the element
2214 types; for example targets are often limited to 2 total integer
2217 </div>
2218 <!-- _______________________________________________________________________ -->
2222 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2223 </pre>
2226 transfer to a different basic block in the current function. There are
2227 two forms of this instruction, corresponding to a conditional branch
2228 and an unconditional branch.</p>
2240 <pre>Test:<br> %cond = <a href="#i_icmp">icmp</a> eq, i32 %a, %b<br> br i1 %cond, label %IfEqual, label %IfUnequal<br>IfEqual:<br> <a
2242 </div>
2243 <!-- _______________________________________________________________________ -->
2246 </div>
2251 <pre>
2252 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2253 </pre>
2259 instruction, allowing a branch to occur to one of many possible
2260 destinations.</p>
2268 table is not allowed to contain duplicate constant entries.</p>
2274 table is searched for the given value. If the value is found, control flow is
2275 transfered to the corresponding destination; otherwise, control flow is
2276 transfered to the default destination.</p>
2280 <p>Depending on properties of the target machine and the particular
2282 ways. For example, it could be generated as a series of chained conditional
2283 branches or with a lookup table.</p>
2287 <pre>
2290 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2293 switch i32 0, label %dest [ ]
2296 switch i32 %val, label %otherwise [ i32 0, label %onzero
2297 i32 1, label %onone
2298 i32 2, label %ontwo ]
2299 </pre>
2300 </div>
2302 <!-- _______________________________________________________________________ -->
2305 </div>
2311 <pre>
2312 <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>]
2314 </pre>
2319 function, with the possibility of control flow transfer to either the
2331 <ol>
2332 <li>
2334 convention</a> the call should use. If none is specified, the call defaults
2335 to using C calling conventions.
2336 </li>
2343 function value being invoked. In most cases, this is a direct function
2345 an arbitrary pointer to function value.
2346 </li>
2349 function to be invoked. </li>
2352 signature argument types. If the function signature indicates the function
2353 accepts a variable number of arguments, the extra arguments can be
2354 specified. </li>
2365 </ol>
2371 difference is that it establishes an association with a label, which is used by
2372 the runtime library to unwind the stack.</p>
2374 <p>This instruction is used in languages with destructors to ensure that proper
2376 exception. Additionally, this is important for implementation of
2380 <pre>
2381 %retval = invoke i32 @Test(i32 15) to label %Continue
2385 </pre>
2386 </div>
2389 <!-- _______________________________________________________________________ -->
2392 Instruction</a> </div>
2397 <pre>
2398 unwind
2399 </pre>
2404 at the first callee in the dynamic call stack which used an <a
2406 primarily used to implement exception handling.</p>
2411 immediately halt. The dynamic call stack is then searched for the first <a
2413 execution continues at the "exceptional" destination block specified by the
2415 dynamic call chain, undefined behavior results.</p>
2416 </div>
2418 <!-- _______________________________________________________________________ -->
2421 Instruction</a> </div>
2426 <pre>
2427 unreachable
2428 </pre>
2433 instruction is used to inform the optimizer that a particular portion of the
2434 code is not reachable. This can be used to indicate that the code after a
2435 no-return function cannot be reached, and other facts.</p>
2440 </div>
2444 <!-- ======================================================================= -->
2447 <p>Binary operators are used to do most of the computation in a
2448 program. They require two operands of the same type, execute an operation on them, and
2449 produce a single value. The operands might represent
2451 The result value has the same type as its operands.</p>
2453 </div>
2454 <!-- _______________________________________________________________________ -->
2457 </div>
2463 <pre>
2465 </pre>
2476 types.</p>
2480 <p>The value produced is the integer or floating point sum of the two
2481 operands.</p>
2483 <p>If an integer sum has unsigned overflow, the result returned is the
2485 the result.</p>
2487 <p>Because LLVM integers use a two's complement representation, this
2488 instruction is appropriate for both signed and unsigned integers.</p>
2492 <pre>
2494 </pre>
2495 </div>
2496 <!-- _______________________________________________________________________ -->
2499 </div>
2505 <pre>
2507 </pre>
2512 operands.</p>
2516 representations.</p>
2523 types.</p>
2527 <p>The value produced is the integer or floating point difference of
2528 the two operands.</p>
2530 <p>If an integer difference has unsigned overflow, the result returned is the
2532 the result.</p>
2534 <p>Because LLVM integers use a two's complement representation, this
2535 instruction is appropriate for both signed and unsigned integers.</p>
2538 <pre>
2541 </pre>
2542 </div>
2544 <!-- _______________________________________________________________________ -->
2547 </div>
2553 </pre>
2556 operands.</p>
2563 types.</p>
2567 <p>The value produced is the integer or floating point product of the
2568 two operands.</p>
2570 <p>If the result of an integer multiplication has unsigned overflow,
2571 the result returned is the mathematical result modulo
2573 <p>Because LLVM integers use a two's complement representation, and the
2574 result is the same width as the operands, this instruction returns the
2575 correct result for both signed and unsigned integers. If a full product
2577 should be sign-extended or zero-extended as appropriate to the
2578 width of the full product.</p>
2581 </pre>
2582 </div>
2584 <!-- _______________________________________________________________________ -->
2586 </a></div>
2590 </pre>
2593 operands.</p>
2599 values. Both arguments must have identical types.</p>
2604 <p>Note that unsigned integer division and signed integer division are distinct
2609 </pre>
2610 </div>
2611 <!-- _______________________________________________________________________ -->
2613 </a> </div>
2616 <pre>
2618 </pre>
2623 operands.</p>
2629 values. Both arguments must have identical types.</p>
2632 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2633 <p>Note that signed integer division and unsigned integer division are distinct
2635 <p>Division by zero leads to undefined behavior. Overflow also leads to
2636 undefined behavior; this is a rare case, but can occur, for example,
2640 </pre>
2641 </div>
2642 <!-- _______________________________________________________________________ -->
2644 Instruction</a> </div>
2647 <pre>
2649 </pre>
2653 operands.</p>
2659 of floating point values. Both arguments must have identical types.</p>
2667 <pre>
2669 </pre>
2670 </div>
2672 <!-- _______________________________________________________________________ -->
2674 </div>
2678 </pre>
2681 unsigned division of its two arguments.</p>
2685 values. Both arguments must have identical types.</p>
2688 This instruction always performs an unsigned division to get the remainder.</p>
2689 <p>Note that unsigned integer remainder and signed integer remainder are
2694 </pre>
2696 </div>
2697 <!-- _______________________________________________________________________ -->
2700 </div>
2706 <pre>
2708 </pre>
2713 signed division of its two operands. This instruction can also take
2715 the elements must be integers.</p>
2721 values. Both arguments must have identical types.</p>
2728 a value. For more information about the difference, see <a
2730 Math Forum</a>. For a table of how this is implemented in various languages,
2733 <p>Note that signed integer remainder and unsigned integer remainder are
2735 <p>Taking the remainder of a division by zero leads to undefined behavior.
2736 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2738 (The remainder doesn't actually overflow, but this rule lets srem be
2739 implemented using instructions that return both the result of the division
2740 and the remainder.)</p>
2743 </pre>
2745 </div>
2746 <!-- _______________________________________________________________________ -->
2754 </pre>
2757 division of its two operands.</p>
2761 of floating point values. Both arguments must have identical types.</p>
2766 The remainder has the same sign as the dividend.</p>
2770 <pre>
2772 </pre>
2773 </div>
2775 <!-- ======================================================================= -->
2777 Operations</a> </div>
2779 <p>Bitwise binary operators are used to do various forms of
2780 bit-twiddling in a program. They are generally very efficient
2781 instructions and can commonly be strength reduced from other
2782 instructions. They require two operands of the same type, execute an operation on them,
2783 and produce a single value. The resulting value is the same type as its operands.</p>
2784 </div>
2786 <!-- _______________________________________________________________________ -->
2788 Instruction</a> </div>
2792 </pre>
2797 the left a specified number of bits.</p>
2818 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
2819 </pre>
2820 </div>
2821 <!-- _______________________________________________________________________ -->
2823 Instruction</a> </div>
2827 </pre>
2831 operand shifted to the right a specified number of bits with zero fill.</p>
2840 <p>This instruction always performs a logical shift right operation. The most
2841 significant bits of the result will be filled with zero bits after the
2848 <pre>
2854 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
2855 </pre>
2856 </div>
2858 <!-- _______________________________________________________________________ -->
2860 Instruction</a> </div>
2865 </pre>
2869 operand shifted to the right a specified number of bits with sign extension.</p>
2877 <p>This instruction always performs an arithmetic shift right operation,
2878 The most significant bits of the result will be filled with the sign bit
2885 <pre>
2891 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
2892 </pre>
2893 </div>
2895 <!-- _______________________________________________________________________ -->
2897 Instruction</a> </div>
2903 <pre>
2905 </pre>
2910 its two operands.</p>
2916 values. Both arguments must have identical types.</p>
2920 <p> </p>
2921 <div>
2923 <tbody>
2924 <tr>
2928 </tr>
2929 <tr>
2933 </tr>
2934 <tr>
2938 </tr>
2939 <tr>
2943 </tr>
2944 <tr>
2948 </tr>
2949 </tbody>
2950 </table>
2951 </div>
2953 <pre>
2957 </pre>
2958 </div>
2959 <!-- _______________________________________________________________________ -->
2964 </pre>
2967 or of its two operands.</p>
2972 values. Both arguments must have identical types.</p>
2975 <p> </p>
2976 <div>
2978 <tbody>
2979 <tr>
2983 </tr>
2984 <tr>
2988 </tr>
2989 <tr>
2993 </tr>
2994 <tr>
2998 </tr>
2999 <tr>
3003 </tr>
3004 </tbody>
3005 </table>
3006 </div>
3011 </pre>
3012 </div>
3013 <!-- _______________________________________________________________________ -->
3015 Instruction</a> </div>
3019 </pre>
3027 values. Both arguments must have identical types.</p>
3032 <p> </p>
3033 <div>
3035 <tbody>
3036 <tr>
3040 </tr>
3041 <tr>
3045 </tr>
3046 <tr>
3050 </tr>
3051 <tr>
3055 </tr>
3056 <tr>
3060 </tr>
3061 </tbody>
3062 </table>
3063 </div>
3064 <p> </p>
3070 </pre>
3071 </div>
3073 <!-- ======================================================================= -->
3076 </div>
3080 <p>LLVM supports several instructions to represent vector operations in a
3081 target-independent manner. These instructions cover the element-access and
3082 vector-specific operations needed to process vectors effectively. While LLVM
3083 does directly support these vector operations, many sophisticated algorithms
3084 will want to use target-specific intrinsics to take full advantage of a specific
3085 target.</p>
3087 </div>
3089 <!-- _______________________________________________________________________ -->
3092 </div>
3098 <pre>
3099 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3100 </pre>
3104 <p>
3106 element from a vector at a specified index.
3107 </p>
3112 <p>
3115 an index indicating the position from which to extract the element.
3116 The index may be a variable.</p>
3120 <p>
3121 The result is a scalar of the same type as the element type of
3124 results are undefined.
3125 </p>
3129 <pre>
3131 </pre>
3132 </div>
3135 <!-- _______________________________________________________________________ -->
3138 </div>
3144 <pre>
3145 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3146 </pre>
3150 <p>
3152 element into a vector at a specified index.
3153 </p>
3158 <p>
3161 scalar value whose type must equal the element type of the first
3162 operand. The third operand is an index indicating the position at
3163 which to insert the value. The index may be a variable.</p>
3167 <p>
3172 </p>
3176 <pre>
3178 </pre>
3179 </div>
3181 <!-- _______________________________________________________________________ -->
3184 </div>
3190 <pre>
3191 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3192 </pre>
3196 <p>
3198 from two input vectors, returning a vector with the same element type as
3199 the input and length that is the same as the shuffle mask.
3200 </p>
3204 <p>
3206 with types that match each other. The third argument is a shuffle mask whose
3207 element type is always 'i32'. The result of the instruction is a vector whose
3208 length is the same as the shuffle mask and whose element type is the same as
3209 the element type of the first two operands.
3210 </p>
3212 <p>
3213 The shuffle mask operand is required to be a constant vector with either
3214 constant integer or undef values.
3215 </p>
3219 <p>
3220 The elements of the two input vectors are numbered from left to right across
3221 both of the vectors. The shuffle mask operand specifies, for each element of
3222 the result vector, which element of the two input vectors the result element
3223 gets. The element selector may be undef (meaning "don't care") and the second
3224 operand may be undef if performing a shuffle from only one vector.
3225 </p>
3229 <pre>
3233 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3237 <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>
3238 </pre>
3239 </div>
3242 <!-- ======================================================================= -->
3245 </div>
3249 <p>LLVM supports several instructions for working with aggregate values.
3250 </p>
3252 </div>
3254 <!-- _______________________________________________________________________ -->
3257 </div>
3263 <pre>
3265 </pre>
3269 <p>
3271 or array element from an aggregate value.
3272 </p>
3277 <p>
3280 type. The operands are constant indices to specify which value to extract
3281 in a similar manner as indices in a
3283 </p>
3287 <p>
3288 The result is the value at the position in the aggregate specified by
3289 the index operands.
3290 </p>
3294 <pre>
3296 </pre>
3297 </div>
3300 <!-- _______________________________________________________________________ -->
3303 </div>
3309 <pre>
3310 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3311 </pre>
3315 <p>
3317 into a struct field or array element in an aggregate.
3318 </p>
3323 <p>
3326 The second operand is a first-class value to insert.
3327 The following operands are constant indices
3328 indicating the position at which to insert the value in a similar manner as
3329 indices in a
3331 The value to insert must have the same type as the value identified
3332 by the indices.
3333 </p>
3337 <p>
3341 </p>
3345 <pre>
3347 </pre>
3348 </div>
3351 <!-- ======================================================================= -->
3354 </div>
3358 <p>A key design point of an SSA-based representation is how it
3359 represents memory. In LLVM, no memory locations are in SSA form, which
3360 makes things very simple. This section describes how to read, write,
3361 allocate, and free memory in LLVM.</p>
3363 </div>
3365 <!-- _______________________________________________________________________ -->
3368 </div>
3374 <pre>
3375 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3376 </pre>
3381 heap and returns a pointer to it. The object is always allocated in the generic
3382 address space (address space zero).</p>
3388 bytes of memory from the operating system and returns a pointer of the
3389 appropriate type to the program. If "NumElements" is specified, it is the
3390 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3391 If a constant alignment is specified, the value result of the allocation is guaranteed to
3392 be aligned to at least that boundary. If not specified, or if zero, the target can
3393 choose to align the allocation on any convenient boundary.</p>
3400 a pointer is returned. The result of a zero byte allocation is undefined. The
3401 result is null if there is insufficient memory available.</p>
3405 <pre>
3413 </pre>
3415 <p>Note that the code generator does not yet respect the
3416 alignment value.</p>
3418 </div>
3420 <!-- _______________________________________________________________________ -->
3423 </div>
3429 <pre>
3431 </pre>
3436 memory heap to be reallocated in the future.</p>
3442 instruction.</p>
3446 <p>Access to the memory pointed to by the pointer is no longer defined
3447 after this instruction executes. If the pointer is null, the operation
3448 is a noop.</p>
3452 <pre>
3454 free [4 x i8]* %array
3455 </pre>
3456 </div>
3458 <!-- _______________________________________________________________________ -->
3461 </div>
3467 <pre>
3468 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3469 </pre>
3474 currently executing function, to be automatically released when this function
3475 returns to its caller. The object is always allocated in the generic address
3476 space (address space zero).</p>
3481 bytes of memory on the runtime stack, returning a pointer of the
3482 appropriate type to the program. If "NumElements" is specified, it is the
3483 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3484 If a constant alignment is specified, the value result of the allocation is guaranteed
3485 to be aligned to at least that boundary. If not specified, or if zero, the target
3486 can choose to align the allocation on any convenient boundary.</p>
3492 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3495 instruction is commonly used to represent automatic variables that must
3496 have an address available. When the function returns (either with the <tt><a
3498 instructions), the memory is reclaimed. Allocating zero bytes
3499 is legal, but the result is undefined.</p>
3503 <pre>
3508 </pre>
3509 </div>
3511 <!-- _______________________________________________________________________ -->
3513 Instruction</a> </div>
3516 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3521 address from which to load. The pointer must point to a <a
3526 instructions. </p>
3527 <p>
3528 The optional constant "align" argument specifies the alignment of the operation
3529 (that is, the alignment of the memory address). A value of 0 or an
3530 omitted "align" argument means that the operation has the preferential
3531 alignment for the target. It is the responsibility of the code emitter
3532 to ensure that the alignment information is correct. Overestimating
3533 the alignment results in an undefined behavior. Underestimating the
3534 alignment may produce less efficient code. An alignment of 1 is always
3535 safe.
3536 </p>
3538 <p>The location of memory pointed to is loaded. If the value being loaded
3539 is of scalar type then the number of bytes read does not exceed the minimum
3540 number of bytes needed to hold all bits of the type. For example, loading an
3543 is undefined if the value was not originally written using a store of the
3544 same type.</p>
3547 <a
3550 </pre>
3551 </div>
3552 <!-- _______________________________________________________________________ -->
3554 Instruction</a> </div>
3557 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3558 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3559 </pre>
3568 optimizer is not allowed to modify the number or order of execution of
3571 <p>
3572 The optional constant "align" argument specifies the alignment of the operation
3573 (that is, the alignment of the memory address). A value of 0 or an
3574 omitted "align" argument means that the operation has the preferential
3575 alignment for the target. It is the responsibility of the code emitter
3576 to ensure that the alignment information is correct. Overestimating
3577 the alignment results in an undefined behavior. Underestimating the
3578 alignment may produce less efficient code. An alignment of 1 is always
3579 safe.
3580 </p>
3585 written does not exceed the minimum number of bytes needed to hold all
3588 size that is not an integral number of bytes, it is unspecified what
3589 happens to the extra bits that do not belong to the type, but they will
3590 typically be overwritten.</p>
3595 </pre>
3596 </div>
3598 <!-- _______________________________________________________________________ -->
3601 </div>
3605 <pre>
3607 </pre>
3611 <p>
3613 subelement of an aggregate data structure. It performs address calculation only
3614 and does not access memory.</p>
3618 <p>The first argument is always a pointer, and forms the basis of the
3619 calculation. The remaining arguments are indices, that indicate which of the
3620 elements of the aggregate object are indexed. The interpretation of each index
3621 is dependent on the type being indexed into. The first index always indexes the
3622 pointer value given as the first argument, the second index indexes a value of
3623 the type pointed to (not necessarily the value directly pointed to, since the
3624 first index can be non-zero), etc. The first type indexed into must be a pointer
3625 value, subsequent types can be arrays, vectors and structs. Note that subsequent
3626 types being indexed into can never be pointers, since that would require loading
3627 the pointer before continuing calculation.</p>
3629 <p>The type of each index argument depends on the type it is indexing into.
3632 integers of any width are allowed (also non-constants).</p>
3634 <p>For example, let's consider a C code fragment and how it gets
3635 compiled to LLVM:</p>
3638 <pre>
3639 struct RT {
3640 char A;
3642 char C;
3643 };
3644 struct ST {
3645 int X;
3646 double Y;
3647 struct RT Z;
3648 };
3650 int *foo(struct ST *s) {
3652 }
3653 </pre>
3654 </div>
3659 <pre>
3663 define i32* %foo(%ST* %s) {
3664 entry:
3666 ret i32* %reg
3667 }
3668 </pre>
3669 </div>
3675 }</tt>' type, a structure. The second index indexes into the third element of
3677 i8 }</tt>' type, another structure. The third index indexes into the second
3679 array. The two dimensions of the array are subscripted into, yielding an
3683 <p>Note that it is perfectly legal to index partially through a
3684 structure, returning a pointer to an inner element. Because of this,
3685 the LLVM code for the given testcase is equivalent to:</p>
3687 <pre>
3688 define i32* %foo(%ST* %s) {
3694 ret i32* %t5
3695 }
3696 </pre>
3698 <p>Note that it is undefined to access an array out of bounds: array
3699 and pointer indexes must always be within the defined bounds of the
3700 array type when accessed with an instruction that dereferences the
3701 pointer (e.g. a load or store instruction). The one exception for
3702 this rule is zero length arrays. These arrays are defined to be
3703 accessible as variable length arrays, which requires access beyond the
3704 zero'th element.</p>
3706 <p>The getelementptr instruction is often confusing. For some more insight
3712 <pre>
3721 </pre>
3722 </div>
3724 <!-- ======================================================================= -->
3726 </div>
3728 <p>The instructions in this category are the conversion instructions (casting)
3729 which all take a single operand and a type. They perform various bit conversions
3730 on the operand.</p>
3731 </div>
3733 <!-- _______________________________________________________________________ -->
3736 </div>
3740 <pre>
3742 </pre>
3745 <p>
3747 </p>
3750 <p>
3758 <p>
3762 It will always truncate bits.</p>
3765 <pre>
3769 </pre>
3770 </div>
3772 <!-- _______________________________________________________________________ -->
3775 </div>
3779 <pre>
3781 </pre>
3802 <pre>
3805 </pre>
3806 </div>
3808 <!-- _______________________________________________________________________ -->
3811 </div>
3815 <pre>
3817 </pre>
3823 <p>
3831 <p>
3839 <pre>
3842 </pre>
3843 </div>
3845 <!-- _______________________________________________________________________ -->
3848 </div>
3854 <pre>
3856 </pre>
3877 <pre>
3880 </pre>
3881 </div>
3883 <!-- _______________________________________________________________________ -->
3886 </div>
3890 <pre>
3892 </pre>
3896 floating point value.</p>
3902 type must be smaller than the destination type.</p>
3912 <pre>
3915 </pre>
3916 </div>
3918 <!-- _______________________________________________________________________ -->
3921 </div>
3925 <pre>
3927 </pre>
3932 </p>
3945 the results are undefined.</p>
3948 <pre>
3952 </pre>
3953 </div>
3955 <!-- _______________________________________________________________________ -->
3958 </div>
3962 <pre>
3964 </pre>
3969 </p>
3982 the results are undefined.</p>
3985 <pre>
3989 </pre>
3990 </div>
3992 <!-- _______________________________________________________________________ -->
3995 </div>
3999 <pre>
4001 </pre>
4016 integer quantity and converts it to the corresponding floating point value. If
4017 the value cannot fit in the floating point value, the results are undefined.</p>
4020 <pre>
4023 </pre>
4024 </div>
4026 <!-- _______________________________________________________________________ -->
4029 </div>
4033 <pre>
4035 </pre>
4050 integer quantity and converts it to the corresponding floating point value. If
4051 the value cannot fit in the floating point value, the results are undefined.</p>
4054 <pre>
4057 </pre>
4058 </div>
4060 <!-- _______________________________________________________________________ -->
4063 </div>
4067 <pre>
4069 </pre>
4083 truncating or zero extending that value to the size of the integer type. If
4087 change.</p>
4090 <pre>
4093 </pre>
4094 </div>
4096 <!-- _______________________________________________________________________ -->
4099 </div>
4103 <pre>
4105 </pre>
4113 value to cast, and a type to cast it to, which must be a
4121 the size of a pointer then a zero extension is done. If they are the same size,
4125 <pre>
4129 </pre>
4130 </div>
4132 <!-- _______________________________________________________________________ -->
4135 </div>
4139 <pre>
4141 </pre>
4151 a non-aggregate first class value, and a type to cast it to, which must also be
4155 type is a pointer, the destination type must also be a pointer. This
4156 instruction supports bitwise conversion of vectors to integers and to vectors
4157 of other types (as long as they have the same size).</p>
4164 converted to other pointer types with this instruction. To convert pointers to
4169 <pre>
4173 </pre>
4174 </div>
4176 <!-- ======================================================================= -->
4180 instructions, which defy better classification.</p>
4181 </div>
4183 <!-- _______________________________________________________________________ -->
4185 </div>
4188 <pre> <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4189 </pre>
4192 a vector of boolean values based on comparison
4193 of its two integer, integer vector, or pointer operands.</p>
4196 the condition code indicating the kind of comparison to perform. It is not
4197 a value, just a keyword. The possible condition code are:
4198 </p>
4199 <ol>
4210 </ol>
4214 They must also be identical types.</p>
4218 yields either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt> result, as follows:
4219 </p>
4220 <ol>
4223 </li>
4242 </ol>
4244 values are compared as if they were integers.</p>
4245 <p>If the operands are integer vectors, then they are compared
4247 the same number of elements as the values being compared.
4249 </p>
4258 </pre>
4260 <p>Note that the code generator does not yet support vector types with
4263 </div>
4265 <!-- _______________________________________________________________________ -->
4267 </div>
4270 <pre> <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4271 </pre>
4274 or vector of boolean values based on comparison
4275 of its operands.</p>
4276 <p>
4277 If the operands are floating point scalars, then the result
4279 </p>
4280 <p>If the operands are floating point vectors, then the result type
4281 is a vector of boolean with the same number of elements as the
4282 operands being compared.</p>
4285 the condition code indicating the kind of comparison to perform. It is not
4286 a value, just a keyword. The possible condition code are:</p>
4287 <ol>
4304 </ol>
4310 They must have identical types.</p>
4314 If the operands are vectors, then the vectors are compared
4315 element by element.
4316 Each comparison performed
4318 <ol>
4347 </ol>
4354 </pre>
4356 <p>Note that the code generator does not yet support vector types with
4359 </div>
4361 <!-- _______________________________________________________________________ -->
4364 </div>
4367 <pre> <result> = vicmp <cond> <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4368 </pre>
4371 element-wise comparison of its two integer vector operands.</p>
4374 the condition code indicating the kind of comparison to perform. It is not
4375 a value, just a keyword. The possible condition code are:</p>
4376 <ol>
4387 </ol>
4394 identical type as the values being compared. The most significant bit in each
4396 otherwise. All other bits of the result are undefined. The condition codes
4401 <pre>
4402 <result> = vicmp eq <2 x i32> < i32 4, i32 0>, < i32 5, i32 0> <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
4403 <result> = vicmp ult <2 x i8 > < i8 1, i8 2>, < i8 2, i8 2 > <i>; yields: result=<2 x i8> < i8 -1, i8 0 ></i>
4404 </pre>
4405 </div>
4407 <!-- _______________________________________________________________________ -->
4410 </div>
4416 element-wise comparison of its two floating point vector operands. The output
4417 elements have the same width as the input elements.</p>
4420 the condition code indicating the kind of comparison to perform. It is not
4421 a value, just a keyword. The possible condition code are:</p>
4422 <ol>
4439 </ol>
4442 types.</p>
4447 an identical number of elements as the values being compared, and each element
4448 having identical with to the width of the floating point elements. The most
4449 significant bit in each element is 1 if the element-wise comparison evaluates to
4450 true, and is 0 otherwise. All other bits of the result are undefined. The
4451 condition codes are evaluated identically to the
4455 <pre>
4457 <result> = vfcmp oeq <2 x float> < float 4, float 0 >, < float 5, float 0 >
4460 <result> = vfcmp ult <2 x double> < double 1, double 2 >, < double 2, double 2>
4461 </pre>
4462 </div>
4464 <!-- _______________________________________________________________________ -->
4467 </div>
4476 the SSA graph representing the function.</p>
4479 <p>The type of the incoming values is specified with the first type
4481 as arguments, with one pair for each predecessor basic block of the
4483 type may be used as the value arguments to the PHI node. Only labels
4484 may be used as the label arguments.</p>
4486 <p>There must be no non-phi instructions between the start of a basic
4487 block and the PHI instructions: i.e. PHI instructions must be first in
4488 a basic block.</p>
4493 specified by the pair corresponding to the predecessor basic block that executed
4494 just prior to the current block.</p>
4497 <pre>
4498 Loop: ; Infinite loop that counts from 0 on up...
4499 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4500 %nextindvar = add i32 %indvar, 1
4501 br label %Loop
4502 </pre>
4503 </div>
4505 <!-- _______________________________________________________________________ -->
4508 </div>
4514 <pre>
4515 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4518 </pre>
4522 <p>
4524 condition, without branching.
4525 </p>
4530 <p>
4532 a vector of 'i1' values indicating the
4534 type. If the val1/val2 are vectors and
4535 the condition is a scalar, then entire vectors are selected, not
4536 individual elements.
4537 </p>
4541 <p>
4542 If the condition is an i1 and it evaluates to 1, the instruction returns the first
4543 value argument; otherwise, it returns the second value argument.
4544 </p>
4545 <p>
4546 If the condition is a vector of i1, then the value arguments must
4547 be vectors of the same size, and the selection is done element
4548 by element.
4549 </p>
4553 <pre>
4555 </pre>
4557 <p>Note that the code generator does not yet support conditions
4558 with vector type.</p>
4560 </div>
4563 <!-- _______________________________________________________________________ -->
4566 </div>
4571 <pre>
4572 <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>]
4573 </pre>
4583 <ol>
4584 <li>
4586 any allocas or varargs in the caller. If the "tail" marker is present, the
4587 function call is eligible for tail call optimization. Note that calls may
4590 </li>
4591 <li>
4593 convention</a> the call should use. If none is specified, the call defaults
4594 to using C calling conventions.</p>
4595 </li>
4597 <li>
4601 </li>
4603 <li>
4605 the type of the return value. Functions that return no value are marked
4607 </li>
4608 <li>
4610 value being invoked. The argument types must match the types implied by
4611 this signature. This type can be omitted if the function is not varargs
4612 and if the function type does not return a pointer to a function.</p>
4613 </li>
4614 <li>
4616 be invoked. In most cases, this is a direct function invocation, but
4618 to function value.</p>
4619 </li>
4620 <li>
4622 function signature argument types. All arguments must be of
4624 indicates the function accepts a variable number of arguments, the extra
4625 arguments can be specified.</p>
4626 </li>
4627 <li>
4631 </li>
4632 </ol>
4637 transfer to a specified function, with its incoming arguments bound to
4639 instruction in the called function, control flow continues with the
4640 instruction after the function call, and the return value of the
4641 function is bound to the result argument.</p>
4645 <pre>
4646 %retval = call i32 @test(i32 %argc)
4650 call void %foo(i8 97 signext)
4652 %struct.A = type { i32, i8 }
4658 </pre>
4660 </div>
4662 <!-- _______________________________________________________________________ -->
4665 </div>
4671 <pre>
4673 </pre>
4678 the "variable argument" area of a function call. It is used to implement the
4684 the argument. It returns a value of the specified argument type and
4696 <p>It is legal for this instruction to be called in a function which does not
4698 function.</p>
4702 argument.</p>
4708 <p>Note that the code generator does not yet fully support va_arg
4709 on many targets. Also, it does not currently support va_arg with
4710 aggregate types on any target.</p>
4712 </div>
4714 <!-- *********************************************************************** -->
4716 <!-- *********************************************************************** -->
4721 well known names and semantics and are required to follow certain restrictions.
4722 Overall, these intrinsics represent an extension mechanism for the LLVM
4723 language that does not require changing all of the transformations in LLVM when
4724 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4727 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4728 begin with this prefix. Intrinsic functions must always be external functions:
4729 you cannot define the body of intrinsic functions. Intrinsic functions may
4730 only be used in call or invoke instructions: it is illegal to take the address
4731 of an intrinsic function. Additionally, because intrinsic functions are part
4732 of the LLVM language, it is required if any are added that they be documented
4733 here.</p>
4735 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4736 a family of functions that perform the same operation but on different data
4737 types. Because LLVM can represent over 8 million different integer types,
4738 overloading is used commonly to allow an intrinsic function to operate on any
4739 integer type. One or more of the argument types or the result type can be
4740 overloaded to accept any integer type. Argument types may also be defined as
4741 exactly matching a previous argument's type or the result type. This allows an
4742 intrinsic function which accepts multiple arguments, but needs all of them to
4743 be of the same type, to only be overloaded with respect to a single argument or
4744 the result.</p>
4746 <p>Overloaded intrinsics will have the names of its overloaded argument types
4747 encoded into its function name, each preceded by a period. Only those types
4748 which are overloaded result in a name suffix. Arguments whose type is matched
4750 take an integer of any width and returns an integer of exactly the same integer
4751 width. This leads to a family of functions such as
4753 Only one type, the return type, is overloaded, and only one type suffix is
4754 required. Because the argument's type is matched against the return type, it
4755 does not require its own name suffix.</p>
4757 <p>To learn how to add an intrinsic function, please see the
4759 </p>
4761 </div>
4763 <!-- ======================================================================= -->
4766 </div>
4772 intrinsic functions. These functions are related to the similarly
4775 <p>All of these functions operate on arguments that use a
4776 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4777 language reference manual does not define what this type is, so all
4778 transformations should be prepared to handle these functions regardless of
4779 the type used.</p>
4782 instruction and the variable argument handling intrinsic functions are
4783 used.</p>
4786 <pre>
4787 define i32 @test(i32 %X, ...) {
4788 ; Initialize variable argument processing
4789 %ap = alloca i8*
4790 %ap2 = bitcast i8** %ap to i8*
4791 call void @llvm.va_start(i8* %ap2)
4793 ; Read a single integer argument
4794 %tmp = va_arg i8** %ap, i32
4796 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4797 %aq = alloca i8*
4798 %aq2 = bitcast i8** %aq to i8*
4799 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4800 call void @llvm.va_end(i8* %aq2)
4802 ; Stop processing of arguments.
4803 call void @llvm.va_end(i8* %ap2)
4804 ret i32 %tmp
4805 }
4807 declare void @llvm.va_start(i8*)
4808 declare void @llvm.va_copy(i8*, i8*)
4809 declare void @llvm.va_end(i8*)
4810 </pre>
4811 </div>
4813 </div>
4815 <!-- _______________________________________________________________________ -->
4818 </div>
4836 macro available in C. In a target-dependent way, it initializes the
4840 last argument of the function as the compiler can figure that out.</p>
4842 </div>
4844 <!-- _______________________________________________________________________ -->
4847 </div>
4865 macro available in C. In a target-dependent way, it destroys the
4871 </div>
4873 <!-- _______________________________________________________________________ -->
4876 </div>
4882 <pre>
4884 </pre>
4889 from the source argument list to the destination argument list.</p>
4900 macro available in C. In a target-dependent way, it copies the source
4903 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4904 example, memory allocation.</p>
4906 </div>
4908 <!-- ======================================================================= -->
4911 </div>
4915 <p>
4917 Collection</a> (GC) requires the implementation and generation of these
4918 intrinsics.
4922 Front-ends for type-safe garbage collected languages should generate these
4923 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4925 </p>
4927 <p>The garbage collection intrinsics only operate on objects in the generic
4928 address space (address space zero).</p>
4930 </div>
4932 <!-- _______________________________________________________________________ -->
4935 </div>
4941 <pre>
4942 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4943 </pre>
4948 the code generator, and allows some metadata to be associated with it.</p>
4952 <p>The first argument specifies the address of a stack object that contains the
4953 root pointer. The second pointer (which must be either a constant or a global
4954 value address) contains the meta-data to be associated with the root.</p>
4959 location. At compile-time, the code generator generates information to allow
4964 </div>
4967 <!-- _______________________________________________________________________ -->
4970 </div>
4976 <pre>
4977 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4978 </pre>
4983 locations, allowing garbage collector implementations that require read
4984 barriers.</p>
4988 <p>The second argument is the address to read from, which should be an address
4989 allocated from the garbage collector. The first object is a pointer to the
4990 start of the referenced object, if needed by the language runtime (otherwise
4991 null).</p>
4996 instruction, but may be replaced with substantially more complex code by the
5001 </div>
5004 <!-- _______________________________________________________________________ -->
5007 </div>
5013 <pre>
5014 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5015 </pre>
5020 locations, allowing garbage collector implementations that require write
5021 barriers (such as generational or reference counting collectors).</p>
5025 <p>The first argument is the reference to store, the second is the start of the
5026 object to store it to, and the third is the address of the field of Obj to
5027 store to. If the runtime does not require a pointer to the object, Obj may be
5028 null.</p>
5033 instruction, but may be replaced with substantially more complex code by the
5038 </div>
5042 <!-- ======================================================================= -->
5045 </div>
5048 <p>
5049 These intrinsics are provided by LLVM to expose special features that may only
5050 be implemented with code generator support.
5051 </p>
5053 </div>
5055 <!-- _______________________________________________________________________ -->
5058 </div>
5063 <pre>
5065 </pre>
5069 <p>
5071 target-specific value indicating the return address of the current function
5072 or one of its callers.
5073 </p>
5077 <p>
5078 The argument to this intrinsic indicates which function to return the address
5079 for. Zero indicates the calling function, one indicates its caller, etc. The
5081 </p>
5085 <p>
5087 the return address of the specified call frame, or zero if it cannot be
5088 identified. The value returned by this intrinsic is likely to be incorrect or 0
5089 for arguments other than zero, so it should only be used for debugging purposes.
5090 </p>
5092 <p>
5093 Note that calling this intrinsic does not prevent function inlining or other
5094 aggressive transformations, so the value returned may not be that of the obvious
5095 source-language caller.
5096 </p>
5097 </div>
5100 <!-- _______________________________________________________________________ -->
5103 </div>
5108 <pre>
5110 </pre>
5114 <p>
5116 target-specific frame pointer value for the specified stack frame.
5117 </p>
5121 <p>
5122 The argument to this intrinsic indicates which function to return the frame
5123 pointer for. Zero indicates the calling function, one indicates its caller,
5125 </p>
5129 <p>
5131 the frame address of the specified call frame, or zero if it cannot be
5132 identified. The value returned by this intrinsic is likely to be incorrect or 0
5133 for arguments other than zero, so it should only be used for debugging purposes.
5134 </p>
5136 <p>
5137 Note that calling this intrinsic does not prevent function inlining or other
5138 aggressive transformations, so the value returned may not be that of the obvious
5139 source-language caller.
5140 </p>
5141 </div>
5143 <!-- _______________________________________________________________________ -->
5146 </div>
5151 <pre>
5152 declare i8 *@llvm.stacksave()
5153 </pre>
5157 <p>
5161 features like scoped automatic variable sized arrays in C99.
5162 </p>
5166 <p>
5167 This intrinsic returns a opaque pointer value that can be passed to <a
5174 </p>
5176 </div>
5178 <!-- _______________________________________________________________________ -->
5181 </div>
5186 <pre>
5187 declare void @llvm.stackrestore(i8 * %ptr)
5188 </pre>
5192 <p>
5194 the function stack to the state it was in when the corresponding <a
5196 useful for implementing language features like scoped automatic variable sized
5197 arrays in C99.
5198 </p>
5202 <p>
5204 </p>
5206 </div>
5209 <!-- _______________________________________________________________________ -->
5212 </div>
5217 <pre>
5219 </pre>
5224 <p>
5226 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
5227 no
5228 effect on the behavior of the program but can change its performance
5229 characteristics.
5230 </p>
5234 <p>
5240 </p>
5244 <p>
5245 This intrinsic does not modify the behavior of the program. In particular,
5246 prefetches cannot trap and do not produce a value. On targets that support this
5247 intrinsic, the prefetch can provide hints to the processor cache for better
5248 performance.
5249 </p>
5251 </div>
5253 <!-- _______________________________________________________________________ -->
5256 </div>
5261 <pre>
5263 </pre>
5268 <p>
5270 (PC) in a region of
5271 code to simulators and other tools. The method is target specific, but it is
5272 expected that the marker will use exported symbols to transmit the PC of the
5273 marker.
5274 The marker makes no guarantees that it will remain with any specific instruction
5275 after optimizations. It is possible that the presence of a marker will inhibit
5276 optimizations. The intended use is to be inserted after optimizations to allow
5277 correlations of simulation runs.
5278 </p>
5282 <p>
5284 </p>
5288 <p>
5289 This intrinsic does not modify the behavior of the program. Backends that do not
5290 support this intrinisic may ignore it.
5291 </p>
5293 </div>
5295 <!-- _______________________________________________________________________ -->
5298 </div>
5303 <pre>
5304 declare i64 @llvm.readcyclecounter( )
5305 </pre>
5310 <p>
5312 counter register (or similar low latency, high accuracy clocks) on those targets
5313 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
5314 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
5315 should only be used for small timings.
5316 </p>
5320 <p>
5321 When directly supported, reading the cycle counter should not modify any memory.
5322 Implementations are allowed to either return a application specific value or a
5323 system wide value. On backends without support, this is lowered to a constant 0.
5324 </p>
5326 </div>
5328 <!-- ======================================================================= -->
5331 </div>
5334 <p>
5335 LLVM provides intrinsics for a few important standard C library functions.
5336 These intrinsics allow source-language front-ends to pass information about the
5337 alignment of the pointer arguments to the code generator, providing opportunity
5338 for more efficient code generation.
5339 </p>
5341 </div>
5343 <!-- _______________________________________________________________________ -->
5346 </div>
5351 <p>This is an overloaded intrinsic. You can use llvm.memcpy on any integer bit
5352 width. Not all targets support all bit widths however.</p>
5353 <pre>
5362 </pre>
5366 <p>
5368 location to the destination location.
5369 </p>
5371 <p>
5373 intrinsics do not return a value, and takes an extra alignment argument.
5374 </p>
5378 <p>
5379 The first argument is a pointer to the destination, the second is a pointer to
5380 the source. The third argument is an integer argument
5381 specifying the number of bytes to copy, and the fourth argument is the alignment
5382 of the source and destination locations.
5383 </p>
5385 <p>
5387 the caller guarantees that both the source and destination pointers are aligned
5388 to that boundary.
5389 </p>
5393 <p>
5395 location to the destination location, which are not allowed to overlap. It
5396 copies "len" bytes of memory over. If the argument is known to be aligned to
5397 some boundary, this can be specified as the fourth argument, otherwise it should
5399 </p>
5400 </div>
5403 <!-- _______________________________________________________________________ -->
5406 </div>
5411 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5412 width. Not all targets support all bit widths however.</p>
5413 <pre>
5422 </pre>
5426 <p>
5428 location to the destination location. It is similar to the
5430 </p>
5432 <p>
5434 intrinsics do not return a value, and takes an extra alignment argument.
5435 </p>
5439 <p>
5440 The first argument is a pointer to the destination, the second is a pointer to
5441 the source. The third argument is an integer argument
5442 specifying the number of bytes to copy, and the fourth argument is the alignment
5443 of the source and destination locations.
5444 </p>
5446 <p>
5448 the caller guarantees that the source and destination pointers are aligned to
5449 that boundary.
5450 </p>
5454 <p>
5456 location to the destination location, which may overlap. It
5457 copies "len" bytes of memory over. If the argument is known to be aligned to
5458 some boundary, this can be specified as the fourth argument, otherwise it should
5460 </p>
5461 </div>
5464 <!-- _______________________________________________________________________ -->
5467 </div>
5472 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5473 width. Not all targets support all bit widths however.</p>
5474 <pre>
5483 </pre>
5487 <p>
5489 byte value.
5490 </p>
5492 <p>
5494 does not return a value, and takes an extra alignment argument.
5495 </p>
5499 <p>
5500 The first argument is a pointer to the destination to fill, the second is the
5501 byte value to fill it with, the third argument is an integer
5502 argument specifying the number of bytes to fill, and the fourth argument is the
5503 known alignment of destination location.
5504 </p>
5506 <p>
5508 the caller guarantees that the destination pointer is aligned to that boundary.
5509 </p>
5513 <p>
5515 the
5516 destination location. If the argument is known to be aligned to some boundary,
5517 this can be specified as the fourth argument, otherwise it should be set to 0 or
5518 1.
5519 </p>
5520 </div>
5523 <!-- _______________________________________________________________________ -->
5526 </div>
5532 floating point or vector of floating point type. Not all targets support all
5533 types however.</p>
5534 <pre>
5535 declare float @llvm.sqrt.f32(float %Val)
5536 declare double @llvm.sqrt.f64(double %Val)
5537 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5538 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5539 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5540 </pre>
5544 <p>
5548 negative numbers other than -0.0 (which allows for better optimization, because
5550 defined to return -0.0 like IEEE sqrt.
5551 </p>
5555 <p>
5556 The argument and return value are floating point numbers of the same type.
5557 </p>
5561 <p>
5562 This function returns the sqrt of the specified operand if it is a nonnegative
5563 floating point number.
5564 </p>
5565 </div>
5567 <!-- _______________________________________________________________________ -->
5570 </div>
5576 floating point or vector of floating point type. Not all targets support all
5577 types however.</p>
5578 <pre>
5579 declare float @llvm.powi.f32(float %Val, i32 %power)
5580 declare double @llvm.powi.f64(double %Val, i32 %power)
5581 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5582 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5583 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5584 </pre>
5588 <p>
5590 specified (positive or negative) power. The order of evaluation of
5591 multiplications is not defined. When a vector of floating point type is
5592 used, the second argument remains a scalar integer value.
5593 </p>
5597 <p>
5598 The second argument is an integer power, and the first is a value to raise to
5599 that power.
5600 </p>
5604 <p>
5605 This function returns the first value raised to the second power with an
5606 unspecified sequence of rounding operations.</p>
5607 </div>
5609 <!-- _______________________________________________________________________ -->
5612 </div>
5618 floating point or vector of floating point type. Not all targets support all
5619 types however.</p>
5620 <pre>
5621 declare float @llvm.sin.f32(float %Val)
5622 declare double @llvm.sin.f64(double %Val)
5623 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5624 declare fp128 @llvm.sin.f128(fp128 %Val)
5625 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5626 </pre>
5630 <p>
5632 </p>
5636 <p>
5637 The argument and return value are floating point numbers of the same type.
5638 </p>
5642 <p>
5643 This function returns the sine of the specified operand, returning the
5645 conditions in the same way.</p>
5646 </div>
5648 <!-- _______________________________________________________________________ -->
5651 </div>
5657 floating point or vector of floating point type. Not all targets support all
5658 types however.</p>
5659 <pre>
5660 declare float @llvm.cos.f32(float %Val)
5661 declare double @llvm.cos.f64(double %Val)
5662 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5663 declare fp128 @llvm.cos.f128(fp128 %Val)
5664 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5665 </pre>
5669 <p>
5671 </p>
5675 <p>
5676 The argument and return value are floating point numbers of the same type.
5677 </p>
5681 <p>
5682 This function returns the cosine of the specified operand, returning the
5684 conditions in the same way.</p>
5685 </div>
5687 <!-- _______________________________________________________________________ -->
5690 </div>
5696 floating point or vector of floating point type. Not all targets support all
5697 types however.</p>
5698 <pre>
5699 declare float @llvm.pow.f32(float %Val, float %Power)
5700 declare double @llvm.pow.f64(double %Val, double %Power)
5701 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5702 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5703 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5704 </pre>
5708 <p>
5710 specified (positive or negative) power.
5711 </p>
5715 <p>
5716 The second argument is a floating point power, and the first is a value to
5717 raise to that power.
5718 </p>
5722 <p>
5723 This function returns the first value raised to the second power,
5724 returning the
5726 conditions in the same way.</p>
5727 </div>
5730 <!-- ======================================================================= -->
5733 </div>
5736 <p>
5737 LLVM provides intrinsics for a few important bit manipulation operations.
5738 These allow efficient code generation for some algorithms.
5739 </p>
5741 </div>
5743 <!-- _______________________________________________________________________ -->
5746 </div>
5751 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5753 <pre>
5757 </pre>
5761 <p>
5763 values with an even number of bytes (positive multiple of 16 bits). These are
5764 useful for performing operations on data that is not in the target's native
5765 byte order.
5766 </p>
5770 <p>
5773 intrinsic returns an i32 value that has the four bytes of the input i32
5778 </p>
5780 </div>
5782 <!-- _______________________________________________________________________ -->
5785 </div>
5790 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5791 width. Not all targets support all bit widths however.</p>
5792 <pre>
5798 </pre>
5802 <p>
5804 value.
5805 </p>
5809 <p>
5810 The only argument is the value to be counted. The argument may be of any
5811 integer type. The return type must match the argument type.
5812 </p>
5816 <p>
5818 </p>
5819 </div>
5821 <!-- _______________________________________________________________________ -->
5824 </div>
5830 integer bit width. Not all targets support all bit widths however.</p>
5831 <pre>
5837 </pre>
5841 <p>
5843 leading zeros in a variable.
5844 </p>
5848 <p>
5849 The only argument is the value to be counted. The argument may be of any
5850 integer type. The return type must match the argument type.
5851 </p>
5855 <p>
5857 in a variable. If the src == 0 then the result is the size in bits of the type
5859 </p>
5860 </div>
5864 <!-- _______________________________________________________________________ -->
5867 </div>
5873 integer bit width. Not all targets support all bit widths however.</p>
5874 <pre>
5880 </pre>
5884 <p>
5886 trailing zeros.
5887 </p>
5891 <p>
5892 The only argument is the value to be counted. The argument may be of any
5893 integer type. The return type must match the argument type.
5894 </p>
5898 <p>
5900 in a variable. If the src == 0 then the result is the size in bits of the type
5902 </p>
5903 </div>
5905 <!-- _______________________________________________________________________ -->
5908 </div>
5914 on any integer bit width.</p>
5915 <pre>
5916 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5917 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5918 </pre>
5922 range of bits from an integer value and returns them in the same bit width as
5923 the original value.</p>
5927 any bit width but they must have the same bit width. The second and third
5934 operates in forward mode.</p>
5938 <ol>
5942 to determine the number of bits to retain.</li>
5945 </ol>
5946 <p>In reverse mode, a similar computation is made except that the bits are
5951 </div>
5955 </div>
5961 on any integer bit width.</p>
5962 <pre>
5963 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5964 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5965 </pre>
5969 of bits in an integer value with another integer value. It returns the integer
5970 with the replaced bits.</p>
5977 type since they specify only a bit index.</p>
5983 operates in forward mode.</p>
5986 truncating it down to the size of the replacement area or zero extending it
5987 up to that size.</p>
5994 <p>In reverse mode, a similar computation is made except that the bits are
6000 <pre>
6006 </pre>
6008 </div>
6010 <!-- ======================================================================= -->
6013 </div>
6016 <p>
6017 LLVM provides intrinsics for some arithmetic with overflow operations.
6018 </p>
6020 </div>
6022 <!-- _______________________________________________________________________ -->
6025 </div>
6032 on any integer bit width.</p>
6034 <pre>
6035 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6036 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6037 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6038 </pre>
6043 a signed addition of the two arguments, and indicate whether an overflow
6044 occurred during the signed summation.</p>
6048 <p>The arguments (%a and %b) and the first element of the result structure may
6049 be of integer types of any bit width, but they must have the same bit width. The
6056 a signed addition of the two variables. They return a structure — the
6057 first element of which is the signed summation, and the second element of which
6058 is a bit specifying if the signed summation resulted in an overflow.</p>
6061 <pre>
6062 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6063 %sum = extractvalue {i32, i1} %res, 0
6064 %obit = extractvalue {i32, i1} %res, 1
6065 br i1 %obit, label %overflow, label %normal
6066 </pre>
6068 </div>
6070 <!-- _______________________________________________________________________ -->
6073 </div>
6080 on any integer bit width.</p>
6082 <pre>
6083 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6084 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6085 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6086 </pre>
6091 an unsigned addition of the two arguments, and indicate whether a carry occurred
6092 during the unsigned summation.</p>
6096 <p>The arguments (%a and %b) and the first element of the result structure may
6097 be of integer types of any bit width, but they must have the same bit width. The
6104 an unsigned addition of the two arguments. They return a structure — the
6105 first element of which is the sum, and the second element of which is a bit
6106 specifying if the unsigned summation resulted in a carry.</p>
6109 <pre>
6110 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6111 %sum = extractvalue {i32, i1} %res, 0
6112 %obit = extractvalue {i32, i1} %res, 1
6113 br i1 %obit, label %carry, label %normal
6114 </pre>
6116 </div>
6118 <!-- _______________________________________________________________________ -->
6121 </div>
6128 on any integer bit width.</p>
6130 <pre>
6131 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6132 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6133 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6134 </pre>
6139 a signed subtraction of the two arguments, and indicate whether an overflow
6140 occurred during the signed subtraction.</p>
6144 <p>The arguments (%a and %b) and the first element of the result structure may
6145 be of integer types of any bit width, but they must have the same bit width. The
6152 a signed subtraction of the two arguments. They return a structure — the
6153 first element of which is the subtraction, and the second element of which is a bit
6154 specifying if the signed subtraction resulted in an overflow.</p>
6157 <pre>
6158 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6159 %sum = extractvalue {i32, i1} %res, 0
6160 %obit = extractvalue {i32, i1} %res, 1
6161 br i1 %obit, label %overflow, label %normal
6162 </pre>
6164 </div>
6166 <!-- _______________________________________________________________________ -->
6169 </div>
6176 on any integer bit width.</p>
6178 <pre>
6179 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6180 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6181 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6182 </pre>
6187 an unsigned subtraction of the two arguments, and indicate whether an overflow
6188 occurred during the unsigned subtraction.</p>
6192 <p>The arguments (%a and %b) and the first element of the result structure may
6193 be of integer types of any bit width, but they must have the same bit width. The
6200 an unsigned subtraction of the two arguments. They return a structure — the
6201 first element of which is the subtraction, and the second element of which is a bit
6202 specifying if the unsigned subtraction resulted in an overflow.</p>
6205 <pre>
6206 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6207 %sum = extractvalue {i32, i1} %res, 0
6208 %obit = extractvalue {i32, i1} %res, 1
6209 br i1 %obit, label %overflow, label %normal
6210 </pre>
6212 </div>
6214 <!-- _______________________________________________________________________ -->
6217 </div>
6224 on any integer bit width.</p>
6226 <pre>
6227 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6228 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6229 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6230 </pre>
6235 a signed multiplication of the two arguments, and indicate whether an overflow
6236 occurred during the signed multiplication.</p>
6240 <p>The arguments (%a and %b) and the first element of the result structure may
6241 be of integer types of any bit width, but they must have the same bit width. The
6248 a signed multiplication of the two arguments. They return a structure —
6249 the first element of which is the multiplication, and the second element of
6250 which is a bit specifying if the signed multiplication resulted in an
6251 overflow.</p>
6254 <pre>
6255 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6256 %sum = extractvalue {i32, i1} %res, 0
6257 %obit = extractvalue {i32, i1} %res, 1
6258 br i1 %obit, label %overflow, label %normal
6259 </pre>
6261 </div>
6263 <!-- _______________________________________________________________________ -->
6266 </div>
6273 on any integer bit width.</p>
6275 <pre>
6276 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6277 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6278 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6279 </pre>
6284 actively being fixed, but it should not currently be used!</i></p>
6287 a unsigned multiplication of the two arguments, and indicate whether an overflow
6288 occurred during the unsigned multiplication.</p>
6292 <p>The arguments (%a and %b) and the first element of the result structure may
6293 be of integer types of any bit width, but they must have the same bit width. The
6296 multiplication.</p>
6301 an unsigned multiplication of the two arguments. They return a structure —
6302 the first element of which is the multiplication, and the second element of
6303 which is a bit specifying if the unsigned multiplication resulted in an
6304 overflow.</p>
6307 <pre>
6308 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6309 %sum = extractvalue {i32, i1} %res, 0
6310 %obit = extractvalue {i32, i1} %res, 1
6311 br i1 %obit, label %overflow, label %normal
6312 </pre>
6314 </div>
6316 <!-- ======================================================================= -->
6319 </div>
6322 <p>
6324 are described in the <a
6326 Debugging</a> document.
6327 </p>
6328 </div>
6331 <!-- ======================================================================= -->
6334 </div>
6337 <p> The LLVM exception handling intrinsics (which all start with
6341 </div>
6343 <!-- ======================================================================= -->
6346 </div>
6349 <p>
6350 This intrinsic makes it possible to excise one parameter, marked with
6352 function pointer lacking the nest parameter - the caller does not need
6353 to provide a value for it. Instead, the value to use is stored in
6354 advance in a "trampoline", a block of memory usually allocated
6355 on the stack, which also contains code to splice the nest value into the
6356 argument list. This is used to implement the GCC nested function address
6357 extension.
6358 </p>
6359 <p>
6360 For example, if the function is
6363 <pre>
6366 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6367 %fp = bitcast i8* %p to i32 (i32, i32)*
6368 </pre>
6371 </div>
6373 <!-- _______________________________________________________________________ -->
6376 </div>
6379 <pre>
6381 </pre>
6383 <p>
6385 and returns a function pointer suitable for executing it.
6386 </p>
6388 <p>
6391 and sufficiently aligned block of memory; this memory is written to by the
6392 intrinsic. Note that the size and the alignment are target-specific - LLVM
6393 currently provides no portable way of determining them, so a front-end that
6394 generates this intrinsic needs to have some target-specific knowledge.
6396 </p>
6398 <p>
6400 dependent code, turning it into a function. A pointer to this function is
6401 returned, but needs to be bitcast to an
6403 before being called. The new function's signature is the same as that of
6406 of pointer type. Calling the new function is equivalent to calling
6410 modified, then the effect of any later call to the returned function pointer is
6411 undefined.
6412 </p>
6413 </div>
6415 <!-- ======================================================================= -->
6418 </div>
6421 <p>
6422 These intrinsic functions expand the "universal IR" of LLVM to represent
6423 hardware constructs for atomic operations and memory synchronization. This
6424 provides an interface to the hardware, not an interface to the programmer. It
6425 is aimed at a low enough level to allow any programming models or APIs
6426 (Application Programming Interfaces) which
6427 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
6428 hardware behavior. Just as hardware provides a "universal IR" for source
6429 languages, it also provides a starting point for developing a "universal"
6430 atomic operation and synchronization IR.
6431 </p>
6432 <p>
6434 software transaction memory systems, atomic primitives, and intrinsic
6435 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6436 application libraries. The hardware interface provided by LLVM should allow
6437 a clean implementation of all of these APIs and parallel programming models.
6438 No one model or paradigm should be selected above others unless the hardware
6439 itself ubiquitously does so.
6441 </p>
6442 </div>
6444 <!-- _______________________________________________________________________ -->
6447 </div>
6450 <pre>
6454 </pre>
6456 <p>
6458 specific pairs of memory access types.
6459 </p>
6461 <p>
6463 The first four arguments enables a specific barrier as listed below. The fith
6464 argument specifies that the barrier applies to io or device or uncached memory.
6466 </p>
6467 <ul>
6473 </ul>
6475 <p>
6476 This intrinsic causes the system to enforce some ordering constraints upon
6477 the loads and stores of the program. This barrier does not indicate
6479 which they occur. For any of the specified pairs of load and store operations
6480 (f.ex. load-load, or store-load), all of the first operations preceding the
6481 barrier will complete before any of the second operations succeeding the
6482 barrier begin. Specifically the semantics for each pairing is as follows:
6483 </p>
6484 <ul>
6486 after the barrier begins.</li>
6489 store after the barrier begins.</li>
6491 store after the barrier begins.</li>
6493 load after the barrier begins.</li>
6494 </ul>
6495 <p>
6496 These semantics are applied with a logical "and" behavior when more than one
6497 is enabled in a single memory barrier intrinsic.
6498 </p>
6499 <p>
6500 Backends may implement stronger barriers than those requested when they do not
6501 support as fine grained a barrier as requested. Some architectures do not
6502 need all types of barriers and on such architectures, these become noops.
6503 </p>
6505 <pre>
6506 %ptr = malloc i32
6507 store i32 4, %ptr
6510 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6513 </pre>
6514 </div>
6516 <!-- _______________________________________________________________________ -->
6519 </div>
6522 <p>
6524 any integer bit width and for different address spaces. Not all targets
6525 support all bit widths however.</p>
6527 <pre>
6529 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6530 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6531 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6533 </pre>
6535 <p>
6536 This loads a value in memory and compares it to a given value. If they are
6537 equal, it stores a new value into the memory.
6538 </p>
6540 <p>
6544 this integer type. While any bit width integer may be used, targets may only
6545 lower representations they support in hardware.
6547 </p>
6549 <p>
6550 This entire intrinsic must be executed atomically. It first loads the value
6553 loaded value is yielded in all cases. This provides the equivalent of an
6554 atomic compare-and-swap operation within the SSA framework.
6555 </p>
6558 <pre>
6559 %ptr = malloc i32
6560 store i32 4, %ptr
6563 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6569 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6574 </pre>
6575 </div>
6577 <!-- _______________________________________________________________________ -->
6580 </div>
6584 <p>
6586 integer bit width. Not all targets support all bit widths however.</p>
6587 <pre>
6593 </pre>
6595 <p>
6599 </p>
6602 <p>
6606 integer type. The targets may only lower integer representations they
6607 support.
6608 </p>
6610 <p>
6613 equivalent of an atomic swap operation within the SSA framework.
6615 </p>
6617 <pre>
6618 %ptr = malloc i32
6619 store i32 4, %ptr
6622 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6628 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6633 </pre>
6634 </div>
6636 <!-- _______________________________________________________________________ -->
6640 </div>
6643 <p>
6645 integer bit width. Not all targets support all bit widths however.</p>
6646 <pre>
6652 </pre>
6654 <p>
6657 </p>
6659 <p>
6661 The intrinsic takes two arguments, the first a pointer to an integer value
6662 and the second an integer value. The result is also an integer value. These
6663 integer types can have any bit width, but they must all have the same bit
6664 width. The targets may only lower integer representations they support.
6665 </p>
6667 <p>
6668 This intrinsic does a series of operations atomically. It first loads the
6671 </p>
6674 <pre>
6675 %ptr = malloc i32
6676 store i32 4, %ptr
6677 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6679 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6681 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6684 </pre>
6685 </div>
6687 <!-- _______________________________________________________________________ -->
6691 </div>
6694 <p>
6696 any integer bit width and for different address spaces. Not all targets
6697 support all bit widths however.</p>
6698 <pre>
6704 </pre>
6706 <p>
6709 </p>
6711 <p>
6713 The intrinsic takes two arguments, the first a pointer to an integer value
6714 and the second an integer value. The result is also an integer value. These
6715 integer types can have any bit width, but they must all have the same bit
6716 width. The targets may only lower integer representations they support.
6717 </p>
6719 <p>
6720 This intrinsic does a series of operations atomically. It first loads the
6723 </p>
6726 <pre>
6727 %ptr = malloc i32
6728 store i32 8, %ptr
6729 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6731 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6733 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6736 </pre>
6737 </div>
6739 <!-- _______________________________________________________________________ -->
6746 </div>
6749 <p>
6753 address spaces. Not all targets support all bit widths however.</p>
6754 <pre>
6760 </pre>
6762 <pre>
6768 </pre>
6770 <pre>
6776 </pre>
6778 <pre>
6784 </pre>
6786 <p>
6790 </p>
6792 <p>
6794 These intrinsics take two arguments, the first a pointer to an integer value
6795 and the second an integer value. The result is also an integer value. These
6796 integer types can have any bit width, but they must all have the same bit
6797 width. The targets may only lower integer representations they support.
6798 </p>
6800 <p>
6801 These intrinsics does a series of operations atomically. They first load the
6805 </p>
6808 <pre>
6809 %ptr = malloc i32
6810 store i32 0x0F0F, %ptr
6811 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6813 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6815 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6817 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6820 </pre>
6821 </div>
6824 <!-- _______________________________________________________________________ -->
6831 </div>
6834 <p>
6838 address spaces. Not all targets
6839 support all bit widths however.</p>
6840 <pre>
6846 </pre>
6848 <pre>
6854 </pre>
6856 <pre>
6862 </pre>
6864 <pre>
6870 </pre>
6872 <p>
6873 These intrinsics takes the signed or unsigned minimum or maximum of
6876 </p>
6878 <p>
6880 These intrinsics take two arguments, the first a pointer to an integer value
6881 and the second an integer value. The result is also an integer value. These
6882 integer types can have any bit width, but they must all have the same bit
6883 width. The targets may only lower integer representations they support.
6884 </p>
6886 <p>
6887 These intrinsics does a series of operations atomically. They first load the
6891 </p>
6894 <pre>
6895 %ptr = malloc i32
6896 store i32 7, %ptr
6897 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6899 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6901 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6903 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6906 </pre>
6907 </div>
6909 <!-- ======================================================================= -->
6912 </div>
6915 <p> This class of intrinsics is designed to be generic and has
6916 no specific purpose. </p>
6917 </div>
6919 <!-- _______________________________________________________________________ -->
6922 </div>
6927 <pre>
6928 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6929 </pre>
6933 <p>
6935 </p>
6939 <p>
6940 The first argument is a pointer to a value, the second is a pointer to a
6941 global string, the third is a pointer to a global string which is the source
6942 file name, and the last argument is the line number.
6943 </p>
6947 <p>
6948 This intrinsic allows annotation of local variables with arbitrary strings.
6949 This can be useful for special purpose optimizations that want to look for these
6950 annotations. These have no other defined use, they are ignored by code
6951 generation and optimization.
6952 </p>
6953 </div>
6955 <!-- _______________________________________________________________________ -->
6958 </div>
6964 any integer bit width.
6965 </p>
6966 <pre>
6967 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6968 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6969 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6970 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6971 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6972 </pre>
6976 <p>
6978 </p>
6982 <p>
6983 The first argument is an integer value (result of some expression),
6984 the second is a pointer to a global string, the third is a pointer to a global
6985 string which is the source file name, and the last argument is the line number.
6986 It returns the value of the first argument.
6987 </p>
6991 <p>
6992 This intrinsic allows annotations to be put on arbitrary expressions
6993 with arbitrary strings. This can be useful for special purpose optimizations
6994 that want to look for these annotations. These have no other defined use, they
6995 are ignored by code generation and optimization.
6996 </p>
6997 </div>
6999 <!-- _______________________________________________________________________ -->
7002 </div>
7007 <pre>
7008 declare void @llvm.trap()
7009 </pre>
7013 <p>
7015 </p>
7019 <p>
7020 None
7021 </p>
7025 <p>
7026 This intrinsics is lowered to the target dependent trap instruction. If the
7027 target does not have a trap instruction, this intrinsic will be lowered to the
7028 call of the abort() function.
7029 </p>
7030 </div>
7032 <!-- _______________________________________________________________________ -->
7035 </div>
7038 <pre>
7041 </pre>
7043 <p>
7046 it is placed on the stack before local variables.
7047 </p>
7049 <p>
7051 first argument is the value loaded from the stack guard
7053 has enough space to hold the value of the guard.
7054 </p>
7056 <p>
7057 This intrinsic causes the prologue/epilogue inserter to force the position of
7059 stack. This is to ensure that if a local variable on the stack is overwritten,
7060 it will destroy the value of the guard. When the function exits, the guard on
7061 the stack is checked against the original guard. If they're different, then
7063 </p>
7064 </div>
7066 <!-- *********************************************************************** -->
7067 <hr>
7068 <address>
7076 Last modified: $Date$
7077 </address>
7079 </body>
7080 </html>