| blob | f15f139aeb384cbd216331638de5c219a76783f6 |
3 <html>
4 <head>
11 </head>
13 <body>
16 <ol>
21 <ol>
24 <ol>
28 <li><a href="#linkage_linker_private_weak_def_auto">'<tt>linker_private_weak_def_auto</tt>' Linkage</a></li>
41 </ol>
42 </li>
56 </ol>
57 </li>
59 <ol>
62 <ol>
69 </ol>
70 </li>
72 <ol>
74 <ol>
79 </ol>
80 </li>
84 </ol>
85 </li>
87 </ol>
88 </li>
90 <ol>
98 </ol>
99 </li>
101 <ol>
104 </ol>
105 </li>
107 <ol>
110 Global Variable</a></li>
112 Global Variable</a></li>
114 Global Variable</a></li>
115 </ol>
116 </li>
118 <ol>
120 <ol>
128 </ol>
129 </li>
131 <ol>
144 </ol>
145 </li>
147 <ol>
154 </ol>
155 </li>
157 <ol>
161 </ol>
162 </li>
164 <ol>
167 </ol>
168 </li>
170 <ol>
175 </ol>
176 </li>
178 <ol>
191 </ol>
192 </li>
194 <ol>
201 </ol>
202 </li>
203 </ol>
204 </li>
206 <ol>
208 <ol>
212 </ol>
213 </li>
215 <ol>
219 </ol>
220 </li>
222 <ol>
230 </ol>
231 </li>
233 <ol>
242 </ol>
243 </li>
245 <ol>
250 </ol>
251 </li>
253 <ol>
260 </ol>
261 </li>
263 <ol>
266 </ol>
267 </li>
271 <ol>
273 </ol>
274 </li>
276 <ol>
290 </ol>
291 </li>
293 <ol>
298 </ol>
299 </li>
301 <ol>
312 </ol>
313 </li>
314 </ol>
315 </li>
316 </ol>
321 </div>
323 <!-- *********************************************************************** -->
325 <!-- *********************************************************************** -->
329 <p>This document is a reference manual for the LLVM assembly language. LLVM is
330 a Static Single Assignment (SSA) based representation that provides type
331 safety, low-level operations, flexibility, and the capability of representing
332 'all' high-level languages cleanly. It is the common code representation
333 used throughout all phases of the LLVM compilation strategy.</p>
335 </div>
337 <!-- *********************************************************************** -->
339 <!-- *********************************************************************** -->
343 <p>The LLVM code representation is designed to be used in three different forms:
344 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
345 for fast loading by a Just-In-Time compiler), and as a human readable
346 assembly language representation. This allows LLVM to provide a powerful
347 intermediate representation for efficient compiler transformations and
348 analysis, while providing a natural means to debug and visualize the
349 transformations. The three different forms of LLVM are all equivalent. This
350 document describes the human readable representation and notation.</p>
352 <p>The LLVM representation aims to be light-weight and low-level while being
353 expressive, typed, and extensible at the same time. It aims to be a
354 "universal IR" of sorts, by being at a low enough level that high-level ideas
355 may be cleanly mapped to it (similar to how microprocessors are "universal
356 IR's", allowing many source languages to be mapped to them). By providing
357 type information, LLVM can be used as the target of optimizations: for
358 example, through pointer analysis, it can be proven that a C automatic
359 variable is never accessed outside of the current function, allowing it to
360 be promoted to a simple SSA value instead of a memory location.</p>
362 </div>
364 <!-- _______________________________________________________________________ -->
369 <p>It is important to note that this document describes 'well formed' LLVM
370 assembly language. There is a difference between what the parser accepts and
371 what is considered 'well formed'. For example, the following instruction is
372 syntactically okay, but not well formed:</p>
376 </pre>
379 LLVM infrastructure provides a verification pass that may be used to verify
380 that an LLVM module is well formed. This pass is automatically run by the
381 parser after parsing input assembly and by the optimizer before it outputs
382 bitcode. The violations pointed out by the verifier pass indicate bugs in
383 transformation passes or input to the parser.</p>
385 </div>
387 <!-- Describe the typesetting conventions here. -->
389 <!-- *********************************************************************** -->
391 <!-- *********************************************************************** -->
395 <p>LLVM identifiers come in two basic types: global and local. Global
397 character. Local identifiers (register names, types) begin with
399 for identifiers, for different purposes:</p>
401 <ol>
402 <li>Named values are represented as a string of characters with their prefix.
406 other characters in their names can be surrounded with quotes. Special
408 ASCII code for the character in hexadecimal. In this way, any character
409 can be used in a name value, even quotes themselves.</li>
411 <li>Unnamed values are represented as an unsigned numeric value with their
416 </ol>
418 <p>LLVM requires that values start with a prefix for two reasons: Compilers
419 don't need to worry about name clashes with reserved words, and the set of
420 reserved words may be expanded in the future without penalty. Additionally,
421 unnamed identifiers allow a compiler to quickly come up with a temporary
422 variable without having to avoid symbol table conflicts.</p>
424 <p>Reserved words in LLVM are very similar to reserved words in other
425 languages. There are keywords for different opcodes
431 reserved words cannot conflict with variable names, because none of them
434 <p>Here is an example of LLVM code to multiply the integer variable
441 </pre>
447 </pre>
455 </pre>
458 lexical features of LLVM:</p>
460 <ol>
462 line.</li>
464 <li>Unnamed temporaries are created when the result of a computation is not
465 assigned to a named value.</li>
468 </ol>
470 <p>It also shows a convention that we follow in this document. When
471 demonstrating instructions, we will follow an instruction with a comment that
472 defines the type and name of value produced. Comments are shown in italic
473 text.</p>
475 </div>
477 <!-- *********************************************************************** -->
479 <!-- *********************************************************************** -->
481 <!-- ======================================================================= -->
483 </div>
488 of the input programs. Each module consists of functions, global variables,
489 and symbol table entries. Modules may be combined together with the LLVM
490 linker, which merges function (and global variable) definitions, resolves
491 forward declarations, and merges symbol table entries. Here is an example of
496 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
504 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8*</i>
509 }
513 !foo = !{!1, null}
514 </pre>
522 <p>In general, a module is made up of a list of global values, where both
523 functions and global variables are global values. Global values are
524 represented by a pointer to a memory location (in this case, a pointer to an
525 array of char, and a pointer to a function), and have one of the
528 </div>
530 <!-- ======================================================================= -->
533 </div>
537 <p>All Global Variables and Functions have one of the following types of
538 linkage:</p>
540 <dl>
543 by objects in the current module. In particular, linking code into a
544 module with an private global value may cause the private to be renamed as
545 necessary to avoid collisions. Because the symbol is private to the
546 module, all references can be updated. This doesn't show up in any symbol
547 table in the object file.</dd>
551 assembler and evaluated by the linker. Unlike normal strong symbols, they
552 are removed by the linker from the final linked image (executable or
553 dynamic library).</dd>
558 linker. The symbols are removed by the linker from the final linked image
559 (executable or dynamic library).</dd>
561 <dt><tt><b><a name="linkage_linker_private_weak_def_auto">linker_private_weak_def_auto</a></b></tt></dt>
563 of the object is not taken. For instance, functions that had an inline
564 definition, but the compiler decided not to inline it. Note,
567 visibility. The symbols are removed by the linker from the final linked
568 image (executable or dynamic library).</dd>
571 <dd>Similar to private, but the value shows as a local symbol
577 into the object file corresponding to the LLVM module. They exist to
578 allow inlining and other optimizations to take place given knowledge of
579 the definition of the global, which is known to be somewhere outside the
582 This linkage type is only allowed on definitions, not declarations.</dd>
586 the same name when linkage occurs. This can be used to implement
587 some forms of inline functions, templates, or other code which must be
588 generated in each translation unit that uses it, but where the body may
589 be overridden with a more definitive definition later. Unreferenced
592 inline the body of this function into callers because it doesn't know if
593 this definition of the function is the definitive definition within the
594 program or whether it will be overridden by a stronger definition.
595 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
596 linkage.</dd>
606 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
607 global scope.
608 Symbols with "<tt>common</tt>" linkage are merged in the same way as
611 must have a zero initializer, and may not be marked '<a
613 have common linkage.</dd>
618 pointer to array type. When two global variables with appending linkage
619 are linked together, the two global arrays are appended together. This is
620 the LLVM, typesafe, equivalent of having the system linker append together
624 <dd>The semantics of this linkage follow the ELF object file model: the symbol
625 is weak until linked, if not linked, the symbol becomes null instead of
626 being an undefined reference.</dd>
630 <dd>Some languages allow differing globals to be merged, such as two functions
632 that only equivalent globals are ever merged (the "one definition rule"
635 be merged with equivalent globals. These linkage types are otherwise the
639 <dd>If none of the above identifiers are used, the global is externally
640 visible, meaning that it participates in linkage and can be used to
641 resolve external symbol references.</dd>
642 </dl>
644 <p>The next two types of linkage are targeted for Microsoft Windows platform
645 only. They are designed to support importing (exporting) symbols from (to)
646 DLLs (Dynamic Link Libraries).</p>
648 <dl>
651 or variable via a global pointer to a pointer that is set up by the DLL
652 exporting the symbol. On Microsoft Windows targets, the pointer name is
654 name.</dd>
658 pointer to a pointer in a DLL, so that it can be referenced with the
661 variable name.</dd>
662 </dl>
665 another module defined a "<tt>.LC0</tt>" variable and was linked with this
666 one, one of the two would be renamed, preventing a collision. Since
668 declarations), they are accessible outside of the current module.</p>
677 </div>
679 <!-- ======================================================================= -->
682 </div>
688 convention specified for the call. The calling convention of any pair of
689 dynamic caller/callee must match, or the behavior of the program is
690 undefined. The following calling conventions are supported by LLVM, and more
691 may be added in the future:</p>
693 <dl>
695 <dd>This calling convention (the default if no other calling convention is
696 specified) matches the target C calling conventions. This calling
697 convention supports varargs function calls and tolerates some mismatch in
698 the declared prototype and implemented declaration of the function (as
699 does normal C).</dd>
702 <dd>This calling convention attempts to make calls as fast as possible
703 (e.g. by passing things in registers). This calling convention allows the
704 target to use whatever tricks it wants to produce fast code for the
705 target, without having to conform to an externally specified ABI
706 (Application Binary Interface).
708 when this or the GHC convention is used.</a> This calling convention
709 does not support varargs and requires the prototype of all callees to
710 exactly match the prototype of the function definition.</dd>
713 <dd>This calling convention attempts to make code in the caller as efficient
714 as possible under the assumption that the call is not commonly executed.
715 As such, these calls often preserve all registers so that the call does
716 not break any live ranges in the caller side. This calling convention
717 does not support varargs and requires the prototype of all callees to
718 exactly match the prototype of the function definition.</dd>
721 <dd>This calling convention has been implemented specifically for use by the
723 It passes everything in registers, going to extremes to achieve this by
724 disabling callee save registers. This calling convention should not be
725 used lightly but only for specific situations such as an alternative to
727 implementing functional programming languages.At the moment only X86
728 supports this convention and it has the following limitations:
729 <ul>
731 floating point types are supported.</li>
734 </ul>
735 This calling convention supports
737 requires both the caller and callee are using it.
738 </dd>
741 <dd>Any calling convention may be specified by number, allowing
742 target-specific calling conventions to be used. Target specific calling
744 </dl>
746 <p>More calling conventions can be added/defined on an as-needed basis, to
747 support Pascal conventions or any other well-known target-independent
748 convention.</p>
750 </div>
752 <!-- ======================================================================= -->
755 </div>
759 <p>All Global Variables and Functions have one of the following visibility
760 styles:</p>
762 <dl>
764 <dd>On targets that use the ELF object file format, default visibility means
765 that the declaration is visible to other modules and, in shared libraries,
766 means that the declared entity may be overridden. On Darwin, default
767 visibility means that the declaration is visible to other modules. Default
771 <dd>Two declarations of an object with hidden visibility refer to the same
772 object if they are in the same shared object. Usually, hidden visibility
773 indicates that the symbol will not be placed into the dynamic symbol
774 table, so no other module (executable or shared library) can reference it
775 directly.</dd>
778 <dd>On ELF, protected visibility indicates that the symbol will be placed in
779 the dynamic symbol table, but that references within the defining module
780 will bind to the local symbol. That is, the symbol cannot be overridden by
781 another module.</dd>
782 </dl>
784 </div>
786 <!-- ======================================================================= -->
789 </div>
793 <p>LLVM IR allows you to specify name aliases for certain types. This can make
794 it easier to read the IR and make the IR more condensed (particularly when
795 recursive types are involved). An example of a name specification is:</p>
798 %mytype = type { %mytype*, i32 }
799 </pre>
805 <p>Note that type names are aliases for the structural type that they indicate,
806 and that you can therefore specify multiple names for the same type. This
807 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
808 uses structural typing, the name is not part of the type. When printing out
810 particular shape. This means that if you have code where two different
811 source types end up having the same LLVM type, that the dumper will sometimes
812 print the "wrong" or unexpected type. This is an important design point and
813 isn't going to change.</p>
815 </div>
817 <!-- ======================================================================= -->
820 </div>
824 <p>Global variables define regions of memory allocated at compilation time
825 instead of run-time. Global variables may optionally be initialized, may
826 have an explicit section to be placed in, and may have an optional explicit
827 alignment specified. A variable may be defined as "thread_local", which
828 means that it will not be shared by threads (each thread will have a
829 separated copy of the variable). A variable may be defined as a global
830 "constant," which indicates that the contents of the variable
832 global data to be placed in the read-only section of an executable, etc).
833 Note that variables that need runtime initialization cannot be marked
837 constant, even if the final definition of the global is not. This capability
838 can be used to enable slightly better optimization of the program, but
839 requires the language definition to guarantee that optimizations based on the
840 'constantness' are valid for the translation units that do not include the
841 definition.</p>
843 <p>As SSA values, global variables define pointer values that are in scope
844 (i.e. they dominate) all basic blocks in the program. Global variables
845 always define a pointer to their "content" type because they describe a
846 region of memory, and all memory objects in LLVM are accessed through
847 pointers.</p>
849 <p>A global variable may be declared to reside in a target-specific numbered
850 address space. For targets that support them, address spaces may affect how
851 optimizations are performed and/or what target instructions are used to
852 access the variable. The default address space is zero. The address space
853 qualifier must precede any other attributes.</p>
855 <p>LLVM allows an explicit section to be specified for globals. If the target
856 supports it, it will emit globals to the section specified.</p>
858 <p>An explicit alignment may be specified for a global, which must be a power
859 of 2. If not present, or if the alignment is set to zero, the alignment of
860 the global is set by the target to whatever it feels convenient. If an
861 explicit alignment is specified, the global is forced to have exactly that
862 alignment. Targets and optimizers are not allowed to over-align the global
863 if the global has an assigned section. In this case, the extra alignment
864 could be observable: for example, code could assume that the globals are
865 densely packed in their section and try to iterate over them as an array,
866 alignment padding would break this iteration.</p>
868 <p>For example, the following defines a global in a numbered address space with
869 an initializer, section, and alignment:</p>
873 </pre>
875 </div>
878 <!-- ======================================================================= -->
881 </div>
890 name, a (possibly empty) argument list (each with optional
894 curly brace, a list of basic blocks, and a closing curly brace.</p>
901 name, a possibly empty list of arguments, an optional alignment, and an
904 <p>A function definition contains a list of basic blocks, forming the CFG
905 (Control Flow Graph) for the function. Each basic block may optionally start
906 with a label (giving the basic block a symbol table entry), contains a list
908 instruction (such as a branch or function return).</p>
910 <p>The first basic block in a function is special in two ways: it is immediately
911 executed on entrance to the function, and it is not allowed to have
912 predecessor basic blocks (i.e. there can not be any branches to the entry
913 block of a function). Because the block can have no predecessors, it also
916 <p>LLVM allows an explicit section to be specified for functions. If the target
917 supports it, it will emit functions to the section specified.</p>
919 <p>An explicit alignment may be specified for a function. If not present, or if
920 the alignment is set to zero, the alignment of the function is set by the
921 target to whatever it feels convenient. If an explicit alignment is
922 specified, the function is forced to have at least that much alignment. All
932 </pre>
934 </div>
936 <!-- ======================================================================= -->
939 </div>
944 function, global variable, another alias or bitcast of global value). Aliases
951 </pre>
953 </div>
955 <!-- ======================================================================= -->
958 </div>
963 nodes</a> (but not metadata strings) are the only valid operands for
964 a named metadata.</p>
968 ; Some unnamed metadata nodes, which are referenced by the named metadata.
972 ; A named metadata.
974 </pre>
976 </div>
978 <!-- ======================================================================= -->
983 <p>The return type and each parameter of a function type may have a set of
985 used to communicate additional information about the result or parameters of
986 a function. Parameter attributes are considered to be part of the function,
987 not of the function type, so functions with different parameter attributes
988 can have the same function type.</p>
990 <p>Parameter attributes are simple keywords that follow the type specified. If
991 multiple parameter attributes are needed, they are space separated. For
992 example:</p>
995 declare i32 @printf(i8* noalias nocapture, ...)
996 declare i32 @atoi(i8 zeroext)
997 declare signext i8 @returns_signed_char()
998 </pre>
1005 <dl>
1007 <dd>This indicates to the code generator that the parameter or return value
1008 should be zero-extended to a 32-bit value by the caller (for a parameter)
1009 or the callee (for a return value).</dd>
1012 <dd>This indicates to the code generator that the parameter or return value
1013 should be sign-extended to a 32-bit value by the caller (for a parameter)
1014 or the callee (for a return value).</dd>
1017 <dd>This indicates that this parameter or return value should be treated in a
1018 special target-dependent fashion during while emitting code for a function
1019 call or return (usually, by putting it in a register as opposed to memory,
1020 though some targets use it to distinguish between two different kinds of
1021 registers). Use of this attribute is target-specific.</dd>
1024 <dd>This indicates that the pointer parameter should really be passed by value
1025 to the function. The attribute implies that a hidden copy of the pointee
1026 is made between the caller and the callee, so the callee is unable to
1027 modify the value in the callee. This attribute is only valid on LLVM
1028 pointer arguments. It is generally used to pass structs and arrays by
1029 value, but is also valid on pointers to scalars. The copy is considered
1030 to belong to the caller not the callee (for example,
1033 values. The byval attribute also supports specifying an alignment with
1034 the align attribute. This has a target-specific effect on the code
1035 generator that usually indicates a desired alignment for the synthesized
1036 stack slot.</dd>
1039 <dd>This indicates that the pointer parameter specifies the address of a
1040 structure that is the return value of the function in the source program.
1041 This pointer must be guaranteed by the caller to be valid: loads and
1042 stores to the structure may be assumed by the callee to not to trap. This
1043 may only be applied to the first parameter. This is not a valid attribute
1044 for return values. </dd>
1047 <dd>This indicates that pointer values
1050 ignoring certain "irrelevant" dependencies.
1051 For a call to the parent function, dependencies between memory
1052 references from before or after the call and from those during the call
1054 return value used in that call.
1055 The caller shares the responsibility with the callee for ensuring that
1056 these requirements are met.
1057 For further details, please see the discussion of the NoAlias response in
1059 <br>
1062 arguments, though it is slightly weaker.
1063 <br>
1066 </dd>
1069 <dd>This indicates that the callee does not make any copies of the pointer
1070 that outlive the callee itself. This is not a valid attribute for return
1071 values.</dd>
1074 <dd>This indicates that the pointer parameter can be excised using the
1076 attribute for return values.</dd>
1077 </dl>
1079 </div>
1081 <!-- ======================================================================= -->
1084 </div>
1088 <p>Each function may specify a garbage collector name, which is simply a
1089 string:</p>
1092 define void @f() gc "name" { ... }
1093 </pre>
1096 collector which will cause the compiler to alter its output in order to
1097 support the named garbage collection algorithm.</p>
1099 </div>
1101 <!-- ======================================================================= -->
1104 </div>
1108 <p>Function attributes are set to communicate additional information about a
1109 function. Function attributes are considered to be part of the function, not
1110 of the function type, so functions with different parameter attributes can
1111 have the same function type.</p>
1113 <p>Function attributes are simple keywords that follow the type specified. If
1114 multiple attributes are needed, they are space separated. For example:</p>
1117 define void @f() noinline { ... }
1118 define void @f() alwaysinline { ... }
1119 define void @f() alwaysinline optsize { ... }
1120 define void @f() optsize { ... }
1121 </pre>
1123 <dl>
1125 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1126 the backend should forcibly align the stack pointer. Specify the
1127 desired alignment, which must be a power of two, in parentheses.
1130 <dd>This attribute indicates that the inliner should attempt to inline this
1131 function into callers whenever possible, ignoring any active inlining size
1132 threshold for this caller.</dd>
1135 <dd>This attribute indicates that the function should be 'hotpatchable',
1136 meaning the function can be patched and/or hooked even while it is
1137 loaded into memory. On x86, the function prologue will be preceded
1138 by six bytes of padding and will begin with a two-byte instruction.
1139 Most of the functions in the Windows system DLLs in Windows XP SP2 or
1140 higher were compiled in this fashion.</dd>
1143 <dd>This attribute indicates that the source code contained a hint that inlining
1144 this function is desirable (such as the "inline" keyword in C/C++). It
1145 is just a hint; it imposes no requirements on the inliner.</dd>
1148 <dd>This attribute disables prologue / epilogue emission for the function.
1149 This can have very system-specific consequences.</dd>
1155 <dd>This attribute indicates that the inliner should never inline this
1156 function in any situation. This attribute may not be used together with
1160 <dd>This attribute indicates that the code generator should not use a red
1161 zone, even if the target-specific ABI normally permits it.</dd>
1164 <dd>This function attribute indicates that the function never returns
1165 normally. This produces undefined behavior at runtime if the function
1166 ever does dynamically return.</dd>
1169 <dd>This function attribute indicates that the function never returns with an
1170 unwind or exceptional control flow. If the function does unwind, its
1171 runtime behavior is undefined.</dd>
1174 <dd>This attribute suggests that optimization passes and code generator passes
1175 make choices that keep the code size of this function low, and otherwise
1176 do optimizations specifically to reduce code size.</dd>
1179 <dd>This attribute indicates that the function computes its result (or decides
1180 to unwind an exception) based strictly on its arguments, without
1181 dereferencing any pointer arguments or otherwise accessing any mutable
1182 state (e.g. memory, control registers, etc) visible to caller functions.
1183 It does not write through any pointer arguments
1185 changes any state visible to callers. This means that it cannot unwind
1190 <dd>This attribute indicates that the function does not write through any
1192 arguments) or otherwise modify any state (e.g. memory, control registers,
1193 etc) visible to caller functions. It may dereference pointer arguments
1194 and read state that may be set in the caller. A readonly function always
1195 returns the same value (or unwinds an exception identically) when called
1196 with the same set of arguments and global state. It cannot unwind an
1201 <dd>This attribute indicates that the function should emit a stack smashing
1203 the stack before the local variables that's checked upon return from the
1204 function to see if it has been overwritten. A heuristic is used to
1205 determine if a function needs stack protectors or not.<br>
1206 <br>
1213 stack smashing protector. This overrides
1215 <br>
1220 </dl>
1222 </div>
1224 <!-- ======================================================================= -->
1227 </div>
1232 the GCC "file scope inline asm" blocks. These blocks are internally
1233 concatenated by LLVM and treated as a single unit, but may be separated in
1237 module asm "inline asm code goes here"
1238 module asm "more can go here"
1239 </pre>
1241 <p>The strings can contain any character by escaping non-printable characters.
1243 for the number.</p>
1245 <p>The inline asm code is simply printed to the machine code .s file when
1246 assembly code is generated.</p>
1248 </div>
1250 <!-- ======================================================================= -->
1253 </div>
1257 <p>A module may specify a target specific data layout string that specifies how
1258 data is to be laid out in memory. The syntax for the data layout is
1259 simply:</p>
1262 target datalayout = "<i>layout specification</i>"
1263 </pre>
1266 separated by the minus sign character ('-'). Each specification starts with
1267 a letter and may include other information after the letter to define some
1268 aspect of the data layout. The specifications accepted are as follows:</p>
1270 <dl>
1272 <dd>Specifies that the target lays out data in big-endian form. That is, the
1273 bits with the most significance have the lowest address location.</dd>
1276 <dd>Specifies that the target lays out data in little-endian form. That is,
1277 the bits with the least significance have the lowest address
1278 location.</dd>
1287 <dd>This specifies the alignment for an integer type of a given bit
1291 <dd>This specifies the alignment for a vector type of a given bit
1295 <dd>This specifies the alignment for a floating point type of a given bit
1299 targets.
1302 <dd>This specifies the alignment for an aggregate type of a given bit
1306 <dd>This specifies the alignment for a stack object of a given bit
1310 <dd>This specifies a set of native integer widths for the target CPU
1313 this set are considered to support most general arithmetic
1314 operations efficiently.</dd>
1315 </dl>
1317 <p>When constructing the data layout for a given target, LLVM starts with a
1318 default set of specifications which are then (possibly) overridden by the
1320 are given in this list:</p>
1322 <ul>
1337 </ul>
1339 <p>When LLVM is determining the alignment for a given type, it uses the
1340 following rules:</p>
1342 <ol>
1343 <li>If the type sought is an exact match for one of the specifications, that
1344 specification is used.</li>
1346 <li>If no match is found, and the type sought is an integer type, then the
1347 smallest integer type that is larger than the bitwidth of the sought type
1348 is used. If none of the specifications are larger than the bitwidth then
1349 the the largest integer type is used. For example, given the default
1350 specifications above, the i7 type will use the alignment of i8 (next
1351 largest) while both i65 and i256 will use the alignment of i64 (largest
1352 specified).</li>
1354 <li>If no match is found, and the type sought is a vector type, then the
1355 largest vector type that is smaller than the sought vector type will be
1358 </ol>
1360 </div>
1362 <!-- ======================================================================= -->
1365 </div>
1369 <p>Any memory access must be done through a pointer value associated
1370 with an address range of the memory access, otherwise the behavior
1371 is undefined. Pointer values are associated with address ranges
1372 according to the following rules:</p>
1374 <ul>
1375 <li>A pointer value is associated with the addresses associated with
1377 <li>An address of a global variable is associated with the address
1378 range of the variable's storage.</li>
1379 <li>The result value of an allocation instruction is associated with
1380 the address range of the allocated storage.</li>
1381 <li>A null pointer in the default address-space is associated with
1382 no address.</li>
1383 <li>An integer constant other than zero or a pointer value returned
1384 from a function not defined within LLVM may be associated with address
1385 ranges allocated through mechanisms other than those provided by
1386 LLVM. Such ranges shall not overlap with any ranges of addresses
1387 allocated by mechanisms provided by LLVM.</li>
1388 </ul>
1391 to the following rules:</p>
1393 <ul>
1394 <li>A pointer value formed from a
1397 <li>The result value of a
1400 <li>A pointer value formed by an
1402 pointer values that contribute (directly or indirectly) to the
1403 computation of the pointer's value.</li>
1405 </ul>
1409 slightly weaker.</p>
1411 <p>LLVM IR does not associate types with memory. The result type of a
1413 alignment of the memory from which to load, as well as the
1414 interpretation of the value. The first operand type of a
1416 and alignment of the store.</p>
1418 <p>Consequently, type-based alias analysis, aka TBAA, aka
1421 additional information which specialized optimization passes may use
1422 to implement type-based alias analysis.</p>
1424 </div>
1426 <!-- ======================================================================= -->
1429 </div>
1436 The optimizers must not change the number of volatile operations or change their
1437 order of execution relative to other volatile operations. The optimizers
1439 operations. This is not Java's "volatile" and has no cross-thread
1440 synchronization behavior.</p>
1442 </div>
1444 <!-- *********************************************************************** -->
1446 <!-- *********************************************************************** -->
1450 <p>The LLVM type system is one of the most important features of the
1451 intermediate representation. Being typed enables a number of optimizations
1452 to be performed on the intermediate representation directly, without having
1453 to do extra analyses on the side before the transformation. A strong type
1454 system makes it easier to read the generated code and enables novel analyses
1455 and transformations that are not feasible to perform on normal three address
1456 code representations.</p>
1458 </div>
1460 <!-- ======================================================================= -->
1462 Classifications</a> </div>
1469 <tbody>
1471 <tr>
1474 </tr>
1475 <tr>
1478 </tr>
1479 <tr>
1489 </td>
1490 </tr>
1491 <tr>
1498 </tr>
1499 <tr>
1508 </td>
1509 </tr>
1510 </tbody>
1511 </table>
1514 important. Values of these types are the only ones which can be produced by
1515 instructions.</p>
1517 </div>
1519 <!-- ======================================================================= -->
1524 <p>The primitive types are the fundamental building blocks of the LLVM
1525 system.</p>
1527 </div>
1529 <!-- _______________________________________________________________________ -->
1535 <p>The integer type is a very simple type that simply specifies an arbitrary
1536 bit width for the integer type desired. Any bit width from 1 bit to
1540 <pre>
1541 iN
1542 </pre>
1545 value.</p>
1552 </tr>
1556 </tr>
1560 </tr>
1561 </table>
1563 </div>
1565 <!-- _______________________________________________________________________ -->
1570 <table>
1571 <tbody>
1578 </tbody>
1579 </table>
1581 </div>
1583 <!-- _______________________________________________________________________ -->
1589 <p>The x86mmx type represents a value held in an MMX register on an x86 machine. The operations allowed on it are quite limited: parameters and return values, load and store, and bitcast. User-specified MMX instructions are represented as intrinsic or asm calls with arguments and/or results of this type. There are no arrays, vectors or constants of this type.</p>
1592 <pre>
1593 x86mmx
1594 </pre>
1596 </div>
1598 <!-- _______________________________________________________________________ -->
1607 <pre>
1608 void
1609 </pre>
1611 </div>
1613 <!-- _______________________________________________________________________ -->
1622 <pre>
1623 label
1624 </pre>
1626 </div>
1628 <!-- _______________________________________________________________________ -->
1634 <p>The metadata type represents embedded metadata. No derived types may be
1636 arguments.
1639 <pre>
1640 metadata
1641 </pre>
1643 </div>
1646 <!-- ======================================================================= -->
1651 <p>The real power in LLVM comes from the derived types in the system. This is
1652 what allows a programmer to represent arrays, functions, pointers, and other
1653 useful types. Each of these types contain one or more element types which
1654 may be a primitive type, or another derived type. For example, it is
1655 possible to have a two dimensional array, using an array as the element type
1656 of another array.</p>
1659 </div>
1661 <!-- _______________________________________________________________________ -->
1666 <p>Aggregate Types are a subset of derived types that can contain multiple
1669 aggregate types.</p>
1671 </div>
1673 <!-- _______________________________________________________________________ -->
1679 <p>The array type is a very simple derived type that arranges elements
1680 sequentially in memory. The array type requires a size (number of elements)
1681 and an underlying data type.</p>
1684 <pre>
1686 </pre>
1689 be any type with a size.</p>
1696 </tr>
1700 </tr>
1704 </tr>
1705 </table>
1711 </tr>
1715 </tr>
1719 </tr>
1720 </table>
1722 <p>There is no restriction on indexing beyond the end of the array implied by
1723 a static type (though there are restrictions on indexing beyond the bounds
1724 of an allocated object in some cases). This means that single-dimension
1725 'variable sized array' addressing can be implemented in LLVM with a zero
1726 length array type. An implementation of 'pascal style arrays' in LLVM could
1729 </div>
1731 <!-- _______________________________________________________________________ -->
1737 <p>The function type can be thought of as a function signature. It consists of
1738 a return type and a list of formal parameter types. The return type of a
1739 function type is a first class type or a void type.</p>
1742 <pre>
1744 </pre>
1748 which indicates that the function takes a variable number of arguments.
1749 Variable argument functions can access their arguments with
1759 </td>
1762 </tt></td>
1766 </td>
1772 LLVM.
1773 </td>
1778 </td>
1779 </tr>
1780 </table>
1782 </div>
1784 <!-- _______________________________________________________________________ -->
1790 <p>The structure type is used to represent a collection of data members together
1791 in memory. The packing of the field types is defined to match the ABI of the
1792 underlying processor. The elements of a structure may be any type that has a
1793 size.</p>
1798 Structures in registers are accessed using the
1802 <pre>
1804 </pre>
1817 </tr>
1818 </table>
1820 </div>
1822 <!-- _______________________________________________________________________ -->
1824 </div>
1829 <p>The packed structure type is used to represent a collection of data members
1830 together in memory. There is no padding between fields. Further, the
1831 alignment of a packed structure is 1 byte. The elements of a packed
1832 structure may be any type that has a size.</p>
1839 <pre>
1841 </pre>
1855 </tr>
1856 </table>
1858 </div>
1860 <!-- _______________________________________________________________________ -->
1866 <p>The pointer type is used to specify memory locations.
1867 Pointers are commonly used to reference objects in memory.</p>
1869 <p>Pointer types may have an optional address space attribute defining the
1870 numbered address space where the pointed-to object resides. The default
1871 address space is number zero. The semantics of non-zero address
1872 spaces are target-specific.</p>
1878 <pre>
1880 </pre>
1888 </tr>
1894 </tr>
1899 </tr>
1900 </table>
1902 </div>
1904 <!-- _______________________________________________________________________ -->
1910 <p>A vector type is a simple derived type that represents a vector of elements.
1911 Vector types are used when multiple primitive data are operated in parallel
1912 using a single instruction (SIMD). A vector type requires a size (number of
1913 elements) and an underlying primitive data type. Vector types are considered
1917 <pre>
1919 </pre>
1922 may be any integer or floating point type. Vectors of size zero are not
1923 allowed, and pointers are not allowed as the element type.</p>
1930 </tr>
1934 </tr>
1938 </tr>
1939 </table>
1941 </div>
1943 <!-- _______________________________________________________________________ -->
1948 <p>Opaque types are used to represent unknown types in the system. This
1949 corresponds (for example) to the C notion of a forward declared structure
1950 type. In LLVM, opaque types can eventually be resolved to any type (not just
1951 a structure type).</p>
1954 <pre>
1955 opaque
1956 </pre>
1963 </tr>
1964 </table>
1966 </div>
1968 <!-- ======================================================================= -->
1971 </div>
1977 requiring it to have a name. For instance, a structure declaration may
1978 contain a pointer to any of the types it is lexically a member of. Example
1979 of up references (with their equivalent as named type declarations)
1980 include:</p>
1982 <pre>
1983 { \2 * } %x = type { %x* }
1984 { \2 }* %y = type { %y }*
1985 \1* %z = type %z*
1986 </pre>
1988 <p>An up reference is needed by the asmprinter for printing out cyclic types
1989 when there is no declared name for a type in the cycle. Because the
1990 asmprinter does not want to print out an infinite type string, it needs a
1991 syntax to handle recursive types that have no names (all names are optional
1992 in llvm IR).</p>
1995 <pre>
1997 </pre>
2006 </tr>
2010 structure.</td>
2011 </tr>
2012 </table>
2014 </div>
2016 <!-- *********************************************************************** -->
2018 <!-- *********************************************************************** -->
2022 <p>LLVM has several different basic types of constants. This section describes
2023 them all and their syntax.</p>
2025 </div>
2027 <!-- ======================================================================= -->
2032 <dl>
2040 with integer types.</dd>
2044 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2045 notation (see below). The assembler requires the exact decimal value of a
2046 floating-point constant. For example, the assembler accepts 1.25 but
2053 </dl>
2055 <p>The one non-intuitive notation for constants is the hexadecimal form of
2056 floating point constants. For example, the form '<tt>double
2059 constants are required (and the only time that they are generated by the
2060 disassembler) is when a floating point constant must be emitted but it cannot
2061 be represented as a decimal floating point number in a reasonable number of
2062 digits. For example, NaN's, infinities, and other special values are
2063 represented in their IEEE hexadecimal format so that assembly and disassembly
2064 do not cause any bits to change in the constants.</p>
2066 <p>When using the hexadecimal form, constants of types float and double are
2067 represented using the 16-digit form shown above (which matches the IEEE754
2068 representation for double); float values must, however, be exactly
2069 representable as IEE754 single precision. Hexadecimal format is always used
2070 for long double, and there are three forms of long double. The 80-bit format
2072 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2075 currently supported target uses this format. Long doubles will only work if
2076 they match the long double format on your target. All hexadecimal formats
2077 are big-endian (sign bit at the left).</p>
2080 </div>
2082 <!-- ======================================================================= -->
2086 </div>
2090 <p>Complex constants are a (potentially recursive) combination of simple
2091 constants and smaller complex constants.</p>
2093 <dl>
2095 <dd>Structure constants are represented with notation similar to structure
2096 type definitions (a comma separated list of elements, surrounded by braces
2100 the number and types of elements must match those specified by the
2101 type.</dd>
2104 <dd>Array constants are represented with notation similar to array type
2105 definitions (a comma separated list of elements, surrounded by square
2108 the number and types of elements must match those specified by the
2109 type.</dd>
2112 <dd>Vector constants are represented with notation similar to vector type
2113 definitions (a comma separated list of elements, surrounded by
2115 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2117 elements must match those specified by the type.</dd>
2123 This is often used to avoid having to print large zero initializers
2124 (e.g. for large arrays) and is always exactly equivalent to using explicit
2125 zero initializers.</dd>
2128 <dd>A metadata node is a structure-like constant with
2131 be interpreted as part of the instruction stream, metadata is a place to
2132 attach additional information such as debug info.</dd>
2133 </dl>
2135 </div>
2137 <!-- ======================================================================= -->
2140 </div>
2146 (link-time) constants. These constants are explicitly referenced when
2149 legal LLVM file:</p>
2152 @X = global i32 17
2153 @Y = global i32 42
2154 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2155 </pre>
2157 </div>
2159 <!-- ======================================================================= -->
2164 indicates that the user of the value may receive an unspecified bit-pattern.
2168 <p>Undefined values are useful because they indicate to the compiler that the
2169 program is well defined no matter what value is used. This gives the
2170 compiler more freedom to optimize. Here are some examples of (potentially
2171 surprising) transformations that are valid (in pseudo IR):</p>
2175 %A = add %X, undef
2176 %B = sub %X, undef
2177 %C = xor %X, undef
2178 Safe:
2179 %A = undef
2180 %B = undef
2181 %C = undef
2182 </pre>
2184 <p>This is safe because all of the output bits are affected by the undef bits.
2185 Any output bit can have a zero or one depending on the input bits.</p>
2188 %A = or %X, undef
2189 %B = and %X, undef
2190 Safe:
2191 %A = -1
2192 %B = 0
2193 Unsafe:
2194 %A = undef
2195 %B = undef
2196 </pre>
2198 <p>These logical operations have bits that are not always affected by the input.
2209 %A = select undef, %X, %Y
2210 %B = select undef, 42, %Y
2211 %C = select %X, %Y, undef
2212 Safe:
2213 %A = %X (or %Y)
2214 %B = 42 (or %Y)
2215 %C = %Y
2216 Unsafe:
2217 %A = undef
2218 %B = undef
2219 %C = undef
2220 </pre>
2229 eliminated.</p>
2232 %A = xor undef, undef
2234 %B = undef
2235 %C = xor %B, %B
2237 %D = undef
2238 %E = icmp lt %D, 4
2239 %F = icmp gte %D, 4
2241 Safe:
2242 %A = undef
2243 %B = undef
2244 %C = undef
2245 %D = undef
2246 %E = undef
2247 %F = undef
2248 </pre>
2251 necessarily the same. This can be surprising to people (and also matches C
2252 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2255 its value over its "live range". This is true because the variable doesn't
2257 from arbitrary registers that happen to be around when needed, so the value
2259 need to have the same semantics or the core LLVM "replace all uses with"
2260 concept would not hold.</p>
2263 %A = fdiv undef, %X
2264 %B = fdiv %X, undef
2265 Safe:
2266 %A = undef
2267 b: unreachable
2268 </pre>
2275 defined on SNaN's. However, in the second example, we can make a more
2277 arbitrary value, we are allowed to assume that it could be zero. Since a
2279 the operation does not execute at all. This allows us to delete the divide and
2280 all code after it. Because the undefined operation "can't happen", the
2281 optimizer can assume that it occurs in dead code.</p>
2284 a: store undef -> %X
2285 b: store %X -> undef
2286 Safe:
2288 b: unreachable
2289 </pre>
2292 undefined value can be assumed to not have any effect; we can assume that the
2293 value is overwritten with bits that happen to match what was already there.
2295 memory, therefore, it has undefined behavior.</p>
2297 </div>
2299 <!-- ======================================================================= -->
2304 instead of representing an unspecified bit pattern, they represent the
2305 fact that an instruction or constant expression which cannot evoke side
2306 effects has nevertheless detected a condition which results in undefined
2307 behavior.</p>
2309 <p>There is currently no way of representing a trap value in the IR; they
2310 only exist when produced by operations such as
2315 <ul>
2317 their operands.</li>
2320 to their dynamic predecessor basic block.</li>
2322 <li>Function arguments depend on the corresponding actual argument values in
2323 the dynamic callers of their functions.</li>
2327 control back to them.</li>
2331 or exception-throwing call instructions that dynamically transfer control
2332 back to them.</li>
2334 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2335 referenced memory addresses, following the order in the IR
2336 (including loads and stores implied by intrinsics such as
2339 <!-- TODO: In the case of multiple threads, this only applies if the store
2340 "happens-before" the load or store. -->
2342 <!-- TODO: floating-point exception state -->
2344 <li>An instruction with externally visible side effects depends on the most
2345 recent preceding instruction with externally visible side effects, following
2346 the order in the IR. (This includes
2351 if the terminator instruction has multiple successors and the instruction
2352 is always executed when control transfers to one of the successors, and
2353 may not be executed when control is transfered to another.</li>
2357 </ul>
2359 <p>Whenever a trap value is generated, all values which depend on it evaluate
2360 to trap. If they have side effects, the evoke their side effects as if each
2361 operand with a trap value were undef. If they have externally-visible side
2362 effects, the behavior is undefined.</p>
2367 entry:
2370 %trap_yet_again = getelementptr i32* @h, i32 %still_trap
2371 store i32 0, i32* %trap_yet_again ; undefined behavior
2373 store i32 %trap, i32* @g ; Trap value conceptually stored to memory.
2374 %trap2 = load i32* @g ; Returns a trap value, not just undef.
2376 volatile store i32 %trap, i32* @g ; External observation; undefined behavior.
2378 %narrowaddr = bitcast i32* @g to i16*
2379 %wideaddr = bitcast i32* @g to i64*
2380 %trap3 = load 16* %narrowaddr ; Returns a trap value.
2381 %trap4 = load i64* %widaddr ; Returns a trap value.
2383 %cmp = icmp i32 slt %trap, 0 ; Returns a trap value.
2384 %br i1 %cmp, %true, %end ; Branch to either destination.
2386 true:
2387 volatile store i32 0, i32* @g ; This is control-dependent on %cmp, so
2388 ; it has undefined behavior.
2389 br label %end
2391 end:
2393 ; Both edges into this PHI are
2394 ; control-dependent on %cmp, so this
2395 ; always results in a trap value.
2397 volatile store i32 0, i32* @g ; %end is control-equivalent to %entry
2398 ; so this is defined (ignoring earlier
2399 ; undefined behavior in this example).
2400 </pre>
2402 </div>
2404 <!-- ======================================================================= -->
2406 Blocks</a></div>
2412 basic block in the specified function, and always has an i8* type. Taking
2413 the address of the entry block is illegal.</p>
2415 <p>This value only has defined behavior when used as an operand to the
2417 comparisons against null. Pointer equality tests between labels addresses
2418 results in undefined behavior — though, again, comparison against null
2419 is ok, and no label is equal to the null pointer. This may be passed around
2420 as an opaque pointer sized value as long as the bits are not inspected. This
2423 instruction.</p>
2425 <p>Finally, some targets may provide defined semantics when using the value as
2426 the operand to an inline assembly, but that is target specific.</p>
2428 </div>
2431 <!-- ======================================================================= -->
2433 </div>
2437 <p>Constant expressions are used to allow expressions involving other constants
2438 to be used as constants. Constant expressions may be of
2440 operation that does not have side effects (e.g. load and call are not
2441 supported). The following is the syntax for constant expressions:</p>
2443 <dl>
2445 <dd>Truncate a constant to another type. The bit size of CST must be larger
2446 than the bit size of TYPE. Both types must be integers.</dd>
2449 <dd>Zero extend a constant to another type. The bit size of CST must be
2450 smaller than the bit size of TYPE. Both types must be integers.</dd>
2453 <dd>Sign extend a constant to another type. The bit size of CST must be
2454 smaller than the bit size of TYPE. Both types must be integers.</dd>
2457 <dd>Truncate a floating point constant to another floating point type. The
2458 size of CST must be larger than the size of TYPE. Both types must be
2459 floating point.</dd>
2462 <dd>Floating point extend a constant to another type. The size of CST must be
2463 smaller or equal to the size of TYPE. Both types must be floating
2464 point.</dd>
2467 <dd>Convert a floating point constant to the corresponding unsigned integer
2468 constant. TYPE must be a scalar or vector integer type. CST must be of
2469 scalar or vector floating point type. Both CST and TYPE must be scalars,
2470 or vectors of the same number of elements. If the value won't fit in the
2471 integer type, the results are undefined.</dd>
2474 <dd>Convert a floating point constant to the corresponding signed integer
2475 constant. TYPE must be a scalar or vector integer type. CST must be of
2476 scalar or vector floating point type. Both CST and TYPE must be scalars,
2477 or vectors of the same number of elements. If the value won't fit in the
2478 integer type, the results are undefined.</dd>
2481 <dd>Convert an unsigned integer constant to the corresponding floating point
2482 constant. TYPE must be a scalar or vector floating point type. CST must be
2483 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2484 vectors of the same number of elements. If the value won't fit in the
2485 floating point type, the results are undefined.</dd>
2488 <dd>Convert a signed integer constant to the corresponding floating point
2489 constant. TYPE must be a scalar or vector floating point type. CST must be
2490 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2491 vectors of the same number of elements. If the value won't fit in the
2492 floating point type, the results are undefined.</dd>
2495 <dd>Convert a pointer typed constant to the corresponding integer constant
2501 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2502 type. CST must be of integer type. The CST value is zero extended,
2503 truncated, or unchanged to make it fit in a pointer size. This one is
2507 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2515 instruction, the index list may have zero or more indexes, which are
2529 constants.</dd>
2533 constants.</dd>
2537 constants.</dd>
2541 constants. The index list is interpreted in a similar manner as indices in
2543 index value must be specified.</dd>
2547 constants. The index list is interpreted in a similar manner as indices in
2549 index value must be specified.</dd>
2552 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2555 on operands are the same as those for the corresponding instruction
2556 (e.g. no bitwise operations on floating point values are allowed).</dd>
2557 </dl>
2559 </div>
2561 <!-- *********************************************************************** -->
2563 <!-- *********************************************************************** -->
2565 <!-- ======================================================================= -->
2568 </div>
2572 <p>LLVM supports inline assembler expressions (as opposed
2574 a special value. This value represents the inline assembler as a string
2575 (containing the instructions to emit), a list of operand constraints (stored
2576 as a string), a flag that indicates whether or not the inline asm
2577 expression has side effects, and a flag indicating whether the function
2578 containing the asm needs to align its stack conservatively. An example
2579 inline assembler expression is:</p>
2583 </pre>
2587 have:</p>
2591 </pre>
2593 <p>Inline asms with side effects not visible in the constraint list must be
2594 marked as having side effects. This is done through the use of the
2599 </pre>
2601 <p>In some cases inline asms will contain code that will not work unless the
2602 stack is aligned in some way, such as calls or SSE instructions on x86,
2603 yet will not contain code that does that alignment within the asm.
2604 The compiler should make conservative assumptions about what the asm might
2605 contain and should generate its usual stack alignment code in the prologue
2610 </pre>
2613 first.</p>
2615 <p>TODO: The format of the asm and constraints string still need to be
2616 documented here. Constraints on what can be done (e.g. duplication, moving,
2617 etc need to be documented). This is probably best done by reference to
2618 another document that covers inline asm from a holistic perspective.</p>
2619 </div>
2623 </div>
2628 attached to it that contains a constant integer. If present, the code
2629 generator will use the integer as the location cookie value when report
2630 errors through the LLVMContext error reporting mechanisms. This allows a
2631 front-end to correlate backend errors that occur with inline asm back to the
2632 source code that produced it. For example:</p>
2636 ...
2638 </pre>
2640 <p>It is up to the front-end to make sense of the magic numbers it places in the
2641 IR.</p>
2643 </div>
2645 <!-- ======================================================================= -->
2647 Strings</a>
2648 </div>
2652 <p>LLVM IR allows metadata to be attached to instructions in the program that
2653 can convey extra information about the code to the optimizers and code
2654 generator. One example application of metadata is source-level debug
2655 information. There are two metadata primitives: strings and nodes. All
2659 <p>A metadata string is a string surrounded by double quotes. It can contain
2663 <p>Metadata nodes are represented with notation similar to structure constants
2664 (a comma separated list of elements, surrounded by braces and preceded by an
2669 metadata nodes, which can be looked up in the module symbol table. For
2670 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2673 function is using two metadata arguments.</p>
2677 </pre>
2684 </pre>
2685 </div>
2688 <!-- *********************************************************************** -->
2691 </div>
2692 <!-- *********************************************************************** -->
2695 code generation or other IR semantics. These are documented here. All globals
2696 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2697 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2698 by LLVM.</p>
2700 <!-- ======================================================================= -->
2703 </div>
2709 pointers to global variables and functions which may optionally have a pointer
2710 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2712 <pre>
2713 @X = global i8 4
2714 @Y = global i32 123
2716 @llvm.used = appending global [2 x i8*] [
2717 i8* @X,
2718 i8* bitcast (i32* @Y to i8*)
2719 ], section "llvm.metadata"
2720 </pre>
2723 compiler, assembler, and linker are required to treat the symbol as if there is
2724 a reference to the global that it cannot see. For example, if a variable has
2726 list, it cannot be deleted. This is commonly used to represent references from
2727 inline asms and other things the compiler cannot "see", and corresponds to
2730 <p>On some targets, the code generator must emit a directive to the assembler or
2731 object file to prevent the assembler and linker from molesting the symbol.</p>
2733 </div>
2735 <!-- ======================================================================= -->
2738 </div>
2744 touching the symbol. On targets that support it, this allows an intelligent
2745 linker to optimize references to the symbol without being impeded as it would be
2748 <p>This is a rare construct that should only be used in rare circumstances, and
2749 should not be exposed to source languages.</p>
2751 </div>
2753 <!-- ======================================================================= -->
2756 </div>
2759 <pre>
2760 %0 = type { i32, void ()* }
2762 </pre>
2763 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor functions and associated priorities. The functions referenced by this array will be called in ascending order of priority (i.e. lowest first) when the module is loaded. The order of functions with the same priority is not defined.
2764 </p>
2766 </div>
2768 <!-- ======================================================================= -->
2771 </div>
2774 <pre>
2775 %0 = type { i32, void ()* }
2777 </pre>
2779 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions and associated priorities. The functions referenced by this array will be called in descending order of priority (i.e. highest first) when the module is loaded. The order of functions with the same priority is not defined.
2780 </p>
2782 </div>
2785 <!-- *********************************************************************** -->
2787 <!-- *********************************************************************** -->
2791 <p>The LLVM instruction set consists of several different classifications of
2798 </div>
2800 <!-- ======================================================================= -->
2802 Instructions</a> </div>
2807 in a program ends with a "Terminator" instruction, which indicates which
2808 block should be executed after the current block is finished. These
2810 control flow, not values (the one exception being the
2813 <p>There are seven different terminator instructions: the
2822 </div>
2824 <!-- _______________________________________________________________________ -->
2826 Instruction</a> </div>
2831 <pre>
2834 </pre>
2838 a value) from a function back to the caller.</p>
2841 value and then causes control flow, and one that just causes control flow to
2842 occur.</p>
2846 return value. The type of the return value must be a
2851 value or a return value with a type that does not match its type, or if it
2853 return value.</p>
2857 the calling function's context. If the caller is a
2859 instruction after the call. If the caller was an
2861 the beginning of the "normal" destination block. If the instruction returns
2862 a value, that value shall set the call or invoke instruction's return
2863 value.</p>
2866 <pre>
2870 </pre>
2872 </div>
2873 <!-- _______________________________________________________________________ -->
2879 <pre>
2880 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2881 </pre>
2885 different basic block in the current function. There are two forms of this
2886 instruction, corresponding to a conditional branch and an unconditional
2887 branch.</p>
2893 target.</p>
2902 <pre>
2903 Test:
2905 br i1 %cond, label %IfEqual, label %IfUnequal
2906 IfEqual:
2908 IfUnequal:
2910 </pre>
2912 </div>
2914 <!-- _______________________________________________________________________ -->
2917 </div>
2922 <pre>
2923 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2924 </pre>
2929 instruction, allowing a branch to occur to one of many possible
2930 destinations.</p>
2936 The table is not allowed to contain duplicate constant entries.</p>
2941 is searched for the given value. If the value is found, control flow is
2942 transferred to the corresponding destination; otherwise, control flow is
2943 transferred to the default destination.</p>
2946 <p>Depending on properties of the target machine and the particular
2948 different ways. For example, it could be generated as a series of chained
2949 conditional branches or with a lookup table.</p>
2952 <pre>
2955 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2958 switch i32 0, label %dest [ ]
2961 switch i32 %val, label %otherwise [ i32 0, label %onzero
2962 i32 1, label %onone
2963 i32 2, label %ontwo ]
2964 </pre>
2966 </div>
2969 <!-- _______________________________________________________________________ -->
2972 </div>
2977 <pre>
2979 </pre>
2984 within the current function, whose address is specified by
2991 rest of the arguments indicate the full set of possible destinations that the
2992 address may point to. Blocks are allowed to occur multiple times in the
2993 destination list, though this isn't particularly useful.</p>
2995 <p>This destination list is required so that dataflow analysis has an accurate
2996 understanding of the CFG.</p>
3000 <p>Control transfers to the block specified in the address argument. All
3001 possible destination blocks must be listed in the label list, otherwise this
3002 instruction has undefined behavior. This implies that jumps to labels
3003 defined in other functions have undefined behavior as well.</p>
3010 <pre>
3011 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3012 </pre>
3014 </div>
3017 <!-- _______________________________________________________________________ -->
3020 </div>
3025 <pre>
3026 <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>]
3028 </pre>
3032 function, with the possibility of control flow transfer to either the
3035 control flow will return to the "normal" label. If the callee (or any
3037 instruction, control is interrupted and continued at the dynamically nearest
3043 <ol>
3045 convention</a> the call should use. If none is specified, the call
3046 defaults to using C calling conventions.</li>
3053 function value being invoked. In most cases, this is a direct function
3055 off an arbitrary pointer to function value.</li>
3058 function to be invoked. </li>
3061 signature argument types and parameter attributes. All arguments must be
3063 signature indicates the function accepts a variable number of arguments,
3064 the extra arguments can be specified.</li>
3075 </ol>
3078 <p>This instruction is designed to operate as a standard
3080 primary difference is that it establishes an association with a label, which
3081 is used by the runtime library to unwind the stack.</p>
3083 <p>This instruction is used in languages with destructors to ensure that proper
3085 exception. Additionally, this is important for implementation of
3088 <p>For the purposes of the SSA form, the definition of the value returned by the
3090 block to the "normal" label. If the callee unwinds then no return value is
3091 available.</p>
3093 <p>Note that the code generator does not yet completely support unwind, and
3094 that the invoke/unwind semantics are likely to change in future versions.</p>
3097 <pre>
3098 %retval = invoke i32 @Test(i32 15) to label %Continue
3102 </pre>
3104 </div>
3106 <!-- _______________________________________________________________________ -->
3109 Instruction</a> </div>
3114 <pre>
3115 unwind
3116 </pre>
3120 at the first callee in the dynamic call stack which used
3122 This is primarily used to implement exception handling.</p>
3126 immediately halt. The dynamic call stack is then searched for the
3128 Once found, execution continues at the "exceptional" destination block
3130 instruction in the dynamic call chain, undefined behavior results.</p>
3132 <p>Note that the code generator does not yet completely support unwind, and
3133 that the invoke/unwind semantics are likely to change in future versions.</p>
3135 </div>
3137 <!-- _______________________________________________________________________ -->
3140 Instruction</a> </div>
3145 <pre>
3146 unreachable
3147 </pre>
3151 instruction is used to inform the optimizer that a particular portion of the
3152 code is not reachable. This can be used to indicate that the code after a
3153 no-return function cannot be reached, and other facts.</p>
3158 </div>
3160 <!-- ======================================================================= -->
3165 <p>Binary operators are used to do most of the computation in a program. They
3166 require two operands of the same type, execute an operation on them, and
3167 produce a single value. The operands might represent multiple data, as is
3169 has the same type as its operands.</p>
3173 </div>
3175 <!-- _______________________________________________________________________ -->
3178 </div>
3183 <pre>
3188 </pre>
3196 integer values. Both arguments must have identical types.</p>
3201 <p>If the sum has unsigned overflow, the result returned is the mathematical
3204 <p>Because LLVM integers use a two's complement representation, this instruction
3205 is appropriate for both signed and unsigned integers.</p>
3211 respectively, occurs.</p>
3214 <pre>
3216 </pre>
3218 </div>
3220 <!-- _______________________________________________________________________ -->
3223 </div>
3228 <pre>
3230 </pre>
3238 floating point values. Both arguments must have identical types.</p>
3244 <pre>
3246 </pre>
3248 </div>
3250 <!-- _______________________________________________________________________ -->
3253 </div>
3258 <pre>
3263 </pre>
3267 operands.</p>
3271 representations.</p>
3276 integer values. Both arguments must have identical types.</p>
3281 <p>If the difference has unsigned overflow, the result returned is the
3283 result.</p>
3285 <p>Because LLVM integers use a two's complement representation, this instruction
3286 is appropriate for both signed and unsigned integers.</p>
3292 respectively, occurs.</p>
3295 <pre>
3298 </pre>
3300 </div>
3302 <!-- _______________________________________________________________________ -->
3305 </div>
3310 <pre>
3312 </pre>
3316 operands.</p>
3320 representations.</p>
3325 floating point values. Both arguments must have identical types.</p>
3331 <pre>
3334 </pre>
3336 </div>
3338 <!-- _______________________________________________________________________ -->
3341 </div>
3346 <pre>
3351 </pre>
3359 integer values. Both arguments must have identical types.</p>
3364 <p>If the result of the multiplication has unsigned overflow, the result
3366 width of the result.</p>
3368 <p>Because LLVM integers use a two's complement representation, and the result
3369 is the same width as the operands, this instruction returns the correct
3370 result for both signed and unsigned integers. If a full product
3372 be sign-extended or zero-extended as appropriate to the width of the full
3373 product.</p>
3379 respectively, occurs.</p>
3382 <pre>
3384 </pre>
3386 </div>
3388 <!-- _______________________________________________________________________ -->
3391 </div>
3396 <pre>
3398 </pre>
3406 floating point values. Both arguments must have identical types.</p>
3412 <pre>
3414 </pre>
3416 </div>
3418 <!-- _______________________________________________________________________ -->
3420 </a></div>
3425 <pre>
3427 </pre>
3435 values. Both arguments must have identical types.</p>
3440 <p>Note that unsigned integer division and signed integer division are distinct
3446 <pre>
3448 </pre>
3450 </div>
3452 <!-- _______________________________________________________________________ -->
3454 </a> </div>
3459 <pre>
3462 </pre>
3470 values. Both arguments must have identical types.</p>
3473 <p>The value produced is the signed integer quotient of the two operands rounded
3474 towards zero.</p>
3476 <p>Note that signed integer division and unsigned integer division are distinct
3479 <p>Division by zero leads to undefined behavior. Overflow also leads to
3480 undefined behavior; this is a rare case, but can occur, for example, by doing
3485 be rounded.</p>
3488 <pre>
3490 </pre>
3492 </div>
3494 <!-- _______________________________________________________________________ -->
3496 Instruction</a> </div>
3501 <pre>
3503 </pre>
3511 floating point values. Both arguments must have identical types.</p>
3517 <pre>
3519 </pre>
3521 </div>
3523 <!-- _______________________________________________________________________ -->
3525 </div>
3530 <pre>
3532 </pre>
3536 division of its two arguments.</p>
3541 values. Both arguments must have identical types.</p>
3545 This instruction always performs an unsigned division to get the
3546 remainder.</p>
3548 <p>Note that unsigned integer remainder and signed integer remainder are
3554 <pre>
3556 </pre>
3558 </div>
3560 <!-- _______________________________________________________________________ -->
3563 </div>
3568 <pre>
3570 </pre>
3574 division of its two operands. This instruction can also take
3576 elements must be integers.</p>
3581 values. Both arguments must have identical types.</p>
3587 a value. For more information about the difference,
3589 Math Forum</a>. For a table of how this is implemented in various languages,
3593 <p>Note that signed integer remainder and unsigned integer remainder are
3596 <p>Taking the remainder of a division by zero leads to undefined behavior.
3597 Overflow also leads to undefined behavior; this is a rare case, but can
3598 occur, for example, by taking the remainder of a 32-bit division of
3600 lets srem be implemented using instructions that return both the result of
3601 the division and the remainder.)</p>
3604 <pre>
3606 </pre>
3608 </div>
3610 <!-- _______________________________________________________________________ -->
3617 <pre>
3619 </pre>
3623 its two operands.</p>
3628 floating point values. Both arguments must have identical types.</p>
3632 has the same sign as the dividend.</p>
3635 <pre>
3637 </pre>
3639 </div>
3641 <!-- ======================================================================= -->
3643 Operations</a> </div>
3647 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3648 program. They are generally very efficient instructions and can commonly be
3649 strength reduced from other instructions. They require two operands of the
3650 same type, execute an operation on them, and produce a single value. The
3651 resulting value is the same type as its operands.</p>
3653 </div>
3655 <!-- _______________________________________________________________________ -->
3657 Instruction</a> </div>
3662 <pre>
3664 </pre>
3668 a specified number of bits.</p>
3678 is (statically or dynamically) negative or equal to or larger than the number
3684 <pre>
3689 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3690 </pre>
3692 </div>
3694 <!-- _______________________________________________________________________ -->
3696 Instruction</a> </div>
3701 <pre>
3703 </pre>
3707 operand shifted to the right a specified number of bits with zero fill.</p>
3715 <p>This instruction always performs a logical shift right operation. The most
3716 significant bits of the result will be filled with zero bits after the shift.
3723 <pre>
3729 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3730 </pre>
3732 </div>
3734 <!-- _______________________________________________________________________ -->
3736 Instruction</a> </div>
3740 <pre>
3742 </pre>
3746 operand shifted to the right a specified number of bits with sign
3747 extension.</p>
3755 <p>This instruction always performs an arithmetic shift right operation, The
3756 most significant bits of the result will be filled with the sign bit
3763 <pre>
3769 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3770 </pre>
3772 </div>
3774 <!-- _______________________________________________________________________ -->
3776 Instruction</a> </div>
3781 <pre>
3783 </pre>
3787 operands.</p>
3792 values. Both arguments must have identical types.</p>
3798 <tbody>
3799 <tr>
3803 </tr>
3804 <tr>
3808 </tr>
3809 <tr>
3813 </tr>
3814 <tr>
3818 </tr>
3819 <tr>
3823 </tr>
3824 </tbody>
3825 </table>
3828 <pre>
3832 </pre>
3833 </div>
3834 <!-- _______________________________________________________________________ -->
3840 <pre>
3842 </pre>
3846 two operands.</p>
3851 values. Both arguments must have identical types.</p>
3857 <tbody>
3858 <tr>
3862 </tr>
3863 <tr>
3867 </tr>
3868 <tr>
3872 </tr>
3873 <tr>
3877 </tr>
3878 <tr>
3882 </tr>
3883 </tbody>
3884 </table>
3887 <pre>
3891 </pre>
3893 </div>
3895 <!-- _______________________________________________________________________ -->
3897 Instruction</a> </div>
3902 <pre>
3904 </pre>
3914 values. Both arguments must have identical types.</p>
3920 <tbody>
3921 <tr>
3925 </tr>
3926 <tr>
3930 </tr>
3931 <tr>
3935 </tr>
3936 <tr>
3940 </tr>
3941 <tr>
3945 </tr>
3946 </tbody>
3947 </table>
3950 <pre>
3955 </pre>
3957 </div>
3959 <!-- ======================================================================= -->
3962 </div>
3966 <p>LLVM supports several instructions to represent vector operations in a
3967 target-independent manner. These instructions cover the element-access and
3968 vector-specific operations needed to process vectors effectively. While LLVM
3969 does directly support these vector operations, many sophisticated algorithms
3970 will want to use target-specific intrinsics to take full advantage of a
3971 specific target.</p>
3973 </div>
3975 <!-- _______________________________________________________________________ -->
3978 </div>
3983 <pre>
3984 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3985 </pre>
3989 from a vector at a specified index.</p>
3995 indicating the position from which to extract the element. The index may be
3996 a variable.</p>
3999 <p>The result is a scalar of the same type as the element type of
4002 results are undefined.</p>
4005 <pre>
4007 </pre>
4009 </div>
4011 <!-- _______________________________________________________________________ -->
4014 </div>
4019 <pre>
4020 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4021 </pre>
4025 vector at a specified index.</p>
4030 whose type must equal the element type of the first operand. The third
4031 operand is an index indicating the position at which to insert the value.
4032 The index may be a variable.</p>
4038 results are undefined.</p>
4041 <pre>
4042 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4043 </pre>
4045 </div>
4047 <!-- _______________________________________________________________________ -->
4050 </div>
4055 <pre>
4056 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4057 </pre>
4061 from two input vectors, returning a vector with the same element type as the
4062 input and length that is the same as the shuffle mask.</p>
4066 with types that match each other. The third argument is a shuffle mask whose
4067 element type is always 'i32'. The result of the instruction is a vector
4068 whose length is the same as the shuffle mask and whose element type is the
4069 same as the element type of the first two operands.</p>
4071 <p>The shuffle mask operand is required to be a constant vector with either
4072 constant integer or undef values.</p>
4075 <p>The elements of the two input vectors are numbered from left to right across
4076 both of the vectors. The shuffle mask operand specifies, for each element of
4077 the result vector, which element of the two input vectors the result element
4078 gets. The element selector may be undef (meaning "don't care") and the
4079 second operand may be undef if performing a shuffle from only one vector.</p>
4082 <pre>
4086 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4090 <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>
4091 </pre>
4093 </div>
4095 <!-- ======================================================================= -->
4098 </div>
4102 <p>LLVM supports several instructions for working with
4105 </div>
4107 <!-- _______________________________________________________________________ -->
4110 </div>
4115 <pre>
4117 </pre>
4127 specify which value to extract in a similar manner as indices in a
4131 <p>The result is the value at the position in the aggregate specified by the
4132 index operands.</p>
4135 <pre>
4137 </pre>
4139 </div>
4141 <!-- _______________________________________________________________________ -->
4144 </div>
4149 <pre>
4150 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx> <i>; yields <aggregate type></i>
4151 </pre>
4161 value to insert. The following operands are constant indices indicating
4162 the position at which to insert the value in a similar manner as indices in a
4164 value to insert must have the same type as the value identified by the
4165 indices.</p>
4173 <pre>
4176 </pre>
4178 </div>
4181 <!-- ======================================================================= -->
4184 </div>
4188 <p>A key design point of an SSA-based representation is how it represents
4189 memory. In LLVM, no memory locations are in SSA form, which makes things
4190 very simple. This section describes how to read, write, and allocate
4191 memory in LLVM.</p>
4193 </div>
4195 <!-- _______________________________________________________________________ -->
4198 </div>
4203 <pre>
4204 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4205 </pre>
4209 currently executing function, to be automatically released when this function
4210 returns to its caller. The object is always allocated in the generic address
4211 space (address space zero).</p>
4216 runtime stack, returning a pointer of the appropriate type to the program.
4217 If "NumElements" is specified, it is the number of elements allocated,
4218 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4219 specified, the value result of the allocation is guaranteed to be aligned to
4220 at least that boundary. If not specified, or if zero, the target can choose
4221 to align the allocation on any convenient boundary compatible with the
4222 type.</p>
4227 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4229 memory is automatically released when the function returns. The
4231 variables that must have an address available. When the function returns
4234 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4237 <pre>
4242 </pre>
4244 </div>
4246 <!-- _______________________________________________________________________ -->
4248 Instruction</a> </div>
4253 <pre>
4254 <result> = load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4255 <result> = volatile load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4257 </pre>
4264 from which to load. The pointer must point to
4271 operation (that is, the alignment of the memory address). A value of 0 or an
4273 alignment for the target. It is the responsibility of the code emitter to
4274 ensure that the alignment information is correct. Overestimating the
4275 alignment results in undefined behavior. Underestimating the alignment may
4282 and code generator that this load is not expected to be reused in the cache.
4283 The code generator may select special instructions to save cache bandwidth,
4287 <p>The location of memory pointed to is loaded. If the value being loaded is of
4288 scalar type then the number of bytes read does not exceed the minimum number
4289 of bytes needed to hold all bits of the type. For example, loading an
4292 is undefined if the value was not originally written using a store of the
4293 same type.</p>
4296 <pre>
4300 </pre>
4302 </div>
4304 <!-- _______________________________________________________________________ -->
4306 Instruction</a> </div>
4311 <pre>
4312 store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4313 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4314 </pre>
4321 and an address at which to store it. The type of the
4330 operation (that is, the alignment of the memory address). A value of 0 or an
4331 omitted "align" argument means that the operation has the preferential
4332 alignment for the target. It is the responsibility of the code emitter to
4333 ensure that the alignment information is correct. Overestimating the
4334 alignment results in an undefined behavior. Underestimating the alignment may
4337 <p>The optional !nontemporal metadata must reference a single metatadata
4339 value 1. The existence of the !nontemporal metatadata on the
4340 instruction tells the optimizer and code generator that this load is
4341 not expected to be reused in the cache. The code generator may
4342 select special instructions to save cache bandwidth, such as the
4343 MOVNT instruction on x86.</p>
4350 does not exceed the minimum number of bytes needed to hold all bits of the
4353 integral number of bytes, it is unspecified what happens to the extra bits
4354 that do not belong to the type, but they will typically be overwritten.</p>
4357 <pre>
4361 </pre>
4363 </div>
4365 <!-- _______________________________________________________________________ -->
4368 </div>
4373 <pre>
4376 </pre>
4381 It performs address calculation only and does not access memory.</p>
4384 <p>The first argument is always a pointer, and forms the basis of the
4385 calculation. The remaining arguments are indices that indicate which of the
4386 elements of the aggregate object are indexed. The interpretation of each
4387 index is dependent on the type being indexed into. The first index always
4388 indexes the pointer value given as the first argument, the second index
4389 indexes a value of the type pointed to (not necessarily the value directly
4390 pointed to, since the first index can be non-zero), etc. The first type
4391 indexed into must be a pointer value, subsequent types can be arrays,
4392 vectors, and structs. Note that subsequent types being indexed into
4393 can never be pointers, since that would require loading the pointer before
4394 continuing calculation.</p>
4396 <p>The type of each index argument depends on the type it is indexing into.
4399 or vector, integers of any width are allowed, and they are not required to be
4400 constant.</p>
4402 <p>For example, let's consider a C code fragment and how it gets compiled to
4403 LLVM:</p>
4406 struct RT {
4407 char A;
4409 char C;
4410 };
4411 struct ST {
4412 int X;
4413 double Y;
4414 struct RT Z;
4415 };
4417 int *foo(struct ST *s) {
4419 }
4420 </pre>
4428 define i32* @foo(%ST* %s) {
4429 entry:
4431 ret i32* %reg
4432 }
4433 </pre>
4438 }</tt>' type, a structure. The second index indexes into the third element
4440 i8 }</tt>' type, another structure. The third index indexes into the second
4442 array. The two dimensions of the array are subscripted into, yielding an
4446 <p>Note that it is perfectly legal to index partially through a structure,
4447 returning a pointer to an inner element. Because of this, the LLVM code for
4448 the given testcase is equivalent to:</p>
4450 <pre>
4451 define i32* @foo(%ST* %s) {
4457 ret i32* %t5
4458 }
4459 </pre>
4464 or if any of the addresses that would be formed by successive addition of
4465 the offsets implied by the indices to the base address with infinitely
4468 the addresses that point into the object, plus the address one byte past
4469 the end.</p>
4472 the base address with silently-wrapping two's complement arithmetic, and
4474 pointed to by the base pointer. The result value may not necessarily be
4475 used to access memory though, even if it happens to point into allocated
4477 section for more information.</p>
4479 <p>The getelementptr instruction is often confusing. For some more insight into
4483 <pre>
4492 </pre>
4494 </div>
4496 <!-- ======================================================================= -->
4498 </div>
4502 <p>The instructions in this category are the conversion instructions (casting)
4503 which all take a single operand and a type. They perform various bit
4504 conversions on the operand.</p>
4506 </div>
4508 <!-- _______________________________________________________________________ -->
4511 </div>
4515 <pre>
4517 </pre>
4526 size and type of the result, which must be
4529 allowed.</p>
4538 <pre>
4542 </pre>
4544 </div>
4546 <!-- _______________________________________________________________________ -->
4549 </div>
4553 <pre>
4555 </pre>
4576 <pre>
4579 </pre>
4581 </div>
4583 <!-- _______________________________________________________________________ -->
4586 </div>
4590 <pre>
4592 </pre>
4612 <pre>
4615 </pre>
4617 </div>
4619 <!-- _______________________________________________________________________ -->
4622 </div>
4627 <pre>
4629 </pre>
4647 undefined.</p>
4650 <pre>
4653 </pre>
4655 </div>
4657 <!-- _______________________________________________________________________ -->
4660 </div>
4664 <pre>
4666 </pre>
4670 floating point value.</p>
4676 type must be smaller than the destination type.</p>
4686 <pre>
4689 </pre>
4691 </div>
4693 <!-- _______________________________________________________________________ -->
4696 </div>
4700 <pre>
4702 </pre>
4718 towards zero) unsigned integer value. If the value cannot fit
4722 <pre>
4726 </pre>
4728 </div>
4730 <!-- _______________________________________________________________________ -->
4733 </div>
4737 <pre>
4739 </pre>
4757 the results are undefined.</p>
4760 <pre>
4764 </pre>
4766 </div>
4768 <!-- _______________________________________________________________________ -->
4771 </div>
4775 <pre>
4777 </pre>
4792 integer quantity and converts it to the corresponding floating point
4793 value. If the value cannot fit in the floating point value, the results are
4794 undefined.</p>
4797 <pre>
4800 </pre>
4802 </div>
4804 <!-- _______________________________________________________________________ -->
4807 </div>
4811 <pre>
4813 </pre>
4828 quantity and converts it to the corresponding floating point value. If the
4829 value cannot fit in the floating point value, the results are undefined.</p>
4832 <pre>
4835 </pre>
4837 </div>
4839 <!-- _______________________________________________________________________ -->
4842 </div>
4846 <pre>
4848 </pre>
4862 truncating or zero extending that value to the size of the integer type. If
4866 change.</p>
4869 <pre>
4872 </pre>
4874 </div>
4876 <!-- _______________________________________________________________________ -->
4879 </div>
4883 <pre>
4885 </pre>
4893 value to cast, and a type to cast it to, which must be a
4901 than the size of a pointer then a zero extension is done. If they are the
4905 <pre>
4909 </pre>
4911 </div>
4913 <!-- _______________________________________________________________________ -->
4916 </div>
4920 <pre>
4922 </pre>
4930 non-aggregate first class value, and a type to cast it to, which must also be
4933 identical. If the source type is a pointer, the destination type must also be
4934 a pointer. This instruction supports bitwise conversion of vectors to
4935 integers and to vectors of other types (as long as they have the same
4936 size).</p>
4943 be converted to other pointer types with this instruction. To convert
4948 <pre>
4952 </pre>
4954 </div>
4956 <!-- ======================================================================= -->
4962 defy better classification.</p>
4964 </div>
4966 <!-- _______________________________________________________________________ -->
4968 </div>
4973 <pre>
4974 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4975 </pre>
4979 boolean values based on comparison of its two integer, integer vector, or
4980 pointer operands.</p>
4984 the condition code indicating the kind of comparison to perform. It is not a
4985 value, just a keyword. The possible condition code are:</p>
4987 <ol>
4998 </ol>
5002 typed. They must also be identical types.</p>
5008 result, as follows:</p>
5010 <ol>
5013 performed.</li>
5017 performed.</li>
5044 </ol>
5047 values are compared as if they were integers.</p>
5049 <p>If the operands are integer vectors, then they are compared element by
5054 <pre>
5061 </pre>
5063 <p>Note that the code generator does not yet support vector types with
5066 </div>
5068 <!-- _______________________________________________________________________ -->
5070 </div>
5075 <pre>
5076 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5077 </pre>
5081 values based on comparison of its operands.</p>
5083 <p>If the operands are floating point scalars, then the result type is a boolean
5086 <p>If the operands are floating point vectors, then the result type is a vector
5087 of boolean with the same number of elements as the operands being
5088 compared.</p>
5092 the condition code indicating the kind of comparison to perform. It is not a
5093 value, just a keyword. The possible condition code are:</p>
5095 <ol>
5112 </ol>
5120 identical types.</p>
5125 vectors, then the vectors are compared element by element. Each comparison
5127 follows:</p>
5129 <ol>
5173 </ol>
5176 <pre>
5181 </pre>
5183 <p>Note that the code generator does not yet support vector types with
5186 </div>
5188 <!-- _______________________________________________________________________ -->
5191 </div>
5196 <pre>
5198 </pre>
5202 SSA graph representing the function.</p>
5205 <p>The type of the incoming values is specified with the first type field. After
5207 one pair for each predecessor basic block of the current block. Only values
5209 arguments to the PHI node. Only labels may be used as the label
5210 arguments.</p>
5212 <p>There must be no non-phi instructions between the start of a basic block and
5213 the PHI instructions: i.e. PHI instructions must be first in a basic
5214 block.</p>
5216 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5217 occur on the edge from the corresponding predecessor block to the current
5219 value on the same edge).</p>
5223 specified by the pair corresponding to the predecessor basic block that
5224 executed just prior to the current block.</p>
5227 <pre>
5228 Loop: ; Infinite loop that counts from 0 on up...
5229 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5230 %nextindvar = add i32 %indvar, 1
5231 br label %Loop
5232 </pre>
5234 </div>
5236 <!-- _______________________________________________________________________ -->
5239 </div>
5244 <pre>
5245 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5248 </pre>
5252 condition, without branching.</p>
5257 values indicating the condition, and two values of the
5259 vectors and the condition is a scalar, then entire vectors are selected, not
5260 individual elements.</p>
5264 first value argument; otherwise, it returns the second value argument.</p>
5266 <p>If the condition is a vector of i1, then the value arguments must be vectors
5267 of the same size, and the selection is done element by element.</p>
5270 <pre>
5272 </pre>
5274 <p>Note that the code generator does not yet support conditions
5275 with vector type.</p>
5277 </div>
5279 <!-- _______________________________________________________________________ -->
5282 </div>
5287 <pre>
5288 <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>]
5289 </pre>
5297 <ol>
5299 access any allocas or varargs in the caller. Note that calls may be
5300 marked "tail" even if they do not occur before
5302 present, the function call is eligible for tail call optimization,
5304 optimized into a jump</a>. The code generator may optimize calls marked
5306 sibling call optimization</a> when the caller and callee have
5307 matching signatures, or 2) forced tail call optimization when the
5308 following extra requirements are met:
5309 <ul>
5310 <li>Caller and callee both have the calling
5312 <li>The call is in tail position (ret immediately follows call and ret
5313 uses value of call or is void).</li>
5317 constraints are met.</a></li>
5318 </ul>
5319 </li>
5322 convention</a> the call should use. If none is specified, the call
5323 defaults to using C calling conventions. The calling convention of the
5324 call must match the calling convention of the target function, or else the
5325 behavior is undefined.</li>
5332 type of the return value. Functions that return no value are marked
5336 being invoked. The argument types must match the types implied by this
5337 signature. This type can be omitted if the function is not varargs and if
5338 the function type does not return a pointer to a function.</li>
5341 be invoked. In most cases, this is a direct function invocation, but
5343 to function value.</li>
5346 signature argument types and parameter attributes. All arguments must be
5348 signature indicates the function accepts a variable number of arguments,
5349 the extra arguments can be specified.</li>
5354 </ol>
5358 a specified function, with its incoming arguments bound to the specified
5360 function, control flow continues with the instruction after the function
5361 call, and the return value of the function is bound to the result
5362 argument.</p>
5365 <pre>
5366 %retval = call i32 @test(i32 %argc)
5370 call void %foo(i8 97 signext)
5372 %struct.A = type { i32, i8 }
5378 </pre>
5380 <p>llvm treats calls to some functions with names and arguments that match the
5381 standard C99 library as being the C99 library functions, and may perform
5382 optimizations or generate code for them under that assumption. This is
5383 something we'd like to change in the future to provide better support for
5384 freestanding environments and non-C-based languages.</p>
5386 </div>
5388 <!-- _______________________________________________________________________ -->
5391 </div>
5396 <pre>
5398 </pre>
5402 the "variable argument" area of a function call. It is used to implement the
5407 argument. It returns a value of the specified argument type and increments
5414 to the next argument. For more information, see the variable argument
5417 <p>It is legal for this instruction to be called in a function which does not
5419 function.</p>
5423 argument.</p>
5428 <p>Note that the code generator does not yet fully support va_arg on many
5429 targets. Also, it does not currently support va_arg with aggregate types on
5430 any target.</p>
5432 </div>
5434 <!-- *********************************************************************** -->
5436 <!-- *********************************************************************** -->
5441 well known names and semantics and are required to follow certain
5442 restrictions. Overall, these intrinsics represent an extension mechanism for
5443 the LLVM language that does not require changing all of the transformations
5444 in LLVM when adding to the language (or the bitcode reader/writer, the
5445 parser, etc...).</p>
5448 prefix is reserved in LLVM for intrinsic names; thus, function names may not
5449 begin with this prefix. Intrinsic functions must always be external
5450 functions: you cannot define the body of intrinsic functions. Intrinsic
5451 functions may only be used in call or invoke instructions: it is illegal to
5452 take the address of an intrinsic function. Additionally, because intrinsic
5453 functions are part of the LLVM language, it is required if any are added that
5454 they be documented here.</p>
5456 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
5457 family of functions that perform the same operation but on different data
5458 types. Because LLVM can represent over 8 million different integer types,
5459 overloading is used commonly to allow an intrinsic function to operate on any
5460 integer type. One or more of the argument types or the result type can be
5461 overloaded to accept any integer type. Argument types may also be defined as
5462 exactly matching a previous argument's type or the result type. This allows
5463 an intrinsic function which accepts multiple arguments, but needs all of them
5464 to be of the same type, to only be overloaded with respect to a single
5465 argument or the result.</p>
5467 <p>Overloaded intrinsics will have the names of its overloaded argument types
5468 encoded into its function name, each preceded by a period. Only those types
5469 which are overloaded result in a name suffix. Arguments whose type is matched
5471 can take an integer of any width and returns an integer of exactly the same
5472 integer width. This leads to a family of functions such as
5474 %val)</tt>. Only one type, the return type, is overloaded, and only one type
5475 suffix is required. Because the argument's type is matched against the return
5476 type, it does not require its own name suffix.</p>
5478 <p>To learn how to add an intrinsic function, please see the
5481 </div>
5483 <!-- ======================================================================= -->
5486 </div>
5490 <p>Variable argument support is defined in LLVM with
5492 intrinsic functions. These functions are related to the similarly named
5495 <p>All of these functions operate on arguments that use a target-specific value
5496 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
5497 not define what this type is, so all transformations should be prepared to
5498 handle these functions regardless of the type used.</p>
5501 instruction and the variable argument handling intrinsic functions are
5502 used.</p>
5505 define i32 @test(i32 %X, ...) {
5506 ; Initialize variable argument processing
5507 %ap = alloca i8*
5508 %ap2 = bitcast i8** %ap to i8*
5509 call void @llvm.va_start(i8* %ap2)
5511 ; Read a single integer argument
5512 %tmp = va_arg i8** %ap, i32
5514 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5515 %aq = alloca i8*
5516 %aq2 = bitcast i8** %aq to i8*
5517 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5518 call void @llvm.va_end(i8* %aq2)
5520 ; Stop processing of arguments.
5521 call void @llvm.va_end(i8* %ap2)
5522 ret i32 %tmp
5523 }
5525 declare void @llvm.va_start(i8*)
5526 declare void @llvm.va_copy(i8*, i8*)
5527 declare void @llvm.va_end(i8*)
5528 </pre>
5530 </div>
5532 <!-- _______________________________________________________________________ -->
5535 </div>
5541 <pre>
5543 </pre>
5554 macro available in C. In a target-dependent way, it initializes
5558 need to know the last argument of the function as the compiler can figure
5559 that out.</p>
5561 </div>
5563 <!-- _______________________________________________________________________ -->
5566 </div>
5571 <pre>
5573 </pre>
5577 which has been initialized previously
5586 macro available in C. In a target-dependent way, it destroys
5592 </div>
5594 <!-- _______________________________________________________________________ -->
5597 </div>
5602 <pre>
5604 </pre>
5608 from the source argument list to the destination argument list.</p>
5613 from.</p>
5617 macro available in C. In a target-dependent way, it copies the
5619 element. This intrinsic is necessary because
5621 arbitrarily complex and require, for example, memory allocation.</p>
5623 </div>
5625 <!-- ======================================================================= -->
5628 </div>
5633 Collection</a> (GC) requires the implementation and generation of these
5635 roots on the stack</a>, as well as garbage collector implementations that
5637 barriers. Front-ends for type-safe garbage collected languages should generate
5638 these intrinsics to make use of the LLVM garbage collectors. For more details,
5642 <p>The garbage collection intrinsics only operate on objects in the generic
5643 address space (address space zero).</p>
5645 </div>
5647 <!-- _______________________________________________________________________ -->
5650 </div>
5655 <pre>
5656 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5657 </pre>
5661 the code generator, and allows some metadata to be associated with it.</p>
5664 <p>The first argument specifies the address of a stack object that contains the
5665 root pointer. The second pointer (which must be either a constant or a
5666 global value address) contains the meta-data to be associated with the
5667 root.</p>
5671 location. At compile-time, the code generator generates information to allow
5676 </div>
5678 <!-- _______________________________________________________________________ -->
5681 </div>
5686 <pre>
5687 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5688 </pre>
5692 locations, allowing garbage collector implementations that require read
5693 barriers.</p>
5696 <p>The second argument is the address to read from, which should be an address
5697 allocated from the garbage collector. The first object is a pointer to the
5698 start of the referenced object, if needed by the language runtime (otherwise
5699 null).</p>
5703 instruction, but may be replaced with substantially more complex code by the
5708 </div>
5710 <!-- _______________________________________________________________________ -->
5713 </div>
5718 <pre>
5719 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5720 </pre>
5724 locations, allowing garbage collector implementations that require write
5725 barriers (such as generational or reference counting collectors).</p>
5728 <p>The first argument is the reference to store, the second is the start of the
5729 object to store it to, and the third is the address of the field of Obj to
5730 store to. If the runtime does not require a pointer to the object, Obj may
5731 be null.</p>
5735 instruction, but may be replaced with substantially more complex code by the
5740 </div>
5742 <!-- ======================================================================= -->
5745 </div>
5749 <p>These intrinsics are provided by LLVM to expose special features that may
5750 only be implemented with code generator support.</p>
5752 </div>
5754 <!-- _______________________________________________________________________ -->
5757 </div>
5762 <pre>
5764 </pre>
5768 target-specific value indicating the return address of the current function
5769 or one of its callers.</p>
5772 <p>The argument to this intrinsic indicates which function to return the address
5773 for. Zero indicates the calling function, one indicates its caller, etc.
5778 indicating the return address of the specified call frame, or zero if it
5779 cannot be identified. The value returned by this intrinsic is likely to be
5780 incorrect or 0 for arguments other than zero, so it should only be used for
5781 debugging purposes.</p>
5783 <p>Note that calling this intrinsic does not prevent function inlining or other
5784 aggressive transformations, so the value returned may not be that of the
5785 obvious source-language caller.</p>
5787 </div>
5789 <!-- _______________________________________________________________________ -->
5792 </div>
5797 <pre>
5799 </pre>
5803 target-specific frame pointer value for the specified stack frame.</p>
5806 <p>The argument to this intrinsic indicates which function to return the frame
5807 pointer for. Zero indicates the calling function, one indicates its caller,
5812 indicating the frame address of the specified call frame, or zero if it
5813 cannot be identified. The value returned by this intrinsic is likely to be
5814 incorrect or 0 for arguments other than zero, so it should only be used for
5815 debugging purposes.</p>
5817 <p>Note that calling this intrinsic does not prevent function inlining or other
5818 aggressive transformations, so the value returned may not be that of the
5819 obvious source-language caller.</p>
5821 </div>
5823 <!-- _______________________________________________________________________ -->
5826 </div>
5831 <pre>
5832 declare i8* @llvm.stacksave()
5833 </pre>
5837 of the function stack, for use
5839 useful for implementing language features like scoped automatic variable
5840 sized arrays in C99.</p>
5843 <p>This intrinsic returns a opaque pointer value that can be passed
5851 </div>
5853 <!-- _______________________________________________________________________ -->
5856 </div>
5861 <pre>
5862 declare void @llvm.stackrestore(i8* %ptr)
5863 </pre>
5867 the function stack to the state it was in when the
5869 executed. This is useful for implementing language features like scoped
5870 automatic variable sized arrays in C99.</p>
5873 <p>See the description
5876 </div>
5878 <!-- _______________________________________________________________________ -->
5881 </div>
5886 <pre>
5888 </pre>
5892 insert a prefetch instruction if supported; otherwise, it is a noop.
5893 Prefetches have no effect on the behavior of the program but can change its
5894 performance characteristics.</p>
5904 <p>This intrinsic does not modify the behavior of the program. In particular,
5905 prefetches cannot trap and do not produce a value. On targets that support
5906 this intrinsic, the prefetch can provide hints to the processor cache for
5907 better performance.</p>
5909 </div>
5911 <!-- _______________________________________________________________________ -->
5914 </div>
5919 <pre>
5921 </pre>
5925 Counter (PC) in a region of code to simulators and other tools. The method
5926 is target specific, but it is expected that the marker will use exported
5927 symbols to transmit the PC of the marker. The marker makes no guarantees
5928 that it will remain with any specific instruction after optimizations. It is
5929 possible that the presence of a marker will inhibit optimizations. The
5930 intended use is to be inserted after optimizations to allow correlations of
5931 simulation runs.</p>
5937 <p>This intrinsic does not modify the behavior of the program. Backends that do
5938 not support this intrinsic may ignore it.</p>
5940 </div>
5942 <!-- _______________________________________________________________________ -->
5945 </div>
5950 <pre>
5951 declare i64 @llvm.readcyclecounter()
5952 </pre>
5956 counter register (or similar low latency, high accuracy clocks) on those
5957 targets that support it. On X86, it should map to RDTSC. On Alpha, it
5958 should map to RPCC. As the backing counters overflow quickly (on the order
5962 <p>When directly supported, reading the cycle counter should not modify any
5963 memory. Implementations are allowed to either return a application specific
5964 value or a system wide value. On backends without support, this is lowered
5967 </div>
5969 <!-- ======================================================================= -->
5972 </div>
5976 <p>LLVM provides intrinsics for a few important standard C library functions.
5977 These intrinsics allow source-language front-ends to pass information about
5978 the alignment of the pointer arguments to the code generator, providing
5979 opportunity for more efficient code generation.</p>
5981 </div>
5983 <!-- _______________________________________________________________________ -->
5986 </div>
5992 integer bit width and for different address spaces. Not all targets support
5993 all bit widths however.</p>
5995 <pre>
6000 </pre>
6004 source location to the destination location.</p>
6007 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6008 and the pointers can be in specified address spaces.</p>
6012 <p>The first argument is a pointer to the destination, the second is a pointer
6013 to the source. The third argument is an integer argument specifying the
6014 number of bytes to copy, the fourth argument is the alignment of the
6015 source and destination locations, and the fifth is a boolean indicating a
6016 volatile access.</p>
6019 then the caller guarantees that both the source and destination pointers are
6020 aligned to that boundary.</p>
6024 The detailed access behavior is not very cleanly specified and it is unwise
6025 to depend on it.</p>
6030 source location to the destination location, which are not allowed to
6031 overlap. It copies "len" bytes of memory over. If the argument is known to
6032 be aligned to some boundary, this can be specified as the fourth argument,
6035 </div>
6037 <!-- _______________________________________________________________________ -->
6040 </div>
6045 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
6046 width and for different address space. Not all targets support all bit
6047 widths however.</p>
6049 <pre>
6054 </pre>
6058 source location to the destination location. It is similar to the
6060 overlap.</p>
6063 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6064 and the pointers can be in specified address spaces.</p>
6068 <p>The first argument is a pointer to the destination, the second is a pointer
6069 to the source. The third argument is an integer argument specifying the
6070 number of bytes to copy, the fourth argument is the alignment of the
6071 source and destination locations, and the fifth is a boolean indicating a
6072 volatile access.</p>
6075 then the caller guarantees that the source and destination pointers are
6076 aligned to that boundary.</p>
6080 The detailed access behavior is not very cleanly specified and it is unwise
6081 to depend on it.</p>
6086 source location to the destination location, which may overlap. It copies
6087 "len" bytes of memory over. If the argument is known to be aligned to some
6088 boundary, this can be specified as the fourth argument, otherwise it should
6091 </div>
6093 <!-- _______________________________________________________________________ -->
6096 </div>
6101 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
6102 width and for different address spaces. However, not all targets support all
6103 bit widths.</p>
6105 <pre>
6110 </pre>
6114 particular byte value.</p>
6117 intrinsic does not return a value and takes extra alignment/volatile
6118 arguments. Also, the destination can be in an arbitrary address space.</p>
6121 <p>The first argument is a pointer to the destination to fill, the second is the
6122 byte value with which to fill it, the third argument is an integer argument
6123 specifying the number of bytes to fill, and the fourth argument is the known
6124 alignment of the destination location.</p>
6127 then the caller guarantees that the destination pointer is aligned to that
6128 boundary.</p>
6132 The detailed access behavior is not very cleanly specified and it is unwise
6133 to depend on it.</p>
6137 at the destination location. If the argument is known to be aligned to some
6138 boundary, this can be specified as the fourth argument, otherwise it should
6141 </div>
6143 <!-- _______________________________________________________________________ -->
6146 </div>
6152 floating point or vector of floating point type. Not all targets support all
6153 types however.</p>
6155 <pre>
6156 declare float @llvm.sqrt.f32(float %Val)
6157 declare double @llvm.sqrt.f64(double %Val)
6158 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6159 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6160 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6161 </pre>
6167 behavior for negative numbers other than -0.0 (which allows for better
6168 optimization, because there is no need to worry about errno being
6172 <p>The argument and return value are floating point numbers of the same
6173 type.</p>
6176 <p>This function returns the sqrt of the specified operand if it is a
6177 nonnegative floating point number.</p>
6179 </div>
6181 <!-- _______________________________________________________________________ -->
6184 </div>
6190 floating point or vector of floating point type. Not all targets support all
6191 types however.</p>
6193 <pre>
6194 declare float @llvm.powi.f32(float %Val, i32 %power)
6195 declare double @llvm.powi.f64(double %Val, i32 %power)
6196 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6197 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6198 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6199 </pre>
6203 specified (positive or negative) power. The order of evaluation of
6204 multiplications is not defined. When a vector of floating point type is
6205 used, the second argument remains a scalar integer value.</p>
6208 <p>The second argument is an integer power, and the first is a value to raise to
6209 that power.</p>
6212 <p>This function returns the first value raised to the second power with an
6213 unspecified sequence of rounding operations.</p>
6215 </div>
6217 <!-- _______________________________________________________________________ -->
6220 </div>
6226 floating point or vector of floating point type. Not all targets support all
6227 types however.</p>
6229 <pre>
6230 declare float @llvm.sin.f32(float %Val)
6231 declare double @llvm.sin.f64(double %Val)
6232 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6233 declare fp128 @llvm.sin.f128(fp128 %Val)
6234 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6235 </pre>
6241 <p>The argument and return value are floating point numbers of the same
6242 type.</p>
6245 <p>This function returns the sine of the specified operand, returning the same
6247 in the same way.</p>
6249 </div>
6251 <!-- _______________________________________________________________________ -->
6254 </div>
6260 floating point or vector of floating point type. Not all targets support all
6261 types however.</p>
6263 <pre>
6264 declare float @llvm.cos.f32(float %Val)
6265 declare double @llvm.cos.f64(double %Val)
6266 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6267 declare fp128 @llvm.cos.f128(fp128 %Val)
6268 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6269 </pre>
6275 <p>The argument and return value are floating point numbers of the same
6276 type.</p>
6279 <p>This function returns the cosine of the specified operand, returning the same
6281 in the same way.</p>
6283 </div>
6285 <!-- _______________________________________________________________________ -->
6288 </div>
6294 floating point or vector of floating point type. Not all targets support all
6295 types however.</p>
6297 <pre>
6298 declare float @llvm.pow.f32(float %Val, float %Power)
6299 declare double @llvm.pow.f64(double %Val, double %Power)
6300 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6301 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6302 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6303 </pre>
6307 specified (positive or negative) power.</p>
6310 <p>The second argument is a floating point power, and the first is a value to
6311 raise to that power.</p>
6314 <p>This function returns the first value raised to the second power, returning
6316 conditions in the same way.</p>
6318 </div>
6320 <!-- ======================================================================= -->
6323 </div>
6327 <p>LLVM provides intrinsics for a few important bit manipulation operations.
6328 These allow efficient code generation for some algorithms.</p>
6330 </div>
6332 <!-- _______________________________________________________________________ -->
6335 </div>
6340 <p>This is an overloaded intrinsic function. You can use bswap on any integer
6343 <pre>
6347 </pre>
6351 values with an even number of bytes (positive multiple of 16 bits). These
6352 are useful for performing operations on data that is not in the target's
6353 native byte order.</p>
6357 and low byte of the input i16 swapped. Similarly,
6363 more, respectively).</p>
6365 </div>
6367 <!-- _______________________________________________________________________ -->
6370 </div>
6375 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
6376 width. Not all targets support all bit widths however.</p>
6378 <pre>
6384 </pre>
6388 in a value.</p>
6391 <p>The only argument is the value to be counted. The argument may be of any
6392 integer type. The return type must match the argument type.</p>
6397 </div>
6399 <!-- _______________________________________________________________________ -->
6402 </div>
6408 integer bit width. Not all targets support all bit widths however.</p>
6410 <pre>
6416 </pre>
6420 leading zeros in a variable.</p>
6423 <p>The only argument is the value to be counted. The argument may be of any
6424 integer type. The return type must match the argument type.</p>
6428 zeros in a variable. If the src == 0 then the result is the size in bits of
6431 </div>
6433 <!-- _______________________________________________________________________ -->
6436 </div>
6442 integer bit width. Not all targets support all bit widths however.</p>
6444 <pre>
6450 </pre>
6454 trailing zeros.</p>
6457 <p>The only argument is the value to be counted. The argument may be of any
6458 integer type. The return type must match the argument type.</p>
6462 zeros in a variable. If the src == 0 then the result is the size in bits of
6465 </div>
6467 <!-- ======================================================================= -->
6470 </div>
6476 </div>
6478 <!-- _______________________________________________________________________ -->
6481 </div>
6487 on any integer bit width.</p>
6489 <pre>
6490 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6491 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6492 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6493 </pre>
6497 a signed addition of the two arguments, and indicate whether an overflow
6498 occurred during the signed summation.</p>
6501 <p>The arguments (%a and %b) and the first element of the result structure may
6502 be of integer types of any bit width, but they must have the same bit
6503 width. The second element of the result structure must be of
6505 undergo signed addition.</p>
6509 a signed addition of the two variables. They return a structure — the
6510 first element of which is the signed summation, and the second element of
6511 which is a bit specifying if the signed summation resulted in an
6512 overflow.</p>
6515 <pre>
6516 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6517 %sum = extractvalue {i32, i1} %res, 0
6518 %obit = extractvalue {i32, i1} %res, 1
6519 br i1 %obit, label %overflow, label %normal
6520 </pre>
6522 </div>
6524 <!-- _______________________________________________________________________ -->
6527 </div>
6533 on any integer bit width.</p>
6535 <pre>
6536 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6537 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6538 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6539 </pre>
6543 an unsigned addition of the two arguments, and indicate whether a carry
6544 occurred during the unsigned summation.</p>
6547 <p>The arguments (%a and %b) and the first element of the result structure may
6548 be of integer types of any bit width, but they must have the same bit
6549 width. The second element of the result structure must be of
6551 undergo unsigned addition.</p>
6555 an unsigned addition of the two arguments. They return a structure —
6556 the first element of which is the sum, and the second element of which is a
6557 bit specifying if the unsigned summation resulted in a carry.</p>
6560 <pre>
6561 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6562 %sum = extractvalue {i32, i1} %res, 0
6563 %obit = extractvalue {i32, i1} %res, 1
6564 br i1 %obit, label %carry, label %normal
6565 </pre>
6567 </div>
6569 <!-- _______________________________________________________________________ -->
6572 </div>
6578 on any integer bit width.</p>
6580 <pre>
6581 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6582 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6583 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6584 </pre>
6588 a signed subtraction of the two arguments, and indicate whether an overflow
6589 occurred during the signed subtraction.</p>
6592 <p>The arguments (%a and %b) and the first element of the result structure may
6593 be of integer types of any bit width, but they must have the same bit
6594 width. The second element of the result structure must be of
6596 undergo signed subtraction.</p>
6600 a signed subtraction of the two arguments. They return a structure —
6601 the first element of which is the subtraction, and the second element of
6602 which is a bit specifying if the signed subtraction resulted in an
6603 overflow.</p>
6606 <pre>
6607 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6608 %sum = extractvalue {i32, i1} %res, 0
6609 %obit = extractvalue {i32, i1} %res, 1
6610 br i1 %obit, label %overflow, label %normal
6611 </pre>
6613 </div>
6615 <!-- _______________________________________________________________________ -->
6618 </div>
6624 on any integer bit width.</p>
6626 <pre>
6627 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6628 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6629 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6630 </pre>
6634 an unsigned subtraction of the two arguments, and indicate whether an
6635 overflow occurred during the unsigned subtraction.</p>
6638 <p>The arguments (%a and %b) and the first element of the result structure may
6639 be of integer types of any bit width, but they must have the same bit
6640 width. The second element of the result structure must be of
6642 undergo unsigned subtraction.</p>
6646 an unsigned subtraction of the two arguments. They return a structure —
6647 the first element of which is the subtraction, and the second element of
6648 which is a bit specifying if the unsigned subtraction resulted in an
6649 overflow.</p>
6652 <pre>
6653 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6654 %sum = extractvalue {i32, i1} %res, 0
6655 %obit = extractvalue {i32, i1} %res, 1
6656 br i1 %obit, label %overflow, label %normal
6657 </pre>
6659 </div>
6661 <!-- _______________________________________________________________________ -->
6664 </div>
6670 on any integer bit width.</p>
6672 <pre>
6673 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6674 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6675 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6676 </pre>
6681 a signed multiplication of the two arguments, and indicate whether an
6682 overflow occurred during the signed multiplication.</p>
6685 <p>The arguments (%a and %b) and the first element of the result structure may
6686 be of integer types of any bit width, but they must have the same bit
6687 width. The second element of the result structure must be of
6689 undergo signed multiplication.</p>
6693 a signed multiplication of the two arguments. They return a structure —
6694 the first element of which is the multiplication, and the second element of
6695 which is a bit specifying if the signed multiplication resulted in an
6696 overflow.</p>
6699 <pre>
6700 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6701 %sum = extractvalue {i32, i1} %res, 0
6702 %obit = extractvalue {i32, i1} %res, 1
6703 br i1 %obit, label %overflow, label %normal
6704 </pre>
6706 </div>
6708 <!-- _______________________________________________________________________ -->
6711 </div>
6717 on any integer bit width.</p>
6719 <pre>
6720 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6721 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6722 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6723 </pre>
6727 a unsigned multiplication of the two arguments, and indicate whether an
6728 overflow occurred during the unsigned multiplication.</p>
6731 <p>The arguments (%a and %b) and the first element of the result structure may
6732 be of integer types of any bit width, but they must have the same bit
6733 width. The second element of the result structure must be of
6735 undergo unsigned multiplication.</p>
6739 an unsigned multiplication of the two arguments. They return a structure
6740 — the first element of which is the multiplication, and the second
6741 element of which is a bit specifying if the unsigned multiplication resulted
6742 in an overflow.</p>
6745 <pre>
6746 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6747 %sum = extractvalue {i32, i1} %res, 0
6748 %obit = extractvalue {i32, i1} %res, 1
6749 br i1 %obit, label %overflow, label %normal
6750 </pre>
6752 </div>
6754 <!-- ======================================================================= -->
6757 </div>
6761 <p>Half precision floating point is a storage-only format. This means that it is
6762 a dense encoding (in memory) but does not support computation in the
6763 format.</p>
6765 <p>This means that code must first load the half-precision floating point
6766 value as an i16, then convert it to float with <a
6768 Computation can then be performed on the float value (including extending to
6769 double etc). To store the value back to memory, it is first converted to
6770 float if needed, then converted to i16 with
6772 storing as an i16 value.</p>
6773 </div>
6775 <!-- _______________________________________________________________________ -->
6778 </div>
6783 <pre>
6784 declare i16 @llvm.convert.to.fp16(f32 %a)
6785 </pre>
6789 a conversion from single precision floating point format to half precision
6790 floating point format.</p>
6793 <p>The intrinsic function contains single argument - the value to be
6794 converted.</p>
6798 a conversion from single precision floating point format to half precision
6800 contains the converted number.</p>
6803 <pre>
6804 %res = call i16 @llvm.convert.to.fp16(f32 %a)
6805 store i16 %res, i16* @x, align 2
6806 </pre>
6808 </div>
6810 <!-- _______________________________________________________________________ -->
6813 </div>
6818 <pre>
6819 declare f32 @llvm.convert.from.fp16(i16 %a)
6820 </pre>
6824 a conversion from half precision floating point format to single precision
6825 floating point format.</p>
6828 <p>The intrinsic function contains single argument - the value to be
6829 converted.</p>
6833 conversion from half single precision floating point format to single
6834 precision floating point format. The input half-float value is represented by
6838 <pre>
6839 %a = load i16* @x, align 2
6840 %res = call f32 @llvm.convert.from.fp16(i16 %a)
6841 </pre>
6843 </div>
6845 <!-- ======================================================================= -->
6848 </div>
6853 prefix), are described in
6857 </div>
6859 <!-- ======================================================================= -->
6862 </div>
6866 <p>The LLVM exception handling intrinsics (which all start with
6871 </div>
6873 <!-- ======================================================================= -->
6876 </div>
6880 <p>This intrinsic makes it possible to excise one parameter, marked with
6882 The result is a callable
6883 function pointer lacking the nest parameter - the caller does not need to
6884 provide a value for it. Instead, the value to use is stored in advance in a
6885 "trampoline", a block of memory usually allocated on the stack, which also
6886 contains code to splice the nest value into the argument list. This is used
6887 to implement the GCC nested function address extension.</p>
6889 <p>For example, if the function is
6892 follows:</p>
6897 %p = call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval)
6898 %fp = bitcast i8* %p to i32 (i32, i32)*
6899 </pre>
6904 </div>
6906 <!-- _______________________________________________________________________ -->
6909 </div>
6914 <pre>
6916 </pre>
6920 function pointer suitable for executing it.</p>
6925 sufficiently aligned block of memory; this memory is written to by the
6926 intrinsic. Note that the size and the alignment are target-specific - LLVM
6927 currently provides no portable way of determining them, so a front-end that
6928 generates this intrinsic needs to have some target-specific knowledge.
6934 dependent code, turning it into a function. A pointer to this function is
6936 function pointer type</a> before being called. The new function's signature
6939 is allowed, and it must be of pointer type. Calling the new function is
6944 returned function pointer is undefined.</p>
6946 </div>
6948 <!-- ======================================================================= -->
6951 </div>
6956 hardware constructs for atomic operations and memory synchronization. This
6957 provides an interface to the hardware, not an interface to the programmer. It
6958 is aimed at a low enough level to allow any programming models or APIs
6959 (Application Programming Interfaces) which need atomic behaviors to map
6960 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
6961 hardware provides a "universal IR" for source languages, it also provides a
6962 starting point for developing a "universal" atomic operation and
6963 synchronization IR.</p>
6966 software transaction memory systems, atomic primitives, and intrinsic
6967 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6968 application libraries. The hardware interface provided by LLVM should allow
6969 a clean implementation of all of these APIs and parallel programming models.
6970 No one model or paradigm should be selected above others unless the hardware
6971 itself ubiquitously does so.</p>
6973 </div>
6975 <!-- _______________________________________________________________________ -->
6978 </div>
6981 <pre>
6982 declare void @llvm.memory.barrier(i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device>)
6983 </pre>
6987 specific pairs of memory access types.</p>
6991 The first four arguments enables a specific barrier as listed below. The
6992 fifth argument specifies that the barrier applies to io or device or uncached
6993 memory.</p>
6995 <ul>
7001 </ul>
7004 <p>This intrinsic causes the system to enforce some ordering constraints upon
7005 the loads and stores of the program. This barrier does not
7008 and store operations (f.ex. load-load, or store-load), all of the first
7009 operations preceding the barrier will complete before any of the second
7010 operations succeeding the barrier begin. Specifically the semantics for each
7011 pairing is as follows:</p>
7013 <ul>
7015 after the barrier begins.</li>
7017 store after the barrier begins.</li>
7019 store after the barrier begins.</li>
7021 load after the barrier begins.</li>
7022 </ul>
7025 is enabled in a single memory barrier intrinsic.</p>
7027 <p>Backends may implement stronger barriers than those requested when they do
7028 not support as fine grained a barrier as requested. Some architectures do
7029 not need all types of barriers and on such architectures, these become
7030 noops.</p>
7033 <pre>
7034 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7035 %ptr = bitcast i8* %mallocP to i32*
7036 store i32 4, %ptr
7039 call void @llvm.memory.barrier(i1 false, i1 true, i1 false, i1 false)
7042 </pre>
7044 </div>
7046 <!-- _______________________________________________________________________ -->
7049 </div>
7055 any integer bit width and for different address spaces. Not all targets
7056 support all bit widths however.</p>
7058 <pre>
7060 declare i16 @llvm.atomic.cmp.swap.i16.p0i16(i16* <ptr>, i16 <cmp>, i16 <val>)
7061 declare i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* <ptr>, i32 <cmp>, i32 <val>)
7062 declare i64 @llvm.atomic.cmp.swap.i64.p0i64(i64* <ptr>, i64 <cmp>, i64 <val>)
7063 </pre>
7066 <p>This loads a value in memory and compares it to a given value. If they are
7067 equal, it stores a new value into the memory.</p>
7073 this integer type. While any bit width integer may be used, targets may only
7074 lower representations they support in hardware.</p>
7077 <p>This entire intrinsic must be executed atomically. It first loads the value
7080 memory. The loaded value is yielded in all cases. This provides the
7081 equivalent of an atomic compare-and-swap operation within the SSA
7082 framework.</p>
7085 <pre>
7086 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7087 %ptr = bitcast i8* %mallocP to i32*
7088 store i32 4, %ptr
7091 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 4, %val1)
7097 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 5, %val2)
7102 </pre>
7104 </div>
7106 <!-- _______________________________________________________________________ -->
7109 </div>
7114 integer bit width. Not all targets support all bit widths however.</p>
7116 <pre>
7121 </pre>
7132 integer type. The targets may only lower integer representations they
7133 support.</p>
7138 equivalent of an atomic swap operation within the SSA framework.</p>
7141 <pre>
7142 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7143 %ptr = bitcast i8* %mallocP to i32*
7144 store i32 4, %ptr
7147 %result1 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val1)
7153 %result2 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val2)
7158 </pre>
7160 </div>
7162 <!-- _______________________________________________________________________ -->
7166 </div>
7172 any integer bit width. Not all targets support all bit widths however.</p>
7174 <pre>
7179 </pre>
7186 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7187 and the second an integer value. The result is also an integer value. These
7188 integer types can have any bit width, but they must all have the same bit
7189 width. The targets may only lower integer representations they support.</p>
7192 <p>This intrinsic does a series of operations atomically. It first loads the
7197 <pre>
7198 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7199 %ptr = bitcast i8* %mallocP to i32*
7200 store i32 4, %ptr
7201 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 4)
7203 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 2)
7205 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 5)
7208 </pre>
7210 </div>
7212 <!-- _______________________________________________________________________ -->
7216 </div>
7222 any integer bit width and for different address spaces. Not all targets
7223 support all bit widths however.</p>
7225 <pre>
7230 </pre>
7237 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7238 and the second an integer value. The result is also an integer value. These
7239 integer types can have any bit width, but they must all have the same bit
7240 width. The targets may only lower integer representations they support.</p>
7243 <p>This intrinsic does a series of operations atomically. It first loads the
7249 <pre>
7250 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7251 %ptr = bitcast i8* %mallocP to i32*
7252 store i32 8, %ptr
7253 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 4)
7255 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 2)
7257 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 5)
7260 </pre>
7262 </div>
7264 <!-- _______________________________________________________________________ -->
7270 </div>
7275 <p>These are overloaded intrinsics. You can
7278 bit width and for different address spaces. Not all targets support all bit
7279 widths however.</p>
7281 <pre>
7286 </pre>
7288 <pre>
7293 </pre>
7295 <pre>
7300 </pre>
7302 <pre>
7307 </pre>
7315 <p>These intrinsics take two arguments, the first a pointer to an integer value
7316 and the second an integer value. The result is also an integer value. These
7317 integer types can have any bit width, but they must all have the same bit
7318 width. The targets may only lower integer representations they support.</p>
7321 <p>These intrinsics does a series of operations atomically. They first load the
7327 <pre>
7328 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7329 %ptr = bitcast i8* %mallocP to i32*
7330 store i32 0x0F0F, %ptr
7331 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32(i32* %ptr, i32 0xFF)
7333 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32(i32* %ptr, i32 0xFF)
7335 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32(i32* %ptr, i32 0F)
7337 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32(i32* %ptr, i32 0F)
7340 </pre>
7342 </div>
7344 <!-- _______________________________________________________________________ -->
7350 </div>
7358 address spaces. Not all targets support all bit widths however.</p>
7360 <pre>
7365 </pre>
7367 <pre>
7372 </pre>
7374 <pre>
7379 </pre>
7381 <pre>
7386 </pre>
7389 <p>These intrinsics takes the signed or unsigned minimum or maximum of
7394 <p>These intrinsics take two arguments, the first a pointer to an integer value
7395 and the second an integer value. The result is also an integer value. These
7396 integer types can have any bit width, but they must all have the same bit
7397 width. The targets may only lower integer representations they support.</p>
7400 <p>These intrinsics does a series of operations atomically. They first load the
7406 <pre>
7407 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7408 %ptr = bitcast i8* %mallocP to i32*
7409 store i32 7, %ptr
7410 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32(i32* %ptr, i32 -2)
7412 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32(i32* %ptr, i32 8)
7414 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32(i32* %ptr, i32 10)
7416 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32(i32* %ptr, i32 30)
7419 </pre>
7421 </div>
7424 <!-- ======================================================================= -->
7427 </div>
7431 <p>This class of intrinsics exists to information about the lifetime of memory
7432 objects and ranges where variables are immutable.</p>
7434 </div>
7436 <!-- _______________________________________________________________________ -->
7439 </div>
7444 <pre>
7446 </pre>
7450 object's lifetime.</p>
7453 <p>The first argument is a constant integer representing the size of the
7454 object, or -1 if it is variable sized. The second argument is a pointer to
7455 the object.</p>
7458 <p>This intrinsic indicates that before this point in the code, the value of the
7460 never be used and has an undefined value. A load from the pointer that
7461 precedes this intrinsic can be replaced with
7464 </div>
7466 <!-- _______________________________________________________________________ -->
7469 </div>
7474 <pre>
7476 </pre>
7480 object's lifetime.</p>
7483 <p>The first argument is a constant integer representing the size of the
7484 object, or -1 if it is variable sized. The second argument is a pointer to
7485 the object.</p>
7488 <p>This intrinsic indicates that after this point in the code, the value of the
7490 never be used and has an undefined value. Any stores into the memory object
7491 following this intrinsic may be removed as dead.
7493 </div>
7495 <!-- _______________________________________________________________________ -->
7498 </div>
7503 <pre>
7505 </pre>
7509 a memory object will not change.</p>
7512 <p>The first argument is a constant integer representing the size of the
7513 object, or -1 if it is variable sized. The second argument is a pointer to
7514 the object.</p>
7518 the return value, the referenced memory location is constant and
7519 unchanging.</p>
7521 </div>
7523 <!-- _______________________________________________________________________ -->
7526 </div>
7531 <pre>
7532 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
7533 </pre>
7537 a memory object are mutable.</p>
7541 The second argument is a constant integer representing the size of the
7542 object, or -1 if it is variable sized and the third argument is a pointer
7543 to the object.</p>
7548 </div>
7550 <!-- ======================================================================= -->
7553 </div>
7557 <p>This class of intrinsics is designed to be generic and has no specific
7558 purpose.</p>
7560 </div>
7562 <!-- _______________________________________________________________________ -->
7565 </div>
7570 <pre>
7571 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
7572 </pre>
7578 <p>The first argument is a pointer to a value, the second is a pointer to a
7579 global string, the third is a pointer to a global string which is the source
7580 file name, and the last argument is the line number.</p>
7583 <p>This intrinsic allows annotation of local variables with arbitrary strings.
7584 This can be useful for special purpose optimizations that want to look for
7585 these annotations. These have no other defined use, they are ignored by code
7586 generation and optimization.</p>
7588 </div>
7590 <!-- _______________________________________________________________________ -->
7593 </div>
7599 any integer bit width.</p>
7601 <pre>
7602 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
7603 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
7604 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
7605 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
7606 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
7607 </pre>
7613 <p>The first argument is an integer value (result of some expression), the
7614 second is a pointer to a global string, the third is a pointer to a global
7615 string which is the source file name, and the last argument is the line
7616 number. It returns the value of the first argument.</p>
7619 <p>This intrinsic allows annotations to be put on arbitrary expressions with
7620 arbitrary strings. This can be useful for special purpose optimizations that
7621 want to look for these annotations. These have no other defined use, they
7622 are ignored by code generation and optimization.</p>
7624 </div>
7626 <!-- _______________________________________________________________________ -->
7629 </div>
7634 <pre>
7635 declare void @llvm.trap()
7636 </pre>
7645 <p>This intrinsics is lowered to the target dependent trap instruction. If the
7646 target does not have a trap instruction, this intrinsic will be lowered to
7649 </div>
7651 <!-- _______________________________________________________________________ -->
7654 </div>
7659 <pre>
7661 </pre>
7666 ensure that it is placed on the stack before local variables.</p>
7670 arguments. The first argument is the value loaded from the stack
7672 that has enough space to hold the value of the guard.</p>
7675 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
7677 stack. This is to ensure that if a local variable on the stack is
7678 overwritten, it will destroy the value of the guard. When the function exits,
7679 the guard on the stack is checked against the original guard. If they are
7681 function.</p>
7683 </div>
7685 <!-- _______________________________________________________________________ -->
7688 </div>
7693 <pre>
7696 </pre>
7700 the optimizers to determine at compile time whether a) an operation (like
7701 memcpy) will overflow a buffer that corresponds to an object, or b) that a
7702 runtime check for overflow isn't necessary. An object in this context means
7703 an allocation of a specific class, structure, array, or other object.</p>
7716 compile time.</p>
7718 </div>
7720 <!-- *********************************************************************** -->
7721 <hr>
7722 <address>
7730 Last modified: $Date$
7731 </address>
7733 </body>
7734 </html>