| blob | 5a53c9aed4e69ed3b5e68bb829be9a12f68040bb |
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>
850 that the address is not significant, only the content. Constants marked
851 like this can be merged with other constants if they have the same
854 constant whose address is significant.</p>
856 <p>A global variable may be declared to reside in a target-specific numbered
857 address space. For targets that support them, address spaces may affect how
858 optimizations are performed and/or what target instructions are used to
859 access the variable. The default address space is zero. The address space
860 qualifier must precede any other attributes.</p>
862 <p>LLVM allows an explicit section to be specified for globals. If the target
863 supports it, it will emit globals to the section specified.</p>
865 <p>An explicit alignment may be specified for a global, which must be a power
866 of 2. If not present, or if the alignment is set to zero, the alignment of
867 the global is set by the target to whatever it feels convenient. If an
868 explicit alignment is specified, the global is forced to have exactly that
869 alignment. Targets and optimizers are not allowed to over-align the global
870 if the global has an assigned section. In this case, the extra alignment
871 could be observable: for example, code could assume that the globals are
872 densely packed in their section and try to iterate over them as an array,
873 alignment padding would break this iteration.</p>
875 <p>For example, the following defines a global in a numbered address space with
876 an initializer, section, and alignment:</p>
880 </pre>
882 </div>
885 <!-- ======================================================================= -->
888 </div>
898 name, a (possibly empty) argument list (each with optional
902 curly brace, a list of basic blocks, and a closing curly brace.</p>
910 name, a possibly empty list of arguments, an optional alignment, and an
913 <p>A function definition contains a list of basic blocks, forming the CFG
914 (Control Flow Graph) for the function. Each basic block may optionally start
915 with a label (giving the basic block a symbol table entry), contains a list
917 instruction (such as a branch or function return).</p>
919 <p>The first basic block in a function is special in two ways: it is immediately
920 executed on entrance to the function, and it is not allowed to have
921 predecessor basic blocks (i.e. there can not be any branches to the entry
922 block of a function). Because the block can have no predecessors, it also
925 <p>LLVM allows an explicit section to be specified for functions. If the target
926 supports it, it will emit functions to the section specified.</p>
928 <p>An explicit alignment may be specified for a function. If not present, or if
929 the alignment is set to zero, the alignment of the function is set by the
930 target to whatever it feels convenient. If an explicit alignment is
931 specified, the function is forced to have at least that much alignment. All
935 be significant and two identical functions can be merged</p>.
944 </pre>
946 </div>
948 <!-- ======================================================================= -->
951 </div>
956 function, global variable, another alias or bitcast of global value). Aliases
963 </pre>
965 </div>
967 <!-- ======================================================================= -->
970 </div>
975 nodes</a> (but not metadata strings) are the only valid operands for
976 a named metadata.</p>
980 ; Some unnamed metadata nodes, which are referenced by the named metadata.
984 ; A named metadata.
986 </pre>
988 </div>
990 <!-- ======================================================================= -->
995 <p>The return type and each parameter of a function type may have a set of
997 used to communicate additional information about the result or parameters of
998 a function. Parameter attributes are considered to be part of the function,
999 not of the function type, so functions with different parameter attributes
1000 can have the same function type.</p>
1002 <p>Parameter attributes are simple keywords that follow the type specified. If
1003 multiple parameter attributes are needed, they are space separated. For
1004 example:</p>
1007 declare i32 @printf(i8* noalias nocapture, ...)
1008 declare i32 @atoi(i8 zeroext)
1009 declare signext i8 @returns_signed_char()
1010 </pre>
1017 <dl>
1019 <dd>This indicates to the code generator that the parameter or return value
1020 should be zero-extended to a 32-bit value by the caller (for a parameter)
1021 or the callee (for a return value).</dd>
1024 <dd>This indicates to the code generator that the parameter or return value
1025 should be sign-extended to a 32-bit value by the caller (for a parameter)
1026 or the callee (for a return value).</dd>
1029 <dd>This indicates that this parameter or return value should be treated in a
1030 special target-dependent fashion during while emitting code for a function
1031 call or return (usually, by putting it in a register as opposed to memory,
1032 though some targets use it to distinguish between two different kinds of
1033 registers). Use of this attribute is target-specific.</dd>
1036 <dd><p>This indicates that the pointer parameter should really be passed by
1037 value to the function. The attribute implies that a hidden copy of the
1038 pointee
1039 is made between the caller and the callee, so the callee is unable to
1040 modify the value in the callee. This attribute is only valid on LLVM
1041 pointer arguments. It is generally used to pass structs and arrays by
1042 value, but is also valid on pointers to scalars. The copy is considered
1043 to belong to the caller not the callee (for example,
1046 values.</p>
1048 <p>The byval attribute also supports specifying an alignment with
1049 the align attribute. It indicates the alignment of the stack slot to
1050 form and the known alignment of the pointer specified to the call site. If
1051 the alignment is not specified, then the code generator makes a
1052 target-specific assumption.</p></dd>
1055 <dd>This indicates that the pointer parameter specifies the address of a
1056 structure that is the return value of the function in the source program.
1057 This pointer must be guaranteed by the caller to be valid: loads and
1058 stores to the structure may be assumed by the callee to not to trap. This
1059 may only be applied to the first parameter. This is not a valid attribute
1060 for return values. </dd>
1063 <dd>This indicates that pointer values
1066 ignoring certain "irrelevant" dependencies.
1067 For a call to the parent function, dependencies between memory
1068 references from before or after the call and from those during the call
1070 return value used in that call.
1071 The caller shares the responsibility with the callee for ensuring that
1072 these requirements are met.
1073 For further details, please see the discussion of the NoAlias response in
1075 <br>
1078 arguments, though it is slightly weaker.
1079 <br>
1082 </dd>
1085 <dd>This indicates that the callee does not make any copies of the pointer
1086 that outlive the callee itself. This is not a valid attribute for return
1087 values.</dd>
1090 <dd>This indicates that the pointer parameter can be excised using the
1092 attribute for return values.</dd>
1093 </dl>
1095 </div>
1097 <!-- ======================================================================= -->
1100 </div>
1104 <p>Each function may specify a garbage collector name, which is simply a
1105 string:</p>
1108 define void @f() gc "name" { ... }
1109 </pre>
1112 collector which will cause the compiler to alter its output in order to
1113 support the named garbage collection algorithm.</p>
1115 </div>
1117 <!-- ======================================================================= -->
1120 </div>
1124 <p>Function attributes are set to communicate additional information about a
1125 function. Function attributes are considered to be part of the function, not
1126 of the function type, so functions with different parameter attributes can
1127 have the same function type.</p>
1129 <p>Function attributes are simple keywords that follow the type specified. If
1130 multiple attributes are needed, they are space separated. For example:</p>
1133 define void @f() noinline { ... }
1134 define void @f() alwaysinline { ... }
1135 define void @f() alwaysinline optsize { ... }
1136 define void @f() optsize { ... }
1137 </pre>
1139 <dl>
1141 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1142 the backend should forcibly align the stack pointer. Specify the
1143 desired alignment, which must be a power of two, in parentheses.
1146 <dd>This attribute indicates that the inliner should attempt to inline this
1147 function into callers whenever possible, ignoring any active inlining size
1148 threshold for this caller.</dd>
1151 <dd>This attribute indicates that the function should be 'hotpatchable',
1152 meaning the function can be patched and/or hooked even while it is
1153 loaded into memory. On x86, the function prologue will be preceded
1154 by six bytes of padding and will begin with a two-byte instruction.
1155 Most of the functions in the Windows system DLLs in Windows XP SP2 or
1156 higher were compiled in this fashion.</dd>
1159 <dd>This attribute indicates that the source code contained a hint that inlining
1160 this function is desirable (such as the "inline" keyword in C/C++). It
1161 is just a hint; it imposes no requirements on the inliner.</dd>
1164 <dd>This attribute disables prologue / epilogue emission for the function.
1165 This can have very system-specific consequences.</dd>
1171 <dd>This attribute indicates that the inliner should never inline this
1172 function in any situation. This attribute may not be used together with
1176 <dd>This attribute indicates that the code generator should not use a red
1177 zone, even if the target-specific ABI normally permits it.</dd>
1180 <dd>This function attribute indicates that the function never returns
1181 normally. This produces undefined behavior at runtime if the function
1182 ever does dynamically return.</dd>
1185 <dd>This function attribute indicates that the function never returns with an
1186 unwind or exceptional control flow. If the function does unwind, its
1187 runtime behavior is undefined.</dd>
1190 <dd>This attribute suggests that optimization passes and code generator passes
1191 make choices that keep the code size of this function low, and otherwise
1192 do optimizations specifically to reduce code size.</dd>
1195 <dd>This attribute indicates that the function computes its result (or decides
1196 to unwind an exception) based strictly on its arguments, without
1197 dereferencing any pointer arguments or otherwise accessing any mutable
1198 state (e.g. memory, control registers, etc) visible to caller functions.
1199 It does not write through any pointer arguments
1201 changes any state visible to callers. This means that it cannot unwind
1206 <dd>This attribute indicates that the function does not write through any
1208 arguments) or otherwise modify any state (e.g. memory, control registers,
1209 etc) visible to caller functions. It may dereference pointer arguments
1210 and read state that may be set in the caller. A readonly function always
1211 returns the same value (or unwinds an exception identically) when called
1212 with the same set of arguments and global state. It cannot unwind an
1217 <dd>This attribute indicates that the function should emit a stack smashing
1219 the stack before the local variables that's checked upon return from the
1220 function to see if it has been overwritten. A heuristic is used to
1221 determine if a function needs stack protectors or not.<br>
1222 <br>
1229 stack smashing protector. This overrides
1231 <br>
1236 </dl>
1238 </div>
1240 <!-- ======================================================================= -->
1243 </div>
1248 the GCC "file scope inline asm" blocks. These blocks are internally
1249 concatenated by LLVM and treated as a single unit, but may be separated in
1253 module asm "inline asm code goes here"
1254 module asm "more can go here"
1255 </pre>
1257 <p>The strings can contain any character by escaping non-printable characters.
1259 for the number.</p>
1261 <p>The inline asm code is simply printed to the machine code .s file when
1262 assembly code is generated.</p>
1264 </div>
1266 <!-- ======================================================================= -->
1269 </div>
1273 <p>A module may specify a target specific data layout string that specifies how
1274 data is to be laid out in memory. The syntax for the data layout is
1275 simply:</p>
1278 target datalayout = "<i>layout specification</i>"
1279 </pre>
1282 separated by the minus sign character ('-'). Each specification starts with
1283 a letter and may include other information after the letter to define some
1284 aspect of the data layout. The specifications accepted are as follows:</p>
1286 <dl>
1288 <dd>Specifies that the target lays out data in big-endian form. That is, the
1289 bits with the most significance have the lowest address location.</dd>
1292 <dd>Specifies that the target lays out data in little-endian form. That is,
1293 the bits with the least significance have the lowest address
1294 location.</dd>
1303 <dd>This specifies the alignment for an integer type of a given bit
1307 <dd>This specifies the alignment for a vector type of a given bit
1311 <dd>This specifies the alignment for a floating point type of a given bit
1315 targets.
1318 <dd>This specifies the alignment for an aggregate type of a given bit
1322 <dd>This specifies the alignment for a stack object of a given bit
1326 <dd>This specifies a set of native integer widths for the target CPU
1329 this set are considered to support most general arithmetic
1330 operations efficiently.</dd>
1331 </dl>
1333 <p>When constructing the data layout for a given target, LLVM starts with a
1334 default set of specifications which are then (possibly) overridden by the
1336 are given in this list:</p>
1338 <ul>
1353 </ul>
1355 <p>When LLVM is determining the alignment for a given type, it uses the
1356 following rules:</p>
1358 <ol>
1359 <li>If the type sought is an exact match for one of the specifications, that
1360 specification is used.</li>
1362 <li>If no match is found, and the type sought is an integer type, then the
1363 smallest integer type that is larger than the bitwidth of the sought type
1364 is used. If none of the specifications are larger than the bitwidth then
1365 the the largest integer type is used. For example, given the default
1366 specifications above, the i7 type will use the alignment of i8 (next
1367 largest) while both i65 and i256 will use the alignment of i64 (largest
1368 specified).</li>
1370 <li>If no match is found, and the type sought is a vector type, then the
1371 largest vector type that is smaller than the sought vector type will be
1374 </ol>
1376 </div>
1378 <!-- ======================================================================= -->
1381 </div>
1385 <p>Any memory access must be done through a pointer value associated
1386 with an address range of the memory access, otherwise the behavior
1387 is undefined. Pointer values are associated with address ranges
1388 according to the following rules:</p>
1390 <ul>
1391 <li>A pointer value is associated with the addresses associated with
1393 <li>An address of a global variable is associated with the address
1394 range of the variable's storage.</li>
1395 <li>The result value of an allocation instruction is associated with
1396 the address range of the allocated storage.</li>
1397 <li>A null pointer in the default address-space is associated with
1398 no address.</li>
1399 <li>An integer constant other than zero or a pointer value returned
1400 from a function not defined within LLVM may be associated with address
1401 ranges allocated through mechanisms other than those provided by
1402 LLVM. Such ranges shall not overlap with any ranges of addresses
1403 allocated by mechanisms provided by LLVM.</li>
1404 </ul>
1407 to the following rules:</p>
1409 <ul>
1410 <li>A pointer value formed from a
1413 <li>The result value of a
1416 <li>A pointer value formed by an
1418 pointer values that contribute (directly or indirectly) to the
1419 computation of the pointer's value.</li>
1421 </ul>
1425 slightly weaker.</p>
1427 <p>LLVM IR does not associate types with memory. The result type of a
1429 alignment of the memory from which to load, as well as the
1430 interpretation of the value. The first operand type of a
1432 and alignment of the store.</p>
1434 <p>Consequently, type-based alias analysis, aka TBAA, aka
1437 additional information which specialized optimization passes may use
1438 to implement type-based alias analysis.</p>
1440 </div>
1442 <!-- ======================================================================= -->
1445 </div>
1452 The optimizers must not change the number of volatile operations or change their
1453 order of execution relative to other volatile operations. The optimizers
1455 operations. This is not Java's "volatile" and has no cross-thread
1456 synchronization behavior.</p>
1458 </div>
1460 <!-- *********************************************************************** -->
1462 <!-- *********************************************************************** -->
1466 <p>The LLVM type system is one of the most important features of the
1467 intermediate representation. Being typed enables a number of optimizations
1468 to be performed on the intermediate representation directly, without having
1469 to do extra analyses on the side before the transformation. A strong type
1470 system makes it easier to read the generated code and enables novel analyses
1471 and transformations that are not feasible to perform on normal three address
1472 code representations.</p>
1474 </div>
1476 <!-- ======================================================================= -->
1478 Classifications</a> </div>
1485 <tbody>
1487 <tr>
1490 </tr>
1491 <tr>
1494 </tr>
1495 <tr>
1505 </td>
1506 </tr>
1507 <tr>
1515 </tr>
1516 <tr>
1525 </td>
1526 </tr>
1527 </tbody>
1528 </table>
1531 important. Values of these types are the only ones which can be produced by
1532 instructions.</p>
1534 </div>
1536 <!-- ======================================================================= -->
1541 <p>The primitive types are the fundamental building blocks of the LLVM
1542 system.</p>
1544 </div>
1546 <!-- _______________________________________________________________________ -->
1552 <p>The integer type is a very simple type that simply specifies an arbitrary
1553 bit width for the integer type desired. Any bit width from 1 bit to
1557 <pre>
1558 iN
1559 </pre>
1562 value.</p>
1569 </tr>
1573 </tr>
1577 </tr>
1578 </table>
1580 </div>
1582 <!-- _______________________________________________________________________ -->
1587 <table>
1588 <tbody>
1595 </tbody>
1596 </table>
1598 </div>
1600 <!-- _______________________________________________________________________ -->
1606 <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>
1609 <pre>
1610 x86mmx
1611 </pre>
1613 </div>
1615 <!-- _______________________________________________________________________ -->
1624 <pre>
1625 void
1626 </pre>
1628 </div>
1630 <!-- _______________________________________________________________________ -->
1639 <pre>
1640 label
1641 </pre>
1643 </div>
1645 <!-- _______________________________________________________________________ -->
1651 <p>The metadata type represents embedded metadata. No derived types may be
1653 arguments.
1656 <pre>
1657 metadata
1658 </pre>
1660 </div>
1663 <!-- ======================================================================= -->
1668 <p>The real power in LLVM comes from the derived types in the system. This is
1669 what allows a programmer to represent arrays, functions, pointers, and other
1670 useful types. Each of these types contain one or more element types which
1671 may be a primitive type, or another derived type. For example, it is
1672 possible to have a two dimensional array, using an array as the element type
1673 of another array.</p>
1676 </div>
1678 <!-- _______________________________________________________________________ -->
1683 <p>Aggregate Types are a subset of derived types that can contain multiple
1686 aggregate types.</p>
1688 </div>
1690 <!-- _______________________________________________________________________ -->
1696 <p>The array type is a very simple derived type that arranges elements
1697 sequentially in memory. The array type requires a size (number of elements)
1698 and an underlying data type.</p>
1701 <pre>
1703 </pre>
1706 be any type with a size.</p>
1713 </tr>
1717 </tr>
1721 </tr>
1722 </table>
1728 </tr>
1732 </tr>
1736 </tr>
1737 </table>
1739 <p>There is no restriction on indexing beyond the end of the array implied by
1740 a static type (though there are restrictions on indexing beyond the bounds
1741 of an allocated object in some cases). This means that single-dimension
1742 'variable sized array' addressing can be implemented in LLVM with a zero
1743 length array type. An implementation of 'pascal style arrays' in LLVM could
1746 </div>
1748 <!-- _______________________________________________________________________ -->
1754 <p>The function type can be thought of as a function signature. It consists of
1755 a return type and a list of formal parameter types. The return type of a
1756 function type is a first class type or a void type.</p>
1759 <pre>
1761 </pre>
1765 which indicates that the function takes a variable number of arguments.
1766 Variable argument functions can access their arguments with
1776 </td>
1779 </tt></td>
1783 </td>
1789 LLVM.
1790 </td>
1795 </td>
1796 </tr>
1797 </table>
1799 </div>
1801 <!-- _______________________________________________________________________ -->
1807 <p>The structure type is used to represent a collection of data members together
1808 in memory. The packing of the field types is defined to match the ABI of the
1809 underlying processor. The elements of a structure may be any type that has a
1810 size.</p>
1815 Structures in registers are accessed using the
1819 <pre>
1821 </pre>
1834 </tr>
1835 </table>
1837 </div>
1839 <!-- _______________________________________________________________________ -->
1841 </div>
1846 <p>The packed structure type is used to represent a collection of data members
1847 together in memory. There is no padding between fields. Further, the
1848 alignment of a packed structure is 1 byte. The elements of a packed
1849 structure may be any type that has a size.</p>
1856 <pre>
1858 </pre>
1872 </tr>
1873 </table>
1875 </div>
1877 <!-- _______________________________________________________________________ -->
1883 <p>The pointer type is used to specify memory locations.
1884 Pointers are commonly used to reference objects in memory.</p>
1886 <p>Pointer types may have an optional address space attribute defining the
1887 numbered address space where the pointed-to object resides. The default
1888 address space is number zero. The semantics of non-zero address
1889 spaces are target-specific.</p>
1895 <pre>
1897 </pre>
1905 </tr>
1911 </tr>
1916 </tr>
1917 </table>
1919 </div>
1921 <!-- _______________________________________________________________________ -->
1927 <p>A vector type is a simple derived type that represents a vector of elements.
1928 Vector types are used when multiple primitive data are operated in parallel
1929 using a single instruction (SIMD). A vector type requires a size (number of
1930 elements) and an underlying primitive data type. Vector types are considered
1934 <pre>
1936 </pre>
1939 may be any integer or floating point type. Vectors of size zero are not
1940 allowed, and pointers are not allowed as the element type.</p>
1947 </tr>
1951 </tr>
1955 </tr>
1956 </table>
1958 </div>
1960 <!-- _______________________________________________________________________ -->
1965 <p>Opaque types are used to represent unknown types in the system. This
1966 corresponds (for example) to the C notion of a forward declared structure
1967 type. In LLVM, opaque types can eventually be resolved to any type (not just
1968 a structure type).</p>
1971 <pre>
1972 opaque
1973 </pre>
1980 </tr>
1981 </table>
1983 </div>
1985 <!-- ======================================================================= -->
1988 </div>
1994 requiring it to have a name. For instance, a structure declaration may
1995 contain a pointer to any of the types it is lexically a member of. Example
1996 of up references (with their equivalent as named type declarations)
1997 include:</p>
1999 <pre>
2000 { \2 * } %x = type { %x* }
2001 { \2 }* %y = type { %y }*
2002 \1* %z = type %z*
2003 </pre>
2005 <p>An up reference is needed by the asmprinter for printing out cyclic types
2006 when there is no declared name for a type in the cycle. Because the
2007 asmprinter does not want to print out an infinite type string, it needs a
2008 syntax to handle recursive types that have no names (all names are optional
2009 in llvm IR).</p>
2012 <pre>
2014 </pre>
2023 </tr>
2027 structure.</td>
2028 </tr>
2029 </table>
2031 </div>
2033 <!-- *********************************************************************** -->
2035 <!-- *********************************************************************** -->
2039 <p>LLVM has several different basic types of constants. This section describes
2040 them all and their syntax.</p>
2042 </div>
2044 <!-- ======================================================================= -->
2049 <dl>
2057 with integer types.</dd>
2061 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2062 notation (see below). The assembler requires the exact decimal value of a
2063 floating-point constant. For example, the assembler accepts 1.25 but
2070 </dl>
2072 <p>The one non-intuitive notation for constants is the hexadecimal form of
2073 floating point constants. For example, the form '<tt>double
2076 constants are required (and the only time that they are generated by the
2077 disassembler) is when a floating point constant must be emitted but it cannot
2078 be represented as a decimal floating point number in a reasonable number of
2079 digits. For example, NaN's, infinities, and other special values are
2080 represented in their IEEE hexadecimal format so that assembly and disassembly
2081 do not cause any bits to change in the constants.</p>
2083 <p>When using the hexadecimal form, constants of types float and double are
2084 represented using the 16-digit form shown above (which matches the IEEE754
2085 representation for double); float values must, however, be exactly
2086 representable as IEE754 single precision. Hexadecimal format is always used
2087 for long double, and there are three forms of long double. The 80-bit format
2089 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2092 currently supported target uses this format. Long doubles will only work if
2093 they match the long double format on your target. All hexadecimal formats
2094 are big-endian (sign bit at the left).</p>
2097 </div>
2099 <!-- ======================================================================= -->
2103 </div>
2107 <p>Complex constants are a (potentially recursive) combination of simple
2108 constants and smaller complex constants.</p>
2110 <dl>
2112 <dd>Structure constants are represented with notation similar to structure
2113 type definitions (a comma separated list of elements, surrounded by braces
2117 the number and types of elements must match those specified by the
2118 type.</dd>
2121 <dd>Array constants are represented with notation similar to array type
2122 definitions (a comma separated list of elements, surrounded by square
2125 the number and types of elements must match those specified by the
2126 type.</dd>
2129 <dd>Vector constants are represented with notation similar to vector type
2130 definitions (a comma separated list of elements, surrounded by
2132 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2134 elements must match those specified by the type.</dd>
2140 This is often used to avoid having to print large zero initializers
2141 (e.g. for large arrays) and is always exactly equivalent to using explicit
2142 zero initializers.</dd>
2145 <dd>A metadata node is a structure-like constant with
2148 be interpreted as part of the instruction stream, metadata is a place to
2149 attach additional information such as debug info.</dd>
2150 </dl>
2152 </div>
2154 <!-- ======================================================================= -->
2157 </div>
2163 (link-time) constants. These constants are explicitly referenced when
2166 legal LLVM file:</p>
2169 @X = global i32 17
2170 @Y = global i32 42
2171 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2172 </pre>
2174 </div>
2176 <!-- ======================================================================= -->
2181 indicates that the user of the value may receive an unspecified bit-pattern.
2185 <p>Undefined values are useful because they indicate to the compiler that the
2186 program is well defined no matter what value is used. This gives the
2187 compiler more freedom to optimize. Here are some examples of (potentially
2188 surprising) transformations that are valid (in pseudo IR):</p>
2192 %A = add %X, undef
2193 %B = sub %X, undef
2194 %C = xor %X, undef
2195 Safe:
2196 %A = undef
2197 %B = undef
2198 %C = undef
2199 </pre>
2201 <p>This is safe because all of the output bits are affected by the undef bits.
2202 Any output bit can have a zero or one depending on the input bits.</p>
2205 %A = or %X, undef
2206 %B = and %X, undef
2207 Safe:
2208 %A = -1
2209 %B = 0
2210 Unsafe:
2211 %A = undef
2212 %B = undef
2213 </pre>
2215 <p>These logical operations have bits that are not always affected by the input.
2226 %A = select undef, %X, %Y
2227 %B = select undef, 42, %Y
2228 %C = select %X, %Y, undef
2229 Safe:
2230 %A = %X (or %Y)
2231 %B = 42 (or %Y)
2232 %C = %Y
2233 Unsafe:
2234 %A = undef
2235 %B = undef
2236 %C = undef
2237 </pre>
2246 eliminated.</p>
2249 %A = xor undef, undef
2251 %B = undef
2252 %C = xor %B, %B
2254 %D = undef
2255 %E = icmp lt %D, 4
2256 %F = icmp gte %D, 4
2258 Safe:
2259 %A = undef
2260 %B = undef
2261 %C = undef
2262 %D = undef
2263 %E = undef
2264 %F = undef
2265 </pre>
2268 necessarily the same. This can be surprising to people (and also matches C
2269 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2272 its value over its "live range". This is true because the variable doesn't
2274 from arbitrary registers that happen to be around when needed, so the value
2276 need to have the same semantics or the core LLVM "replace all uses with"
2277 concept would not hold.</p>
2280 %A = fdiv undef, %X
2281 %B = fdiv %X, undef
2282 Safe:
2283 %A = undef
2284 b: unreachable
2285 </pre>
2292 defined on SNaN's. However, in the second example, we can make a more
2294 arbitrary value, we are allowed to assume that it could be zero. Since a
2296 the operation does not execute at all. This allows us to delete the divide and
2297 all code after it. Because the undefined operation "can't happen", the
2298 optimizer can assume that it occurs in dead code.</p>
2301 a: store undef -> %X
2302 b: store %X -> undef
2303 Safe:
2305 b: unreachable
2306 </pre>
2309 undefined value can be assumed to not have any effect; we can assume that the
2310 value is overwritten with bits that happen to match what was already there.
2312 memory, therefore, it has undefined behavior.</p>
2314 </div>
2316 <!-- ======================================================================= -->
2321 instead of representing an unspecified bit pattern, they represent the
2322 fact that an instruction or constant expression which cannot evoke side
2323 effects has nevertheless detected a condition which results in undefined
2324 behavior.</p>
2326 <p>There is currently no way of representing a trap value in the IR; they
2327 only exist when produced by operations such as
2332 <ul>
2334 their operands.</li>
2337 to their dynamic predecessor basic block.</li>
2339 <li>Function arguments depend on the corresponding actual argument values in
2340 the dynamic callers of their functions.</li>
2344 control back to them.</li>
2348 or exception-throwing call instructions that dynamically transfer control
2349 back to them.</li>
2351 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2352 referenced memory addresses, following the order in the IR
2353 (including loads and stores implied by intrinsics such as
2356 <!-- TODO: In the case of multiple threads, this only applies if the store
2357 "happens-before" the load or store. -->
2359 <!-- TODO: floating-point exception state -->
2361 <li>An instruction with externally visible side effects depends on the most
2362 recent preceding instruction with externally visible side effects, following
2363 the order in the IR. (This includes
2368 if the terminator instruction has multiple successors and the instruction
2369 is always executed when control transfers to one of the successors, and
2370 may not be executed when control is transfered to another.</li>
2374 </ul>
2376 <p>Whenever a trap value is generated, all values which depend on it evaluate
2377 to trap. If they have side effects, the evoke their side effects as if each
2378 operand with a trap value were undef. If they have externally-visible side
2379 effects, the behavior is undefined.</p>
2384 entry:
2387 %trap_yet_again = getelementptr i32* @h, i32 %still_trap
2388 store i32 0, i32* %trap_yet_again ; undefined behavior
2390 store i32 %trap, i32* @g ; Trap value conceptually stored to memory.
2391 %trap2 = load i32* @g ; Returns a trap value, not just undef.
2393 volatile store i32 %trap, i32* @g ; External observation; undefined behavior.
2395 %narrowaddr = bitcast i32* @g to i16*
2396 %wideaddr = bitcast i32* @g to i64*
2397 %trap3 = load 16* %narrowaddr ; Returns a trap value.
2398 %trap4 = load i64* %widaddr ; Returns a trap value.
2400 %cmp = icmp i32 slt %trap, 0 ; Returns a trap value.
2401 %br i1 %cmp, %true, %end ; Branch to either destination.
2403 true:
2404 volatile store i32 0, i32* @g ; This is control-dependent on %cmp, so
2405 ; it has undefined behavior.
2406 br label %end
2408 end:
2410 ; Both edges into this PHI are
2411 ; control-dependent on %cmp, so this
2412 ; always results in a trap value.
2414 volatile store i32 0, i32* @g ; %end is control-equivalent to %entry
2415 ; so this is defined (ignoring earlier
2416 ; undefined behavior in this example).
2417 </pre>
2419 </div>
2421 <!-- ======================================================================= -->
2423 Blocks</a></div>
2429 basic block in the specified function, and always has an i8* type. Taking
2430 the address of the entry block is illegal.</p>
2432 <p>This value only has defined behavior when used as an operand to the
2434 comparisons against null. Pointer equality tests between labels addresses
2435 results in undefined behavior — though, again, comparison against null
2436 is ok, and no label is equal to the null pointer. This may be passed around
2437 as an opaque pointer sized value as long as the bits are not inspected. This
2440 instruction.</p>
2442 <p>Finally, some targets may provide defined semantics when using the value as
2443 the operand to an inline assembly, but that is target specific.</p>
2445 </div>
2448 <!-- ======================================================================= -->
2450 </div>
2454 <p>Constant expressions are used to allow expressions involving other constants
2455 to be used as constants. Constant expressions may be of
2457 operation that does not have side effects (e.g. load and call are not
2458 supported). The following is the syntax for constant expressions:</p>
2460 <dl>
2462 <dd>Truncate a constant to another type. The bit size of CST must be larger
2463 than the bit size of TYPE. Both types must be integers.</dd>
2466 <dd>Zero extend a constant to another type. The bit size of CST must be
2467 smaller than the bit size of TYPE. Both types must be integers.</dd>
2470 <dd>Sign extend a constant to another type. The bit size of CST must be
2471 smaller than the bit size of TYPE. Both types must be integers.</dd>
2474 <dd>Truncate a floating point constant to another floating point type. The
2475 size of CST must be larger than the size of TYPE. Both types must be
2476 floating point.</dd>
2479 <dd>Floating point extend a constant to another type. The size of CST must be
2480 smaller or equal to the size of TYPE. Both types must be floating
2481 point.</dd>
2484 <dd>Convert a floating point constant to the corresponding unsigned integer
2485 constant. TYPE must be a scalar or vector integer type. CST must be of
2486 scalar or vector floating point type. Both CST and TYPE must be scalars,
2487 or vectors of the same number of elements. If the value won't fit in the
2488 integer type, the results are undefined.</dd>
2491 <dd>Convert a floating point constant to the corresponding signed integer
2492 constant. TYPE must be a scalar or vector integer type. CST must be of
2493 scalar or vector floating point type. Both CST and TYPE must be scalars,
2494 or vectors of the same number of elements. If the value won't fit in the
2495 integer type, the results are undefined.</dd>
2498 <dd>Convert an unsigned integer constant to the corresponding floating point
2499 constant. TYPE must be a scalar or vector floating point type. CST must be
2500 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2501 vectors of the same number of elements. If the value won't fit in the
2502 floating point type, the results are undefined.</dd>
2505 <dd>Convert a signed integer constant to the corresponding floating point
2506 constant. TYPE must be a scalar or vector floating point type. CST must be
2507 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2508 vectors of the same number of elements. If the value won't fit in the
2509 floating point type, the results are undefined.</dd>
2512 <dd>Convert a pointer typed constant to the corresponding integer constant
2518 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2519 type. CST must be of integer type. The CST value is zero extended,
2520 truncated, or unchanged to make it fit in a pointer size. This one is
2524 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2532 instruction, the index list may have zero or more indexes, which are
2546 constants.</dd>
2550 constants.</dd>
2554 constants.</dd>
2558 constants. The index list is interpreted in a similar manner as indices in
2560 index value must be specified.</dd>
2564 constants. The index list is interpreted in a similar manner as indices in
2566 index value must be specified.</dd>
2569 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2572 on operands are the same as those for the corresponding instruction
2573 (e.g. no bitwise operations on floating point values are allowed).</dd>
2574 </dl>
2576 </div>
2578 <!-- *********************************************************************** -->
2580 <!-- *********************************************************************** -->
2582 <!-- ======================================================================= -->
2585 </div>
2589 <p>LLVM supports inline assembler expressions (as opposed
2591 a special value. This value represents the inline assembler as a string
2592 (containing the instructions to emit), a list of operand constraints (stored
2593 as a string), a flag that indicates whether or not the inline asm
2594 expression has side effects, and a flag indicating whether the function
2595 containing the asm needs to align its stack conservatively. An example
2596 inline assembler expression is:</p>
2600 </pre>
2604 have:</p>
2608 </pre>
2610 <p>Inline asms with side effects not visible in the constraint list must be
2611 marked as having side effects. This is done through the use of the
2616 </pre>
2618 <p>In some cases inline asms will contain code that will not work unless the
2619 stack is aligned in some way, such as calls or SSE instructions on x86,
2620 yet will not contain code that does that alignment within the asm.
2621 The compiler should make conservative assumptions about what the asm might
2622 contain and should generate its usual stack alignment code in the prologue
2627 </pre>
2630 first.</p>
2632 <p>TODO: The format of the asm and constraints string still need to be
2633 documented here. Constraints on what can be done (e.g. duplication, moving,
2634 etc need to be documented). This is probably best done by reference to
2635 another document that covers inline asm from a holistic perspective.</p>
2636 </div>
2640 </div>
2645 attached to it that contains a list of constant integers. If present, the
2646 code generator will use the integer as the location cookie value when report
2647 errors through the LLVMContext error reporting mechanisms. This allows a
2648 front-end to correlate backend errors that occur with inline asm back to the
2649 source code that produced it. For example:</p>
2653 ...
2655 </pre>
2657 <p>It is up to the front-end to make sense of the magic numbers it places in the
2658 IR. If the MDNode contains multiple constants, the code generator will use
2659 the one that corresponds to the line of the asm that the error occurs on.</p>
2661 </div>
2663 <!-- ======================================================================= -->
2665 Strings</a>
2666 </div>
2670 <p>LLVM IR allows metadata to be attached to instructions in the program that
2671 can convey extra information about the code to the optimizers and code
2672 generator. One example application of metadata is source-level debug
2673 information. There are two metadata primitives: strings and nodes. All
2677 <p>A metadata string is a string surrounded by double quotes. It can contain
2681 <p>Metadata nodes are represented with notation similar to structure constants
2682 (a comma separated list of elements, surrounded by braces and preceded by an
2687 metadata nodes, which can be looked up in the module symbol table. For
2688 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2691 function is using two metadata arguments.</p>
2695 </pre>
2702 </pre>
2703 </div>
2706 <!-- *********************************************************************** -->
2709 </div>
2710 <!-- *********************************************************************** -->
2713 code generation or other IR semantics. These are documented here. All globals
2714 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2715 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2716 by LLVM.</p>
2718 <!-- ======================================================================= -->
2721 </div>
2727 pointers to global variables and functions which may optionally have a pointer
2728 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2730 <pre>
2731 @X = global i8 4
2732 @Y = global i32 123
2734 @llvm.used = appending global [2 x i8*] [
2735 i8* @X,
2736 i8* bitcast (i32* @Y to i8*)
2737 ], section "llvm.metadata"
2738 </pre>
2741 compiler, assembler, and linker are required to treat the symbol as if there is
2742 a reference to the global that it cannot see. For example, if a variable has
2744 list, it cannot be deleted. This is commonly used to represent references from
2745 inline asms and other things the compiler cannot "see", and corresponds to
2748 <p>On some targets, the code generator must emit a directive to the assembler or
2749 object file to prevent the assembler and linker from molesting the symbol.</p>
2751 </div>
2753 <!-- ======================================================================= -->
2756 </div>
2762 touching the symbol. On targets that support it, this allows an intelligent
2763 linker to optimize references to the symbol without being impeded as it would be
2766 <p>This is a rare construct that should only be used in rare circumstances, and
2767 should not be exposed to source languages.</p>
2769 </div>
2771 <!-- ======================================================================= -->
2774 </div>
2777 <pre>
2778 %0 = type { i32, void ()* }
2780 </pre>
2781 <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.
2782 </p>
2784 </div>
2786 <!-- ======================================================================= -->
2789 </div>
2792 <pre>
2793 %0 = type { i32, void ()* }
2795 </pre>
2797 <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.
2798 </p>
2800 </div>
2803 <!-- *********************************************************************** -->
2805 <!-- *********************************************************************** -->
2809 <p>The LLVM instruction set consists of several different classifications of
2816 </div>
2818 <!-- ======================================================================= -->
2820 Instructions</a> </div>
2825 in a program ends with a "Terminator" instruction, which indicates which
2826 block should be executed after the current block is finished. These
2828 control flow, not values (the one exception being the
2831 <p>There are seven different terminator instructions: the
2840 </div>
2842 <!-- _______________________________________________________________________ -->
2844 Instruction</a> </div>
2849 <pre>
2852 </pre>
2856 a value) from a function back to the caller.</p>
2859 value and then causes control flow, and one that just causes control flow to
2860 occur.</p>
2864 return value. The type of the return value must be a
2869 value or a return value with a type that does not match its type, or if it
2871 return value.</p>
2875 the calling function's context. If the caller is a
2877 instruction after the call. If the caller was an
2879 the beginning of the "normal" destination block. If the instruction returns
2880 a value, that value shall set the call or invoke instruction's return
2881 value.</p>
2884 <pre>
2888 </pre>
2890 </div>
2891 <!-- _______________________________________________________________________ -->
2897 <pre>
2898 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2899 </pre>
2903 different basic block in the current function. There are two forms of this
2904 instruction, corresponding to a conditional branch and an unconditional
2905 branch.</p>
2911 target.</p>
2920 <pre>
2921 Test:
2923 br i1 %cond, label %IfEqual, label %IfUnequal
2924 IfEqual:
2926 IfUnequal:
2928 </pre>
2930 </div>
2932 <!-- _______________________________________________________________________ -->
2935 </div>
2940 <pre>
2941 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2942 </pre>
2947 instruction, allowing a branch to occur to one of many possible
2948 destinations.</p>
2954 The table is not allowed to contain duplicate constant entries.</p>
2959 is searched for the given value. If the value is found, control flow is
2960 transferred to the corresponding destination; otherwise, control flow is
2961 transferred to the default destination.</p>
2964 <p>Depending on properties of the target machine and the particular
2966 different ways. For example, it could be generated as a series of chained
2967 conditional branches or with a lookup table.</p>
2970 <pre>
2973 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2976 switch i32 0, label %dest [ ]
2979 switch i32 %val, label %otherwise [ i32 0, label %onzero
2980 i32 1, label %onone
2981 i32 2, label %ontwo ]
2982 </pre>
2984 </div>
2987 <!-- _______________________________________________________________________ -->
2990 </div>
2995 <pre>
2997 </pre>
3002 within the current function, whose address is specified by
3009 rest of the arguments indicate the full set of possible destinations that the
3010 address may point to. Blocks are allowed to occur multiple times in the
3011 destination list, though this isn't particularly useful.</p>
3013 <p>This destination list is required so that dataflow analysis has an accurate
3014 understanding of the CFG.</p>
3018 <p>Control transfers to the block specified in the address argument. All
3019 possible destination blocks must be listed in the label list, otherwise this
3020 instruction has undefined behavior. This implies that jumps to labels
3021 defined in other functions have undefined behavior as well.</p>
3028 <pre>
3029 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3030 </pre>
3032 </div>
3035 <!-- _______________________________________________________________________ -->
3038 </div>
3043 <pre>
3044 <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>]
3046 </pre>
3050 function, with the possibility of control flow transfer to either the
3053 control flow will return to the "normal" label. If the callee (or any
3055 instruction, control is interrupted and continued at the dynamically nearest
3061 <ol>
3063 convention</a> the call should use. If none is specified, the call
3064 defaults to using C calling conventions.</li>
3071 function value being invoked. In most cases, this is a direct function
3073 off an arbitrary pointer to function value.</li>
3076 function to be invoked. </li>
3079 signature argument types and parameter attributes. All arguments must be
3081 signature indicates the function accepts a variable number of arguments,
3082 the extra arguments can be specified.</li>
3093 </ol>
3096 <p>This instruction is designed to operate as a standard
3098 primary difference is that it establishes an association with a label, which
3099 is used by the runtime library to unwind the stack.</p>
3101 <p>This instruction is used in languages with destructors to ensure that proper
3103 exception. Additionally, this is important for implementation of
3106 <p>For the purposes of the SSA form, the definition of the value returned by the
3108 block to the "normal" label. If the callee unwinds then no return value is
3109 available.</p>
3111 <p>Note that the code generator does not yet completely support unwind, and
3112 that the invoke/unwind semantics are likely to change in future versions.</p>
3115 <pre>
3116 %retval = invoke i32 @Test(i32 15) to label %Continue
3120 </pre>
3122 </div>
3124 <!-- _______________________________________________________________________ -->
3127 Instruction</a> </div>
3132 <pre>
3133 unwind
3134 </pre>
3138 at the first callee in the dynamic call stack which used
3140 This is primarily used to implement exception handling.</p>
3144 immediately halt. The dynamic call stack is then searched for the
3146 Once found, execution continues at the "exceptional" destination block
3148 instruction in the dynamic call chain, undefined behavior results.</p>
3150 <p>Note that the code generator does not yet completely support unwind, and
3151 that the invoke/unwind semantics are likely to change in future versions.</p>
3153 </div>
3155 <!-- _______________________________________________________________________ -->
3158 Instruction</a> </div>
3163 <pre>
3164 unreachable
3165 </pre>
3169 instruction is used to inform the optimizer that a particular portion of the
3170 code is not reachable. This can be used to indicate that the code after a
3171 no-return function cannot be reached, and other facts.</p>
3176 </div>
3178 <!-- ======================================================================= -->
3183 <p>Binary operators are used to do most of the computation in a program. They
3184 require two operands of the same type, execute an operation on them, and
3185 produce a single value. The operands might represent multiple data, as is
3187 has the same type as its operands.</p>
3191 </div>
3193 <!-- _______________________________________________________________________ -->
3196 </div>
3201 <pre>
3206 </pre>
3214 integer values. Both arguments must have identical types.</p>
3219 <p>If the sum has unsigned overflow, the result returned is the mathematical
3222 <p>Because LLVM integers use a two's complement representation, this instruction
3223 is appropriate for both signed and unsigned integers.</p>
3229 respectively, occurs.</p>
3232 <pre>
3234 </pre>
3236 </div>
3238 <!-- _______________________________________________________________________ -->
3241 </div>
3246 <pre>
3248 </pre>
3256 floating point values. Both arguments must have identical types.</p>
3262 <pre>
3264 </pre>
3266 </div>
3268 <!-- _______________________________________________________________________ -->
3271 </div>
3276 <pre>
3281 </pre>
3285 operands.</p>
3289 representations.</p>
3294 integer values. Both arguments must have identical types.</p>
3299 <p>If the difference has unsigned overflow, the result returned is the
3301 result.</p>
3303 <p>Because LLVM integers use a two's complement representation, this instruction
3304 is appropriate for both signed and unsigned integers.</p>
3310 respectively, occurs.</p>
3313 <pre>
3316 </pre>
3318 </div>
3320 <!-- _______________________________________________________________________ -->
3323 </div>
3328 <pre>
3330 </pre>
3334 operands.</p>
3338 representations.</p>
3343 floating point values. Both arguments must have identical types.</p>
3349 <pre>
3352 </pre>
3354 </div>
3356 <!-- _______________________________________________________________________ -->
3359 </div>
3364 <pre>
3369 </pre>
3377 integer values. Both arguments must have identical types.</p>
3382 <p>If the result of the multiplication has unsigned overflow, the result
3384 width of the result.</p>
3386 <p>Because LLVM integers use a two's complement representation, and the result
3387 is the same width as the operands, this instruction returns the correct
3388 result for both signed and unsigned integers. If a full product
3390 be sign-extended or zero-extended as appropriate to the width of the full
3391 product.</p>
3397 respectively, occurs.</p>
3400 <pre>
3402 </pre>
3404 </div>
3406 <!-- _______________________________________________________________________ -->
3409 </div>
3414 <pre>
3416 </pre>
3424 floating point values. Both arguments must have identical types.</p>
3430 <pre>
3432 </pre>
3434 </div>
3436 <!-- _______________________________________________________________________ -->
3438 </a></div>
3443 <pre>
3445 </pre>
3453 values. Both arguments must have identical types.</p>
3458 <p>Note that unsigned integer division and signed integer division are distinct
3464 <pre>
3466 </pre>
3468 </div>
3470 <!-- _______________________________________________________________________ -->
3472 </a> </div>
3477 <pre>
3480 </pre>
3488 values. Both arguments must have identical types.</p>
3491 <p>The value produced is the signed integer quotient of the two operands rounded
3492 towards zero.</p>
3494 <p>Note that signed integer division and unsigned integer division are distinct
3497 <p>Division by zero leads to undefined behavior. Overflow also leads to
3498 undefined behavior; this is a rare case, but can occur, for example, by doing
3503 be rounded.</p>
3506 <pre>
3508 </pre>
3510 </div>
3512 <!-- _______________________________________________________________________ -->
3514 Instruction</a> </div>
3519 <pre>
3521 </pre>
3529 floating point values. Both arguments must have identical types.</p>
3535 <pre>
3537 </pre>
3539 </div>
3541 <!-- _______________________________________________________________________ -->
3543 </div>
3548 <pre>
3550 </pre>
3554 division of its two arguments.</p>
3559 values. Both arguments must have identical types.</p>
3563 This instruction always performs an unsigned division to get the
3564 remainder.</p>
3566 <p>Note that unsigned integer remainder and signed integer remainder are
3572 <pre>
3574 </pre>
3576 </div>
3578 <!-- _______________________________________________________________________ -->
3581 </div>
3586 <pre>
3588 </pre>
3592 division of its two operands. This instruction can also take
3594 elements must be integers.</p>
3599 values. Both arguments must have identical types.</p>
3605 a value. For more information about the difference,
3607 Math Forum</a>. For a table of how this is implemented in various languages,
3611 <p>Note that signed integer remainder and unsigned integer remainder are
3614 <p>Taking the remainder of a division by zero leads to undefined behavior.
3615 Overflow also leads to undefined behavior; this is a rare case, but can
3616 occur, for example, by taking the remainder of a 32-bit division of
3618 lets srem be implemented using instructions that return both the result of
3619 the division and the remainder.)</p>
3622 <pre>
3624 </pre>
3626 </div>
3628 <!-- _______________________________________________________________________ -->
3635 <pre>
3637 </pre>
3641 its two operands.</p>
3646 floating point values. Both arguments must have identical types.</p>
3650 has the same sign as the dividend.</p>
3653 <pre>
3655 </pre>
3657 </div>
3659 <!-- ======================================================================= -->
3661 Operations</a> </div>
3665 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3666 program. They are generally very efficient instructions and can commonly be
3667 strength reduced from other instructions. They require two operands of the
3668 same type, execute an operation on them, and produce a single value. The
3669 resulting value is the same type as its operands.</p>
3671 </div>
3673 <!-- _______________________________________________________________________ -->
3675 Instruction</a> </div>
3680 <pre>
3682 </pre>
3686 a specified number of bits.</p>
3696 is (statically or dynamically) negative or equal to or larger than the number
3702 <pre>
3707 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3708 </pre>
3710 </div>
3712 <!-- _______________________________________________________________________ -->
3714 Instruction</a> </div>
3719 <pre>
3721 </pre>
3725 operand shifted to the right a specified number of bits with zero fill.</p>
3733 <p>This instruction always performs a logical shift right operation. The most
3734 significant bits of the result will be filled with zero bits after the shift.
3741 <pre>
3747 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3748 </pre>
3750 </div>
3752 <!-- _______________________________________________________________________ -->
3754 Instruction</a> </div>
3758 <pre>
3760 </pre>
3764 operand shifted to the right a specified number of bits with sign
3765 extension.</p>
3773 <p>This instruction always performs an arithmetic shift right operation, The
3774 most significant bits of the result will be filled with the sign bit
3781 <pre>
3787 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3788 </pre>
3790 </div>
3792 <!-- _______________________________________________________________________ -->
3794 Instruction</a> </div>
3799 <pre>
3801 </pre>
3805 operands.</p>
3810 values. Both arguments must have identical types.</p>
3816 <tbody>
3817 <tr>
3821 </tr>
3822 <tr>
3826 </tr>
3827 <tr>
3831 </tr>
3832 <tr>
3836 </tr>
3837 <tr>
3841 </tr>
3842 </tbody>
3843 </table>
3846 <pre>
3850 </pre>
3851 </div>
3852 <!-- _______________________________________________________________________ -->
3858 <pre>
3860 </pre>
3864 two operands.</p>
3869 values. Both arguments must have identical types.</p>
3875 <tbody>
3876 <tr>
3880 </tr>
3881 <tr>
3885 </tr>
3886 <tr>
3890 </tr>
3891 <tr>
3895 </tr>
3896 <tr>
3900 </tr>
3901 </tbody>
3902 </table>
3905 <pre>
3909 </pre>
3911 </div>
3913 <!-- _______________________________________________________________________ -->
3915 Instruction</a> </div>
3920 <pre>
3922 </pre>
3932 values. Both arguments must have identical types.</p>
3938 <tbody>
3939 <tr>
3943 </tr>
3944 <tr>
3948 </tr>
3949 <tr>
3953 </tr>
3954 <tr>
3958 </tr>
3959 <tr>
3963 </tr>
3964 </tbody>
3965 </table>
3968 <pre>
3973 </pre>
3975 </div>
3977 <!-- ======================================================================= -->
3980 </div>
3984 <p>LLVM supports several instructions to represent vector operations in a
3985 target-independent manner. These instructions cover the element-access and
3986 vector-specific operations needed to process vectors effectively. While LLVM
3987 does directly support these vector operations, many sophisticated algorithms
3988 will want to use target-specific intrinsics to take full advantage of a
3989 specific target.</p>
3991 </div>
3993 <!-- _______________________________________________________________________ -->
3996 </div>
4001 <pre>
4002 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4003 </pre>
4007 from a vector at a specified index.</p>
4013 indicating the position from which to extract the element. The index may be
4014 a variable.</p>
4017 <p>The result is a scalar of the same type as the element type of
4020 results are undefined.</p>
4023 <pre>
4025 </pre>
4027 </div>
4029 <!-- _______________________________________________________________________ -->
4032 </div>
4037 <pre>
4038 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4039 </pre>
4043 vector at a specified index.</p>
4048 whose type must equal the element type of the first operand. The third
4049 operand is an index indicating the position at which to insert the value.
4050 The index may be a variable.</p>
4056 results are undefined.</p>
4059 <pre>
4060 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4061 </pre>
4063 </div>
4065 <!-- _______________________________________________________________________ -->
4068 </div>
4073 <pre>
4074 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4075 </pre>
4079 from two input vectors, returning a vector with the same element type as the
4080 input and length that is the same as the shuffle mask.</p>
4084 with types that match each other. The third argument is a shuffle mask whose
4085 element type is always 'i32'. The result of the instruction is a vector
4086 whose length is the same as the shuffle mask and whose element type is the
4087 same as the element type of the first two operands.</p>
4089 <p>The shuffle mask operand is required to be a constant vector with either
4090 constant integer or undef values.</p>
4093 <p>The elements of the two input vectors are numbered from left to right across
4094 both of the vectors. The shuffle mask operand specifies, for each element of
4095 the result vector, which element of the two input vectors the result element
4096 gets. The element selector may be undef (meaning "don't care") and the
4097 second operand may be undef if performing a shuffle from only one vector.</p>
4100 <pre>
4104 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4108 <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>
4109 </pre>
4111 </div>
4113 <!-- ======================================================================= -->
4116 </div>
4120 <p>LLVM supports several instructions for working with
4123 </div>
4125 <!-- _______________________________________________________________________ -->
4128 </div>
4133 <pre>
4135 </pre>
4145 specify which value to extract in a similar manner as indices in a
4148 <ul>
4149 <li>Since the value being indexed is not a pointer, the first index is
4150 omitted and assumed to be zero.</li>
4152 <li>Not only struct indices but also array indices must be in
4153 bounds.</li>
4154 </ul>
4157 <p>The result is the value at the position in the aggregate specified by the
4158 index operands.</p>
4161 <pre>
4163 </pre>
4165 </div>
4167 <!-- _______________________________________________________________________ -->
4170 </div>
4175 <pre>
4176 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx> <i>; yields <aggregate type></i>
4177 </pre>
4187 value to insert. The following operands are constant indices indicating
4188 the position at which to insert the value in a similar manner as indices in a
4190 value to insert must have the same type as the value identified by the
4191 indices.</p>
4199 <pre>
4202 </pre>
4204 </div>
4207 <!-- ======================================================================= -->
4210 </div>
4214 <p>A key design point of an SSA-based representation is how it represents
4215 memory. In LLVM, no memory locations are in SSA form, which makes things
4216 very simple. This section describes how to read, write, and allocate
4217 memory in LLVM.</p>
4219 </div>
4221 <!-- _______________________________________________________________________ -->
4224 </div>
4229 <pre>
4230 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4231 </pre>
4235 currently executing function, to be automatically released when this function
4236 returns to its caller. The object is always allocated in the generic address
4237 space (address space zero).</p>
4242 runtime stack, returning a pointer of the appropriate type to the program.
4243 If "NumElements" is specified, it is the number of elements allocated,
4244 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4245 specified, the value result of the allocation is guaranteed to be aligned to
4246 at least that boundary. If not specified, or if zero, the target can choose
4247 to align the allocation on any convenient boundary compatible with the
4248 type.</p>
4253 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4255 memory is automatically released when the function returns. The
4257 variables that must have an address available. When the function returns
4260 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4263 <pre>
4268 </pre>
4270 </div>
4272 <!-- _______________________________________________________________________ -->
4274 Instruction</a> </div>
4279 <pre>
4280 <result> = load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4281 <result> = volatile load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4283 </pre>
4290 from which to load. The pointer must point to
4297 operation (that is, the alignment of the memory address). A value of 0 or an
4299 alignment for the target. It is the responsibility of the code emitter to
4300 ensure that the alignment information is correct. Overestimating the
4301 alignment results in undefined behavior. Underestimating the alignment may
4308 and code generator that this load is not expected to be reused in the cache.
4309 The code generator may select special instructions to save cache bandwidth,
4313 <p>The location of memory pointed to is loaded. If the value being loaded is of
4314 scalar type then the number of bytes read does not exceed the minimum number
4315 of bytes needed to hold all bits of the type. For example, loading an
4318 is undefined if the value was not originally written using a store of the
4319 same type.</p>
4322 <pre>
4326 </pre>
4328 </div>
4330 <!-- _______________________________________________________________________ -->
4332 Instruction</a> </div>
4337 <pre>
4338 store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4339 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4340 </pre>
4347 and an address at which to store it. The type of the
4356 operation (that is, the alignment of the memory address). A value of 0 or an
4357 omitted "align" argument means that the operation has the preferential
4358 alignment for the target. It is the responsibility of the code emitter to
4359 ensure that the alignment information is correct. Overestimating the
4360 alignment results in an undefined behavior. Underestimating the alignment may
4363 <p>The optional !nontemporal metadata must reference a single metatadata
4365 value 1. The existence of the !nontemporal metatadata on the
4366 instruction tells the optimizer and code generator that this load is
4367 not expected to be reused in the cache. The code generator may
4368 select special instructions to save cache bandwidth, such as the
4369 MOVNT instruction on x86.</p>
4376 does not exceed the minimum number of bytes needed to hold all bits of the
4379 integral number of bytes, it is unspecified what happens to the extra bits
4380 that do not belong to the type, but they will typically be overwritten.</p>
4383 <pre>
4387 </pre>
4389 </div>
4391 <!-- _______________________________________________________________________ -->
4394 </div>
4399 <pre>
4402 </pre>
4407 It performs address calculation only and does not access memory.</p>
4410 <p>The first argument is always a pointer, and forms the basis of the
4411 calculation. The remaining arguments are indices that indicate which of the
4412 elements of the aggregate object are indexed. The interpretation of each
4413 index is dependent on the type being indexed into. The first index always
4414 indexes the pointer value given as the first argument, the second index
4415 indexes a value of the type pointed to (not necessarily the value directly
4416 pointed to, since the first index can be non-zero), etc. The first type
4417 indexed into must be a pointer value, subsequent types can be arrays,
4418 vectors, and structs. Note that subsequent types being indexed into
4419 can never be pointers, since that would require loading the pointer before
4420 continuing calculation.</p>
4422 <p>The type of each index argument depends on the type it is indexing into.
4425 or vector, integers of any width are allowed, and they are not required to be
4426 constant.</p>
4428 <p>For example, let's consider a C code fragment and how it gets compiled to
4429 LLVM:</p>
4432 struct RT {
4433 char A;
4435 char C;
4436 };
4437 struct ST {
4438 int X;
4439 double Y;
4440 struct RT Z;
4441 };
4443 int *foo(struct ST *s) {
4445 }
4446 </pre>
4454 define i32* @foo(%ST* %s) {
4455 entry:
4457 ret i32* %reg
4458 }
4459 </pre>
4464 }</tt>' type, a structure. The second index indexes into the third element
4466 i8 }</tt>' type, another structure. The third index indexes into the second
4468 array. The two dimensions of the array are subscripted into, yielding an
4472 <p>Note that it is perfectly legal to index partially through a structure,
4473 returning a pointer to an inner element. Because of this, the LLVM code for
4474 the given testcase is equivalent to:</p>
4476 <pre>
4477 define i32* @foo(%ST* %s) {
4483 ret i32* %t5
4484 }
4485 </pre>
4490 or if any of the addresses that would be formed by successive addition of
4491 the offsets implied by the indices to the base address with infinitely
4494 the addresses that point into the object, plus the address one byte past
4495 the end.</p>
4498 the base address with silently-wrapping two's complement arithmetic, and
4500 pointed to by the base pointer. The result value may not necessarily be
4501 used to access memory though, even if it happens to point into allocated
4503 section for more information.</p>
4505 <p>The getelementptr instruction is often confusing. For some more insight into
4509 <pre>
4518 </pre>
4520 </div>
4522 <!-- ======================================================================= -->
4524 </div>
4528 <p>The instructions in this category are the conversion instructions (casting)
4529 which all take a single operand and a type. They perform various bit
4530 conversions on the operand.</p>
4532 </div>
4534 <!-- _______________________________________________________________________ -->
4537 </div>
4541 <pre>
4543 </pre>
4552 size and type of the result, which must be
4555 allowed.</p>
4564 <pre>
4568 </pre>
4570 </div>
4572 <!-- _______________________________________________________________________ -->
4575 </div>
4579 <pre>
4581 </pre>
4602 <pre>
4605 </pre>
4607 </div>
4609 <!-- _______________________________________________________________________ -->
4612 </div>
4616 <pre>
4618 </pre>
4638 <pre>
4641 </pre>
4643 </div>
4645 <!-- _______________________________________________________________________ -->
4648 </div>
4653 <pre>
4655 </pre>
4673 undefined.</p>
4676 <pre>
4679 </pre>
4681 </div>
4683 <!-- _______________________________________________________________________ -->
4686 </div>
4690 <pre>
4692 </pre>
4696 floating point value.</p>
4702 type must be smaller than the destination type.</p>
4712 <pre>
4715 </pre>
4717 </div>
4719 <!-- _______________________________________________________________________ -->
4722 </div>
4726 <pre>
4728 </pre>
4744 towards zero) unsigned integer value. If the value cannot fit
4748 <pre>
4752 </pre>
4754 </div>
4756 <!-- _______________________________________________________________________ -->
4759 </div>
4763 <pre>
4765 </pre>
4783 the results are undefined.</p>
4786 <pre>
4790 </pre>
4792 </div>
4794 <!-- _______________________________________________________________________ -->
4797 </div>
4801 <pre>
4803 </pre>
4818 integer quantity and converts it to the corresponding floating point
4819 value. If the value cannot fit in the floating point value, the results are
4820 undefined.</p>
4823 <pre>
4826 </pre>
4828 </div>
4830 <!-- _______________________________________________________________________ -->
4833 </div>
4837 <pre>
4839 </pre>
4854 quantity and converts it to the corresponding floating point value. If the
4855 value cannot fit in the floating point value, the results are undefined.</p>
4858 <pre>
4861 </pre>
4863 </div>
4865 <!-- _______________________________________________________________________ -->
4868 </div>
4872 <pre>
4874 </pre>
4888 truncating or zero extending that value to the size of the integer type. If
4892 change.</p>
4895 <pre>
4898 </pre>
4900 </div>
4902 <!-- _______________________________________________________________________ -->
4905 </div>
4909 <pre>
4911 </pre>
4919 value to cast, and a type to cast it to, which must be a
4927 than the size of a pointer then a zero extension is done. If they are the
4931 <pre>
4935 </pre>
4937 </div>
4939 <!-- _______________________________________________________________________ -->
4942 </div>
4946 <pre>
4948 </pre>
4956 non-aggregate first class value, and a type to cast it to, which must also be
4959 identical. If the source type is a pointer, the destination type must also be
4960 a pointer. This instruction supports bitwise conversion of vectors to
4961 integers and to vectors of other types (as long as they have the same
4962 size).</p>
4969 be converted to other pointer types with this instruction. To convert
4974 <pre>
4978 </pre>
4980 </div>
4982 <!-- ======================================================================= -->
4988 defy better classification.</p>
4990 </div>
4992 <!-- _______________________________________________________________________ -->
4994 </div>
4999 <pre>
5000 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5001 </pre>
5005 boolean values based on comparison of its two integer, integer vector, or
5006 pointer operands.</p>
5010 the condition code indicating the kind of comparison to perform. It is not a
5011 value, just a keyword. The possible condition code are:</p>
5013 <ol>
5024 </ol>
5028 typed. They must also be identical types.</p>
5034 result, as follows:</p>
5036 <ol>
5039 performed.</li>
5043 performed.</li>
5070 </ol>
5073 values are compared as if they were integers.</p>
5075 <p>If the operands are integer vectors, then they are compared element by
5080 <pre>
5087 </pre>
5089 <p>Note that the code generator does not yet support vector types with
5092 </div>
5094 <!-- _______________________________________________________________________ -->
5096 </div>
5101 <pre>
5102 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5103 </pre>
5107 values based on comparison of its operands.</p>
5109 <p>If the operands are floating point scalars, then the result type is a boolean
5112 <p>If the operands are floating point vectors, then the result type is a vector
5113 of boolean with the same number of elements as the operands being
5114 compared.</p>
5118 the condition code indicating the kind of comparison to perform. It is not a
5119 value, just a keyword. The possible condition code are:</p>
5121 <ol>
5138 </ol>
5146 identical types.</p>
5151 vectors, then the vectors are compared element by element. Each comparison
5153 follows:</p>
5155 <ol>
5199 </ol>
5202 <pre>
5207 </pre>
5209 <p>Note that the code generator does not yet support vector types with
5212 </div>
5214 <!-- _______________________________________________________________________ -->
5217 </div>
5222 <pre>
5224 </pre>
5228 SSA graph representing the function.</p>
5231 <p>The type of the incoming values is specified with the first type field. After
5233 one pair for each predecessor basic block of the current block. Only values
5235 arguments to the PHI node. Only labels may be used as the label
5236 arguments.</p>
5238 <p>There must be no non-phi instructions between the start of a basic block and
5239 the PHI instructions: i.e. PHI instructions must be first in a basic
5240 block.</p>
5242 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5243 occur on the edge from the corresponding predecessor block to the current
5245 value on the same edge).</p>
5249 specified by the pair corresponding to the predecessor basic block that
5250 executed just prior to the current block.</p>
5253 <pre>
5254 Loop: ; Infinite loop that counts from 0 on up...
5255 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5256 %nextindvar = add i32 %indvar, 1
5257 br label %Loop
5258 </pre>
5260 </div>
5262 <!-- _______________________________________________________________________ -->
5265 </div>
5270 <pre>
5271 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5274 </pre>
5278 condition, without branching.</p>
5283 values indicating the condition, and two values of the
5285 vectors and the condition is a scalar, then entire vectors are selected, not
5286 individual elements.</p>
5290 first value argument; otherwise, it returns the second value argument.</p>
5292 <p>If the condition is a vector of i1, then the value arguments must be vectors
5293 of the same size, and the selection is done element by element.</p>
5296 <pre>
5298 </pre>
5300 <p>Note that the code generator does not yet support conditions
5301 with vector type.</p>
5303 </div>
5305 <!-- _______________________________________________________________________ -->
5308 </div>
5313 <pre>
5314 <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>]
5315 </pre>
5323 <ol>
5325 access any allocas or varargs in the caller. Note that calls may be
5326 marked "tail" even if they do not occur before
5328 present, the function call is eligible for tail call optimization,
5330 optimized into a jump</a>. The code generator may optimize calls marked
5332 sibling call optimization</a> when the caller and callee have
5333 matching signatures, or 2) forced tail call optimization when the
5334 following extra requirements are met:
5335 <ul>
5336 <li>Caller and callee both have the calling
5338 <li>The call is in tail position (ret immediately follows call and ret
5339 uses value of call or is void).</li>
5343 constraints are met.</a></li>
5344 </ul>
5345 </li>
5348 convention</a> the call should use. If none is specified, the call
5349 defaults to using C calling conventions. The calling convention of the
5350 call must match the calling convention of the target function, or else the
5351 behavior is undefined.</li>
5358 type of the return value. Functions that return no value are marked
5362 being invoked. The argument types must match the types implied by this
5363 signature. This type can be omitted if the function is not varargs and if
5364 the function type does not return a pointer to a function.</li>
5367 be invoked. In most cases, this is a direct function invocation, but
5369 to function value.</li>
5372 signature argument types and parameter attributes. All arguments must be
5374 signature indicates the function accepts a variable number of arguments,
5375 the extra arguments can be specified.</li>
5380 </ol>
5384 a specified function, with its incoming arguments bound to the specified
5386 function, control flow continues with the instruction after the function
5387 call, and the return value of the function is bound to the result
5388 argument.</p>
5391 <pre>
5392 %retval = call i32 @test(i32 %argc)
5396 call void %foo(i8 97 signext)
5398 %struct.A = type { i32, i8 }
5404 </pre>
5406 <p>llvm treats calls to some functions with names and arguments that match the
5407 standard C99 library as being the C99 library functions, and may perform
5408 optimizations or generate code for them under that assumption. This is
5409 something we'd like to change in the future to provide better support for
5410 freestanding environments and non-C-based languages.</p>
5412 </div>
5414 <!-- _______________________________________________________________________ -->
5417 </div>
5422 <pre>
5424 </pre>
5428 the "variable argument" area of a function call. It is used to implement the
5433 argument. It returns a value of the specified argument type and increments
5440 to the next argument. For more information, see the variable argument
5443 <p>It is legal for this instruction to be called in a function which does not
5445 function.</p>
5449 argument.</p>
5454 <p>Note that the code generator does not yet fully support va_arg on many
5455 targets. Also, it does not currently support va_arg with aggregate types on
5456 any target.</p>
5458 </div>
5460 <!-- *********************************************************************** -->
5462 <!-- *********************************************************************** -->
5467 well known names and semantics and are required to follow certain
5468 restrictions. Overall, these intrinsics represent an extension mechanism for
5469 the LLVM language that does not require changing all of the transformations
5470 in LLVM when adding to the language (or the bitcode reader/writer, the
5471 parser, etc...).</p>
5474 prefix is reserved in LLVM for intrinsic names; thus, function names may not
5475 begin with this prefix. Intrinsic functions must always be external
5476 functions: you cannot define the body of intrinsic functions. Intrinsic
5477 functions may only be used in call or invoke instructions: it is illegal to
5478 take the address of an intrinsic function. Additionally, because intrinsic
5479 functions are part of the LLVM language, it is required if any are added that
5480 they be documented here.</p>
5482 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
5483 family of functions that perform the same operation but on different data
5484 types. Because LLVM can represent over 8 million different integer types,
5485 overloading is used commonly to allow an intrinsic function to operate on any
5486 integer type. One or more of the argument types or the result type can be
5487 overloaded to accept any integer type. Argument types may also be defined as
5488 exactly matching a previous argument's type or the result type. This allows
5489 an intrinsic function which accepts multiple arguments, but needs all of them
5490 to be of the same type, to only be overloaded with respect to a single
5491 argument or the result.</p>
5493 <p>Overloaded intrinsics will have the names of its overloaded argument types
5494 encoded into its function name, each preceded by a period. Only those types
5495 which are overloaded result in a name suffix. Arguments whose type is matched
5497 can take an integer of any width and returns an integer of exactly the same
5498 integer width. This leads to a family of functions such as
5500 %val)</tt>. Only one type, the return type, is overloaded, and only one type
5501 suffix is required. Because the argument's type is matched against the return
5502 type, it does not require its own name suffix.</p>
5504 <p>To learn how to add an intrinsic function, please see the
5507 </div>
5509 <!-- ======================================================================= -->
5512 </div>
5516 <p>Variable argument support is defined in LLVM with
5518 intrinsic functions. These functions are related to the similarly named
5521 <p>All of these functions operate on arguments that use a target-specific value
5522 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
5523 not define what this type is, so all transformations should be prepared to
5524 handle these functions regardless of the type used.</p>
5527 instruction and the variable argument handling intrinsic functions are
5528 used.</p>
5531 define i32 @test(i32 %X, ...) {
5532 ; Initialize variable argument processing
5533 %ap = alloca i8*
5534 %ap2 = bitcast i8** %ap to i8*
5535 call void @llvm.va_start(i8* %ap2)
5537 ; Read a single integer argument
5538 %tmp = va_arg i8** %ap, i32
5540 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5541 %aq = alloca i8*
5542 %aq2 = bitcast i8** %aq to i8*
5543 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5544 call void @llvm.va_end(i8* %aq2)
5546 ; Stop processing of arguments.
5547 call void @llvm.va_end(i8* %ap2)
5548 ret i32 %tmp
5549 }
5551 declare void @llvm.va_start(i8*)
5552 declare void @llvm.va_copy(i8*, i8*)
5553 declare void @llvm.va_end(i8*)
5554 </pre>
5556 </div>
5558 <!-- _______________________________________________________________________ -->
5561 </div>
5567 <pre>
5569 </pre>
5580 macro available in C. In a target-dependent way, it initializes
5584 need to know the last argument of the function as the compiler can figure
5585 that out.</p>
5587 </div>
5589 <!-- _______________________________________________________________________ -->
5592 </div>
5597 <pre>
5599 </pre>
5603 which has been initialized previously
5612 macro available in C. In a target-dependent way, it destroys
5618 </div>
5620 <!-- _______________________________________________________________________ -->
5623 </div>
5628 <pre>
5630 </pre>
5634 from the source argument list to the destination argument list.</p>
5639 from.</p>
5643 macro available in C. In a target-dependent way, it copies the
5645 element. This intrinsic is necessary because
5647 arbitrarily complex and require, for example, memory allocation.</p>
5649 </div>
5651 <!-- ======================================================================= -->
5654 </div>
5659 Collection</a> (GC) requires the implementation and generation of these
5661 roots on the stack</a>, as well as garbage collector implementations that
5663 barriers. Front-ends for type-safe garbage collected languages should generate
5664 these intrinsics to make use of the LLVM garbage collectors. For more details,
5668 <p>The garbage collection intrinsics only operate on objects in the generic
5669 address space (address space zero).</p>
5671 </div>
5673 <!-- _______________________________________________________________________ -->
5676 </div>
5681 <pre>
5682 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5683 </pre>
5687 the code generator, and allows some metadata to be associated with it.</p>
5690 <p>The first argument specifies the address of a stack object that contains the
5691 root pointer. The second pointer (which must be either a constant or a
5692 global value address) contains the meta-data to be associated with the
5693 root.</p>
5697 location. At compile-time, the code generator generates information to allow
5702 </div>
5704 <!-- _______________________________________________________________________ -->
5707 </div>
5712 <pre>
5713 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5714 </pre>
5718 locations, allowing garbage collector implementations that require read
5719 barriers.</p>
5722 <p>The second argument is the address to read from, which should be an address
5723 allocated from the garbage collector. The first object is a pointer to the
5724 start of the referenced object, if needed by the language runtime (otherwise
5725 null).</p>
5729 instruction, but may be replaced with substantially more complex code by the
5734 </div>
5736 <!-- _______________________________________________________________________ -->
5739 </div>
5744 <pre>
5745 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5746 </pre>
5750 locations, allowing garbage collector implementations that require write
5751 barriers (such as generational or reference counting collectors).</p>
5754 <p>The first argument is the reference to store, the second is the start of the
5755 object to store it to, and the third is the address of the field of Obj to
5756 store to. If the runtime does not require a pointer to the object, Obj may
5757 be null.</p>
5761 instruction, but may be replaced with substantially more complex code by the
5766 </div>
5768 <!-- ======================================================================= -->
5771 </div>
5775 <p>These intrinsics are provided by LLVM to expose special features that may
5776 only be implemented with code generator support.</p>
5778 </div>
5780 <!-- _______________________________________________________________________ -->
5783 </div>
5788 <pre>
5790 </pre>
5794 target-specific value indicating the return address of the current function
5795 or one of its callers.</p>
5798 <p>The argument to this intrinsic indicates which function to return the address
5799 for. Zero indicates the calling function, one indicates its caller, etc.
5804 indicating the return address of the specified call frame, or zero if it
5805 cannot be identified. The value returned by this intrinsic is likely to be
5806 incorrect or 0 for arguments other than zero, so it should only be used for
5807 debugging purposes.</p>
5809 <p>Note that calling this intrinsic does not prevent function inlining or other
5810 aggressive transformations, so the value returned may not be that of the
5811 obvious source-language caller.</p>
5813 </div>
5815 <!-- _______________________________________________________________________ -->
5818 </div>
5823 <pre>
5825 </pre>
5829 target-specific frame pointer value for the specified stack frame.</p>
5832 <p>The argument to this intrinsic indicates which function to return the frame
5833 pointer for. Zero indicates the calling function, one indicates its caller,
5838 indicating the frame address of the specified call frame, or zero if it
5839 cannot be identified. The value returned by this intrinsic is likely to be
5840 incorrect or 0 for arguments other than zero, so it should only be used for
5841 debugging purposes.</p>
5843 <p>Note that calling this intrinsic does not prevent function inlining or other
5844 aggressive transformations, so the value returned may not be that of the
5845 obvious source-language caller.</p>
5847 </div>
5849 <!-- _______________________________________________________________________ -->
5852 </div>
5857 <pre>
5858 declare i8* @llvm.stacksave()
5859 </pre>
5863 of the function stack, for use
5865 useful for implementing language features like scoped automatic variable
5866 sized arrays in C99.</p>
5869 <p>This intrinsic returns a opaque pointer value that can be passed
5877 </div>
5879 <!-- _______________________________________________________________________ -->
5882 </div>
5887 <pre>
5888 declare void @llvm.stackrestore(i8* %ptr)
5889 </pre>
5893 the function stack to the state it was in when the
5895 executed. This is useful for implementing language features like scoped
5896 automatic variable sized arrays in C99.</p>
5899 <p>See the description
5902 </div>
5904 <!-- _______________________________________________________________________ -->
5907 </div>
5912 <pre>
5914 </pre>
5918 insert a prefetch instruction if supported; otherwise, it is a noop.
5919 Prefetches have no effect on the behavior of the program but can change its
5920 performance characteristics.</p>
5930 <p>This intrinsic does not modify the behavior of the program. In particular,
5931 prefetches cannot trap and do not produce a value. On targets that support
5932 this intrinsic, the prefetch can provide hints to the processor cache for
5933 better performance.</p>
5935 </div>
5937 <!-- _______________________________________________________________________ -->
5940 </div>
5945 <pre>
5947 </pre>
5951 Counter (PC) in a region of code to simulators and other tools. The method
5952 is target specific, but it is expected that the marker will use exported
5953 symbols to transmit the PC of the marker. The marker makes no guarantees
5954 that it will remain with any specific instruction after optimizations. It is
5955 possible that the presence of a marker will inhibit optimizations. The
5956 intended use is to be inserted after optimizations to allow correlations of
5957 simulation runs.</p>
5963 <p>This intrinsic does not modify the behavior of the program. Backends that do
5964 not support this intrinsic may ignore it.</p>
5966 </div>
5968 <!-- _______________________________________________________________________ -->
5971 </div>
5976 <pre>
5977 declare i64 @llvm.readcyclecounter()
5978 </pre>
5982 counter register (or similar low latency, high accuracy clocks) on those
5983 targets that support it. On X86, it should map to RDTSC. On Alpha, it
5984 should map to RPCC. As the backing counters overflow quickly (on the order
5988 <p>When directly supported, reading the cycle counter should not modify any
5989 memory. Implementations are allowed to either return a application specific
5990 value or a system wide value. On backends without support, this is lowered
5993 </div>
5995 <!-- ======================================================================= -->
5998 </div>
6002 <p>LLVM provides intrinsics for a few important standard C library functions.
6003 These intrinsics allow source-language front-ends to pass information about
6004 the alignment of the pointer arguments to the code generator, providing
6005 opportunity for more efficient code generation.</p>
6007 </div>
6009 <!-- _______________________________________________________________________ -->
6012 </div>
6018 integer bit width and for different address spaces. Not all targets support
6019 all bit widths however.</p>
6021 <pre>
6026 </pre>
6030 source location to the destination location.</p>
6033 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6034 and the pointers can be in specified address spaces.</p>
6038 <p>The first argument is a pointer to the destination, the second is a pointer
6039 to the source. The third argument is an integer argument specifying the
6040 number of bytes to copy, the fourth argument is the alignment of the
6041 source and destination locations, and the fifth is a boolean indicating a
6042 volatile access.</p>
6045 then the caller guarantees that both the source and destination pointers are
6046 aligned to that boundary.</p>
6050 The detailed access behavior is not very cleanly specified and it is unwise
6051 to depend on it.</p>
6056 source location to the destination location, which are not allowed to
6057 overlap. It copies "len" bytes of memory over. If the argument is known to
6058 be aligned to some boundary, this can be specified as the fourth argument,
6061 </div>
6063 <!-- _______________________________________________________________________ -->
6066 </div>
6071 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
6072 width and for different address space. Not all targets support all bit
6073 widths however.</p>
6075 <pre>
6080 </pre>
6084 source location to the destination location. It is similar to the
6086 overlap.</p>
6089 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6090 and the pointers can be in specified address spaces.</p>
6094 <p>The first argument is a pointer to the destination, the second is a pointer
6095 to the source. The third argument is an integer argument specifying the
6096 number of bytes to copy, the fourth argument is the alignment of the
6097 source and destination locations, and the fifth is a boolean indicating a
6098 volatile access.</p>
6101 then the caller guarantees that the source and destination pointers are
6102 aligned to that boundary.</p>
6106 The detailed access behavior is not very cleanly specified and it is unwise
6107 to depend on it.</p>
6112 source location to the destination location, which may overlap. It copies
6113 "len" bytes of memory over. If the argument is known to be aligned to some
6114 boundary, this can be specified as the fourth argument, otherwise it should
6117 </div>
6119 <!-- _______________________________________________________________________ -->
6122 </div>
6127 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
6128 width and for different address spaces. However, not all targets support all
6129 bit widths.</p>
6131 <pre>
6136 </pre>
6140 particular byte value.</p>
6143 intrinsic does not return a value and takes extra alignment/volatile
6144 arguments. Also, the destination can be in an arbitrary address space.</p>
6147 <p>The first argument is a pointer to the destination to fill, the second is the
6148 byte value with which to fill it, the third argument is an integer argument
6149 specifying the number of bytes to fill, and the fourth argument is the known
6150 alignment of the destination location.</p>
6153 then the caller guarantees that the destination pointer is aligned to that
6154 boundary.</p>
6158 The detailed access behavior is not very cleanly specified and it is unwise
6159 to depend on it.</p>
6163 at the destination location. If the argument is known to be aligned to some
6164 boundary, this can be specified as the fourth argument, otherwise it should
6167 </div>
6169 <!-- _______________________________________________________________________ -->
6172 </div>
6178 floating point or vector of floating point type. Not all targets support all
6179 types however.</p>
6181 <pre>
6182 declare float @llvm.sqrt.f32(float %Val)
6183 declare double @llvm.sqrt.f64(double %Val)
6184 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6185 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6186 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6187 </pre>
6193 behavior for negative numbers other than -0.0 (which allows for better
6194 optimization, because there is no need to worry about errno being
6198 <p>The argument and return value are floating point numbers of the same
6199 type.</p>
6202 <p>This function returns the sqrt of the specified operand if it is a
6203 nonnegative floating point number.</p>
6205 </div>
6207 <!-- _______________________________________________________________________ -->
6210 </div>
6216 floating point or vector of floating point type. Not all targets support all
6217 types however.</p>
6219 <pre>
6220 declare float @llvm.powi.f32(float %Val, i32 %power)
6221 declare double @llvm.powi.f64(double %Val, i32 %power)
6222 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6223 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6224 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6225 </pre>
6229 specified (positive or negative) power. The order of evaluation of
6230 multiplications is not defined. When a vector of floating point type is
6231 used, the second argument remains a scalar integer value.</p>
6234 <p>The second argument is an integer power, and the first is a value to raise to
6235 that power.</p>
6238 <p>This function returns the first value raised to the second power with an
6239 unspecified sequence of rounding operations.</p>
6241 </div>
6243 <!-- _______________________________________________________________________ -->
6246 </div>
6252 floating point or vector of floating point type. Not all targets support all
6253 types however.</p>
6255 <pre>
6256 declare float @llvm.sin.f32(float %Val)
6257 declare double @llvm.sin.f64(double %Val)
6258 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6259 declare fp128 @llvm.sin.f128(fp128 %Val)
6260 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6261 </pre>
6267 <p>The argument and return value are floating point numbers of the same
6268 type.</p>
6271 <p>This function returns the sine of the specified operand, returning the same
6273 in the same way.</p>
6275 </div>
6277 <!-- _______________________________________________________________________ -->
6280 </div>
6286 floating point or vector of floating point type. Not all targets support all
6287 types however.</p>
6289 <pre>
6290 declare float @llvm.cos.f32(float %Val)
6291 declare double @llvm.cos.f64(double %Val)
6292 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6293 declare fp128 @llvm.cos.f128(fp128 %Val)
6294 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6295 </pre>
6301 <p>The argument and return value are floating point numbers of the same
6302 type.</p>
6305 <p>This function returns the cosine of the specified operand, returning the same
6307 in the same way.</p>
6309 </div>
6311 <!-- _______________________________________________________________________ -->
6314 </div>
6320 floating point or vector of floating point type. Not all targets support all
6321 types however.</p>
6323 <pre>
6324 declare float @llvm.pow.f32(float %Val, float %Power)
6325 declare double @llvm.pow.f64(double %Val, double %Power)
6326 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6327 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6328 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6329 </pre>
6333 specified (positive or negative) power.</p>
6336 <p>The second argument is a floating point power, and the first is a value to
6337 raise to that power.</p>
6340 <p>This function returns the first value raised to the second power, returning
6342 conditions in the same way.</p>
6344 </div>
6346 <!-- ======================================================================= -->
6349 </div>
6353 <p>LLVM provides intrinsics for a few important bit manipulation operations.
6354 These allow efficient code generation for some algorithms.</p>
6356 </div>
6358 <!-- _______________________________________________________________________ -->
6361 </div>
6366 <p>This is an overloaded intrinsic function. You can use bswap on any integer
6369 <pre>
6373 </pre>
6377 values with an even number of bytes (positive multiple of 16 bits). These
6378 are useful for performing operations on data that is not in the target's
6379 native byte order.</p>
6383 and low byte of the input i16 swapped. Similarly,
6389 more, respectively).</p>
6391 </div>
6393 <!-- _______________________________________________________________________ -->
6396 </div>
6401 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
6402 width. Not all targets support all bit widths however.</p>
6404 <pre>
6410 </pre>
6414 in a value.</p>
6417 <p>The only argument is the value to be counted. The argument may be of any
6418 integer type. The return type must match the argument type.</p>
6423 </div>
6425 <!-- _______________________________________________________________________ -->
6428 </div>
6434 integer bit width. Not all targets support all bit widths however.</p>
6436 <pre>
6442 </pre>
6446 leading zeros in a variable.</p>
6449 <p>The only argument is the value to be counted. The argument may be of any
6450 integer type. The return type must match the argument type.</p>
6454 zeros in a variable. If the src == 0 then the result is the size in bits of
6457 </div>
6459 <!-- _______________________________________________________________________ -->
6462 </div>
6468 integer bit width. Not all targets support all bit widths however.</p>
6470 <pre>
6476 </pre>
6480 trailing zeros.</p>
6483 <p>The only argument is the value to be counted. The argument may be of any
6484 integer type. The return type must match the argument type.</p>
6488 zeros in a variable. If the src == 0 then the result is the size in bits of
6491 </div>
6493 <!-- ======================================================================= -->
6496 </div>
6502 </div>
6504 <!-- _______________________________________________________________________ -->
6507 </div>
6513 on any integer bit width.</p>
6515 <pre>
6516 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6517 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6518 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6519 </pre>
6523 a signed addition of the two arguments, and indicate whether an overflow
6524 occurred during the signed summation.</p>
6527 <p>The arguments (%a and %b) and the first element of the result structure may
6528 be of integer types of any bit width, but they must have the same bit
6529 width. The second element of the result structure must be of
6531 undergo signed addition.</p>
6535 a signed addition of the two variables. They return a structure — the
6536 first element of which is the signed summation, and the second element of
6537 which is a bit specifying if the signed summation resulted in an
6538 overflow.</p>
6541 <pre>
6542 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6543 %sum = extractvalue {i32, i1} %res, 0
6544 %obit = extractvalue {i32, i1} %res, 1
6545 br i1 %obit, label %overflow, label %normal
6546 </pre>
6548 </div>
6550 <!-- _______________________________________________________________________ -->
6553 </div>
6559 on any integer bit width.</p>
6561 <pre>
6562 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6563 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6564 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6565 </pre>
6569 an unsigned addition of the two arguments, and indicate whether a carry
6570 occurred during the unsigned summation.</p>
6573 <p>The arguments (%a and %b) and the first element of the result structure may
6574 be of integer types of any bit width, but they must have the same bit
6575 width. The second element of the result structure must be of
6577 undergo unsigned addition.</p>
6581 an unsigned addition of the two arguments. They return a structure —
6582 the first element of which is the sum, and the second element of which is a
6583 bit specifying if the unsigned summation resulted in a carry.</p>
6586 <pre>
6587 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6588 %sum = extractvalue {i32, i1} %res, 0
6589 %obit = extractvalue {i32, i1} %res, 1
6590 br i1 %obit, label %carry, label %normal
6591 </pre>
6593 </div>
6595 <!-- _______________________________________________________________________ -->
6598 </div>
6604 on any integer bit width.</p>
6606 <pre>
6607 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6608 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6609 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6610 </pre>
6614 a signed subtraction of the two arguments, and indicate whether an overflow
6615 occurred during the signed subtraction.</p>
6618 <p>The arguments (%a and %b) and the first element of the result structure may
6619 be of integer types of any bit width, but they must have the same bit
6620 width. The second element of the result structure must be of
6622 undergo signed subtraction.</p>
6626 a signed subtraction of the two arguments. They return a structure —
6627 the first element of which is the subtraction, and the second element of
6628 which is a bit specifying if the signed subtraction resulted in an
6629 overflow.</p>
6632 <pre>
6633 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6634 %sum = extractvalue {i32, i1} %res, 0
6635 %obit = extractvalue {i32, i1} %res, 1
6636 br i1 %obit, label %overflow, label %normal
6637 </pre>
6639 </div>
6641 <!-- _______________________________________________________________________ -->
6644 </div>
6650 on any integer bit width.</p>
6652 <pre>
6653 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6654 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6655 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6656 </pre>
6660 an unsigned subtraction of the two arguments, and indicate whether an
6661 overflow occurred during the unsigned subtraction.</p>
6664 <p>The arguments (%a and %b) and the first element of the result structure may
6665 be of integer types of any bit width, but they must have the same bit
6666 width. The second element of the result structure must be of
6668 undergo unsigned subtraction.</p>
6672 an unsigned subtraction of the two arguments. They return a structure —
6673 the first element of which is the subtraction, and the second element of
6674 which is a bit specifying if the unsigned subtraction resulted in an
6675 overflow.</p>
6678 <pre>
6679 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6680 %sum = extractvalue {i32, i1} %res, 0
6681 %obit = extractvalue {i32, i1} %res, 1
6682 br i1 %obit, label %overflow, label %normal
6683 </pre>
6685 </div>
6687 <!-- _______________________________________________________________________ -->
6690 </div>
6696 on any integer bit width.</p>
6698 <pre>
6699 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6700 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6701 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6702 </pre>
6707 a signed multiplication of the two arguments, and indicate whether an
6708 overflow occurred during the signed multiplication.</p>
6711 <p>The arguments (%a and %b) and the first element of the result structure may
6712 be of integer types of any bit width, but they must have the same bit
6713 width. The second element of the result structure must be of
6715 undergo signed multiplication.</p>
6719 a signed multiplication of the two arguments. They return a structure —
6720 the first element of which is the multiplication, and the second element of
6721 which is a bit specifying if the signed multiplication resulted in an
6722 overflow.</p>
6725 <pre>
6726 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6727 %sum = extractvalue {i32, i1} %res, 0
6728 %obit = extractvalue {i32, i1} %res, 1
6729 br i1 %obit, label %overflow, label %normal
6730 </pre>
6732 </div>
6734 <!-- _______________________________________________________________________ -->
6737 </div>
6743 on any integer bit width.</p>
6745 <pre>
6746 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6747 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6748 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6749 </pre>
6753 a unsigned multiplication of the two arguments, and indicate whether an
6754 overflow occurred during the unsigned multiplication.</p>
6757 <p>The arguments (%a and %b) and the first element of the result structure may
6758 be of integer types of any bit width, but they must have the same bit
6759 width. The second element of the result structure must be of
6761 undergo unsigned multiplication.</p>
6765 an unsigned multiplication of the two arguments. They return a structure
6766 — the first element of which is the multiplication, and the second
6767 element of which is a bit specifying if the unsigned multiplication resulted
6768 in an overflow.</p>
6771 <pre>
6772 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6773 %sum = extractvalue {i32, i1} %res, 0
6774 %obit = extractvalue {i32, i1} %res, 1
6775 br i1 %obit, label %overflow, label %normal
6776 </pre>
6778 </div>
6780 <!-- ======================================================================= -->
6783 </div>
6787 <p>Half precision floating point is a storage-only format. This means that it is
6788 a dense encoding (in memory) but does not support computation in the
6789 format.</p>
6791 <p>This means that code must first load the half-precision floating point
6792 value as an i16, then convert it to float with <a
6794 Computation can then be performed on the float value (including extending to
6795 double etc). To store the value back to memory, it is first converted to
6796 float if needed, then converted to i16 with
6798 storing as an i16 value.</p>
6799 </div>
6801 <!-- _______________________________________________________________________ -->
6804 </div>
6809 <pre>
6810 declare i16 @llvm.convert.to.fp16(f32 %a)
6811 </pre>
6815 a conversion from single precision floating point format to half precision
6816 floating point format.</p>
6819 <p>The intrinsic function contains single argument - the value to be
6820 converted.</p>
6824 a conversion from single precision floating point format to half precision
6826 contains the converted number.</p>
6829 <pre>
6830 %res = call i16 @llvm.convert.to.fp16(f32 %a)
6831 store i16 %res, i16* @x, align 2
6832 </pre>
6834 </div>
6836 <!-- _______________________________________________________________________ -->
6839 </div>
6844 <pre>
6845 declare f32 @llvm.convert.from.fp16(i16 %a)
6846 </pre>
6850 a conversion from half precision floating point format to single precision
6851 floating point format.</p>
6854 <p>The intrinsic function contains single argument - the value to be
6855 converted.</p>
6859 conversion from half single precision floating point format to single
6860 precision floating point format. The input half-float value is represented by
6864 <pre>
6865 %a = load i16* @x, align 2
6866 %res = call f32 @llvm.convert.from.fp16(i16 %a)
6867 </pre>
6869 </div>
6871 <!-- ======================================================================= -->
6874 </div>
6879 prefix), are described in
6883 </div>
6885 <!-- ======================================================================= -->
6888 </div>
6892 <p>The LLVM exception handling intrinsics (which all start with
6897 </div>
6899 <!-- ======================================================================= -->
6902 </div>
6906 <p>This intrinsic makes it possible to excise one parameter, marked with
6908 The result is a callable
6909 function pointer lacking the nest parameter - the caller does not need to
6910 provide a value for it. Instead, the value to use is stored in advance in a
6911 "trampoline", a block of memory usually allocated on the stack, which also
6912 contains code to splice the nest value into the argument list. This is used
6913 to implement the GCC nested function address extension.</p>
6915 <p>For example, if the function is
6918 follows:</p>
6923 %p = call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval)
6924 %fp = bitcast i8* %p to i32 (i32, i32)*
6925 </pre>
6930 </div>
6932 <!-- _______________________________________________________________________ -->
6935 </div>
6940 <pre>
6942 </pre>
6946 function pointer suitable for executing it.</p>
6951 sufficiently aligned block of memory; this memory is written to by the
6952 intrinsic. Note that the size and the alignment are target-specific - LLVM
6953 currently provides no portable way of determining them, so a front-end that
6954 generates this intrinsic needs to have some target-specific knowledge.
6960 dependent code, turning it into a function. A pointer to this function is
6962 function pointer type</a> before being called. The new function's signature
6965 is allowed, and it must be of pointer type. Calling the new function is
6970 returned function pointer is undefined.</p>
6972 </div>
6974 <!-- ======================================================================= -->
6977 </div>
6982 hardware constructs for atomic operations and memory synchronization. This
6983 provides an interface to the hardware, not an interface to the programmer. It
6984 is aimed at a low enough level to allow any programming models or APIs
6985 (Application Programming Interfaces) which need atomic behaviors to map
6986 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
6987 hardware provides a "universal IR" for source languages, it also provides a
6988 starting point for developing a "universal" atomic operation and
6989 synchronization IR.</p>
6992 software transaction memory systems, atomic primitives, and intrinsic
6993 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6994 application libraries. The hardware interface provided by LLVM should allow
6995 a clean implementation of all of these APIs and parallel programming models.
6996 No one model or paradigm should be selected above others unless the hardware
6997 itself ubiquitously does so.</p>
6999 </div>
7001 <!-- _______________________________________________________________________ -->
7004 </div>
7007 <pre>
7008 declare void @llvm.memory.barrier(i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device>)
7009 </pre>
7013 specific pairs of memory access types.</p>
7017 The first four arguments enables a specific barrier as listed below. The
7018 fifth argument specifies that the barrier applies to io or device or uncached
7019 memory.</p>
7021 <ul>
7027 </ul>
7030 <p>This intrinsic causes the system to enforce some ordering constraints upon
7031 the loads and stores of the program. This barrier does not
7034 and store operations (f.ex. load-load, or store-load), all of the first
7035 operations preceding the barrier will complete before any of the second
7036 operations succeeding the barrier begin. Specifically the semantics for each
7037 pairing is as follows:</p>
7039 <ul>
7041 after the barrier begins.</li>
7043 store after the barrier begins.</li>
7045 store after the barrier begins.</li>
7047 load after the barrier begins.</li>
7048 </ul>
7051 is enabled in a single memory barrier intrinsic.</p>
7053 <p>Backends may implement stronger barriers than those requested when they do
7054 not support as fine grained a barrier as requested. Some architectures do
7055 not need all types of barriers and on such architectures, these become
7056 noops.</p>
7059 <pre>
7060 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7061 %ptr = bitcast i8* %mallocP to i32*
7062 store i32 4, %ptr
7065 call void @llvm.memory.barrier(i1 false, i1 true, i1 false, i1 false)
7068 </pre>
7070 </div>
7072 <!-- _______________________________________________________________________ -->
7075 </div>
7081 any integer bit width and for different address spaces. Not all targets
7082 support all bit widths however.</p>
7084 <pre>
7086 declare i16 @llvm.atomic.cmp.swap.i16.p0i16(i16* <ptr>, i16 <cmp>, i16 <val>)
7087 declare i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* <ptr>, i32 <cmp>, i32 <val>)
7088 declare i64 @llvm.atomic.cmp.swap.i64.p0i64(i64* <ptr>, i64 <cmp>, i64 <val>)
7089 </pre>
7092 <p>This loads a value in memory and compares it to a given value. If they are
7093 equal, it stores a new value into the memory.</p>
7099 this integer type. While any bit width integer may be used, targets may only
7100 lower representations they support in hardware.</p>
7103 <p>This entire intrinsic must be executed atomically. It first loads the value
7106 memory. The loaded value is yielded in all cases. This provides the
7107 equivalent of an atomic compare-and-swap operation within the SSA
7108 framework.</p>
7111 <pre>
7112 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7113 %ptr = bitcast i8* %mallocP to i32*
7114 store i32 4, %ptr
7117 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 4, %val1)
7123 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 5, %val2)
7128 </pre>
7130 </div>
7132 <!-- _______________________________________________________________________ -->
7135 </div>
7140 integer bit width. Not all targets support all bit widths however.</p>
7142 <pre>
7147 </pre>
7158 integer type. The targets may only lower integer representations they
7159 support.</p>
7164 equivalent of an atomic swap operation within the SSA framework.</p>
7167 <pre>
7168 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7169 %ptr = bitcast i8* %mallocP to i32*
7170 store i32 4, %ptr
7173 %result1 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val1)
7179 %result2 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val2)
7184 </pre>
7186 </div>
7188 <!-- _______________________________________________________________________ -->
7192 </div>
7198 any integer bit width. Not all targets support all bit widths however.</p>
7200 <pre>
7205 </pre>
7212 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7213 and the second an integer value. The result is also an integer value. These
7214 integer types can have any bit width, but they must all have the same bit
7215 width. The targets may only lower integer representations they support.</p>
7218 <p>This intrinsic does a series of operations atomically. It first loads the
7223 <pre>
7224 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7225 %ptr = bitcast i8* %mallocP to i32*
7226 store i32 4, %ptr
7227 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 4)
7229 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 2)
7231 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 5)
7234 </pre>
7236 </div>
7238 <!-- _______________________________________________________________________ -->
7242 </div>
7248 any integer bit width and for different address spaces. Not all targets
7249 support all bit widths however.</p>
7251 <pre>
7256 </pre>
7263 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7264 and the second an integer value. The result is also an integer value. These
7265 integer types can have any bit width, but they must all have the same bit
7266 width. The targets may only lower integer representations they support.</p>
7269 <p>This intrinsic does a series of operations atomically. It first loads the
7275 <pre>
7276 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7277 %ptr = bitcast i8* %mallocP to i32*
7278 store i32 8, %ptr
7279 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 4)
7281 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 2)
7283 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 5)
7286 </pre>
7288 </div>
7290 <!-- _______________________________________________________________________ -->
7296 </div>
7301 <p>These are overloaded intrinsics. You can
7304 bit width and for different address spaces. Not all targets support all bit
7305 widths however.</p>
7307 <pre>
7312 </pre>
7314 <pre>
7319 </pre>
7321 <pre>
7326 </pre>
7328 <pre>
7333 </pre>
7341 <p>These intrinsics take two arguments, the first a pointer to an integer value
7342 and the second an integer value. The result is also an integer value. These
7343 integer types can have any bit width, but they must all have the same bit
7344 width. The targets may only lower integer representations they support.</p>
7347 <p>These intrinsics does a series of operations atomically. They first load the
7353 <pre>
7354 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7355 %ptr = bitcast i8* %mallocP to i32*
7356 store i32 0x0F0F, %ptr
7357 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32(i32* %ptr, i32 0xFF)
7359 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32(i32* %ptr, i32 0xFF)
7361 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32(i32* %ptr, i32 0F)
7363 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32(i32* %ptr, i32 0F)
7366 </pre>
7368 </div>
7370 <!-- _______________________________________________________________________ -->
7376 </div>
7384 address spaces. Not all targets support all bit widths however.</p>
7386 <pre>
7391 </pre>
7393 <pre>
7398 </pre>
7400 <pre>
7405 </pre>
7407 <pre>
7412 </pre>
7415 <p>These intrinsics takes the signed or unsigned minimum or maximum of
7420 <p>These intrinsics take two arguments, the first a pointer to an integer value
7421 and the second an integer value. The result is also an integer value. These
7422 integer types can have any bit width, but they must all have the same bit
7423 width. The targets may only lower integer representations they support.</p>
7426 <p>These intrinsics does a series of operations atomically. They first load the
7432 <pre>
7433 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7434 %ptr = bitcast i8* %mallocP to i32*
7435 store i32 7, %ptr
7436 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32(i32* %ptr, i32 -2)
7438 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32(i32* %ptr, i32 8)
7440 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32(i32* %ptr, i32 10)
7442 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32(i32* %ptr, i32 30)
7445 </pre>
7447 </div>
7450 <!-- ======================================================================= -->
7453 </div>
7457 <p>This class of intrinsics exists to information about the lifetime of memory
7458 objects and ranges where variables are immutable.</p>
7460 </div>
7462 <!-- _______________________________________________________________________ -->
7465 </div>
7470 <pre>
7472 </pre>
7476 object's lifetime.</p>
7479 <p>The first argument is a constant integer representing the size of the
7480 object, or -1 if it is variable sized. The second argument is a pointer to
7481 the object.</p>
7484 <p>This intrinsic indicates that before this point in the code, the value of the
7486 never be used and has an undefined value. A load from the pointer that
7487 precedes this intrinsic can be replaced with
7490 </div>
7492 <!-- _______________________________________________________________________ -->
7495 </div>
7500 <pre>
7502 </pre>
7506 object's lifetime.</p>
7509 <p>The first argument is a constant integer representing the size of the
7510 object, or -1 if it is variable sized. The second argument is a pointer to
7511 the object.</p>
7514 <p>This intrinsic indicates that after this point in the code, the value of the
7516 never be used and has an undefined value. Any stores into the memory object
7517 following this intrinsic may be removed as dead.
7519 </div>
7521 <!-- _______________________________________________________________________ -->
7524 </div>
7529 <pre>
7531 </pre>
7535 a memory object will not change.</p>
7538 <p>The first argument is a constant integer representing the size of the
7539 object, or -1 if it is variable sized. The second argument is a pointer to
7540 the object.</p>
7544 the return value, the referenced memory location is constant and
7545 unchanging.</p>
7547 </div>
7549 <!-- _______________________________________________________________________ -->
7552 </div>
7557 <pre>
7558 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
7559 </pre>
7563 a memory object are mutable.</p>
7567 The second argument is a constant integer representing the size of the
7568 object, or -1 if it is variable sized and the third argument is a pointer
7569 to the object.</p>
7574 </div>
7576 <!-- ======================================================================= -->
7579 </div>
7583 <p>This class of intrinsics is designed to be generic and has no specific
7584 purpose.</p>
7586 </div>
7588 <!-- _______________________________________________________________________ -->
7591 </div>
7596 <pre>
7597 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
7598 </pre>
7604 <p>The first argument is a pointer to a value, the second is a pointer to a
7605 global string, the third is a pointer to a global string which is the source
7606 file name, and the last argument is the line number.</p>
7609 <p>This intrinsic allows annotation of local variables with arbitrary strings.
7610 This can be useful for special purpose optimizations that want to look for
7611 these annotations. These have no other defined use, they are ignored by code
7612 generation and optimization.</p>
7614 </div>
7616 <!-- _______________________________________________________________________ -->
7619 </div>
7625 any integer bit width.</p>
7627 <pre>
7628 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
7629 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
7630 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
7631 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
7632 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
7633 </pre>
7639 <p>The first argument is an integer value (result of some expression), the
7640 second is a pointer to a global string, the third is a pointer to a global
7641 string which is the source file name, and the last argument is the line
7642 number. It returns the value of the first argument.</p>
7645 <p>This intrinsic allows annotations to be put on arbitrary expressions with
7646 arbitrary strings. This can be useful for special purpose optimizations that
7647 want to look for these annotations. These have no other defined use, they
7648 are ignored by code generation and optimization.</p>
7650 </div>
7652 <!-- _______________________________________________________________________ -->
7655 </div>
7660 <pre>
7661 declare void @llvm.trap()
7662 </pre>
7671 <p>This intrinsics is lowered to the target dependent trap instruction. If the
7672 target does not have a trap instruction, this intrinsic will be lowered to
7675 </div>
7677 <!-- _______________________________________________________________________ -->
7680 </div>
7685 <pre>
7687 </pre>
7692 ensure that it is placed on the stack before local variables.</p>
7696 arguments. The first argument is the value loaded from the stack
7698 that has enough space to hold the value of the guard.</p>
7701 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
7703 stack. This is to ensure that if a local variable on the stack is
7704 overwritten, it will destroy the value of the guard. When the function exits,
7705 the guard on the stack is checked against the original guard. If they are
7707 function.</p>
7709 </div>
7711 <!-- _______________________________________________________________________ -->
7714 </div>
7719 <pre>
7722 </pre>
7726 the optimizers to determine at compile time whether a) an operation (like
7727 memcpy) will overflow a buffer that corresponds to an object, or b) that a
7728 runtime check for overflow isn't necessary. An object in this context means
7729 an allocation of a specific class, structure, array, or other object.</p>
7742 compile time.</p>
7744 </div>
7746 <!-- *********************************************************************** -->
7747 <hr>
7748 <address>
7756 Last modified: $Date$
7757 </address>
7759 </body>
7760 </html>