| blob | 5959b3d5dca825e110d6327770a682daae73f777 |
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>
83 </ol>
84 </li>
85 </ol>
86 </li>
88 <ol>
96 </ol>
97 </li>
99 <ol>
102 </ol>
103 </li>
105 <ol>
108 Global Variable</a></li>
110 Global Variable</a></li>
112 Global Variable</a></li>
113 </ol>
114 </li>
116 <ol>
118 <ol>
126 </ol>
127 </li>
129 <ol>
142 </ol>
143 </li>
145 <ol>
152 </ol>
153 </li>
155 <ol>
159 </ol>
160 </li>
162 <ol>
165 </ol>
166 </li>
168 <ol>
173 </ol>
174 </li>
176 <ol>
189 </ol>
190 </li>
192 <ol>
199 </ol>
200 </li>
201 </ol>
202 </li>
204 <ol>
206 <ol>
210 </ol>
211 </li>
213 <ol>
217 </ol>
218 </li>
220 <ol>
228 </ol>
229 </li>
231 <ol>
243 </ol>
244 </li>
246 <ol>
251 </ol>
252 </li>
254 <ol>
261 </ol>
262 </li>
264 <ol>
267 </ol>
268 </li>
272 <ol>
274 </ol>
275 </li>
277 <ol>
291 </ol>
292 </li>
294 <ol>
299 </ol>
300 </li>
302 <ol>
313 </ol>
314 </li>
315 </ol>
316 </li>
317 </ol>
322 </div>
324 <!-- *********************************************************************** -->
326 <!-- *********************************************************************** -->
328 <div>
330 <p>This document is a reference manual for the LLVM assembly language. LLVM is
331 a Static Single Assignment (SSA) based representation that provides type
332 safety, low-level operations, flexibility, and the capability of representing
333 'all' high-level languages cleanly. It is the common code representation
334 used throughout all phases of the LLVM compilation strategy.</p>
336 </div>
338 <!-- *********************************************************************** -->
340 <!-- *********************************************************************** -->
342 <div>
344 <p>The LLVM code representation is designed to be used in three different forms:
345 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
346 for fast loading by a Just-In-Time compiler), and as a human readable
347 assembly language representation. This allows LLVM to provide a powerful
348 intermediate representation for efficient compiler transformations and
349 analysis, while providing a natural means to debug and visualize the
350 transformations. The three different forms of LLVM are all equivalent. This
351 document describes the human readable representation and notation.</p>
353 <p>The LLVM representation aims to be light-weight and low-level while being
354 expressive, typed, and extensible at the same time. It aims to be a
355 "universal IR" of sorts, by being at a low enough level that high-level ideas
356 may be cleanly mapped to it (similar to how microprocessors are "universal
357 IR's", allowing many source languages to be mapped to them). By providing
358 type information, LLVM can be used as the target of optimizations: for
359 example, through pointer analysis, it can be proven that a C automatic
360 variable is never accessed outside of the current function, allowing it to
361 be promoted to a simple SSA value instead of a memory location.</p>
363 <!-- _______________________________________________________________________ -->
364 <h4>
366 </h4>
368 <div>
370 <p>It is important to note that this document describes 'well formed' LLVM
371 assembly language. There is a difference between what the parser accepts and
372 what is considered 'well formed'. For example, the following instruction is
373 syntactically okay, but not well formed:</p>
377 </pre>
380 LLVM infrastructure provides a verification pass that may be used to verify
381 that an LLVM module is well formed. This pass is automatically run by the
382 parser after parsing input assembly and by the optimizer before it outputs
383 bitcode. The violations pointed out by the verifier pass indicate bugs in
384 transformation passes or input to the parser.</p>
386 </div>
388 </div>
390 <!-- Describe the typesetting conventions here. -->
392 <!-- *********************************************************************** -->
394 <!-- *********************************************************************** -->
396 <div>
398 <p>LLVM identifiers come in two basic types: global and local. Global
400 character. Local identifiers (register names, types) begin with
402 for identifiers, for different purposes:</p>
404 <ol>
405 <li>Named values are represented as a string of characters with their prefix.
409 other characters in their names can be surrounded with quotes. Special
411 ASCII code for the character in hexadecimal. In this way, any character
412 can be used in a name value, even quotes themselves.</li>
414 <li>Unnamed values are represented as an unsigned numeric value with their
419 </ol>
421 <p>LLVM requires that values start with a prefix for two reasons: Compilers
422 don't need to worry about name clashes with reserved words, and the set of
423 reserved words may be expanded in the future without penalty. Additionally,
424 unnamed identifiers allow a compiler to quickly come up with a temporary
425 variable without having to avoid symbol table conflicts.</p>
427 <p>Reserved words in LLVM are very similar to reserved words in other
428 languages. There are keywords for different opcodes
434 reserved words cannot conflict with variable names, because none of them
437 <p>Here is an example of LLVM code to multiply the integer variable
444 </pre>
450 </pre>
458 </pre>
461 lexical features of LLVM:</p>
463 <ol>
465 line.</li>
467 <li>Unnamed temporaries are created when the result of a computation is not
468 assigned to a named value.</li>
471 </ol>
473 <p>It also shows a convention that we follow in this document. When
474 demonstrating instructions, we will follow an instruction with a comment that
475 defines the type and name of value produced. Comments are shown in italic
476 text.</p>
478 </div>
480 <!-- *********************************************************************** -->
482 <!-- *********************************************************************** -->
483 <div>
484 <!-- ======================================================================= -->
485 <h3>
487 </h3>
489 <div>
492 of the input programs. Each module consists of functions, global variables,
493 and symbol table entries. Modules may be combined together with the LLVM
494 linker, which merges function (and global variable) definitions, resolves
495 forward declarations, and merges symbol table entries. Here is an example of
500 <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>
508 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8*</i>
513 }
517 !foo = !{!1, null}
518 </pre>
526 <p>In general, a module is made up of a list of global values, where both
527 functions and global variables are global values. Global values are
528 represented by a pointer to a memory location (in this case, a pointer to an
529 array of char, and a pointer to a function), and have one of the
532 </div>
534 <!-- ======================================================================= -->
535 <h3>
537 </h3>
539 <div>
541 <p>All Global Variables and Functions have one of the following types of
542 linkage:</p>
544 <dl>
547 by objects in the current module. In particular, linking code into a
548 module with an private global value may cause the private to be renamed as
549 necessary to avoid collisions. Because the symbol is private to the
550 module, all references can be updated. This doesn't show up in any symbol
551 table in the object file.</dd>
555 assembler and evaluated by the linker. Unlike normal strong symbols, they
556 are removed by the linker from the final linked image (executable or
557 dynamic library).</dd>
562 linker. The symbols are removed by the linker from the final linked image
563 (executable or dynamic library).</dd>
565 <dt><tt><b><a name="linkage_linker_private_weak_def_auto">linker_private_weak_def_auto</a></b></tt></dt>
567 of the object is not taken. For instance, functions that had an inline
568 definition, but the compiler decided not to inline it. Note,
571 visibility. The symbols are removed by the linker from the final linked
572 image (executable or dynamic library).</dd>
575 <dd>Similar to private, but the value shows as a local symbol
581 into the object file corresponding to the LLVM module. They exist to
582 allow inlining and other optimizations to take place given knowledge of
583 the definition of the global, which is known to be somewhere outside the
586 This linkage type is only allowed on definitions, not declarations.</dd>
590 the same name when linkage occurs. This can be used to implement
591 some forms of inline functions, templates, or other code which must be
592 generated in each translation unit that uses it, but where the body may
593 be overridden with a more definitive definition later. Unreferenced
596 inline the body of this function into callers because it doesn't know if
597 this definition of the function is the definitive definition within the
598 program or whether it will be overridden by a stronger definition.
599 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
600 linkage.</dd>
610 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
611 global scope.
612 Symbols with "<tt>common</tt>" linkage are merged in the same way as
615 must have a zero initializer, and may not be marked '<a
617 have common linkage.</dd>
622 pointer to array type. When two global variables with appending linkage
623 are linked together, the two global arrays are appended together. This is
624 the LLVM, typesafe, equivalent of having the system linker append together
628 <dd>The semantics of this linkage follow the ELF object file model: the symbol
629 is weak until linked, if not linked, the symbol becomes null instead of
630 being an undefined reference.</dd>
634 <dd>Some languages allow differing globals to be merged, such as two functions
636 that only equivalent globals are ever merged (the "one definition rule"
639 be merged with equivalent globals. These linkage types are otherwise the
643 <dd>If none of the above identifiers are used, the global is externally
644 visible, meaning that it participates in linkage and can be used to
645 resolve external symbol references.</dd>
646 </dl>
648 <p>The next two types of linkage are targeted for Microsoft Windows platform
649 only. They are designed to support importing (exporting) symbols from (to)
650 DLLs (Dynamic Link Libraries).</p>
652 <dl>
655 or variable via a global pointer to a pointer that is set up by the DLL
656 exporting the symbol. On Microsoft Windows targets, the pointer name is
658 name.</dd>
662 pointer to a pointer in a DLL, so that it can be referenced with the
665 variable name.</dd>
666 </dl>
669 another module defined a "<tt>.LC0</tt>" variable and was linked with this
670 one, one of the two would be renamed, preventing a collision. Since
672 declarations), they are accessible outside of the current module.</p>
681 </div>
683 <!-- ======================================================================= -->
684 <h3>
686 </h3>
688 <div>
692 convention specified for the call. The calling convention of any pair of
693 dynamic caller/callee must match, or the behavior of the program is
694 undefined. The following calling conventions are supported by LLVM, and more
695 may be added in the future:</p>
697 <dl>
699 <dd>This calling convention (the default if no other calling convention is
700 specified) matches the target C calling conventions. This calling
701 convention supports varargs function calls and tolerates some mismatch in
702 the declared prototype and implemented declaration of the function (as
703 does normal C).</dd>
706 <dd>This calling convention attempts to make calls as fast as possible
707 (e.g. by passing things in registers). This calling convention allows the
708 target to use whatever tricks it wants to produce fast code for the
709 target, without having to conform to an externally specified ABI
710 (Application Binary Interface).
712 when this or the GHC convention is used.</a> This calling convention
713 does not support varargs and requires the prototype of all callees to
714 exactly match the prototype of the function definition.</dd>
717 <dd>This calling convention attempts to make code in the caller as efficient
718 as possible under the assumption that the call is not commonly executed.
719 As such, these calls often preserve all registers so that the call does
720 not break any live ranges in the caller side. This calling convention
721 does not support varargs and requires the prototype of all callees to
722 exactly match the prototype of the function definition.</dd>
725 <dd>This calling convention has been implemented specifically for use by the
727 It passes everything in registers, going to extremes to achieve this by
728 disabling callee save registers. This calling convention should not be
729 used lightly but only for specific situations such as an alternative to
731 implementing functional programming languages.At the moment only X86
732 supports this convention and it has the following limitations:
733 <ul>
735 floating point types are supported.</li>
738 </ul>
739 This calling convention supports
741 requires both the caller and callee are using it.
742 </dd>
745 <dd>Any calling convention may be specified by number, allowing
746 target-specific calling conventions to be used. Target specific calling
748 </dl>
750 <p>More calling conventions can be added/defined on an as-needed basis, to
751 support Pascal conventions or any other well-known target-independent
752 convention.</p>
754 </div>
756 <!-- ======================================================================= -->
757 <h3>
759 </h3>
761 <div>
763 <p>All Global Variables and Functions have one of the following visibility
764 styles:</p>
766 <dl>
768 <dd>On targets that use the ELF object file format, default visibility means
769 that the declaration is visible to other modules and, in shared libraries,
770 means that the declared entity may be overridden. On Darwin, default
771 visibility means that the declaration is visible to other modules. Default
775 <dd>Two declarations of an object with hidden visibility refer to the same
776 object if they are in the same shared object. Usually, hidden visibility
777 indicates that the symbol will not be placed into the dynamic symbol
778 table, so no other module (executable or shared library) can reference it
779 directly.</dd>
782 <dd>On ELF, protected visibility indicates that the symbol will be placed in
783 the dynamic symbol table, but that references within the defining module
784 will bind to the local symbol. That is, the symbol cannot be overridden by
785 another module.</dd>
786 </dl>
788 </div>
790 <!-- ======================================================================= -->
791 <h3>
793 </h3>
795 <div>
797 <p>LLVM IR allows you to specify name aliases for certain types. This can make
798 it easier to read the IR and make the IR more condensed (particularly when
799 recursive types are involved). An example of a name specification is:</p>
802 %mytype = type { %mytype*, i32 }
803 </pre>
809 <p>Note that type names are aliases for the structural type that they indicate,
810 and that you can therefore specify multiple names for the same type. This
811 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
812 uses structural typing, the name is not part of the type. When printing out
814 particular shape. This means that if you have code where two different
815 source types end up having the same LLVM type, that the dumper will sometimes
816 print the "wrong" or unexpected type. This is an important design point and
817 isn't going to change.</p>
819 </div>
821 <!-- ======================================================================= -->
822 <h3>
824 </h3>
826 <div>
828 <p>Global variables define regions of memory allocated at compilation time
829 instead of run-time. Global variables may optionally be initialized, may
830 have an explicit section to be placed in, and may have an optional explicit
831 alignment specified. A variable may be defined as "thread_local", which
832 means that it will not be shared by threads (each thread will have a
833 separated copy of the variable). A variable may be defined as a global
834 "constant," which indicates that the contents of the variable
836 global data to be placed in the read-only section of an executable, etc).
837 Note that variables that need runtime initialization cannot be marked
841 constant, even if the final definition of the global is not. This capability
842 can be used to enable slightly better optimization of the program, but
843 requires the language definition to guarantee that optimizations based on the
844 'constantness' are valid for the translation units that do not include the
845 definition.</p>
847 <p>As SSA values, global variables define pointer values that are in scope
848 (i.e. they dominate) all basic blocks in the program. Global variables
849 always define a pointer to their "content" type because they describe a
850 region of memory, and all memory objects in LLVM are accessed through
851 pointers.</p>
854 that the address is not significant, only the content. Constants marked
855 like this can be merged with other constants if they have the same
858 constant whose address is significant.</p>
860 <p>A global variable may be declared to reside in a target-specific numbered
861 address space. For targets that support them, address spaces may affect how
862 optimizations are performed and/or what target instructions are used to
863 access the variable. The default address space is zero. The address space
864 qualifier must precede any other attributes.</p>
866 <p>LLVM allows an explicit section to be specified for globals. If the target
867 supports it, it will emit globals to the section specified.</p>
869 <p>An explicit alignment may be specified for a global, which must be a power
870 of 2. If not present, or if the alignment is set to zero, the alignment of
871 the global is set by the target to whatever it feels convenient. If an
872 explicit alignment is specified, the global is forced to have exactly that
873 alignment. Targets and optimizers are not allowed to over-align the global
874 if the global has an assigned section. In this case, the extra alignment
875 could be observable: for example, code could assume that the globals are
876 densely packed in their section and try to iterate over them as an array,
877 alignment padding would break this iteration.</p>
879 <p>For example, the following defines a global in a numbered address space with
880 an initializer, section, and alignment:</p>
884 </pre>
886 </div>
889 <!-- ======================================================================= -->
890 <h3>
892 </h3>
894 <div>
902 name, a (possibly empty) argument list (each with optional
906 curly brace, a list of basic blocks, and a closing curly brace.</p>
914 name, a possibly empty list of arguments, an optional alignment, and an
917 <p>A function definition contains a list of basic blocks, forming the CFG
918 (Control Flow Graph) for the function. Each basic block may optionally start
919 with a label (giving the basic block a symbol table entry), contains a list
921 instruction (such as a branch or function return).</p>
923 <p>The first basic block in a function is special in two ways: it is immediately
924 executed on entrance to the function, and it is not allowed to have
925 predecessor basic blocks (i.e. there can not be any branches to the entry
926 block of a function). Because the block can have no predecessors, it also
929 <p>LLVM allows an explicit section to be specified for functions. If the target
930 supports it, it will emit functions to the section specified.</p>
932 <p>An explicit alignment may be specified for a function. If not present, or if
933 the alignment is set to zero, the alignment of the function is set by the
934 target to whatever it feels convenient. If an explicit alignment is
935 specified, the function is forced to have at least that much alignment. All
939 be significant and two identical functions can be merged</p>.
948 </pre>
950 </div>
952 <!-- ======================================================================= -->
953 <h3>
955 </h3>
957 <div>
960 function, global variable, another alias or bitcast of global value). Aliases
967 </pre>
969 </div>
971 <!-- ======================================================================= -->
972 <h3>
974 </h3>
976 <div>
979 nodes</a> (but not metadata strings) are the only valid operands for
980 a named metadata.</p>
984 ; Some unnamed metadata nodes, which are referenced by the named metadata.
988 ; A named metadata.
990 </pre>
992 </div>
994 <!-- ======================================================================= -->
995 <h3>
997 </h3>
999 <div>
1001 <p>The return type and each parameter of a function type may have a set of
1003 used to communicate additional information about the result or parameters of
1004 a function. Parameter attributes are considered to be part of the function,
1005 not of the function type, so functions with different parameter attributes
1006 can have the same function type.</p>
1008 <p>Parameter attributes are simple keywords that follow the type specified. If
1009 multiple parameter attributes are needed, they are space separated. For
1010 example:</p>
1013 declare i32 @printf(i8* noalias nocapture, ...)
1014 declare i32 @atoi(i8 zeroext)
1015 declare signext i8 @returns_signed_char()
1016 </pre>
1023 <dl>
1025 <dd>This indicates to the code generator that the parameter or return value
1026 should be zero-extended to the extent required by the target's ABI (which
1028 parameter) or the callee (for a return value).</dd>
1031 <dd>This indicates to the code generator that the parameter or return value
1032 should be sign-extended to the extent required by the target's ABI (which
1033 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1034 return value).</dd>
1037 <dd>This indicates that this parameter or return value should be treated in a
1038 special target-dependent fashion during while emitting code for a function
1039 call or return (usually, by putting it in a register as opposed to memory,
1040 though some targets use it to distinguish between two different kinds of
1041 registers). Use of this attribute is target-specific.</dd>
1044 <dd><p>This indicates that the pointer parameter should really be passed by
1045 value to the function. The attribute implies that a hidden copy of the
1046 pointee
1047 is made between the caller and the callee, so the callee is unable to
1048 modify the value in the callee. This attribute is only valid on LLVM
1049 pointer arguments. It is generally used to pass structs and arrays by
1050 value, but is also valid on pointers to scalars. The copy is considered
1051 to belong to the caller not the callee (for example,
1054 values.</p>
1056 <p>The byval attribute also supports specifying an alignment with
1057 the align attribute. It indicates the alignment of the stack slot to
1058 form and the known alignment of the pointer specified to the call site. If
1059 the alignment is not specified, then the code generator makes a
1060 target-specific assumption.</p></dd>
1063 <dd>This indicates that the pointer parameter specifies the address of a
1064 structure that is the return value of the function in the source program.
1065 This pointer must be guaranteed by the caller to be valid: loads and
1066 stores to the structure may be assumed by the callee to not to trap. This
1067 may only be applied to the first parameter. This is not a valid attribute
1068 for return values. </dd>
1071 <dd>This indicates that pointer values
1074 ignoring certain "irrelevant" dependencies.
1075 For a call to the parent function, dependencies between memory
1076 references from before or after the call and from those during the call
1078 return value used in that call.
1079 The caller shares the responsibility with the callee for ensuring that
1080 these requirements are met.
1081 For further details, please see the discussion of the NoAlias response in
1083 <br>
1086 arguments, though it is slightly weaker.
1087 <br>
1090 </dd>
1093 <dd>This indicates that the callee does not make any copies of the pointer
1094 that outlive the callee itself. This is not a valid attribute for return
1095 values.</dd>
1098 <dd>This indicates that the pointer parameter can be excised using the
1100 attribute for return values.</dd>
1101 </dl>
1103 </div>
1105 <!-- ======================================================================= -->
1106 <h3>
1108 </h3>
1110 <div>
1112 <p>Each function may specify a garbage collector name, which is simply a
1113 string:</p>
1116 define void @f() gc "name" { ... }
1117 </pre>
1120 collector which will cause the compiler to alter its output in order to
1121 support the named garbage collection algorithm.</p>
1123 </div>
1125 <!-- ======================================================================= -->
1126 <h3>
1128 </h3>
1130 <div>
1132 <p>Function attributes are set to communicate additional information about a
1133 function. Function attributes are considered to be part of the function, not
1134 of the function type, so functions with different parameter attributes can
1135 have the same function type.</p>
1137 <p>Function attributes are simple keywords that follow the type specified. If
1138 multiple attributes are needed, they are space separated. For example:</p>
1141 define void @f() noinline { ... }
1142 define void @f() alwaysinline { ... }
1143 define void @f() alwaysinline optsize { ... }
1144 define void @f() optsize { ... }
1145 </pre>
1147 <dl>
1149 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1150 the backend should forcibly align the stack pointer. Specify the
1151 desired alignment, which must be a power of two, in parentheses.
1154 <dd>This attribute indicates that the inliner should attempt to inline this
1155 function into callers whenever possible, ignoring any active inlining size
1156 threshold for this caller.</dd>
1159 <dd>This attribute indicates that the function should be 'hotpatchable',
1160 meaning the function can be patched and/or hooked even while it is
1161 loaded into memory. On x86, the function prologue will be preceded
1162 by six bytes of padding and will begin with a two-byte instruction.
1163 Most of the functions in the Windows system DLLs in Windows XP SP2 or
1164 higher were compiled in this fashion.</dd>
1167 <dd>This attribute suppresses lazy symbol binding for the function. This
1168 may make calls to the function faster, at the cost of extra program
1169 startup time if the function is not called during program startup.</dd>
1172 <dd>This attribute indicates that the source code contained a hint that inlining
1173 this function is desirable (such as the "inline" keyword in C/C++). It
1174 is just a hint; it imposes no requirements on the inliner.</dd>
1177 <dd>This attribute disables prologue / epilogue emission for the function.
1178 This can have very system-specific consequences.</dd>
1184 <dd>This attribute indicates that the inliner should never inline this
1185 function in any situation. This attribute may not be used together with
1189 <dd>This attribute indicates that the code generator should not use a red
1190 zone, even if the target-specific ABI normally permits it.</dd>
1193 <dd>This function attribute indicates that the function never returns
1194 normally. This produces undefined behavior at runtime if the function
1195 ever does dynamically return.</dd>
1198 <dd>This function attribute indicates that the function never returns with an
1199 unwind or exceptional control flow. If the function does unwind, its
1200 runtime behavior is undefined.</dd>
1203 <dd>This attribute suggests that optimization passes and code generator passes
1204 make choices that keep the code size of this function low, and otherwise
1205 do optimizations specifically to reduce code size.</dd>
1208 <dd>This attribute indicates that the function computes its result (or decides
1209 to unwind an exception) based strictly on its arguments, without
1210 dereferencing any pointer arguments or otherwise accessing any mutable
1211 state (e.g. memory, control registers, etc) visible to caller functions.
1212 It does not write through any pointer arguments
1214 changes any state visible to callers. This means that it cannot unwind
1219 <dd>This attribute indicates that the function does not write through any
1221 arguments) or otherwise modify any state (e.g. memory, control registers,
1222 etc) visible to caller functions. It may dereference pointer arguments
1223 and read state that may be set in the caller. A readonly function always
1224 returns the same value (or unwinds an exception identically) when called
1225 with the same set of arguments and global state. It cannot unwind an
1230 <dd>This attribute indicates that the function should emit a stack smashing
1232 the stack before the local variables that's checked upon return from the
1233 function to see if it has been overwritten. A heuristic is used to
1234 determine if a function needs stack protectors or not.<br>
1235 <br>
1242 stack smashing protector. This overrides
1244 <br>
1249 </dl>
1251 </div>
1253 <!-- ======================================================================= -->
1254 <h3>
1256 </h3>
1258 <div>
1261 the GCC "file scope inline asm" blocks. These blocks are internally
1262 concatenated by LLVM and treated as a single unit, but may be separated in
1266 module asm "inline asm code goes here"
1267 module asm "more can go here"
1268 </pre>
1270 <p>The strings can contain any character by escaping non-printable characters.
1272 for the number.</p>
1274 <p>The inline asm code is simply printed to the machine code .s file when
1275 assembly code is generated.</p>
1277 </div>
1279 <!-- ======================================================================= -->
1280 <h3>
1282 </h3>
1284 <div>
1286 <p>A module may specify a target specific data layout string that specifies how
1287 data is to be laid out in memory. The syntax for the data layout is
1288 simply:</p>
1291 target datalayout = "<i>layout specification</i>"
1292 </pre>
1295 separated by the minus sign character ('-'). Each specification starts with
1296 a letter and may include other information after the letter to define some
1297 aspect of the data layout. The specifications accepted are as follows:</p>
1299 <dl>
1301 <dd>Specifies that the target lays out data in big-endian form. That is, the
1302 bits with the most significance have the lowest address location.</dd>
1305 <dd>Specifies that the target lays out data in little-endian form. That is,
1306 the bits with the least significance have the lowest address
1307 location.</dd>
1316 <dd>This specifies the alignment for an integer type of a given bit
1320 <dd>This specifies the alignment for a vector type of a given bit
1324 <dd>This specifies the alignment for a floating point type of a given bit
1328 targets.
1331 <dd>This specifies the alignment for an aggregate type of a given bit
1335 <dd>This specifies the alignment for a stack object of a given bit
1339 <dd>This specifies a set of native integer widths for the target CPU
1342 this set are considered to support most general arithmetic
1343 operations efficiently.</dd>
1344 </dl>
1346 <p>When constructing the data layout for a given target, LLVM starts with a
1347 default set of specifications which are then (possibly) overridden by the
1349 are given in this list:</p>
1351 <ul>
1366 </ul>
1368 <p>When LLVM is determining the alignment for a given type, it uses the
1369 following rules:</p>
1371 <ol>
1372 <li>If the type sought is an exact match for one of the specifications, that
1373 specification is used.</li>
1375 <li>If no match is found, and the type sought is an integer type, then the
1376 smallest integer type that is larger than the bitwidth of the sought type
1377 is used. If none of the specifications are larger than the bitwidth then
1378 the the largest integer type is used. For example, given the default
1379 specifications above, the i7 type will use the alignment of i8 (next
1380 largest) while both i65 and i256 will use the alignment of i64 (largest
1381 specified).</li>
1383 <li>If no match is found, and the type sought is a vector type, then the
1384 largest vector type that is smaller than the sought vector type will be
1387 </ol>
1389 </div>
1391 <!-- ======================================================================= -->
1392 <h3>
1394 </h3>
1396 <div>
1398 <p>Any memory access must be done through a pointer value associated
1399 with an address range of the memory access, otherwise the behavior
1400 is undefined. Pointer values are associated with address ranges
1401 according to the following rules:</p>
1403 <ul>
1404 <li>A pointer value is associated with the addresses associated with
1406 <li>An address of a global variable is associated with the address
1407 range of the variable's storage.</li>
1408 <li>The result value of an allocation instruction is associated with
1409 the address range of the allocated storage.</li>
1410 <li>A null pointer in the default address-space is associated with
1411 no address.</li>
1412 <li>An integer constant other than zero or a pointer value returned
1413 from a function not defined within LLVM may be associated with address
1414 ranges allocated through mechanisms other than those provided by
1415 LLVM. Such ranges shall not overlap with any ranges of addresses
1416 allocated by mechanisms provided by LLVM.</li>
1417 </ul>
1420 to the following rules:</p>
1422 <ul>
1423 <li>A pointer value formed from a
1426 <li>The result value of a
1429 <li>A pointer value formed by an
1431 pointer values that contribute (directly or indirectly) to the
1432 computation of the pointer's value.</li>
1434 </ul>
1438 slightly weaker.</p>
1440 <p>LLVM IR does not associate types with memory. The result type of a
1442 alignment of the memory from which to load, as well as the
1443 interpretation of the value. The first operand type of a
1445 and alignment of the store.</p>
1447 <p>Consequently, type-based alias analysis, aka TBAA, aka
1450 additional information which specialized optimization passes may use
1451 to implement type-based alias analysis.</p>
1453 </div>
1455 <!-- ======================================================================= -->
1456 <h3>
1458 </h3>
1460 <div>
1465 The optimizers must not change the number of volatile operations or change their
1466 order of execution relative to other volatile operations. The optimizers
1468 operations. This is not Java's "volatile" and has no cross-thread
1469 synchronization behavior.</p>
1471 </div>
1473 </div>
1475 <!-- *********************************************************************** -->
1477 <!-- *********************************************************************** -->
1479 <div>
1481 <p>The LLVM type system is one of the most important features of the
1482 intermediate representation. Being typed enables a number of optimizations
1483 to be performed on the intermediate representation directly, without having
1484 to do extra analyses on the side before the transformation. A strong type
1485 system makes it easier to read the generated code and enables novel analyses
1486 and transformations that are not feasible to perform on normal three address
1487 code representations.</p>
1489 <!-- ======================================================================= -->
1490 <h3>
1492 </h3>
1494 <div>
1499 <tbody>
1501 <tr>
1504 </tr>
1505 <tr>
1508 </tr>
1509 <tr>
1519 </td>
1520 </tr>
1521 <tr>
1529 </tr>
1530 <tr>
1538 </td>
1539 </tr>
1540 </tbody>
1541 </table>
1544 important. Values of these types are the only ones which can be produced by
1545 instructions.</p>
1547 </div>
1549 <!-- ======================================================================= -->
1550 <h3>
1552 </h3>
1554 <div>
1556 <p>The primitive types are the fundamental building blocks of the LLVM
1557 system.</p>
1559 <!-- _______________________________________________________________________ -->
1560 <h4>
1562 </h4>
1564 <div>
1567 <p>The integer type is a very simple type that simply specifies an arbitrary
1568 bit width for the integer type desired. Any bit width from 1 bit to
1572 <pre>
1573 iN
1574 </pre>
1577 value.</p>
1584 </tr>
1588 </tr>
1592 </tr>
1593 </table>
1595 </div>
1597 <!-- _______________________________________________________________________ -->
1598 <h4>
1600 </h4>
1602 <div>
1604 <table>
1605 <tbody>
1612 </tbody>
1613 </table>
1615 </div>
1617 <!-- _______________________________________________________________________ -->
1618 <h4>
1620 </h4>
1622 <div>
1625 <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>
1628 <pre>
1629 x86mmx
1630 </pre>
1632 </div>
1634 <!-- _______________________________________________________________________ -->
1635 <h4>
1637 </h4>
1639 <div>
1645 <pre>
1646 void
1647 </pre>
1649 </div>
1651 <!-- _______________________________________________________________________ -->
1652 <h4>
1654 </h4>
1656 <div>
1662 <pre>
1663 label
1664 </pre>
1666 </div>
1668 <!-- _______________________________________________________________________ -->
1669 <h4>
1671 </h4>
1673 <div>
1676 <p>The metadata type represents embedded metadata. No derived types may be
1678 arguments.
1681 <pre>
1682 metadata
1683 </pre>
1685 </div>
1687 </div>
1689 <!-- ======================================================================= -->
1690 <h3>
1692 </h3>
1694 <div>
1696 <p>The real power in LLVM comes from the derived types in the system. This is
1697 what allows a programmer to represent arrays, functions, pointers, and other
1698 useful types. Each of these types contain one or more element types which
1699 may be a primitive type, or another derived type. For example, it is
1700 possible to have a two dimensional array, using an array as the element type
1701 of another array.</p>
1703 </div>
1706 <!-- _______________________________________________________________________ -->
1707 <h4>
1709 </h4>
1711 <div>
1713 <p>Aggregate Types are a subset of derived types that can contain multiple
1716 aggregate types.</p>
1718 </div>
1720 <!-- _______________________________________________________________________ -->
1721 <h4>
1723 </h4>
1725 <div>
1728 <p>The array type is a very simple derived type that arranges elements
1729 sequentially in memory. The array type requires a size (number of elements)
1730 and an underlying data type.</p>
1733 <pre>
1735 </pre>
1738 be any type with a size.</p>
1745 </tr>
1749 </tr>
1753 </tr>
1754 </table>
1760 </tr>
1764 </tr>
1768 </tr>
1769 </table>
1771 <p>There is no restriction on indexing beyond the end of the array implied by
1772 a static type (though there are restrictions on indexing beyond the bounds
1773 of an allocated object in some cases). This means that single-dimension
1774 'variable sized array' addressing can be implemented in LLVM with a zero
1775 length array type. An implementation of 'pascal style arrays' in LLVM could
1778 </div>
1780 <!-- _______________________________________________________________________ -->
1781 <h4>
1783 </h4>
1785 <div>
1788 <p>The function type can be thought of as a function signature. It consists of
1789 a return type and a list of formal parameter types. The return type of a
1790 function type is a first class type or a void type.</p>
1793 <pre>
1795 </pre>
1799 which indicates that the function takes a variable number of arguments.
1800 Variable argument functions can access their arguments with
1810 </td>
1813 </tt></td>
1817 </td>
1823 LLVM.
1824 </td>
1829 </td>
1830 </tr>
1831 </table>
1833 </div>
1835 <!-- _______________________________________________________________________ -->
1836 <h4>
1838 </h4>
1840 <div>
1843 <p>The structure type is used to represent a collection of data members together
1844 in memory. The elements of a structure may be any type that has a size.</p>
1849 Structures in registers are accessed using the
1854 alignment of the struct is one byte, and that there is no padding between
1855 the elements. In non-packed structs, padding between field types is defined
1856 by the target data string to match the underlying processor.</p>
1860 are always defined at the top level with a name. Anonmyous types are uniqued
1861 by their contents and can never be recursive since there is no way to write
1862 one. Named types can be recursive.
1863 </p>
1866 <pre>
1869 </pre>
1876 </tr>
1883 </tr>
1887 </tr>
1888 </table>
1890 </div>
1892 <!-- _______________________________________________________________________ -->
1893 <h4>
1895 </h4>
1897 <div>
1900 <p>Opaque types are used to represent named structure types that do not have a
1901 body specified. This corresponds (for example) to the C notion of a forward
1902 declared structure.</p>
1905 <pre>
1906 %X = type opaque
1907 %52 = type opaque
1908 </pre>
1915 </tr>
1916 </table>
1918 </div>
1922 <!-- _______________________________________________________________________ -->
1923 <h4>
1925 </h4>
1927 <div>
1930 <p>The pointer type is used to specify memory locations.
1931 Pointers are commonly used to reference objects in memory.</p>
1933 <p>Pointer types may have an optional address space attribute defining the
1934 numbered address space where the pointed-to object resides. The default
1935 address space is number zero. The semantics of non-zero address
1936 spaces are target-specific.</p>
1942 <pre>
1944 </pre>
1952 </tr>
1958 </tr>
1963 </tr>
1964 </table>
1966 </div>
1968 <!-- _______________________________________________________________________ -->
1969 <h4>
1971 </h4>
1973 <div>
1976 <p>A vector type is a simple derived type that represents a vector of elements.
1977 Vector types are used when multiple primitive data are operated in parallel
1978 using a single instruction (SIMD). A vector type requires a size (number of
1979 elements) and an underlying primitive data type. Vector types are considered
1983 <pre>
1985 </pre>
1988 may be any integer or floating point type. Vectors of size zero are not
1989 allowed, and pointers are not allowed as the element type.</p>
1996 </tr>
2000 </tr>
2004 </tr>
2005 </table>
2007 </div>
2009 <!-- *********************************************************************** -->
2011 <!-- *********************************************************************** -->
2013 <div>
2015 <p>LLVM has several different basic types of constants. This section describes
2016 them all and their syntax.</p>
2018 <!-- ======================================================================= -->
2019 <h3>
2021 </h3>
2023 <div>
2025 <dl>
2033 with integer types.</dd>
2037 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2038 notation (see below). The assembler requires the exact decimal value of a
2039 floating-point constant. For example, the assembler accepts 1.25 but
2046 </dl>
2048 <p>The one non-intuitive notation for constants is the hexadecimal form of
2049 floating point constants. For example, the form '<tt>double
2052 constants are required (and the only time that they are generated by the
2053 disassembler) is when a floating point constant must be emitted but it cannot
2054 be represented as a decimal floating point number in a reasonable number of
2055 digits. For example, NaN's, infinities, and other special values are
2056 represented in their IEEE hexadecimal format so that assembly and disassembly
2057 do not cause any bits to change in the constants.</p>
2059 <p>When using the hexadecimal form, constants of types float and double are
2060 represented using the 16-digit form shown above (which matches the IEEE754
2061 representation for double); float values must, however, be exactly
2062 representable as IEE754 single precision. Hexadecimal format is always used
2063 for long double, and there are three forms of long double. The 80-bit format
2065 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2068 currently supported target uses this format. Long doubles will only work if
2069 they match the long double format on your target. All hexadecimal formats
2070 are big-endian (sign bit at the left).</p>
2073 </div>
2075 <!-- ======================================================================= -->
2076 <h3>
2079 </h3>
2081 <div>
2083 <p>Complex constants are a (potentially recursive) combination of simple
2084 constants and smaller complex constants.</p>
2086 <dl>
2088 <dd>Structure constants are represented with notation similar to structure
2089 type definitions (a comma separated list of elements, surrounded by braces
2093 the number and types of elements must match those specified by the
2094 type.</dd>
2097 <dd>Array constants are represented with notation similar to array type
2098 definitions (a comma separated list of elements, surrounded by square
2101 the number and types of elements must match those specified by the
2102 type.</dd>
2105 <dd>Vector constants are represented with notation similar to vector type
2106 definitions (a comma separated list of elements, surrounded by
2108 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2110 elements must match those specified by the type.</dd>
2116 This is often used to avoid having to print large zero initializers
2117 (e.g. for large arrays) and is always exactly equivalent to using explicit
2118 zero initializers.</dd>
2121 <dd>A metadata node is a structure-like constant with
2124 be interpreted as part of the instruction stream, metadata is a place to
2125 attach additional information such as debug info.</dd>
2126 </dl>
2128 </div>
2130 <!-- ======================================================================= -->
2131 <h3>
2133 </h3>
2135 <div>
2139 (link-time) constants. These constants are explicitly referenced when
2142 legal LLVM file:</p>
2145 @X = global i32 17
2146 @Y = global i32 42
2147 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2148 </pre>
2150 </div>
2152 <!-- ======================================================================= -->
2153 <h3>
2155 </h3>
2157 <div>
2160 indicates that the user of the value may receive an unspecified bit-pattern.
2164 <p>Undefined values are useful because they indicate to the compiler that the
2165 program is well defined no matter what value is used. This gives the
2166 compiler more freedom to optimize. Here are some examples of (potentially
2167 surprising) transformations that are valid (in pseudo IR):</p>
2171 %A = add %X, undef
2172 %B = sub %X, undef
2173 %C = xor %X, undef
2174 Safe:
2175 %A = undef
2176 %B = undef
2177 %C = undef
2178 </pre>
2180 <p>This is safe because all of the output bits are affected by the undef bits.
2181 Any output bit can have a zero or one depending on the input bits.</p>
2184 %A = or %X, undef
2185 %B = and %X, undef
2186 Safe:
2187 %A = -1
2188 %B = 0
2189 Unsafe:
2190 %A = undef
2191 %B = undef
2192 </pre>
2194 <p>These logical operations have bits that are not always affected by the input.
2205 %A = select undef, %X, %Y
2206 %B = select undef, 42, %Y
2207 %C = select %X, %Y, undef
2208 Safe:
2209 %A = %X (or %Y)
2210 %B = 42 (or %Y)
2211 %C = %Y
2212 Unsafe:
2213 %A = undef
2214 %B = undef
2215 %C = undef
2216 </pre>
2225 eliminated.</p>
2228 %A = xor undef, undef
2230 %B = undef
2231 %C = xor %B, %B
2233 %D = undef
2234 %E = icmp lt %D, 4
2235 %F = icmp gte %D, 4
2237 Safe:
2238 %A = undef
2239 %B = undef
2240 %C = undef
2241 %D = undef
2242 %E = undef
2243 %F = undef
2244 </pre>
2247 necessarily the same. This can be surprising to people (and also matches C
2248 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2251 its value over its "live range". This is true because the variable doesn't
2253 from arbitrary registers that happen to be around when needed, so the value
2255 need to have the same semantics or the core LLVM "replace all uses with"
2256 concept would not hold.</p>
2259 %A = fdiv undef, %X
2260 %B = fdiv %X, undef
2261 Safe:
2262 %A = undef
2263 b: unreachable
2264 </pre>
2271 defined on SNaN's. However, in the second example, we can make a more
2273 arbitrary value, we are allowed to assume that it could be zero. Since a
2275 the operation does not execute at all. This allows us to delete the divide and
2276 all code after it. Because the undefined operation "can't happen", the
2277 optimizer can assume that it occurs in dead code.</p>
2280 a: store undef -> %X
2281 b: store %X -> undef
2282 Safe:
2284 b: unreachable
2285 </pre>
2288 undefined value can be assumed to not have any effect; we can assume that the
2289 value is overwritten with bits that happen to match what was already there.
2291 memory, therefore, it has undefined behavior.</p>
2293 </div>
2295 <!-- ======================================================================= -->
2296 <h3>
2298 </h3>
2300 <div>
2303 instead of representing an unspecified bit pattern, they represent the
2304 fact that an instruction or constant expression which cannot evoke side
2305 effects has nevertheless detected a condition which results in undefined
2306 behavior.</p>
2308 <p>There is currently no way of representing a trap value in the IR; they
2309 only exist when produced by operations such as
2314 <ul>
2316 their operands.</li>
2319 to their dynamic predecessor basic block.</li>
2321 <li>Function arguments depend on the corresponding actual argument values in
2322 the dynamic callers of their functions.</li>
2326 control back to them.</li>
2330 or exception-throwing call instructions that dynamically transfer control
2331 back to them.</li>
2333 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2334 referenced memory addresses, following the order in the IR
2335 (including loads and stores implied by intrinsics such as
2338 <!-- TODO: In the case of multiple threads, this only applies if the store
2339 "happens-before" the load or store. -->
2341 <!-- TODO: floating-point exception state -->
2343 <li>An instruction with externally visible side effects depends on the most
2344 recent preceding instruction with externally visible side effects, following
2345 the order in the IR. (This includes
2350 if the terminator instruction has multiple successors and the instruction
2351 is always executed when control transfers to one of the successors, and
2352 may not be executed when control is transferred to another.</li>
2355 instruction if the set of instructions it otherwise depends on would be
2356 different if the terminator had transferred control to a different
2357 successor.</li>
2361 </ul>
2363 <p>Whenever a trap value is generated, all values which depend on it evaluate
2364 to trap. If they have side effects, the evoke their side effects as if each
2365 operand with a trap value were undef. If they have externally-visible side
2366 effects, the behavior is undefined.</p>
2371 entry:
2374 %trap_yet_again = getelementptr i32* @h, i32 %still_trap
2375 store i32 0, i32* %trap_yet_again ; undefined behavior
2377 store i32 %trap, i32* @g ; Trap value conceptually stored to memory.
2378 %trap2 = load i32* @g ; Returns a trap value, not just undef.
2380 volatile store i32 %trap, i32* @g ; External observation; undefined behavior.
2382 %narrowaddr = bitcast i32* @g to i16*
2383 %wideaddr = bitcast i32* @g to i64*
2384 %trap3 = load i16* %narrowaddr ; Returns a trap value.
2385 %trap4 = load i64* %wideaddr ; Returns a trap value.
2387 %cmp = icmp slt i32 %trap, 0 ; Returns a trap value.
2388 br i1 %cmp, label %true, label %end ; Branch to either destination.
2390 true:
2391 volatile store i32 0, i32* @g ; This is control-dependent on %cmp, so
2392 ; it has undefined behavior.
2393 br label %end
2395 end:
2397 ; Both edges into this PHI are
2398 ; control-dependent on %cmp, so this
2399 ; always results in a trap value.
2401 volatile store i32 0, i32* @g ; This would depend on the store in %true
2402 ; if %cmp is true, or the store in %entry
2403 ; otherwise, so this is undefined behavior.
2405 br i1 %cmp, label %second_true, label %second_end
2406 ; The same branch again, but this time the
2407 ; true block doesn't have side effects.
2409 second_true:
2410 ; No side effects!
2411 ret void
2413 second_end:
2414 volatile store i32 0, i32* @g ; This time, the instruction always depends
2415 ; on the store in %end. Also, it is
2416 ; control-equivalent to %end, so this is
2417 ; well-defined (again, ignoring earlier
2418 ; undefined behavior in this example).
2419 </pre>
2421 </div>
2423 <!-- ======================================================================= -->
2424 <h3>
2426 </h3>
2428 <div>
2433 basic block in the specified function, and always has an i8* type. Taking
2434 the address of the entry block is illegal.</p>
2436 <p>This value only has defined behavior when used as an operand to the
2438 comparisons against null. Pointer equality tests between labels addresses
2439 results in undefined behavior — though, again, comparison against null
2440 is ok, and no label is equal to the null pointer. This may be passed around
2441 as an opaque pointer sized value as long as the bits are not inspected. This
2444 instruction.</p>
2446 <p>Finally, some targets may provide defined semantics when using the value as
2447 the operand to an inline assembly, but that is target specific.</p>
2449 </div>
2452 <!-- ======================================================================= -->
2453 <h3>
2455 </h3>
2457 <div>
2459 <p>Constant expressions are used to allow expressions involving other constants
2460 to be used as constants. Constant expressions may be of
2462 operation that does not have side effects (e.g. load and call are not
2463 supported). The following is the syntax for constant expressions:</p>
2465 <dl>
2467 <dd>Truncate a constant to another type. The bit size of CST must be larger
2468 than the bit size of TYPE. Both types must be integers.</dd>
2471 <dd>Zero extend a constant to another type. The bit size of CST must be
2472 smaller than the bit size of TYPE. Both types must be integers.</dd>
2475 <dd>Sign extend a constant to another type. The bit size of CST must be
2476 smaller than the bit size of TYPE. Both types must be integers.</dd>
2479 <dd>Truncate a floating point constant to another floating point type. The
2480 size of CST must be larger than the size of TYPE. Both types must be
2481 floating point.</dd>
2484 <dd>Floating point extend a constant to another type. The size of CST must be
2485 smaller or equal to the size of TYPE. Both types must be floating
2486 point.</dd>
2489 <dd>Convert a floating point constant to the corresponding unsigned integer
2490 constant. TYPE must be a scalar or vector integer type. CST must be of
2491 scalar or vector floating point type. Both CST and TYPE must be scalars,
2492 or vectors of the same number of elements. If the value won't fit in the
2493 integer type, the results are undefined.</dd>
2496 <dd>Convert a floating point constant to the corresponding signed integer
2497 constant. TYPE must be a scalar or vector integer type. CST must be of
2498 scalar or vector floating point type. Both CST and TYPE must be scalars,
2499 or vectors of the same number of elements. If the value won't fit in the
2500 integer type, the results are undefined.</dd>
2503 <dd>Convert an unsigned integer constant to the corresponding floating point
2504 constant. TYPE must be a scalar or vector floating point type. CST must be
2505 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2506 vectors of the same number of elements. If the value won't fit in the
2507 floating point type, the results are undefined.</dd>
2510 <dd>Convert a signed integer constant to the corresponding floating point
2511 constant. TYPE must be a scalar or vector floating point type. CST must be
2512 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2513 vectors of the same number of elements. If the value won't fit in the
2514 floating point type, the results are undefined.</dd>
2517 <dd>Convert a pointer typed constant to the corresponding integer constant
2523 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2524 type. CST must be of integer type. The CST value is zero extended,
2525 truncated, or unchanged to make it fit in a pointer size. This one is
2529 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2537 instruction, the index list may have zero or more indexes, which are
2551 constants.</dd>
2555 constants.</dd>
2559 constants.</dd>
2563 constants. The index list is interpreted in a similar manner as indices in
2565 index value must be specified.</dd>
2569 constants. The index list is interpreted in a similar manner as indices in
2571 index value must be specified.</dd>
2574 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2577 on operands are the same as those for the corresponding instruction
2578 (e.g. no bitwise operations on floating point values are allowed).</dd>
2579 </dl>
2581 </div>
2583 </div>
2585 <!-- *********************************************************************** -->
2587 <!-- *********************************************************************** -->
2588 <div>
2589 <!-- ======================================================================= -->
2590 <h3>
2592 </h3>
2594 <div>
2596 <p>LLVM supports inline assembler expressions (as opposed
2598 a special value. This value represents the inline assembler as a string
2599 (containing the instructions to emit), a list of operand constraints (stored
2600 as a string), a flag that indicates whether or not the inline asm
2601 expression has side effects, and a flag indicating whether the function
2602 containing the asm needs to align its stack conservatively. An example
2603 inline assembler expression is:</p>
2607 </pre>
2611 have:</p>
2615 </pre>
2617 <p>Inline asms with side effects not visible in the constraint list must be
2618 marked as having side effects. This is done through the use of the
2623 </pre>
2625 <p>In some cases inline asms will contain code that will not work unless the
2626 stack is aligned in some way, such as calls or SSE instructions on x86,
2627 yet will not contain code that does that alignment within the asm.
2628 The compiler should make conservative assumptions about what the asm might
2629 contain and should generate its usual stack alignment code in the prologue
2634 </pre>
2637 first.</p>
2639 <p>TODO: The format of the asm and constraints string still need to be
2640 documented here. Constraints on what can be done (e.g. duplication, moving,
2641 etc need to be documented). This is probably best done by reference to
2642 another document that covers inline asm from a holistic perspective.</p>
2644 <h4>
2646 </h4>
2648 <div>
2651 attached to it that contains a list of constant integers. If present, the
2652 code generator will use the integer as the location cookie value when report
2653 errors through the LLVMContext error reporting mechanisms. This allows a
2654 front-end to correlate backend errors that occur with inline asm back to the
2655 source code that produced it. For example:</p>
2659 ...
2661 </pre>
2663 <p>It is up to the front-end to make sense of the magic numbers it places in the
2664 IR. If the MDNode contains multiple constants, the code generator will use
2665 the one that corresponds to the line of the asm that the error occurs on.</p>
2667 </div>
2669 </div>
2671 <!-- ======================================================================= -->
2672 <h3>
2674 </h3>
2676 <div>
2678 <p>LLVM IR allows metadata to be attached to instructions in the program that
2679 can convey extra information about the code to the optimizers and code
2680 generator. One example application of metadata is source-level debug
2681 information. There are two metadata primitives: strings and nodes. All
2685 <p>A metadata string is a string surrounded by double quotes. It can contain
2689 <p>Metadata nodes are represented with notation similar to structure constants
2690 (a comma separated list of elements, surrounded by braces and preceded by an
2695 metadata nodes, which can be looked up in the module symbol table. For
2696 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2699 function is using two metadata arguments.</p>
2702 <pre>
2704 </pre>
2705 </div>
2711 <pre>
2713 </pre>
2714 </div>
2716 </div>
2718 </div>
2720 <!-- *********************************************************************** -->
2721 <h2>
2723 </h2>
2724 <!-- *********************************************************************** -->
2725 <div>
2727 code generation or other IR semantics. These are documented here. All globals
2728 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2729 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2730 by LLVM.</p>
2732 <!-- ======================================================================= -->
2733 <h3>
2735 </h3>
2737 <div>
2741 pointers to global variables and functions which may optionally have a pointer
2742 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2744 <pre>
2745 @X = global i8 4
2746 @Y = global i32 123
2748 @llvm.used = appending global [2 x i8*] [
2749 i8* @X,
2750 i8* bitcast (i32* @Y to i8*)
2751 ], section "llvm.metadata"
2752 </pre>
2755 compiler, assembler, and linker are required to treat the symbol as if there is
2756 a reference to the global that it cannot see. For example, if a variable has
2758 list, it cannot be deleted. This is commonly used to represent references from
2759 inline asms and other things the compiler cannot "see", and corresponds to
2762 <p>On some targets, the code generator must emit a directive to the assembler or
2763 object file to prevent the assembler and linker from molesting the symbol.</p>
2765 </div>
2767 <!-- ======================================================================= -->
2768 <h3>
2771 </a>
2772 </h3>
2774 <div>
2778 touching the symbol. On targets that support it, this allows an intelligent
2779 linker to optimize references to the symbol without being impeded as it would be
2782 <p>This is a rare construct that should only be used in rare circumstances, and
2783 should not be exposed to source languages.</p>
2785 </div>
2787 <!-- ======================================================================= -->
2788 <h3>
2790 </h3>
2792 <div>
2793 <pre>
2794 %0 = type { i32, void ()* }
2796 </pre>
2797 <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.
2798 </p>
2800 </div>
2802 <!-- ======================================================================= -->
2803 <h3>
2805 </h3>
2807 <div>
2808 <pre>
2809 %0 = type { i32, void ()* }
2811 </pre>
2813 <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.
2814 </p>
2816 </div>
2818 </div>
2820 <!-- *********************************************************************** -->
2822 <!-- *********************************************************************** -->
2824 <div>
2826 <p>The LLVM instruction set consists of several different classifications of
2833 <!-- ======================================================================= -->
2834 <h3>
2836 </h3>
2838 <div>
2841 in a program ends with a "Terminator" instruction, which indicates which
2842 block should be executed after the current block is finished. These
2844 control flow, not values (the one exception being the
2847 <p>There are seven different terminator instructions: the
2856 <!-- _______________________________________________________________________ -->
2857 <h4>
2859 </h4>
2861 <div>
2864 <pre>
2867 </pre>
2871 a value) from a function back to the caller.</p>
2874 value and then causes control flow, and one that just causes control flow to
2875 occur.</p>
2879 return value. The type of the return value must be a
2884 value or a return value with a type that does not match its type, or if it
2886 return value.</p>
2890 the calling function's context. If the caller is a
2892 instruction after the call. If the caller was an
2894 the beginning of the "normal" destination block. If the instruction returns
2895 a value, that value shall set the call or invoke instruction's return
2896 value.</p>
2899 <pre>
2903 </pre>
2905 </div>
2906 <!-- _______________________________________________________________________ -->
2907 <h4>
2909 </h4>
2911 <div>
2914 <pre>
2915 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2916 </pre>
2920 different basic block in the current function. There are two forms of this
2921 instruction, corresponding to a conditional branch and an unconditional
2922 branch.</p>
2928 target.</p>
2937 <pre>
2938 Test:
2940 br i1 %cond, label %IfEqual, label %IfUnequal
2941 IfEqual:
2943 IfUnequal:
2945 </pre>
2947 </div>
2949 <!-- _______________________________________________________________________ -->
2950 <h4>
2952 </h4>
2954 <div>
2957 <pre>
2958 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2959 </pre>
2964 instruction, allowing a branch to occur to one of many possible
2965 destinations.</p>
2971 The table is not allowed to contain duplicate constant entries.</p>
2976 is searched for the given value. If the value is found, control flow is
2977 transferred to the corresponding destination; otherwise, control flow is
2978 transferred to the default destination.</p>
2981 <p>Depending on properties of the target machine and the particular
2983 different ways. For example, it could be generated as a series of chained
2984 conditional branches or with a lookup table.</p>
2987 <pre>
2990 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2993 switch i32 0, label %dest [ ]
2996 switch i32 %val, label %otherwise [ i32 0, label %onzero
2997 i32 1, label %onone
2998 i32 2, label %ontwo ]
2999 </pre>
3001 </div>
3004 <!-- _______________________________________________________________________ -->
3005 <h4>
3007 </h4>
3009 <div>
3012 <pre>
3014 </pre>
3019 within the current function, whose address is specified by
3026 rest of the arguments indicate the full set of possible destinations that the
3027 address may point to. Blocks are allowed to occur multiple times in the
3028 destination list, though this isn't particularly useful.</p>
3030 <p>This destination list is required so that dataflow analysis has an accurate
3031 understanding of the CFG.</p>
3035 <p>Control transfers to the block specified in the address argument. All
3036 possible destination blocks must be listed in the label list, otherwise this
3037 instruction has undefined behavior. This implies that jumps to labels
3038 defined in other functions have undefined behavior as well.</p>
3045 <pre>
3046 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3047 </pre>
3049 </div>
3052 <!-- _______________________________________________________________________ -->
3053 <h4>
3055 </h4>
3057 <div>
3060 <pre>
3061 <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>]
3063 </pre>
3067 function, with the possibility of control flow transfer to either the
3070 control flow will return to the "normal" label. If the callee (or any
3072 instruction, control is interrupted and continued at the dynamically nearest
3078 <ol>
3080 convention</a> the call should use. If none is specified, the call
3081 defaults to using C calling conventions.</li>
3088 function value being invoked. In most cases, this is a direct function
3090 off an arbitrary pointer to function value.</li>
3093 function to be invoked. </li>
3096 signature argument types and parameter attributes. All arguments must be
3098 signature indicates the function accepts a variable number of arguments,
3099 the extra arguments can be specified.</li>
3110 </ol>
3113 <p>This instruction is designed to operate as a standard
3115 primary difference is that it establishes an association with a label, which
3116 is used by the runtime library to unwind the stack.</p>
3118 <p>This instruction is used in languages with destructors to ensure that proper
3120 exception. Additionally, this is important for implementation of
3123 <p>For the purposes of the SSA form, the definition of the value returned by the
3125 block to the "normal" label. If the callee unwinds then no return value is
3126 available.</p>
3128 <p>Note that the code generator does not yet completely support unwind, and
3129 that the invoke/unwind semantics are likely to change in future versions.</p>
3132 <pre>
3133 %retval = invoke i32 @Test(i32 15) to label %Continue
3137 </pre>
3139 </div>
3141 <!-- _______________________________________________________________________ -->
3143 <h4>
3145 </h4>
3147 <div>
3150 <pre>
3151 unwind
3152 </pre>
3156 at the first callee in the dynamic call stack which used
3158 This is primarily used to implement exception handling.</p>
3162 immediately halt. The dynamic call stack is then searched for the
3164 Once found, execution continues at the "exceptional" destination block
3166 instruction in the dynamic call chain, undefined behavior results.</p>
3168 <p>Note that the code generator does not yet completely support unwind, and
3169 that the invoke/unwind semantics are likely to change in future versions.</p>
3171 </div>
3173 <!-- _______________________________________________________________________ -->
3175 <h4>
3177 </h4>
3179 <div>
3182 <pre>
3183 unreachable
3184 </pre>
3188 instruction is used to inform the optimizer that a particular portion of the
3189 code is not reachable. This can be used to indicate that the code after a
3190 no-return function cannot be reached, and other facts.</p>
3195 </div>
3197 </div>
3199 <!-- ======================================================================= -->
3200 <h3>
3202 </h3>
3204 <div>
3206 <p>Binary operators are used to do most of the computation in a program. They
3207 require two operands of the same type, execute an operation on them, and
3208 produce a single value. The operands might represent multiple data, as is
3210 has the same type as its operands.</p>
3214 <!-- _______________________________________________________________________ -->
3215 <h4>
3217 </h4>
3219 <div>
3222 <pre>
3227 </pre>
3235 integer values. Both arguments must have identical types.</p>
3240 <p>If the sum has unsigned overflow, the result returned is the mathematical
3243 <p>Because LLVM integers use a two's complement representation, this instruction
3244 is appropriate for both signed and unsigned integers.</p>
3250 respectively, occurs.</p>
3253 <pre>
3255 </pre>
3257 </div>
3259 <!-- _______________________________________________________________________ -->
3260 <h4>
3262 </h4>
3264 <div>
3267 <pre>
3269 </pre>
3277 floating point values. Both arguments must have identical types.</p>
3283 <pre>
3285 </pre>
3287 </div>
3289 <!-- _______________________________________________________________________ -->
3290 <h4>
3292 </h4>
3294 <div>
3297 <pre>
3302 </pre>
3306 operands.</p>
3310 representations.</p>
3315 integer values. Both arguments must have identical types.</p>
3320 <p>If the difference has unsigned overflow, the result returned is the
3322 result.</p>
3324 <p>Because LLVM integers use a two's complement representation, this instruction
3325 is appropriate for both signed and unsigned integers.</p>
3331 respectively, occurs.</p>
3334 <pre>
3337 </pre>
3339 </div>
3341 <!-- _______________________________________________________________________ -->
3342 <h4>
3344 </h4>
3346 <div>
3349 <pre>
3351 </pre>
3355 operands.</p>
3359 representations.</p>
3364 floating point values. Both arguments must have identical types.</p>
3370 <pre>
3373 </pre>
3375 </div>
3377 <!-- _______________________________________________________________________ -->
3378 <h4>
3380 </h4>
3382 <div>
3385 <pre>
3390 </pre>
3398 integer values. Both arguments must have identical types.</p>
3403 <p>If the result of the multiplication has unsigned overflow, the result
3405 width of the result.</p>
3407 <p>Because LLVM integers use a two's complement representation, and the result
3408 is the same width as the operands, this instruction returns the correct
3409 result for both signed and unsigned integers. If a full product
3411 be sign-extended or zero-extended as appropriate to the width of the full
3412 product.</p>
3418 respectively, occurs.</p>
3421 <pre>
3423 </pre>
3425 </div>
3427 <!-- _______________________________________________________________________ -->
3428 <h4>
3430 </h4>
3432 <div>
3435 <pre>
3437 </pre>
3445 floating point values. Both arguments must have identical types.</p>
3451 <pre>
3453 </pre>
3455 </div>
3457 <!-- _______________________________________________________________________ -->
3458 <h4>
3460 </h4>
3462 <div>
3465 <pre>
3468 </pre>
3476 values. Both arguments must have identical types.</p>
3481 <p>Note that unsigned integer division and signed integer division are distinct
3492 <pre>
3494 </pre>
3496 </div>
3498 <!-- _______________________________________________________________________ -->
3499 <h4>
3501 </h4>
3503 <div>
3506 <pre>
3509 </pre>
3517 values. Both arguments must have identical types.</p>
3520 <p>The value produced is the signed integer quotient of the two operands rounded
3521 towards zero.</p>
3523 <p>Note that signed integer division and unsigned integer division are distinct
3526 <p>Division by zero leads to undefined behavior. Overflow also leads to
3527 undefined behavior; this is a rare case, but can occur, for example, by doing
3532 be rounded.</p>
3535 <pre>
3537 </pre>
3539 </div>
3541 <!-- _______________________________________________________________________ -->
3542 <h4>
3544 </h4>
3546 <div>
3549 <pre>
3551 </pre>
3559 floating point values. Both arguments must have identical types.</p>
3565 <pre>
3567 </pre>
3569 </div>
3571 <!-- _______________________________________________________________________ -->
3572 <h4>
3574 </h4>
3576 <div>
3579 <pre>
3581 </pre>
3585 division of its two arguments.</p>
3590 values. Both arguments must have identical types.</p>
3594 This instruction always performs an unsigned division to get the
3595 remainder.</p>
3597 <p>Note that unsigned integer remainder and signed integer remainder are
3603 <pre>
3605 </pre>
3607 </div>
3609 <!-- _______________________________________________________________________ -->
3610 <h4>
3612 </h4>
3614 <div>
3617 <pre>
3619 </pre>
3623 division of its two operands. This instruction can also take
3625 elements must be integers.</p>
3630 values. Both arguments must have identical types.</p>
3637 For more information about the difference,
3639 Math Forum</a>. For a table of how this is implemented in various languages,
3643 <p>Note that signed integer remainder and unsigned integer remainder are
3646 <p>Taking the remainder of a division by zero leads to undefined behavior.
3647 Overflow also leads to undefined behavior; this is a rare case, but can
3648 occur, for example, by taking the remainder of a 32-bit division of
3650 lets srem be implemented using instructions that return both the result of
3651 the division and the remainder.)</p>
3654 <pre>
3656 </pre>
3658 </div>
3660 <!-- _______________________________________________________________________ -->
3661 <h4>
3663 </h4>
3665 <div>
3668 <pre>
3670 </pre>
3674 its two operands.</p>
3679 floating point values. Both arguments must have identical types.</p>
3683 has the same sign as the dividend.</p>
3686 <pre>
3688 </pre>
3690 </div>
3692 </div>
3694 <!-- ======================================================================= -->
3695 <h3>
3697 </h3>
3699 <div>
3701 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3702 program. They are generally very efficient instructions and can commonly be
3703 strength reduced from other instructions. They require two operands of the
3704 same type, execute an operation on them, and produce a single value. The
3705 resulting value is the same type as its operands.</p>
3707 <!-- _______________________________________________________________________ -->
3708 <h4>
3710 </h4>
3712 <div>
3715 <pre>
3720 </pre>
3724 a specified number of bits.</p>
3734 is (statically or dynamically) negative or equal to or larger than the number
3743 with the resultant sign bit. As such, NUW/NSW have the same semantics as
3744 they would if the shift were expressed as a mul instruction with the same
3748 <pre>
3753 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3754 </pre>
3756 </div>
3758 <!-- _______________________________________________________________________ -->
3759 <h4>
3761 </h4>
3763 <div>
3766 <pre>
3769 </pre>
3773 operand shifted to the right a specified number of bits with zero fill.</p>
3781 <p>This instruction always performs a logical shift right operation. The most
3782 significant bits of the result will be filled with zero bits after the shift.
3790 shifted out are non-zero.</p>
3794 <pre>
3800 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3801 </pre>
3803 </div>
3805 <!-- _______________________________________________________________________ -->
3806 <h4>
3808 </h4>
3810 <div>
3813 <pre>
3816 </pre>
3820 operand shifted to the right a specified number of bits with sign
3821 extension.</p>
3829 <p>This instruction always performs an arithmetic shift right operation, The
3830 most significant bits of the result will be filled with the sign bit
3838 shifted out are non-zero.</p>
3841 <pre>
3847 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3848 </pre>
3850 </div>
3852 <!-- _______________________________________________________________________ -->
3853 <h4>
3855 </h4>
3857 <div>
3860 <pre>
3862 </pre>
3866 operands.</p>
3871 values. Both arguments must have identical types.</p>
3877 <tbody>
3878 <tr>
3882 </tr>
3883 <tr>
3887 </tr>
3888 <tr>
3892 </tr>
3893 <tr>
3897 </tr>
3898 <tr>
3902 </tr>
3903 </tbody>
3904 </table>
3907 <pre>
3911 </pre>
3912 </div>
3913 <!-- _______________________________________________________________________ -->
3914 <h4>
3916 </h4>
3918 <div>
3921 <pre>
3923 </pre>
3927 two operands.</p>
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>
3972 </pre>
3974 </div>
3976 <!-- _______________________________________________________________________ -->
3977 <h4>
3979 </h4>
3981 <div>
3984 <pre>
3986 </pre>
3996 values. Both arguments must have identical types.</p>
4002 <tbody>
4003 <tr>
4007 </tr>
4008 <tr>
4012 </tr>
4013 <tr>
4017 </tr>
4018 <tr>
4022 </tr>
4023 <tr>
4027 </tr>
4028 </tbody>
4029 </table>
4032 <pre>
4037 </pre>
4039 </div>
4041 </div>
4043 <!-- ======================================================================= -->
4044 <h3>
4046 </h3>
4048 <div>
4050 <p>LLVM supports several instructions to represent vector operations in a
4051 target-independent manner. These instructions cover the element-access and
4052 vector-specific operations needed to process vectors effectively. While LLVM
4053 does directly support these vector operations, many sophisticated algorithms
4054 will want to use target-specific intrinsics to take full advantage of a
4055 specific target.</p>
4057 <!-- _______________________________________________________________________ -->
4058 <h4>
4060 </h4>
4062 <div>
4065 <pre>
4066 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4067 </pre>
4071 from a vector at a specified index.</p>
4077 indicating the position from which to extract the element. The index may be
4078 a variable.</p>
4081 <p>The result is a scalar of the same type as the element type of
4084 results are undefined.</p>
4087 <pre>
4089 </pre>
4091 </div>
4093 <!-- _______________________________________________________________________ -->
4094 <h4>
4096 </h4>
4098 <div>
4101 <pre>
4102 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4103 </pre>
4107 vector at a specified index.</p>
4112 whose type must equal the element type of the first operand. The third
4113 operand is an index indicating the position at which to insert the value.
4114 The index may be a variable.</p>
4120 results are undefined.</p>
4123 <pre>
4124 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4125 </pre>
4127 </div>
4129 <!-- _______________________________________________________________________ -->
4130 <h4>
4132 </h4>
4134 <div>
4137 <pre>
4138 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4139 </pre>
4143 from two input vectors, returning a vector with the same element type as the
4144 input and length that is the same as the shuffle mask.</p>
4148 with types that match each other. The third argument is a shuffle mask whose
4149 element type is always 'i32'. The result of the instruction is a vector
4150 whose length is the same as the shuffle mask and whose element type is the
4151 same as the element type of the first two operands.</p>
4153 <p>The shuffle mask operand is required to be a constant vector with either
4154 constant integer or undef values.</p>
4157 <p>The elements of the two input vectors are numbered from left to right across
4158 both of the vectors. The shuffle mask operand specifies, for each element of
4159 the result vector, which element of the two input vectors the result element
4160 gets. The element selector may be undef (meaning "don't care") and the
4161 second operand may be undef if performing a shuffle from only one vector.</p>
4164 <pre>
4168 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4172 <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>
4173 </pre>
4175 </div>
4177 </div>
4179 <!-- ======================================================================= -->
4180 <h3>
4182 </h3>
4184 <div>
4186 <p>LLVM supports several instructions for working with
4189 <!-- _______________________________________________________________________ -->
4190 <h4>
4192 </h4>
4194 <div>
4197 <pre>
4199 </pre>
4209 specify which value to extract in a similar manner as indices in a
4212 <ul>
4213 <li>Since the value being indexed is not a pointer, the first index is
4214 omitted and assumed to be zero.</li>
4216 <li>Not only struct indices but also array indices must be in
4217 bounds.</li>
4218 </ul>
4221 <p>The result is the value at the position in the aggregate specified by the
4222 index operands.</p>
4225 <pre>
4227 </pre>
4229 </div>
4231 <!-- _______________________________________________________________________ -->
4232 <h4>
4234 </h4>
4236 <div>
4239 <pre>
4240 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4241 </pre>
4251 value to insert. The following operands are constant indices indicating
4252 the position at which to insert the value in a similar manner as indices in a
4254 value to insert must have the same type as the value identified by the
4255 indices.</p>
4263 <pre>
4267 </pre>
4269 </div>
4271 </div>
4273 <!-- ======================================================================= -->
4274 <h3>
4276 </h3>
4278 <div>
4280 <p>A key design point of an SSA-based representation is how it represents
4281 memory. In LLVM, no memory locations are in SSA form, which makes things
4282 very simple. This section describes how to read, write, and allocate
4283 memory in LLVM.</p>
4285 <!-- _______________________________________________________________________ -->
4286 <h4>
4288 </h4>
4290 <div>
4293 <pre>
4294 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4295 </pre>
4299 currently executing function, to be automatically released when this function
4300 returns to its caller. The object is always allocated in the generic address
4301 space (address space zero).</p>
4306 runtime stack, returning a pointer of the appropriate type to the program.
4307 If "NumElements" is specified, it is the number of elements allocated,
4308 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4309 specified, the value result of the allocation is guaranteed to be aligned to
4310 at least that boundary. If not specified, or if zero, the target can choose
4311 to align the allocation on any convenient boundary compatible with the
4312 type.</p>
4317 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4319 memory is automatically released when the function returns. The
4321 variables that must have an address available. When the function returns
4324 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4327 <pre>
4332 </pre>
4334 </div>
4336 <!-- _______________________________________________________________________ -->
4337 <h4>
4339 </h4>
4341 <div>
4344 <pre>
4345 <result> = load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4346 <result> = volatile load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4348 </pre>
4355 from which to load. The pointer must point to
4362 operation (that is, the alignment of the memory address). A value of 0 or an
4364 alignment for the target. It is the responsibility of the code emitter to
4365 ensure that the alignment information is correct. Overestimating the
4366 alignment results in undefined behavior. Underestimating the alignment may
4373 and code generator that this load is not expected to be reused in the cache.
4374 The code generator may select special instructions to save cache bandwidth,
4378 <p>The location of memory pointed to is loaded. If the value being loaded is of
4379 scalar type then the number of bytes read does not exceed the minimum number
4380 of bytes needed to hold all bits of the type. For example, loading an
4383 is undefined if the value was not originally written using a store of the
4384 same type.</p>
4387 <pre>
4391 </pre>
4393 </div>
4395 <!-- _______________________________________________________________________ -->
4396 <h4>
4398 </h4>
4400 <div>
4403 <pre>
4404 store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4405 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4406 </pre>
4413 and an address at which to store it. The type of the
4422 operation (that is, the alignment of the memory address). A value of 0 or an
4423 omitted "align" argument means that the operation has the preferential
4424 alignment for the target. It is the responsibility of the code emitter to
4425 ensure that the alignment information is correct. Overestimating the
4426 alignment results in an undefined behavior. Underestimating the alignment may
4429 <p>The optional !nontemporal metadata must reference a single metatadata
4431 value 1. The existence of the !nontemporal metatadata on the
4432 instruction tells the optimizer and code generator that this load is
4433 not expected to be reused in the cache. The code generator may
4434 select special instructions to save cache bandwidth, such as the
4435 MOVNT instruction on x86.</p>
4442 does not exceed the minimum number of bytes needed to hold all bits of the
4445 integral number of bytes, it is unspecified what happens to the extra bits
4446 that do not belong to the type, but they will typically be overwritten.</p>
4449 <pre>
4453 </pre>
4455 </div>
4457 <!-- _______________________________________________________________________ -->
4458 <h4>
4460 </h4>
4462 <div>
4465 <pre>
4468 </pre>
4473 It performs address calculation only and does not access memory.</p>
4476 <p>The first argument is always a pointer, and forms the basis of the
4477 calculation. The remaining arguments are indices that indicate which of the
4478 elements of the aggregate object are indexed. The interpretation of each
4479 index is dependent on the type being indexed into. The first index always
4480 indexes the pointer value given as the first argument, the second index
4481 indexes a value of the type pointed to (not necessarily the value directly
4482 pointed to, since the first index can be non-zero), etc. The first type
4483 indexed into must be a pointer value, subsequent types can be arrays,
4484 vectors, and structs. Note that subsequent types being indexed into
4485 can never be pointers, since that would require loading the pointer before
4486 continuing calculation.</p>
4488 <p>The type of each index argument depends on the type it is indexing into.
4491 or vector, integers of any width are allowed, and they are not required to be
4492 constant.</p>
4494 <p>For example, let's consider a C code fragment and how it gets compiled to
4495 LLVM:</p>
4498 struct RT {
4499 char A;
4501 char C;
4502 };
4503 struct ST {
4504 int X;
4505 double Y;
4506 struct RT Z;
4507 };
4509 int *foo(struct ST *s) {
4511 }
4512 </pre>
4520 define i32* @foo(%ST* %s) {
4521 entry:
4523 ret i32* %reg
4524 }
4525 </pre>
4530 }</tt>' type, a structure. The second index indexes into the third element
4532 i8 }</tt>' type, another structure. The third index indexes into the second
4534 array. The two dimensions of the array are subscripted into, yielding an
4538 <p>Note that it is perfectly legal to index partially through a structure,
4539 returning a pointer to an inner element. Because of this, the LLVM code for
4540 the given testcase is equivalent to:</p>
4542 <pre>
4543 define i32* @foo(%ST* %s) {
4549 ret i32* %t5
4550 }
4551 </pre>
4556 or if any of the addresses that would be formed by successive addition of
4557 the offsets implied by the indices to the base address with infinitely
4560 the addresses that point into the object, plus the address one byte past
4561 the end.</p>
4564 the base address with silently-wrapping two's complement arithmetic, and
4566 pointed to by the base pointer. The result value may not necessarily be
4567 used to access memory though, even if it happens to point into allocated
4569 section for more information.</p>
4571 <p>The getelementptr instruction is often confusing. For some more insight into
4575 <pre>
4584 </pre>
4586 </div>
4588 </div>
4590 <!-- ======================================================================= -->
4591 <h3>
4593 </h3>
4595 <div>
4597 <p>The instructions in this category are the conversion instructions (casting)
4598 which all take a single operand and a type. They perform various bit
4599 conversions on the operand.</p>
4601 <!-- _______________________________________________________________________ -->
4602 <h4>
4604 </h4>
4606 <div>
4609 <pre>
4611 </pre>
4620 of the same number of integers.
4623 Equal sized types are not allowed.</p>
4632 <pre>
4636 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
4637 </pre>
4639 </div>
4641 <!-- _______________________________________________________________________ -->
4642 <h4>
4644 </h4>
4646 <div>
4649 <pre>
4651 </pre>
4661 of the same number of integers.
4663 the bit size of the destination type,
4673 <pre>
4676 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
4677 </pre>
4679 </div>
4681 <!-- _______________________________________________________________________ -->
4682 <h4>
4684 </h4>
4686 <div>
4689 <pre>
4691 </pre>
4699 of the same number of integers.
4701 the bit size of the destination type,
4712 <pre>
4715 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
4716 </pre>
4718 </div>
4720 <!-- _______________________________________________________________________ -->
4721 <h4>
4723 </h4>
4725 <div>
4728 <pre>
4730 </pre>
4748 undefined.</p>
4751 <pre>
4754 </pre>
4756 </div>
4758 <!-- _______________________________________________________________________ -->
4759 <h4>
4761 </h4>
4763 <div>
4766 <pre>
4768 </pre>
4772 floating point value.</p>
4778 type must be smaller than the destination type.</p>
4788 <pre>
4791 </pre>
4793 </div>
4795 <!-- _______________________________________________________________________ -->
4796 <h4>
4798 </h4>
4800 <div>
4803 <pre>
4805 </pre>
4821 towards zero) unsigned integer value. If the value cannot fit
4825 <pre>
4829 </pre>
4831 </div>
4833 <!-- _______________________________________________________________________ -->
4834 <h4>
4836 </h4>
4838 <div>
4841 <pre>
4843 </pre>
4861 the results are undefined.</p>
4864 <pre>
4868 </pre>
4870 </div>
4872 <!-- _______________________________________________________________________ -->
4873 <h4>
4875 </h4>
4877 <div>
4880 <pre>
4882 </pre>
4897 integer quantity and converts it to the corresponding floating point
4898 value. If the value cannot fit in the floating point value, the results are
4899 undefined.</p>
4902 <pre>
4905 </pre>
4907 </div>
4909 <!-- _______________________________________________________________________ -->
4910 <h4>
4912 </h4>
4914 <div>
4917 <pre>
4919 </pre>
4934 quantity and converts it to the corresponding floating point value. If the
4935 value cannot fit in the floating point value, the results are undefined.</p>
4938 <pre>
4941 </pre>
4943 </div>
4945 <!-- _______________________________________________________________________ -->
4946 <h4>
4948 </h4>
4950 <div>
4953 <pre>
4955 </pre>
4969 truncating or zero extending that value to the size of the integer type. If
4973 change.</p>
4976 <pre>
4979 </pre>
4981 </div>
4983 <!-- _______________________________________________________________________ -->
4984 <h4>
4986 </h4>
4988 <div>
4991 <pre>
4993 </pre>
5001 value to cast, and a type to cast it to, which must be a
5009 than the size of a pointer then a zero extension is done. If they are the
5013 <pre>
5017 </pre>
5019 </div>
5021 <!-- _______________________________________________________________________ -->
5022 <h4>
5024 </h4>
5026 <div>
5029 <pre>
5031 </pre>
5039 non-aggregate first class value, and a type to cast it to, which must also be
5042 identical. If the source type is a pointer, the destination type must also be
5043 a pointer. This instruction supports bitwise conversion of vectors to
5044 integers and to vectors of other types (as long as they have the same
5045 size).</p>
5052 be converted to other pointer types with this instruction. To convert
5057 <pre>
5061 </pre>
5063 </div>
5065 </div>
5067 <!-- ======================================================================= -->
5068 <h3>
5070 </h3>
5072 <div>
5075 defy better classification.</p>
5077 <!-- _______________________________________________________________________ -->
5078 <h4>
5080 </h4>
5082 <div>
5085 <pre>
5086 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5087 </pre>
5091 boolean values based on comparison of its two integer, integer vector, or
5092 pointer operands.</p>
5096 the condition code indicating the kind of comparison to perform. It is not a
5097 value, just a keyword. The possible condition code are:</p>
5099 <ol>
5110 </ol>
5114 typed. They must also be identical types.</p>
5120 result, as follows:</p>
5122 <ol>
5125 performed.</li>
5129 performed.</li>
5156 </ol>
5159 values are compared as if they were integers.</p>
5161 <p>If the operands are integer vectors, then they are compared element by
5166 <pre>
5173 </pre>
5175 <p>Note that the code generator does not yet support vector types with
5178 </div>
5180 <!-- _______________________________________________________________________ -->
5181 <h4>
5183 </h4>
5185 <div>
5188 <pre>
5189 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5190 </pre>
5194 values based on comparison of its operands.</p>
5196 <p>If the operands are floating point scalars, then the result type is a boolean
5199 <p>If the operands are floating point vectors, then the result type is a vector
5200 of boolean with the same number of elements as the operands being
5201 compared.</p>
5205 the condition code indicating the kind of comparison to perform. It is not a
5206 value, just a keyword. The possible condition code are:</p>
5208 <ol>
5225 </ol>
5233 identical types.</p>
5238 vectors, then the vectors are compared element by element. Each comparison
5240 follows:</p>
5242 <ol>
5286 </ol>
5289 <pre>
5294 </pre>
5296 <p>Note that the code generator does not yet support vector types with
5299 </div>
5301 <!-- _______________________________________________________________________ -->
5302 <h4>
5304 </h4>
5306 <div>
5309 <pre>
5311 </pre>
5315 SSA graph representing the function.</p>
5318 <p>The type of the incoming values is specified with the first type field. After
5320 one pair for each predecessor basic block of the current block. Only values
5322 arguments to the PHI node. Only labels may be used as the label
5323 arguments.</p>
5325 <p>There must be no non-phi instructions between the start of a basic block and
5326 the PHI instructions: i.e. PHI instructions must be first in a basic
5327 block.</p>
5329 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5330 occur on the edge from the corresponding predecessor block to the current
5332 value on the same edge).</p>
5336 specified by the pair corresponding to the predecessor basic block that
5337 executed just prior to the current block.</p>
5340 <pre>
5341 Loop: ; Infinite loop that counts from 0 on up...
5342 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5343 %nextindvar = add i32 %indvar, 1
5344 br label %Loop
5345 </pre>
5347 </div>
5349 <!-- _______________________________________________________________________ -->
5350 <h4>
5352 </h4>
5354 <div>
5357 <pre>
5358 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5361 </pre>
5365 condition, without branching.</p>
5370 values indicating the condition, and two values of the
5372 vectors and the condition is a scalar, then entire vectors are selected, not
5373 individual elements.</p>
5377 first value argument; otherwise, it returns the second value argument.</p>
5379 <p>If the condition is a vector of i1, then the value arguments must be vectors
5380 of the same size, and the selection is done element by element.</p>
5383 <pre>
5385 </pre>
5387 <p>Note that the code generator does not yet support conditions
5388 with vector type.</p>
5390 </div>
5392 <!-- _______________________________________________________________________ -->
5393 <h4>
5395 </h4>
5397 <div>
5400 <pre>
5401 <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>]
5402 </pre>
5410 <ol>
5412 access any allocas or varargs in the caller. Note that calls may be
5413 marked "tail" even if they do not occur before
5415 present, the function call is eligible for tail call optimization,
5417 optimized into a jump</a>. The code generator may optimize calls marked
5419 sibling call optimization</a> when the caller and callee have
5420 matching signatures, or 2) forced tail call optimization when the
5421 following extra requirements are met:
5422 <ul>
5423 <li>Caller and callee both have the calling
5425 <li>The call is in tail position (ret immediately follows call and ret
5426 uses value of call or is void).</li>
5430 constraints are met.</a></li>
5431 </ul>
5432 </li>
5435 convention</a> the call should use. If none is specified, the call
5436 defaults to using C calling conventions. The calling convention of the
5437 call must match the calling convention of the target function, or else the
5438 behavior is undefined.</li>
5445 type of the return value. Functions that return no value are marked
5449 being invoked. The argument types must match the types implied by this
5450 signature. This type can be omitted if the function is not varargs and if
5451 the function type does not return a pointer to a function.</li>
5454 be invoked. In most cases, this is a direct function invocation, but
5456 to function value.</li>
5459 signature argument types and parameter attributes. All arguments must be
5461 signature indicates the function accepts a variable number of arguments,
5462 the extra arguments can be specified.</li>
5467 </ol>
5471 a specified function, with its incoming arguments bound to the specified
5473 function, control flow continues with the instruction after the function
5474 call, and the return value of the function is bound to the result
5475 argument.</p>
5478 <pre>
5479 %retval = call i32 @test(i32 %argc)
5483 call void %foo(i8 97 signext)
5485 %struct.A = type { i32, i8 }
5491 </pre>
5493 <p>llvm treats calls to some functions with names and arguments that match the
5494 standard C99 library as being the C99 library functions, and may perform
5495 optimizations or generate code for them under that assumption. This is
5496 something we'd like to change in the future to provide better support for
5497 freestanding environments and non-C-based languages.</p>
5499 </div>
5501 <!-- _______________________________________________________________________ -->
5502 <h4>
5504 </h4>
5506 <div>
5509 <pre>
5511 </pre>
5515 the "variable argument" area of a function call. It is used to implement the
5520 argument. It returns a value of the specified argument type and increments
5527 to the next argument. For more information, see the variable argument
5530 <p>It is legal for this instruction to be called in a function which does not
5532 function.</p>
5536 argument.</p>
5541 <p>Note that the code generator does not yet fully support va_arg on many
5542 targets. Also, it does not currently support va_arg with aggregate types on
5543 any target.</p>
5545 </div>
5547 </div>
5549 </div>
5551 <!-- *********************************************************************** -->
5553 <!-- *********************************************************************** -->
5555 <div>
5558 well known names and semantics and are required to follow certain
5559 restrictions. Overall, these intrinsics represent an extension mechanism for
5560 the LLVM language that does not require changing all of the transformations
5561 in LLVM when adding to the language (or the bitcode reader/writer, the
5562 parser, etc...).</p>
5565 prefix is reserved in LLVM for intrinsic names; thus, function names may not
5566 begin with this prefix. Intrinsic functions must always be external
5567 functions: you cannot define the body of intrinsic functions. Intrinsic
5568 functions may only be used in call or invoke instructions: it is illegal to
5569 take the address of an intrinsic function. Additionally, because intrinsic
5570 functions are part of the LLVM language, it is required if any are added that
5571 they be documented here.</p>
5573 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
5574 family of functions that perform the same operation but on different data
5575 types. Because LLVM can represent over 8 million different integer types,
5576 overloading is used commonly to allow an intrinsic function to operate on any
5577 integer type. One or more of the argument types or the result type can be
5578 overloaded to accept any integer type. Argument types may also be defined as
5579 exactly matching a previous argument's type or the result type. This allows
5580 an intrinsic function which accepts multiple arguments, but needs all of them
5581 to be of the same type, to only be overloaded with respect to a single
5582 argument or the result.</p>
5584 <p>Overloaded intrinsics will have the names of its overloaded argument types
5585 encoded into its function name, each preceded by a period. Only those types
5586 which are overloaded result in a name suffix. Arguments whose type is matched
5588 can take an integer of any width and returns an integer of exactly the same
5589 integer width. This leads to a family of functions such as
5591 %val)</tt>. Only one type, the return type, is overloaded, and only one type
5592 suffix is required. Because the argument's type is matched against the return
5593 type, it does not require its own name suffix.</p>
5595 <p>To learn how to add an intrinsic function, please see the
5598 <!-- ======================================================================= -->
5599 <h3>
5601 </h3>
5603 <div>
5605 <p>Variable argument support is defined in LLVM with
5607 intrinsic functions. These functions are related to the similarly named
5610 <p>All of these functions operate on arguments that use a target-specific value
5611 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
5612 not define what this type is, so all transformations should be prepared to
5613 handle these functions regardless of the type used.</p>
5616 instruction and the variable argument handling intrinsic functions are
5617 used.</p>
5620 define i32 @test(i32 %X, ...) {
5621 ; Initialize variable argument processing
5622 %ap = alloca i8*
5623 %ap2 = bitcast i8** %ap to i8*
5624 call void @llvm.va_start(i8* %ap2)
5626 ; Read a single integer argument
5627 %tmp = va_arg i8** %ap, i32
5629 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5630 %aq = alloca i8*
5631 %aq2 = bitcast i8** %aq to i8*
5632 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5633 call void @llvm.va_end(i8* %aq2)
5635 ; Stop processing of arguments.
5636 call void @llvm.va_end(i8* %ap2)
5637 ret i32 %tmp
5638 }
5640 declare void @llvm.va_start(i8*)
5641 declare void @llvm.va_copy(i8*, i8*)
5642 declare void @llvm.va_end(i8*)
5643 </pre>
5645 <!-- _______________________________________________________________________ -->
5646 <h4>
5648 </h4>
5651 <div>
5654 <pre>
5656 </pre>
5667 macro available in C. In a target-dependent way, it initializes
5671 need to know the last argument of the function as the compiler can figure
5672 that out.</p>
5674 </div>
5676 <!-- _______________________________________________________________________ -->
5677 <h4>
5679 </h4>
5681 <div>
5684 <pre>
5686 </pre>
5690 which has been initialized previously
5699 macro available in C. In a target-dependent way, it destroys
5705 </div>
5707 <!-- _______________________________________________________________________ -->
5708 <h4>
5710 </h4>
5712 <div>
5715 <pre>
5717 </pre>
5721 from the source argument list to the destination argument list.</p>
5726 from.</p>
5730 macro available in C. In a target-dependent way, it copies the
5732 element. This intrinsic is necessary because
5734 arbitrarily complex and require, for example, memory allocation.</p>
5736 </div>
5738 </div>
5740 <!-- ======================================================================= -->
5741 <h3>
5743 </h3>
5745 <div>
5748 Collection</a> (GC) requires the implementation and generation of these
5750 roots on the stack</a>, as well as garbage collector implementations that
5752 barriers. Front-ends for type-safe garbage collected languages should generate
5753 these intrinsics to make use of the LLVM garbage collectors. For more details,
5757 <p>The garbage collection intrinsics only operate on objects in the generic
5758 address space (address space zero).</p>
5760 <!-- _______________________________________________________________________ -->
5761 <h4>
5763 </h4>
5765 <div>
5768 <pre>
5769 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5770 </pre>
5774 the code generator, and allows some metadata to be associated with it.</p>
5777 <p>The first argument specifies the address of a stack object that contains the
5778 root pointer. The second pointer (which must be either a constant or a
5779 global value address) contains the meta-data to be associated with the
5780 root.</p>
5784 location. At compile-time, the code generator generates information to allow
5789 </div>
5791 <!-- _______________________________________________________________________ -->
5792 <h4>
5794 </h4>
5796 <div>
5799 <pre>
5800 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5801 </pre>
5805 locations, allowing garbage collector implementations that require read
5806 barriers.</p>
5809 <p>The second argument is the address to read from, which should be an address
5810 allocated from the garbage collector. The first object is a pointer to the
5811 start of the referenced object, if needed by the language runtime (otherwise
5812 null).</p>
5816 instruction, but may be replaced with substantially more complex code by the
5821 </div>
5823 <!-- _______________________________________________________________________ -->
5824 <h4>
5826 </h4>
5828 <div>
5831 <pre>
5832 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5833 </pre>
5837 locations, allowing garbage collector implementations that require write
5838 barriers (such as generational or reference counting collectors).</p>
5841 <p>The first argument is the reference to store, the second is the start of the
5842 object to store it to, and the third is the address of the field of Obj to
5843 store to. If the runtime does not require a pointer to the object, Obj may
5844 be null.</p>
5848 instruction, but may be replaced with substantially more complex code by the
5853 </div>
5855 </div>
5857 <!-- ======================================================================= -->
5858 <h3>
5860 </h3>
5862 <div>
5864 <p>These intrinsics are provided by LLVM to expose special features that may
5865 only be implemented with code generator support.</p>
5867 <!-- _______________________________________________________________________ -->
5868 <h4>
5870 </h4>
5872 <div>
5875 <pre>
5877 </pre>
5881 target-specific value indicating the return address of the current function
5882 or one of its callers.</p>
5885 <p>The argument to this intrinsic indicates which function to return the address
5886 for. Zero indicates the calling function, one indicates its caller, etc.
5891 indicating the return address of the specified call frame, or zero if it
5892 cannot be identified. The value returned by this intrinsic is likely to be
5893 incorrect or 0 for arguments other than zero, so it should only be used for
5894 debugging purposes.</p>
5896 <p>Note that calling this intrinsic does not prevent function inlining or other
5897 aggressive transformations, so the value returned may not be that of the
5898 obvious source-language caller.</p>
5900 </div>
5902 <!-- _______________________________________________________________________ -->
5903 <h4>
5905 </h4>
5907 <div>
5910 <pre>
5912 </pre>
5916 target-specific frame pointer value for the specified stack frame.</p>
5919 <p>The argument to this intrinsic indicates which function to return the frame
5920 pointer for. Zero indicates the calling function, one indicates its caller,
5925 indicating the frame address of the specified call frame, or zero if it
5926 cannot be identified. The value returned by this intrinsic is likely to be
5927 incorrect or 0 for arguments other than zero, so it should only be used for
5928 debugging purposes.</p>
5930 <p>Note that calling this intrinsic does not prevent function inlining or other
5931 aggressive transformations, so the value returned may not be that of the
5932 obvious source-language caller.</p>
5934 </div>
5936 <!-- _______________________________________________________________________ -->
5937 <h4>
5939 </h4>
5941 <div>
5944 <pre>
5945 declare i8* @llvm.stacksave()
5946 </pre>
5950 of the function stack, for use
5952 useful for implementing language features like scoped automatic variable
5953 sized arrays in C99.</p>
5956 <p>This intrinsic returns a opaque pointer value that can be passed
5964 </div>
5966 <!-- _______________________________________________________________________ -->
5967 <h4>
5969 </h4>
5971 <div>
5974 <pre>
5975 declare void @llvm.stackrestore(i8* %ptr)
5976 </pre>
5980 the function stack to the state it was in when the
5982 executed. This is useful for implementing language features like scoped
5983 automatic variable sized arrays in C99.</p>
5986 <p>See the description
5989 </div>
5991 <!-- _______________________________________________________________________ -->
5992 <h4>
5994 </h4>
5996 <div>
5999 <pre>
6000 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6001 </pre>
6005 insert a prefetch instruction if supported; otherwise, it is a noop.
6006 Prefetches have no effect on the behavior of the program but can change its
6007 performance characteristics.</p>
6016 must be constant integers.</p>
6019 <p>This intrinsic does not modify the behavior of the program. In particular,
6020 prefetches cannot trap and do not produce a value. On targets that support
6021 this intrinsic, the prefetch can provide hints to the processor cache for
6022 better performance.</p>
6024 </div>
6026 <!-- _______________________________________________________________________ -->
6027 <h4>
6029 </h4>
6031 <div>
6034 <pre>
6036 </pre>
6040 Counter (PC) in a region of code to simulators and other tools. The method
6041 is target specific, but it is expected that the marker will use exported
6042 symbols to transmit the PC of the marker. The marker makes no guarantees
6043 that it will remain with any specific instruction after optimizations. It is
6044 possible that the presence of a marker will inhibit optimizations. The
6045 intended use is to be inserted after optimizations to allow correlations of
6046 simulation runs.</p>
6052 <p>This intrinsic does not modify the behavior of the program. Backends that do
6053 not support this intrinsic may ignore it.</p>
6055 </div>
6057 <!-- _______________________________________________________________________ -->
6058 <h4>
6060 </h4>
6062 <div>
6065 <pre>
6066 declare i64 @llvm.readcyclecounter()
6067 </pre>
6071 counter register (or similar low latency, high accuracy clocks) on those
6072 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6073 should map to RPCC. As the backing counters overflow quickly (on the order
6077 <p>When directly supported, reading the cycle counter should not modify any
6078 memory. Implementations are allowed to either return a application specific
6079 value or a system wide value. On backends without support, this is lowered
6082 </div>
6084 </div>
6086 <!-- ======================================================================= -->
6087 <h3>
6089 </h3>
6091 <div>
6093 <p>LLVM provides intrinsics for a few important standard C library functions.
6094 These intrinsics allow source-language front-ends to pass information about
6095 the alignment of the pointer arguments to the code generator, providing
6096 opportunity for more efficient code generation.</p>
6098 <!-- _______________________________________________________________________ -->
6099 <h4>
6101 </h4>
6103 <div>
6107 integer bit width and for different address spaces. Not all targets support
6108 all bit widths however.</p>
6110 <pre>
6115 </pre>
6119 source location to the destination location.</p>
6122 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6123 and the pointers can be in specified address spaces.</p>
6127 <p>The first argument is a pointer to the destination, the second is a pointer
6128 to the source. The third argument is an integer argument specifying the
6129 number of bytes to copy, the fourth argument is the alignment of the
6130 source and destination locations, and the fifth is a boolean indicating a
6131 volatile access.</p>
6134 then the caller guarantees that both the source and destination pointers are
6135 aligned to that boundary.</p>
6139 The detailed access behavior is not very cleanly specified and it is unwise
6140 to depend on it.</p>
6145 source location to the destination location, which are not allowed to
6146 overlap. It copies "len" bytes of memory over. If the argument is known to
6147 be aligned to some boundary, this can be specified as the fourth argument,
6150 </div>
6152 <!-- _______________________________________________________________________ -->
6153 <h4>
6155 </h4>
6157 <div>
6160 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
6161 width and for different address space. Not all targets support all bit
6162 widths however.</p>
6164 <pre>
6169 </pre>
6173 source location to the destination location. It is similar to the
6175 overlap.</p>
6178 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6179 and the pointers can be in specified address spaces.</p>
6183 <p>The first argument is a pointer to the destination, the second is a pointer
6184 to the source. The third argument is an integer argument specifying the
6185 number of bytes to copy, the fourth argument is the alignment of the
6186 source and destination locations, and the fifth is a boolean indicating a
6187 volatile access.</p>
6190 then the caller guarantees that the source and destination pointers are
6191 aligned to that boundary.</p>
6195 The detailed access behavior is not very cleanly specified and it is unwise
6196 to depend on it.</p>
6201 source location to the destination location, which may overlap. It copies
6202 "len" bytes of memory over. If the argument is known to be aligned to some
6203 boundary, this can be specified as the fourth argument, otherwise it should
6206 </div>
6208 <!-- _______________________________________________________________________ -->
6209 <h4>
6211 </h4>
6213 <div>
6216 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
6217 width and for different address spaces. However, not all targets support all
6218 bit widths.</p>
6220 <pre>
6225 </pre>
6229 particular byte value.</p>
6232 intrinsic does not return a value and takes extra alignment/volatile
6233 arguments. Also, the destination can be in an arbitrary address space.</p>
6236 <p>The first argument is a pointer to the destination to fill, the second is the
6237 byte value with which to fill it, the third argument is an integer argument
6238 specifying the number of bytes to fill, and the fourth argument is the known
6239 alignment of the destination location.</p>
6242 then the caller guarantees that the destination pointer is aligned to that
6243 boundary.</p>
6247 The detailed access behavior is not very cleanly specified and it is unwise
6248 to depend on it.</p>
6252 at the destination location. If the argument is known to be aligned to some
6253 boundary, this can be specified as the fourth argument, otherwise it should
6256 </div>
6258 <!-- _______________________________________________________________________ -->
6259 <h4>
6261 </h4>
6263 <div>
6267 floating point or vector of floating point type. Not all targets support all
6268 types however.</p>
6270 <pre>
6271 declare float @llvm.sqrt.f32(float %Val)
6272 declare double @llvm.sqrt.f64(double %Val)
6273 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6274 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6275 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6276 </pre>
6282 behavior for negative numbers other than -0.0 (which allows for better
6283 optimization, because there is no need to worry about errno being
6287 <p>The argument and return value are floating point numbers of the same
6288 type.</p>
6291 <p>This function returns the sqrt of the specified operand if it is a
6292 nonnegative floating point number.</p>
6294 </div>
6296 <!-- _______________________________________________________________________ -->
6297 <h4>
6299 </h4>
6301 <div>
6305 floating point or vector of floating point type. Not all targets support all
6306 types however.</p>
6308 <pre>
6309 declare float @llvm.powi.f32(float %Val, i32 %power)
6310 declare double @llvm.powi.f64(double %Val, i32 %power)
6311 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6312 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6313 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6314 </pre>
6318 specified (positive or negative) power. The order of evaluation of
6319 multiplications is not defined. When a vector of floating point type is
6320 used, the second argument remains a scalar integer value.</p>
6323 <p>The second argument is an integer power, and the first is a value to raise to
6324 that power.</p>
6327 <p>This function returns the first value raised to the second power with an
6328 unspecified sequence of rounding operations.</p>
6330 </div>
6332 <!-- _______________________________________________________________________ -->
6333 <h4>
6335 </h4>
6337 <div>
6341 floating point or vector of floating point type. Not all targets support all
6342 types however.</p>
6344 <pre>
6345 declare float @llvm.sin.f32(float %Val)
6346 declare double @llvm.sin.f64(double %Val)
6347 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6348 declare fp128 @llvm.sin.f128(fp128 %Val)
6349 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6350 </pre>
6356 <p>The argument and return value are floating point numbers of the same
6357 type.</p>
6360 <p>This function returns the sine of the specified operand, returning the same
6362 in the same way.</p>
6364 </div>
6366 <!-- _______________________________________________________________________ -->
6367 <h4>
6369 </h4>
6371 <div>
6375 floating point or vector of floating point type. Not all targets support all
6376 types however.</p>
6378 <pre>
6379 declare float @llvm.cos.f32(float %Val)
6380 declare double @llvm.cos.f64(double %Val)
6381 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6382 declare fp128 @llvm.cos.f128(fp128 %Val)
6383 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6384 </pre>
6390 <p>The argument and return value are floating point numbers of the same
6391 type.</p>
6394 <p>This function returns the cosine of the specified operand, returning the same
6396 in the same way.</p>
6398 </div>
6400 <!-- _______________________________________________________________________ -->
6401 <h4>
6403 </h4>
6405 <div>
6409 floating point or vector of floating point type. Not all targets support all
6410 types however.</p>
6412 <pre>
6413 declare float @llvm.pow.f32(float %Val, float %Power)
6414 declare double @llvm.pow.f64(double %Val, double %Power)
6415 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6416 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6417 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6418 </pre>
6422 specified (positive or negative) power.</p>
6425 <p>The second argument is a floating point power, and the first is a value to
6426 raise to that power.</p>
6429 <p>This function returns the first value raised to the second power, returning
6431 conditions in the same way.</p>
6433 </div>
6435 </div>
6437 <!-- _______________________________________________________________________ -->
6438 <h4>
6440 </h4>
6442 <div>
6446 floating point or vector of floating point type. Not all targets support all
6447 types however.</p>
6449 <pre>
6450 declare float @llvm.exp.f32(float %Val)
6451 declare double @llvm.exp.f64(double %Val)
6452 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
6453 declare fp128 @llvm.exp.f128(fp128 %Val)
6454 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
6455 </pre>
6461 <p>The argument and return value are floating point numbers of the same
6462 type.</p>
6466 would, and handles error conditions in the same way.</p>
6468 </div>
6470 <!-- _______________________________________________________________________ -->
6471 <h4>
6473 </h4>
6475 <div>
6479 floating point or vector of floating point type. Not all targets support all
6480 types however.</p>
6482 <pre>
6483 declare float @llvm.log.f32(float %Val)
6484 declare double @llvm.log.f64(double %Val)
6485 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
6486 declare fp128 @llvm.log.f128(fp128 %Val)
6487 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
6488 </pre>
6494 <p>The argument and return value are floating point numbers of the same
6495 type.</p>
6499 would, and handles error conditions in the same way.</p>
6501 <h4>
6503 </h4>
6505 <div>
6509 floating point or vector of floating point type. Not all targets support all
6510 types however.</p>
6512 <pre>
6513 declare float @llvm.fma.f32(float %a, float %b, float %c)
6514 declare double @llvm.fma.f64(double %a, double %b, double %c)
6515 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
6516 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
6517 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
6518 </pre>
6522 operation.</p>
6525 <p>The argument and return value are floating point numbers of the same
6526 type.</p>
6530 would.</p>
6532 </div>
6534 <!-- ======================================================================= -->
6535 <h3>
6537 </h3>
6539 <div>
6541 <p>LLVM provides intrinsics for a few important bit manipulation operations.
6542 These allow efficient code generation for some algorithms.</p>
6544 <!-- _______________________________________________________________________ -->
6545 <h4>
6547 </h4>
6549 <div>
6552 <p>This is an overloaded intrinsic function. You can use bswap on any integer
6555 <pre>
6559 </pre>
6563 values with an even number of bytes (positive multiple of 16 bits). These
6564 are useful for performing operations on data that is not in the target's
6565 native byte order.</p>
6569 and low byte of the input i16 swapped. Similarly,
6575 more, respectively).</p>
6577 </div>
6579 <!-- _______________________________________________________________________ -->
6580 <h4>
6582 </h4>
6584 <div>
6587 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
6588 width, or on any vector with integer elements. Not all targets support all
6589 bit widths or vector types, however.</p>
6591 <pre>
6598 </pre>
6602 in a value.</p>
6605 <p>The only argument is the value to be counted. The argument may be of any
6606 integer type, or a vector with integer elements.
6607 The return type must match the argument type.</p>
6611 element of a vector.</p>
6613 </div>
6615 <!-- _______________________________________________________________________ -->
6616 <h4>
6618 </h4>
6620 <div>
6624 integer bit width, or any vector whose elements are integers. Not all
6625 targets support all bit widths or vector types, however.</p>
6627 <pre>
6634 </pre>
6638 leading zeros in a variable.</p>
6641 <p>The only argument is the value to be counted. The argument may be of any
6642 integer type, or any vector type with integer element type.
6643 The return type must match the argument type.</p>
6647 zeros in a variable, or within each element of the vector if the operation
6648 is of vector type. If the src == 0 then the result is the size in bits of
6651 </div>
6653 <!-- _______________________________________________________________________ -->
6654 <h4>
6656 </h4>
6658 <div>
6662 integer bit width, or any vector of integer elements. Not all targets
6663 support all bit widths or vector types, however.</p>
6665 <pre>
6672 </pre>
6676 trailing zeros.</p>
6679 <p>The only argument is the value to be counted. The argument may be of any
6680 integer type, or a vectory with integer element type.. The return type
6681 must match the argument type.</p>
6685 zeros in a variable, or within each element of a vector.
6686 If the src == 0 then the result is the size in bits of
6689 </div>
6691 </div>
6693 <!-- ======================================================================= -->
6694 <h3>
6696 </h3>
6698 <div>
6702 <!-- _______________________________________________________________________ -->
6703 <h4>
6706 </a>
6707 </h4>
6709 <div>
6713 on any integer bit width.</p>
6715 <pre>
6716 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6717 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6718 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6719 </pre>
6723 a signed addition of the two arguments, and indicate whether an overflow
6724 occurred during the signed summation.</p>
6727 <p>The arguments (%a and %b) and the first element of the result structure may
6728 be of integer types of any bit width, but they must have the same bit
6729 width. The second element of the result structure must be of
6731 undergo signed addition.</p>
6735 a signed addition of the two variables. They return a structure — the
6736 first element of which is the signed summation, and the second element of
6737 which is a bit specifying if the signed summation resulted in an
6738 overflow.</p>
6741 <pre>
6742 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6743 %sum = extractvalue {i32, i1} %res, 0
6744 %obit = extractvalue {i32, i1} %res, 1
6745 br i1 %obit, label %overflow, label %normal
6746 </pre>
6748 </div>
6750 <!-- _______________________________________________________________________ -->
6751 <h4>
6754 </a>
6755 </h4>
6757 <div>
6761 on any integer bit width.</p>
6763 <pre>
6764 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6765 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6766 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6767 </pre>
6771 an unsigned addition of the two arguments, and indicate whether a carry
6772 occurred during the unsigned summation.</p>
6775 <p>The arguments (%a and %b) and the first element of the result structure may
6776 be of integer types of any bit width, but they must have the same bit
6777 width. The second element of the result structure must be of
6779 undergo unsigned addition.</p>
6783 an unsigned addition of the two arguments. They return a structure —
6784 the first element of which is the sum, and the second element of which is a
6785 bit specifying if the unsigned summation resulted in a carry.</p>
6788 <pre>
6789 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6790 %sum = extractvalue {i32, i1} %res, 0
6791 %obit = extractvalue {i32, i1} %res, 1
6792 br i1 %obit, label %carry, label %normal
6793 </pre>
6795 </div>
6797 <!-- _______________________________________________________________________ -->
6798 <h4>
6801 </a>
6802 </h4>
6804 <div>
6808 on any integer bit width.</p>
6810 <pre>
6811 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6812 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6813 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6814 </pre>
6818 a signed subtraction of the two arguments, and indicate whether an overflow
6819 occurred during the signed subtraction.</p>
6822 <p>The arguments (%a and %b) and the first element of the result structure may
6823 be of integer types of any bit width, but they must have the same bit
6824 width. The second element of the result structure must be of
6826 undergo signed subtraction.</p>
6830 a signed subtraction of the two arguments. They return a structure —
6831 the first element of which is the subtraction, and the second element of
6832 which is a bit specifying if the signed subtraction resulted in an
6833 overflow.</p>
6836 <pre>
6837 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6838 %sum = extractvalue {i32, i1} %res, 0
6839 %obit = extractvalue {i32, i1} %res, 1
6840 br i1 %obit, label %overflow, label %normal
6841 </pre>
6843 </div>
6845 <!-- _______________________________________________________________________ -->
6846 <h4>
6849 </a>
6850 </h4>
6852 <div>
6856 on any integer bit width.</p>
6858 <pre>
6859 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6860 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6861 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6862 </pre>
6866 an unsigned subtraction of the two arguments, and indicate whether an
6867 overflow occurred during the unsigned subtraction.</p>
6870 <p>The arguments (%a and %b) and the first element of the result structure may
6871 be of integer types of any bit width, but they must have the same bit
6872 width. The second element of the result structure must be of
6874 undergo unsigned subtraction.</p>
6878 an unsigned subtraction of the two arguments. They return a structure —
6879 the first element of which is the subtraction, and the second element of
6880 which is a bit specifying if the unsigned subtraction resulted in an
6881 overflow.</p>
6884 <pre>
6885 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6886 %sum = extractvalue {i32, i1} %res, 0
6887 %obit = extractvalue {i32, i1} %res, 1
6888 br i1 %obit, label %overflow, label %normal
6889 </pre>
6891 </div>
6893 <!-- _______________________________________________________________________ -->
6894 <h4>
6897 </a>
6898 </h4>
6900 <div>
6904 on any integer bit width.</p>
6906 <pre>
6907 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6908 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6909 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6910 </pre>
6915 a signed multiplication of the two arguments, and indicate whether an
6916 overflow occurred during the signed multiplication.</p>
6919 <p>The arguments (%a and %b) and the first element of the result structure may
6920 be of integer types of any bit width, but they must have the same bit
6921 width. The second element of the result structure must be of
6923 undergo signed multiplication.</p>
6927 a signed multiplication of the two arguments. They return a structure —
6928 the first element of which is the multiplication, and the second element of
6929 which is a bit specifying if the signed multiplication resulted in an
6930 overflow.</p>
6933 <pre>
6934 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6935 %sum = extractvalue {i32, i1} %res, 0
6936 %obit = extractvalue {i32, i1} %res, 1
6937 br i1 %obit, label %overflow, label %normal
6938 </pre>
6940 </div>
6942 <!-- _______________________________________________________________________ -->
6943 <h4>
6946 </a>
6947 </h4>
6949 <div>
6953 on any integer bit width.</p>
6955 <pre>
6956 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6957 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6958 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6959 </pre>
6963 a unsigned multiplication of the two arguments, and indicate whether an
6964 overflow occurred during the unsigned multiplication.</p>
6967 <p>The arguments (%a and %b) and the first element of the result structure may
6968 be of integer types of any bit width, but they must have the same bit
6969 width. The second element of the result structure must be of
6971 undergo unsigned multiplication.</p>
6975 an unsigned multiplication of the two arguments. They return a structure
6976 — the first element of which is the multiplication, and the second
6977 element of which is a bit specifying if the unsigned multiplication resulted
6978 in an overflow.</p>
6981 <pre>
6982 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6983 %sum = extractvalue {i32, i1} %res, 0
6984 %obit = extractvalue {i32, i1} %res, 1
6985 br i1 %obit, label %overflow, label %normal
6986 </pre>
6988 </div>
6990 </div>
6992 <!-- ======================================================================= -->
6993 <h3>
6995 </h3>
6997 <div>
6999 <p>Half precision floating point is a storage-only format. This means that it is
7000 a dense encoding (in memory) but does not support computation in the
7001 format.</p>
7003 <p>This means that code must first load the half-precision floating point
7004 value as an i16, then convert it to float with <a
7006 Computation can then be performed on the float value (including extending to
7007 double etc). To store the value back to memory, it is first converted to
7008 float if needed, then converted to i16 with
7010 storing as an i16 value.</p>
7012 <!-- _______________________________________________________________________ -->
7013 <h4>
7016 </a>
7017 </h4>
7019 <div>
7022 <pre>
7023 declare i16 @llvm.convert.to.fp16(f32 %a)
7024 </pre>
7028 a conversion from single precision floating point format to half precision
7029 floating point format.</p>
7032 <p>The intrinsic function contains single argument - the value to be
7033 converted.</p>
7037 a conversion from single precision floating point format to half precision
7039 contains the converted number.</p>
7042 <pre>
7043 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7044 store i16 %res, i16* @x, align 2
7045 </pre>
7047 </div>
7049 <!-- _______________________________________________________________________ -->
7050 <h4>
7053 </a>
7054 </h4>
7056 <div>
7059 <pre>
7060 declare f32 @llvm.convert.from.fp16(i16 %a)
7061 </pre>
7065 a conversion from half precision floating point format to single precision
7066 floating point format.</p>
7069 <p>The intrinsic function contains single argument - the value to be
7070 converted.</p>
7074 conversion from half single precision floating point format to single
7075 precision floating point format. The input half-float value is represented by
7079 <pre>
7080 %a = load i16* @x, align 2
7081 %res = call f32 @llvm.convert.from.fp16(i16 %a)
7082 </pre>
7084 </div>
7086 </div>
7088 <!-- ======================================================================= -->
7089 <h3>
7091 </h3>
7093 <div>
7096 prefix), are described in
7100 </div>
7102 <!-- ======================================================================= -->
7103 <h3>
7105 </h3>
7107 <div>
7109 <p>The LLVM exception handling intrinsics (which all start with
7114 </div>
7116 <!-- ======================================================================= -->
7117 <h3>
7119 </h3>
7121 <div>
7123 <p>This intrinsic makes it possible to excise one parameter, marked with
7125 The result is a callable
7126 function pointer lacking the nest parameter - the caller does not need to
7127 provide a value for it. Instead, the value to use is stored in advance in a
7128 "trampoline", a block of memory usually allocated on the stack, which also
7129 contains code to splice the nest value into the argument list. This is used
7130 to implement the GCC nested function address extension.</p>
7132 <p>For example, if the function is
7135 follows:</p>
7140 %p = call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval)
7141 %fp = bitcast i8* %p to i32 (i32, i32)*
7142 </pre>
7147 <!-- _______________________________________________________________________ -->
7148 <h4>
7151 </a>
7152 </h4>
7154 <div>
7157 <pre>
7159 </pre>
7163 function pointer suitable for executing it.</p>
7168 sufficiently aligned block of memory; this memory is written to by the
7169 intrinsic. Note that the size and the alignment are target-specific - LLVM
7170 currently provides no portable way of determining them, so a front-end that
7171 generates this intrinsic needs to have some target-specific knowledge.
7177 dependent code, turning it into a function. A pointer to this function is
7179 function pointer type</a> before being called. The new function's signature
7182 is allowed, and it must be of pointer type. Calling the new function is
7187 returned function pointer is undefined.</p>
7189 </div>
7191 </div>
7193 <!-- ======================================================================= -->
7194 <h3>
7196 </h3>
7198 <div>
7201 hardware constructs for atomic operations and memory synchronization. This
7202 provides an interface to the hardware, not an interface to the programmer. It
7203 is aimed at a low enough level to allow any programming models or APIs
7204 (Application Programming Interfaces) which need atomic behaviors to map
7205 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
7206 hardware provides a "universal IR" for source languages, it also provides a
7207 starting point for developing a "universal" atomic operation and
7208 synchronization IR.</p>
7211 software transaction memory systems, atomic primitives, and intrinsic
7212 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
7213 application libraries. The hardware interface provided by LLVM should allow
7214 a clean implementation of all of these APIs and parallel programming models.
7215 No one model or paradigm should be selected above others unless the hardware
7216 itself ubiquitously does so.</p>
7218 <!-- _______________________________________________________________________ -->
7219 <h4>
7221 </h4>
7223 <div>
7225 <pre>
7226 declare void @llvm.memory.barrier(i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device>)
7227 </pre>
7231 specific pairs of memory access types.</p>
7235 The first four arguments enables a specific barrier as listed below. The
7236 fifth argument specifies that the barrier applies to io or device or uncached
7237 memory.</p>
7239 <ul>
7245 </ul>
7248 <p>This intrinsic causes the system to enforce some ordering constraints upon
7249 the loads and stores of the program. This barrier does not
7252 and store operations (f.ex. load-load, or store-load), all of the first
7253 operations preceding the barrier will complete before any of the second
7254 operations succeeding the barrier begin. Specifically the semantics for each
7255 pairing is as follows:</p>
7257 <ul>
7259 after the barrier begins.</li>
7261 store after the barrier begins.</li>
7263 store after the barrier begins.</li>
7265 load after the barrier begins.</li>
7266 </ul>
7269 is enabled in a single memory barrier intrinsic.</p>
7271 <p>Backends may implement stronger barriers than those requested when they do
7272 not support as fine grained a barrier as requested. Some architectures do
7273 not need all types of barriers and on such architectures, these become
7274 noops.</p>
7277 <pre>
7278 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7279 %ptr = bitcast i8* %mallocP to i32*
7280 store i32 4, %ptr
7283 call void @llvm.memory.barrier(i1 false, i1 true, i1 false, i1 false, i1 true)
7286 </pre>
7288 </div>
7290 <!-- _______________________________________________________________________ -->
7291 <h4>
7293 </h4>
7295 <div>
7299 any integer bit width and for different address spaces. Not all targets
7300 support all bit widths however.</p>
7302 <pre>
7304 declare i16 @llvm.atomic.cmp.swap.i16.p0i16(i16* <ptr>, i16 <cmp>, i16 <val>)
7305 declare i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* <ptr>, i32 <cmp>, i32 <val>)
7306 declare i64 @llvm.atomic.cmp.swap.i64.p0i64(i64* <ptr>, i64 <cmp>, i64 <val>)
7307 </pre>
7310 <p>This loads a value in memory and compares it to a given value. If they are
7311 equal, it stores a new value into the memory.</p>
7317 this integer type. While any bit width integer may be used, targets may only
7318 lower representations they support in hardware.</p>
7321 <p>This entire intrinsic must be executed atomically. It first loads the value
7324 memory. The loaded value is yielded in all cases. This provides the
7325 equivalent of an atomic compare-and-swap operation within the SSA
7326 framework.</p>
7329 <pre>
7330 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7331 %ptr = bitcast i8* %mallocP to i32*
7332 store i32 4, %ptr
7335 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 4, %val1)
7341 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 5, %val2)
7346 </pre>
7348 </div>
7350 <!-- _______________________________________________________________________ -->
7351 <h4>
7353 </h4>
7355 <div>
7359 integer bit width. Not all targets support all bit widths however.</p>
7361 <pre>
7366 </pre>
7377 integer type. The targets may only lower integer representations they
7378 support.</p>
7383 equivalent of an atomic swap operation within the SSA framework.</p>
7386 <pre>
7387 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7388 %ptr = bitcast i8* %mallocP to i32*
7389 store i32 4, %ptr
7392 %result1 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val1)
7398 %result2 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val2)
7403 </pre>
7405 </div>
7407 <!-- _______________________________________________________________________ -->
7408 <h4>
7410 </h4>
7412 <div>
7416 any integer bit width. Not all targets support all bit widths however.</p>
7418 <pre>
7423 </pre>
7430 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7431 and the second an integer value. The result is also an integer value. These
7432 integer types can have any bit width, but they must all have the same bit
7433 width. The targets may only lower integer representations they support.</p>
7436 <p>This intrinsic does a series of operations atomically. It first loads the
7441 <pre>
7442 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7443 %ptr = bitcast i8* %mallocP to i32*
7444 store i32 4, %ptr
7445 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 4)
7447 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 2)
7449 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 5)
7452 </pre>
7454 </div>
7456 <!-- _______________________________________________________________________ -->
7457 <h4>
7459 </h4>
7461 <div>
7465 any integer bit width and for different address spaces. Not all targets
7466 support all bit widths however.</p>
7468 <pre>
7473 </pre>
7480 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7481 and the second an integer value. The result is also an integer value. These
7482 integer types can have any bit width, but they must all have the same bit
7483 width. The targets may only lower integer representations they support.</p>
7486 <p>This intrinsic does a series of operations atomically. It first loads the
7492 <pre>
7493 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7494 %ptr = bitcast i8* %mallocP to i32*
7495 store i32 8, %ptr
7496 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 4)
7498 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 2)
7500 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 5)
7503 </pre>
7505 </div>
7507 <!-- _______________________________________________________________________ -->
7508 <h4>
7511 </a>
7512 <br>
7515 </a>
7516 <br>
7519 </a>
7520 <br>
7523 </a>
7524 </h4>
7526 <div>
7529 <p>These are overloaded intrinsics. You can
7532 bit width and for different address spaces. Not all targets support all bit
7533 widths however.</p>
7535 <pre>
7540 </pre>
7542 <pre>
7547 </pre>
7549 <pre>
7554 </pre>
7556 <pre>
7561 </pre>
7569 <p>These intrinsics take two arguments, the first a pointer to an integer value
7570 and the second an integer value. The result is also an integer value. These
7571 integer types can have any bit width, but they must all have the same bit
7572 width. The targets may only lower integer representations they support.</p>
7575 <p>These intrinsics does a series of operations atomically. They first load the
7581 <pre>
7582 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7583 %ptr = bitcast i8* %mallocP to i32*
7584 store i32 0x0F0F, %ptr
7585 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32(i32* %ptr, i32 0xFF)
7587 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32(i32* %ptr, i32 0xFF)
7589 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32(i32* %ptr, i32 0F)
7591 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32(i32* %ptr, i32 0F)
7594 </pre>
7596 </div>
7598 <!-- _______________________________________________________________________ -->
7599 <h4>
7602 </a>
7603 <br>
7606 </a>
7607 <br>
7610 </a>
7611 <br>
7614 </a>
7615 </h4>
7617 <div>
7623 address spaces. Not all targets support all bit widths however.</p>
7625 <pre>
7630 </pre>
7632 <pre>
7637 </pre>
7639 <pre>
7644 </pre>
7646 <pre>
7651 </pre>
7654 <p>These intrinsics takes the signed or unsigned minimum or maximum of
7659 <p>These intrinsics take two arguments, the first a pointer to an integer value
7660 and the second an integer value. The result is also an integer value. These
7661 integer types can have any bit width, but they must all have the same bit
7662 width. The targets may only lower integer representations they support.</p>
7665 <p>These intrinsics does a series of operations atomically. They first load the
7671 <pre>
7672 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7673 %ptr = bitcast i8* %mallocP to i32*
7674 store i32 7, %ptr
7675 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32(i32* %ptr, i32 -2)
7677 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32(i32* %ptr, i32 8)
7679 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32(i32* %ptr, i32 10)
7681 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32(i32* %ptr, i32 30)
7684 </pre>
7686 </div>
7688 </div>
7690 <!-- ======================================================================= -->
7691 <h3>
7693 </h3>
7695 <div>
7697 <p>This class of intrinsics exists to information about the lifetime of memory
7698 objects and ranges where variables are immutable.</p>
7700 <!-- _______________________________________________________________________ -->
7701 <h4>
7703 </h4>
7705 <div>
7708 <pre>
7710 </pre>
7714 object's lifetime.</p>
7717 <p>The first argument is a constant integer representing the size of the
7718 object, or -1 if it is variable sized. The second argument is a pointer to
7719 the object.</p>
7722 <p>This intrinsic indicates that before this point in the code, the value of the
7724 never be used and has an undefined value. A load from the pointer that
7725 precedes this intrinsic can be replaced with
7728 </div>
7730 <!-- _______________________________________________________________________ -->
7731 <h4>
7733 </h4>
7735 <div>
7738 <pre>
7740 </pre>
7744 object's lifetime.</p>
7747 <p>The first argument is a constant integer representing the size of the
7748 object, or -1 if it is variable sized. The second argument is a pointer to
7749 the object.</p>
7752 <p>This intrinsic indicates that after this point in the code, the value of the
7754 never be used and has an undefined value. Any stores into the memory object
7755 following this intrinsic may be removed as dead.
7757 </div>
7759 <!-- _______________________________________________________________________ -->
7760 <h4>
7762 </h4>
7764 <div>
7767 <pre>
7769 </pre>
7773 a memory object will not change.</p>
7776 <p>The first argument is a constant integer representing the size of the
7777 object, or -1 if it is variable sized. The second argument is a pointer to
7778 the object.</p>
7782 the return value, the referenced memory location is constant and
7783 unchanging.</p>
7785 </div>
7787 <!-- _______________________________________________________________________ -->
7788 <h4>
7790 </h4>
7792 <div>
7795 <pre>
7796 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
7797 </pre>
7801 a memory object are mutable.</p>
7805 The second argument is a constant integer representing the size of the
7806 object, or -1 if it is variable sized and the third argument is a pointer
7807 to the object.</p>
7812 </div>
7814 </div>
7816 <!-- ======================================================================= -->
7817 <h3>
7819 </h3>
7821 <div>
7823 <p>This class of intrinsics is designed to be generic and has no specific
7824 purpose.</p>
7826 <!-- _______________________________________________________________________ -->
7827 <h4>
7829 </h4>
7831 <div>
7834 <pre>
7835 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
7836 </pre>
7842 <p>The first argument is a pointer to a value, the second is a pointer to a
7843 global string, the third is a pointer to a global string which is the source
7844 file name, and the last argument is the line number.</p>
7847 <p>This intrinsic allows annotation of local variables with arbitrary strings.
7848 This can be useful for special purpose optimizations that want to look for
7849 these annotations. These have no other defined use, they are ignored by code
7850 generation and optimization.</p>
7852 </div>
7854 <!-- _______________________________________________________________________ -->
7855 <h4>
7857 </h4>
7859 <div>
7863 any integer bit width.</p>
7865 <pre>
7866 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
7867 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
7868 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
7869 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
7870 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
7871 </pre>
7877 <p>The first argument is an integer value (result of some expression), the
7878 second is a pointer to a global string, the third is a pointer to a global
7879 string which is the source file name, and the last argument is the line
7880 number. It returns the value of the first argument.</p>
7883 <p>This intrinsic allows annotations to be put on arbitrary expressions with
7884 arbitrary strings. This can be useful for special purpose optimizations that
7885 want to look for these annotations. These have no other defined use, they
7886 are ignored by code generation and optimization.</p>
7888 </div>
7890 <!-- _______________________________________________________________________ -->
7891 <h4>
7893 </h4>
7895 <div>
7898 <pre>
7899 declare void @llvm.trap()
7900 </pre>
7909 <p>This intrinsics is lowered to the target dependent trap instruction. If the
7910 target does not have a trap instruction, this intrinsic will be lowered to
7913 </div>
7915 <!-- _______________________________________________________________________ -->
7916 <h4>
7918 </h4>
7920 <div>
7923 <pre>
7925 </pre>
7930 ensure that it is placed on the stack before local variables.</p>
7934 arguments. The first argument is the value loaded from the stack
7936 that has enough space to hold the value of the guard.</p>
7939 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
7941 stack. This is to ensure that if a local variable on the stack is
7942 overwritten, it will destroy the value of the guard. When the function exits,
7943 the guard on the stack is checked against the original guard. If they are
7945 function.</p>
7947 </div>
7949 <!-- _______________________________________________________________________ -->
7950 <h4>
7952 </h4>
7954 <div>
7957 <pre>
7960 </pre>
7964 the optimizers to determine at compile time whether a) an operation (like
7965 memcpy) will overflow a buffer that corresponds to an object, or b) that a
7966 runtime check for overflow isn't necessary. An object in this context means
7967 an allocation of a specific class, structure, array, or other object.</p>
7980 compile time.</p>
7982 </div>
7984 </div>
7986 </div>
7988 <!-- *********************************************************************** -->
7989 <hr>
7990 <address>
7998 Last modified: $Date$
7999 </address>
8001 </body>
8002 </html>