| blob | bf63884357159ff46fb084ca16f1be2a116c682f |
3 <html>
4 <head>
11 </head>
13 <body>
16 <ol>
21 <ol>
28 </ol>
29 </li>
31 <ol>
33 <ol>
35 </ol>
36 </li>
38 <ol>
45 </ol>
46 </li>
47 </ol>
48 </li>
50 <ol>
56 </ol>
57 </li>
59 <ol>
61 </ol>
62 </li>
64 <ol>
66 <ol>
73 </ol>
74 </li>
76 <ol>
83 </ol>
84 </li>
86 <ol>
92 </ol>
93 </li>
95 <ol>
102 </ol>
103 </li>
105 <ol>
112 </ol>
113 </li>
115 <ol>
121 </ol>
122 </li>
123 </ol>
124 </li>
126 <ol>
128 <ol>
132 </ol>
133 </li>
135 <ol>
139 </ol>
140 </li>
142 <ol>
150 </ol>
151 </li>
153 <ol>
160 </ol>
161 </li>
163 <ol>
168 </ol>
169 </li>
171 </ol>
172 </li>
173 </ol>
178 </div>
180 <!-- *********************************************************************** -->
182 <!-- *********************************************************************** -->
185 <p>This document is a reference manual for the LLVM assembly language.
186 LLVM is an SSA based representation that provides type safety,
187 low-level operations, flexibility, and the capability of representing
188 'all' high-level languages cleanly. It is the common code
189 representation used throughout all phases of the LLVM compilation
190 strategy.</p>
191 </div>
193 <!-- *********************************************************************** -->
195 <!-- *********************************************************************** -->
199 <p>The LLVM code representation is designed to be used in three
200 different forms: as an in-memory compiler IR, as an on-disk bytecode
201 representation (suitable for fast loading by a Just-In-Time compiler),
202 and as a human readable assembly language representation. This allows
203 LLVM to provide a powerful intermediate representation for efficient
204 compiler transformations and analysis, while providing a natural means
205 to debug and visualize the transformations. The three different forms
206 of LLVM are all equivalent. This document describes the human readable
207 representation and notation.</p>
209 <p>The LLVM representation aims to be light-weight and low-level
210 while being expressive, typed, and extensible at the same time. It
211 aims to be a "universal IR" of sorts, by being at a low enough level
212 that high-level ideas may be cleanly mapped to it (similar to how
213 microprocessors are "universal IR's", allowing many source languages to
214 be mapped to them). By providing type information, LLVM can be used as
215 the target of optimizations: for example, through pointer analysis, it
216 can be proven that a C automatic variable is never accessed outside of
217 the current function... allowing it to be promoted to a simple SSA
218 value instead of a memory location.</p>
220 </div>
222 <!-- _______________________________________________________________________ -->
227 <p>It is important to note that this document describes 'well formed'
228 LLVM assembly language. There is a difference between what the parser
229 accepts and what is considered 'well formed'. For example, the
230 following instruction is syntactically okay, but not well formed:</p>
232 <pre>
234 </pre>
237 its uses. The LLVM infrastructure provides a verification pass that may
238 be used to verify that an LLVM module is well formed. This pass is
239 automatically run by the parser after parsing input assembly and by
240 the optimizer before it outputs bytecode. The violations pointed out
241 by the verifier pass indicate bugs in transformation passes or input to
242 the parser.</p>
246 <!-- *********************************************************************** -->
248 <!-- *********************************************************************** -->
252 <p>LLVM uses three different forms of identifiers, for different
253 purposes:</p>
255 <ol>
256 <li>Named values are represented as a string of characters with a '%' prefix.
257 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
259 Identifiers which require other characters in their names can be surrounded
261 in a name.</li>
263 <li>Unnamed values are represented as an unsigned numeric value with a '%'
264 prefix. For example, %12, %2, %44.</li>
267 constants</a>, below.</li>
268 </ol>
270 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
271 don't need to worry about name clashes with reserved words, and the set of
272 reserved words may be expanded in the future without penalty. Additionally,
273 unnamed identifiers allow a compiler to quickly come up with a temporary
274 variable without having to avoid symbol table conflicts.</p>
276 <p>Reserved words in LLVM are very similar to reserved words in other
277 languages. There are keywords for different opcodes ('<tt><a
281 and others. These reserved words cannot conflict with variable names, because
282 none of them start with a '%' character.</p>
284 <p>Here is an example of LLVM code to multiply the integer variable
285 '<tt>%X</tt>' by 8:</p>
287 <p>The easy way:</p>
289 <pre>
291 </pre>
293 <p>After strength reduction:</p>
295 <pre>
297 </pre>
299 <p>And the hard way:</p>
301 <pre>
305 </pre>
307 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
308 important lexical features of LLVM:</p>
310 <ol>
312 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
313 line.</li>
315 <li>Unnamed temporaries are created when the result of a computation is not
316 assigned to a named value.</li>
318 <li>Unnamed temporaries are numbered sequentially</li>
320 </ol>
322 <p>...and it also shows a convention that we follow in this document. When
323 demonstrating instructions, we will follow an instruction with a comment that
324 defines the type and name of value produced. Comments are shown in italic
325 text.</p>
327 </div>
329 <!-- *********************************************************************** -->
331 <!-- *********************************************************************** -->
333 <!-- ======================================================================= -->
335 </div>
340 translation unit of the input programs. Each module consists of
341 functions, global variables, and symbol table entries. Modules may be
342 combined together with the LLVM linker, which merges function (and
343 global variable) definitions, resolves forward declarations, and merges
346 <pre><i>; Declare the string constant as a global constant...</i>
348 href="#globalvars">constant</a> <a href="#t_array">[13 x sbyte]</a> c"hello world\0A\00" <i>; [13 x sbyte]*</i>
350 <i>; External declaration of the puts function</i>
353 <i>; Definition of main function</i>
354 int %main() { <i>; int()* </i>
355 <i>; Convert [13x sbyte]* to sbyte *...</i>
356 %cast210 = <a
359 <i>; Call puts function to write out the string to stdout...</i>
360 <a
362 <a
370 <p>In general, a module is made up of a list of global values,
371 where both functions and global variables are global values. Global values are
372 represented by a pointer to a memory location (in this case, a pointer to an
373 array of char, and a pointer to a function), and have one of the following <a
376 </div>
378 <!-- ======================================================================= -->
381 </div>
385 <p>
386 All Global Variables and Functions have one of the following types of linkage:
387 </p>
389 <dl>
393 <dd>Global values with internal linkage are only directly accessible by
394 objects in the current module. In particular, linking code into a module with
395 an internal global value may cause the internal to be renamed as necessary to
396 avoid collisions. Because the symbol is internal to the module, all
397 references can be updated. This corresponds to the notion of the
399 </dd>
404 the twist that linking together two modules defining the same
405 <tt>linkonce</tt> globals will cause one of the globals to be discarded. This
406 is typically used to implement inline functions. Unreferenced
407 <tt>linkonce</tt> globals are allowed to be discarded.
408 </dd>
413 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
415 </dd>
420 pointer to array type. When two global variables with appending linkage are
421 linked together, the two global arrays are appended together. This is the
422 LLVM, typesafe, equivalent of having the system linker append together
424 </dd>
428 <dd>If none of the above identifiers are used, the global is externally
429 visible, meaning that it participates in linkage and can be used to resolve
430 external symbol references.
431 </dd>
432 </dl>
436 variable and was linked with this one, one of the two would be renamed,
438 external (i.e., lacking any linkage declarations), they are accessible
439 outside of the current module. It is illegal for a function <i>declaration</i>
442 </div>
444 <!-- ======================================================================= -->
447 </div>
453 specified for the call. The calling convention of any pair of dynamic
454 caller/callee must match, or the behavior of the program is undefined. The
455 following calling conventions are supported by LLVM, and more may be added in
456 the future:</p>
458 <dl>
461 <dd>This calling convention (the default if no other calling convention is
462 specified) matches the target C calling conventions. This calling convention
463 supports varargs function calls and tolerates some mismatch in the declared
464 prototype and implemented declaration of the function (as does normal C).
465 </dd>
469 <dd>This calling convention attempts to make calls as fast as possible
470 (e.g. by passing things in registers). This calling convention allows the
471 target to use whatever tricks it wants to produce fast code for the target,
472 without having to conform to an externally specified ABI. Implementations of
473 this convention should allow arbitrary tail call optimization to be supported.
474 This calling convention does not support varargs and requires the prototype of
475 all callees to exactly match the prototype of the function definition.
476 </dd>
480 <dd>This calling convention attempts to make code in the caller as efficient
481 as possible under the assumption that the call is not commonly executed. As
482 such, these calls often preserve all registers so that the call does not break
483 any live ranges in the caller side. This calling convention does not support
484 varargs and requires the prototype of all callees to exactly match the
485 prototype of the function definition.
486 </dd>
490 <dd>Any calling convention may be specified by number, allowing
491 target-specific calling conventions to be used. Target specific calling
492 conventions start at 64.
493 </dd>
494 </dl>
496 <p>More calling conventions can be added/defined on an as-needed basis, to
497 support pascal conventions or any other well-known target-independent
498 convention.</p>
500 </div>
502 <!-- ======================================================================= -->
505 </div>
509 <p>Global variables define regions of memory allocated at compilation time
510 instead of run-time. Global variables may optionally be initialized, may have
511 an explicit section to be placed in, and may
512 have an optional explicit alignment specified. A
514 contents of the variable will <b>never</b> be modified (enabling better
515 optimization, allowing the global data to be placed in the read-only section of
516 an executable, etc). Note that variables that need runtime initialization
519 <p>
520 LLVM explicitly allows <em>declarations</em> of global variables to be marked
521 constant, even if the final definition of the global is not. This capability
522 can be used to enable slightly better optimization of the program, but requires
523 the language definition to guarantee that optimizations based on the
524 'constantness' are valid for the translation units that do not include the
525 definition.
526 </p>
528 <p>As SSA values, global variables define pointer values that are in
529 scope (i.e. they dominate) all basic blocks in the program. Global
531 describe a region of memory, and all memory objects in LLVM are
532 accessed through pointers.</p>
534 <p>LLVM allows an explicit section to be specified for globals. If the target
535 supports it, it will emit globals to the section specified.</p>
537 <p>An explicit alignment may be specified for a global. If not present, or if
538 the alignment is set to zero, the alignment of the global is set by the target
539 to whatever it feels convenient. If an explicit alignment is specified, the
540 global is forced to have at least that much alignment. All alignments must be
541 a power of 2.</p>
543 </div>
546 <!-- ======================================================================= -->
549 </div>
555 type, a function name, a (possibly empty) argument list, an optional section,
556 an optional alignment, an opening curly brace,
557 a list of basic blocks, and a closing curly brace. LLVM function declarations
560 a possibly empty list of arguments, and an optional alignment.</p>
562 <p>A function definition contains a list of basic blocks, forming the CFG for
563 the function. Each basic block may optionally start with a label (giving the
564 basic block a symbol table entry), contains a list of instructions, and ends
566 function return).</p>
568 <p>The first basic block in a program is special in two ways: it is immediately
569 executed on entrance to the function, and it is not allowed to have predecessor
570 basic blocks (i.e. there can not be any branches to the entry block of a
571 function). Because the block can have no predecessors, it also cannot have any
574 <p>LLVM functions are identified by their name and type signature. Hence, two
575 functions with the same name but different parameter lists or return values are
576 considered different functions, and LLVM will resolve references to each
577 appropriately.</p>
579 <p>LLVM allows an explicit section to be specified for functions. If the target
580 supports it, it will emit functions to the section specified.</p>
582 <p>An explicit alignment may be specified for a function. If not present, or if
583 the alignment is set to zero, the alignment of the function is set by the target
584 to whatever it feels convenient. If an explicit alignment is specified, the
585 function is forced to have at least that much alignment. All alignments must be
586 a power of 2.</p>
588 </div>
590 <!-- ======================================================================= -->
593 </div>
596 <p>
599 LLVM and treated as a single unit, but may be separated in the .ll file if
600 desired. The syntax is very simple:
601 </p>
606 </pre></div>
608 <p>The strings can contain any character by escaping non-printable characters.
610 for the number.
611 </p>
613 <p>
614 The inline asm code is simply printed to the machine code .s file when
615 assembly code is generated.
616 </p>
617 </div>
620 <!-- *********************************************************************** -->
622 <!-- *********************************************************************** -->
626 <p>The LLVM type system is one of the most important features of the
627 intermediate representation. Being typed enables a number of
628 optimizations to be performed on the IR directly, without having to do
629 extra analyses on the side before the transformation. A strong type
630 system makes it easier to read the generated code and enables novel
631 analyses and transformations that are not feasible to perform on normal
632 three address code representations.</p>
634 </div>
636 <!-- ======================================================================= -->
639 <p>The primitive types are the fundamental building blocks of the LLVM
640 system. The current set of primitive types is as follows:</p>
645 <table>
646 <tbody>
647 <tr><th>Type</th><th>Description</th></tr>
648 <tr><td><tt>void</tt></td><td>No value</td></tr>
649 <tr><td><tt>ubyte</tt></td><td>Unsigned 8-bit value</td></tr>
650 <tr><td><tt>ushort</tt></td><td>Unsigned 16-bit value</td></tr>
651 <tr><td><tt>uint</tt></td><td>Unsigned 32-bit value</td></tr>
652 <tr><td><tt>ulong</tt></td><td>Unsigned 64-bit value</td></tr>
653 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
654 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
655 </tbody>
656 </table>
657 </td>
659 <table>
660 <tbody>
661 <tr><th>Type</th><th>Description</th></tr>
662 <tr><td><tt>bool</tt></td><td>True or False value</td></tr>
663 <tr><td><tt>sbyte</tt></td><td>Signed 8-bit value</td></tr>
664 <tr><td><tt>short</tt></td><td>Signed 16-bit value</td></tr>
665 <tr><td><tt>int</tt></td><td>Signed 32-bit value</td></tr>
666 <tr><td><tt>long</tt></td><td>Signed 64-bit value</td></tr>
667 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
668 </tbody>
669 </table>
670 </td>
671 </tr>
672 </table>
673 </div>
675 <!-- _______________________________________________________________________ -->
677 Classifications</a> </div>
679 <p>These different primitive types fall into a few useful
680 classifications:</p>
683 <tbody>
684 <tr><th>Classification</th><th>Types</th></tr>
685 <tr>
687 <td><tt>sbyte, short, int, long, float, double</tt></td>
688 </tr>
689 <tr>
691 <td><tt>ubyte, ushort, uint, ulong</tt></td>
692 </tr>
693 <tr>
695 <td><tt>ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
696 </tr>
697 <tr>
699 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long</tt>
700 </td>
701 </tr>
702 <tr>
704 <td><tt>float, double</tt></td>
705 </tr>
706 <tr>
708 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long,<br>
711 </tr>
712 </tbody>
713 </table>
716 most important. Values of these types are the only ones which can be
717 produced by instructions, passed as arguments, or used as operands to
718 instructions. This means that all structures and arrays must be
719 manipulated either by pointer or by component.</p>
720 </div>
722 <!-- ======================================================================= -->
727 <p>The real power in LLVM comes from the derived types in the system.
728 This is what allows a programmer to represent arrays, functions,
729 pointers, and other useful types. Note that these derived types may be
730 recursive: For example, it is possible to have a two dimensional array.</p>
732 </div>
734 <!-- _______________________________________________________________________ -->
739 <h5>Overview:</h5>
741 <p>The array type is a very simple derived type that arranges elements
742 sequentially in memory. The array type requires a size (number of
743 elements) and an underlying data type.</p>
745 <h5>Syntax:</h5>
747 <pre>
748 [<# elements> x <elementtype>]
749 </pre>
751 <p>The number of elements is a constant integer value; elementtype may
752 be any type with a size.</p>
754 <h5>Examples:</h5>
758 <tt>[40 x int ]</tt><br/>
759 <tt>[41 x int ]</tt><br/>
760 <tt>[40 x uint]</tt><br/>
761 </td>
763 Array of 40 integer values.<br/>
764 Array of 41 integer values.<br/>
765 Array of 40 unsigned integer values.<br/>
766 </td>
767 </tr>
768 </table>
769 <p>Here are some examples of multidimensional arrays:</p>
773 <tt>[3 x [4 x int]]</tt><br/>
774 <tt>[12 x [10 x float]]</tt><br/>
775 <tt>[2 x [3 x [4 x uint]]]</tt><br/>
776 </td>
778 3x4 array of integer values.<br/>
779 12x10 array of single precision floating point values.<br/>
780 2x3x4 array of unsigned integer values.<br/>
781 </td>
782 </tr>
783 </table>
785 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
786 length array. Normally, accesses past the end of an array are undefined in
787 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
788 As a special case, however, zero length arrays are recognized to be variable
789 length. This allows implementation of 'pascal style arrays' with the LLVM
792 </div>
794 <!-- _______________________________________________________________________ -->
797 <h5>Overview:</h5>
798 <p>The function type can be thought of as a function signature. It
799 consists of a return type and a list of formal parameter types.
800 Function types are usually used to build virtual function tables
801 (which are structures of pointers to functions), for indirect function
802 calls, and when defining a function.</p>
803 <p>
804 The return type of a function type cannot be an aggregate type.
805 </p>
806 <h5>Syntax:</h5>
807 <pre> <returntype> (<parameter list>)<br></pre>
808 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
809 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
810 which indicates that the function takes a variable number of arguments.
811 Variable argument functions can access their arguments with the <a
813 <h5>Examples:</h5>
817 <tt>int (int)</tt> <br/>
818 <tt>float (int, int *) *</tt><br/>
819 <tt>int (sbyte *, ...)</tt><br/>
820 </td>
822 function taking an <tt>int</tt>, returning an <tt>int</tt><br/>
825 returning <tt>float</tt>.<br/>
827 to <tt>sbyte</tt> (signed char in C), which returns an integer. This is
828 the signature for <tt>printf</tt> in LLVM.<br/>
829 </td>
830 </tr>
831 </table>
833 </div>
834 <!-- _______________________________________________________________________ -->
837 <h5>Overview:</h5>
838 <p>The structure type is used to represent a collection of data members
839 together in memory. The packing of the field types is defined to match
840 the ABI of the underlying processor. The elements of a structure may
841 be any type that has a size.</p>
845 instruction.</p>
846 <h5>Syntax:</h5>
847 <pre> { <type list> }<br></pre>
848 <h5>Examples:</h5>
852 <tt>{ int, int, int }</tt><br/>
853 <tt>{ float, int (int) * }</tt><br/>
854 </td>
856 a triple of three <tt>int</tt> values<br/>
857 A pair, where the first element is a <tt>float</tt> and the second element
859 that takes an <tt>int</tt>, returning an <tt>int</tt>.<br/>
860 </td>
861 </tr>
862 </table>
863 </div>
865 <!-- _______________________________________________________________________ -->
868 <h5>Overview:</h5>
869 <p>As in many languages, the pointer type represents a pointer or
870 reference to another object, which must live in memory.</p>
871 <h5>Syntax:</h5>
872 <pre> <type> *<br></pre>
873 <h5>Examples:</h5>
877 <tt>[4x int]*</tt><br/>
878 <tt>int (int *) *</tt><br/>
879 </td>
882 four <tt>int</tt> values<br/>
885 <tt>int</tt>.<br/>
886 </td>
887 </tr>
888 </table>
889 </div>
891 <!-- _______________________________________________________________________ -->
895 <h5>Overview:</h5>
897 <p>A packed type is a simple derived type that represents a vector
898 of elements. Packed types are used when multiple primitive data
899 are operated in parallel using a single instruction (SIMD).
900 A packed type requires a size (number of
901 elements) and an underlying primitive data type. Vectors must have a power
902 of two length (1, 2, 4, 8, 16 ...). Packed types are
905 <h5>Syntax:</h5>
907 <pre>
908 < <# elements> x <elementtype> >
909 </pre>
911 <p>The number of elements is a constant integer value; elementtype may
912 be any integral or floating point type.</p>
914 <h5>Examples:</h5>
919 <tt><4 x int></tt><br/>
920 <tt><8 x float></tt><br/>
921 <tt><2 x uint></tt><br/>
922 </td>
924 Packed vector of 4 integer values.<br/>
925 Packed vector of 8 floating-point values.<br/>
926 Packed vector of 2 unsigned integer values.<br/>
927 </td>
928 </tr>
929 </table>
930 </div>
932 <!-- _______________________________________________________________________ -->
936 <h5>Overview:</h5>
938 <p>Opaque types are used to represent unknown types in the system. This
939 corresponds (for example) to the C notion of a foward declared structure type.
940 In LLVM, opaque types can eventually be resolved to any type (not just a
941 structure type).</p>
943 <h5>Syntax:</h5>
945 <pre>
946 opaque
947 </pre>
949 <h5>Examples:</h5>
954 <tt>opaque</tt>
955 </td>
957 An opaque type.<br/>
958 </td>
959 </tr>
960 </table>
961 </div>
964 <!-- *********************************************************************** -->
966 <!-- *********************************************************************** -->
970 <p>LLVM has several different basic types of constants. This section describes
971 them all and their syntax.</p>
973 </div>
975 <!-- ======================================================================= -->
980 <dl>
981 <dt><b>Boolean constants</b></dt>
983 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
985 </dd>
987 <dt><b>Integer constants</b></dt>
989 <dd>Standard integers (such as '4') are constants of the <a
991 integer types.
992 </dd>
994 <dt><b>Floating point constants</b></dt>
996 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
997 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
998 notation (see below). Floating point constants must have a <a
1001 <dt><b>Null pointer constants</b></dt>
1003 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1006 </dl>
1008 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1009 of floating point constants. For example, the form '<tt>double
1010 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1011 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1012 (and the only time that they are generated by the disassembler) is when a
1013 floating point constant must be emitted but it cannot be represented as a
1014 decimal floating point number. For example, NaN's, infinities, and other
1015 special values are represented in their IEEE hexadecimal format so that
1016 assembly and disassembly do not cause any bits to change in the constants.</p>
1018 </div>
1020 <!-- ======================================================================= -->
1022 </div>
1025 <p>Aggregate constants arise from aggregation of simple constants
1026 and smaller aggregate constants.</p>
1028 <dl>
1029 <dt><b>Structure constants</b></dt>
1031 <dd>Structure constants are represented with notation similar to structure
1032 type definitions (a comma separated list of elements, surrounded by braces
1036 types of elements must match those specified by the type.
1037 </dd>
1039 <dt><b>Array constants</b></dt>
1041 <dd>Array constants are represented with notation similar to array type
1042 definitions (a comma separated list of elements, surrounded by square brackets
1045 types of elements must match those specified by the type.
1046 </dd>
1048 <dt><b>Packed constants</b></dt>
1050 <dd>Packed constants are represented with notation similar to packed type
1051 definitions (a comma separated list of elements, surrounded by
1055 match those specified by the type.
1056 </dd>
1058 <dt><b>Zero initialization</b></dt>
1060 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1061 value to zero of <em>any</em> type, including scalar and aggregate types.
1062 This is often used to avoid having to print large zero initializers (e.g. for
1063 large arrays) and is always exactly equivalent to using explicit zero
1064 initializers.
1065 </dd>
1066 </dl>
1068 </div>
1070 <!-- ======================================================================= -->
1073 </div>
1079 constants. These constants are explicitly referenced when the <a
1082 file:</p>
1084 <pre>
1085 %X = global int 17
1086 %Y = global int 42
1087 %Z = global [2 x int*] [ int* %X, int* %Y ]
1088 </pre>
1090 </div>
1092 <!-- ======================================================================= -->
1095 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1096 no specific value. Undefined values may be of any type and be used anywhere
1097 a constant is permitted.</p>
1099 <p>Undefined values indicate to the compiler that the program is well defined
1100 no matter what value is used, giving the compiler more freedom to optimize.
1101 </p>
1102 </div>
1104 <!-- ======================================================================= -->
1106 </div>
1110 <p>Constant expressions are used to allow expressions involving other constants
1111 to be used as constants. Constant expressions may be of any <a
1113 that does not have side effects (e.g. load and call are not supported). The
1114 following is the syntax for constant expressions:</p>
1116 <dl>
1117 <dt><b><tt>cast ( CST to TYPE )</tt></b></dt>
1119 <dd>Cast a constant to another type.</dd>
1121 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1125 instruction, the index list may have zero or more indexes, which are required
1128 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1131 constants.
1133 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1136 operation</a> on constants.
1138 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1141 operation</a> on constants.
1144 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1147 operation</a> on constants.
1149 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1151 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1153 binary</a> operations. The constraints on operands are the same as those for
1154 the corresponding instruction (e.g. no bitwise operations on floating point
1155 values are allowed).</dd>
1156 </dl>
1157 </div>
1159 <!-- *********************************************************************** -->
1161 <!-- *********************************************************************** -->
1163 <!-- ======================================================================= -->
1166 </div>
1170 <p>
1172 Module-Level Inline Assembly</a>) through the use of a special value. This
1173 value represents the inline assembler as a string (containing the instructions
1174 to emit), a list of operand constraints (stored as a string), and a flag that
1175 indicates whether or not the inline asm expression has side effects. An example
1176 inline assembler expression is:
1177 </p>
1179 <pre>
1181 </pre>
1183 <p>
1184 Inline assembler expressions may <b>only</b> be used as the callee operand of
1186 </p>
1188 <pre>
1190 </pre>
1192 <p>
1193 Inline asms with side effects not visible in the constraint list must be marked
1194 as having side effects. This is done through the use of the
1195 '<tt>sideeffect</tt>' keyword, like so:
1196 </p>
1198 <pre>
1200 </pre>
1202 <p>TODO: The format of the asm and constraints string still need to be
1203 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1204 need to be documented).
1205 </p>
1207 </div>
1209 <!-- *********************************************************************** -->
1211 <!-- *********************************************************************** -->
1215 <p>The LLVM instruction set consists of several different
1220 instructions</a>.</p>
1222 </div>
1224 <!-- ======================================================================= -->
1226 Instructions</a> </div>
1232 indicates which block should be executed after the current block is
1233 finished. These terminator instructions typically yield a '<tt>void</tt>'
1234 value: they produce control flow, not values (the one exception being
1236 <p>There are six different terminator instructions: the '<a
1243 </div>
1245 <!-- _______________________________________________________________________ -->
1247 Instruction</a> </div>
1249 <h5>Syntax:</h5>
1250 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1251 ret void <i>; Return from void function</i>
1252 </pre>
1253 <h5>Overview:</h5>
1254 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1255 value) from a function back to the caller.</p>
1256 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1257 returns a value and then causes control flow, and one that just causes
1258 control flow to occur.</p>
1259 <h5>Arguments:</h5>
1260 <p>The '<tt>ret</tt>' instruction may return any '<a
1263 instruction inside of the function that returns a value that does not
1264 match the return type of the function.</p>
1265 <h5>Semantics:</h5>
1266 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1272 returns a value, that value shall set the call or invoke instruction's
1273 return value.</p>
1274 <h5>Example:</h5>
1275 <pre> ret int 5 <i>; Return an integer value of 5</i>
1276 ret void <i>; Return from a void function</i>
1277 </pre>
1278 </div>
1279 <!-- _______________________________________________________________________ -->
1282 <h5>Syntax:</h5>
1283 <pre> br bool <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1284 </pre>
1285 <h5>Overview:</h5>
1286 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1287 transfer to a different basic block in the current function. There are
1288 two forms of this instruction, corresponding to a conditional branch
1289 and an unconditional branch.</p>
1290 <h5>Arguments:</h5>
1291 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1292 single '<tt>bool</tt>' value and two '<tt>label</tt>' values. The
1293 unconditional form of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>'
1294 value as a target.</p>
1295 <h5>Semantics:</h5>
1296 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>bool</tt>'
1297 argument is evaluated. If the value is <tt>true</tt>, control flows
1299 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1300 <h5>Example:</h5>
1301 <pre>Test:<br> %cond = <a href="#i_setcc">seteq</a> int %a, %b<br> br bool %cond, label %IfEqual, label %IfUnequal<br>IfEqual:<br> <a
1303 </div>
1304 <!-- _______________________________________________________________________ -->
1307 </div>
1310 <h5>Syntax:</h5>
1312 <pre>
1313 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1314 </pre>
1316 <h5>Overview:</h5>
1318 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1319 several different places. It is a generalization of the '<tt>br</tt>'
1320 instruction, allowing a branch to occur to one of many possible
1321 destinations.</p>
1324 <h5>Arguments:</h5>
1326 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1327 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1328 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1329 table is not allowed to contain duplicate constant entries.</p>
1331 <h5>Semantics:</h5>
1333 <p>The <tt>switch</tt> instruction specifies a table of values and
1334 destinations. When the '<tt>switch</tt>' instruction is executed, this
1335 table is searched for the given value. If the value is found, control flow is
1336 transfered to the corresponding destination; otherwise, control flow is
1337 transfered to the default destination.</p>
1339 <h5>Implementation:</h5>
1341 <p>Depending on properties of the target machine and the particular
1342 <tt>switch</tt> instruction, this instruction may be code generated in different
1343 ways. For example, it could be generated as a series of chained conditional
1344 branches or with a lookup table.</p>
1346 <h5>Example:</h5>
1348 <pre>
1349 <i>; Emulate a conditional br instruction</i>
1351 switch int %Val, label %truedest [int 0, label %falsedest ]
1353 <i>; Emulate an unconditional br instruction</i>
1354 switch uint 0, label %dest [ ]
1356 <i>; Implement a jump table:</i>
1357 switch uint %val, label %otherwise [ uint 0, label %onzero
1358 uint 1, label %onone
1359 uint 2, label %ontwo ]
1360 </pre>
1361 </div>
1363 <!-- _______________________________________________________________________ -->
1366 </div>
1370 <h5>Syntax:</h5>
1372 <pre>
1373 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1374 to label <normal label> except label <exception label>
1375 </pre>
1377 <h5>Overview:</h5>
1379 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1380 function, with the possibility of control flow transfer to either the
1381 '<tt>normal</tt>' label or the
1382 '<tt>exception</tt>' label. If the callee function returns with the
1388 <h5>Arguments:</h5>
1390 <p>This instruction requires several arguments:</p>
1392 <ol>
1393 <li>
1395 convention</a> the call should use. If none is specified, the call defaults
1396 to using C calling conventions.
1397 </li>
1398 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1399 function value being invoked. In most cases, this is a direct function
1400 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1401 an arbitrary pointer to function value.
1402 </li>
1404 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1405 function to be invoked. </li>
1407 <li>'<tt>function args</tt>': argument list whose types match the function
1408 signature argument types. If the function signature indicates the function
1409 accepts a variable number of arguments, the extra arguments can be
1410 specified. </li>
1412 <li>'<tt>normal label</tt>': the label reached when the called function
1415 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1418 </ol>
1420 <h5>Semantics:</h5>
1422 <p>This instruction is designed to operate as a standard '<tt><a
1424 difference is that it establishes an association with a label, which is used by
1425 the runtime library to unwind the stack.</p>
1427 <p>This instruction is used in languages with destructors to ensure that proper
1428 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1429 exception. Additionally, this is important for implementation of
1430 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1432 <h5>Example:</h5>
1433 <pre>
1434 %retval = invoke int %Test(int 15) to label %Continue
1435 except label %TestCleanup <i>; {int}:retval set</i>
1437 except label %TestCleanup <i>; {int}:retval set</i>
1438 </pre>
1439 </div>
1442 <!-- _______________________________________________________________________ -->
1445 Instruction</a> </div>
1449 <h5>Syntax:</h5>
1450 <pre>
1451 unwind
1452 </pre>
1454 <h5>Overview:</h5>
1456 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1457 at the first callee in the dynamic call stack which used an <a
1459 primarily used to implement exception handling.</p>
1461 <h5>Semantics:</h5>
1463 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1464 immediately halt. The dynamic call stack is then searched for the first <a
1467 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1468 dynamic call chain, undefined behavior results.</p>
1469 </div>
1471 <!-- _______________________________________________________________________ -->
1474 Instruction</a> </div>
1478 <h5>Syntax:</h5>
1479 <pre>
1480 unreachable
1481 </pre>
1483 <h5>Overview:</h5>
1485 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1486 instruction is used to inform the optimizer that a particular portion of the
1487 code is not reachable. This can be used to indicate that the code after a
1488 no-return function cannot be reached, and other facts.</p>
1490 <h5>Semantics:</h5>
1492 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1493 </div>
1497 <!-- ======================================================================= -->
1500 <p>Binary operators are used to do most of the computation in a
1501 program. They require two operands, execute an operation on them, and
1502 produce a single value. The operands might represent
1504 The result value of a binary operator is not
1505 necessarily the same type as its operands.</p>
1506 <p>There are several different binary operators:</p>
1507 </div>
1508 <!-- _______________________________________________________________________ -->
1510 Instruction</a> </div>
1512 <h5>Syntax:</h5>
1513 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1514 </pre>
1515 <h5>Overview:</h5>
1516 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1517 <h5>Arguments:</h5>
1518 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1521 Both arguments must have identical types.</p>
1522 <h5>Semantics:</h5>
1523 <p>The value produced is the integer or floating point sum of the two
1524 operands.</p>
1525 <h5>Example:</h5>
1526 <pre> <result> = add int 4, %var <i>; yields {int}:result = 4 + %var</i>
1527 </pre>
1528 </div>
1529 <!-- _______________________________________________________________________ -->
1531 Instruction</a> </div>
1533 <h5>Syntax:</h5>
1534 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1535 </pre>
1536 <h5>Overview:</h5>
1537 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1538 operands.</p>
1539 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1540 instruction present in most other intermediate representations.</p>
1541 <h5>Arguments:</h5>
1542 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1544 values.
1546 Both arguments must have identical types.</p>
1547 <h5>Semantics:</h5>
1548 <p>The value produced is the integer or floating point difference of
1549 the two operands.</p>
1550 <h5>Example:</h5>
1551 <pre> <result> = sub int 4, %var <i>; yields {int}:result = 4 - %var</i>
1552 <result> = sub int 0, %val <i>; yields {int}:result = -%var</i>
1553 </pre>
1554 </div>
1555 <!-- _______________________________________________________________________ -->
1557 Instruction</a> </div>
1559 <h5>Syntax:</h5>
1560 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1561 </pre>
1562 <h5>Overview:</h5>
1563 <p>The '<tt>mul</tt>' instruction returns the product of its two
1564 operands.</p>
1565 <h5>Arguments:</h5>
1566 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1568 values.
1570 Both arguments must have identical types.</p>
1571 <h5>Semantics:</h5>
1572 <p>The value produced is the integer or floating point product of the
1573 two operands.</p>
1574 <p>There is no signed vs unsigned multiplication. The appropriate
1575 action is taken based on the type of the operand.</p>
1576 <h5>Example:</h5>
1577 <pre> <result> = mul int 4, %var <i>; yields {int}:result = 4 * %var</i>
1578 </pre>
1579 </div>
1580 <!-- _______________________________________________________________________ -->
1582 Instruction</a> </div>
1584 <h5>Syntax:</h5>
1585 <pre> <result> = div <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1586 </pre>
1587 <h5>Overview:</h5>
1588 <p>The '<tt>div</tt>' instruction returns the quotient of its two
1589 operands.</p>
1590 <h5>Arguments:</h5>
1591 <p>The two arguments to the '<tt>div</tt>' instruction must be either <a
1593 values.
1595 Both arguments must have identical types.</p>
1596 <h5>Semantics:</h5>
1597 <p>The value produced is the integer or floating point quotient of the
1598 two operands.</p>
1599 <h5>Example:</h5>
1600 <pre> <result> = div int 4, %var <i>; yields {int}:result = 4 / %var</i>
1601 </pre>
1602 </div>
1603 <!-- _______________________________________________________________________ -->
1605 Instruction</a> </div>
1607 <h5>Syntax:</h5>
1608 <pre> <result> = rem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1609 </pre>
1610 <h5>Overview:</h5>
1611 <p>The '<tt>rem</tt>' instruction returns the remainder from the
1612 division of its two operands.</p>
1613 <h5>Arguments:</h5>
1614 <p>The two arguments to the '<tt>rem</tt>' instruction must be either <a
1616 values.
1618 Both arguments must have identical types.</p>
1619 <h5>Semantics:</h5>
1620 <p>This returns the <i>remainder</i> of a division (where the result
1621 has the same sign as the divisor), not the <i>modulus</i> (where the
1622 result has the same sign as the dividend) of a value. For more
1623 information about the difference, see <a
1625 Math Forum</a>.</p>
1626 <h5>Example:</h5>
1627 <pre> <result> = rem int 4, %var <i>; yields {int}:result = 4 % %var</i>
1628 </pre>
1630 </div>
1631 <!-- _______________________________________________________________________ -->
1633 Instructions</a> </div>
1635 <h5>Syntax:</h5>
1636 <pre> <result> = seteq <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1637 <result> = setne <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1638 <result> = setlt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1639 <result> = setgt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1640 <result> = setle <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1641 <result> = setge <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1642 </pre>
1643 <h5>Overview:</h5>
1644 <p>The '<tt>set<i>cc</i></tt>' family of instructions returns a boolean
1645 value based on a comparison of their two operands.</p>
1646 <h5>Arguments:</h5>
1647 <p>The two arguments to the '<tt>set<i>cc</i></tt>' instructions must
1649 to compare '<tt>label</tt>'s, '<tt>array</tt>'s, '<tt>structure</tt>'
1650 or '<tt>void</tt>' values, etc...). Both arguments must have identical
1651 types.</p>
1652 <h5>Semantics:</h5>
1653 <p>The '<tt>seteq</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1654 value if both operands are equal.<br>
1655 The '<tt>setne</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1656 value if both operands are unequal.<br>
1657 The '<tt>setlt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1658 value if the first operand is less than the second operand.<br>
1659 The '<tt>setgt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1660 value if the first operand is greater than the second operand.<br>
1661 The '<tt>setle</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1662 value if the first operand is less than or equal to the second operand.<br>
1663 The '<tt>setge</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1664 value if the first operand is greater than or equal to the second
1665 operand.</p>
1666 <h5>Example:</h5>
1667 <pre> <result> = seteq int 4, 5 <i>; yields {bool}:result = false</i>
1668 <result> = setne float 4, 5 <i>; yields {bool}:result = true</i>
1669 <result> = setlt uint 4, 5 <i>; yields {bool}:result = true</i>
1670 <result> = setgt sbyte 4, 5 <i>; yields {bool}:result = false</i>
1671 <result> = setle sbyte 4, 5 <i>; yields {bool}:result = true</i>
1672 <result> = setge sbyte 4, 5 <i>; yields {bool}:result = false</i>
1673 </pre>
1674 </div>
1676 <!-- ======================================================================= -->
1678 Operations</a> </div>
1680 <p>Bitwise binary operators are used to do various forms of
1681 bit-twiddling in a program. They are generally very efficient
1682 instructions and can commonly be strength reduced from other
1683 instructions. They require two operands, execute an operation on them,
1684 and produce a single value. The resulting value of the bitwise binary
1685 operators is always the same type as its first operand.</p>
1686 </div>
1687 <!-- _______________________________________________________________________ -->
1689 Instruction</a> </div>
1691 <h5>Syntax:</h5>
1692 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1693 </pre>
1694 <h5>Overview:</h5>
1695 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
1696 its two operands.</p>
1697 <h5>Arguments:</h5>
1698 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
1700 identical types.</p>
1701 <h5>Semantics:</h5>
1702 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
1703 <p> </p>
1706 <tbody>
1707 <tr>
1708 <td>In0</td>
1709 <td>In1</td>
1710 <td>Out</td>
1711 </tr>
1712 <tr>
1713 <td>0</td>
1714 <td>0</td>
1715 <td>0</td>
1716 </tr>
1717 <tr>
1718 <td>0</td>
1719 <td>1</td>
1720 <td>0</td>
1721 </tr>
1722 <tr>
1723 <td>1</td>
1724 <td>0</td>
1725 <td>0</td>
1726 </tr>
1727 <tr>
1728 <td>1</td>
1729 <td>1</td>
1730 <td>1</td>
1731 </tr>
1732 </tbody>
1733 </table>
1734 </div>
1735 <h5>Example:</h5>
1736 <pre> <result> = and int 4, %var <i>; yields {int}:result = 4 & %var</i>
1737 <result> = and int 15, 40 <i>; yields {int}:result = 8</i>
1738 <result> = and int 4, 8 <i>; yields {int}:result = 0</i>
1739 </pre>
1740 </div>
1741 <!-- _______________________________________________________________________ -->
1744 <h5>Syntax:</h5>
1745 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1746 </pre>
1747 <h5>Overview:</h5>
1748 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
1749 or of its two operands.</p>
1750 <h5>Arguments:</h5>
1751 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
1753 identical types.</p>
1754 <h5>Semantics:</h5>
1755 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
1756 <p> </p>
1759 <tbody>
1760 <tr>
1761 <td>In0</td>
1762 <td>In1</td>
1763 <td>Out</td>
1764 </tr>
1765 <tr>
1766 <td>0</td>
1767 <td>0</td>
1768 <td>0</td>
1769 </tr>
1770 <tr>
1771 <td>0</td>
1772 <td>1</td>
1773 <td>1</td>
1774 </tr>
1775 <tr>
1776 <td>1</td>
1777 <td>0</td>
1778 <td>1</td>
1779 </tr>
1780 <tr>
1781 <td>1</td>
1782 <td>1</td>
1783 <td>1</td>
1784 </tr>
1785 </tbody>
1786 </table>
1787 </div>
1788 <h5>Example:</h5>
1789 <pre> <result> = or int 4, %var <i>; yields {int}:result = 4 | %var</i>
1790 <result> = or int 15, 40 <i>; yields {int}:result = 47</i>
1791 <result> = or int 4, 8 <i>; yields {int}:result = 12</i>
1792 </pre>
1793 </div>
1794 <!-- _______________________________________________________________________ -->
1796 Instruction</a> </div>
1798 <h5>Syntax:</h5>
1799 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1800 </pre>
1801 <h5>Overview:</h5>
1802 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
1803 or of its two operands. The <tt>xor</tt> is used to implement the
1805 <h5>Arguments:</h5>
1806 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
1808 identical types.</p>
1809 <h5>Semantics:</h5>
1810 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
1811 <p> </p>
1814 <tbody>
1815 <tr>
1816 <td>In0</td>
1817 <td>In1</td>
1818 <td>Out</td>
1819 </tr>
1820 <tr>
1821 <td>0</td>
1822 <td>0</td>
1823 <td>0</td>
1824 </tr>
1825 <tr>
1826 <td>0</td>
1827 <td>1</td>
1828 <td>1</td>
1829 </tr>
1830 <tr>
1831 <td>1</td>
1832 <td>0</td>
1833 <td>1</td>
1834 </tr>
1835 <tr>
1836 <td>1</td>
1837 <td>1</td>
1838 <td>0</td>
1839 </tr>
1840 </tbody>
1841 </table>
1842 </div>
1843 <p> </p>
1844 <h5>Example:</h5>
1845 <pre> <result> = xor int 4, %var <i>; yields {int}:result = 4 ^ %var</i>
1846 <result> = xor int 15, 40 <i>; yields {int}:result = 39</i>
1847 <result> = xor int 4, 8 <i>; yields {int}:result = 12</i>
1848 <result> = xor int %V, -1 <i>; yields {int}:result = ~%V</i>
1849 </pre>
1850 </div>
1851 <!-- _______________________________________________________________________ -->
1853 Instruction</a> </div>
1855 <h5>Syntax:</h5>
1856 <pre> <result> = shl <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1857 </pre>
1858 <h5>Overview:</h5>
1859 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
1860 the left a specified number of bits.</p>
1861 <h5>Arguments:</h5>
1862 <p>The first argument to the '<tt>shl</tt>' instruction must be an <a
1864 type.</p>
1865 <h5>Semantics:</h5>
1866 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
1867 <h5>Example:</h5>
1868 <pre> <result> = shl int 4, ubyte %var <i>; yields {int}:result = 4 << %var</i>
1869 <result> = shl int 4, ubyte 2 <i>; yields {int}:result = 16</i>
1870 <result> = shl int 1, ubyte 10 <i>; yields {int}:result = 1024</i>
1871 </pre>
1872 </div>
1873 <!-- _______________________________________________________________________ -->
1875 Instruction</a> </div>
1877 <h5>Syntax:</h5>
1878 <pre> <result> = shr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1879 </pre>
1880 <h5>Overview:</h5>
1881 <p>The '<tt>shr</tt>' instruction returns the first operand shifted to
1882 the right a specified number of bits.</p>
1883 <h5>Arguments:</h5>
1884 <p>The first argument to the '<tt>shr</tt>' instruction must be an <a
1886 type.</p>
1887 <h5>Semantics:</h5>
1889 most significant bit is duplicated in the newly free'd bit positions.
1890 If the first argument is unsigned, zero bits shall fill the empty
1891 positions.</p>
1892 <h5>Example:</h5>
1893 <pre> <result> = shr int 4, ubyte %var <i>; yields {int}:result = 4 >> %var</i>
1894 <result> = shr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
1895 <result> = shr int 4, ubyte 2 <i>; yields {int}:result = 1</i>
1896 <result> = shr sbyte 4, ubyte 3 <i>; yields {sbyte}:result = 0</i>
1897 <result> = shr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = -1</i>
1898 </pre>
1899 </div>
1901 <!-- ======================================================================= -->
1904 </div>
1908 <p>LLVM supports several instructions to represent vector operations in a
1909 target-independent manner. This instructions cover the element-access and
1910 vector-specific operations needed to process vectors effectively. While LLVM
1911 does directly support these vector operations, many sophisticated algorithms
1912 will want to use target-specific intrinsics to take full advantage of a specific
1913 target.</p>
1915 </div>
1917 <!-- _______________________________________________________________________ -->
1920 </div>
1924 <h5>Syntax:</h5>
1926 <pre>
1927 <result> = extractelement <n x <ty>> <val>, uint <idx> <i>; yields <ty></i>
1928 </pre>
1930 <h5>Overview:</h5>
1932 <p>
1933 The '<tt>extractelement</tt>' instruction extracts a single scalar
1934 element from a packed vector at a specified index.
1935 </p>
1938 <h5>Arguments:</h5>
1940 <p>
1941 The first operand of an '<tt>extractelement</tt>' instruction is a
1943 an index indicating the position from which to extract the element.
1944 The index may be a variable.</p>
1946 <h5>Semantics:</h5>
1948 <p>
1949 The result is a scalar of the same type as the element type of
1950 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
1951 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
1952 results are undefined.
1953 </p>
1955 <h5>Example:</h5>
1957 <pre>
1958 %result = extractelement <4 x int> %vec, uint 0 <i>; yields int</i>
1959 </pre>
1960 </div>
1963 <!-- _______________________________________________________________________ -->
1966 </div>
1970 <h5>Syntax:</h5>
1972 <pre>
1973 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, uint <idx> <i>; yields <n x <ty>></i>
1974 </pre>
1976 <h5>Overview:</h5>
1978 <p>
1979 The '<tt>insertelement</tt>' instruction inserts a scalar
1980 element into a packed vector at a specified index.
1981 </p>
1984 <h5>Arguments:</h5>
1986 <p>
1987 The first operand of an '<tt>insertelement</tt>' instruction is a
1989 scalar value whose type must equal the element type of the first
1990 operand. The third operand is an index indicating the position at
1991 which to insert the value. The index may be a variable.</p>
1993 <h5>Semantics:</h5>
1995 <p>
1996 The result is a packed vector of the same type as <tt>val</tt>. Its
1997 element values are those of <tt>val</tt> except at position
1998 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
1999 exceeds the length of <tt>val</tt>, the results are undefined.
2000 </p>
2002 <h5>Example:</h5>
2004 <pre>
2005 %result = insertelement <4 x int> %vec, int 1, uint 0 <i>; yields <4 x int></i>
2006 </pre>
2007 </div>
2009 <!-- _______________________________________________________________________ -->
2012 </div>
2016 <h5>Syntax:</h5>
2018 <pre>
2019 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x uint> <mask> <i>; yields <n x <ty>></i>
2020 </pre>
2022 <h5>Overview:</h5>
2024 <p>
2025 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2026 from two input vectors, returning a vector of the same type.
2027 </p>
2029 <h5>Arguments:</h5>
2031 <p>
2032 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2033 with types that match each other and types that match the result of the
2034 instruction. The third argument is a shuffle mask, which has the same number
2035 of elements as the other vector type, but whose element type is always 'uint'.
2036 </p>
2038 <p>
2039 The shuffle mask operand is required to be a constant vector with either
2040 constant integer or undef values.
2041 </p>
2043 <h5>Semantics:</h5>
2045 <p>
2046 The elements of the two input vectors are numbered from left to right across
2047 both of the vectors. The shuffle mask operand specifies, for each element of
2048 the result vector, which element of the two input registers the result element
2050 operand may be undef if performing a shuffle from only one vector.
2051 </p>
2053 <h5>Example:</h5>
2055 <pre>
2056 %result = shufflevector <4 x int> %v1, <4 x int> %v2,
2057 <4 x uint> <uint 0, uint 4, uint 1, uint 5> <i>; yields <4 x int></i>
2058 %result = shufflevector <4 x int> %v1, <4 x int> undef,
2059 <4 x uint> <uint 0, uint 1, uint 2, uint 3> <i>; yields <4 x int></i> - Identity shuffle.
2060 </pre>
2061 </div>
2064 <!-- _______________________________________________________________________ -->
2066 Instruction</a> </div>
2068 <h5>Syntax:</h5>
2069 <pre><result> = vsetint <op>, <n x <ty>> <var1>, <var2> <i>; yields <n x bool></i>
2070 </pre>
2072 <h5>Overview:</h5>
2074 <p>The '<tt>vsetint</tt>' instruction takes two integer vectors and
2075 returns a vector of boolean values representing, at each position, the
2076 result of the comparison between the values at that position in the
2077 two operands.</p>
2079 <h5>Arguments:</h5>
2081 <p>The arguments to a '<tt>vsetint</tt>' instruction are a comparison
2082 operation and two value arguments. The value arguments must be of <a
2084 and they must have identical types. The operation argument must be
2085 one of <tt>eq</tt>, <tt>ne</tt>, <tt>slt</tt>, <tt>sgt</tt>,
2086 <tt>sle</tt>, <tt>sge</tt>, <tt>ult</tt>, <tt>ugt</tt>, <tt>ule</tt>,
2087 <tt>uge</tt>, <tt>true</tt>, and <tt>false</tt>. The result is a
2088 packed <tt>bool</tt> value with the same length as each operand.</p>
2090 <h5>Semantics:</h5>
2092 <p>The following table shows the semantics of '<tt>vsetint</tt>'. For
2093 each position of the result, the comparison is done on the
2094 corresponding positions of the two value arguments. Note that the
2095 signedness of the comparison depends on the comparison opcode and
2096 <i>not</i> on the signedness of the value operands. E.g., <tt>vsetint
2097 slt <4 x unsigned> %x, %y</tt> does an elementwise <i>signed</i>
2098 comparison of <tt>%x</tt> and <tt>%y</tt>.</p>
2101 <tbody>
2102 <tr><th>Operation</th><th>Result is true iff</th><th>Comparison is</th></tr>
2103 <tr><td><tt>eq</tt></td><td>var1 == var2</td><td>--</td></tr>
2104 <tr><td><tt>ne</tt></td><td>var1 != var2</td><td>--</td></tr>
2105 <tr><td><tt>slt</tt></td><td>var1 < var2</td><td>signed</td></tr>
2106 <tr><td><tt>sgt</tt></td><td>var1 > var2</td><td>signed</td></tr>
2107 <tr><td><tt>sle</tt></td><td>var1 <= var2</td><td>signed</td></tr>
2108 <tr><td><tt>sge</tt></td><td>var1 >= var2</td><td>signed</td></tr>
2109 <tr><td><tt>ult</tt></td><td>var1 < var2</td><td>unsigned</td></tr>
2110 <tr><td><tt>ugt</tt></td><td>var1 > var2</td><td>unsigned</td></tr>
2111 <tr><td><tt>ule</tt></td><td>var1 <= var2</td><td>unsigned</td></tr>
2112 <tr><td><tt>uge</tt></td><td>var1 >= var2</td><td>unsigned</td></tr>
2113 <tr><td><tt>true</tt></td><td>always</td><td>--</td></tr>
2114 <tr><td><tt>false</tt></td><td>never</td><td>--</td></tr>
2115 </tbody>
2116 </table>
2118 <h5>Example:</h5>
2119 <pre> <result> = vsetint eq <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = false, false</i>
2120 <result> = vsetint ne <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = true, true</i>
2121 <result> = vsetint slt <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = true, false</i>
2122 <result> = vsetint sgt <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = false, true</i>
2123 <result> = vsetint sle <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = true, false</i>
2124 <result> = vsetint sge <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = false, true</i>
2125 </pre>
2126 </div>
2128 <!-- _______________________________________________________________________ -->
2130 Instruction</a> </div>
2132 <h5>Syntax:</h5>
2133 <pre><result> = vsetfp <op>, <n x <ty>> <var1>, <var2> <i>; yields <n x bool></i>
2134 </pre>
2136 <h5>Overview:</h5>
2138 <p>The '<tt>vsetfp</tt>' instruction takes two floating point vector
2139 arguments and returns a vector of boolean values representing, at each
2140 position, the result of the comparison between the values at that
2141 position in the two operands.</p>
2143 <h5>Arguments:</h5>
2145 <p>The arguments to a '<tt>vsetfp</tt>' instruction are a comparison
2146 operation and two value arguments. The value arguments must be of <a
2148 type, and they must have identical types. The operation argument must
2149 be one of <tt>eq</tt>, <tt>ne</tt>, <tt>lt</tt>, <tt>gt</tt>,
2150 <tt>le</tt>, <tt>ge</tt>, <tt>oeq</tt>, <tt>one</tt>, <tt>olt</tt>,
2151 <tt>ogt</tt>, <tt>ole</tt>, <tt>oge</tt>, <tt>ueq</tt>, <tt>une</tt>,
2152 <tt>ult</tt>, <tt>ugt</tt>, <tt>ule</tt>, <tt>uge</tt>, <tt>o</tt>,
2153 <tt>u</tt>, <tt>true</tt>, and <tt>false</tt>. The result is a packed
2154 <tt>bool</tt> value with the same length as each operand.</p>
2156 <h5>Semantics:</h5>
2158 <p>The following table shows the semantics of '<tt>vsetfp</tt>' for
2159 floating point types. If either operand is a floating point Not a
2160 Number (NaN) value, the operation is unordered, and the value in the
2161 first column below is produced at that position. Otherwise, the
2162 operation is ordered, and the value in the second column is
2163 produced.</p>
2166 <tbody>
2167 <tr><th>Operation</th><th>If unordered<th>Otherwise true iff</th></tr>
2168 <tr><td><tt>eq</tt></td><td>undefined</td><td>var1 == var2</td></tr>
2169 <tr><td><tt>ne</tt></td><td>undefined</td><td>var1 != var2</td></tr>
2170 <tr><td><tt>lt</tt></td><td>undefined</td><td>var1 < var2</td></tr>
2171 <tr><td><tt>gt</tt></td><td>undefined</td><td>var1 > var2</td></tr>
2172 <tr><td><tt>le</tt></td><td>undefined</td><td>var1 <= var2</td></tr>
2173 <tr><td><tt>ge</tt></td><td>undefined</td><td>var1 >= var2</td></tr>
2174 <tr><td><tt>oeq</tt></td><td>false</td><td>var1 == var2</td></tr>
2175 <tr><td><tt>one</tt></td><td>false</td><td>var1 != var2</td></tr>
2176 <tr><td><tt>olt</tt></td><td>false</td><td>var1 < var2</td></tr>
2177 <tr><td><tt>ogt</tt></td><td>false</td><td>var1 > var2</td></tr>
2178 <tr><td><tt>ole</tt></td><td>false</td><td>var1 <= var2</td></tr>
2179 <tr><td><tt>oge</tt></td><td>false</td><td>var1 >= var2</td></tr>
2180 <tr><td><tt>ueq</tt></td><td>true</td><td>var1 == var2</td></tr>
2181 <tr><td><tt>une</tt></td><td>true</td><td>var1 != var2</td></tr>
2182 <tr><td><tt>ult</tt></td><td>true</td><td>var1 < var2</td></tr>
2183 <tr><td><tt>ugt</tt></td><td>true</td><td>var1 > var2</td></tr>
2184 <tr><td><tt>ule</tt></td><td>true</td><td>var1 <= var2</td></tr>
2185 <tr><td><tt>uge</tt></td><td>true</td><td>var1 >= var2</td></tr>
2186 <tr><td><tt>o</tt></td><td>false</td><td>always</td></tr>
2187 <tr><td><tt>u</tt></td><td>true</td><td>never</td></tr>
2188 <tr><td><tt>true</tt></td><td>true</td><td>always</td></tr>
2189 <tr><td><tt>false</tt></td><td>false</td><td>never</td></tr>
2190 </tbody>
2191 </table>
2193 <h5>Example:</h5>
2194 <pre> <result> = vsetfp eq <2 x float> <float 0.0, float 1.0>, <float 1.0, float 0.0> <i>; yields {<2 x bool>}:result = false, false</i>
2195 <result> = vsetfp ne <2 x float> <float 0.0, float 1.0>, <float 1.0, float 0.0> <i>; yields {<2 x bool>}:result = true, true</i>
2196 <result> = vsetfp lt <2 x float> <float 0.0, float 1.0>, <float 1.0, float 0.0> <i>; yields {<2 x bool>}:result = true, false</i>
2197 <result> = vsetfp gt <2 x float> <float 0.0, float 1.0>, <float 1.0, float 0.0> <i>; yields {<2 x bool>}:result = false, true</i>
2198 <result> = vsetfp le <2 x float> <float 0.0, float 1.0>, <float 1.0, float 0.0> <i>; yields {<2 x bool>}:result = true, false</i>
2199 <result> = vsetfp ge <2 x float> <float 0.0, float 1.0>, <float 1.0, float 0.0> <i>; yields {<2 x bool>}:result = false, true</i>
2200 </pre>
2201 </div>
2203 <!-- _______________________________________________________________________ -->
2206 </div>
2210 <h5>Syntax:</h5>
2212 <pre>
2213 <result> = vselect <n x bool> <cond>, <n x <ty>> <val1>, <n x <ty>> <val2> <i>; yields <n x <ty>></i>
2214 </pre>
2216 <h5>Overview:</h5>
2218 <p>
2219 The '<tt>vselect</tt>' instruction chooses one value at each position
2220 of a vector based on a condition.
2221 </p>
2224 <h5>Arguments:</h5>
2226 <p>
2227 The '<tt>vselect</tt>' instruction requires a <a
2229 condition at each vector position, and two values of the same packed
2230 type. All three operands must have the same length. The type of the
2231 result is the same as the type of the two value operands.</p>
2233 <h5>Semantics:</h5>
2235 <p>
2236 At each position where the <tt>bool</tt> vector is true, that position
2237 of the result gets its value from the first value argument; otherwise,
2238 it gets its value from the second value argument.
2239 </p>
2241 <h5>Example:</h5>
2243 <pre>
2244 %X = vselect bool <2 x bool> <bool true, bool false>, <2 x ubyte> <ubyte 17, ubyte 17>,
2245 <2 x ubyte> <ubyte 42, ubyte 42> <i>; yields <2 x ubyte>:17, 42</i>
2246 </pre>
2247 </div>
2251 <!-- ======================================================================= -->
2254 </div>
2258 <p>A key design point of an SSA-based representation is how it
2259 represents memory. In LLVM, no memory locations are in SSA form, which
2260 makes things very simple. This section describes how to read, write,
2261 allocate, and free memory in LLVM.</p>
2263 </div>
2265 <!-- _______________________________________________________________________ -->
2268 </div>
2272 <h5>Syntax:</h5>
2274 <pre>
2275 <result> = malloc <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2276 </pre>
2278 <h5>Overview:</h5>
2280 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2281 heap and returns a pointer to it.</p>
2283 <h5>Arguments:</h5>
2285 <p>The '<tt>malloc</tt>' instruction allocates
2286 <tt>sizeof(<type>)*NumElements</tt>
2287 bytes of memory from the operating system and returns a pointer of the
2289 number of elements allocated. If an alignment is specified, the value result
2290 of the allocation is guaranteed to be aligned to at least that boundary. If
2291 not specified, or if zero, the target can choose to align the allocation on any
2292 convenient boundary.</p>
2294 <p>'<tt>type</tt>' must be a sized type.</p>
2296 <h5>Semantics:</h5>
2299 a pointer is returned.</p>
2301 <h5>Example:</h5>
2303 <pre>
2304 %array = malloc [4 x ubyte ] <i>; yields {[%4 x ubyte]*}:array</i>
2307 %array1 = malloc ubyte, uint 4 <i>; yields {ubyte*}:array1</i>
2308 %array2 = malloc [12 x ubyte], uint %size <i>; yields {[12 x ubyte]*}:array2</i>
2309 %array3 = malloc int, uint 4, align 1024 <i>; yields {int*}:array3</i>
2310 %array4 = malloc int, align 1024 <i>; yields {int*}:array4</i>
2311 </pre>
2312 </div>
2314 <!-- _______________________________________________________________________ -->
2317 </div>
2321 <h5>Syntax:</h5>
2323 <pre>
2324 free <type> <value> <i>; yields {void}</i>
2325 </pre>
2327 <h5>Overview:</h5>
2329 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2330 memory heap to be reallocated in the future.</p>
2332 <h5>Arguments:</h5>
2334 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2336 instruction.</p>
2338 <h5>Semantics:</h5>
2340 <p>Access to the memory pointed to by the pointer is no longer defined
2341 after this instruction executes.</p>
2343 <h5>Example:</h5>
2345 <pre>
2347 free [4 x ubyte]* %array
2348 </pre>
2349 </div>
2351 <!-- _______________________________________________________________________ -->
2354 </div>
2358 <h5>Syntax:</h5>
2360 <pre>
2361 <result> = alloca <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2362 </pre>
2364 <h5>Overview:</h5>
2366 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
2367 stack frame of the procedure that is live until the current function
2368 returns to its caller.</p>
2370 <h5>Arguments:</h5>
2372 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2373 bytes of memory on the runtime stack, returning a pointer of the
2375 number of elements allocated. If an alignment is specified, the value result
2376 of the allocation is guaranteed to be aligned to at least that boundary. If
2377 not specified, or if zero, the target can choose to align the allocation on any
2378 convenient boundary.</p>
2380 <p>'<tt>type</tt>' may be any sized type.</p>
2382 <h5>Semantics:</h5>
2384 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2385 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2386 instruction is commonly used to represent automatic variables that must
2387 have an address available. When the function returns (either with the <tt><a
2389 instructions), the memory is reclaimed.</p>
2391 <h5>Example:</h5>
2393 <pre>
2394 %ptr = alloca int <i>; yields {int*}:ptr</i>
2395 %ptr = alloca int, uint 4 <i>; yields {int*}:ptr</i>
2396 %ptr = alloca int, uint 4, align 1024 <i>; yields {int*}:ptr</i>
2397 %ptr = alloca int, align 1024 <i>; yields {int*}:ptr</i>
2398 </pre>
2399 </div>
2401 <!-- _______________________________________________________________________ -->
2403 Instruction</a> </div>
2405 <h5>Syntax:</h5>
2406 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
2407 <h5>Overview:</h5>
2408 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2409 <h5>Arguments:</h5>
2410 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2411 address from which to load. The pointer must point to a <a
2413 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2414 the number or order of execution of this <tt>load</tt> with other
2416 instructions. </p>
2417 <h5>Semantics:</h5>
2418 <p>The location of memory pointed to is loaded.</p>
2419 <h5>Examples:</h5>
2421 <a
2423 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2424 </pre>
2425 </div>
2426 <!-- _______________________________________________________________________ -->
2428 Instruction</a> </div>
2429 <h5>Syntax:</h5>
2430 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2431 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2432 </pre>
2433 <h5>Overview:</h5>
2434 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2435 <h5>Arguments:</h5>
2436 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2437 to store and an address in which to store it. The type of the '<tt><pointer></tt>'
2438 operand must be a pointer to the type of the '<tt><value></tt>'
2439 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2440 optimizer is not allowed to modify the number or order of execution of
2441 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2443 <h5>Semantics:</h5>
2444 <p>The contents of memory are updated to contain '<tt><value></tt>'
2445 at the location specified by the '<tt><pointer></tt>' operand.</p>
2446 <h5>Example:</h5>
2448 <a
2450 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2451 </pre>
2452 <!-- _______________________________________________________________________ -->
2455 </div>
2458 <h5>Syntax:</h5>
2459 <pre>
2460 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2461 </pre>
2463 <h5>Overview:</h5>
2465 <p>
2466 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2467 subelement of an aggregate data structure.</p>
2469 <h5>Arguments:</h5>
2471 <p>This instruction takes a list of integer constants that indicate what
2472 elements of the aggregate object to index to. The actual types of the arguments
2473 provided depend on the type of the first pointer argument. The
2474 '<tt>getelementptr</tt>' instruction is used to index down through the type
2475 levels of a structure or to a specific index in an array. When indexing into a
2476 structure, only <tt>uint</tt>
2477 integer constants are allowed. When indexing into an array or pointer,
2478 <tt>int</tt> and <tt>long</tt> indexes are allowed of any sign.</p>
2480 <p>For example, let's consider a C code fragment and how it gets
2481 compiled to LLVM:</p>
2483 <pre>
2484 struct RT {
2485 char A;
2486 int B[10][20];
2487 char C;
2488 };
2489 struct ST {
2490 int X;
2491 double Y;
2492 struct RT Z;
2493 };
2495 int *foo(struct ST *s) {
2496 return &s[1].Z.B[5][13];
2497 }
2498 </pre>
2500 <p>The LLVM code generated by the GCC frontend is:</p>
2502 <pre>
2503 %RT = type { sbyte, [10 x [20 x int]], sbyte }
2504 %ST = type { int, double, %RT }
2506 implementation
2508 int* %foo(%ST* %s) {
2509 entry:
2510 %reg = getelementptr %ST* %s, int 1, uint 2, uint 1, int 5, int 13
2511 ret int* %reg
2512 }
2513 </pre>
2515 <h5>Semantics:</h5>
2517 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2521 types require <tt>uint</tt> <b>constants</b>.</p>
2523 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2524 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ int, double, %RT
2525 }</tt>' type, a structure. The second index indexes into the third element of
2526 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ sbyte, [10 x [20 x int]],
2527 sbyte }</tt>' type, another structure. The third index indexes into the second
2528 element of the structure, yielding a '<tt>[10 x [20 x int]]</tt>' type, an
2529 array. The two dimensions of the array are subscripted into, yielding an
2530 '<tt>int</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2531 to this element, thus computing a value of '<tt>int*</tt>' type.</p>
2533 <p>Note that it is perfectly legal to index partially through a
2534 structure, returning a pointer to an inner element. Because of this,
2535 the LLVM code for the given testcase is equivalent to:</p>
2537 <pre>
2538 int* %foo(%ST* %s) {
2539 %t1 = getelementptr %ST* %s, int 1 <i>; yields %ST*:%t1</i>
2540 %t2 = getelementptr %ST* %t1, int 0, uint 2 <i>; yields %RT*:%t2</i>
2541 %t3 = getelementptr %RT* %t2, int 0, uint 1 <i>; yields [10 x [20 x int]]*:%t3</i>
2542 %t4 = getelementptr [10 x [20 x int]]* %t3, int 0, int 5 <i>; yields [20 x int]*:%t4</i>
2543 %t5 = getelementptr [20 x int]* %t4, int 0, int 13 <i>; yields int*:%t5</i>
2544 ret int* %t5
2545 }
2546 </pre>
2548 <p>Note that it is undefined to access an array out of bounds: array and
2549 pointer indexes must always be within the defined bounds of the array type.
2550 The one exception for this rules is zero length arrays. These arrays are
2551 defined to be accessible as variable length arrays, which requires access
2552 beyond the zero'th element.</p>
2554 <h5>Example:</h5>
2556 <pre>
2557 <i>; yields [12 x ubyte]*:aptr</i>
2558 %aptr = getelementptr {int, [12 x ubyte]}* %sptr, long 0, uint 1
2559 </pre>
2561 </div>
2562 <!-- ======================================================================= -->
2566 instructions, which defy better classification.</p>
2567 </div>
2568 <!-- _______________________________________________________________________ -->
2570 Instruction</a> </div>
2572 <h5>Syntax:</h5>
2573 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
2574 <h5>Overview:</h5>
2575 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
2576 the SSA graph representing the function.</p>
2577 <h5>Arguments:</h5>
2578 <p>The type of the incoming values are specified with the first type
2579 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
2580 as arguments, with one pair for each predecessor basic block of the
2582 type may be used as the value arguments to the PHI node. Only labels
2583 may be used as the label arguments.</p>
2584 <p>There must be no non-phi instructions between the start of a basic
2585 block and the PHI instructions: i.e. PHI instructions must be first in
2586 a basic block.</p>
2587 <h5>Semantics:</h5>
2588 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
2589 value specified by the parameter, depending on which basic block we
2591 <h5>Example:</h5>
2592 <pre>Loop: ; Infinite loop that counts from 0 on up...<br> %indvar = phi uint [ 0, %LoopHeader ], [ %nextindvar, %Loop ]<br> %nextindvar = add uint %indvar, 1<br> br label %Loop<br></pre>
2593 </div>
2595 <!-- _______________________________________________________________________ -->
2598 </div>
2602 <h5>Syntax:</h5>
2604 <pre>
2605 <result> = cast <ty> <value> to <ty2> <i>; yields ty2</i>
2606 </pre>
2608 <h5>Overview:</h5>
2610 <p>
2611 The '<tt>cast</tt>' instruction is used as the primitive means to convert
2612 integers to floating point, change data type sizes, and break type safety (by
2613 casting pointers).
2614 </p>
2617 <h5>Arguments:</h5>
2619 <p>
2620 The '<tt>cast</tt>' instruction takes a value to cast, which must be a first
2621 class value, and a type to cast it to, which must also be a <a
2623 </p>
2625 <h5>Semantics:</h5>
2627 <p>
2628 This instruction follows the C rules for explicit casts when determining how the
2629 data being cast must change to fit in its new container.
2630 </p>
2632 <p>
2633 When casting to bool, any value that would be considered true in the context of
2634 a C '<tt>if</tt>' condition is converted to the boolean '<tt>true</tt>' values,
2635 all else are '<tt>false</tt>'.
2636 </p>
2638 <p>
2639 When extending an integral value from a type of one signness to another (for
2640 example '<tt>sbyte</tt>' to '<tt>ulong</tt>'), the value is sign-extended if the
2641 <b>source</b> value is signed, and zero-extended if the source value is
2642 unsigned. <tt>bool</tt> values are always zero extended into either zero or
2643 one.
2644 </p>
2646 <h5>Example:</h5>
2648 <pre>
2649 %X = cast int 257 to ubyte <i>; yields ubyte:1</i>
2650 %Y = cast int 123 to bool <i>; yields bool:true</i>
2651 </pre>
2652 </div>
2654 <!-- _______________________________________________________________________ -->
2657 </div>
2661 <h5>Syntax:</h5>
2663 <pre>
2664 <result> = select bool <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
2665 </pre>
2667 <h5>Overview:</h5>
2669 <p>
2670 The '<tt>select</tt>' instruction is used to choose one value based on a
2671 condition, without branching.
2672 </p>
2675 <h5>Arguments:</h5>
2677 <p>
2678 The '<tt>select</tt>' instruction requires a boolean value indicating the condition, and two values of the same <a href="#t_firstclass">first class</a> type.
2679 </p>
2681 <h5>Semantics:</h5>
2683 <p>
2684 If the boolean condition evaluates to true, the instruction returns the first
2685 value argument; otherwise, it returns the second value argument.
2686 </p>
2688 <h5>Example:</h5>
2690 <pre>
2691 %X = select bool true, ubyte 17, ubyte 42 <i>; yields ubyte:17</i>
2692 </pre>
2693 </div>
2696 <!-- _______________________________________________________________________ -->
2699 </div>
2703 <h5>Syntax:</h5>
2704 <pre>
2705 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
2706 </pre>
2708 <h5>Overview:</h5>
2710 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
2712 <h5>Arguments:</h5>
2714 <p>This instruction requires several arguments:</p>
2716 <ol>
2717 <li>
2720 function call is eligible for tail call optimization. Note that calls may
2723 </li>
2724 <li>
2726 convention</a> the call should use. If none is specified, the call defaults
2727 to using C calling conventions.
2728 </li>
2729 <li>
2730 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
2731 being invoked. The argument types must match the types implied by this
2732 signature. This type can be omitted if the function is not varargs and
2733 if the function type does not return a pointer to a function.</p>
2734 </li>
2735 <li>
2736 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
2737 be invoked. In most cases, this is a direct function invocation, but
2738 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
2739 to function value.</p>
2740 </li>
2741 <li>
2742 <p>'<tt>function args</tt>': argument list whose types match the
2743 function signature argument types. All arguments must be of
2745 indicates the function accepts a variable number of arguments, the extra
2746 arguments can be specified.</p>
2747 </li>
2748 </ol>
2750 <h5>Semantics:</h5>
2752 <p>The '<tt>call</tt>' instruction is used to cause control flow to
2753 transfer to a specified function, with its incoming arguments bound to
2755 instruction in the called function, control flow continues with the
2756 instruction after the function call, and the return value of the
2757 function is bound to the result argument. This is a simpler case of
2760 <h5>Example:</h5>
2762 <pre>
2763 %retval = call int %test(int %argc)
2764 call int(sbyte*, ...) *%printf(sbyte* %msg, int 12, sbyte 42);
2765 %X = tail call int %foo()
2767 </pre>
2769 </div>
2771 <!-- _______________________________________________________________________ -->
2774 </div>
2778 <h5>Syntax:</h5>
2780 <pre>
2781 <resultval> = va_arg <va_list*> <arglist>, <argty>
2782 </pre>
2784 <h5>Overview:</h5>
2786 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
2788 <tt>va_arg</tt> macro in C.</p>
2790 <h5>Arguments:</h5>
2792 <p>This instruction takes a <tt>va_list*</tt> value and the type of
2793 the argument. It returns a value of the specified argument type and
2794 increments the <tt>va_list</tt> to point to the next argument. Again, the
2795 actual type of <tt>va_list</tt> is target specific.</p>
2797 <h5>Semantics:</h5>
2799 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
2800 type from the specified <tt>va_list</tt> and causes the
2801 <tt>va_list</tt> to point to the next argument. For more information,
2803 Functions</a>.</p>
2805 <p>It is legal for this instruction to be called in a function which does not
2806 take a variable number of arguments, for example, the <tt>vfprintf</tt>
2807 function.</p>
2809 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
2811 argument.</p>
2813 <h5>Example:</h5>
2817 </div>
2819 <!-- *********************************************************************** -->
2821 <!-- *********************************************************************** -->
2826 well known names and semantics and are required to follow certain
2827 restrictions. Overall, these instructions represent an extension mechanism for
2828 the LLVM language that does not require changing all of the transformations in
2829 LLVM to add to the language (or the bytecode reader/writer, the parser,
2830 etc...).</p>
2833 prefix is reserved in LLVM for intrinsic names; thus, functions may not be named
2834 this. Intrinsic functions must always be external functions: you cannot define
2835 the body of intrinsic functions. Intrinsic functions may only be used in call
2836 or invoke instructions: it is illegal to take the address of an intrinsic
2837 function. Additionally, because intrinsic functions are part of the LLVM
2838 language, it is required that they all be documented here if any are added.</p>
2841 <p>To learn how to add an intrinsic function, please see the <a
2843 </p>
2845 </div>
2847 <!-- ======================================================================= -->
2850 </div>
2854 <p>Variable argument support is defined in LLVM with the <a
2856 intrinsic functions. These functions are related to the similarly
2857 named macros defined in the <tt><stdarg.h></tt> header file.</p>
2859 <p>All of these functions operate on arguments that use a
2861 language reference manual does not define what this type is, so all
2862 transformations should be prepared to handle intrinsics with any type
2863 used.</p>
2866 instruction and the variable argument handling intrinsic functions are
2867 used.</p>
2869 <pre>
2870 int %test(int %X, ...) {
2871 ; Initialize variable argument processing
2872 %ap = alloca sbyte*
2875 ; Read a single integer argument
2876 %tmp = va_arg sbyte** %ap, int
2878 ; Demonstrate usage of llvm.va_copy and llvm.va_end
2879 %aq = alloca sbyte*
2883 ; Stop processing of arguments.
2885 ret int %tmp
2886 }
2887 </pre>
2888 </div>
2890 <!-- _______________________________________________________________________ -->
2893 </div>
2897 <h5>Syntax:</h5>
2898 <pre> declare void %llvm.va_start(<va_list>* <arglist>)<br></pre>
2899 <h5>Overview:</h5>
2900 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
2901 <tt>*<arglist></tt> for subsequent use by <tt><a
2904 <h5>Arguments:</h5>
2906 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
2908 <h5>Semantics:</h5>
2910 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
2911 macro available in C. In a target-dependent way, it initializes the
2912 <tt>va_list</tt> element the argument points to, so that the next call to
2913 <tt>va_arg</tt> will produce the first variable argument passed to the function.
2914 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
2915 last argument of the function, the compiler can figure that out.</p>
2917 </div>
2919 <!-- _______________________________________________________________________ -->
2922 </div>
2925 <h5>Syntax:</h5>
2926 <pre> declare void %llvm.va_end(<va_list*> <arglist>)<br></pre>
2927 <h5>Overview:</h5>
2928 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
2931 <h5>Arguments:</h5>
2932 <p>The argument is a <tt>va_list</tt> to destroy.</p>
2933 <h5>Semantics:</h5>
2934 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
2935 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
2938 with calls to <tt>llvm.va_end</tt>.</p>
2939 </div>
2941 <!-- _______________________________________________________________________ -->
2944 </div>
2948 <h5>Syntax:</h5>
2950 <pre>
2951 declare void %llvm.va_copy(<va_list>* <destarglist>,
2952 <va_list>* <srcarglist>)
2953 </pre>
2955 <h5>Overview:</h5>
2957 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position from
2958 the source argument list to the destination argument list.</p>
2960 <h5>Arguments:</h5>
2962 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
2963 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
2966 <h5>Semantics:</h5>
2968 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> macro
2969 available in C. In a target-dependent way, it copies the source
2970 <tt>va_list</tt> element into the destination list. This intrinsic is necessary
2972 arbitrarily complex and require memory allocation, for example.</p>
2974 </div>
2976 <!-- ======================================================================= -->
2979 </div>
2983 <p>
2985 Collection</a> requires the implementation and generation of these intrinsics.
2987 stack</a>, as well as garbage collector implementations that require <a
2989 Front-ends for type-safe garbage collected languages should generate these
2990 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
2992 </p>
2993 </div>
2995 <!-- _______________________________________________________________________ -->
2998 </div>
3002 <h5>Syntax:</h5>
3004 <pre>
3005 declare void %llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
3006 </pre>
3008 <h5>Overview:</h5>
3010 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
3011 the code generator, and allows some metadata to be associated with it.</p>
3013 <h5>Arguments:</h5>
3015 <p>The first argument specifies the address of a stack object that contains the
3016 root pointer. The second pointer (which must be either a constant or a global
3017 value address) contains the meta-data to be associated with the root.</p>
3019 <h5>Semantics:</h5>
3022 location. At compile-time, the code generator generates information to allow
3023 the runtime to find the pointer at GC safe points.
3024 </p>
3026 </div>
3029 <!-- _______________________________________________________________________ -->
3032 </div>
3036 <h5>Syntax:</h5>
3038 <pre>
3039 declare sbyte* %llvm.gcread(sbyte* %ObjPtr, sbyte** %Ptr)
3040 </pre>
3042 <h5>Overview:</h5>
3044 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
3045 locations, allowing garbage collector implementations that require read
3046 barriers.</p>
3048 <h5>Arguments:</h5>
3050 <p>The second argument is the address to read from, which should be an address
3051 allocated from the garbage collector. The first object is a pointer to the
3052 start of the referenced object, if needed by the language runtime (otherwise
3053 null).</p>
3055 <h5>Semantics:</h5>
3057 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
3058 instruction, but may be replaced with substantially more complex code by the
3059 garbage collector runtime, as needed.</p>
3061 </div>
3064 <!-- _______________________________________________________________________ -->
3067 </div>
3071 <h5>Syntax:</h5>
3073 <pre>
3074 declare void %llvm.gcwrite(sbyte* %P1, sbyte* %Obj, sbyte** %P2)
3075 </pre>
3077 <h5>Overview:</h5>
3079 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
3080 locations, allowing garbage collector implementations that require write
3081 barriers (such as generational or reference counting collectors).</p>
3083 <h5>Arguments:</h5>
3085 <p>The first argument is the reference to store, the second is the start of the
3086 object to store it to, and the third is the address of the field of Obj to
3087 store to. If the runtime does not require a pointer to the object, Obj may be
3088 null.</p>
3090 <h5>Semantics:</h5>
3092 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
3093 instruction, but may be replaced with substantially more complex code by the
3094 garbage collector runtime, as needed.</p>
3096 </div>
3100 <!-- ======================================================================= -->
3103 </div>
3106 <p>
3107 These intrinsics are provided by LLVM to expose special features that may only
3108 be implemented with code generator support.
3109 </p>
3111 </div>
3113 <!-- _______________________________________________________________________ -->
3116 </div>
3120 <h5>Syntax:</h5>
3121 <pre>
3122 declare sbyte *%llvm.returnaddress(uint <level>)
3123 </pre>
3125 <h5>Overview:</h5>
3127 <p>
3128 The '<tt>llvm.returnaddress</tt>' intrinsic returns a target-specific value
3129 indicating the return address of the current function or one of its callers.
3130 </p>
3132 <h5>Arguments:</h5>
3134 <p>
3135 The argument to this intrinsic indicates which function to return the address
3136 for. Zero indicates the calling function, one indicates its caller, etc. The
3137 argument is <b>required</b> to be a constant integer value.
3138 </p>
3140 <h5>Semantics:</h5>
3142 <p>
3143 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
3144 the return address of the specified call frame, or zero if it cannot be
3145 identified. The value returned by this intrinsic is likely to be incorrect or 0
3146 for arguments other than zero, so it should only be used for debugging purposes.
3147 </p>
3149 <p>
3150 Note that calling this intrinsic does not prevent function inlining or other
3151 aggressive transformations, so the value returned may not be that of the obvious
3152 source-language caller.
3153 </p>
3154 </div>
3157 <!-- _______________________________________________________________________ -->
3160 </div>
3164 <h5>Syntax:</h5>
3165 <pre>
3166 declare sbyte *%llvm.frameaddress(uint <level>)
3167 </pre>
3169 <h5>Overview:</h5>
3171 <p>
3172 The '<tt>llvm.frameaddress</tt>' intrinsic returns the target-specific frame
3173 pointer value for the specified stack frame.
3174 </p>
3176 <h5>Arguments:</h5>
3178 <p>
3179 The argument to this intrinsic indicates which function to return the frame
3180 pointer for. Zero indicates the calling function, one indicates its caller,
3181 etc. The argument is <b>required</b> to be a constant integer value.
3182 </p>
3184 <h5>Semantics:</h5>
3186 <p>
3187 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
3188 the frame address of the specified call frame, or zero if it cannot be
3189 identified. The value returned by this intrinsic is likely to be incorrect or 0
3190 for arguments other than zero, so it should only be used for debugging purposes.
3191 </p>
3193 <p>
3194 Note that calling this intrinsic does not prevent function inlining or other
3195 aggressive transformations, so the value returned may not be that of the obvious
3196 source-language caller.
3197 </p>
3198 </div>
3200 <!-- _______________________________________________________________________ -->
3203 </div>
3207 <h5>Syntax:</h5>
3208 <pre>
3209 declare sbyte *%llvm.stacksave()
3210 </pre>
3212 <h5>Overview:</h5>
3214 <p>
3215 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
3217 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
3218 features like scoped automatic variable sized arrays in C99.
3219 </p>
3221 <h5>Semantics:</h5>
3223 <p>
3224 This intrinsic returns a opaque pointer value that can be passed to <a
3226 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
3227 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
3228 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
3230 that were allocated after the <tt>llvm.stacksave</tt> was executed.
3231 </p>
3233 </div>
3235 <!-- _______________________________________________________________________ -->
3238 </div>
3242 <h5>Syntax:</h5>
3243 <pre>
3244 declare void %llvm.stackrestore(sbyte* %ptr)
3245 </pre>
3247 <h5>Overview:</h5>
3249 <p>
3250 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
3251 the function stack to the state it was in when the corresponding <a
3253 useful for implementing language features like scoped automatic variable sized
3254 arrays in C99.
3255 </p>
3257 <h5>Semantics:</h5>
3259 <p>
3261 </p>
3263 </div>
3266 <!-- _______________________________________________________________________ -->
3269 </div>
3273 <h5>Syntax:</h5>
3274 <pre>
3275 declare void %llvm.prefetch(sbyte * <address>,
3276 uint <rw>, uint <locality>)
3277 </pre>
3279 <h5>Overview:</h5>
3282 <p>
3283 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
3284 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
3285 no
3286 effect on the behavior of the program but can change its performance
3287 characteristics.
3288 </p>
3290 <h5>Arguments:</h5>
3292 <p>
3293 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
3294 determining if the fetch should be for a read (0) or write (1), and
3295 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
3296 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
3297 <tt>locality</tt> arguments must be constant integers.
3298 </p>
3300 <h5>Semantics:</h5>
3302 <p>
3303 This intrinsic does not modify the behavior of the program. In particular,
3304 prefetches cannot trap and do not produce a value. On targets that support this
3305 intrinsic, the prefetch can provide hints to the processor cache for better
3306 performance.
3307 </p>
3309 </div>
3311 <!-- _______________________________________________________________________ -->
3314 </div>
3318 <h5>Syntax:</h5>
3319 <pre>
3320 declare void %llvm.pcmarker( uint <id> )
3321 </pre>
3323 <h5>Overview:</h5>
3326 <p>
3327 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
3328 (PC) in a region of
3329 code to simulators and other tools. The method is target specific, but it is
3330 expected that the marker will use exported symbols to transmit the PC of the marker.
3331 The marker makes no guarantees that it will remain with any specific instruction
3332 after optimizations. It is possible that the presence of a marker will inhibit
3333 optimizations. The intended use is to be inserted after optimizations to allow
3334 correlations of simulation runs.
3335 </p>
3337 <h5>Arguments:</h5>
3339 <p>
3340 <tt>id</tt> is a numerical id identifying the marker.
3341 </p>
3343 <h5>Semantics:</h5>
3345 <p>
3346 This intrinsic does not modify the behavior of the program. Backends that do not
3347 support this intrinisic may ignore it.
3348 </p>
3350 </div>
3352 <!-- _______________________________________________________________________ -->
3355 </div>
3359 <h5>Syntax:</h5>
3360 <pre>
3361 declare ulong %llvm.readcyclecounter( )
3362 </pre>
3364 <h5>Overview:</h5>
3367 <p>
3368 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
3369 counter register (or similar low latency, high accuracy clocks) on those targets
3370 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
3371 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
3372 should only be used for small timings.
3373 </p>
3375 <h5>Semantics:</h5>
3377 <p>
3378 When directly supported, reading the cycle counter should not modify any memory.
3379 Implementations are allowed to either return a application specific value or a
3380 system wide value. On backends without support, this is lowered to a constant 0.
3381 </p>
3383 </div>
3385 <!-- ======================================================================= -->
3388 </div>
3391 <p>
3392 LLVM provides intrinsics for a few important standard C library functions.
3393 These intrinsics allow source-language front-ends to pass information about the
3394 alignment of the pointer arguments to the code generator, providing opportunity
3395 for more efficient code generation.
3396 </p>
3398 </div>
3400 <!-- _______________________________________________________________________ -->
3403 </div>
3407 <h5>Syntax:</h5>
3408 <pre>
3409 declare void %llvm.memcpy.i32(sbyte* <dest>, sbyte* <src>,
3410 uint <len>, uint <align>)
3411 declare void %llvm.memcpy.i64(sbyte* <dest>, sbyte* <src>,
3412 ulong <len>, uint <align>)
3413 </pre>
3415 <h5>Overview:</h5>
3417 <p>
3418 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
3419 location to the destination location.
3420 </p>
3422 <p>
3423 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
3424 intrinsics do not return a value, and takes an extra alignment argument.
3425 </p>
3427 <h5>Arguments:</h5>
3429 <p>
3430 The first argument is a pointer to the destination, the second is a pointer to
3431 the source. The third argument is an integer argument
3432 specifying the number of bytes to copy, and the fourth argument is the alignment
3433 of the source and destination locations.
3434 </p>
3436 <p>
3437 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3438 the caller guarantees that both the source and destination pointers are aligned
3439 to that boundary.
3440 </p>
3442 <h5>Semantics:</h5>
3444 <p>
3445 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
3446 location to the destination location, which are not allowed to overlap. It
3448 some boundary, this can be specified as the fourth argument, otherwise it should
3449 be set to 0 or 1.
3450 </p>
3451 </div>
3454 <!-- _______________________________________________________________________ -->
3457 </div>
3461 <h5>Syntax:</h5>
3462 <pre>
3463 declare void %llvm.memmove.i32(sbyte* <dest>, sbyte* <src>,
3464 uint <len>, uint <align>)
3465 declare void %llvm.memmove.i64(sbyte* <dest>, sbyte* <src>,
3466 ulong <len>, uint <align>)
3467 </pre>
3469 <h5>Overview:</h5>
3471 <p>
3472 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
3473 location to the destination location. It is similar to the
3474 '<tt>llvm.memcmp</tt>' intrinsic but allows the two memory locations to overlap.
3475 </p>
3477 <p>
3478 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
3479 intrinsics do not return a value, and takes an extra alignment argument.
3480 </p>
3482 <h5>Arguments:</h5>
3484 <p>
3485 The first argument is a pointer to the destination, the second is a pointer to
3486 the source. The third argument is an integer argument
3487 specifying the number of bytes to copy, and the fourth argument is the alignment
3488 of the source and destination locations.
3489 </p>
3491 <p>
3492 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3493 the caller guarantees that the source and destination pointers are aligned to
3494 that boundary.
3495 </p>
3497 <h5>Semantics:</h5>
3499 <p>
3500 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
3501 location to the destination location, which may overlap. It
3503 some boundary, this can be specified as the fourth argument, otherwise it should
3504 be set to 0 or 1.
3505 </p>
3506 </div>
3509 <!-- _______________________________________________________________________ -->
3512 </div>
3516 <h5>Syntax:</h5>
3517 <pre>
3518 declare void %llvm.memset.i32(sbyte* <dest>, ubyte <val>,
3519 uint <len>, uint <align>)
3520 declare void %llvm.memset.i64(sbyte* <dest>, ubyte <val>,
3521 ulong <len>, uint <align>)
3522 </pre>
3524 <h5>Overview:</h5>
3526 <p>
3527 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
3528 byte value.
3529 </p>
3531 <p>
3532 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
3533 does not return a value, and takes an extra alignment argument.
3534 </p>
3536 <h5>Arguments:</h5>
3538 <p>
3539 The first argument is a pointer to the destination to fill, the second is the
3540 byte value to fill it with, the third argument is an integer
3541 argument specifying the number of bytes to fill, and the fourth argument is the
3542 known alignment of destination location.
3543 </p>
3545 <p>
3546 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3547 the caller guarantees that the destination pointer is aligned to that boundary.
3548 </p>
3550 <h5>Semantics:</h5>
3552 <p>
3554 the
3555 destination location. If the argument is known to be aligned to some boundary,
3556 this can be specified as the fourth argument, otherwise it should be set to 0 or
3557 1.
3558 </p>
3559 </div>
3562 <!-- _______________________________________________________________________ -->
3565 </div>
3569 <h5>Syntax:</h5>
3570 <pre>
3571 declare bool %llvm.isunordered.f32(float Val1, float Val2)
3572 declare bool %llvm.isunordered.f64(double Val1, double Val2)
3573 </pre>
3575 <h5>Overview:</h5>
3577 <p>
3578 The '<tt>llvm.isunordered</tt>' intrinsics return true if either or both of the
3579 specified floating point values is a NAN.
3580 </p>
3582 <h5>Arguments:</h5>
3584 <p>
3585 The arguments are floating point numbers of the same type.
3586 </p>
3588 <h5>Semantics:</h5>
3590 <p>
3591 If either or both of the arguments is a SNAN or QNAN, it returns true, otherwise
3592 false.
3593 </p>
3594 </div>
3597 <!-- _______________________________________________________________________ -->
3600 </div>
3604 <h5>Syntax:</h5>
3605 <pre>
3606 declare double %llvm.sqrt.f32(float Val)
3607 declare double %llvm.sqrt.f64(double Val)
3608 </pre>
3610 <h5>Overview:</h5>
3612 <p>
3613 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
3614 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
3615 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
3616 negative numbers (which allows for better optimization).
3617 </p>
3619 <h5>Arguments:</h5>
3621 <p>
3622 The argument and return value are floating point numbers of the same type.
3623 </p>
3625 <h5>Semantics:</h5>
3627 <p>
3628 This function returns the sqrt of the specified operand if it is a positive
3629 floating point number.
3630 </p>
3631 </div>
3633 <!-- ======================================================================= -->
3636 </div>
3639 <p>
3640 LLVM provides intrinsics for a few important bit manipulation operations.
3641 These allow efficient code generation for some algorithms.
3642 </p>
3644 </div>
3646 <!-- _______________________________________________________________________ -->
3649 </div>
3653 <h5>Syntax:</h5>
3654 <pre>
3655 declare ushort %llvm.bswap.i16(ushort <id>)
3656 declare uint %llvm.bswap.i32(uint <id>)
3657 declare ulong %llvm.bswap.i64(ulong <id>)
3658 </pre>
3660 <h5>Overview:</h5>
3662 <p>
3663 The '<tt>llvm.bwsap</tt>' family of intrinsics is used to byteswap a 16, 32 or
3664 64 bit quantity. These are useful for performing operations on data that is not
3665 in the target's native byte order.
3666 </p>
3668 <h5>Semantics:</h5>
3670 <p>
3671 The <tt>llvm.bswap.16</tt> intrinsic returns a ushort value that has the high and low
3672 byte of the input ushort swapped. Similarly, the <tt>llvm.bswap.i32</tt> intrinsic
3673 returns a uint value that has the four bytes of the input uint swapped, so that
3674 if the input bytes are numbered 0, 1, 2, 3 then the returned uint will have its
3675 bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i64</tt> intrinsic extends this concept
3676 to 64 bits.
3677 </p>
3679 </div>
3681 <!-- _______________________________________________________________________ -->
3684 </div>
3688 <h5>Syntax:</h5>
3689 <pre>
3690 declare ubyte %llvm.ctpop.i8 (ubyte <src>)
3691 declare ushort %llvm.ctpop.i16(ushort <src>)
3692 declare uint %llvm.ctpop.i32(uint <src>)
3693 declare ulong %llvm.ctpop.i64(ulong <src>)
3694 </pre>
3696 <h5>Overview:</h5>
3698 <p>
3699 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
3700 value.
3701 </p>
3703 <h5>Arguments:</h5>
3705 <p>
3706 The only argument is the value to be counted. The argument may be of any
3707 unsigned integer type. The return type must match the argument type.
3708 </p>
3710 <h5>Semantics:</h5>
3712 <p>
3713 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
3714 </p>
3715 </div>
3717 <!-- _______________________________________________________________________ -->
3720 </div>
3724 <h5>Syntax:</h5>
3725 <pre>
3726 declare ubyte %llvm.ctlz.i8 (ubyte <src>)
3727 declare ushort %llvm.ctlz.i16(ushort <src>)
3728 declare uint %llvm.ctlz.i32(uint <src>)
3729 declare ulong %llvm.ctlz.i64(ulong <src>)
3730 </pre>
3732 <h5>Overview:</h5>
3734 <p>
3735 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
3736 leading zeros in a variable.
3737 </p>
3739 <h5>Arguments:</h5>
3741 <p>
3742 The only argument is the value to be counted. The argument may be of any
3743 unsigned integer type. The return type must match the argument type.
3744 </p>
3746 <h5>Semantics:</h5>
3748 <p>
3749 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
3750 in a variable. If the src == 0 then the result is the size in bits of the type
3751 of src. For example, <tt>llvm.cttz(int 2) = 30</tt>.
3752 </p>
3753 </div>
3757 <!-- _______________________________________________________________________ -->
3760 </div>
3764 <h5>Syntax:</h5>
3765 <pre>
3766 declare ubyte %llvm.cttz.i8 (ubyte <src>)
3767 declare ushort %llvm.cttz.i16(ushort <src>)
3768 declare uint %llvm.cttz.i32(uint <src>)
3769 declare ulong %llvm.cttz.i64(ulong <src>)
3770 </pre>
3772 <h5>Overview:</h5>
3774 <p>
3775 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
3776 trailing zeros.
3777 </p>
3779 <h5>Arguments:</h5>
3781 <p>
3782 The only argument is the value to be counted. The argument may be of any
3783 unsigned integer type. The return type must match the argument type.
3784 </p>
3786 <h5>Semantics:</h5>
3788 <p>
3789 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
3790 in a variable. If the src == 0 then the result is the size in bits of the type
3791 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
3792 </p>
3793 </div>
3795 <!-- ======================================================================= -->
3798 </div>
3801 <p>
3802 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
3803 are described in the <a
3805 Debugging</a> document.
3806 </p>
3807 </div>
3810 <!-- *********************************************************************** -->
3811 <hr>
3812 <address>
3820 Last modified: $Date$
3821 </address>
3822 </body>
3823 </html>