1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <div class="doc_title"> LLVM Language Reference Manual </div>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a></li>
24 <li><a href="#callingconv">Calling Conventions</a></li>
25 <li><a href="#globalvars">Global Variables</a></li>
26 <li><a href="#functionstructure">Functions</a></li>
27 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
30 <li><a href="#typesystem">Type System</a>
32 <li><a href="#t_primitive">Primitive Types</a>
34 <li><a href="#t_classifications">Type Classifications</a></li>
37 <li><a href="#t_derived">Derived Types</a>
39 <li><a href="#t_array">Array Type</a></li>
40 <li><a href="#t_function">Function Type</a></li>
41 <li><a href="#t_pointer">Pointer Type</a></li>
42 <li><a href="#t_struct">Structure Type</a></li>
43 <li><a href="#t_packed">Packed Type</a></li>
44 <li><a href="#t_opaque">Opaque Type</a></li>
49 <li><a href="#constants">Constants</a>
51 <li><a href="#simpleconstants">Simple Constants</a>
52 <li><a href="#aggregateconstants">Aggregate Constants</a>
53 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
54 <li><a href="#undefvalues">Undefined Values</a>
55 <li><a href="#constantexprs">Constant Expressions</a>
58 <li><a href="#othervalues">Other Values</a>
60 <li><a href="#inlineasm">Inline Assembler Expressions</a>
63 <li><a href="#instref">Instruction Reference</a>
65 <li><a href="#terminators">Terminator Instructions</a>
67 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
68 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
69 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
70 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
71 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
72 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
75 <li><a href="#binaryops">Binary Operations</a>
77 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
78 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
79 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
80 <li><a href="#i_div">'<tt>div</tt>' Instruction</a></li>
81 <li><a href="#i_rem">'<tt>rem</tt>' Instruction</a></li>
82 <li><a href="#i_setcc">'<tt>set<i>cc</i></tt>' Instructions</a></li>
85 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
87 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
88 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
89 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
90 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
91 <li><a href="#i_shr">'<tt>shr</tt>' Instruction</a></li>
94 <li><a href="#vectorops">Vector Operations</a>
96 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
97 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
98 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
101 <li><a href="#memoryops">Memory Access Operations</a>
103 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
104 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
105 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
106 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
107 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
108 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
111 <li><a href="#otherops">Other Operations</a>
113 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
114 <li><a href="#i_cast">'<tt>cast .. to</tt>' Instruction</a></li>
115 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
116 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
117 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
122 <li><a href="#intrinsics">Intrinsic Functions</a>
124 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
126 <li><a href="#i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
127 <li><a href="#i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
128 <li><a href="#i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
131 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
133 <li><a href="#i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
134 <li><a href="#i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
135 <li><a href="#i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
138 <li><a href="#int_codegen">Code Generator Intrinsics</a>
140 <li><a href="#i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
141 <li><a href="#i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
142 <li><a href="#i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
143 <li><a href="#i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
144 <li><a href="#i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
145 <li><a href="#i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
146 <li><a href="#i_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
149 <li><a href="#int_libc">Standard C Library Intrinsics</a>
151 <li><a href="#i_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
152 <li><a href="#i_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
153 <li><a href="#i_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
154 <li><a href="#i_isunordered">'<tt>llvm.isunordered.*</tt>' Intrinsic</a></li>
155 <li><a href="#i_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
159 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
161 <li><a href="#i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
162 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
163 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
164 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
167 <li><a href="#int_debugger">Debugger intrinsics</a></li>
172 <div class="doc_author">
173 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
174 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
177 <!-- *********************************************************************** -->
178 <div class="doc_section"> <a name="abstract">Abstract </a></div>
179 <!-- *********************************************************************** -->
181 <div class="doc_text">
182 <p>This document is a reference manual for the LLVM assembly language.
183 LLVM is an SSA based representation that provides type safety,
184 low-level operations, flexibility, and the capability of representing
185 'all' high-level languages cleanly. It is the common code
186 representation used throughout all phases of the LLVM compilation
190 <!-- *********************************************************************** -->
191 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
192 <!-- *********************************************************************** -->
194 <div class="doc_text">
196 <p>The LLVM code representation is designed to be used in three
197 different forms: as an in-memory compiler IR, as an on-disk bytecode
198 representation (suitable for fast loading by a Just-In-Time compiler),
199 and as a human readable assembly language representation. This allows
200 LLVM to provide a powerful intermediate representation for efficient
201 compiler transformations and analysis, while providing a natural means
202 to debug and visualize the transformations. The three different forms
203 of LLVM are all equivalent. This document describes the human readable
204 representation and notation.</p>
206 <p>The LLVM representation aims to be light-weight and low-level
207 while being expressive, typed, and extensible at the same time. It
208 aims to be a "universal IR" of sorts, by being at a low enough level
209 that high-level ideas may be cleanly mapped to it (similar to how
210 microprocessors are "universal IR's", allowing many source languages to
211 be mapped to them). By providing type information, LLVM can be used as
212 the target of optimizations: for example, through pointer analysis, it
213 can be proven that a C automatic variable is never accessed outside of
214 the current function... allowing it to be promoted to a simple SSA
215 value instead of a memory location.</p>
219 <!-- _______________________________________________________________________ -->
220 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
222 <div class="doc_text">
224 <p>It is important to note that this document describes 'well formed'
225 LLVM assembly language. There is a difference between what the parser
226 accepts and what is considered 'well formed'. For example, the
227 following instruction is syntactically okay, but not well formed:</p>
230 %x = <a href="#i_add">add</a> int 1, %x
233 <p>...because the definition of <tt>%x</tt> does not dominate all of
234 its uses. The LLVM infrastructure provides a verification pass that may
235 be used to verify that an LLVM module is well formed. This pass is
236 automatically run by the parser after parsing input assembly and by
237 the optimizer before it outputs bytecode. The violations pointed out
238 by the verifier pass indicate bugs in transformation passes or input to
241 <!-- Describe the typesetting conventions here. --> </div>
243 <!-- *********************************************************************** -->
244 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
245 <!-- *********************************************************************** -->
247 <div class="doc_text">
249 <p>LLVM uses three different forms of identifiers, for different
253 <li>Named values are represented as a string of characters with a '%' prefix.
254 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
255 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
256 Identifiers which require other characters in their names can be surrounded
257 with quotes. In this way, anything except a <tt>"</tt> character can be used
260 <li>Unnamed values are represented as an unsigned numeric value with a '%'
261 prefix. For example, %12, %2, %44.</li>
263 <li>Constants, which are described in a <a href="#constants">section about
264 constants</a>, below.</li>
267 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
268 don't need to worry about name clashes with reserved words, and the set of
269 reserved words may be expanded in the future without penalty. Additionally,
270 unnamed identifiers allow a compiler to quickly come up with a temporary
271 variable without having to avoid symbol table conflicts.</p>
273 <p>Reserved words in LLVM are very similar to reserved words in other
274 languages. There are keywords for different opcodes ('<tt><a
275 href="#i_add">add</a></tt>', '<tt><a href="#i_cast">cast</a></tt>', '<tt><a
276 href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
277 href="#t_void">void</a></tt>', '<tt><a href="#t_uint">uint</a></tt>', etc...),
278 and others. These reserved words cannot conflict with variable names, because
279 none of them start with a '%' character.</p>
281 <p>Here is an example of LLVM code to multiply the integer variable
282 '<tt>%X</tt>' by 8:</p>
287 %result = <a href="#i_mul">mul</a> uint %X, 8
290 <p>After strength reduction:</p>
293 %result = <a href="#i_shl">shl</a> uint %X, ubyte 3
296 <p>And the hard way:</p>
299 <a href="#i_add">add</a> uint %X, %X <i>; yields {uint}:%0</i>
300 <a href="#i_add">add</a> uint %0, %0 <i>; yields {uint}:%1</i>
301 %result = <a href="#i_add">add</a> uint %1, %1
304 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
305 important lexical features of LLVM:</p>
309 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
312 <li>Unnamed temporaries are created when the result of a computation is not
313 assigned to a named value.</li>
315 <li>Unnamed temporaries are numbered sequentially</li>
319 <p>...and it also shows a convention that we follow in this document. When
320 demonstrating instructions, we will follow an instruction with a comment that
321 defines the type and name of value produced. Comments are shown in italic
326 <!-- *********************************************************************** -->
327 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
328 <!-- *********************************************************************** -->
330 <!-- ======================================================================= -->
331 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
334 <div class="doc_text">
336 <p>LLVM programs are composed of "Module"s, each of which is a
337 translation unit of the input programs. Each module consists of
338 functions, global variables, and symbol table entries. Modules may be
339 combined together with the LLVM linker, which merges function (and
340 global variable) definitions, resolves forward declarations, and merges
341 symbol table entries. Here is an example of the "hello world" module:</p>
343 <pre><i>; Declare the string constant as a global constant...</i>
344 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a
345 href="#globalvars">constant</a> <a href="#t_array">[13 x sbyte]</a> c"hello world\0A\00" <i>; [13 x sbyte]*</i>
347 <i>; External declaration of the puts function</i>
348 <a href="#functionstructure">declare</a> int %puts(sbyte*) <i>; int(sbyte*)* </i>
350 <i>; Definition of main function</i>
351 int %main() { <i>; int()* </i>
352 <i>; Convert [13x sbyte]* to sbyte *...</i>
354 href="#i_getelementptr">getelementptr</a> [13 x sbyte]* %.LC0, long 0, long 0 <i>; sbyte*</i>
356 <i>; Call puts function to write out the string to stdout...</i>
358 href="#i_call">call</a> int %puts(sbyte* %cast210) <i>; int</i>
360 href="#i_ret">ret</a> int 0<br>}<br></pre>
362 <p>This example is made up of a <a href="#globalvars">global variable</a>
363 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
364 function, and a <a href="#functionstructure">function definition</a>
365 for "<tt>main</tt>".</p>
367 <p>In general, a module is made up of a list of global values,
368 where both functions and global variables are global values. Global values are
369 represented by a pointer to a memory location (in this case, a pointer to an
370 array of char, and a pointer to a function), and have one of the following <a
371 href="#linkage">linkage types</a>.</p>
375 <!-- ======================================================================= -->
376 <div class="doc_subsection">
377 <a name="linkage">Linkage Types</a>
380 <div class="doc_text">
383 All Global Variables and Functions have one of the following types of linkage:
388 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
390 <dd>Global values with internal linkage are only directly accessible by
391 objects in the current module. In particular, linking code into a module with
392 an internal global value may cause the internal to be renamed as necessary to
393 avoid collisions. Because the symbol is internal to the module, all
394 references can be updated. This corresponds to the notion of the
395 '<tt>static</tt>' keyword in C, or the idea of "anonymous namespaces" in C++.
398 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
400 <dd>"<tt>linkonce</tt>" linkage is similar to <tt>internal</tt> linkage, with
401 the twist that linking together two modules defining the same
402 <tt>linkonce</tt> globals will cause one of the globals to be discarded. This
403 is typically used to implement inline functions. Unreferenced
404 <tt>linkonce</tt> globals are allowed to be discarded.
407 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
409 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
410 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
411 used to implement constructs in C such as "<tt>int X;</tt>" at global scope.
414 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
416 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
417 pointer to array type. When two global variables with appending linkage are
418 linked together, the two global arrays are appended together. This is the
419 LLVM, typesafe, equivalent of having the system linker append together
420 "sections" with identical names when .o files are linked.
423 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
425 <dd>If none of the above identifiers are used, the global is externally
426 visible, meaning that it participates in linkage and can be used to resolve
427 external symbol references.
431 <p><a name="linkage_external">For example, since the "<tt>.LC0</tt>"
432 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
433 variable and was linked with this one, one of the two would be renamed,
434 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
435 external (i.e., lacking any linkage declarations), they are accessible
436 outside of the current module. It is illegal for a function <i>declaration</i>
437 to have any linkage type other than "externally visible".</a></p>
441 <!-- ======================================================================= -->
442 <div class="doc_subsection">
443 <a name="callingconv">Calling Conventions</a>
446 <div class="doc_text">
448 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
449 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
450 specified for the call. The calling convention of any pair of dynamic
451 caller/callee must match, or the behavior of the program is undefined. The
452 following calling conventions are supported by LLVM, and more may be added in
456 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
458 <dd>This calling convention (the default if no other calling convention is
459 specified) matches the target C calling conventions. This calling convention
460 supports varargs function calls and tolerates some mismatch in the declared
461 prototype and implemented declaration of the function (as does normal C).
464 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
466 <dd>This calling convention attempts to make calls as fast as possible
467 (e.g. by passing things in registers). This calling convention allows the
468 target to use whatever tricks it wants to produce fast code for the target,
469 without having to conform to an externally specified ABI. Implementations of
470 this convention should allow arbitrary tail call optimization to be supported.
471 This calling convention does not support varargs and requires the prototype of
472 all callees to exactly match the prototype of the function definition.
475 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
477 <dd>This calling convention attempts to make code in the caller as efficient
478 as possible under the assumption that the call is not commonly executed. As
479 such, these calls often preserve all registers so that the call does not break
480 any live ranges in the caller side. This calling convention does not support
481 varargs and requires the prototype of all callees to exactly match the
482 prototype of the function definition.
485 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
487 <dd>Any calling convention may be specified by number, allowing
488 target-specific calling conventions to be used. Target specific calling
489 conventions start at 64.
493 <p>More calling conventions can be added/defined on an as-needed basis, to
494 support pascal conventions or any other well-known target-independent
499 <!-- ======================================================================= -->
500 <div class="doc_subsection">
501 <a name="globalvars">Global Variables</a>
504 <div class="doc_text">
506 <p>Global variables define regions of memory allocated at compilation time
507 instead of run-time. Global variables may optionally be initialized, may have
508 an explicit section to be placed in, and may
509 have an optional explicit alignment specified. A
510 variable may be defined as a global "constant," which indicates that the
511 contents of the variable will <b>never</b> be modified (enabling better
512 optimization, allowing the global data to be placed in the read-only section of
513 an executable, etc). Note that variables that need runtime initialization
514 cannot be marked "constant" as there is a store to the variable.</p>
517 LLVM explicitly allows <em>declarations</em> of global variables to be marked
518 constant, even if the final definition of the global is not. This capability
519 can be used to enable slightly better optimization of the program, but requires
520 the language definition to guarantee that optimizations based on the
521 'constantness' are valid for the translation units that do not include the
525 <p>As SSA values, global variables define pointer values that are in
526 scope (i.e. they dominate) all basic blocks in the program. Global
527 variables always define a pointer to their "content" type because they
528 describe a region of memory, and all memory objects in LLVM are
529 accessed through pointers.</p>
531 <p>LLVM allows an explicit section to be specified for globals. If the target
532 supports it, it will emit globals to the section specified.</p>
534 <p>An explicit alignment may be specified for a global. If not present, or if
535 the alignment is set to zero, the alignment of the global is set by the target
536 to whatever it feels convenient. If an explicit alignment is specified, the
537 global is forced to have at least that much alignment. All alignments must be
543 <!-- ======================================================================= -->
544 <div class="doc_subsection">
545 <a name="functionstructure">Functions</a>
548 <div class="doc_text">
550 <p>LLVM function definitions consist of an optional <a href="#linkage">linkage
551 type</a>, an optional <a href="#callingconv">calling convention</a>, a return
552 type, a function name, a (possibly empty) argument list, an optional section,
553 an optional alignment, an opening curly brace,
554 a list of basic blocks, and a closing curly brace. LLVM function declarations
555 are defined with the "<tt>declare</tt>" keyword, an optional <a
556 href="#callingconv">calling convention</a>, a return type, a function name,
557 a possibly empty list of arguments, and an optional alignment.</p>
559 <p>A function definition contains a list of basic blocks, forming the CFG for
560 the function. Each basic block may optionally start with a label (giving the
561 basic block a symbol table entry), contains a list of instructions, and ends
562 with a <a href="#terminators">terminator</a> instruction (such as a branch or
563 function return).</p>
565 <p>The first basic block in a program is special in two ways: it is immediately
566 executed on entrance to the function, and it is not allowed to have predecessor
567 basic blocks (i.e. there can not be any branches to the entry block of a
568 function). Because the block can have no predecessors, it also cannot have any
569 <a href="#i_phi">PHI nodes</a>.</p>
571 <p>LLVM functions are identified by their name and type signature. Hence, two
572 functions with the same name but different parameter lists or return values are
573 considered different functions, and LLVM will resolve references to each
576 <p>LLVM allows an explicit section to be specified for functions. If the target
577 supports it, it will emit functions to the section specified.</p>
579 <p>An explicit alignment may be specified for a function. If not present, or if
580 the alignment is set to zero, the alignment of the function is set by the target
581 to whatever it feels convenient. If an explicit alignment is specified, the
582 function is forced to have at least that much alignment. All alignments must be
587 <!-- ======================================================================= -->
588 <div class="doc_subsection">
589 <a name="moduleasm">Module-Level Inline Assembly</a>
592 <div class="doc_text">
594 Modules may contain "module-level inline asm" blocks, which corresponds to the
595 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
596 LLVM and treated as a single unit, but may be separated in the .ll file if
597 desired. The syntax is very simple:
600 <div class="doc_code"><pre>
601 module asm "inline asm code goes here"
602 module asm "more can go here"
605 <p>The strings can contain any character by escaping non-printable characters.
606 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
611 The inline asm code is simply printed to the machine code .s file when
612 assembly code is generated.
617 <!-- *********************************************************************** -->
618 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
619 <!-- *********************************************************************** -->
621 <div class="doc_text">
623 <p>The LLVM type system is one of the most important features of the
624 intermediate representation. Being typed enables a number of
625 optimizations to be performed on the IR directly, without having to do
626 extra analyses on the side before the transformation. A strong type
627 system makes it easier to read the generated code and enables novel
628 analyses and transformations that are not feasible to perform on normal
629 three address code representations.</p>
633 <!-- ======================================================================= -->
634 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
635 <div class="doc_text">
636 <p>The primitive types are the fundamental building blocks of the LLVM
637 system. The current set of primitive types is as follows:</p>
639 <table class="layout">
644 <tr><th>Type</th><th>Description</th></tr>
645 <tr><td><tt>void</tt></td><td>No value</td></tr>
646 <tr><td><tt>ubyte</tt></td><td>Unsigned 8-bit value</td></tr>
647 <tr><td><tt>ushort</tt></td><td>Unsigned 16-bit value</td></tr>
648 <tr><td><tt>uint</tt></td><td>Unsigned 32-bit value</td></tr>
649 <tr><td><tt>ulong</tt></td><td>Unsigned 64-bit value</td></tr>
650 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
651 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
658 <tr><th>Type</th><th>Description</th></tr>
659 <tr><td><tt>bool</tt></td><td>True or False value</td></tr>
660 <tr><td><tt>sbyte</tt></td><td>Signed 8-bit value</td></tr>
661 <tr><td><tt>short</tt></td><td>Signed 16-bit value</td></tr>
662 <tr><td><tt>int</tt></td><td>Signed 32-bit value</td></tr>
663 <tr><td><tt>long</tt></td><td>Signed 64-bit value</td></tr>
664 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
672 <!-- _______________________________________________________________________ -->
673 <div class="doc_subsubsection"> <a name="t_classifications">Type
674 Classifications</a> </div>
675 <div class="doc_text">
676 <p>These different primitive types fall into a few useful
679 <table border="1" cellspacing="0" cellpadding="4">
681 <tr><th>Classification</th><th>Types</th></tr>
683 <td><a name="t_signed">signed</a></td>
684 <td><tt>sbyte, short, int, long, float, double</tt></td>
687 <td><a name="t_unsigned">unsigned</a></td>
688 <td><tt>ubyte, ushort, uint, ulong</tt></td>
691 <td><a name="t_integer">integer</a></td>
692 <td><tt>ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
695 <td><a name="t_integral">integral</a></td>
696 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long</tt>
700 <td><a name="t_floating">floating point</a></td>
701 <td><tt>float, double</tt></td>
704 <td><a name="t_firstclass">first class</a></td>
705 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long,<br>
706 float, double, <a href="#t_pointer">pointer</a>,
707 <a href="#t_packed">packed</a></tt></td>
712 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
713 most important. Values of these types are the only ones which can be
714 produced by instructions, passed as arguments, or used as operands to
715 instructions. This means that all structures and arrays must be
716 manipulated either by pointer or by component.</p>
719 <!-- ======================================================================= -->
720 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
722 <div class="doc_text">
724 <p>The real power in LLVM comes from the derived types in the system.
725 This is what allows a programmer to represent arrays, functions,
726 pointers, and other useful types. Note that these derived types may be
727 recursive: For example, it is possible to have a two dimensional array.</p>
731 <!-- _______________________________________________________________________ -->
732 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
734 <div class="doc_text">
738 <p>The array type is a very simple derived type that arranges elements
739 sequentially in memory. The array type requires a size (number of
740 elements) and an underlying data type.</p>
745 [<# elements> x <elementtype>]
748 <p>The number of elements is a constant integer value; elementtype may
749 be any type with a size.</p>
752 <table class="layout">
755 <tt>[40 x int ]</tt><br/>
756 <tt>[41 x int ]</tt><br/>
757 <tt>[40 x uint]</tt><br/>
760 Array of 40 integer values.<br/>
761 Array of 41 integer values.<br/>
762 Array of 40 unsigned integer values.<br/>
766 <p>Here are some examples of multidimensional arrays:</p>
767 <table class="layout">
770 <tt>[3 x [4 x int]]</tt><br/>
771 <tt>[12 x [10 x float]]</tt><br/>
772 <tt>[2 x [3 x [4 x uint]]]</tt><br/>
775 3x4 array of integer values.<br/>
776 12x10 array of single precision floating point values.<br/>
777 2x3x4 array of unsigned integer values.<br/>
782 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
783 length array. Normally, accesses past the end of an array are undefined in
784 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
785 As a special case, however, zero length arrays are recognized to be variable
786 length. This allows implementation of 'pascal style arrays' with the LLVM
787 type "{ int, [0 x float]}", for example.</p>
791 <!-- _______________________________________________________________________ -->
792 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
793 <div class="doc_text">
795 <p>The function type can be thought of as a function signature. It
796 consists of a return type and a list of formal parameter types.
797 Function types are usually used to build virtual function tables
798 (which are structures of pointers to functions), for indirect function
799 calls, and when defining a function.</p>
801 The return type of a function type cannot be an aggregate type.
804 <pre> <returntype> (<parameter list>)<br></pre>
805 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
806 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
807 which indicates that the function takes a variable number of arguments.
808 Variable argument functions can access their arguments with the <a
809 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
811 <table class="layout">
814 <tt>int (int)</tt> <br/>
815 <tt>float (int, int *) *</tt><br/>
816 <tt>int (sbyte *, ...)</tt><br/>
819 function taking an <tt>int</tt>, returning an <tt>int</tt><br/>
820 <a href="#t_pointer">Pointer</a> to a function that takes an
821 <tt>int</tt> and a <a href="#t_pointer">pointer</a> to <tt>int</tt>,
822 returning <tt>float</tt>.<br/>
823 A vararg function that takes at least one <a href="#t_pointer">pointer</a>
824 to <tt>sbyte</tt> (signed char in C), which returns an integer. This is
825 the signature for <tt>printf</tt> in LLVM.<br/>
831 <!-- _______________________________________________________________________ -->
832 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
833 <div class="doc_text">
835 <p>The structure type is used to represent a collection of data members
836 together in memory. The packing of the field types is defined to match
837 the ABI of the underlying processor. The elements of a structure may
838 be any type that has a size.</p>
839 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
840 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
841 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
844 <pre> { <type list> }<br></pre>
846 <table class="layout">
849 <tt>{ int, int, int }</tt><br/>
850 <tt>{ float, int (int) * }</tt><br/>
853 a triple of three <tt>int</tt> values<br/>
854 A pair, where the first element is a <tt>float</tt> and the second element
855 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
856 that takes an <tt>int</tt>, returning an <tt>int</tt>.<br/>
862 <!-- _______________________________________________________________________ -->
863 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
864 <div class="doc_text">
866 <p>As in many languages, the pointer type represents a pointer or
867 reference to another object, which must live in memory.</p>
869 <pre> <type> *<br></pre>
871 <table class="layout">
874 <tt>[4x int]*</tt><br/>
875 <tt>int (int *) *</tt><br/>
878 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
879 four <tt>int</tt> values<br/>
880 A <a href="#t_pointer">pointer</a> to a <a
881 href="#t_function">function</a> that takes an <tt>int*</tt>, returning an
888 <!-- _______________________________________________________________________ -->
889 <div class="doc_subsubsection"> <a name="t_packed">Packed Type</a> </div>
890 <div class="doc_text">
894 <p>A packed type is a simple derived type that represents a vector
895 of elements. Packed types are used when multiple primitive data
896 are operated in parallel using a single instruction (SIMD).
897 A packed type requires a size (number of
898 elements) and an underlying primitive data type. Vectors must have a power
899 of two length (1, 2, 4, 8, 16 ...). Packed types are
900 considered <a href="#t_firstclass">first class</a>.</p>
905 < <# elements> x <elementtype> >
908 <p>The number of elements is a constant integer value; elementtype may
909 be any integral or floating point type.</p>
913 <table class="layout">
916 <tt><4 x int></tt><br/>
917 <tt><8 x float></tt><br/>
918 <tt><2 x uint></tt><br/>
921 Packed vector of 4 integer values.<br/>
922 Packed vector of 8 floating-point values.<br/>
923 Packed vector of 2 unsigned integer values.<br/>
929 <!-- _______________________________________________________________________ -->
930 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
931 <div class="doc_text">
935 <p>Opaque types are used to represent unknown types in the system. This
936 corresponds (for example) to the C notion of a foward declared structure type.
937 In LLVM, opaque types can eventually be resolved to any type (not just a
948 <table class="layout">
961 <!-- *********************************************************************** -->
962 <div class="doc_section"> <a name="constants">Constants</a> </div>
963 <!-- *********************************************************************** -->
965 <div class="doc_text">
967 <p>LLVM has several different basic types of constants. This section describes
968 them all and their syntax.</p>
972 <!-- ======================================================================= -->
973 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
975 <div class="doc_text">
978 <dt><b>Boolean constants</b></dt>
980 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
981 constants of the <tt><a href="#t_primitive">bool</a></tt> type.
984 <dt><b>Integer constants</b></dt>
986 <dd>Standard integers (such as '4') are constants of the <a
987 href="#t_integer">integer</a> type. Negative numbers may be used with signed
991 <dt><b>Floating point constants</b></dt>
993 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
994 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
995 notation (see below). Floating point constants must have a <a
996 href="#t_floating">floating point</a> type. </dd>
998 <dt><b>Null pointer constants</b></dt>
1000 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1001 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1005 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1006 of floating point constants. For example, the form '<tt>double
1007 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1008 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1009 (and the only time that they are generated by the disassembler) is when a
1010 floating point constant must be emitted but it cannot be represented as a
1011 decimal floating point number. For example, NaN's, infinities, and other
1012 special values are represented in their IEEE hexadecimal format so that
1013 assembly and disassembly do not cause any bits to change in the constants.</p>
1017 <!-- ======================================================================= -->
1018 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1021 <div class="doc_text">
1022 <p>Aggregate constants arise from aggregation of simple constants
1023 and smaller aggregate constants.</p>
1026 <dt><b>Structure constants</b></dt>
1028 <dd>Structure constants are represented with notation similar to structure
1029 type definitions (a comma separated list of elements, surrounded by braces
1030 (<tt>{}</tt>)). For example: "<tt>{ int 4, float 17.0, int* %G }</tt>",
1031 where "<tt>%G</tt>" is declared as "<tt>%G = external global int</tt>". Structure constants
1032 must have <a href="#t_struct">structure type</a>, and the number and
1033 types of elements must match those specified by the type.
1036 <dt><b>Array constants</b></dt>
1038 <dd>Array constants are represented with notation similar to array type
1039 definitions (a comma separated list of elements, surrounded by square brackets
1040 (<tt>[]</tt>)). For example: "<tt>[ int 42, int 11, int 74 ]</tt>". Array
1041 constants must have <a href="#t_array">array type</a>, and the number and
1042 types of elements must match those specified by the type.
1045 <dt><b>Packed constants</b></dt>
1047 <dd>Packed constants are represented with notation similar to packed type
1048 definitions (a comma separated list of elements, surrounded by
1049 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< int 42,
1050 int 11, int 74, int 100 ></tt>". Packed constants must have <a
1051 href="#t_packed">packed type</a>, and the number and types of elements must
1052 match those specified by the type.
1055 <dt><b>Zero initialization</b></dt>
1057 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1058 value to zero of <em>any</em> type, including scalar and aggregate types.
1059 This is often used to avoid having to print large zero initializers (e.g. for
1060 large arrays) and is always exactly equivalent to using explicit zero
1067 <!-- ======================================================================= -->
1068 <div class="doc_subsection">
1069 <a name="globalconstants">Global Variable and Function Addresses</a>
1072 <div class="doc_text">
1074 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1075 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1076 constants. These constants are explicitly referenced when the <a
1077 href="#identifiers">identifier for the global</a> is used and always have <a
1078 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1084 %Z = global [2 x int*] [ int* %X, int* %Y ]
1089 <!-- ======================================================================= -->
1090 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1091 <div class="doc_text">
1092 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1093 no specific value. Undefined values may be of any type and be used anywhere
1094 a constant is permitted.</p>
1096 <p>Undefined values indicate to the compiler that the program is well defined
1097 no matter what value is used, giving the compiler more freedom to optimize.
1101 <!-- ======================================================================= -->
1102 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1105 <div class="doc_text">
1107 <p>Constant expressions are used to allow expressions involving other constants
1108 to be used as constants. Constant expressions may be of any <a
1109 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1110 that does not have side effects (e.g. load and call are not supported). The
1111 following is the syntax for constant expressions:</p>
1114 <dt><b><tt>cast ( CST to TYPE )</tt></b></dt>
1116 <dd>Cast a constant to another type.</dd>
1118 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1120 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1121 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1122 instruction, the index list may have zero or more indexes, which are required
1123 to make sense for the type of "CSTPTR".</dd>
1125 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1127 <dd>Perform the <a href="#i_select">select operation</a> on
1130 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1132 <dd>Perform the <a href="#i_extractelement">extractelement
1133 operation</a> on constants.
1135 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1137 <dd>Perform the <a href="#i_insertelement">insertelement
1138 operation</a> on constants.
1141 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1143 <dd>Perform the <a href="#i_shufflevector">shufflevector
1144 operation</a> on constants.
1146 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1148 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1149 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1150 binary</a> operations. The constraints on operands are the same as those for
1151 the corresponding instruction (e.g. no bitwise operations on floating point
1152 values are allowed).</dd>
1156 <!-- *********************************************************************** -->
1157 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1158 <!-- *********************************************************************** -->
1160 <!-- ======================================================================= -->
1161 <div class="doc_subsection">
1162 <a name="inlineasm">Inline Assembler Expressions</a>
1165 <div class="doc_text">
1168 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1169 Module-Level Inline Assembly</a>) through the use of a special value. This
1170 value represents the inline assembler as a string (containing the instructions
1171 to emit), a list of operand constraints (stored as a string), and a flag that
1172 indicates whether or not the inline asm expression has side effects. An example
1173 inline assembler expression is:
1177 int(int) asm "bswap $0", "=r,r"
1181 Inline assembler expressions may <b>only</b> be used as the callee operand of
1182 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1186 %X = call int asm "<a href="#i_bswap">bswap</a> $0", "=r,r"(int %Y)
1190 Inline asms with side effects not visible in the constraint list must be marked
1191 as having side effects. This is done through the use of the
1192 '<tt>sideeffect</tt>' keyword, like so:
1196 call void asm sideeffect "eieio", ""()
1199 <p>TODO: The format of the asm and constraints string still need to be
1200 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1201 need to be documented).
1206 <!-- *********************************************************************** -->
1207 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1208 <!-- *********************************************************************** -->
1210 <div class="doc_text">
1212 <p>The LLVM instruction set consists of several different
1213 classifications of instructions: <a href="#terminators">terminator
1214 instructions</a>, <a href="#binaryops">binary instructions</a>,
1215 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1216 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1217 instructions</a>.</p>
1221 <!-- ======================================================================= -->
1222 <div class="doc_subsection"> <a name="terminators">Terminator
1223 Instructions</a> </div>
1225 <div class="doc_text">
1227 <p>As mentioned <a href="#functionstructure">previously</a>, every
1228 basic block in a program ends with a "Terminator" instruction, which
1229 indicates which block should be executed after the current block is
1230 finished. These terminator instructions typically yield a '<tt>void</tt>'
1231 value: they produce control flow, not values (the one exception being
1232 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1233 <p>There are six different terminator instructions: the '<a
1234 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1235 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1236 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1237 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1238 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1242 <!-- _______________________________________________________________________ -->
1243 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1244 Instruction</a> </div>
1245 <div class="doc_text">
1247 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1248 ret void <i>; Return from void function</i>
1251 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1252 value) from a function back to the caller.</p>
1253 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1254 returns a value and then causes control flow, and one that just causes
1255 control flow to occur.</p>
1257 <p>The '<tt>ret</tt>' instruction may return any '<a
1258 href="#t_firstclass">first class</a>' type. Notice that a function is
1259 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1260 instruction inside of the function that returns a value that does not
1261 match the return type of the function.</p>
1263 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1264 returns back to the calling function's context. If the caller is a "<a
1265 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1266 the instruction after the call. If the caller was an "<a
1267 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1268 at the beginning of the "normal" destination block. If the instruction
1269 returns a value, that value shall set the call or invoke instruction's
1272 <pre> ret int 5 <i>; Return an integer value of 5</i>
1273 ret void <i>; Return from a void function</i>
1276 <!-- _______________________________________________________________________ -->
1277 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1278 <div class="doc_text">
1280 <pre> br bool <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1283 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1284 transfer to a different basic block in the current function. There are
1285 two forms of this instruction, corresponding to a conditional branch
1286 and an unconditional branch.</p>
1288 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1289 single '<tt>bool</tt>' value and two '<tt>label</tt>' values. The
1290 unconditional form of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>'
1291 value as a target.</p>
1293 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>bool</tt>'
1294 argument is evaluated. If the value is <tt>true</tt>, control flows
1295 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1296 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1298 <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
1299 href="#i_ret">ret</a> int 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> int 0<br></pre>
1301 <!-- _______________________________________________________________________ -->
1302 <div class="doc_subsubsection">
1303 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1306 <div class="doc_text">
1310 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1315 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1316 several different places. It is a generalization of the '<tt>br</tt>'
1317 instruction, allowing a branch to occur to one of many possible
1323 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1324 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1325 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1326 table is not allowed to contain duplicate constant entries.</p>
1330 <p>The <tt>switch</tt> instruction specifies a table of values and
1331 destinations. When the '<tt>switch</tt>' instruction is executed, this
1332 table is searched for the given value. If the value is found, control flow is
1333 transfered to the corresponding destination; otherwise, control flow is
1334 transfered to the default destination.</p>
1336 <h5>Implementation:</h5>
1338 <p>Depending on properties of the target machine and the particular
1339 <tt>switch</tt> instruction, this instruction may be code generated in different
1340 ways. For example, it could be generated as a series of chained conditional
1341 branches or with a lookup table.</p>
1346 <i>; Emulate a conditional br instruction</i>
1347 %Val = <a href="#i_cast">cast</a> bool %value to int
1348 switch int %Val, label %truedest [int 0, label %falsedest ]
1350 <i>; Emulate an unconditional br instruction</i>
1351 switch uint 0, label %dest [ ]
1353 <i>; Implement a jump table:</i>
1354 switch uint %val, label %otherwise [ uint 0, label %onzero
1355 uint 1, label %onone
1356 uint 2, label %ontwo ]
1360 <!-- _______________________________________________________________________ -->
1361 <div class="doc_subsubsection">
1362 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1365 <div class="doc_text">
1370 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1371 to label <normal label> except label <exception label>
1376 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1377 function, with the possibility of control flow transfer to either the
1378 '<tt>normal</tt>' label or the
1379 '<tt>exception</tt>' label. If the callee function returns with the
1380 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1381 "normal" label. If the callee (or any indirect callees) returns with the "<a
1382 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1383 continued at the dynamically nearest "exception" label.</p>
1387 <p>This instruction requires several arguments:</p>
1391 The optional "cconv" marker indicates which <a href="callingconv">calling
1392 convention</a> the call should use. If none is specified, the call defaults
1393 to using C calling conventions.
1395 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1396 function value being invoked. In most cases, this is a direct function
1397 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1398 an arbitrary pointer to function value.
1401 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1402 function to be invoked. </li>
1404 <li>'<tt>function args</tt>': argument list whose types match the function
1405 signature argument types. If the function signature indicates the function
1406 accepts a variable number of arguments, the extra arguments can be
1409 <li>'<tt>normal label</tt>': the label reached when the called function
1410 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1412 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1413 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1419 <p>This instruction is designed to operate as a standard '<tt><a
1420 href="#i_call">call</a></tt>' instruction in most regards. The primary
1421 difference is that it establishes an association with a label, which is used by
1422 the runtime library to unwind the stack.</p>
1424 <p>This instruction is used in languages with destructors to ensure that proper
1425 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1426 exception. Additionally, this is important for implementation of
1427 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1431 %retval = invoke int %Test(int 15) to label %Continue
1432 except label %TestCleanup <i>; {int}:retval set</i>
1433 %retval = invoke <a href="#callingconv">coldcc</a> int %Test(int 15) to label %Continue
1434 except label %TestCleanup <i>; {int}:retval set</i>
1439 <!-- _______________________________________________________________________ -->
1441 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1442 Instruction</a> </div>
1444 <div class="doc_text">
1453 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1454 at the first callee in the dynamic call stack which used an <a
1455 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1456 primarily used to implement exception handling.</p>
1460 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1461 immediately halt. The dynamic call stack is then searched for the first <a
1462 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1463 execution continues at the "exceptional" destination block specified by the
1464 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1465 dynamic call chain, undefined behavior results.</p>
1468 <!-- _______________________________________________________________________ -->
1470 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1471 Instruction</a> </div>
1473 <div class="doc_text">
1482 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1483 instruction is used to inform the optimizer that a particular portion of the
1484 code is not reachable. This can be used to indicate that the code after a
1485 no-return function cannot be reached, and other facts.</p>
1489 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1494 <!-- ======================================================================= -->
1495 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1496 <div class="doc_text">
1497 <p>Binary operators are used to do most of the computation in a
1498 program. They require two operands, execute an operation on them, and
1499 produce a single value. The operands might represent
1500 multiple data, as is the case with the <a href="#t_packed">packed</a> data type.
1501 The result value of a binary operator is not
1502 necessarily the same type as its operands.</p>
1503 <p>There are several different binary operators:</p>
1505 <!-- _______________________________________________________________________ -->
1506 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1507 Instruction</a> </div>
1508 <div class="doc_text">
1510 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1513 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1515 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1516 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1517 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1518 Both arguments must have identical types.</p>
1520 <p>The value produced is the integer or floating point sum of the two
1523 <pre> <result> = add int 4, %var <i>; yields {int}:result = 4 + %var</i>
1526 <!-- _______________________________________________________________________ -->
1527 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1528 Instruction</a> </div>
1529 <div class="doc_text">
1531 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1534 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1536 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1537 instruction present in most other intermediate representations.</p>
1539 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1540 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1542 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1543 Both arguments must have identical types.</p>
1545 <p>The value produced is the integer or floating point difference of
1546 the two operands.</p>
1548 <pre> <result> = sub int 4, %var <i>; yields {int}:result = 4 - %var</i>
1549 <result> = sub int 0, %val <i>; yields {int}:result = -%var</i>
1552 <!-- _______________________________________________________________________ -->
1553 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1554 Instruction</a> </div>
1555 <div class="doc_text">
1557 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1560 <p>The '<tt>mul</tt>' instruction returns the product of its two
1563 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1564 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1566 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1567 Both arguments must have identical types.</p>
1569 <p>The value produced is the integer or floating point product of the
1571 <p>There is no signed vs unsigned multiplication. The appropriate
1572 action is taken based on the type of the operand.</p>
1574 <pre> <result> = mul int 4, %var <i>; yields {int}:result = 4 * %var</i>
1577 <!-- _______________________________________________________________________ -->
1578 <div class="doc_subsubsection"> <a name="i_div">'<tt>div</tt>'
1579 Instruction</a> </div>
1580 <div class="doc_text">
1582 <pre> <result> = div <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1585 <p>The '<tt>div</tt>' instruction returns the quotient of its two
1588 <p>The two arguments to the '<tt>div</tt>' instruction must be either <a
1589 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1591 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1592 Both arguments must have identical types.</p>
1594 <p>The value produced is the integer or floating point quotient of the
1597 <pre> <result> = div int 4, %var <i>; yields {int}:result = 4 / %var</i>
1600 <!-- _______________________________________________________________________ -->
1601 <div class="doc_subsubsection"> <a name="i_rem">'<tt>rem</tt>'
1602 Instruction</a> </div>
1603 <div class="doc_text">
1605 <pre> <result> = rem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1608 <p>The '<tt>rem</tt>' instruction returns the remainder from the
1609 division of its two operands.</p>
1611 <p>The two arguments to the '<tt>rem</tt>' instruction must be either <a
1612 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1614 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1615 Both arguments must have identical types.</p>
1617 <p>This returns the <i>remainder</i> of a division (where the result
1618 has the same sign as the divisor), not the <i>modulus</i> (where the
1619 result has the same sign as the dividend) of a value. For more
1620 information about the difference, see <a
1621 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
1624 <pre> <result> = rem int 4, %var <i>; yields {int}:result = 4 % %var</i>
1628 <!-- _______________________________________________________________________ -->
1629 <div class="doc_subsubsection"> <a name="i_setcc">'<tt>set<i>cc</i></tt>'
1630 Instructions</a> </div>
1631 <div class="doc_text">
1633 <pre> <result> = seteq <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1634 <result> = setne <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1635 <result> = setlt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1636 <result> = setgt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1637 <result> = setle <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1638 <result> = setge <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1641 <p>The '<tt>set<i>cc</i></tt>' family of instructions returns a boolean
1642 value based on a comparison of their two operands.</p>
1644 <p>The two arguments to the '<tt>set<i>cc</i></tt>' instructions must
1645 be of <a href="#t_firstclass">first class</a> type (it is not possible
1646 to compare '<tt>label</tt>'s, '<tt>array</tt>'s, '<tt>structure</tt>'
1647 or '<tt>void</tt>' values, etc...). Both arguments must have identical
1650 <p>The '<tt>seteq</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1651 value if both operands are equal.<br>
1652 The '<tt>setne</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1653 value if both operands are unequal.<br>
1654 The '<tt>setlt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1655 value if the first operand is less than the second operand.<br>
1656 The '<tt>setgt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1657 value if the first operand is greater than the second operand.<br>
1658 The '<tt>setle</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1659 value if the first operand is less than or equal to the second operand.<br>
1660 The '<tt>setge</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1661 value if the first operand is greater than or equal to the second
1664 <pre> <result> = seteq int 4, 5 <i>; yields {bool}:result = false</i>
1665 <result> = setne float 4, 5 <i>; yields {bool}:result = true</i>
1666 <result> = setlt uint 4, 5 <i>; yields {bool}:result = true</i>
1667 <result> = setgt sbyte 4, 5 <i>; yields {bool}:result = false</i>
1668 <result> = setle sbyte 4, 5 <i>; yields {bool}:result = true</i>
1669 <result> = setge sbyte 4, 5 <i>; yields {bool}:result = false</i>
1673 <!-- ======================================================================= -->
1674 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
1675 Operations</a> </div>
1676 <div class="doc_text">
1677 <p>Bitwise binary operators are used to do various forms of
1678 bit-twiddling in a program. They are generally very efficient
1679 instructions and can commonly be strength reduced from other
1680 instructions. They require two operands, execute an operation on them,
1681 and produce a single value. The resulting value of the bitwise binary
1682 operators is always the same type as its first operand.</p>
1684 <!-- _______________________________________________________________________ -->
1685 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
1686 Instruction</a> </div>
1687 <div class="doc_text">
1689 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1692 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
1693 its two operands.</p>
1695 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
1696 href="#t_integral">integral</a> values. Both arguments must have
1697 identical types.</p>
1699 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
1701 <div style="align: center">
1702 <table border="1" cellspacing="0" cellpadding="4">
1733 <pre> <result> = and int 4, %var <i>; yields {int}:result = 4 & %var</i>
1734 <result> = and int 15, 40 <i>; yields {int}:result = 8</i>
1735 <result> = and int 4, 8 <i>; yields {int}:result = 0</i>
1738 <!-- _______________________________________________________________________ -->
1739 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
1740 <div class="doc_text">
1742 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1745 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
1746 or of its two operands.</p>
1748 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
1749 href="#t_integral">integral</a> values. Both arguments must have
1750 identical types.</p>
1752 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
1754 <div style="align: center">
1755 <table border="1" cellspacing="0" cellpadding="4">
1786 <pre> <result> = or int 4, %var <i>; yields {int}:result = 4 | %var</i>
1787 <result> = or int 15, 40 <i>; yields {int}:result = 47</i>
1788 <result> = or int 4, 8 <i>; yields {int}:result = 12</i>
1791 <!-- _______________________________________________________________________ -->
1792 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
1793 Instruction</a> </div>
1794 <div class="doc_text">
1796 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1799 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
1800 or of its two operands. The <tt>xor</tt> is used to implement the
1801 "one's complement" operation, which is the "~" operator in C.</p>
1803 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
1804 href="#t_integral">integral</a> values. Both arguments must have
1805 identical types.</p>
1807 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
1809 <div style="align: center">
1810 <table border="1" cellspacing="0" cellpadding="4">
1842 <pre> <result> = xor int 4, %var <i>; yields {int}:result = 4 ^ %var</i>
1843 <result> = xor int 15, 40 <i>; yields {int}:result = 39</i>
1844 <result> = xor int 4, 8 <i>; yields {int}:result = 12</i>
1845 <result> = xor int %V, -1 <i>; yields {int}:result = ~%V</i>
1848 <!-- _______________________________________________________________________ -->
1849 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
1850 Instruction</a> </div>
1851 <div class="doc_text">
1853 <pre> <result> = shl <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1856 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
1857 the left a specified number of bits.</p>
1859 <p>The first argument to the '<tt>shl</tt>' instruction must be an <a
1860 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1863 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
1865 <pre> <result> = shl int 4, ubyte %var <i>; yields {int}:result = 4 << %var</i>
1866 <result> = shl int 4, ubyte 2 <i>; yields {int}:result = 16</i>
1867 <result> = shl int 1, ubyte 10 <i>; yields {int}:result = 1024</i>
1870 <!-- _______________________________________________________________________ -->
1871 <div class="doc_subsubsection"> <a name="i_shr">'<tt>shr</tt>'
1872 Instruction</a> </div>
1873 <div class="doc_text">
1875 <pre> <result> = shr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1878 <p>The '<tt>shr</tt>' instruction returns the first operand shifted to
1879 the right a specified number of bits.</p>
1881 <p>The first argument to the '<tt>shr</tt>' instruction must be an <a
1882 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1885 <p>If the first argument is a <a href="#t_signed">signed</a> type, the
1886 most significant bit is duplicated in the newly free'd bit positions.
1887 If the first argument is unsigned, zero bits shall fill the empty
1890 <pre> <result> = shr int 4, ubyte %var <i>; yields {int}:result = 4 >> %var</i>
1891 <result> = shr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
1892 <result> = shr int 4, ubyte 2 <i>; yields {int}:result = 1</i>
1893 <result> = shr sbyte 4, ubyte 3 <i>; yields {sbyte}:result = 0</i>
1894 <result> = shr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = -1</i>
1898 <!-- ======================================================================= -->
1899 <div class="doc_subsection">
1900 <a name="vectorops">Vector Operations</a>
1903 <div class="doc_text">
1905 <p>LLVM supports several instructions to represent vector operations in a
1906 target-independent manner. This instructions cover the element-access and
1907 vector-specific operations needed to process vectors effectively. While LLVM
1908 does directly support these vector operations, many sophisticated algorithms
1909 will want to use target-specific intrinsics to take full advantage of a specific
1914 <!-- _______________________________________________________________________ -->
1915 <div class="doc_subsubsection">
1916 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
1919 <div class="doc_text">
1924 <result> = extractelement <n x <ty>> <val>, uint <idx> <i>; yields <ty></i>
1930 The '<tt>extractelement</tt>' instruction extracts a single scalar
1931 element from a packed vector at a specified index.
1938 The first operand of an '<tt>extractelement</tt>' instruction is a
1939 value of <a href="#t_packed">packed</a> type. The second operand is
1940 an index indicating the position from which to extract the element.
1941 The index may be a variable.</p>
1946 The result is a scalar of the same type as the element type of
1947 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
1948 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
1949 results are undefined.
1955 %result = extractelement <4 x int> %vec, uint 0 <i>; yields int</i>
1960 <!-- _______________________________________________________________________ -->
1961 <div class="doc_subsubsection">
1962 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
1965 <div class="doc_text">
1970 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, uint <idx> <i>; yields <n x <ty>></i>
1976 The '<tt>insertelement</tt>' instruction inserts a scalar
1977 element into a packed vector at a specified index.
1984 The first operand of an '<tt>insertelement</tt>' instruction is a
1985 value of <a href="#t_packed">packed</a> type. The second operand is a
1986 scalar value whose type must equal the element type of the first
1987 operand. The third operand is an index indicating the position at
1988 which to insert the value. The index may be a variable.</p>
1993 The result is a packed vector of the same type as <tt>val</tt>. Its
1994 element values are those of <tt>val</tt> except at position
1995 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
1996 exceeds the length of <tt>val</tt>, the results are undefined.
2002 %result = insertelement <4 x int> %vec, int 1, uint 0 <i>; yields <4 x int></i>
2006 <!-- _______________________________________________________________________ -->
2007 <div class="doc_subsubsection">
2008 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2011 <div class="doc_text">
2016 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x uint> <mask> <i>; yields <n x <ty>></i>
2022 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2023 from two input vectors, returning a vector of the same type.
2029 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2030 with types that match each other and types that match the result of the
2031 instruction. The third argument is a shuffle mask, which has the same number
2032 of elements as the other vector type, but whose element type is always 'uint'.
2036 The shuffle mask operand is required to be a constant vector with either
2037 constant integer or undef values.
2043 The elements of the two input vectors are numbered from left to right across
2044 both of the vectors. The shuffle mask operand specifies, for each element of
2045 the result vector, which element of the two input registers the result element
2046 gets. The element selector may be undef (meaning "don't care") and the second
2047 operand may be undef if performing a shuffle from only one vector.
2053 %result = shufflevector <4 x int> %v1, <4 x int> %v2,
2054 <4 x uint> <uint 0, uint 4, uint 1, uint 5> <i>; yields <4 x int></i>
2055 %result = shufflevector <4 x int> %v1, <4 x int> undef,
2056 <4 x uint> <uint 0, uint 1, uint 2, uint 3> <i>; yields <4 x int></i> - Identity shuffle.
2060 <!-- ======================================================================= -->
2061 <div class="doc_subsection">
2062 <a name="memoryops">Memory Access Operations</a>
2065 <div class="doc_text">
2067 <p>A key design point of an SSA-based representation is how it
2068 represents memory. In LLVM, no memory locations are in SSA form, which
2069 makes things very simple. This section describes how to read, write,
2070 allocate, and free memory in LLVM.</p>
2074 <!-- _______________________________________________________________________ -->
2075 <div class="doc_subsubsection">
2076 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2079 <div class="doc_text">
2084 <result> = malloc <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2089 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2090 heap and returns a pointer to it.</p>
2094 <p>The '<tt>malloc</tt>' instruction allocates
2095 <tt>sizeof(<type>)*NumElements</tt>
2096 bytes of memory from the operating system and returns a pointer of the
2097 appropriate type to the program. If "NumElements" is specified, it is the
2098 number of elements allocated. If an alignment is specified, the value result
2099 of the allocation is guaranteed to be aligned to at least that boundary. If
2100 not specified, or if zero, the target can choose to align the allocation on any
2101 convenient boundary.</p>
2103 <p>'<tt>type</tt>' must be a sized type.</p>
2107 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2108 a pointer is returned.</p>
2113 %array = malloc [4 x ubyte ] <i>; yields {[%4 x ubyte]*}:array</i>
2115 %size = <a href="#i_add">add</a> uint 2, 2 <i>; yields {uint}:size = uint 4</i>
2116 %array1 = malloc ubyte, uint 4 <i>; yields {ubyte*}:array1</i>
2117 %array2 = malloc [12 x ubyte], uint %size <i>; yields {[12 x ubyte]*}:array2</i>
2118 %array3 = malloc int, uint 4, align 1024 <i>; yields {int*}:array3</i>
2119 %array4 = malloc int, align 1024 <i>; yields {int*}:array4</i>
2123 <!-- _______________________________________________________________________ -->
2124 <div class="doc_subsubsection">
2125 <a name="i_free">'<tt>free</tt>' Instruction</a>
2128 <div class="doc_text">
2133 free <type> <value> <i>; yields {void}</i>
2138 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2139 memory heap to be reallocated in the future.</p>
2143 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2144 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2149 <p>Access to the memory pointed to by the pointer is no longer defined
2150 after this instruction executes.</p>
2155 %array = <a href="#i_malloc">malloc</a> [4 x ubyte] <i>; yields {[4 x ubyte]*}:array</i>
2156 free [4 x ubyte]* %array
2160 <!-- _______________________________________________________________________ -->
2161 <div class="doc_subsubsection">
2162 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2165 <div class="doc_text">
2170 <result> = alloca <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2175 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
2176 stack frame of the procedure that is live until the current function
2177 returns to its caller.</p>
2181 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2182 bytes of memory on the runtime stack, returning a pointer of the
2183 appropriate type to the program. If "NumElements" is specified, it is the
2184 number of elements allocated. If an alignment is specified, the value result
2185 of the allocation is guaranteed to be aligned to at least that boundary. If
2186 not specified, or if zero, the target can choose to align the allocation on any
2187 convenient boundary.</p>
2189 <p>'<tt>type</tt>' may be any sized type.</p>
2193 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2194 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2195 instruction is commonly used to represent automatic variables that must
2196 have an address available. When the function returns (either with the <tt><a
2197 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2198 instructions), the memory is reclaimed.</p>
2203 %ptr = alloca int <i>; yields {int*}:ptr</i>
2204 %ptr = alloca int, uint 4 <i>; yields {int*}:ptr</i>
2205 %ptr = alloca int, uint 4, align 1024 <i>; yields {int*}:ptr</i>
2206 %ptr = alloca int, align 1024 <i>; yields {int*}:ptr</i>
2210 <!-- _______________________________________________________________________ -->
2211 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2212 Instruction</a> </div>
2213 <div class="doc_text">
2215 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
2217 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2219 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2220 address from which to load. The pointer must point to a <a
2221 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2222 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2223 the number or order of execution of this <tt>load</tt> with other
2224 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2227 <p>The location of memory pointed to is loaded.</p>
2229 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
2231 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2232 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2235 <!-- _______________________________________________________________________ -->
2236 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2237 Instruction</a> </div>
2239 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2240 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2243 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2245 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2246 to store and an address in which to store it. The type of the '<tt><pointer></tt>'
2247 operand must be a pointer to the type of the '<tt><value></tt>'
2248 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2249 optimizer is not allowed to modify the number or order of execution of
2250 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2251 href="#i_store">store</a></tt> instructions.</p>
2253 <p>The contents of memory are updated to contain '<tt><value></tt>'
2254 at the location specified by the '<tt><pointer></tt>' operand.</p>
2256 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
2258 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2259 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2261 <!-- _______________________________________________________________________ -->
2262 <div class="doc_subsubsection">
2263 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2266 <div class="doc_text">
2269 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2275 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2276 subelement of an aggregate data structure.</p>
2280 <p>This instruction takes a list of integer constants that indicate what
2281 elements of the aggregate object to index to. The actual types of the arguments
2282 provided depend on the type of the first pointer argument. The
2283 '<tt>getelementptr</tt>' instruction is used to index down through the type
2284 levels of a structure or to a specific index in an array. When indexing into a
2285 structure, only <tt>uint</tt>
2286 integer constants are allowed. When indexing into an array or pointer,
2287 <tt>int</tt> and <tt>long</tt> indexes are allowed of any sign.</p>
2289 <p>For example, let's consider a C code fragment and how it gets
2290 compiled to LLVM:</p>
2304 int *foo(struct ST *s) {
2305 return &s[1].Z.B[5][13];
2309 <p>The LLVM code generated by the GCC frontend is:</p>
2312 %RT = type { sbyte, [10 x [20 x int]], sbyte }
2313 %ST = type { int, double, %RT }
2317 int* %foo(%ST* %s) {
2319 %reg = getelementptr %ST* %s, int 1, uint 2, uint 1, int 5, int 13
2326 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2327 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2328 and <a href="#t_array">array</a> types require <tt>uint</tt>, <tt>int</tt>,
2329 <tt>ulong</tt>, or <tt>long</tt> values, and <a href="#t_struct">structure</a>
2330 types require <tt>uint</tt> <b>constants</b>.</p>
2332 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2333 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ int, double, %RT
2334 }</tt>' type, a structure. The second index indexes into the third element of
2335 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ sbyte, [10 x [20 x int]],
2336 sbyte }</tt>' type, another structure. The third index indexes into the second
2337 element of the structure, yielding a '<tt>[10 x [20 x int]]</tt>' type, an
2338 array. The two dimensions of the array are subscripted into, yielding an
2339 '<tt>int</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2340 to this element, thus computing a value of '<tt>int*</tt>' type.</p>
2342 <p>Note that it is perfectly legal to index partially through a
2343 structure, returning a pointer to an inner element. Because of this,
2344 the LLVM code for the given testcase is equivalent to:</p>
2347 int* %foo(%ST* %s) {
2348 %t1 = getelementptr %ST* %s, int 1 <i>; yields %ST*:%t1</i>
2349 %t2 = getelementptr %ST* %t1, int 0, uint 2 <i>; yields %RT*:%t2</i>
2350 %t3 = getelementptr %RT* %t2, int 0, uint 1 <i>; yields [10 x [20 x int]]*:%t3</i>
2351 %t4 = getelementptr [10 x [20 x int]]* %t3, int 0, int 5 <i>; yields [20 x int]*:%t4</i>
2352 %t5 = getelementptr [20 x int]* %t4, int 0, int 13 <i>; yields int*:%t5</i>
2357 <p>Note that it is undefined to access an array out of bounds: array and
2358 pointer indexes must always be within the defined bounds of the array type.
2359 The one exception for this rules is zero length arrays. These arrays are
2360 defined to be accessible as variable length arrays, which requires access
2361 beyond the zero'th element.</p>
2366 <i>; yields [12 x ubyte]*:aptr</i>
2367 %aptr = getelementptr {int, [12 x ubyte]}* %sptr, long 0, uint 1
2371 <!-- ======================================================================= -->
2372 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
2373 <div class="doc_text">
2374 <p>The instructions in this category are the "miscellaneous"
2375 instructions, which defy better classification.</p>
2377 <!-- _______________________________________________________________________ -->
2378 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
2379 Instruction</a> </div>
2380 <div class="doc_text">
2382 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
2384 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
2385 the SSA graph representing the function.</p>
2387 <p>The type of the incoming values are specified with the first type
2388 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
2389 as arguments, with one pair for each predecessor basic block of the
2390 current block. Only values of <a href="#t_firstclass">first class</a>
2391 type may be used as the value arguments to the PHI node. Only labels
2392 may be used as the label arguments.</p>
2393 <p>There must be no non-phi instructions between the start of a basic
2394 block and the PHI instructions: i.e. PHI instructions must be first in
2397 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
2398 value specified by the parameter, depending on which basic block we
2399 came from in the last <a href="#terminators">terminator</a> instruction.</p>
2401 <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>
2404 <!-- _______________________________________________________________________ -->
2405 <div class="doc_subsubsection">
2406 <a name="i_cast">'<tt>cast .. to</tt>' Instruction</a>
2409 <div class="doc_text">
2414 <result> = cast <ty> <value> to <ty2> <i>; yields ty2</i>
2420 The '<tt>cast</tt>' instruction is used as the primitive means to convert
2421 integers to floating point, change data type sizes, and break type safety (by
2429 The '<tt>cast</tt>' instruction takes a value to cast, which must be a first
2430 class value, and a type to cast it to, which must also be a <a
2431 href="#t_firstclass">first class</a> type.
2437 This instruction follows the C rules for explicit casts when determining how the
2438 data being cast must change to fit in its new container.
2442 When casting to bool, any value that would be considered true in the context of
2443 a C '<tt>if</tt>' condition is converted to the boolean '<tt>true</tt>' values,
2444 all else are '<tt>false</tt>'.
2448 When extending an integral value from a type of one signness to another (for
2449 example '<tt>sbyte</tt>' to '<tt>ulong</tt>'), the value is sign-extended if the
2450 <b>source</b> value is signed, and zero-extended if the source value is
2451 unsigned. <tt>bool</tt> values are always zero extended into either zero or
2458 %X = cast int 257 to ubyte <i>; yields ubyte:1</i>
2459 %Y = cast int 123 to bool <i>; yields bool:true</i>
2463 <!-- _______________________________________________________________________ -->
2464 <div class="doc_subsubsection">
2465 <a name="i_select">'<tt>select</tt>' Instruction</a>
2468 <div class="doc_text">
2473 <result> = select bool <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
2479 The '<tt>select</tt>' instruction is used to choose one value based on a
2480 condition, without branching.
2487 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.
2493 If the boolean condition evaluates to true, the instruction returns the first
2494 value argument; otherwise, it returns the second value argument.
2500 %X = select bool true, ubyte 17, ubyte 42 <i>; yields ubyte:17</i>
2505 <!-- _______________________________________________________________________ -->
2506 <div class="doc_subsubsection">
2507 <a name="i_call">'<tt>call</tt>' Instruction</a>
2510 <div class="doc_text">
2514 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
2519 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
2523 <p>This instruction requires several arguments:</p>
2527 <p>The optional "tail" marker indicates whether the callee function accesses
2528 any allocas or varargs in the caller. If the "tail" marker is present, the
2529 function call is eligible for tail call optimization. Note that calls may
2530 be marked "tail" even if they do not occur before a <a
2531 href="#i_ret"><tt>ret</tt></a> instruction.
2534 <p>The optional "cconv" marker indicates which <a href="callingconv">calling
2535 convention</a> the call should use. If none is specified, the call defaults
2536 to using C calling conventions.
2539 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
2540 being invoked. The argument types must match the types implied by this
2541 signature. This type can be omitted if the function is not varargs and
2542 if the function type does not return a pointer to a function.</p>
2545 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
2546 be invoked. In most cases, this is a direct function invocation, but
2547 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
2548 to function value.</p>
2551 <p>'<tt>function args</tt>': argument list whose types match the
2552 function signature argument types. All arguments must be of
2553 <a href="#t_firstclass">first class</a> type. If the function signature
2554 indicates the function accepts a variable number of arguments, the extra
2555 arguments can be specified.</p>
2561 <p>The '<tt>call</tt>' instruction is used to cause control flow to
2562 transfer to a specified function, with its incoming arguments bound to
2563 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
2564 instruction in the called function, control flow continues with the
2565 instruction after the function call, and the return value of the
2566 function is bound to the result argument. This is a simpler case of
2567 the <a href="#i_invoke">invoke</a> instruction.</p>
2572 %retval = call int %test(int %argc)
2573 call int(sbyte*, ...) *%printf(sbyte* %msg, int 12, sbyte 42);
2574 %X = tail call int %foo()
2575 %Y = tail call <a href="#callingconv">fastcc</a> int %foo()
2580 <!-- _______________________________________________________________________ -->
2581 <div class="doc_subsubsection">
2582 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
2585 <div class="doc_text">
2590 <resultval> = va_arg <va_list*> <arglist>, <argty>
2595 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
2596 the "variable argument" area of a function call. It is used to implement the
2597 <tt>va_arg</tt> macro in C.</p>
2601 <p>This instruction takes a <tt>va_list*</tt> value and the type of
2602 the argument. It returns a value of the specified argument type and
2603 increments the <tt>va_list</tt> to point to the next argument. Again, the
2604 actual type of <tt>va_list</tt> is target specific.</p>
2608 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
2609 type from the specified <tt>va_list</tt> and causes the
2610 <tt>va_list</tt> to point to the next argument. For more information,
2611 see the variable argument handling <a href="#int_varargs">Intrinsic
2614 <p>It is legal for this instruction to be called in a function which does not
2615 take a variable number of arguments, for example, the <tt>vfprintf</tt>
2618 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
2619 href="#intrinsics">intrinsic function</a> because it takes a type as an
2624 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
2628 <!-- *********************************************************************** -->
2629 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
2630 <!-- *********************************************************************** -->
2632 <div class="doc_text">
2634 <p>LLVM supports the notion of an "intrinsic function". These functions have
2635 well known names and semantics and are required to follow certain
2636 restrictions. Overall, these instructions represent an extension mechanism for
2637 the LLVM language that does not require changing all of the transformations in
2638 LLVM to add to the language (or the bytecode reader/writer, the parser,
2641 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
2642 prefix is reserved in LLVM for intrinsic names; thus, functions may not be named
2643 this. Intrinsic functions must always be external functions: you cannot define
2644 the body of intrinsic functions. Intrinsic functions may only be used in call
2645 or invoke instructions: it is illegal to take the address of an intrinsic
2646 function. Additionally, because intrinsic functions are part of the LLVM
2647 language, it is required that they all be documented here if any are added.</p>
2650 <p>To learn how to add an intrinsic function, please see the <a
2651 href="ExtendingLLVM.html">Extending LLVM Guide</a>.
2656 <!-- ======================================================================= -->
2657 <div class="doc_subsection">
2658 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
2661 <div class="doc_text">
2663 <p>Variable argument support is defined in LLVM with the <a
2664 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
2665 intrinsic functions. These functions are related to the similarly
2666 named macros defined in the <tt><stdarg.h></tt> header file.</p>
2668 <p>All of these functions operate on arguments that use a
2669 target-specific value type "<tt>va_list</tt>". The LLVM assembly
2670 language reference manual does not define what this type is, so all
2671 transformations should be prepared to handle intrinsics with any type
2674 <p>This example shows how the <a href="#i_vanext"><tt>vanext</tt></a>
2675 instruction and the variable argument handling intrinsic functions are
2679 int %test(int %X, ...) {
2680 ; Initialize variable argument processing
2682 call void %<a href="#i_va_start">llvm.va_start</a>(sbyte** %ap)
2684 ; Read a single integer argument
2685 %tmp = va_arg sbyte** %ap, int
2687 ; Demonstrate usage of llvm.va_copy and llvm.va_end
2689 call void %<a href="#i_va_copy">llvm.va_copy</a>(sbyte** %aq, sbyte** %ap)
2690 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %aq)
2692 ; Stop processing of arguments.
2693 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %ap)
2699 <!-- _______________________________________________________________________ -->
2700 <div class="doc_subsubsection">
2701 <a name="i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
2705 <div class="doc_text">
2707 <pre> declare void %llvm.va_start(<va_list>* <arglist>)<br></pre>
2709 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
2710 <tt>*<arglist></tt> for subsequent use by <tt><a
2711 href="#i_va_arg">va_arg</a></tt>.</p>
2715 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
2719 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
2720 macro available in C. In a target-dependent way, it initializes the
2721 <tt>va_list</tt> element the argument points to, so that the next call to
2722 <tt>va_arg</tt> will produce the first variable argument passed to the function.
2723 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
2724 last argument of the function, the compiler can figure that out.</p>
2728 <!-- _______________________________________________________________________ -->
2729 <div class="doc_subsubsection">
2730 <a name="i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
2733 <div class="doc_text">
2735 <pre> declare void %llvm.va_end(<va_list*> <arglist>)<br></pre>
2737 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
2738 which has been initialized previously with <tt><a href="#i_va_start">llvm.va_start</a></tt>
2739 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
2741 <p>The argument is a <tt>va_list</tt> to destroy.</p>
2743 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
2744 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
2745 Calls to <a href="#i_va_start"><tt>llvm.va_start</tt></a> and <a
2746 href="#i_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly
2747 with calls to <tt>llvm.va_end</tt>.</p>
2750 <!-- _______________________________________________________________________ -->
2751 <div class="doc_subsubsection">
2752 <a name="i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
2755 <div class="doc_text">
2760 declare void %llvm.va_copy(<va_list>* <destarglist>,
2761 <va_list>* <srcarglist>)
2766 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position from
2767 the source argument list to the destination argument list.</p>
2771 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
2772 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
2777 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> macro
2778 available in C. In a target-dependent way, it copies the source
2779 <tt>va_list</tt> element into the destination list. This intrinsic is necessary
2780 because the <tt><a href="i_va_begin">llvm.va_begin</a></tt> intrinsic may be
2781 arbitrarily complex and require memory allocation, for example.</p>
2785 <!-- ======================================================================= -->
2786 <div class="doc_subsection">
2787 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
2790 <div class="doc_text">
2793 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
2794 Collection</a> requires the implementation and generation of these intrinsics.
2795 These intrinsics allow identification of <a href="#i_gcroot">GC roots on the
2796 stack</a>, as well as garbage collector implementations that require <a
2797 href="#i_gcread">read</a> and <a href="#i_gcwrite">write</a> barriers.
2798 Front-ends for type-safe garbage collected languages should generate these
2799 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
2800 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
2804 <!-- _______________________________________________________________________ -->
2805 <div class="doc_subsubsection">
2806 <a name="i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
2809 <div class="doc_text">
2814 declare void %llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
2819 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
2820 the code generator, and allows some metadata to be associated with it.</p>
2824 <p>The first argument specifies the address of a stack object that contains the
2825 root pointer. The second pointer (which must be either a constant or a global
2826 value address) contains the meta-data to be associated with the root.</p>
2830 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
2831 location. At compile-time, the code generator generates information to allow
2832 the runtime to find the pointer at GC safe points.
2838 <!-- _______________________________________________________________________ -->
2839 <div class="doc_subsubsection">
2840 <a name="i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
2843 <div class="doc_text">
2848 declare sbyte* %llvm.gcread(sbyte* %ObjPtr, sbyte** %Ptr)
2853 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
2854 locations, allowing garbage collector implementations that require read
2859 <p>The second argument is the address to read from, which should be an address
2860 allocated from the garbage collector. The first object is a pointer to the
2861 start of the referenced object, if needed by the language runtime (otherwise
2866 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
2867 instruction, but may be replaced with substantially more complex code by the
2868 garbage collector runtime, as needed.</p>
2873 <!-- _______________________________________________________________________ -->
2874 <div class="doc_subsubsection">
2875 <a name="i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
2878 <div class="doc_text">
2883 declare void %llvm.gcwrite(sbyte* %P1, sbyte* %Obj, sbyte** %P2)
2888 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
2889 locations, allowing garbage collector implementations that require write
2890 barriers (such as generational or reference counting collectors).</p>
2894 <p>The first argument is the reference to store, the second is the start of the
2895 object to store it to, and the third is the address of the field of Obj to
2896 store to. If the runtime does not require a pointer to the object, Obj may be
2901 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
2902 instruction, but may be replaced with substantially more complex code by the
2903 garbage collector runtime, as needed.</p>
2909 <!-- ======================================================================= -->
2910 <div class="doc_subsection">
2911 <a name="int_codegen">Code Generator Intrinsics</a>
2914 <div class="doc_text">
2916 These intrinsics are provided by LLVM to expose special features that may only
2917 be implemented with code generator support.
2922 <!-- _______________________________________________________________________ -->
2923 <div class="doc_subsubsection">
2924 <a name="i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
2927 <div class="doc_text">
2931 declare sbyte *%llvm.returnaddress(uint <level>)
2937 The '<tt>llvm.returnaddress</tt>' intrinsic returns a target-specific value
2938 indicating the return address of the current function or one of its callers.
2944 The argument to this intrinsic indicates which function to return the address
2945 for. Zero indicates the calling function, one indicates its caller, etc. The
2946 argument is <b>required</b> to be a constant integer value.
2952 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
2953 the return address of the specified call frame, or zero if it cannot be
2954 identified. The value returned by this intrinsic is likely to be incorrect or 0
2955 for arguments other than zero, so it should only be used for debugging purposes.
2959 Note that calling this intrinsic does not prevent function inlining or other
2960 aggressive transformations, so the value returned may not be that of the obvious
2961 source-language caller.
2966 <!-- _______________________________________________________________________ -->
2967 <div class="doc_subsubsection">
2968 <a name="i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
2971 <div class="doc_text">
2975 declare sbyte *%llvm.frameaddress(uint <level>)
2981 The '<tt>llvm.frameaddress</tt>' intrinsic returns the target-specific frame
2982 pointer value for the specified stack frame.
2988 The argument to this intrinsic indicates which function to return the frame
2989 pointer for. Zero indicates the calling function, one indicates its caller,
2990 etc. The argument is <b>required</b> to be a constant integer value.
2996 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
2997 the frame address of the specified call frame, or zero if it cannot be
2998 identified. The value returned by this intrinsic is likely to be incorrect or 0
2999 for arguments other than zero, so it should only be used for debugging purposes.
3003 Note that calling this intrinsic does not prevent function inlining or other
3004 aggressive transformations, so the value returned may not be that of the obvious
3005 source-language caller.
3009 <!-- _______________________________________________________________________ -->
3010 <div class="doc_subsubsection">
3011 <a name="i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
3014 <div class="doc_text">
3018 declare sbyte *%llvm.stacksave()
3024 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
3025 the function stack, for use with <a href="#i_stackrestore">
3026 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
3027 features like scoped automatic variable sized arrays in C99.
3033 This intrinsic returns a opaque pointer value that can be passed to <a
3034 href="#i_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
3035 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
3036 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
3037 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
3038 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
3039 that were allocated after the <tt>llvm.stacksave</tt> was executed.
3044 <!-- _______________________________________________________________________ -->
3045 <div class="doc_subsubsection">
3046 <a name="i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
3049 <div class="doc_text">
3053 declare void %llvm.stackrestore(sbyte* %ptr)
3059 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
3060 the function stack to the state it was in when the corresponding <a
3061 href="#llvm.stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
3062 useful for implementing language features like scoped automatic variable sized
3069 See the description for <a href="#i_stacksave"><tt>llvm.stacksave</tt></a>.
3075 <!-- _______________________________________________________________________ -->
3076 <div class="doc_subsubsection">
3077 <a name="i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
3080 <div class="doc_text">
3084 declare void %llvm.prefetch(sbyte * <address>,
3085 uint <rw>, uint <locality>)
3092 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
3093 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
3095 effect on the behavior of the program but can change its performance
3102 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
3103 determining if the fetch should be for a read (0) or write (1), and
3104 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
3105 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
3106 <tt>locality</tt> arguments must be constant integers.
3112 This intrinsic does not modify the behavior of the program. In particular,
3113 prefetches cannot trap and do not produce a value. On targets that support this
3114 intrinsic, the prefetch can provide hints to the processor cache for better
3120 <!-- _______________________________________________________________________ -->
3121 <div class="doc_subsubsection">
3122 <a name="i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
3125 <div class="doc_text">
3129 declare void %llvm.pcmarker( uint <id> )
3136 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
3138 code to simulators and other tools. The method is target specific, but it is
3139 expected that the marker will use exported symbols to transmit the PC of the marker.
3140 The marker makes no guarantees that it will remain with any specific instruction
3141 after optimizations. It is possible that the presence of a marker will inhibit
3142 optimizations. The intended use is to be inserted after optimizations to allow
3143 correlations of simulation runs.
3149 <tt>id</tt> is a numerical id identifying the marker.
3155 This intrinsic does not modify the behavior of the program. Backends that do not
3156 support this intrinisic may ignore it.
3161 <!-- _______________________________________________________________________ -->
3162 <div class="doc_subsubsection">
3163 <a name="i_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
3166 <div class="doc_text">
3170 declare ulong %llvm.readcyclecounter( )
3177 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
3178 counter register (or similar low latency, high accuracy clocks) on those targets
3179 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
3180 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
3181 should only be used for small timings.
3187 When directly supported, reading the cycle counter should not modify any memory.
3188 Implementations are allowed to either return a application specific value or a
3189 system wide value. On backends without support, this is lowered to a constant 0.
3194 <!-- ======================================================================= -->
3195 <div class="doc_subsection">
3196 <a name="int_libc">Standard C Library Intrinsics</a>
3199 <div class="doc_text">
3201 LLVM provides intrinsics for a few important standard C library functions.
3202 These intrinsics allow source-language front-ends to pass information about the
3203 alignment of the pointer arguments to the code generator, providing opportunity
3204 for more efficient code generation.
3209 <!-- _______________________________________________________________________ -->
3210 <div class="doc_subsubsection">
3211 <a name="i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
3214 <div class="doc_text">
3218 declare void %llvm.memcpy.i32(sbyte* <dest>, sbyte* <src>,
3219 uint <len>, uint <align>)
3220 declare void %llvm.memcpy.i64(sbyte* <dest>, sbyte* <src>,
3221 ulong <len>, uint <align>)
3227 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
3228 location to the destination location.
3232 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
3233 intrinsics do not return a value, and takes an extra alignment argument.
3239 The first argument is a pointer to the destination, the second is a pointer to
3240 the source. The third argument is an integer argument
3241 specifying the number of bytes to copy, and the fourth argument is the alignment
3242 of the source and destination locations.
3246 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3247 the caller guarantees that both the source and destination pointers are aligned
3254 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
3255 location to the destination location, which are not allowed to overlap. It
3256 copies "len" bytes of memory over. If the argument is known to be aligned to
3257 some boundary, this can be specified as the fourth argument, otherwise it should
3263 <!-- _______________________________________________________________________ -->
3264 <div class="doc_subsubsection">
3265 <a name="i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
3268 <div class="doc_text">
3272 declare void %llvm.memmove.i32(sbyte* <dest>, sbyte* <src>,
3273 uint <len>, uint <align>)
3274 declare void %llvm.memmove.i64(sbyte* <dest>, sbyte* <src>,
3275 ulong <len>, uint <align>)
3281 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
3282 location to the destination location. It is similar to the
3283 '<tt>llvm.memcmp</tt>' intrinsic but allows the two memory locations to overlap.
3287 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
3288 intrinsics do not return a value, and takes an extra alignment argument.
3294 The first argument is a pointer to the destination, the second is a pointer to
3295 the source. The third argument is an integer argument
3296 specifying the number of bytes to copy, and the fourth argument is the alignment
3297 of the source and destination locations.
3301 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3302 the caller guarantees that the source and destination pointers are aligned to
3309 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
3310 location to the destination location, which may overlap. It
3311 copies "len" bytes of memory over. If the argument is known to be aligned to
3312 some boundary, this can be specified as the fourth argument, otherwise it should
3318 <!-- _______________________________________________________________________ -->
3319 <div class="doc_subsubsection">
3320 <a name="i_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
3323 <div class="doc_text">
3327 declare void %llvm.memset.i32(sbyte* <dest>, ubyte <val>,
3328 uint <len>, uint <align>)
3329 declare void %llvm.memset.i64(sbyte* <dest>, ubyte <val>,
3330 ulong <len>, uint <align>)
3336 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
3341 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
3342 does not return a value, and takes an extra alignment argument.
3348 The first argument is a pointer to the destination to fill, the second is the
3349 byte value to fill it with, the third argument is an integer
3350 argument specifying the number of bytes to fill, and the fourth argument is the
3351 known alignment of destination location.
3355 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3356 the caller guarantees that the destination pointer is aligned to that boundary.
3362 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
3364 destination location. If the argument is known to be aligned to some boundary,
3365 this can be specified as the fourth argument, otherwise it should be set to 0 or
3371 <!-- _______________________________________________________________________ -->
3372 <div class="doc_subsubsection">
3373 <a name="i_isunordered">'<tt>llvm.isunordered.*</tt>' Intrinsic</a>
3376 <div class="doc_text">
3380 declare bool %llvm.isunordered.f32(float Val1, float Val2)
3381 declare bool %llvm.isunordered.f64(double Val1, double Val2)
3387 The '<tt>llvm.isunordered</tt>' intrinsics return true if either or both of the
3388 specified floating point values is a NAN.
3394 The arguments are floating point numbers of the same type.
3400 If either or both of the arguments is a SNAN or QNAN, it returns true, otherwise
3406 <!-- _______________________________________________________________________ -->
3407 <div class="doc_subsubsection">
3408 <a name="i_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
3411 <div class="doc_text">
3415 declare double %llvm.sqrt.f32(float Val)
3416 declare double %llvm.sqrt.f64(double Val)
3422 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
3423 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
3424 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
3425 negative numbers (which allows for better optimization).
3431 The argument and return value are floating point numbers of the same type.
3437 This function returns the sqrt of the specified operand if it is a positive
3438 floating point number.
3442 <!-- ======================================================================= -->
3443 <div class="doc_subsection">
3444 <a name="int_manip">Bit Manipulation Intrinsics</a>
3447 <div class="doc_text">
3449 LLVM provides intrinsics for a few important bit manipulation operations.
3450 These allow efficient code generation for some algorithms.
3455 <!-- _______________________________________________________________________ -->
3456 <div class="doc_subsubsection">
3457 <a name="i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
3460 <div class="doc_text">
3464 declare ushort %llvm.bswap.i16(ushort <id>)
3465 declare uint %llvm.bswap.i32(uint <id>)
3466 declare ulong %llvm.bswap.i64(ulong <id>)
3472 The '<tt>llvm.bwsap</tt>' family of intrinsics is used to byteswap a 16, 32 or
3473 64 bit quantity. These are useful for performing operations on data that is not
3474 in the target's native byte order.
3480 The <tt>llvm.bswap.16</tt> intrinsic returns a ushort value that has the high and low
3481 byte of the input ushort swapped. Similarly, the <tt>llvm.bswap.i32</tt> intrinsic
3482 returns a uint value that has the four bytes of the input uint swapped, so that
3483 if the input bytes are numbered 0, 1, 2, 3 then the returned uint will have its
3484 bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i64</tt> intrinsic extends this concept
3490 <!-- _______________________________________________________________________ -->
3491 <div class="doc_subsubsection">
3492 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
3495 <div class="doc_text">
3499 declare ubyte %llvm.ctpop.i8 (ubyte <src>)
3500 declare ushort %llvm.ctpop.i16(ushort <src>)
3501 declare uint %llvm.ctpop.i32(uint <src>)
3502 declare ulong %llvm.ctpop.i64(ulong <src>)
3508 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
3515 The only argument is the value to be counted. The argument may be of any
3516 unsigned integer type. The return type must match the argument type.
3522 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
3526 <!-- _______________________________________________________________________ -->
3527 <div class="doc_subsubsection">
3528 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
3531 <div class="doc_text">
3535 declare ubyte %llvm.ctlz.i8 (ubyte <src>)
3536 declare ushort %llvm.ctlz.i16(ushort <src>)
3537 declare uint %llvm.ctlz.i32(uint <src>)
3538 declare ulong %llvm.ctlz.i64(ulong <src>)
3544 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
3545 leading zeros in a variable.
3551 The only argument is the value to be counted. The argument may be of any
3552 unsigned integer type. The return type must match the argument type.
3558 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
3559 in a variable. If the src == 0 then the result is the size in bits of the type
3560 of src. For example, <tt>llvm.cttz(int 2) = 30</tt>.
3566 <!-- _______________________________________________________________________ -->
3567 <div class="doc_subsubsection">
3568 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
3571 <div class="doc_text">
3575 declare ubyte %llvm.cttz.i8 (ubyte <src>)
3576 declare ushort %llvm.cttz.i16(ushort <src>)
3577 declare uint %llvm.cttz.i32(uint <src>)
3578 declare ulong %llvm.cttz.i64(ulong <src>)
3584 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
3591 The only argument is the value to be counted. The argument may be of any
3592 unsigned integer type. The return type must match the argument type.
3598 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
3599 in a variable. If the src == 0 then the result is the size in bits of the type
3600 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
3604 <!-- ======================================================================= -->
3605 <div class="doc_subsection">
3606 <a name="int_debugger">Debugger Intrinsics</a>
3609 <div class="doc_text">
3611 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
3612 are described in the <a
3613 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
3614 Debugging</a> document.
3619 <!-- *********************************************************************** -->
3622 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3623 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
3624 <a href="http://validator.w3.org/check/referer"><img
3625 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
3627 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3628 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
3629 Last modified: $Date$