Gitiles
Code Review Sign In
gerrit-public.fairphone.software / fp2-dev / platform / external / llvm / 624dc1d4abf26a3ccd474f85a39058a99a9053ca / . / docs / CodeGenerator.html
blob: 50036f2ed73f94a9fce340e3900a75eb7ec06415 [file] [log] [blame]
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
3<html>
4<head>
Jim Laskeyb744c252006-12-15 10:40:48 +0000[diff] [blame]5 <meta http-equiv="content-type" content="text/html; charset=utf-8">
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]6 <title>The LLVM Target-Independent Code Generator</title>
7 <link rel="stylesheet" href="llvm.css" type="text/css">
Benjamin Kramer943beeb2010-10-30 21:07:28 +0000[diff] [blame]8
9 <style type="text/css">
10 .unknown { background-color: #C0C0C0; text-align: center; }
11 .unknown:before { content: "?" }
12 .no { background-color: #C11B17 }
13 .no:before { content: "N" }
14 .partial { background-color: #F88017 }
15 .yes { background-color: #0F0; }
16 .yes:before { content: "Y" }
17 </style>
18
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]19</head>
20<body>
21
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]22<h1>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]23 The LLVM Target-Independent Code Generator
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]24</h1>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]25
26<ol>
27 <li><a href="#introduction">Introduction</a>
28 <ul>
29 <li><a href="#required">Required components in the code generator</a></li>
Chris Lattnere35d3bb2005-10-16 00:36:38 +0000[diff] [blame]30 <li><a href="#high-level-design">The high-level design of the code
31 generator</a></li>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]32 <li><a href="#tablegen">Using TableGen for target description</a></li>
33 </ul>
34 </li>
35 <li><a href="#targetdesc">Target description classes</a>
36 <ul>
37 <li><a href="#targetmachine">The <tt>TargetMachine</tt> class</a></li>
38 <li><a href="#targetdata">The <tt>TargetData</tt> class</a></li>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]39 <li><a href="#targetlowering">The <tt>TargetLowering</tt> class</a></li>
Dan Gohman6f0d0242008-02-10 18:45:23 +0000[diff] [blame]40 <li><a href="#targetregisterinfo">The <tt>TargetRegisterInfo</tt> class</a></li>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]41 <li><a href="#targetinstrinfo">The <tt>TargetInstrInfo</tt> class</a></li>
42 <li><a href="#targetframeinfo">The <tt>TargetFrameInfo</tt> class</a></li>
Chris Lattner47adebb2005-10-16 17:06:07 +0000[diff] [blame]43 <li><a href="#targetsubtarget">The <tt>TargetSubtarget</tt> class</a></li>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]44 <li><a href="#targetjitinfo">The <tt>TargetJITInfo</tt> class</a></li>
45 </ul>
46 </li>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]47 <li><a href="#codegendesc">The "Machine" Code Generator classes</a>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]48 <ul>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]49 <li><a href="#machineinstr">The <tt>MachineInstr</tt> class</a></li>
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]50 <li><a href="#machinebasicblock">The <tt>MachineBasicBlock</tt>
51 class</a></li>
52 <li><a href="#machinefunction">The <tt>MachineFunction</tt> class</a></li>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]53 </ul>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]54 </li>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]55 <li><a href="#mc">The "MC" Layer</a>
56 <ul>
57 <li><a href="#mcstreamer">The <tt>MCStreamer</tt> API</a></li>
58 <li><a href="#mccontext">The <tt>MCContext</tt> class</a>
59 <li><a href="#mcsymbol">The <tt>MCSymbol</tt> class</a></li>
60 <li><a href="#mcsection">The <tt>MCSection</tt> class</a></li>
61 <li><a href="#mcinst">The <tt>MCInst</tt> class</a></li>
62 </ul>
63 </li>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]64 <li><a href="#codegenalgs">Target-independent code generation algorithms</a>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]65 <ul>
66 <li><a href="#instselect">Instruction Selection</a>
67 <ul>
68 <li><a href="#selectiondag_intro">Introduction to SelectionDAGs</a></li>
69 <li><a href="#selectiondag_process">SelectionDAG Code Generation
70 Process</a></li>
71 <li><a href="#selectiondag_build">Initial SelectionDAG
72 Construction</a></li>
Dan Gohman641b2792008-11-24 16:27:17 +0000[diff] [blame]73 <li><a href="#selectiondag_legalize_types">SelectionDAG LegalizeTypes Phase</a></li>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]74 <li><a href="#selectiondag_legalize">SelectionDAG Legalize Phase</a></li>
75 <li><a href="#selectiondag_optimize">SelectionDAG Optimization
Chris Lattnere35d3bb2005-10-16 00:36:38 +0000[diff] [blame]76 Phase: the DAG Combiner</a></li>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]77 <li><a href="#selectiondag_select">SelectionDAG Select Phase</a></li>
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]78 <li><a href="#selectiondag_sched">SelectionDAG Scheduling and Formation
Chris Lattnere35d3bb2005-10-16 00:36:38 +0000[diff] [blame]79 Phase</a></li>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]80 <li><a href="#selectiondag_future">Future directions for the
81 SelectionDAG</a></li>
82 </ul></li>
Bill Wendling3fc488d2006-09-06 18:42:41 +0000[diff] [blame]83 <li><a href="#liveintervals">Live Intervals</a>
Bill Wendling2f87a882006-09-04 23:35:52 +0000[diff] [blame]84 <ul>
85 <li><a href="#livevariable_analysis">Live Variable Analysis</a></li>
Bill Wendling3fc488d2006-09-06 18:42:41 +0000[diff] [blame]86 <li><a href="#liveintervals_analysis">Live Intervals Analysis</a></li>
Bill Wendling2f87a882006-09-04 23:35:52 +0000[diff] [blame]87 </ul></li>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]88 <li><a href="#regalloc">Register Allocation</a>
89 <ul>
90 <li><a href="#regAlloc_represent">How registers are represented in
91 LLVM</a></li>
92 <li><a href="#regAlloc_howTo">Mapping virtual registers to physical
93 registers</a></li>
94 <li><a href="#regAlloc_twoAddr">Handling two address instructions</a></li>
95 <li><a href="#regAlloc_ssaDecon">The SSA deconstruction phase</a></li>
96 <li><a href="#regAlloc_fold">Instruction folding</a></li>
97 <li><a href="#regAlloc_builtIn">Built in register allocators</a></li>
98 </ul></li>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]99 <li><a href="#codeemit">Code Emission</a></li>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]100 </ul>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]101 </li>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]102 <li><a href="#nativeassembler">Implementing a Native Assembler</a></li>
103
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]104 <li><a href="#targetimpls">Target-specific Implementation Notes</a>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]105 <ul>
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]106 <li><a href="#targetfeatures">Target Feature Matrix</a></li>
Arnold Schwaighofer9097d142008-05-14 09:17:12 +0000[diff] [blame]107 <li><a href="#tailcallopt">Tail call optimization</a></li>
Evan Chengdc444e92010-03-08 21:05:02 +0000[diff] [blame]108 <li><a href="#sibcallopt">Sibling call optimization</a></li>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]109 <li><a href="#x86">The X86 backend</a></li>
Jim Laskeyb744c252006-12-15 10:40:48 +0000[diff] [blame]110 <li><a href="#ppc">The PowerPC backend</a>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]111 <ul>
112 <li><a href="#ppc_abi">LLVM PowerPC ABI</a></li>
113 <li><a href="#ppc_frame">Frame Layout</a></li>
114 <li><a href="#ppc_prolog">Prolog/Epilog</a></li>
115 <li><a href="#ppc_dynamic">Dynamic Allocation</a></li>
Jim Laskeyb744c252006-12-15 10:40:48 +0000[diff] [blame]116 </ul></li>
117 </ul></li>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]118
119</ol>
120
121<div class="doc_author">
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]122 <p>Written by the LLVM Team.</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]123</div>
124
Chris Lattner10d68002004-06-01 17:18:11 +0000[diff] [blame]125<div class="doc_warning">
126 <p>Warning: This is a work in progress.</p>
127</div>
128
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]129<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]130<h2>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]131 <a name="introduction">Introduction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]132</h2>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]133<!-- *********************************************************************** -->
134
135<div class="doc_text">
136
137<p>The LLVM target-independent code generator is a framework that provides a
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]138 suite of reusable components for translating the LLVM internal representation
139 to the machine code for a specified target&mdash;either in assembly form
140 (suitable for a static compiler) or in binary machine code format (usable for
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]141 a JIT compiler). The LLVM target-independent code generator consists of six
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]142 main components:</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]143
144<ol>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]145 <li><a href="#targetdesc">Abstract target description</a> interfaces which
146 capture important properties about various aspects of the machine,
147 independently of how they will be used. These interfaces are defined in
148 <tt>include/llvm/Target/</tt>.</li>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]149
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]150 <li>Classes used to represent the <a href="#codegendesc">code being
151 generated</a> for a target. These classes are intended to be abstract
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]152 enough to represent the machine code for <i>any</i> target machine. These
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]153 classes are defined in <tt>include/llvm/CodeGen/</tt>. At this level,
154 concepts like "constant pool entries" and "jump tables" are explicitly
155 exposed.</li>
156
157 <li>Classes and algorithms used to represent code as the object file level,
158 the <a href="#mc">MC Layer</a>. These classes represent assembly level
159 constructs like labels, sections, and instructions. At this level,
160 concepts like "constant pool entries" and "jump tables" don't exist.</li>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]161
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]162 <li><a href="#codegenalgs">Target-independent algorithms</a> used to implement
163 various phases of native code generation (register allocation, scheduling,
164 stack frame representation, etc). This code lives
165 in <tt>lib/CodeGen/</tt>.</li>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]166
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]167 <li><a href="#targetimpls">Implementations of the abstract target description
168 interfaces</a> for particular targets. These machine descriptions make
169 use of the components provided by LLVM, and can optionally provide custom
170 target-specific passes, to build complete code generators for a specific
171 target. Target descriptions live in <tt>lib/Target/</tt>.</li>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]172
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]173 <li><a href="#jit">The target-independent JIT components</a>. The LLVM JIT is
174 completely target independent (it uses the <tt>TargetJITInfo</tt>
175 structure to interface for target-specific issues. The code for the
176 target-independent JIT lives in <tt>lib/ExecutionEngine/JIT</tt>.</li>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]177</ol>
178
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]179<p>Depending on which part of the code generator you are interested in working
180 on, different pieces of this will be useful to you. In any case, you should
181 be familiar with the <a href="#targetdesc">target description</a>
182 and <a href="#codegendesc">machine code representation</a> classes. If you
183 want to add a backend for a new target, you will need
184 to <a href="#targetimpls">implement the target description</a> classes for
185 your new target and understand the <a href="LangRef.html">LLVM code
186 representation</a>. If you are interested in implementing a
187 new <a href="#codegenalgs">code generation algorithm</a>, it should only
188 depend on the target-description and machine code representation classes,
189 ensuring that it is portable.</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]190
191</div>
192
193<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]194<h3>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]195 <a name="required">Required components in the code generator</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]196</h3>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]197
198<div class="doc_text">
199
200<p>The two pieces of the LLVM code generator are the high-level interface to the
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]201 code generator and the set of reusable components that can be used to build
202 target-specific backends. The two most important interfaces
203 (<a href="#targetmachine"><tt>TargetMachine</tt></a>
204 and <a href="#targetdata"><tt>TargetData</tt></a>) are the only ones that are
205 required to be defined for a backend to fit into the LLVM system, but the
206 others must be defined if the reusable code generator components are going to
207 be used.</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]208
209<p>This design has two important implications. The first is that LLVM can
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]210 support completely non-traditional code generation targets. For example, the
211 C backend does not require register allocation, instruction selection, or any
212 of the other standard components provided by the system. As such, it only
213 implements these two interfaces, and does its own thing. Another example of
214 a code generator like this is a (purely hypothetical) backend that converts
215 LLVM to the GCC RTL form and uses GCC to emit machine code for a target.</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]216
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]217<p>This design also implies that it is possible to design and implement
218 radically different code generators in the LLVM system that do not make use
219 of any of the built-in components. Doing so is not recommended at all, but
220 could be required for radically different targets that do not fit into the
221 LLVM machine description model: FPGAs for example.</p>
Chris Lattner900bf8c2004-06-02 07:06:06 +0000[diff] [blame]222
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]223</div>
224
225<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]226<h3>
Chris Lattner10d68002004-06-01 17:18:11 +0000[diff] [blame]227 <a name="high-level-design">The high-level design of the code generator</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]228</h3>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]229
230<div class="doc_text">
231
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]232<p>The LLVM target-independent code generator is designed to support efficient
233 and quality code generation for standard register-based microprocessors.
234 Code generation in this model is divided into the following stages:</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]235
236<ol>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]237 <li><b><a href="#instselect">Instruction Selection</a></b> &mdash; This phase
238 determines an efficient way to express the input LLVM code in the target
239 instruction set. This stage produces the initial code for the program in
240 the target instruction set, then makes use of virtual registers in SSA
241 form and physical registers that represent any required register
242 assignments due to target constraints or calling conventions. This step
243 turns the LLVM code into a DAG of target instructions.</li>
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]244
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]245 <li><b><a href="#selectiondag_sched">Scheduling and Formation</a></b> &mdash;
246 This phase takes the DAG of target instructions produced by the
247 instruction selection phase, determines an ordering of the instructions,
248 then emits the instructions
249 as <tt><a href="#machineinstr">MachineInstr</a></tt>s with that ordering.
250 Note that we describe this in the <a href="#instselect">instruction
251 selection section</a> because it operates on
252 a <a href="#selectiondag_intro">SelectionDAG</a>.</li>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]253
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]254 <li><b><a href="#ssamco">SSA-based Machine Code Optimizations</a></b> &mdash;
255 This optional stage consists of a series of machine-code optimizations
256 that operate on the SSA-form produced by the instruction selector.
257 Optimizations like modulo-scheduling or peephole optimization work
258 here.</li>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]259
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]260 <li><b><a href="#regalloc">Register Allocation</a></b> &mdash; The target code
261 is transformed from an infinite virtual register file in SSA form to the
262 concrete register file used by the target. This phase introduces spill
263 code and eliminates all virtual register references from the program.</li>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]264
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]265 <li><b><a href="#proepicode">Prolog/Epilog Code Insertion</a></b> &mdash; Once
266 the machine code has been generated for the function and the amount of
267 stack space required is known (used for LLVM alloca's and spill slots),
268 the prolog and epilog code for the function can be inserted and "abstract
269 stack location references" can be eliminated. This stage is responsible
270 for implementing optimizations like frame-pointer elimination and stack
271 packing.</li>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]272
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]273 <li><b><a href="#latemco">Late Machine Code Optimizations</a></b> &mdash;
274 Optimizations that operate on "final" machine code can go here, such as
275 spill code scheduling and peephole optimizations.</li>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]276
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]277 <li><b><a href="#codeemit">Code Emission</a></b> &mdash; The final stage
278 actually puts out the code for the current function, either in the target
279 assembler format or in machine code.</li>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]280</ol>
281
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]282<p>The code generator is based on the assumption that the instruction selector
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]283 will use an optimal pattern matching selector to create high-quality
284 sequences of native instructions. Alternative code generator designs based
285 on pattern expansion and aggressive iterative peephole optimization are much
286 slower. This design permits efficient compilation (important for JIT
287 environments) and aggressive optimization (used when generating code offline)
288 by allowing components of varying levels of sophistication to be used for any
289 step of compilation.</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]290
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]291<p>In addition to these stages, target implementations can insert arbitrary
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]292 target-specific passes into the flow. For example, the X86 target uses a
293 special pass to handle the 80x87 floating point stack architecture. Other
294 targets with unusual requirements can be supported with custom passes as
295 needed.</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]296
297</div>
298
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]299<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]300<h3>
Chris Lattner10d68002004-06-01 17:18:11 +0000[diff] [blame]301 <a name="tablegen">Using TableGen for target description</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]302</h3>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]303
304<div class="doc_text">
305
Chris Lattner5489e932004-06-01 18:35:00 +0000[diff] [blame]306<p>The target description classes require a detailed description of the target
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]307 architecture. These target descriptions often have a large amount of common
308 information (e.g., an <tt>add</tt> instruction is almost identical to a
309 <tt>sub</tt> instruction). In order to allow the maximum amount of
310 commonality to be factored out, the LLVM code generator uses
311 the <a href="TableGenFundamentals.html">TableGen</a> tool to describe big
312 chunks of the target machine, which allows the use of domain-specific and
313 target-specific abstractions to reduce the amount of repetition.</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]314
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]315<p>As LLVM continues to be developed and refined, we plan to move more and more
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]316 of the target description to the <tt>.td</tt> form. Doing so gives us a
317 number of advantages. The most important is that it makes it easier to port
318 LLVM because it reduces the amount of C++ code that has to be written, and
319 the surface area of the code generator that needs to be understood before
320 someone can get something working. Second, it makes it easier to change
321 things. In particular, if tables and other things are all emitted
322 by <tt>tblgen</tt>, we only need a change in one place (<tt>tblgen</tt>) to
323 update all of the targets to a new interface.</p>
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]324
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]325</div>
326
327<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]328<h2>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]329 <a name="targetdesc">Target description classes</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]330</h2>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]331<!-- *********************************************************************** -->
332
333<div class="doc_text">
334
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]335<p>The LLVM target description classes (located in the
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]336 <tt>include/llvm/Target</tt> directory) provide an abstract description of
337 the target machine independent of any particular client. These classes are
338 designed to capture the <i>abstract</i> properties of the target (such as the
339 instructions and registers it has), and do not incorporate any particular
340 pieces of code generation algorithms.</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]341
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]342<p>All of the target description classes (except the
343 <tt><a href="#targetdata">TargetData</a></tt> class) are designed to be
344 subclassed by the concrete target implementation, and have virtual methods
345 implemented. To get to these implementations, the
346 <tt><a href="#targetmachine">TargetMachine</a></tt> class provides accessors
347 that should be implemented by the target.</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]348
349</div>
350
351<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]352<h3>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]353 <a name="targetmachine">The <tt>TargetMachine</tt> class</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]354</h3>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]355
356<div class="doc_text">
357
358<p>The <tt>TargetMachine</tt> class provides virtual methods that are used to
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]359 access the target-specific implementations of the various target description
360 classes via the <tt>get*Info</tt> methods (<tt>getInstrInfo</tt>,
361 <tt>getRegisterInfo</tt>, <tt>getFrameInfo</tt>, etc.). This class is
362 designed to be specialized by a concrete target implementation
363 (e.g., <tt>X86TargetMachine</tt>) which implements the various virtual
364 methods. The only required target description class is
365 the <a href="#targetdata"><tt>TargetData</tt></a> class, but if the code
366 generator components are to be used, the other interfaces should be
367 implemented as well.</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]368
369</div>
370
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]371<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]372<h3>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]373 <a name="targetdata">The <tt>TargetData</tt> class</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]374</h3>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]375
376<div class="doc_text">
377
378<p>The <tt>TargetData</tt> class is the only required target description class,
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]379 and it is the only class that is not extensible (you cannot derived a new
380 class from it). <tt>TargetData</tt> specifies information about how the
381 target lays out memory for structures, the alignment requirements for various
382 data types, the size of pointers in the target, and whether the target is
383 little-endian or big-endian.</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]384
385</div>
386
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]387<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]388<h3>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]389 <a name="targetlowering">The <tt>TargetLowering</tt> class</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]390</h3>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]391
392<div class="doc_text">
393
394<p>The <tt>TargetLowering</tt> class is used by SelectionDAG based instruction
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]395 selectors primarily to describe how LLVM code should be lowered to
396 SelectionDAG operations. Among other things, this class indicates:</p>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]397
398<ul>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]399 <li>an initial register class to use for various <tt>ValueType</tt>s,</li>
400
401 <li>which operations are natively supported by the target machine,</li>
402
403 <li>the return type of <tt>setcc</tt> operations,</li>
404
405 <li>the type to use for shift amounts, and</li>
406
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]407 <li>various high-level characteristics, like whether it is profitable to turn
408 division by a constant into a multiplication sequence</li>
Jim Laskeyb744c252006-12-15 10:40:48 +0000[diff] [blame]409</ul>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]410
411</div>
412
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]413<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]414<h3>
Dan Gohman6f0d0242008-02-10 18:45:23 +0000[diff] [blame]415 <a name="targetregisterinfo">The <tt>TargetRegisterInfo</tt> class</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]416</h3>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]417
418<div class="doc_text">
419
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]420<p>The <tt>TargetRegisterInfo</tt> class is used to describe the register file
421 of the target and any interactions between the registers.</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]422
423<p>Registers in the code generator are represented in the code generator by
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]424 unsigned integers. Physical registers (those that actually exist in the
425 target description) are unique small numbers, and virtual registers are
426 generally large. Note that register #0 is reserved as a flag value.</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]427
428<p>Each register in the processor description has an associated
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]429 <tt>TargetRegisterDesc</tt> entry, which provides a textual name for the
430 register (used for assembly output and debugging dumps) and a set of aliases
431 (used to indicate whether one register overlaps with another).</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]432
Dan Gohman6f0d0242008-02-10 18:45:23 +0000[diff] [blame]433<p>In addition to the per-register description, the <tt>TargetRegisterInfo</tt>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]434 class exposes a set of processor specific register classes (instances of the
435 <tt>TargetRegisterClass</tt> class). Each register class contains sets of
436 registers that have the same properties (for example, they are all 32-bit
437 integer registers). Each SSA virtual register created by the instruction
438 selector has an associated register class. When the register allocator runs,
439 it replaces virtual registers with a physical register in the set.</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]440
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]441<p>The target-specific implementations of these classes is auto-generated from
442 a <a href="TableGenFundamentals.html">TableGen</a> description of the
443 register file.</p>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]444
445</div>
446
447<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]448<h3>
Chris Lattner10d68002004-06-01 17:18:11 +0000[diff] [blame]449 <a name="targetinstrinfo">The <tt>TargetInstrInfo</tt> class</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]450</h3>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]451
Reid Spencer627cd002005-07-19 01:36:35 +0000[diff] [blame]452<div class="doc_text">
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]453
454<p>The <tt>TargetInstrInfo</tt> class is used to describe the machine
455 instructions supported by the target. It is essentially an array of
456 <tt>TargetInstrDescriptor</tt> objects, each of which describes one
457 instruction the target supports. Descriptors define things like the mnemonic
458 for the opcode, the number of operands, the list of implicit register uses
459 and defs, whether the instruction has certain target-independent properties
460 (accesses memory, is commutable, etc), and holds any target-specific
461 flags.</p>
462
Reid Spencer627cd002005-07-19 01:36:35 +0000[diff] [blame]463</div>
464
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]465<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]466<h3>
Chris Lattner10d68002004-06-01 17:18:11 +0000[diff] [blame]467 <a name="targetframeinfo">The <tt>TargetFrameInfo</tt> class</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]468</h3>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]469
Reid Spencer627cd002005-07-19 01:36:35 +0000[diff] [blame]470<div class="doc_text">
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]471
472<p>The <tt>TargetFrameInfo</tt> class is used to provide information about the
473 stack frame layout of the target. It holds the direction of stack growth, the
474 known stack alignment on entry to each function, and the offset to the local
475 area. The offset to the local area is the offset from the stack pointer on
476 function entry to the first location where function data (local variables,
477 spill locations) can be stored.</p>
478
Reid Spencer627cd002005-07-19 01:36:35 +0000[diff] [blame]479</div>
Chris Lattner47adebb2005-10-16 17:06:07 +0000[diff] [blame]480
481<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]482<h3>
Chris Lattner47adebb2005-10-16 17:06:07 +0000[diff] [blame]483 <a name="targetsubtarget">The <tt>TargetSubtarget</tt> class</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]484</h3>
Chris Lattner47adebb2005-10-16 17:06:07 +0000[diff] [blame]485
486<div class="doc_text">
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]487
488<p>The <tt>TargetSubtarget</tt> class is used to provide information about the
489 specific chip set being targeted. A sub-target informs code generation of
490 which instructions are supported, instruction latencies and instruction
491 execution itinerary; i.e., which processing units are used, in what order,
492 and for how long.</p>
493
Chris Lattner47adebb2005-10-16 17:06:07 +0000[diff] [blame]494</div>
495
496
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]497<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]498<h3>
Chris Lattner10d68002004-06-01 17:18:11 +0000[diff] [blame]499 <a name="targetjitinfo">The <tt>TargetJITInfo</tt> class</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]500</h3>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]501
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]502<div class="doc_text">
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]503
504<p>The <tt>TargetJITInfo</tt> class exposes an abstract interface used by the
505 Just-In-Time code generator to perform target-specific activities, such as
506 emitting stubs. If a <tt>TargetMachine</tt> supports JIT code generation, it
507 should provide one of these objects through the <tt>getJITInfo</tt>
508 method.</p>
509
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]510</div>
511
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]512<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]513<h2>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]514 <a name="codegendesc">Machine code description classes</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]515</h2>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]516<!-- *********************************************************************** -->
517
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]518<div class="doc_text">
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]519
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]520<p>At the high-level, LLVM code is translated to a machine specific
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]521 representation formed out of
522 <a href="#machinefunction"><tt>MachineFunction</tt></a>,
523 <a href="#machinebasicblock"><tt>MachineBasicBlock</tt></a>,
524 and <a href="#machineinstr"><tt>MachineInstr</tt></a> instances (defined
525 in <tt>include/llvm/CodeGen</tt>). This representation is completely target
526 agnostic, representing instructions in their most abstract form: an opcode
527 and a series of operands. This representation is designed to support both an
528 SSA representation for machine code, as well as a register allocated, non-SSA
529 form.</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]530
531</div>
532
533<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]534<h3>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]535 <a name="machineinstr">The <tt>MachineInstr</tt> class</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]536</h3>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]537
538<div class="doc_text">
539
540<p>Target machine instructions are represented as instances of the
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]541 <tt>MachineInstr</tt> class. This class is an extremely abstract way of
542 representing machine instructions. In particular, it only keeps track of an
543 opcode number and a set of operands.</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]544
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]545<p>The opcode number is a simple unsigned integer that only has meaning to a
546 specific backend. All of the instructions for a target should be defined in
547 the <tt>*InstrInfo.td</tt> file for the target. The opcode enum values are
548 auto-generated from this description. The <tt>MachineInstr</tt> class does
549 not have any information about how to interpret the instruction (i.e., what
550 the semantics of the instruction are); for that you must refer to the
551 <tt><a href="#targetinstrinfo">TargetInstrInfo</a></tt> class.</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]552
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]553<p>The operands of a machine instruction can be of several different types: a
554 register reference, a constant integer, a basic block reference, etc. In
555 addition, a machine operand should be marked as a def or a use of the value
556 (though only registers are allowed to be defs).</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]557
558<p>By convention, the LLVM code generator orders instruction operands so that
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]559 all register definitions come before the register uses, even on architectures
560 that are normally printed in other orders. For example, the SPARC add
561 instruction: "<tt>add %i1, %i2, %i3</tt>" adds the "%i1", and "%i2" registers
562 and stores the result into the "%i3" register. In the LLVM code generator,
563 the operands should be stored as "<tt>%i3, %i1, %i2</tt>": with the
564 destination first.</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]565
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]566<p>Keeping destination (definition) operands at the beginning of the operand
567 list has several advantages. In particular, the debugging printer will print
568 the instruction like this:</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]569
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]570<div class="doc_code">
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]571<pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]572%r3 = add %i1, %i2
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]573</pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]574</div>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]575
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]576<p>Also if the first operand is a def, it is easier to <a href="#buildmi">create
577 instructions</a> whose only def is the first operand.</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]578
579</div>
580
581<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]582<h4>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]583 <a name="buildmi">Using the <tt>MachineInstrBuilder.h</tt> functions</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]584</h4>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]585
586<div class="doc_text">
587
588<p>Machine instructions are created by using the <tt>BuildMI</tt> functions,
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]589 located in the <tt>include/llvm/CodeGen/MachineInstrBuilder.h</tt> file. The
590 <tt>BuildMI</tt> functions make it easy to build arbitrary machine
591 instructions. Usage of the <tt>BuildMI</tt> functions look like this:</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]592
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]593<div class="doc_code">
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]594<pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]595// Create a 'DestReg = mov 42' (rendered in X86 assembly as 'mov DestReg, 42')
596// instruction. The '1' specifies how many operands will be added.
597MachineInstr *MI = BuildMI(X86::MOV32ri, 1, DestReg).addImm(42);
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]598
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]599// Create the same instr, but insert it at the end of a basic block.
600MachineBasicBlock &amp;MBB = ...
601BuildMI(MBB, X86::MOV32ri, 1, DestReg).addImm(42);
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]602
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]603// Create the same instr, but insert it before a specified iterator point.
604MachineBasicBlock::iterator MBBI = ...
605BuildMI(MBB, MBBI, X86::MOV32ri, 1, DestReg).addImm(42);
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]606
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]607// Create a 'cmp Reg, 0' instruction, no destination reg.
608MI = BuildMI(X86::CMP32ri, 2).addReg(Reg).addImm(0);
609// Create an 'sahf' instruction which takes no operands and stores nothing.
610MI = BuildMI(X86::SAHF, 0);
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]611
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]612// Create a self looping branch instruction.
613BuildMI(MBB, X86::JNE, 1).addMBB(&amp;MBB);
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]614</pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]615</div>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]616
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]617<p>The key thing to remember with the <tt>BuildMI</tt> functions is that you
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]618 have to specify the number of operands that the machine instruction will
619 take. This allows for efficient memory allocation. You also need to specify
620 if operands default to be uses of values, not definitions. If you need to
621 add a definition operand (other than the optional destination register), you
622 must explicitly mark it as such:</p>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]623
624<div class="doc_code">
625<pre>
Bill Wendling587daed2009-05-13 21:33:08 +0000[diff] [blame]626MI.addReg(Reg, RegState::Define);
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]627</pre>
628</div>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]629
630</div>
631
632<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]633<h4>
Reid Spencerad1f0cd2005-04-24 20:56:18 +0000[diff] [blame]634 <a name="fixedregs">Fixed (preassigned) registers</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]635</h4>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]636
637<div class="doc_text">
638
639<p>One important issue that the code generator needs to be aware of is the
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]640 presence of fixed registers. In particular, there are often places in the
641 instruction stream where the register allocator <em>must</em> arrange for a
642 particular value to be in a particular register. This can occur due to
643 limitations of the instruction set (e.g., the X86 can only do a 32-bit divide
644 with the <tt>EAX</tt>/<tt>EDX</tt> registers), or external factors like
645 calling conventions. In any case, the instruction selector should emit code
646 that copies a virtual register into or out of a physical register when
647 needed.</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]648
649<p>For example, consider this simple LLVM example:</p>
650
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]651<div class="doc_code">
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]652<pre>
Matthijs Kooijman61399af2008-06-04 15:46:35 +0000[diff] [blame]653define i32 @test(i32 %X, i32 %Y) {
654 %Z = udiv i32 %X, %Y
655 ret i32 %Z
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]656}
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]657</pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]658</div>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]659
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]660<p>The X86 instruction selector produces this machine code for the <tt>div</tt>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]661 and <tt>ret</tt> (use "<tt>llc X.bc -march=x86 -print-machineinstrs</tt>" to
662 get this):</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]663
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]664<div class="doc_code">
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]665<pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]666;; Start of div
667%EAX = mov %reg1024 ;; Copy X (in reg1024) into EAX
668%reg1027 = sar %reg1024, 31
669%EDX = mov %reg1027 ;; Sign extend X into EDX
670idiv %reg1025 ;; Divide by Y (in reg1025)
671%reg1026 = mov %EAX ;; Read the result (Z) out of EAX
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]672
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]673;; Start of ret
674%EAX = mov %reg1026 ;; 32-bit return value goes in EAX
675ret
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]676</pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]677</div>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]678
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]679<p>By the end of code generation, the register allocator has coalesced the
680 registers and deleted the resultant identity moves producing the following
681 code:</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]682
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]683<div class="doc_code">
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]684<pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]685;; X is in EAX, Y is in ECX
686mov %EAX, %EDX
687sar %EDX, 31
688idiv %ECX
689ret
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]690</pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]691</div>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]692
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]693<p>This approach is extremely general (if it can handle the X86 architecture, it
694 can handle anything!) and allows all of the target specific knowledge about
695 the instruction stream to be isolated in the instruction selector. Note that
696 physical registers should have a short lifetime for good code generation, and
697 all physical registers are assumed dead on entry to and exit from basic
698 blocks (before register allocation). Thus, if you need a value to be live
699 across basic block boundaries, it <em>must</em> live in a virtual
700 register.</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]701
702</div>
703
704<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]705<h4>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]706 <a name="ssa">Machine code in SSA form</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]707</h4>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]708
709<div class="doc_text">
710
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]711<p><tt>MachineInstr</tt>'s are initially selected in SSA-form, and are
712 maintained in SSA-form until register allocation happens. For the most part,
713 this is trivially simple since LLVM is already in SSA form; LLVM PHI nodes
714 become machine code PHI nodes, and virtual registers are only allowed to have
715 a single definition.</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]716
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]717<p>After register allocation, machine code is no longer in SSA-form because
718 there are no virtual registers left in the code.</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]719
720</div>
721
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]722<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]723<h3>
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]724 <a name="machinebasicblock">The <tt>MachineBasicBlock</tt> class</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]725</h3>
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]726
727<div class="doc_text">
728
729<p>The <tt>MachineBasicBlock</tt> class contains a list of machine instructions
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]730 (<tt><a href="#machineinstr">MachineInstr</a></tt> instances). It roughly
731 corresponds to the LLVM code input to the instruction selector, but there can
732 be a one-to-many mapping (i.e. one LLVM basic block can map to multiple
733 machine basic blocks). The <tt>MachineBasicBlock</tt> class has a
734 "<tt>getBasicBlock</tt>" method, which returns the LLVM basic block that it
735 comes from.</p>
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]736
737</div>
738
739<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]740<h3>
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]741 <a name="machinefunction">The <tt>MachineFunction</tt> class</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]742</h3>
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]743
744<div class="doc_text">
745
746<p>The <tt>MachineFunction</tt> class contains a list of machine basic blocks
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]747 (<tt><a href="#machinebasicblock">MachineBasicBlock</a></tt> instances). It
748 corresponds one-to-one with the LLVM function input to the instruction
749 selector. In addition to a list of basic blocks,
750 the <tt>MachineFunction</tt> contains a a <tt>MachineConstantPool</tt>,
751 a <tt>MachineFrameInfo</tt>, a <tt>MachineFunctionInfo</tt>, and a
752 <tt>MachineRegisterInfo</tt>. See
753 <tt>include/llvm/CodeGen/MachineFunction.h</tt> for more information.</p>
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]754
755</div>
756
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]757
758<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]759<h2>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]760 <a name="mc">The "MC" Layer</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]761</h2>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]762<!-- *********************************************************************** -->
763
764<div class="doc_text">
765
766<p>
767The MC Layer is used to represent and process code at the raw machine code
768level, devoid of "high level" information like "constant pools", "jump tables",
769"global variables" or anything like that. At this level, LLVM handles things
770like label names, machine instructions, and sections in the object file. The
771code in this layer is used for a number of important purposes: the tail end of
772the code generator uses it to write a .s or .o file, and it is also used by the
Jay Foadd61895a2011-04-13 13:03:56 +0000[diff] [blame]773llvm-mc tool to implement standalone machine code assemblers and disassemblers.
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]774</p>
775
776<p>
777This section describes some of the important classes. There are also a number
778of important subsystems that interact at this layer, they are described later
779in this manual.
780</p>
781
782</div>
783
784
785<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]786<h3>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]787 <a name="mcstreamer">The <tt>MCStreamer</tt> API</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]788</h3>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]789
790<div class="doc_text">
791
792<p>
793MCStreamer is best thought of as an assembler API. It is an abstract API which
794is <em>implemented</em> in different ways (e.g. to output a .s file, output an
795ELF .o file, etc) but whose API correspond directly to what you see in a .s
796file. MCStreamer has one method per directive, such as EmitLabel,
797EmitSymbolAttribute, SwitchSection, EmitValue (for .byte, .word), etc, which
798directly correspond to assembly level directives. It also has an
799EmitInstruction method, which is used to output an MCInst to the streamer.
800</p>
801
802<p>
803This API is most important for two clients: the llvm-mc stand-alone assembler is
804effectively a parser that parses a line, then invokes a method on MCStreamer. In
805the code generator, the <a href="#codeemit">Code Emission</a> phase of the code
806generator lowers higher level LLVM IR and Machine* constructs down to the MC
807layer, emitting directives through MCStreamer.</p>
808
809<p>
810On the implementation side of MCStreamer, there are two major implementations:
811one for writing out a .s file (MCAsmStreamer), and one for writing out a .o
812file (MCObjectStreamer). MCAsmStreamer is a straight-forward implementation
813that prints out a directive for each method (e.g. EmitValue -&gt; .byte), but
814MCObjectStreamer implements a full assembler.
815</p>
816
817</div>
818
819<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]820<h3>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]821 <a name="mccontext">The <tt>MCContext</tt> class</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]822</h3>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]823
824<div class="doc_text">
825
826<p>
827The MCContext class is the owner of a variety of uniqued data structures at the
828MC layer, including symbols, sections, etc. As such, this is the class that you
829interact with to create symbols and sections. This class can not be subclassed.
830</p>
831
832</div>
833
834<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]835<h3>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]836 <a name="mcsymbol">The <tt>MCSymbol</tt> class</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]837</h3>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]838
839<div class="doc_text">
840
841<p>
842The MCSymbol class represents a symbol (aka label) in the assembly file. There
843are two interesting kinds of symbols: assembler temporary symbols, and normal
844symbols. Assembler temporary symbols are used and processed by the assembler
845but are discarded when the object file is produced. The distinction is usually
846represented by adding a prefix to the label, for example "L" labels are
847assembler temporary labels in MachO.
848</p>
849
850<p>MCSymbols are created by MCContext and uniqued there. This means that
851MCSymbols can be compared for pointer equivalence to find out if they are the
852same symbol. Note that pointer inequality does not guarantee the labels will
853end up at different addresses though. It's perfectly legal to output something
854like this to the .s file:<p>
855
856<pre>
857 foo:
858 bar:
859 .byte 4
860</pre>
861
862<p>In this case, both the foo and bar symbols will have the same address.</p>
863
864</div>
865
866<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]867<h3>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]868 <a name="mcsection">The <tt>MCSection</tt> class</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]869</h3>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]870
871<div class="doc_text">
872
873<p>
874The MCSection class represents an object-file specific section. It is subclassed
875by object file specific implementations (e.g. <tt>MCSectionMachO</tt>,
876<tt>MCSectionCOFF</tt>, <tt>MCSectionELF</tt>) and these are created and uniqued
877by MCContext. The MCStreamer has a notion of the current section, which can be
878changed with the SwitchToSection method (which corresponds to a ".section"
879directive in a .s file).
880</p>
881
882</div>
883
884<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]885<h3>
Benjamin Kramer943beeb2010-10-30 21:07:28 +0000[diff] [blame]886 <a name="mcinst">The <tt>MCInst</tt> class</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]887</h3>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]888
889<div class="doc_text">
890
891<p>
892The MCInst class is a target-independent representation of an instruction. It
893is a simple class (much more so than <a href="#machineinstr">MachineInstr</a>)
894that holds a target-specific opcode and a vector of MCOperands. MCOperand, in
895turn, is a simple discriminated union of three cases: 1) a simple immediate,
8962) a target register ID, 3) a symbolic expression (e.g. "Lfoo-Lbar+42") as an
897MCExpr.
898</p>
899
900<p>MCInst is the common currency used to represent machine instructions at the
901MC layer. It is the type used by the instruction encoder, the instruction
902printer, and the type generated by the assembly parser and disassembler.
903</p>
904
905</div>
906
907
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]908<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]909<h2>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]910 <a name="codegenalgs">Target-independent code generation algorithms</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]911</h2>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]912<!-- *********************************************************************** -->
913
914<div class="doc_text">
915
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]916<p>This section documents the phases described in the
917 <a href="#high-level-design">high-level design of the code generator</a>.
918 It explains how they work and some of the rationale behind their design.</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]919
920</div>
921
922<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]923<h3>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]924 <a name="instselect">Instruction Selection</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]925</h3>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]926
927<div class="doc_text">
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]928
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]929<p>Instruction Selection is the process of translating LLVM code presented to
930 the code generator into target-specific machine instructions. There are
931 several well-known ways to do this in the literature. LLVM uses a
932 SelectionDAG based instruction selector.</p>
933
934<p>Portions of the DAG instruction selector are generated from the target
935 description (<tt>*.td</tt>) files. Our goal is for the entire instruction
936 selector to be generated from these <tt>.td</tt> files, though currently
937 there are still things that require custom C++ code.</p>
938
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]939</div>
940
941<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]942<h4>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]943 <a name="selectiondag_intro">Introduction to SelectionDAGs</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]944</h4>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]945
946<div class="doc_text">
947
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]948<p>The SelectionDAG provides an abstraction for code representation in a way
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]949 that is amenable to instruction selection using automatic techniques
950 (e.g. dynamic-programming based optimal pattern matching selectors). It is
951 also well-suited to other phases of code generation; in particular,
952 instruction scheduling (SelectionDAG's are very close to scheduling DAGs
953 post-selection). Additionally, the SelectionDAG provides a host
954 representation where a large variety of very-low-level (but
955 target-independent) <a href="#selectiondag_optimize">optimizations</a> may be
956 performed; ones which require extensive information about the instructions
957 efficiently supported by the target.</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]958
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]959<p>The SelectionDAG is a Directed-Acyclic-Graph whose nodes are instances of the
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]960 <tt>SDNode</tt> class. The primary payload of the <tt>SDNode</tt> is its
961 operation code (Opcode) that indicates what operation the node performs and
962 the operands to the operation. The various operation node types are
963 described at the top of the <tt>include/llvm/CodeGen/SelectionDAGNodes.h</tt>
964 file.</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]965
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]966<p>Although most operations define a single value, each node in the graph may
967 define multiple values. For example, a combined div/rem operation will
968 define both the dividend and the remainder. Many other situations require
969 multiple values as well. Each node also has some number of operands, which
970 are edges to the node defining the used value. Because nodes may define
971 multiple values, edges are represented by instances of the <tt>SDValue</tt>
972 class, which is a <tt>&lt;SDNode, unsigned&gt;</tt> pair, indicating the node
973 and result value being used, respectively. Each value produced by
974 an <tt>SDNode</tt> has an associated <tt>MVT</tt> (Machine Value Type)
975 indicating what the type of the value is.</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]976
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]977<p>SelectionDAGs contain two different kinds of values: those that represent
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]978 data flow and those that represent control flow dependencies. Data values
979 are simple edges with an integer or floating point value type. Control edges
980 are represented as "chain" edges which are of type <tt>MVT::Other</tt>.
981 These edges provide an ordering between nodes that have side effects (such as
982 loads, stores, calls, returns, etc). All nodes that have side effects should
983 take a token chain as input and produce a new one as output. By convention,
984 token chain inputs are always operand #0, and chain results are always the
985 last value produced by an operation.</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]986
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]987<p>A SelectionDAG has designated "Entry" and "Root" nodes. The Entry node is
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]988 always a marker node with an Opcode of <tt>ISD::EntryToken</tt>. The Root
989 node is the final side-effecting node in the token chain. For example, in a
990 single basic block function it would be the return node.</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]991
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]992<p>One important concept for SelectionDAGs is the notion of a "legal" vs.
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]993 "illegal" DAG. A legal DAG for a target is one that only uses supported
994 operations and supported types. On a 32-bit PowerPC, for example, a DAG with
995 a value of type i1, i8, i16, or i64 would be illegal, as would a DAG that
996 uses a SREM or UREM operation. The
997 <a href="#selectinodag_legalize_types">legalize types</a> and
998 <a href="#selectiondag_legalize">legalize operations</a> phases are
999 responsible for turning an illegal DAG into a legal DAG.</p>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1000
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1001</div>
1002
1003<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1004<h4>
Reid Spencerad1f0cd2005-04-24 20:56:18 +0000[diff] [blame]1005 <a name="selectiondag_process">SelectionDAG Instruction Selection Process</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1006</h4>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1007
1008<div class="doc_text">
1009
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1010<p>SelectionDAG-based instruction selection consists of the following steps:</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1011
1012<ol>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1013 <li><a href="#selectiondag_build">Build initial DAG</a> &mdash; This stage
1014 performs a simple translation from the input LLVM code to an illegal
1015 SelectionDAG.</li>
1016
1017 <li><a href="#selectiondag_optimize">Optimize SelectionDAG</a> &mdash; This
1018 stage performs simple optimizations on the SelectionDAG to simplify it,
1019 and recognize meta instructions (like rotates
1020 and <tt>div</tt>/<tt>rem</tt> pairs) for targets that support these meta
1021 operations. This makes the resultant code more efficient and
1022 the <a href="#selectiondag_select">select instructions from DAG</a> phase
1023 (below) simpler.</li>
1024
1025 <li><a href="#selectiondag_legalize_types">Legalize SelectionDAG Types</a>
1026 &mdash; This stage transforms SelectionDAG nodes to eliminate any types
1027 that are unsupported on the target.</li>
1028
1029 <li><a href="#selectiondag_optimize">Optimize SelectionDAG</a> &mdash; The
1030 SelectionDAG optimizer is run to clean up redundancies exposed by type
1031 legalization.</li>
1032
Chris Lattner71388632010-12-12 02:42:57 +0000[diff] [blame]1033 <li><a href="#selectiondag_legalize">Legalize SelectionDAG Ops</a> &mdash;
Chris Lattner4c247f62010-12-13 00:17:12 +0000[diff] [blame]1034 This stage transforms SelectionDAG nodes to eliminate any operations
1035 that are unsupported on the target.</li>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1036
1037 <li><a href="#selectiondag_optimize">Optimize SelectionDAG</a> &mdash; The
1038 SelectionDAG optimizer is run to eliminate inefficiencies introduced by
1039 operation legalization.</li>
1040
1041 <li><a href="#selectiondag_select">Select instructions from DAG</a> &mdash;
1042 Finally, the target instruction selector matches the DAG operations to
1043 target instructions. This process translates the target-independent input
1044 DAG into another DAG of target instructions.</li>
1045
1046 <li><a href="#selectiondag_sched">SelectionDAG Scheduling and Formation</a>
1047 &mdash; The last phase assigns a linear order to the instructions in the
1048 target-instruction DAG and emits them into the MachineFunction being
1049 compiled. This step uses traditional prepass scheduling techniques.</li>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1050</ol>
1051
1052<p>After all of these steps are complete, the SelectionDAG is destroyed and the
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1053 rest of the code generation passes are run.</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1054
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1055<p>One great way to visualize what is going on here is to take advantage of a
1056 few LLC command line options. The following options pop up a window
1057 displaying the SelectionDAG at specific times (if you only get errors printed
1058 to the console while using this, you probably
1059 <a href="ProgrammersManual.html#ViewGraph">need to configure your system</a>
1060 to add support for it).</p>
Dan Gohman8c9c55f2008-09-10 22:23:41 +0000[diff] [blame]1061
1062<ul>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1063 <li><tt>-view-dag-combine1-dags</tt> displays the DAG after being built,
1064 before the first optimization pass.</li>
1065
1066 <li><tt>-view-legalize-dags</tt> displays the DAG before Legalization.</li>
1067
1068 <li><tt>-view-dag-combine2-dags</tt> displays the DAG before the second
1069 optimization pass.</li>
1070
1071 <li><tt>-view-isel-dags</tt> displays the DAG before the Select phase.</li>
1072
1073 <li><tt>-view-sched-dags</tt> displays the DAG before Scheduling.</li>
Dan Gohman8c9c55f2008-09-10 22:23:41 +0000[diff] [blame]1074</ul>
1075
1076<p>The <tt>-view-sunit-dags</tt> displays the Scheduler's dependency graph.
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1077 This graph is based on the final SelectionDAG, with nodes that must be
1078 scheduled together bundled into a single scheduling-unit node, and with
1079 immediate operands and other nodes that aren't relevant for scheduling
1080 omitted.</p>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1081
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1082</div>
1083
1084<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1085<h4>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1086 <a name="selectiondag_build">Initial SelectionDAG Construction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1087</h4>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1088
1089<div class="doc_text">
1090
Bill Wendling16448772006-08-28 03:04:05 +0000[diff] [blame]1091<p>The initial SelectionDAG is na&iuml;vely peephole expanded from the LLVM
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1092 input by the <tt>SelectionDAGLowering</tt> class in the
1093 <tt>lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp</tt> file. The intent of
1094 this pass is to expose as much low-level, target-specific details to the
1095 SelectionDAG as possible. This pass is mostly hard-coded (e.g. an
1096 LLVM <tt>add</tt> turns into an <tt>SDNode add</tt> while a
1097 <tt>getelementptr</tt> is expanded into the obvious arithmetic). This pass
1098 requires target-specific hooks to lower calls, returns, varargs, etc. For
1099 these features, the <tt><a href="#targetlowering">TargetLowering</a></tt>
1100 interface is used.</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1101
1102</div>
1103
1104<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1105<h4>
Dan Gohman641b2792008-11-24 16:27:17 +0000[diff] [blame]1106 <a name="selectiondag_legalize_types">SelectionDAG LegalizeTypes Phase</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1107</h4>
Dan Gohman641b2792008-11-24 16:27:17 +0000[diff] [blame]1108
1109<div class="doc_text">
1110
1111<p>The Legalize phase is in charge of converting a DAG to only use the types
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1112 that are natively supported by the target.</p>
Dan Gohman641b2792008-11-24 16:27:17 +0000[diff] [blame]1113
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1114<p>There are two main ways of converting values of unsupported scalar types to
1115 values of supported types: converting small types to larger types
1116 ("promoting"), and breaking up large integer types into smaller ones
1117 ("expanding"). For example, a target might require that all f32 values are
1118 promoted to f64 and that all i1/i8/i16 values are promoted to i32. The same
1119 target might require that all i64 values be expanded into pairs of i32
1120 values. These changes can insert sign and zero extensions as needed to make
1121 sure that the final code has the same behavior as the input.</p>
Dan Gohman641b2792008-11-24 16:27:17 +0000[diff] [blame]1122
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1123<p>There are two main ways of converting values of unsupported vector types to
1124 value of supported types: splitting vector types, multiple times if
1125 necessary, until a legal type is found, and extending vector types by adding
1126 elements to the end to round them out to legal types ("widening"). If a
1127 vector gets split all the way down to single-element parts with no supported
1128 vector type being found, the elements are converted to scalars
1129 ("scalarizing").</p>
Dan Gohman641b2792008-11-24 16:27:17 +0000[diff] [blame]1130
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1131<p>A target implementation tells the legalizer which types are supported (and
1132 which register class to use for them) by calling the
Dan Gohman641b2792008-11-24 16:27:17 +0000[diff] [blame]1133 <tt>addRegisterClass</tt> method in its TargetLowering constructor.</p>
1134
1135</div>
1136
1137<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1138<h4>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1139 <a name="selectiondag_legalize">SelectionDAG Legalize Phase</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1140</h4>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1141
1142<div class="doc_text">
1143
Dan Gohman641b2792008-11-24 16:27:17 +0000[diff] [blame]1144<p>The Legalize phase is in charge of converting a DAG to only use the
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1145 operations that are natively supported by the target.</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1146
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1147<p>Targets often have weird constraints, such as not supporting every operation
1148 on every supported datatype (e.g. X86 does not support byte conditional moves
1149 and PowerPC does not support sign-extending loads from a 16-bit memory
1150 location). Legalize takes care of this by open-coding another sequence of
1151 operations to emulate the operation ("expansion"), by promoting one type to a
1152 larger type that supports the operation ("promotion"), or by using a
1153 target-specific hook to implement the legalization ("custom").</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1154
Dan Gohman641b2792008-11-24 16:27:17 +0000[diff] [blame]1155<p>A target implementation tells the legalizer which operations are not
1156 supported (and which of the above three actions to take) by calling the
1157 <tt>setOperationAction</tt> method in its <tt>TargetLowering</tt>
1158 constructor.</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1159
Dan Gohman641b2792008-11-24 16:27:17 +0000[diff] [blame]1160<p>Prior to the existence of the Legalize passes, we required that every target
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1161 <a href="#selectiondag_optimize">selector</a> supported and handled every
1162 operator and type even if they are not natively supported. The introduction
1163 of the Legalize phases allows all of the canonicalization patterns to be
1164 shared across targets, and makes it very easy to optimize the canonicalized
1165 code because it is still in the form of a DAG.</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1166
1167</div>
1168
1169<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1170<h4>
1171 <a name="selectiondag_optimize">
1172 SelectionDAG Optimization Phase: the DAG Combiner
1173 </a>
1174</h4>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1175
1176<div class="doc_text">
1177
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1178<p>The SelectionDAG optimization phase is run multiple times for code
1179 generation, immediately after the DAG is built and once after each
1180 legalization. The first run of the pass allows the initial code to be
1181 cleaned up (e.g. performing optimizations that depend on knowing that the
1182 operators have restricted type inputs). Subsequent runs of the pass clean up
1183 the messy code generated by the Legalize passes, which allows Legalize to be
1184 very simple (it can focus on making code legal instead of focusing on
1185 generating <em>good</em> and legal code).</p>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1186
1187<p>One important class of optimizations performed is optimizing inserted sign
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1188 and zero extension instructions. We currently use ad-hoc techniques, but
1189 could move to more rigorous techniques in the future. Here are some good
1190 papers on the subject:</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1191
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1192<p>"<a href="http://www.eecs.harvard.edu/~nr/pubs/widen-abstract.html">Widening
1193 integer arithmetic</a>"<br>
1194 Kevin Redwine and Norman Ramsey<br>
1195 International Conference on Compiler Construction (CC) 2004</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1196
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1197<p>"<a href="http://portal.acm.org/citation.cfm?doid=512529.512552">Effective
1198 sign extension elimination</a>"<br>
1199 Motohiro Kawahito, Hideaki Komatsu, and Toshio Nakatani<br>
1200 Proceedings of the ACM SIGPLAN 2002 Conference on Programming Language Design
1201 and Implementation.</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1202
1203</div>
1204
1205<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1206<h4>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1207 <a name="selectiondag_select">SelectionDAG Select Phase</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1208</h4>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1209
1210<div class="doc_text">
1211
1212<p>The Select phase is the bulk of the target-specific code for instruction
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1213 selection. This phase takes a legal SelectionDAG as input, pattern matches
1214 the instructions supported by the target to this DAG, and produces a new DAG
1215 of target code. For example, consider the following LLVM fragment:</p>
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1216
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1217<div class="doc_code">
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1218<pre>
Dan Gohmana9445e12010-03-02 01:11:08 +0000[diff] [blame]1219%t1 = fadd float %W, %X
1220%t2 = fmul float %t1, %Y
1221%t3 = fadd float %t2, %Z
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1222</pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1223</div>
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1224
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1225<p>This LLVM code corresponds to a SelectionDAG that looks basically like
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1226 this:</p>
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1227
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1228<div class="doc_code">
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1229<pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1230(fadd:f32 (fmul:f32 (fadd:f32 W, X), Y), Z)
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1231</pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1232</div>
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1233
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1234<p>If a target supports floating point multiply-and-add (FMA) operations, one of
1235 the adds can be merged with the multiply. On the PowerPC, for example, the
1236 output of the instruction selector might look like this DAG:</p>
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1237
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1238<div class="doc_code">
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1239<pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1240(FMADDS (FADDS W, X), Y, Z)
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1241</pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1242</div>
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1243
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1244<p>The <tt>FMADDS</tt> instruction is a ternary instruction that multiplies its
1245first two operands and adds the third (as single-precision floating-point
1246numbers). The <tt>FADDS</tt> instruction is a simple binary single-precision
1247add instruction. To perform this pattern match, the PowerPC backend includes
1248the following instruction definitions:</p>
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1249
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1250<div class="doc_code">
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1251<pre>
1252def FMADDS : AForm_1&lt;59, 29,
1253 (ops F4RC:$FRT, F4RC:$FRA, F4RC:$FRC, F4RC:$FRB),
1254 "fmadds $FRT, $FRA, $FRC, $FRB",
1255 [<b>(set F4RC:$FRT, (fadd (fmul F4RC:$FRA, F4RC:$FRC),
1256 F4RC:$FRB))</b>]&gt;;
1257def FADDS : AForm_2&lt;59, 21,
1258 (ops F4RC:$FRT, F4RC:$FRA, F4RC:$FRB),
1259 "fadds $FRT, $FRA, $FRB",
1260 [<b>(set F4RC:$FRT, (fadd F4RC:$FRA, F4RC:$FRB))</b>]&gt;;
1261</pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1262</div>
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1263
1264<p>The portion of the instruction definition in bold indicates the pattern used
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1265 to match the instruction. The DAG operators
1266 (like <tt>fmul</tt>/<tt>fadd</tt>) are defined in
Dan Gohman6a4824c2010-03-25 00:03:04 +0000[diff] [blame]1267 the <tt>include/llvm/Target/TargetSelectionDAG.td</tt> file. "
1268 <tt>F4RC</tt>" is the register class of the input and result values.</p>
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1269
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1270<p>The TableGen DAG instruction selector generator reads the instruction
1271 patterns in the <tt>.td</tt> file and automatically builds parts of the
1272 pattern matching code for your target. It has the following strengths:</p>
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1273
1274<ul>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1275 <li>At compiler-compiler time, it analyzes your instruction patterns and tells
1276 you if your patterns make sense or not.</li>
1277
1278 <li>It can handle arbitrary constraints on operands for the pattern match. In
1279 particular, it is straight-forward to say things like "match any immediate
1280 that is a 13-bit sign-extended value". For examples, see the
1281 <tt>immSExt16</tt> and related <tt>tblgen</tt> classes in the PowerPC
1282 backend.</li>
1283
1284 <li>It knows several important identities for the patterns defined. For
1285 example, it knows that addition is commutative, so it allows the
1286 <tt>FMADDS</tt> pattern above to match "<tt>(fadd X, (fmul Y, Z))</tt>" as
1287 well as "<tt>(fadd (fmul X, Y), Z)</tt>", without the target author having
1288 to specially handle this case.</li>
1289
1290 <li>It has a full-featured type-inferencing system. In particular, you should
1291 rarely have to explicitly tell the system what type parts of your patterns
1292 are. In the <tt>FMADDS</tt> case above, we didn't have to tell
1293 <tt>tblgen</tt> that all of the nodes in the pattern are of type 'f32'.
1294 It was able to infer and propagate this knowledge from the fact that
1295 <tt>F4RC</tt> has type 'f32'.</li>
1296
1297 <li>Targets can define their own (and rely on built-in) "pattern fragments".
1298 Pattern fragments are chunks of reusable patterns that get inlined into
1299 your patterns during compiler-compiler time. For example, the integer
1300 "<tt>(not x)</tt>" operation is actually defined as a pattern fragment
1301 that expands as "<tt>(xor x, -1)</tt>", since the SelectionDAG does not
1302 have a native '<tt>not</tt>' operation. Targets can define their own
1303 short-hand fragments as they see fit. See the definition of
1304 '<tt>not</tt>' and '<tt>ineg</tt>' for examples.</li>
1305
1306 <li>In addition to instructions, targets can specify arbitrary patterns that
1307 map to one or more instructions using the 'Pat' class. For example, the
1308 PowerPC has no way to load an arbitrary integer immediate into a register
1309 in one instruction. To tell tblgen how to do this, it defines:
1310 <br>
1311 <br>
1312<div class="doc_code">
1313<pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1314// Arbitrary immediate support. Implement in terms of LIS/ORI.
1315def : Pat&lt;(i32 imm:$imm),
1316 (ORI (LIS (HI16 imm:$imm)), (LO16 imm:$imm))&gt;;
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1317</pre>
1318</div>
1319 <br>
1320 If none of the single-instruction patterns for loading an immediate into a
1321 register match, this will be used. This rule says "match an arbitrary i32
1322 immediate, turning it into an <tt>ORI</tt> ('or a 16-bit immediate') and
1323 an <tt>LIS</tt> ('load 16-bit immediate, where the immediate is shifted to
1324 the left 16 bits') instruction". To make this work, the
1325 <tt>LO16</tt>/<tt>HI16</tt> node transformations are used to manipulate
1326 the input immediate (in this case, take the high or low 16-bits of the
1327 immediate).</li>
1328
1329 <li>While the system does automate a lot, it still allows you to write custom
1330 C++ code to match special cases if there is something that is hard to
1331 express.</li>
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1332</ul>
1333
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]1334<p>While it has many strengths, the system currently has some limitations,
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1335 primarily because it is a work in progress and is not yet finished:</p>
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1336
1337<ul>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1338 <li>Overall, there is no way to define or match SelectionDAG nodes that define
Dan Gohmane370c802009-04-22 15:55:31 +0000[diff] [blame]1339 multiple values (e.g. <tt>SMUL_LOHI</tt>, <tt>LOAD</tt>, <tt>CALL</tt>,
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1340 etc). This is the biggest reason that you currently still <em>have
1341 to</em> write custom C++ code for your instruction selector.</li>
1342
1343 <li>There is no great way to support matching complex addressing modes yet.
1344 In the future, we will extend pattern fragments to allow them to define
1345 multiple values (e.g. the four operands of the <a href="#x86_memory">X86
1346 addressing mode</a>, which are currently matched with custom C++ code).
1347 In addition, we'll extend fragments so that a fragment can match multiple
1348 different patterns.</li>
1349
1350 <li>We don't automatically infer flags like isStore/isLoad yet.</li>
1351
1352 <li>We don't automatically generate the set of supported registers and
1353 operations for the <a href="#selectiondag_legalize">Legalizer</a>
1354 yet.</li>
1355
1356 <li>We don't have a way of tying in custom legalized nodes yet.</li>
Chris Lattner7d6915c2005-10-17 04:18:41 +0000[diff] [blame]1357</ul>
Chris Lattner7a025c82005-10-16 20:02:19 +0000[diff] [blame]1358
1359<p>Despite these limitations, the instruction selector generator is still quite
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1360 useful for most of the binary and logical operations in typical instruction
1361 sets. If you run into any problems or can't figure out how to do something,
1362 please let Chris know!</p>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1363
1364</div>
1365
1366<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1367<h4>
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]1368 <a name="selectiondag_sched">SelectionDAG Scheduling and Formation Phase</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1369</h4>
Chris Lattnere35d3bb2005-10-16 00:36:38 +0000[diff] [blame]1370
1371<div class="doc_text">
1372
1373<p>The scheduling phase takes the DAG of target instructions from the selection
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1374 phase and assigns an order. The scheduler can pick an order depending on
1375 various constraints of the machines (i.e. order for minimal register pressure
1376 or try to cover instruction latencies). Once an order is established, the
1377 DAG is converted to a list
1378 of <tt><a href="#machineinstr">MachineInstr</a></tt>s and the SelectionDAG is
1379 destroyed.</p>
Chris Lattnere35d3bb2005-10-16 00:36:38 +0000[diff] [blame]1380
Jeff Cohen0b81cda2005-10-24 16:54:55 +0000[diff] [blame]1381<p>Note that this phase is logically separate from the instruction selection
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1382 phase, but is tied to it closely in the code because it operates on
1383 SelectionDAGs.</p>
Chris Lattnerc38959f2005-10-17 03:09:31 +0000[diff] [blame]1384
Chris Lattnere35d3bb2005-10-16 00:36:38 +0000[diff] [blame]1385</div>
1386
1387<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1388<h4>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1389 <a name="selectiondag_future">Future directions for the SelectionDAG</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1390</h4>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1391
1392<div class="doc_text">
1393
1394<ol>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1395 <li>Optional function-at-a-time selection.</li>
1396
1397 <li>Auto-generate entire selector from <tt>.td</tt> file.</li>
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]1398</ol>
1399
1400</div>
Reid Spencerad1f0cd2005-04-24 20:56:18 +0000[diff] [blame]1401
1402<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1403<h3>
Reid Spencerad1f0cd2005-04-24 20:56:18 +0000[diff] [blame]1404 <a name="ssamco">SSA-based Machine Code Optimizations</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1405</h3>
Reid Spencerad1f0cd2005-04-24 20:56:18 +0000[diff] [blame]1406<div class="doc_text"><p>To Be Written</p></div>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1407
Reid Spencerad1f0cd2005-04-24 20:56:18 +0000[diff] [blame]1408<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1409<h3>
Bill Wendling3fc488d2006-09-06 18:42:41 +0000[diff] [blame]1410 <a name="liveintervals">Live Intervals</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1411</h3>
Bill Wendling2f87a882006-09-04 23:35:52 +0000[diff] [blame]1412
1413<div class="doc_text">
1414
Bill Wendling3fc488d2006-09-06 18:42:41 +0000[diff] [blame]1415<p>Live Intervals are the ranges (intervals) where a variable is <i>live</i>.
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1416 They are used by some <a href="#regalloc">register allocator</a> passes to
1417 determine if two or more virtual registers which require the same physical
1418 register are live at the same point in the program (i.e., they conflict).
1419 When this situation occurs, one virtual register must be <i>spilled</i>.</p>
Bill Wendling2f87a882006-09-04 23:35:52 +0000[diff] [blame]1420
1421</div>
1422
1423<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1424<h4>
Bill Wendling2f87a882006-09-04 23:35:52 +0000[diff] [blame]1425 <a name="livevariable_analysis">Live Variable Analysis</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1426</h4>
Bill Wendling2f87a882006-09-04 23:35:52 +0000[diff] [blame]1427
1428<div class="doc_text">
1429
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1430<p>The first step in determining the live intervals of variables is to calculate
1431 the set of registers that are immediately dead after the instruction (i.e.,
1432 the instruction calculates the value, but it is never used) and the set of
1433 registers that are used by the instruction, but are never used after the
1434 instruction (i.e., they are killed). Live variable information is computed
1435 for each <i>virtual</i> register and <i>register allocatable</i> physical
1436 register in the function. This is done in a very efficient manner because it
1437 uses SSA to sparsely compute lifetime information for virtual registers
1438 (which are in SSA form) and only has to track physical registers within a
1439 block. Before register allocation, LLVM can assume that physical registers
1440 are only live within a single basic block. This allows it to do a single,
1441 local analysis to resolve physical register lifetimes within each basic
1442 block. If a physical register is not register allocatable (e.g., a stack
1443 pointer or condition codes), it is not tracked.</p>
Bill Wendling2f87a882006-09-04 23:35:52 +0000[diff] [blame]1444
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1445<p>Physical registers may be live in to or out of a function. Live in values are
1446 typically arguments in registers. Live out values are typically return values
1447 in registers. Live in values are marked as such, and are given a dummy
1448 "defining" instruction during live intervals analysis. If the last basic
1449 block of a function is a <tt>return</tt>, then it's marked as using all live
1450 out values in the function.</p>
Bill Wendling2f87a882006-09-04 23:35:52 +0000[diff] [blame]1451
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1452<p><tt>PHI</tt> nodes need to be handled specially, because the calculation of
1453 the live variable information from a depth first traversal of the CFG of the
1454 function won't guarantee that a virtual register used by the <tt>PHI</tt>
1455 node is defined before it's used. When a <tt>PHI</tt> node is encountered,
1456 only the definition is handled, because the uses will be handled in other
1457 basic blocks.</p>
Bill Wendling2f87a882006-09-04 23:35:52 +0000[diff] [blame]1458
1459<p>For each <tt>PHI</tt> node of the current basic block, we simulate an
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1460 assignment at the end of the current basic block and traverse the successor
1461 basic blocks. If a successor basic block has a <tt>PHI</tt> node and one of
1462 the <tt>PHI</tt> node's operands is coming from the current basic block, then
1463 the variable is marked as <i>alive</i> within the current basic block and all
1464 of its predecessor basic blocks, until the basic block with the defining
1465 instruction is encountered.</p>
Bill Wendling2f87a882006-09-04 23:35:52 +0000[diff] [blame]1466
1467</div>
1468
Bill Wendling3fc488d2006-09-06 18:42:41 +0000[diff] [blame]1469<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1470<h4>
Bill Wendling3fc488d2006-09-06 18:42:41 +0000[diff] [blame]1471 <a name="liveintervals_analysis">Live Intervals Analysis</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1472</h4>
Bill Wendling2f87a882006-09-04 23:35:52 +0000[diff] [blame]1473
Bill Wendling3fc488d2006-09-06 18:42:41 +0000[diff] [blame]1474<div class="doc_text">
Bill Wendling3cd5ca62006-10-11 06:30:10 +0000[diff] [blame]1475
Bill Wendling82e2eea2006-10-11 18:00:22 +0000[diff] [blame]1476<p>We now have the information available to perform the live intervals analysis
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1477 and build the live intervals themselves. We start off by numbering the basic
1478 blocks and machine instructions. We then handle the "live-in" values. These
1479 are in physical registers, so the physical register is assumed to be killed
1480 by the end of the basic block. Live intervals for virtual registers are
1481 computed for some ordering of the machine instructions <tt>[1, N]</tt>. A
1482 live interval is an interval <tt>[i, j)</tt>, where <tt>1 &lt;= i &lt;= j
1483 &lt; N</tt>, for which a variable is live.</p>
Bill Wendling3cd5ca62006-10-11 06:30:10 +0000[diff] [blame]1484
Bill Wendling82e2eea2006-10-11 18:00:22 +0000[diff] [blame]1485<p><i><b>More to come...</b></i></p>
1486
Bill Wendling3fc488d2006-09-06 18:42:41 +0000[diff] [blame]1487</div>
Bill Wendling2f87a882006-09-04 23:35:52 +0000[diff] [blame]1488
1489<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1490<h3>
Reid Spencerad1f0cd2005-04-24 20:56:18 +0000[diff] [blame]1491 <a name="regalloc">Register Allocation</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1492</h3>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1493
1494<div class="doc_text">
1495
Bill Wendling3cd5ca62006-10-11 06:30:10 +0000[diff] [blame]1496<p>The <i>Register Allocation problem</i> consists in mapping a program
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1497 <i>P<sub>v</sub></i>, that can use an unbounded number of virtual registers,
1498 to a program <i>P<sub>p</sub></i> that contains a finite (possibly small)
1499 number of physical registers. Each target architecture has a different number
1500 of physical registers. If the number of physical registers is not enough to
1501 accommodate all the virtual registers, some of them will have to be mapped
1502 into memory. These virtuals are called <i>spilled virtuals</i>.</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1503
1504</div>
1505
1506<!-- _______________________________________________________________________ -->
1507
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1508<h4>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1509 <a name="regAlloc_represent">How registers are represented in LLVM</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1510</h4>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1511
1512<div class="doc_text">
1513
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1514<p>In LLVM, physical registers are denoted by integer numbers that normally
1515 range from 1 to 1023. To see how this numbering is defined for a particular
1516 architecture, you can read the <tt>GenRegisterNames.inc</tt> file for that
1517 architecture. For instance, by
1518 inspecting <tt>lib/Target/X86/X86GenRegisterNames.inc</tt> we see that the
1519 32-bit register <tt>EAX</tt> is denoted by 15, and the MMX register
1520 <tt>MM0</tt> is mapped to 48.</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1521
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1522<p>Some architectures contain registers that share the same physical location. A
1523 notable example is the X86 platform. For instance, in the X86 architecture,
1524 the registers <tt>EAX</tt>, <tt>AX</tt> and <tt>AL</tt> share the first eight
1525 bits. These physical registers are marked as <i>aliased</i> in LLVM. Given a
1526 particular architecture, you can check which registers are aliased by
1527 inspecting its <tt>RegisterInfo.td</tt> file. Moreover, the method
1528 <tt>TargetRegisterInfo::getAliasSet(p_reg)</tt> returns an array containing
1529 all the physical registers aliased to the register <tt>p_reg</tt>.</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1530
1531<p>Physical registers, in LLVM, are grouped in <i>Register Classes</i>.
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1532 Elements in the same register class are functionally equivalent, and can be
1533 interchangeably used. Each virtual register can only be mapped to physical
1534 registers of a particular class. For instance, in the X86 architecture, some
1535 virtuals can only be allocated to 8 bit registers. A register class is
1536 described by <tt>TargetRegisterClass</tt> objects. To discover if a virtual
1537 register is compatible with a given physical, this code can be used:</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1538
1539<div class="doc_code">
1540<pre>
Jim Laskeyb744c252006-12-15 10:40:48 +0000[diff] [blame]1541bool RegMapping_Fer::compatible_class(MachineFunction &amp;mf,
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1542 unsigned v_reg,
1543 unsigned p_reg) {
Dan Gohman6f0d0242008-02-10 18:45:23 +0000[diff] [blame]1544 assert(TargetRegisterInfo::isPhysicalRegister(p_reg) &amp;&amp;
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1545 "Target register must be physical");
Chris Lattner534bcfb2007-12-31 04:16:08 +0000[diff] [blame]1546 const TargetRegisterClass *trc = mf.getRegInfo().getRegClass(v_reg);
1547 return trc-&gt;contains(p_reg);
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1548}
1549</pre>
1550</div>
1551
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1552<p>Sometimes, mostly for debugging purposes, it is useful to change the number
1553 of physical registers available in the target architecture. This must be done
1554 statically, inside the <tt>TargetRegsterInfo.td</tt> file. Just <tt>grep</tt>
1555 for <tt>RegisterClass</tt>, the last parameter of which is a list of
1556 registers. Just commenting some out is one simple way to avoid them being
1557 used. A more polite way is to explicitly exclude some registers from
Dan Gohmand2cb3d22009-07-24 00:30:09 +0000[diff] [blame]1558 the <i>allocation order</i>. See the definition of the <tt>GR8</tt> register
1559 class in <tt>lib/Target/X86/X86RegisterInfo.td</tt> for an example of this.
1560 </p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1561
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1562<p>Virtual registers are also denoted by integer numbers. Contrary to physical
Jakob Stoklund Olesen3ca21022011-01-08 23:10:59 +0000[diff] [blame]1563 registers, different virtual registers never share the same number. Whereas
1564 physical registers are statically defined in a <tt>TargetRegisterInfo.td</tt>
1565 file and cannot be created by the application developer, that is not the case
1566 with virtual registers. In order to create new virtual registers, use the
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1567 method <tt>MachineRegisterInfo::createVirtualRegister()</tt>. This method
Jakob Stoklund Olesen3ca21022011-01-08 23:10:59 +0000[diff] [blame]1568 will return a new virtual register. Use an <tt>IndexedMap&lt;Foo,
1569 VirtReg2IndexFunctor&gt;</tt> to hold information per virtual register. If you
1570 need to enumerate all virtual registers, use the function
1571 <tt>TargetRegisterInfo::index2VirtReg()</tt> to find the virtual register
1572 numbers:</p>
1573
1574<div class="doc_code">
1575<pre>
1576 for (unsigned i = 0, e = MRI->getNumVirtRegs(); i != e; ++i) {
1577 unsigned VirtReg = TargetRegisterInfo::index2VirtReg(i);
1578 stuff(VirtReg);
1579 }
1580</pre>
1581</div>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1582
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1583<p>Before register allocation, the operands of an instruction are mostly virtual
1584 registers, although physical registers may also be used. In order to check if
1585 a given machine operand is a register, use the boolean
1586 function <tt>MachineOperand::isRegister()</tt>. To obtain the integer code of
1587 a register, use <tt>MachineOperand::getReg()</tt>. An instruction may define
1588 or use a register. For instance, <tt>ADD reg:1026 := reg:1025 reg:1024</tt>
1589 defines the registers 1024, and uses registers 1025 and 1026. Given a
1590 register operand, the method <tt>MachineOperand::isUse()</tt> informs if that
1591 register is being used by the instruction. The
1592 method <tt>MachineOperand::isDef()</tt> informs if that registers is being
1593 defined.</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1594
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1595<p>We will call physical registers present in the LLVM bitcode before register
1596 allocation <i>pre-colored registers</i>. Pre-colored registers are used in
1597 many different situations, for instance, to pass parameters of functions
1598 calls, and to store results of particular instructions. There are two types
1599 of pre-colored registers: the ones <i>implicitly</i> defined, and
1600 those <i>explicitly</i> defined. Explicitly defined registers are normal
1601 operands, and can be accessed
1602 with <tt>MachineInstr::getOperand(int)::getReg()</tt>. In order to check
1603 which registers are implicitly defined by an instruction, use
1604 the <tt>TargetInstrInfo::get(opcode)::ImplicitDefs</tt>,
1605 where <tt>opcode</tt> is the opcode of the target instruction. One important
1606 difference between explicit and implicit physical registers is that the
1607 latter are defined statically for each instruction, whereas the former may
1608 vary depending on the program being compiled. For example, an instruction
1609 that represents a function call will always implicitly define or use the same
1610 set of physical registers. To read the registers implicitly used by an
1611 instruction,
1612 use <tt>TargetInstrInfo::get(opcode)::ImplicitUses</tt>. Pre-colored
1613 registers impose constraints on any register allocation algorithm. The
Bob Wilson04738682010-04-09 18:39:54 +0000[diff] [blame]1614 register allocator must make sure that none of them are overwritten by
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1615 the values of virtual registers while still alive.</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1616
1617</div>
1618
1619<!-- _______________________________________________________________________ -->
1620
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1621<h4>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1622 <a name="regAlloc_howTo">Mapping virtual registers to physical registers</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1623</h4>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1624
1625<div class="doc_text">
1626
1627<p>There are two ways to map virtual registers to physical registers (or to
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1628 memory slots). The first way, that we will call <i>direct mapping</i>, is
1629 based on the use of methods of the classes <tt>TargetRegisterInfo</tt>,
1630 and <tt>MachineOperand</tt>. The second way, that we will call <i>indirect
1631 mapping</i>, relies on the <tt>VirtRegMap</tt> class in order to insert loads
1632 and stores sending and getting values to and from memory.</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1633
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1634<p>The direct mapping provides more flexibility to the developer of the register
1635 allocator; however, it is more error prone, and demands more implementation
1636 work. Basically, the programmer will have to specify where load and store
1637 instructions should be inserted in the target function being compiled in
1638 order to get and store values in memory. To assign a physical register to a
1639 virtual register present in a given operand,
1640 use <tt>MachineOperand::setReg(p_reg)</tt>. To insert a store instruction,
Jakob Stoklund Olesen297907f2010-08-31 22:01:07 +0000[diff] [blame]1641 use <tt>TargetInstrInfo::storeRegToStackSlot(...)</tt>, and to insert a
1642 load instruction, use <tt>TargetInstrInfo::loadRegFromStackSlot</tt>.</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1643
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1644<p>The indirect mapping shields the application developer from the complexities
1645 of inserting load and store instructions. In order to map a virtual register
1646 to a physical one, use <tt>VirtRegMap::assignVirt2Phys(vreg, preg)</tt>. In
1647 order to map a certain virtual register to memory,
1648 use <tt>VirtRegMap::assignVirt2StackSlot(vreg)</tt>. This method will return
1649 the stack slot where <tt>vreg</tt>'s value will be located. If it is
1650 necessary to map another virtual register to the same stack slot,
1651 use <tt>VirtRegMap::assignVirt2StackSlot(vreg, stack_location)</tt>. One
1652 important point to consider when using the indirect mapping, is that even if
1653 a virtual register is mapped to memory, it still needs to be mapped to a
1654 physical register. This physical register is the location where the virtual
1655 register is supposed to be found before being stored or after being
1656 reloaded.</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1657
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1658<p>If the indirect strategy is used, after all the virtual registers have been
1659 mapped to physical registers or stack slots, it is necessary to use a spiller
1660 object to place load and store instructions in the code. Every virtual that
1661 has been mapped to a stack slot will be stored to memory after been defined
1662 and will be loaded before being used. The implementation of the spiller tries
1663 to recycle load/store instructions, avoiding unnecessary instructions. For an
1664 example of how to invoke the spiller,
1665 see <tt>RegAllocLinearScan::runOnMachineFunction</tt>
1666 in <tt>lib/CodeGen/RegAllocLinearScan.cpp</tt>.</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1667
1668</div>
1669
1670<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1671<h4>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1672 <a name="regAlloc_twoAddr">Handling two address instructions</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1673</h4>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1674
1675<div class="doc_text">
1676
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1677<p>With very rare exceptions (e.g., function calls), the LLVM machine code
1678 instructions are three address instructions. That is, each instruction is
1679 expected to define at most one register, and to use at most two registers.
1680 However, some architectures use two address instructions. In this case, the
1681 defined register is also one of the used register. For instance, an
1682 instruction such as <tt>ADD %EAX, %EBX</tt>, in X86 is actually equivalent
1683 to <tt>%EAX = %EAX + %EBX</tt>.</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1684
1685<p>In order to produce correct code, LLVM must convert three address
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1686 instructions that represent two address instructions into true two address
1687 instructions. LLVM provides the pass <tt>TwoAddressInstructionPass</tt> for
1688 this specific purpose. It must be run before register allocation takes
1689 place. After its execution, the resulting code may no longer be in SSA
1690 form. This happens, for instance, in situations where an instruction such
1691 as <tt>%a = ADD %b %c</tt> is converted to two instructions such as:</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1692
1693<div class="doc_code">
1694<pre>
1695%a = MOVE %b
Dan Gohman03e58572008-06-13 17:55:57 +0000[diff] [blame]1696%a = ADD %a %c
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1697</pre>
1698</div>
1699
1700<p>Notice that, internally, the second instruction is represented as
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1701 <tt>ADD %a[def/use] %c</tt>. I.e., the register operand <tt>%a</tt> is both
1702 used and defined by the instruction.</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1703
1704</div>
1705
1706<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1707<h4>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1708 <a name="regAlloc_ssaDecon">The SSA deconstruction phase</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1709</h4>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1710
1711<div class="doc_text">
1712
1713<p>An important transformation that happens during register allocation is called
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1714 the <i>SSA Deconstruction Phase</i>. The SSA form simplifies many analyses
1715 that are performed on the control flow graph of programs. However,
1716 traditional instruction sets do not implement PHI instructions. Thus, in
1717 order to generate executable code, compilers must replace PHI instructions
1718 with other instructions that preserve their semantics.</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1719
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1720<p>There are many ways in which PHI instructions can safely be removed from the
1721 target code. The most traditional PHI deconstruction algorithm replaces PHI
1722 instructions with copy instructions. That is the strategy adopted by
1723 LLVM. The SSA deconstruction algorithm is implemented
1724 in <tt>lib/CodeGen/PHIElimination.cpp</tt>. In order to invoke this pass, the
1725 identifier <tt>PHIEliminationID</tt> must be marked as required in the code
1726 of the register allocator.</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1727
1728</div>
1729
1730<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1731<h4>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1732 <a name="regAlloc_fold">Instruction folding</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1733</h4>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1734
1735<div class="doc_text">
1736
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1737<p><i>Instruction folding</i> is an optimization performed during register
1738 allocation that removes unnecessary copy instructions. For instance, a
1739 sequence of instructions such as:</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1740
1741<div class="doc_code">
1742<pre>
1743%EBX = LOAD %mem_address
1744%EAX = COPY %EBX
1745</pre>
1746</div>
1747
Dan Gohmana7ab2bf2008-11-24 16:35:31 +0000[diff] [blame]1748<p>can be safely substituted by the single instruction:</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1749
1750<div class="doc_code">
1751<pre>
1752%EAX = LOAD %mem_address
1753</pre>
1754</div>
1755
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1756<p>Instructions can be folded with
1757 the <tt>TargetRegisterInfo::foldMemoryOperand(...)</tt> method. Care must be
1758 taken when folding instructions; a folded instruction can be quite different
1759 from the original
1760 instruction. See <tt>LiveIntervals::addIntervalsForSpills</tt>
1761 in <tt>lib/CodeGen/LiveIntervalAnalysis.cpp</tt> for an example of its
1762 use.</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1763
1764</div>
1765
1766<!-- _______________________________________________________________________ -->
1767
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1768<h4>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1769 <a name="regAlloc_builtIn">Built in register allocators</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1770</h4>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1771
1772<div class="doc_text">
1773
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1774<p>The LLVM infrastructure provides the application developer with three
1775 different register allocators:</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1776
1777<ul>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1778 <li><i>Linear Scan</i> &mdash; <i>The default allocator</i>. This is the
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1779 well-know linear scan register allocator. Whereas the
1780 <i>Simple</i> and <i>Local</i> algorithms use a direct mapping
1781 implementation technique, the <i>Linear Scan</i> implementation
1782 uses a spiller in order to place load and stores.</li>
Jakob Stoklund Olesen8a3eab92010-06-15 21:58:33 +0000[diff] [blame]1783
1784 <li><i>Fast</i> &mdash; This register allocator is the default for debug
1785 builds. It allocates registers on a basic block level, attempting to keep
1786 values in registers and reusing registers as appropriate.</li>
1787
1788 <li><i>PBQP</i> &mdash; A Partitioned Boolean Quadratic Programming (PBQP)
1789 based register allocator. This allocator works by constructing a PBQP
1790 problem representing the register allocation problem under consideration,
1791 solving this using a PBQP solver, and mapping the solution back to a
1792 register assignment.</li>
1793
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1794</ul>
1795
1796<p>The type of register allocator used in <tt>llc</tt> can be chosen with the
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]1797 command line option <tt>-regalloc=...</tt>:</p>
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1798
1799<div class="doc_code">
1800<pre>
Dan Gohman0cabaa52009-08-25 15:54:01 +0000[diff] [blame]1801$ llc -regalloc=linearscan file.bc -o ln.s;
Jakob Stoklund Olesen8a3eab92010-06-15 21:58:33 +0000[diff] [blame]1802$ llc -regalloc=fast file.bc -o fa.s;
1803$ llc -regalloc=pbqp file.bc -o pbqp.s;
Bill Wendlinga396ee82006-09-01 21:46:00 +0000[diff] [blame]1804</pre>
1805</div>
1806
1807</div>
1808
Reid Spencerad1f0cd2005-04-24 20:56:18 +0000[diff] [blame]1809<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1810<h3>
Reid Spencerad1f0cd2005-04-24 20:56:18 +0000[diff] [blame]1811 <a name="proepicode">Prolog/Epilog Code Insertion</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1812</h3>
Reid Spencerad1f0cd2005-04-24 20:56:18 +0000[diff] [blame]1813<div class="doc_text"><p>To Be Written</p></div>
1814<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1815<h3>
Reid Spencerad1f0cd2005-04-24 20:56:18 +0000[diff] [blame]1816 <a name="latemco">Late Machine Code Optimizations</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1817</h3>
Reid Spencerad1f0cd2005-04-24 20:56:18 +0000[diff] [blame]1818<div class="doc_text"><p>To Be Written</p></div>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]1819
Reid Spencerad1f0cd2005-04-24 20:56:18 +0000[diff] [blame]1820<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1821<h3>
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]1822 <a name="codeemit">Code Emission</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1823</h3>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]1824
1825<div class="doc_text">
1826
1827<p>The code emission step of code generation is responsible for lowering from
1828the code generator abstractions (like <a
1829href="#machinefunction">MachineFunction</a>, <a
1830href="#machineinstr">MachineInstr</a>, etc) down
1831to the abstractions used by the MC layer (<a href="#mcinst">MCInst</a>,
1832<a href="#mcstreamer">MCStreamer</a>, etc). This is
1833done with a combination of several different classes: the (misnamed)
1834target-independent AsmPrinter class, target-specific subclasses of AsmPrinter
1835(such as SparcAsmPrinter), and the TargetLoweringObjectFile class.</p>
1836
1837<p>Since the MC layer works at the level of abstraction of object files, it
1838doesn't have a notion of functions, global variables etc. Instead, it thinks
1839about labels, directives, and instructions. A key class used at this time is
1840the MCStreamer class. This is an abstract API that is implemented in different
1841ways (e.g. to output a .s file, output an ELF .o file, etc) that is effectively
1842an "assembler API". MCStreamer has one method per directive, such as EmitLabel,
1843EmitSymbolAttribute, SwitchSection, etc, which directly correspond to assembly
1844level directives.
1845</p>
1846
1847<p>If you are interested in implementing a code generator for a target, there
1848are three important things that you have to implement for your target:</p>
1849
1850<ol>
1851<li>First, you need a subclass of AsmPrinter for your target. This class
1852implements the general lowering process converting MachineFunction's into MC
1853label constructs. The AsmPrinter base class provides a number of useful methods
1854and routines, and also allows you to override the lowering process in some
1855important ways. You should get much of the lowering for free if you are
1856implementing an ELF, COFF, or MachO target, because the TargetLoweringObjectFile
1857class implements much of the common logic.</li>
1858
1859<li>Second, you need to implement an instruction printer for your target. The
1860instruction printer takes an <a href="#mcinst">MCInst</a> and renders it to a
1861raw_ostream as text. Most of this is automatically generated from the .td file
1862(when you specify something like "<tt>add $dst, $src1, $src2</tt>" in the
1863instructions), but you need to implement routines to print operands.</li>
1864
1865<li>Third, you need to implement code that lowers a <a
1866href="#machineinstr">MachineInstr</a> to an MCInst, usually implemented in
1867"&lt;target&gt;MCInstLower.cpp". This lowering process is often target
1868specific, and is responsible for turning jump table entries, constant pool
1869indices, global variable addresses, etc into MCLabels as appropriate. This
1870translation layer is also responsible for expanding pseudo ops used by the code
1871generator into the actual machine instructions they correspond to. The MCInsts
1872that are generated by this are fed into the instruction printer or the encoder.
1873</li>
1874
1875</ol>
1876
1877<p>Finally, at your choosing, you can also implement an subclass of
1878MCCodeEmitter which lowers MCInst's into machine code bytes and relocations.
1879This is important if you want to support direct .o file emission, or would like
1880to implement an assembler for your target.</p>
1881
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]1882</div>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]1883
1884
Chris Lattner22481f22010-09-21 04:03:39 +0000[diff] [blame]1885<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1886<h2>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]1887 <a name="nativeassembler">Implementing a Native Assembler</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1888</h2>
Chris Lattner22481f22010-09-21 04:03:39 +0000[diff] [blame]1889<!-- *********************************************************************** -->
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]1890
1891<div class="doc_text">
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]1892
Chris Lattner22481f22010-09-21 04:03:39 +0000[diff] [blame]1893<p>Though you're probably reading this because you want to write or maintain a
1894compiler backend, LLVM also fully supports building a native assemblers too.
1895We've tried hard to automate the generation of the assembler from the .td files
1896(in particular the instruction syntax and encodings), which means that a large
1897part of the manual and repetitive data entry can be factored and shared with the
1898compiler.</p>
1899
Chris Lattner674c1dc2010-10-30 17:36:36 +0000[diff] [blame]1900</div>
Chris Lattner22481f22010-09-21 04:03:39 +0000[diff] [blame]1901
Chris Lattner674c1dc2010-10-30 17:36:36 +0000[diff] [blame]1902<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1903<h3 id="na_instparsing">Instruction Parsing</h3>
Chris Lattner674c1dc2010-10-30 17:36:36 +0000[diff] [blame]1904
1905<div class="doc_text"><p>To Be Written</p></div>
1906
1907
1908<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1909<h3 id="na_instaliases">
Chris Lattner674c1dc2010-10-30 17:36:36 +0000[diff] [blame]1910 Instruction Alias Processing
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1911</h3>
Chris Lattner674c1dc2010-10-30 17:36:36 +0000[diff] [blame]1912
1913<div class="doc_text">
1914<p>Once the instruction is parsed, it enters the MatchInstructionImpl function.
1915The MatchInstructionImpl function performs alias processing and then does
1916actual matching.</p>
1917
Chris Lattner693173f2010-10-30 19:23:13 +0000[diff] [blame]1918<p>Alias processing is the phase that canonicalizes different lexical forms of
Chris Lattner674c1dc2010-10-30 17:36:36 +0000[diff] [blame]1919the same instructions down to one representation. There are several different
1920kinds of alias that are possible to implement and they are listed below in the
1921order that they are processed (which is in order from simplest/weakest to most
1922complex/powerful). Generally you want to use the first alias mechanism that
1923meets the needs of your instruction, because it will allow a more concise
1924description.</p>
1925
Chris Lattner50e59722010-10-30 20:21:00 +0000[diff] [blame]1926</div>
1927
Chris Lattner674c1dc2010-10-30 17:36:36 +0000[diff] [blame]1928<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1929<h4>Mnemonic Aliases</h4>
Chris Lattner674c1dc2010-10-30 17:36:36 +0000[diff] [blame]1930
1931<div class="doc_text">
1932
Chris Lattner8cf8bcc2010-10-30 19:47:49 +0000[diff] [blame]1933<p>The first phase of alias processing is simple instruction mnemonic
Chris Lattner674c1dc2010-10-30 17:36:36 +0000[diff] [blame]1934remapping for classes of instructions which are allowed with two different
Chris Lattner693173f2010-10-30 19:23:13 +0000[diff] [blame]1935mnemonics. This phase is a simple and unconditionally remapping from one input
Chris Lattner674c1dc2010-10-30 17:36:36 +0000[diff] [blame]1936mnemonic to one output mnemonic. It isn't possible for this form of alias to
1937look at the operands at all, so the remapping must apply for all forms of a
1938given mnemonic. Mnemonic aliases are defined simply, for example X86 has:
1939</p>
1940
1941<div class="doc_code">
1942<pre>
1943def : MnemonicAlias&lt;"cbw", "cbtw"&gt;;
1944def : MnemonicAlias&lt;"smovq", "movsq"&gt;;
1945def : MnemonicAlias&lt;"fldcww", "fldcw"&gt;;
1946def : MnemonicAlias&lt;"fucompi", "fucomip"&gt;;
1947def : MnemonicAlias&lt;"ud2a", "ud2"&gt;;
1948</pre>
1949</div>
1950
1951<p>... and many others. With a MnemonicAlias definition, the mnemonic is
Chris Lattner693173f2010-10-30 19:23:13 +0000[diff] [blame]1952remapped simply and directly. Though MnemonicAlias's can't look at any aspect
1953of the instruction (such as the operands) they can depend on global modes (the
1954same ones supported by the matcher), through a Requires clause:</p>
1955
1956<div class="doc_code">
1957<pre>
1958def : MnemonicAlias&lt;"pushf", "pushfq"&gt;, Requires&lt;[In64BitMode]&gt;;
1959def : MnemonicAlias&lt;"pushf", "pushfl"&gt;, Requires&lt;[In32BitMode]&gt;;
1960</pre>
1961</div>
1962
1963<p>In this example, the mnemonic gets mapped into different a new one depending
1964on the current instruction set.</p>
Chris Lattnere1b83452010-09-11 23:02:10 +0000[diff] [blame]1965
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]1966</div>
1967
Chris Lattnerc7a03fb2010-11-06 08:30:26 +0000[diff] [blame]1968<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]1969<h4>Instruction Aliases</h4>
Chris Lattnerc7a03fb2010-11-06 08:30:26 +0000[diff] [blame]1970
1971<div class="doc_text">
1972
1973<p>The most general phase of alias processing occurs while matching is
1974happening: it provides new forms for the matcher to match along with a specific
1975instruction to generate. An instruction alias has two parts: the string to
1976match and the instruction to generate. For example:
1977</p>
1978
1979<div class="doc_code">
1980<pre>
1981def : InstAlias&lt;"movsx $src, $dst", (MOVSX16rr8W GR16:$dst, GR8 :$src)&gt;;
1982def : InstAlias&lt;"movsx $src, $dst", (MOVSX16rm8W GR16:$dst, i8mem:$src)&gt;;
1983def : InstAlias&lt;"movsx $src, $dst", (MOVSX32rr8 GR32:$dst, GR8 :$src)&gt;;
1984def : InstAlias&lt;"movsx $src, $dst", (MOVSX32rr16 GR32:$dst, GR16 :$src)&gt;;
1985def : InstAlias&lt;"movsx $src, $dst", (MOVSX64rr8 GR64:$dst, GR8 :$src)&gt;;
1986def : InstAlias&lt;"movsx $src, $dst", (MOVSX64rr16 GR64:$dst, GR16 :$src)&gt;;
1987def : InstAlias&lt;"movsx $src, $dst", (MOVSX64rr32 GR64:$dst, GR32 :$src)&gt;;
1988</pre>
1989</div>
1990
1991<p>This shows a powerful example of the instruction aliases, matching the
1992same mnemonic in multiple different ways depending on what operands are present
1993in the assembly. The result of instruction aliases can include operands in a
1994different order than the destination instruction, and can use an input
1995multiple times, for example:</p>
1996
1997<div class="doc_code">
1998<pre>
1999def : InstAlias&lt;"clrb $reg", (XOR8rr GR8 :$reg, GR8 :$reg)&gt;;
2000def : InstAlias&lt;"clrw $reg", (XOR16rr GR16:$reg, GR16:$reg)&gt;;
2001def : InstAlias&lt;"clrl $reg", (XOR32rr GR32:$reg, GR32:$reg)&gt;;
2002def : InstAlias&lt;"clrq $reg", (XOR64rr GR64:$reg, GR64:$reg)&gt;;
2003</pre>
2004</div>
2005
2006<p>This example also shows that tied operands are only listed once. In the X86
2007backend, XOR8rr has two input GR8's and one output GR8 (where an input is tied
2008to the output). InstAliases take a flattened operand list without duplicates
Chris Lattner90fd7972010-11-06 19:57:21 +0000[diff] [blame]2009for tied operands. The result of an instruction alias can also use immediates
2010and fixed physical registers which are added as simple immediate operands in the
2011result, for example:</p>
Chris Lattner98c870f2010-11-06 19:25:43 +0000[diff] [blame]2012
2013<div class="doc_code">
2014<pre>
Chris Lattner90fd7972010-11-06 19:57:21 +0000[diff] [blame]2015// Fixed Immediate operand.
Chris Lattner98c870f2010-11-06 19:25:43 +0000[diff] [blame]2016def : InstAlias&lt;"aad", (AAD8i8 10)&gt;;
Chris Lattner90fd7972010-11-06 19:57:21 +0000[diff] [blame]2017
2018// Fixed register operand.
2019def : InstAlias&lt;"fcomi", (COM_FIr ST1)&gt;;
2020
2021// Simple alias.
2022def : InstAlias&lt;"fcomi $reg", (COM_FIr RST:$reg)&gt;;
Chris Lattner98c870f2010-11-06 19:25:43 +0000[diff] [blame]2023</pre>
2024</div>
2025
Chris Lattnerc7a03fb2010-11-06 08:30:26 +0000[diff] [blame]2026
2027<p>Instruction aliases can also have a Requires clause to make them
2028subtarget specific.</p>
2029
2030</div>
2031
2032
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]2033
Chris Lattner22481f22010-09-21 04:03:39 +0000[diff] [blame]2034<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2035<h3 id="na_matching">Instruction Matching</h3>
Chris Lattner674c1dc2010-10-30 17:36:36 +0000[diff] [blame]2036
Chris Lattner22481f22010-09-21 04:03:39 +0000[diff] [blame]2037<div class="doc_text"><p>To Be Written</p></div>
2038
2039
2040
2041
Chris Lattneraa5bcb52005-01-28 17:22:53 +0000[diff] [blame]2042<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2043<h2>
Chris Lattner32e89f22005-10-16 18:31:08 +0000[diff] [blame]2044 <a name="targetimpls">Target-specific Implementation Notes</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2045</h2>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2046<!-- *********************************************************************** -->
2047
2048<div class="doc_text">
2049
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2050<p>This section of the document explains features or design decisions that are
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2051 specific to the code generator for a particular target. First we start
2052 with a table that summarizes what features are supported by each target.</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2053
2054</div>
2055
Arnold Schwaighofer9097d142008-05-14 09:17:12 +0000[diff] [blame]2056<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2057<h3>
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2058 <a name="targetfeatures">Target Feature Matrix</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2059</h3>
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2060
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2061<div class="doc_text">
2062
2063<p>Note that this table does not include the C backend or Cpp backends, since
2064they do not use the target independent code generator infrastructure. It also
2065doesn't list features that are not supported fully by any target yet. It
2066considers a feature to be supported if at least one subtarget supports it. A
2067feature being supported means that it is useful and works for most cases, it
2068does not indicate that there are zero known bugs in the implementation. Here
2069is the key:</p>
2070
2071
2072<table border="1" cellspacing="0">
2073 <tr>
2074 <th>Unknown</th>
2075 <th>No support</th>
2076 <th>Partial Support</th>
2077 <th>Complete Support</th>
2078 </tr>
2079 <tr>
2080 <td class="unknown"></td>
2081 <td class="no"></td>
2082 <td class="partial"></td>
2083 <td class="yes"></td>
2084 </tr>
2085</table>
2086
2087<p>Here is the table:</p>
2088
2089<table width="689" border="1" cellspacing="0">
2090<tr><td></td>
Benjamin Kramer943beeb2010-10-30 21:07:28 +0000[diff] [blame]2091<td colspan="13" align="center" style="background-color:#ffc">Target</td>
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2092</tr>
2093 <tr>
2094 <th>Feature</th>
2095 <th>ARM</th>
2096 <th>Alpha</th>
2097 <th>Blackfin</th>
2098 <th>CellSPU</th>
2099 <th>MBlaze</th>
2100 <th>MSP430</th>
2101 <th>Mips</th>
2102 <th>PTX</th>
2103 <th>PowerPC</th>
2104 <th>Sparc</th>
2105 <th>SystemZ</th>
2106 <th>X86</th>
2107 <th>XCore</th>
2108 </tr>
2109
2110<tr>
2111 <td><a href="#feat_reliable">is generally reliable</a></td>
2112 <td class="yes"></td> <!-- ARM -->
2113 <td class="unknown"></td> <!-- Alpha -->
Jakob Stoklund Olesen4e136122010-10-24 20:04:05 +0000[diff] [blame]2114 <td class="no"></td> <!-- Blackfin -->
Kalle Raiskila94cc4fe2010-10-25 08:57:30 +0000[diff] [blame]2115 <td class="no"></td> <!-- CellSPU -->
Wesley Peckc6a45242010-10-24 18:50:12 +0000[diff] [blame]2116 <td class="no"></td> <!-- MBlaze -->
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2117 <td class="unknown"></td> <!-- MSP430 -->
Bruno Cardoso Lopes48461f62010-12-19 22:41:43 +0000[diff] [blame]2118 <td class="no"></td> <!-- Mips -->
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2119 <td class="no"></td> <!-- PTX -->
2120 <td class="yes"></td> <!-- PowerPC -->
2121 <td class="yes"></td> <!-- Sparc -->
2122 <td class="unknown"></td> <!-- SystemZ -->
2123 <td class="yes"></td> <!-- X86 -->
2124 <td class="unknown"></td> <!-- XCore -->
2125</tr>
2126
2127<tr>
2128 <td><a href="#feat_asmparser">assembly parser</a></td>
2129 <td class="no"></td> <!-- ARM -->
2130 <td class="no"></td> <!-- Alpha -->
2131 <td class="no"></td> <!-- Blackfin -->
2132 <td class="no"></td> <!-- CellSPU -->
Wesley Peckd5fe3ef2010-12-20 21:54:50 +0000[diff] [blame]2133 <td class="yes"></td> <!-- MBlaze -->
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2134 <td class="no"></td> <!-- MSP430 -->
2135 <td class="no"></td> <!-- Mips -->
2136 <td class="no"></td> <!-- PTX -->
2137 <td class="no"></td> <!-- PowerPC -->
2138 <td class="no"></td> <!-- Sparc -->
2139 <td class="no"></td> <!-- SystemZ -->
2140 <td class="yes"></td> <!-- X86 -->
2141 <td class="no"></td> <!-- XCore -->
2142</tr>
2143
2144<tr>
2145 <td><a href="#feat_disassembler">disassembler</a></td>
2146 <td class="yes"></td> <!-- ARM -->
2147 <td class="no"></td> <!-- Alpha -->
2148 <td class="no"></td> <!-- Blackfin -->
2149 <td class="no"></td> <!-- CellSPU -->
Wesley Peckd5fe3ef2010-12-20 21:54:50 +0000[diff] [blame]2150 <td class="yes"></td> <!-- MBlaze -->
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2151 <td class="no"></td> <!-- MSP430 -->
2152 <td class="no"></td> <!-- Mips -->
2153 <td class="no"></td> <!-- PTX -->
2154 <td class="no"></td> <!-- PowerPC -->
2155 <td class="no"></td> <!-- Sparc -->
2156 <td class="no"></td> <!-- SystemZ -->
2157 <td class="yes"></td> <!-- X86 -->
2158 <td class="no"></td> <!-- XCore -->
2159</tr>
2160
2161<tr>
2162 <td><a href="#feat_inlineasm">inline asm</a></td>
2163 <td class="yes"></td> <!-- ARM -->
2164 <td class="unknown"></td> <!-- Alpha -->
Jakob Stoklund Olesen4e136122010-10-24 20:04:05 +0000[diff] [blame]2165 <td class="yes"></td> <!-- Blackfin -->
Kalle Raiskila94cc4fe2010-10-25 08:57:30 +0000[diff] [blame]2166 <td class="no"></td> <!-- CellSPU -->
Wesley Peckd5fe3ef2010-12-20 21:54:50 +0000[diff] [blame]2167 <td class="yes"></td> <!-- MBlaze -->
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2168 <td class="unknown"></td> <!-- MSP430 -->
Bruno Cardoso Lopes48461f62010-12-19 22:41:43 +0000[diff] [blame]2169 <td class="no"></td> <!-- Mips -->
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2170 <td class="unknown"></td> <!-- PTX -->
2171 <td class="yes"></td> <!-- PowerPC -->
2172 <td class="unknown"></td> <!-- Sparc -->
2173 <td class="unknown"></td> <!-- SystemZ -->
2174 <td class="yes"><a href="#feat_inlineasm_x86">*</a></td> <!-- X86 -->
2175 <td class="unknown"></td> <!-- XCore -->
2176</tr>
2177
2178<tr>
2179 <td><a href="#feat_jit">jit</a></td>
2180 <td class="partial"><a href="#feat_jit_arm">*</a></td> <!-- ARM -->
Chris Lattnerac3031a2010-11-14 18:25:50 +0000[diff] [blame]2181 <td class="no"></td> <!-- Alpha -->
Jakob Stoklund Olesen4e136122010-10-24 20:04:05 +0000[diff] [blame]2182 <td class="no"></td> <!-- Blackfin -->
Kalle Raiskila94cc4fe2010-10-25 08:57:30 +0000[diff] [blame]2183 <td class="no"></td> <!-- CellSPU -->
Wesley Peckc6a45242010-10-24 18:50:12 +0000[diff] [blame]2184 <td class="no"></td> <!-- MBlaze -->
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2185 <td class="unknown"></td> <!-- MSP430 -->
Bruno Cardoso Lopes48461f62010-12-19 22:41:43 +0000[diff] [blame]2186 <td class="no"></td> <!-- Mips -->
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2187 <td class="unknown"></td> <!-- PTX -->
2188 <td class="yes"></td> <!-- PowerPC -->
2189 <td class="unknown"></td> <!-- Sparc -->
2190 <td class="unknown"></td> <!-- SystemZ -->
2191 <td class="yes"></td> <!-- X86 -->
2192 <td class="unknown"></td> <!-- XCore -->
2193</tr>
2194
2195<tr>
2196 <td><a href="#feat_objectwrite">.o&nbsp;file writing</a></td>
2197 <td class="no"></td> <!-- ARM -->
2198 <td class="no"></td> <!-- Alpha -->
2199 <td class="no"></td> <!-- Blackfin -->
2200 <td class="no"></td> <!-- CellSPU -->
Wesley Peckd5fe3ef2010-12-20 21:54:50 +0000[diff] [blame]2201 <td class="yes"></td> <!-- MBlaze -->
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2202 <td class="no"></td> <!-- MSP430 -->
2203 <td class="no"></td> <!-- Mips -->
2204 <td class="no"></td> <!-- PTX -->
2205 <td class="no"></td> <!-- PowerPC -->
2206 <td class="no"></td> <!-- Sparc -->
2207 <td class="no"></td> <!-- SystemZ -->
2208 <td class="yes"></td> <!-- X86 -->
2209 <td class="no"></td> <!-- XCore -->
2210</tr>
2211
2212<tr>
2213 <td><a href="#feat_tailcall">tail calls</a></td>
2214 <td class="yes"></td> <!-- ARM -->
2215 <td class="unknown"></td> <!-- Alpha -->
Jakob Stoklund Olesen4e136122010-10-24 20:04:05 +0000[diff] [blame]2216 <td class="no"></td> <!-- Blackfin -->
Kalle Raiskila94cc4fe2010-10-25 08:57:30 +0000[diff] [blame]2217 <td class="no"></td> <!-- CellSPU -->
Wesley Peckc6a45242010-10-24 18:50:12 +0000[diff] [blame]2218 <td class="no"></td> <!-- MBlaze -->
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2219 <td class="unknown"></td> <!-- MSP430 -->
Bruno Cardoso Lopes48461f62010-12-19 22:41:43 +0000[diff] [blame]2220 <td class="no"></td> <!-- Mips -->
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2221 <td class="unknown"></td> <!-- PTX -->
2222 <td class="yes"></td> <!-- PowerPC -->
2223 <td class="unknown"></td> <!-- Sparc -->
2224 <td class="unknown"></td> <!-- SystemZ -->
2225 <td class="yes"></td> <!-- X86 -->
2226 <td class="unknown"></td> <!-- XCore -->
2227</tr>
2228
2229
2230</table>
2231
2232</div>
2233
2234<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2235<h4 id="feat_reliable">Is Generally Reliable</h4>
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2236
2237<div class="doc_text">
2238<p>This box indicates whether the target is considered to be production quality.
2239This indicates that the target has been used as a static compiler to
2240compile large amounts of code by a variety of different people and is in
2241continuous use.</p>
2242</div>
2243
2244<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2245<h4 id="feat_asmparser">Assembly Parser</h4>
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2246
2247<div class="doc_text">
2248<p>This box indicates whether the target supports parsing target specific .s
2249files by implementing the MCAsmParser interface. This is required for llvm-mc
2250to be able to act as a native assembler and is required for inline assembly
2251support in the native .o file writer.</p>
2252
2253</div>
2254
2255
2256<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2257<h4 id="feat_disassembler">Disassembler</h4>
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2258
2259<div class="doc_text">
2260<p>This box indicates whether the target supports the MCDisassembler API for
2261disassembling machine opcode bytes into MCInst's.</p>
2262
2263</div>
2264
2265<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2266<h4 id="feat_inlineasm">Inline Asm</h4>
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2267
2268<div class="doc_text">
2269<p>This box indicates whether the target supports most popular inline assembly
2270constraints and modifiers.</p>
2271
2272<p id="feat_inlineasm_x86">X86 lacks reliable support for inline assembly
2273constraints relating to the X86 floating point stack.</p>
2274
2275</div>
2276
2277<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2278<h4 id="feat_jit">JIT Support</h4>
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2279
2280<div class="doc_text">
2281<p>This box indicates whether the target supports the JIT compiler through
2282the ExecutionEngine interface.</p>
2283
Chris Lattner6fb99552010-10-24 16:24:22 +0000[diff] [blame]2284<p id="feat_jit_arm">The ARM backend has basic support for integer code
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2285in ARM codegen mode, but lacks NEON and full Thumb support.</p>
2286
2287</div>
2288
2289<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2290<h4 id="feat_objectwrite">.o File Writing</h4>
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2291
2292<div class="doc_text">
2293
2294<p>This box indicates whether the target supports writing .o files (e.g. MachO,
2295ELF, and/or COFF) files directly from the target. Note that the target also
2296must include an assembly parser and general inline assembly support for full
2297inline assembly support in the .o writer.</p>
2298
Chris Lattner219ddf52010-10-28 02:22:02 +0000[diff] [blame]2299<p>Targets that don't support this feature can obviously still write out .o
2300files, they just rely on having an external assembler to translate from a .s
2301file to a .o file (as is the case for many C compilers).</p>
2302
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2303</div>
2304
2305<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2306<h4 id="feat_tailcall">Tail Calls</h4>
Chris Lattner68de6022010-10-24 16:18:00 +0000[diff] [blame]2307
2308<div class="doc_text">
2309
2310<p>This box indicates whether the target supports guaranteed tail calls. These
2311are calls marked "<a href="LangRef.html#i_call">tail</a>" and use the fastcc
2312calling convention. Please see the <a href="#tailcallopt">tail call section
2313more more details</a>.</p>
2314
2315</div>
2316
2317
2318
2319
2320<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2321<h3>
Arnold Schwaighofer9097d142008-05-14 09:17:12 +0000[diff] [blame]2322 <a name="tailcallopt">Tail call optimization</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2323</h3>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2324
Arnold Schwaighofer9097d142008-05-14 09:17:12 +0000[diff] [blame]2325<div class="doc_text">
Arnold Schwaighofer9097d142008-05-14 09:17:12 +0000[diff] [blame]2326
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2327<p>Tail call optimization, callee reusing the stack of the caller, is currently
2328 supported on x86/x86-64 and PowerPC. It is performed if:</p>
2329
2330<ul>
Chris Lattner29689432010-03-11 00:22:57 +0000[diff] [blame]2331 <li>Caller and callee have the calling convention <tt>fastcc</tt> or
2332 <tt>cc 10</tt> (GHC call convention).</li>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2333
2334 <li>The call is a tail call - in tail position (ret immediately follows call
2335 and ret uses value of call or is void).</li>
2336
2337 <li>Option <tt>-tailcallopt</tt> is enabled.</li>
2338
2339 <li>Platform specific constraints are met.</li>
2340</ul>
2341
2342<p>x86/x86-64 constraints:</p>
2343
2344<ul>
2345 <li>No variable argument lists are used.</li>
2346
2347 <li>On x86-64 when generating GOT/PIC code only module-local calls (visibility
2348 = hidden or protected) are supported.</li>
2349</ul>
2350
2351<p>PowerPC constraints:</p>
2352
2353<ul>
2354 <li>No variable argument lists are used.</li>
2355
2356 <li>No byval parameters are used.</li>
2357
2358 <li>On ppc32/64 GOT/PIC only module-local calls (visibility = hidden or protected) are supported.</li>
2359</ul>
2360
2361<p>Example:</p>
2362
2363<p>Call as <tt>llc -tailcallopt test.ll</tt>.</p>
2364
2365<div class="doc_code">
2366<pre>
Arnold Schwaighofer9097d142008-05-14 09:17:12 +0000[diff] [blame]2367declare fastcc i32 @tailcallee(i32 inreg %a1, i32 inreg %a2, i32 %a3, i32 %a4)
2368
2369define fastcc i32 @tailcaller(i32 %in1, i32 %in2) {
2370 %l1 = add i32 %in1, %in2
2371 %tmp = tail call fastcc i32 @tailcallee(i32 %in1 inreg, i32 %in2 inreg, i32 %in1, i32 %l1)
2372 ret i32 %tmp
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2373}
2374</pre>
2375</div>
2376
2377<p>Implications of <tt>-tailcallopt</tt>:</p>
2378
2379<p>To support tail call optimization in situations where the callee has more
2380 arguments than the caller a 'callee pops arguments' convention is used. This
2381 currently causes each <tt>fastcc</tt> call that is not tail call optimized
2382 (because one or more of above constraints are not met) to be followed by a
2383 readjustment of the stack. So performance might be worse in such cases.</p>
2384
Arnold Schwaighofer9097d142008-05-14 09:17:12 +0000[diff] [blame]2385</div>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2386<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2387<h3>
Evan Chengdc444e92010-03-08 21:05:02 +0000[diff] [blame]2388 <a name="sibcallopt">Sibling call optimization</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2389</h3>
Evan Chengdc444e92010-03-08 21:05:02 +0000[diff] [blame]2390
2391<div class="doc_text">
2392
2393<p>Sibling call optimization is a restricted form of tail call optimization.
2394 Unlike tail call optimization described in the previous section, it can be
2395 performed automatically on any tail calls when <tt>-tailcallopt</tt> option
2396 is not specified.</p>
2397
2398<p>Sibling call optimization is currently performed on x86/x86-64 when the
2399 following constraints are met:</p>
2400
2401<ul>
2402 <li>Caller and callee have the same calling convention. It can be either
2403 <tt>c</tt> or <tt>fastcc</tt>.
2404
2405 <li>The call is a tail call - in tail position (ret immediately follows call
2406 and ret uses value of call or is void).</li>
2407
2408 <li>Caller and callee have matching return type or the callee result is not
2409 used.
2410
2411 <li>If any of the callee arguments are being passed in stack, they must be
2412 available in caller's own incoming argument stack and the frame offsets
2413 must be the same.
2414</ul>
2415
2416<p>Example:</p>
2417<div class="doc_code">
2418<pre>
2419declare i32 @bar(i32, i32)
2420
2421define i32 @foo(i32 %a, i32 %b, i32 %c) {
2422entry:
2423 %0 = tail call i32 @bar(i32 %a, i32 %b)
2424 ret i32 %0
2425}
2426</pre>
2427</div>
2428
2429</div>
2430<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2431<h3>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2432 <a name="x86">The X86 backend</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2433</h3>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2434
2435<div class="doc_text">
2436
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]2437<p>The X86 code generator lives in the <tt>lib/Target/X86</tt> directory. This
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2438 code generator is capable of targeting a variety of x86-32 and x86-64
2439 processors, and includes support for ISA extensions such as MMX and SSE.</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2440
2441</div>
2442
2443<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2444<h4>
Nate Begeman34509842009-01-26 02:54:45 +0000[diff] [blame]2445 <a name="x86_tt">X86 Target Triples supported</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2446</h4>
Chris Lattner9b988be2005-07-12 00:20:49 +0000[diff] [blame]2447
2448<div class="doc_text">
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]2449
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2450<p>The following are the known target triples that are supported by the X86
2451 backend. This is not an exhaustive list, and it would be useful to add those
2452 that people test.</p>
Chris Lattner9b988be2005-07-12 00:20:49 +0000[diff] [blame]2453
2454<ul>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2455 <li><b>i686-pc-linux-gnu</b> &mdash; Linux</li>
2456
2457 <li><b>i386-unknown-freebsd5.3</b> &mdash; FreeBSD 5.3</li>
2458
2459 <li><b>i686-pc-cygwin</b> &mdash; Cygwin on Win32</li>
2460
2461 <li><b>i686-pc-mingw32</b> &mdash; MingW on Win32</li>
2462
2463 <li><b>i386-pc-mingw32msvc</b> &mdash; MingW crosscompiler on Linux</li>
2464
2465 <li><b>i686-apple-darwin*</b> &mdash; Apple Darwin on X86</li>
Torok Edwinc457b652009-06-15 12:17:44 +0000[diff] [blame]2466
2467 <li><b>x86_64-unknown-linux-gnu</b> &mdash; Linux</li>
Chris Lattner9b988be2005-07-12 00:20:49 +0000[diff] [blame]2468</ul>
2469
2470</div>
2471
2472<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2473<h4>
Anton Korobeynikovbcb97702006-09-17 20:25:45 +0000[diff] [blame]2474 <a name="x86_cc">X86 Calling Conventions supported</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2475</h4>
Anton Korobeynikovbcb97702006-09-17 20:25:45 +0000[diff] [blame]2476
2477
2478<div class="doc_text">
2479
Dan Gohman641b2792008-11-24 16:27:17 +0000[diff] [blame]2480<p>The following target-specific calling conventions are known to backend:</p>
Anton Korobeynikovbcb97702006-09-17 20:25:45 +0000[diff] [blame]2481
2482<ul>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2483 <li><b>x86_StdCall</b> &mdash; stdcall calling convention seen on Microsoft
2484 Windows platform (CC ID = 64).</li>
2485
2486 <li><b>x86_FastCall</b> &mdash; fastcall calling convention seen on Microsoft
2487 Windows platform (CC ID = 65).</li>
Anton Korobeynikovbcb97702006-09-17 20:25:45 +0000[diff] [blame]2488</ul>
2489
2490</div>
2491
2492<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2493<h4>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2494 <a name="x86_memory">Representing X86 addressing modes in MachineInstrs</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2495</h4>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2496
2497<div class="doc_text">
2498
Misha Brukman600df452005-02-17 22:22:24 +0000[diff] [blame]2499<p>The x86 has a very flexible way of accessing memory. It is capable of
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2500 forming memory addresses of the following expression directly in integer
2501 instructions (which use ModR/M addressing):</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2502
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]2503<div class="doc_code">
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2504<pre>
Chris Lattnerb91227d2009-10-10 21:30:55 +0000[diff] [blame]2505SegmentReg: Base + [1,2,4,8] * IndexReg + Disp32
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2506</pre>
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]2507</div>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2508
Chris Lattnerb91227d2009-10-10 21:30:55 +0000[diff] [blame]2509<p>In order to represent this, LLVM tracks no less than 5 operands for each
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2510 memory operand of this form. This means that the "load" form of
2511 '<tt>mov</tt>' has the following <tt>MachineOperand</tt>s in this order:</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2512
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2513<div class="doc_code">
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2514<pre>
Chris Lattnerb91227d2009-10-10 21:30:55 +0000[diff] [blame]2515Index: 0 | 1 2 3 4 5
2516Meaning: DestReg, | BaseReg, Scale, IndexReg, Displacement Segment
2517OperandTy: VirtReg, | VirtReg, UnsImm, VirtReg, SignExtImm PhysReg
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2518</pre>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2519</div>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2520
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2521<p>Stores, and all other instructions, treat the four memory operands in the
Chris Lattnerb91227d2009-10-10 21:30:55 +0000[diff] [blame]2522 same way and in the same order. If the segment register is unspecified
2523 (regno = 0), then no segment override is generated. "Lea" operations do not
2524 have a segment register specified, so they only have 4 operands for their
2525 memory reference.</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2526
2527</div>
2528
2529<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2530<h4>
Nate Begeman34509842009-01-26 02:54:45 +0000[diff] [blame]2531 <a name="x86_memory">X86 address spaces supported</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2532</h4>
Nate Begeman34509842009-01-26 02:54:45 +0000[diff] [blame]2533
2534<div class="doc_text">
2535
Jay Foadcb88ec32011-04-06 07:55:30 +0000[diff] [blame]2536<p>x86 has a feature which provides
Dan Gohmand26795a2009-05-05 20:48:47 +0000[diff] [blame]2537 the ability to perform loads and stores to different address spaces
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2538 via the x86 segment registers. A segment override prefix byte on an
2539 instruction causes the instruction's memory access to go to the specified
2540 segment. LLVM address space 0 is the default address space, which includes
2541 the stack, and any unqualified memory accesses in a program. Address spaces
2542 1-255 are currently reserved for user-defined code. The GS-segment is
Chris Lattner1777d0c2009-05-05 18:52:19 +0000[diff] [blame]2543 represented by address space 256, while the FS-segment is represented by
2544 address space 257. Other x86 segments have yet to be allocated address space
2545 numbers.</p>
Nate Begeman34509842009-01-26 02:54:45 +0000[diff] [blame]2546
Dan Gohmand26795a2009-05-05 20:48:47 +0000[diff] [blame]2547<p>While these address spaces may seem similar to TLS via the
2548 <tt>thread_local</tt> keyword, and often use the same underlying hardware,
2549 there are some fundamental differences.</p>
2550
2551<p>The <tt>thread_local</tt> keyword applies to global variables and
2552 specifies that they are to be allocated in thread-local memory. There are
2553 no type qualifiers involved, and these variables can be pointed to with
2554 normal pointers and accessed with normal loads and stores.
2555 The <tt>thread_local</tt> keyword is target-independent at the LLVM IR
2556 level (though LLVM doesn't yet have implementations of it for some
2557 configurations).<p>
2558
2559<p>Special address spaces, in contrast, apply to static types. Every
2560 load and store has a particular address space in its address operand type,
2561 and this is what determines which address space is accessed.
2562 LLVM ignores these special address space qualifiers on global variables,
2563 and does not provide a way to directly allocate storage in them.
2564 At the LLVM IR level, the behavior of these special address spaces depends
2565 in part on the underlying OS or runtime environment, and they are specific
2566 to x86 (and LLVM doesn't yet handle them correctly in some cases).</p>
2567
2568<p>Some operating systems and runtime environments use (or may in the future
2569 use) the FS/GS-segment registers for various low-level purposes, so care
2570 should be taken when considering them.</p>
Nate Begeman34509842009-01-26 02:54:45 +0000[diff] [blame]2571
2572</div>
2573
2574<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2575<h4>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2576 <a name="x86_names">Instruction naming</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2577</h4>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2578
2579<div class="doc_text">
2580
Bill Wendling91e10c42006-08-28 02:26:32 +0000[diff] [blame]2581<p>An instruction name consists of the base name, a default operand size, and a
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2582 a character per operand with an optional special size. For example:</p>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2583
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2584<div class="doc_code">
2585<pre>
2586ADD8rr -&gt; add, 8-bit register, 8-bit register
2587IMUL16rmi -&gt; imul, 16-bit register, 16-bit memory, 16-bit immediate
2588IMUL16rmi8 -&gt; imul, 16-bit register, 16-bit memory, 8-bit immediate
2589MOVSX32rm16 -&gt; movsx, 32-bit register, 16-bit memory
2590</pre>
2591</div>
Chris Lattnerec94f802004-06-04 00:16:02 +0000[diff] [blame]2592
2593</div>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]2594
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2595<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2596<h3>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2597 <a name="ppc">The PowerPC backend</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2598</h3>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2599
2600<div class="doc_text">
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2601
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2602<p>The PowerPC code generator lives in the lib/Target/PowerPC directory. The
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2603 code generation is retargetable to several variations or <i>subtargets</i> of
2604 the PowerPC ISA; including ppc32, ppc64 and altivec.</p>
2605
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2606</div>
2607
2608<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2609<h4>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2610 <a name="ppc_abi">LLVM PowerPC ABI</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2611</h4>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2612
2613<div class="doc_text">
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2614
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2615<p>LLVM follows the AIX PowerPC ABI, with two deviations. LLVM uses a PC
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2616 relative (PIC) or static addressing for accessing global values, so no TOC
2617 (r2) is used. Second, r31 is used as a frame pointer to allow dynamic growth
2618 of a stack frame. LLVM takes advantage of having no TOC to provide space to
2619 save the frame pointer in the PowerPC linkage area of the caller frame.
2620 Other details of PowerPC ABI can be found at <a href=
2621 "http://developer.apple.com/documentation/DeveloperTools/Conceptual/LowLevelABI/Articles/32bitPowerPC.html"
2622 >PowerPC ABI.</a> Note: This link describes the 32 bit ABI. The 64 bit ABI
2623 is similar except space for GPRs are 8 bytes wide (not 4) and r13 is reserved
2624 for system use.</p>
2625
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2626</div>
2627
2628<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2629<h4>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2630 <a name="ppc_frame">Frame Layout</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2631</h4>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2632
2633<div class="doc_text">
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2634
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2635<p>The size of a PowerPC frame is usually fixed for the duration of a
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2636 function's invocation. Since the frame is fixed size, all references
2637 into the frame can be accessed via fixed offsets from the stack pointer. The
2638 exception to this is when dynamic alloca or variable sized arrays are
2639 present, then a base pointer (r31) is used as a proxy for the stack pointer
2640 and stack pointer is free to grow or shrink. A base pointer is also used if
2641 llvm-gcc is not passed the -fomit-frame-pointer flag. The stack pointer is
2642 always aligned to 16 bytes, so that space allocated for altivec vectors will
2643 be properly aligned.</p>
2644
Dan Gohman641b2792008-11-24 16:27:17 +0000[diff] [blame]2645<p>An invocation frame is laid out as follows (low memory at top);</p>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2646
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2647<table class="layout">
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2648 <tr>
2649 <td>Linkage<br><br></td>
2650 </tr>
2651 <tr>
2652 <td>Parameter area<br><br></td>
2653 </tr>
2654 <tr>
2655 <td>Dynamic area<br><br></td>
2656 </tr>
2657 <tr>
2658 <td>Locals area<br><br></td>
2659 </tr>
2660 <tr>
2661 <td>Saved registers area<br><br></td>
2662 </tr>
2663 <tr style="border-style: none hidden none hidden;">
2664 <td><br></td>
2665 </tr>
2666 <tr>
2667 <td>Previous Frame<br><br></td>
2668 </tr>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2669</table>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2670
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2671<p>The <i>linkage</i> area is used by a callee to save special registers prior
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2672 to allocating its own frame. Only three entries are relevant to LLVM. The
2673 first entry is the previous stack pointer (sp), aka link. This allows
2674 probing tools like gdb or exception handlers to quickly scan the frames in
2675 the stack. A function epilog can also use the link to pop the frame from the
2676 stack. The third entry in the linkage area is used to save the return
2677 address from the lr register. Finally, as mentioned above, the last entry is
2678 used to save the previous frame pointer (r31.) The entries in the linkage
2679 area are the size of a GPR, thus the linkage area is 24 bytes long in 32 bit
2680 mode and 48 bytes in 64 bit mode.</p>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2681
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2682<p>32 bit linkage area</p>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2683
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2684<table class="layout">
2685 <tr>
2686 <td>0</td>
2687 <td>Saved SP (r1)</td>
2688 </tr>
2689 <tr>
2690 <td>4</td>
2691 <td>Saved CR</td>
2692 </tr>
2693 <tr>
2694 <td>8</td>
2695 <td>Saved LR</td>
2696 </tr>
2697 <tr>
2698 <td>12</td>
2699 <td>Reserved</td>
2700 </tr>
2701 <tr>
2702 <td>16</td>
2703 <td>Reserved</td>
2704 </tr>
2705 <tr>
2706 <td>20</td>
2707 <td>Saved FP (r31)</td>
2708 </tr>
2709</table>
2710
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2711<p>64 bit linkage area</p>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2712
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2713<table class="layout">
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2714 <tr>
2715 <td>0</td>
2716 <td>Saved SP (r1)</td>
2717 </tr>
2718 <tr>
2719 <td>8</td>
2720 <td>Saved CR</td>
2721 </tr>
2722 <tr>
2723 <td>16</td>
2724 <td>Saved LR</td>
2725 </tr>
2726 <tr>
2727 <td>24</td>
2728 <td>Reserved</td>
2729 </tr>
2730 <tr>
2731 <td>32</td>
2732 <td>Reserved</td>
2733 </tr>
2734 <tr>
2735 <td>40</td>
2736 <td>Saved FP (r31)</td>
2737 </tr>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2738</table>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2739
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2740<p>The <i>parameter area</i> is used to store arguments being passed to a callee
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2741 function. Following the PowerPC ABI, the first few arguments are actually
2742 passed in registers, with the space in the parameter area unused. However,
2743 if there are not enough registers or the callee is a thunk or vararg
2744 function, these register arguments can be spilled into the parameter area.
2745 Thus, the parameter area must be large enough to store all the parameters for
2746 the largest call sequence made by the caller. The size must also be
2747 minimally large enough to spill registers r3-r10. This allows callees blind
2748 to the call signature, such as thunks and vararg functions, enough space to
2749 cache the argument registers. Therefore, the parameter area is minimally 32
2750 bytes (64 bytes in 64 bit mode.) Also note that since the parameter area is
2751 a fixed offset from the top of the frame, that a callee can access its spilt
2752 arguments using fixed offsets from the stack pointer (or base pointer.)</p>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2753
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2754<p>Combining the information about the linkage, parameter areas and alignment. A
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2755 stack frame is minimally 64 bytes in 32 bit mode and 128 bytes in 64 bit
2756 mode.</p>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2757
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2758<p>The <i>dynamic area</i> starts out as size zero. If a function uses dynamic
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2759 alloca then space is added to the stack, the linkage and parameter areas are
2760 shifted to top of stack, and the new space is available immediately below the
2761 linkage and parameter areas. The cost of shifting the linkage and parameter
2762 areas is minor since only the link value needs to be copied. The link value
2763 can be easily fetched by adding the original frame size to the base pointer.
2764 Note that allocations in the dynamic space need to observe 16 byte
2765 alignment.</p>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2766
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2767<p>The <i>locals area</i> is where the llvm compiler reserves space for local
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2768 variables.</p>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2769
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2770<p>The <i>saved registers area</i> is where the llvm compiler spills callee
2771 saved registers on entry to the callee.</p>
2772
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2773</div>
2774
2775<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2776<h4>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2777 <a name="ppc_prolog">Prolog/Epilog</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2778</h4>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2779
2780<div class="doc_text">
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2781
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2782<p>The llvm prolog and epilog are the same as described in the PowerPC ABI, with
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2783 the following exceptions. Callee saved registers are spilled after the frame
2784 is created. This allows the llvm epilog/prolog support to be common with
2785 other targets. The base pointer callee saved register r31 is saved in the
2786 TOC slot of linkage area. This simplifies allocation of space for the base
2787 pointer and makes it convenient to locate programatically and during
2788 debugging.</p>
2789
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2790</div>
2791
2792<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2793<h4>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2794 <a name="ppc_dynamic">Dynamic Allocation</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000[diff] [blame]2795</h4>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2796
2797<div class="doc_text">
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2798
Jim Laskeyb744c252006-12-15 10:40:48 +0000[diff] [blame]2799<p><i>TODO - More to come.</i></p>
Bill Wendling80118802009-04-15 02:12:37 +0000[diff] [blame]2800
Jim Laskeyb744c252006-12-15 10:40:48 +0000[diff] [blame]2801</div>
Jim Laskey762b6cb2006-12-14 17:19:50 +0000[diff] [blame]2802
2803
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]2804<!-- *********************************************************************** -->
2805<hr>
2806<address>
2807 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
Misha Brukman44408702008-12-11 17:34:48 +0000[diff] [blame]2808 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]2809 <a href="http://validator.w3.org/check/referer"><img
Misha Brukmanf00ddb02008-12-11 18:23:24 +0000[diff] [blame]2810 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]2811
2812 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
NAKAMURA Takumib9a33632011-04-09 02:13:37 +0000[diff] [blame]2813 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
Chris Lattnerce52b7e2004-06-01 06:48:00 +0000[diff] [blame]2814 Last modified: $Date$
2815</address>
2816
2817</body>
2818</html>