Chris Lattner | ce52b7e | 2004-06-01 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> |
| 5 | <title>The LLVM Target-Independent Code Generator</title> |
| 6 | <link rel="stylesheet" href="llvm.css" type="text/css"> |
| 7 | </head> |
| 8 | <body> |
| 9 | |
| 10 | <div class="doc_title"> |
| 11 | The LLVM Target-Independent Code Generator |
| 12 | </div> |
| 13 | |
| 14 | <ol> |
| 15 | <li><a href="#introduction">Introduction</a> |
| 16 | <ul> |
| 17 | <li><a href="#required">Required components in the code generator</a></li> |
| 18 | <li><a href="#high-level-design">The high-level design of the code generator</a></li> |
| 19 | <li><a href="#tablegen">Using TableGen for target description</a></li> |
| 20 | </ul> |
| 21 | </li> |
| 22 | <li><a href="#targetdesc">Target description classes</a> |
| 23 | <ul> |
| 24 | <li><a href="#targetmachine">The <tt>TargetMachine</tt> class</a></li> |
| 25 | <li><a href="#targetdata">The <tt>TargetData</tt> class</a></li> |
| 26 | <li><a href="#mregisterinfo">The <tt>MRegisterInfo</tt> class</a></li> |
| 27 | <li><a href="#targetinstrinfo">The <tt>TargetInstrInfo</tt> class</a></li> |
| 28 | <li><a href="#targetframeinfo">The <tt>TargetFrameInfo</tt> class</a></li> |
| 29 | <li><a href="#targetjitinfo">The <tt>TargetJITInfo</tt> class</a></li> |
| 30 | </ul> |
| 31 | </li> |
| 32 | <li><a href="#codegendesc">Machine code description classes</a> |
Chris Lattner | ec94f80 | 2004-06-04 00:16:02 +0000 | [diff] [blame^] | 33 | <ul> |
| 34 | <li><a href="#machineinstr">The <tt>MachineInstr</tt> class</a></li> |
| 35 | </ul> |
Chris Lattner | ce52b7e | 2004-06-01 06:48:00 +0000 | [diff] [blame] | 36 | </li> |
| 37 | <li><a href="#codegenalgs">Target-independent code generation algorithms</a> |
| 38 | </li> |
| 39 | <li><a href="#targetimpls">Target description implementations</a> |
| 40 | <ul> |
| 41 | <li><a href="#x86">The X86 backend</a></li> |
Chris Lattner | 10d6800 | 2004-06-01 17:18:11 +0000 | [diff] [blame] | 42 | </ul> |
Chris Lattner | ce52b7e | 2004-06-01 06:48:00 +0000 | [diff] [blame] | 43 | </li> |
| 44 | |
| 45 | </ol> |
| 46 | |
| 47 | <div class="doc_author"> |
| 48 | <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p> |
| 49 | </div> |
| 50 | |
Chris Lattner | 10d6800 | 2004-06-01 17:18:11 +0000 | [diff] [blame] | 51 | <div class="doc_warning"> |
| 52 | <p>Warning: This is a work in progress.</p> |
| 53 | </div> |
| 54 | |
Chris Lattner | ce52b7e | 2004-06-01 06:48:00 +0000 | [diff] [blame] | 55 | <!-- *********************************************************************** --> |
| 56 | <div class="doc_section"> |
| 57 | <a name="introduction">Introduction</a> |
| 58 | </div> |
| 59 | <!-- *********************************************************************** --> |
| 60 | |
| 61 | <div class="doc_text"> |
| 62 | |
| 63 | <p>The LLVM target-independent code generator is a framework that provides a |
| 64 | suite of reusable components for translating the LLVM internal representation to |
| 65 | the machine code for a specified target -- either in assembly form (suitable for |
| 66 | a static compiler) or in binary machine code format (usable for a JIT compiler). |
Chris Lattner | ec94f80 | 2004-06-04 00:16:02 +0000 | [diff] [blame^] | 67 | The LLVM target-independent code generator consists of five main components:</p> |
Chris Lattner | ce52b7e | 2004-06-01 06:48:00 +0000 | [diff] [blame] | 68 | |
| 69 | <ol> |
| 70 | <li><a href="#targetdesc">Abstract target description</a> interfaces which |
| 71 | capture improtant properties about various aspects of the machine independently |
| 72 | of how they will be used. These interfaces are defined in |
| 73 | <tt>include/llvm/Target/</tt>.</li> |
| 74 | |
| 75 | <li>Classes used to represent the <a href="#codegendesc">machine code</a> being |
| 76 | generator for a target. These classes are intended to be abstract enough to |
| 77 | represent the machine code for <i>any</i> target machine. These classes are |
| 78 | defined in <tt>include/llvm/CodeGen/</tt>.</li> |
| 79 | |
| 80 | <li><a href="#codegenalgs">Target-independent algorithms</a> used to implement |
| 81 | various phases of native code generation (register allocation, scheduling, stack |
| 82 | frame representation, etc). This code lives in <tt>lib/CodeGen/</tt>.</li> |
| 83 | |
| 84 | <li><a href="#targetimpls">Implementations of the abstract target description |
| 85 | interfaces</a> for particular targets. These machine descriptions make use of |
| 86 | the components provided by LLVM, and can optionally provide custom |
| 87 | target-specific passes, to build complete code generators for a specific target. |
| 88 | Target descriptions live in <tt>lib/Target/</tt>.</li> |
| 89 | |
Chris Lattner | ec94f80 | 2004-06-04 00:16:02 +0000 | [diff] [blame^] | 90 | <li><a href="#jit">The target-independent JIT components</a>. The LLVM JIT is |
| 91 | completely target independent (it uses the <tt>TargetJITInfo</tt> structure to |
| 92 | interface for target-specific issues. The code for the target-independent |
| 93 | JIT lives in <tt>lib/ExecutionEngine/JIT</tt>.</li> |
| 94 | |
Chris Lattner | ce52b7e | 2004-06-01 06:48:00 +0000 | [diff] [blame] | 95 | </ol> |
| 96 | |
| 97 | <p> |
| 98 | Depending on which part of the code generator you are interested in working on, |
| 99 | different pieces of this will be useful to you. In any case, you should be |
| 100 | familiar with the <a href="#targetdesc">target description</a> and <a |
| 101 | href="#codegendesc">machine code representation</a> classes. If you want to add |
| 102 | a backend for a new target, you will need <a href="#targetimpls">implement the |
| 103 | targe description</a> classes for your new target and understand the <a |
| 104 | href="LangRef.html">LLVM code representation</a>. If you are interested in |
| 105 | implementing a new <a href="#codegenalgs">code generation algorithm</a>, it |
| 106 | should only depend on the target-description and machine code representation |
| 107 | classes, ensuring that it is portable. |
| 108 | </p> |
| 109 | |
| 110 | </div> |
| 111 | |
| 112 | <!-- ======================================================================= --> |
| 113 | <div class="doc_subsection"> |
| 114 | <a name="required">Required components in the code generator</a> |
| 115 | </div> |
| 116 | |
| 117 | <div class="doc_text"> |
| 118 | |
| 119 | <p>The two pieces of the LLVM code generator are the high-level interface to the |
| 120 | code generator and the set of reusable components that can be used to build |
| 121 | target-specific backends. The two most important interfaces (<a |
| 122 | href="#targetmachine"><tt>TargetMachine</tt></a> and <a |
| 123 | href="#targetdata"><tt>TargetData</tt></a> classes) are the only ones that are |
| 124 | required to be defined for a backend to fit into the LLVM system, but the others |
| 125 | must be defined if the reusable code generator components are going to be |
| 126 | used.</p> |
| 127 | |
| 128 | <p>This design has two important implications. The first is that LLVM can |
| 129 | support completely non-traditional code generation targets. For example, the C |
| 130 | backend does not require register allocation, instruction selection, or any of |
| 131 | the other standard components provided by the system. As such, it only |
| 132 | implements these two interfaces, and does its own thing. Another example of a |
| 133 | code generator like this is a (purely hypothetical) backend that converts LLVM |
| 134 | to the GCC RTL form and uses GCC to emit machine code for a target.</p> |
| 135 | |
| 136 | <p>The other implication of this design is that it is possible to design and |
| 137 | implement radically different code generators in the LLVM system that do not |
| 138 | make use of any of the built-in components. Doing so is not recommended at all, |
| 139 | but could be required for radically different targets that do not fit into the |
| 140 | LLVM machine description model: programmable FPGAs for example.</p> |
Chris Lattner | 900bf8c | 2004-06-02 07:06:06 +0000 | [diff] [blame] | 141 | |
| 142 | <p><b>Important Note:</b> For historical reasons, the LLVM SparcV9 code |
| 143 | generator uses almost entirely different code paths than described in this |
| 144 | document. For this reason, there are some deprecated interfaces (such as |
| 145 | <tt>TargetRegInfo</tt> and <tt>TargetSchedInfo</tt>), which are only used by the |
| 146 | V9 backend and should not be used by any other targets. Also, all code in the |
| 147 | <tt>lib/Target/SparcV9</tt> directory and subdirectories should be considered |
| 148 | deprecated, and should not be used as the basis for future code generator work. |
Misha Brukman | f3709d6 | 2004-06-03 16:55:57 +0000 | [diff] [blame] | 149 | The SparcV9 backend is slowly being merged into the rest of the |
| 150 | target-independent code generators, but this is a low-priority process with no |
Chris Lattner | 900bf8c | 2004-06-02 07:06:06 +0000 | [diff] [blame] | 151 | predictable completion date.</p> |
| 152 | |
Chris Lattner | ce52b7e | 2004-06-01 06:48:00 +0000 | [diff] [blame] | 153 | </div> |
| 154 | |
| 155 | <!-- ======================================================================= --> |
| 156 | <div class="doc_subsection"> |
Chris Lattner | 10d6800 | 2004-06-01 17:18:11 +0000 | [diff] [blame] | 157 | <a name="high-level-design">The high-level design of the code generator</a> |
Chris Lattner | ce52b7e | 2004-06-01 06:48:00 +0000 | [diff] [blame] | 158 | </div> |
| 159 | |
| 160 | <div class="doc_text"> |
| 161 | |
| 162 | <p>The LLVM target-indendent code generator is designed to support efficient and |
| 163 | quality code generation for standard register-based microprocessors. Code |
| 164 | generation in this model is divided into the following stages:</p> |
| 165 | |
| 166 | <ol> |
| 167 | <li><b>Instruction Selection</b> - Determining a efficient implementation of the |
| 168 | input LLVM code in the target instruction set. This stage produces the initial |
| 169 | code for the program in the target instruction set the makes use of virtual |
| 170 | registers in SSA form and physical registers that represent any required |
| 171 | register assignments due to target constraints or calling conventions.</li> |
| 172 | |
| 173 | <li><b>SSA-based Machine Code Optimizations</b> - This (optional) stage consists |
| 174 | of a series of machine-code optimizations that operate on the SSA-form produced |
| 175 | by the instruction selector. Optimizations like modulo-scheduling, normal |
| 176 | scheduling, or peephole optimization work here.</li> |
| 177 | |
| 178 | <li><b>Register Allocation</b> - The target code is transformed from an infinite |
| 179 | virtual register file in SSA form to the concrete register file used by the |
| 180 | target. This phase introduces spill code and eliminates all virtual register |
| 181 | references from the program.</li> |
| 182 | |
| 183 | <li><b>Prolog/Epilog Code Insertion</b> - Once the machine code has been |
| 184 | generated for the function and the amount of stack space required is known (used |
| 185 | for LLVM alloca's and spill slots), the prolog and epilog code for the function |
| 186 | can be inserted and "abstract stack location references" can be eliminated. |
| 187 | This stage is responsible for implementing optimizations like frame-pointer |
| 188 | elimination and stack packing.</li> |
| 189 | |
| 190 | <li><b>Late Machine Code Optimizations</b> - Optimizations that operate on |
| 191 | "final" machine code can go here, such as spill code scheduling and peephole |
| 192 | optimizations.</li> |
| 193 | |
| 194 | <li><b>Code Emission</b> - The final stage actually outputs the machine code for |
| 195 | the current function, either in the target assembler format or in machine |
| 196 | code.</li> |
| 197 | |
| 198 | </ol> |
| 199 | |
| 200 | <p> |
| 201 | The code generator is based on the assumption that the instruction selector will |
| 202 | use an optimal pattern matching selector to create high-quality sequences of |
| 203 | native code. Alternative code generator designs based on pattern expansion and |
| 204 | aggressive iterative peephole optimization are much slower. This design is |
| 205 | designed to permit efficient compilation (important for JIT environments) and |
| 206 | aggressive optimization (used when generate code offline) by allowing components |
| 207 | of varying levels of sophisication to be used for any step of compilation.</p> |
| 208 | |
| 209 | <p> |
| 210 | In addition to these stages, target implementations can insert arbitrary |
| 211 | target-specific passes into the flow. For example, the X86 target uses a |
| 212 | special pass to handle the 80x87 floating point stack architecture. Other |
| 213 | targets with unusual requirements can be supported with custom passes as needed. |
| 214 | </p> |
| 215 | |
| 216 | </div> |
| 217 | |
| 218 | |
| 219 | <!-- ======================================================================= --> |
| 220 | <div class="doc_subsection"> |
Chris Lattner | 10d6800 | 2004-06-01 17:18:11 +0000 | [diff] [blame] | 221 | <a name="tablegen">Using TableGen for target description</a> |
Chris Lattner | ce52b7e | 2004-06-01 06:48:00 +0000 | [diff] [blame] | 222 | </div> |
| 223 | |
| 224 | <div class="doc_text"> |
| 225 | |
Chris Lattner | 5489e93 | 2004-06-01 18:35:00 +0000 | [diff] [blame] | 226 | <p>The target description classes require a detailed description of the target |
Chris Lattner | ce52b7e | 2004-06-01 06:48:00 +0000 | [diff] [blame] | 227 | architecture. These target descriptions often have a large amount of common |
| 228 | information (e.g., an add instruction is almost identical to a sub instruction). |
| 229 | In order to allow the maximum amount of commonality to be factored out, the LLVM |
| 230 | code generator uses the <a href="TableGenFundamentals.html">TableGen</a> tool to |
Chris Lattner | 5489e93 | 2004-06-01 18:35:00 +0000 | [diff] [blame] | 231 | describe big chunks of the target machine, which allows the use of domain- and |
| 232 | target-specific abstractions to reduce the amount of repetition. |
Chris Lattner | ce52b7e | 2004-06-01 06:48:00 +0000 | [diff] [blame] | 233 | </p> |
| 234 | |
| 235 | </div> |
| 236 | |
| 237 | <!-- *********************************************************************** --> |
| 238 | <div class="doc_section"> |
| 239 | <a name="targetdesc">Target description classes</a> |
| 240 | </div> |
| 241 | <!-- *********************************************************************** --> |
| 242 | |
| 243 | <div class="doc_text"> |
| 244 | |
| 245 | <p>The LLVM target description classes (which are located in the |
| 246 | <tt>include/llvm/Target</tt> directory) provide an abstract description of the |
| 247 | target machine, independent of any particular client. These classes are |
| 248 | designed to capture the <i>abstract</i> properties of the target (such as what |
| 249 | instruction and registers it has), and do not incorporate any particular pieces |
| 250 | of code generation algorithms (these interfaces do not take interference graphs |
| 251 | as inputs or other algorithm-specific data structures).</p> |
| 252 | |
| 253 | <p>All of the target description classes (except the <tt><a |
| 254 | href="#targetdata">TargetData</a></tt> class) are designed to be subclassed by |
| 255 | the concrete target implementation, and have virtual methods implemented. To |
| 256 | get to these implementations, <tt><a |
| 257 | href="#targetmachine">TargetMachine</a></tt> class provides accessors that |
| 258 | should be implemented by the target.</p> |
| 259 | |
| 260 | </div> |
| 261 | |
| 262 | <!-- ======================================================================= --> |
| 263 | <div class="doc_subsection"> |
| 264 | <a name="targetmachine">The <tt>TargetMachine</tt> class</a> |
| 265 | </div> |
| 266 | |
| 267 | <div class="doc_text"> |
| 268 | |
| 269 | <p>The <tt>TargetMachine</tt> class provides virtual methods that are used to |
| 270 | access the target-specific implementations of the various target description |
| 271 | classes (with the <tt>getInstrInfo</tt>, <tt>getRegisterInfo</tt>, |
| 272 | <tt>getFrameInfo</tt>, ... methods). This class is designed to be subclassed by |
| 273 | a concrete target implementation (e.g., <tt>X86TargetMachine</tt>) which |
| 274 | implements the various virtual methods. The only required target description |
| 275 | class is the <a href="#targetdata"><tt>TargetData</tt></a> class, but if the |
| 276 | code generator components are to be used, the other interfaces should be |
| 277 | implemented as well.</p> |
| 278 | |
| 279 | </div> |
| 280 | |
| 281 | |
| 282 | <!-- ======================================================================= --> |
| 283 | <div class="doc_subsection"> |
| 284 | <a name="targetdata">The <tt>TargetData</tt> class</a> |
| 285 | </div> |
| 286 | |
| 287 | <div class="doc_text"> |
| 288 | |
| 289 | <p>The <tt>TargetData</tt> class is the only required target description class, |
| 290 | and it is the only class that is not extensible (it cannot be derived from). It |
| 291 | specifies information about how the target lays out memory for structures, the |
| 292 | alignment requirements for various data types, the size of pointers in the |
| 293 | target, and whether the target is little- or big-endian.</p> |
| 294 | |
| 295 | </div> |
| 296 | |
| 297 | |
| 298 | <!-- ======================================================================= --> |
| 299 | <div class="doc_subsection"> |
Chris Lattner | 10d6800 | 2004-06-01 17:18:11 +0000 | [diff] [blame] | 300 | <a name="mregisterinfo">The <tt>MRegisterInfo</tt> class</a> |
Chris Lattner | ce52b7e | 2004-06-01 06:48:00 +0000 | [diff] [blame] | 301 | </div> |
| 302 | |
| 303 | <div class="doc_text"> |
| 304 | |
| 305 | <p>The <tt>MRegisterInfo</tt> class (which will eventually be renamed to |
| 306 | <tt>TargetRegisterInfo</tt>) is used to describe the register file of the |
| 307 | target and any interactions between the registers.</p> |
| 308 | |
| 309 | <p>Registers in the code generator are represented in the code generator by |
| 310 | unsigned numbers. Physical registers (those that actually exist in the target |
| 311 | description) are unique small numbers, and virtual registers are generally |
| 312 | large.</p> |
| 313 | |
| 314 | <p>Each register in the processor description has an associated |
| 315 | <tt>MRegisterDesc</tt> entry, which provides a textual name for the register |
| 316 | (used for assembly output and debugging dumps), a set of aliases (used to |
| 317 | indicate that one register overlaps with another), and some flag bits. |
| 318 | </p> |
| 319 | |
| 320 | <p>In addition to the per-register description, the <tt>MRegisterInfo</tt> class |
| 321 | exposes a set of processor specific register classes (instances of the |
| 322 | <tt>TargetRegisterClass</tt> class). Each register class contains sets of |
| 323 | registers that have the same properties (for example, they are all 32-bit |
| 324 | integer registers). Each SSA virtual register created by the instruction |
| 325 | selector has an associated register class. When the register allocator runs, it |
| 326 | replaces virtual registers with a physical register in the set.</p> |
| 327 | |
| 328 | <p> |
| 329 | The target-specific implementations of these classes is auto-generated from a <a |
| 330 | href="TableGenFundamentals.html">TableGen</a> description of the register file. |
| 331 | </p> |
| 332 | |
| 333 | </div> |
| 334 | |
| 335 | <!-- ======================================================================= --> |
| 336 | <div class="doc_subsection"> |
Chris Lattner | 10d6800 | 2004-06-01 17:18:11 +0000 | [diff] [blame] | 337 | <a name="targetinstrinfo">The <tt>TargetInstrInfo</tt> class</a> |
Chris Lattner | ce52b7e | 2004-06-01 06:48:00 +0000 | [diff] [blame] | 338 | </div> |
| 339 | |
| 340 | <!-- ======================================================================= --> |
| 341 | <div class="doc_subsection"> |
Chris Lattner | 10d6800 | 2004-06-01 17:18:11 +0000 | [diff] [blame] | 342 | <a name="targetframeinfo">The <tt>TargetFrameInfo</tt> class</a> |
Chris Lattner | ce52b7e | 2004-06-01 06:48:00 +0000 | [diff] [blame] | 343 | </div> |
| 344 | |
| 345 | <!-- ======================================================================= --> |
| 346 | <div class="doc_subsection"> |
Chris Lattner | 10d6800 | 2004-06-01 17:18:11 +0000 | [diff] [blame] | 347 | <a name="targetjitinfo">The <tt>TargetJITInfo</tt> class</a> |
Chris Lattner | ce52b7e | 2004-06-01 06:48:00 +0000 | [diff] [blame] | 348 | </div> |
| 349 | |
| 350 | <!-- *********************************************************************** --> |
| 351 | <div class="doc_section"> |
| 352 | <a name="codegendesc">Machine code description classes</a> |
| 353 | </div> |
| 354 | <!-- *********************************************************************** --> |
| 355 | |
Chris Lattner | ec94f80 | 2004-06-04 00:16:02 +0000 | [diff] [blame^] | 356 | <div class="doc_text"> |
Chris Lattner | ce52b7e | 2004-06-01 06:48:00 +0000 | [diff] [blame] | 357 | |
Chris Lattner | ec94f80 | 2004-06-04 00:16:02 +0000 | [diff] [blame^] | 358 | <p> |
| 359 | At the high-level, LLVM code is translated to a machine specific representation |
| 360 | formed out of MachineFunction, MachineBasicBlock, and <a |
| 361 | href="#machineinstr"><tt>MachineInstr</tt></a> instances |
| 362 | (defined in include/llvm/CodeGen). This representation is completely target |
| 363 | agnostic, representing instructions in their most abstract form: an opcode and a |
| 364 | series of operands. This representation is designed to support both SSA |
| 365 | representation for machine code, as well as a register allocated, non-SSA form. |
| 366 | </p> |
| 367 | |
| 368 | </div> |
| 369 | |
| 370 | <!-- ======================================================================= --> |
| 371 | <div class="doc_subsection"> |
| 372 | <a name="machineinstr">The <tt>MachineInstr</tt> class</a> |
| 373 | </div> |
| 374 | |
| 375 | <div class="doc_text"> |
| 376 | |
| 377 | <p>Target machine instructions are represented as instances of the |
| 378 | <tt>MachineInstr</tt> class. This class is an extremely abstract way of |
| 379 | representing machine instructions. In particular, all it keeps track of is |
| 380 | an opcode number and some number of operands.</p> |
| 381 | |
| 382 | <p>The opcode number is an simple unsigned number that only has meaning to a |
| 383 | specific backend. All of the instructions for a target should be defined in |
| 384 | the <tt>*InstrInfo.td</tt> file for the target, and the opcode enum values |
| 385 | are autogenerated from this description. The <tt>MachineInstr</tt> class does |
| 386 | not have any information about how to intepret the instruction (i.e., what the |
| 387 | semantics of the instruction are): for that you must refer to the |
| 388 | <tt><a href="#targetinstrinfo">TargetInstrInfo</a></tt> class.</p> |
| 389 | |
| 390 | <p>The operands of a machine instruction can be of several different types: |
| 391 | they can be a register reference, constant integer, basic block reference, etc. |
| 392 | In addition, a machine operand should be marked as a def or a use of the value |
| 393 | (though only registers are allowed to be defs).</p> |
| 394 | |
| 395 | <p>By convention, the LLVM code generator orders instruction operands so that |
| 396 | all register definitions come before the register uses, even on architectures |
| 397 | that are normally printed in other orders. For example, the sparc add |
| 398 | instruction: "<tt>add %i1, %i2, %i3</tt>" adds the "%i1", and "%i2" registers |
| 399 | and stores the result into the "%i3" register. In the LLVM code generator, |
| 400 | the operands should be stored as "<tt>%i3, %i1, %i2</tt>": with the destination |
| 401 | first.</p> |
| 402 | |
| 403 | <p>Keeping destination operands at the beginning of the operand list has several |
| 404 | advantages. In particular, the debugging printer will print the instruction |
| 405 | like this:</p> |
| 406 | |
| 407 | <pre> |
| 408 | %r3 = add %i1, %i2 |
| 409 | </pre> |
| 410 | |
| 411 | <p>If the first operand is a def, and it is also easier to <a |
| 412 | href="#buildmi">create instructions</a> whose only def is the first |
| 413 | operand.</p> |
| 414 | |
| 415 | </div> |
| 416 | |
| 417 | <!-- _______________________________________________________________________ --> |
| 418 | <div class="doc_subsubsection"> |
| 419 | <a name="buildmi">Using the <tt>MachineInstrBuilder.h</tt> functions</a> |
| 420 | </div> |
| 421 | |
| 422 | <div class="doc_text"> |
| 423 | |
| 424 | <p>Machine instructions are created by using the <tt>BuildMI</tt> functions, |
| 425 | located in the <tt>include/llvm/CodeGen/MachineInstrBuilder.h</tt> file. The |
| 426 | <tt>BuildMI</tt> functions make it easy to build arbitrary machine |
| 427 | instructions. Usage of the <tt>BuildMI</tt> functions look like this: |
| 428 | </p> |
| 429 | |
| 430 | <pre> |
| 431 | // Create a 'DestReg = mov 42' (rendered in X86 assembly as 'mov DestReg, 42') |
| 432 | // instruction. The '1' specifies how many operands will be added. |
| 433 | MachineInstr *MI = BuildMI(X86::MOV32ri, 1, DestReg).addImm(42); |
| 434 | |
| 435 | // Create the same instr, but insert it at the end of a basic block. |
| 436 | MachineBasicBlock &MBB = ... |
| 437 | BuildMI(MBB, X86::MOV32ri, 1, DestReg).addImm(42); |
| 438 | |
| 439 | // Create the same instr, but insert it before a specified iterator point. |
| 440 | MachineBasicBlock::iterator MBBI = ... |
| 441 | BuildMI(MBB, MBBI, X86::MOV32ri, 1, DestReg).addImm(42); |
| 442 | |
| 443 | // Create a 'cmp Reg, 0' instruction, no destination reg. |
| 444 | MI = BuildMI(X86::CMP32ri, 2).addReg(Reg).addImm(0); |
| 445 | // Create an 'sahf' instruction which takes no operands and stores nothing. |
| 446 | MI = BuildMI(X86::SAHF, 0); |
| 447 | |
| 448 | // Create a self looping branch instruction. |
| 449 | BuildMI(MBB, X86::JNE, 1).addMBB(&MBB); |
| 450 | </pre> |
| 451 | |
| 452 | <p> |
| 453 | The key thing to remember with the <tt>BuildMI</tt> functions is that you have |
| 454 | to specify the number of operands that the machine instruction will take |
| 455 | (allowing efficient memory allocation). Also, if operands default to be uses |
| 456 | of values, not definitions. If you need to add a definition operand (other |
| 457 | than the optional destination register), you must explicitly mark it as such. |
| 458 | </p> |
| 459 | |
| 460 | </div> |
| 461 | |
| 462 | <!-- _______________________________________________________________________ --> |
| 463 | <div class="doc_subsubsection"> |
| 464 | <a name="fixedregs">Fixed (aka preassigned) registers</a> |
| 465 | </div> |
| 466 | |
| 467 | <div class="doc_text"> |
| 468 | |
| 469 | <p>One important issue that the code generator needs to be aware of is the |
| 470 | presence of fixed registers. In particular, there are often places in the |
| 471 | instruction stream where the register allocator <em>must</em> arrange for a |
| 472 | particular value to be in a particular register. This can occur due to |
| 473 | limitations in the instruction set (e.g., the X86 can only do a 32-bit divide |
| 474 | with the <tt>EAX</tt>/<tt>EDX</tt> registers), or external factors like calling |
| 475 | conventions. In any case, the instruction selector should emit code that |
| 476 | copies a virtual register into or out of a physical register when needed.</p> |
| 477 | |
| 478 | <p>For example, consider this simple LLVM example:</p> |
| 479 | |
| 480 | <pre> |
| 481 | int %test(int %X, int %Y) { |
| 482 | %Z = div int %X, %Y |
| 483 | ret int %Z |
| 484 | } |
| 485 | </pre> |
| 486 | |
| 487 | <p>The X86 instruction selector produces this machine code for the div |
| 488 | and ret (use |
| 489 | "<tt>llc X.bc -march=x86 -print-machineinstrs</tt>" to get this):</p> |
| 490 | |
| 491 | <pre> |
| 492 | ;; Start of div |
| 493 | %EAX = mov %reg1024 ;; Copy X (in reg1024) into EAX |
| 494 | %reg1027 = sar %reg1024, 31 |
| 495 | %EDX = mov %reg1027 ;; Sign extend X into EDX |
| 496 | idiv %reg1025 ;; Divide by Y (in reg1025) |
| 497 | %reg1026 = mov %EAX ;; Read the result (Z) out of EAX |
| 498 | |
| 499 | ;; Start of ret |
| 500 | %EAX = mov %reg1026 ;; 32-bit return value goes in EAX |
| 501 | ret |
| 502 | </pre> |
| 503 | |
| 504 | <p>By the end of code generation, the register allocator has coallesced |
| 505 | the registers and deleted the resultant identity moves, producing the |
| 506 | following code:</p> |
| 507 | |
| 508 | <pre> |
| 509 | ;; X is in EAX, Y is in ECX |
| 510 | mov %EAX, %EDX |
| 511 | sar %EDX, 31 |
| 512 | idiv %ECX |
| 513 | ret |
| 514 | </pre> |
| 515 | |
| 516 | <p>This approach is extremely general (if it can handle the X86 architecture, |
| 517 | it can handle anything!) and allows all of the target specific |
| 518 | knowledge about the instruction stream to be isolated in the instruction |
| 519 | selector. Note that physical registers should have a short lifetime for good |
| 520 | code generation, and all physical registers are assumed dead on entry and |
| 521 | exit of basic blocks (before register allocation). Thus if you need a value |
| 522 | to be live across basic block boundaries, it <em>must</em> live in a virtual |
| 523 | register.</p> |
| 524 | |
| 525 | </div> |
| 526 | |
| 527 | <!-- _______________________________________________________________________ --> |
| 528 | <div class="doc_subsubsection"> |
| 529 | <a name="ssa">Machine code SSA form</a> |
| 530 | </div> |
| 531 | |
| 532 | <div class="doc_text"> |
| 533 | |
| 534 | <p><tt>MachineInstr</tt>'s are initially instruction selected in SSA-form, and |
| 535 | are maintained in SSA-form until register allocation happens. For the most |
| 536 | part, this is trivially simple since LLVM is already in SSA form: LLVM PHI nodes |
| 537 | become machine code PHI nodes, and virtual registers are only allowed to have a |
| 538 | single definition.</p> |
| 539 | |
| 540 | <p>After register allocation, machine code is no longer in SSA-form, as there |
| 541 | are no virtual registers left in the code.</p> |
| 542 | |
| 543 | </div> |
| 544 | |
| 545 | <!-- *********************************************************************** --> |
| 546 | <div class="doc_section"> |
| 547 | <a name="targetimpls">Target description implementations</a> |
| 548 | </div> |
| 549 | <!-- *********************************************************************** --> |
| 550 | |
| 551 | <div class="doc_text"> |
| 552 | |
| 553 | <p>This section of the document explains any features or design decisions that |
| 554 | are specific to the code generator for a particular target.</p> |
| 555 | |
| 556 | </div> |
| 557 | |
| 558 | |
| 559 | <!-- ======================================================================= --> |
| 560 | <div class="doc_subsection"> |
| 561 | <a name="x86">The X86 backend</a> |
| 562 | </div> |
| 563 | |
| 564 | <div class="doc_text"> |
| 565 | |
| 566 | <p> |
| 567 | The X86 code generator lives in the <tt>lib/Target/X86</tt> directory. This |
| 568 | code generator currently targets a generic P6-like processor. As such, it |
| 569 | produces a few P6-and-above instructions (like conditional moves), but it does |
| 570 | not make use of newer features like MMX or SSE. In the future, the X86 backend |
| 571 | will have subtarget support added for specific processor families and |
| 572 | implementations.</p> |
| 573 | |
| 574 | </div> |
| 575 | |
| 576 | <!-- _______________________________________________________________________ --> |
| 577 | <div class="doc_subsubsection"> |
| 578 | <a name="x86_memory">Representing X86 addressing modes in MachineInstrs</a> |
| 579 | </div> |
| 580 | |
| 581 | <div class="doc_text"> |
| 582 | |
| 583 | <p> |
| 584 | The x86 has a very, uhm, flexible, way of accessing memory. It is capable of |
| 585 | forming memory addresses of the following expression directly in integer |
| 586 | instructions (which use ModR/M addressing):</p> |
| 587 | |
| 588 | <pre> |
| 589 | Base+[1,2,4,8]*IndexReg+Disp32 |
| 590 | </pre> |
| 591 | |
| 592 | <p>Wow, that's crazy. In order to represent this, LLVM tracks no less that 4 |
| 593 | operands for each memory operand of this form. This means that the "load" form |
| 594 | of 'mov' has the following "Operands" in this order:</p> |
| 595 | |
| 596 | <pre> |
| 597 | Index: 0 | 1 2 3 4 |
| 598 | Meaning: DestReg, | BaseReg, Scale, IndexReg, Displacement |
| 599 | OperandTy: VirtReg, | VirtReg, UnsImm, VirtReg, SignExtImm |
| 600 | </pre> |
| 601 | |
| 602 | <p>Stores and all other instructions treat the four memory operands in the same |
| 603 | way, in the same order.</p> |
| 604 | </p> |
| 605 | |
| 606 | </div> |
| 607 | |
| 608 | <!-- _______________________________________________________________________ --> |
| 609 | <div class="doc_subsubsection"> |
| 610 | <a name="x86_names">Instruction naming</a> |
| 611 | </div> |
| 612 | |
| 613 | <div class="doc_text"> |
| 614 | |
| 615 | <p> |
| 616 | An instruction name consists of the base name, a default operand size |
| 617 | followed by a character per operand with an optional special size. For |
| 618 | example:</p> |
| 619 | |
| 620 | <p> |
| 621 | <tt>ADD8rr</tt> -> add, 8-bit register, 8-bit register<br> |
| 622 | <tt>IMUL16rmi</tt> -> imul, 16-bit register, 16-bit memory, 16-bit immediate<br> |
| 623 | <tt>IMUL16rmi8</tt> -> imul, 16-bit register, 16-bit memory, 8-bit immediate<br> |
| 624 | <tt>MOVSX32rm16</tt> -> movsx, 32-bit register, 16-bit memory |
| 625 | </p> |
| 626 | |
| 627 | </div> |
Chris Lattner | ce52b7e | 2004-06-01 06:48:00 +0000 | [diff] [blame] | 628 | |
| 629 | <!-- *********************************************************************** --> |
| 630 | <hr> |
| 631 | <address> |
| 632 | <a href="http://jigsaw.w3.org/css-validator/check/referer"><img |
| 633 | src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a> |
| 634 | <a href="http://validator.w3.org/check/referer"><img |
| 635 | src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a> |
| 636 | |
| 637 | <a href="mailto:sabre@nondot.org">Chris Lattner</a><br> |
| 638 | <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a><br> |
| 639 | Last modified: $Date$ |
| 640 | </address> |
| 641 | |
| 642 | </body> |
| 643 | </html> |