1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" 2 "http://www.w3.org/TR/html4/strict.dtd"> 3 <html> 4 <head> 5 <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> 6 <title>LLVM: Frequently Asked Questions</title> 7 <style type="text/css"> 8 @import url("llvm.css"); 9 .question { font-weight: bold } 10 .answer { margin-left: 2em } 11 </style> 12 </head> 13 <body> 14 15 <h1> 16 LLVM: Frequently Asked Questions 17 </h1> 18 19 <ol> 20 <li><a href="#license">License</a> 21 <ol> 22 <li>Why are the LLVM source code and the front-end distributed under 23 different licenses?</li> 24 25 <li>Does the University of Illinois Open Source License really qualify as an 26 "open source" license?</li> 27 28 <li>Can I modify LLVM source code and redistribute the modified source?</li> 29 30 <li>Can I modify LLVM source code and redistribute binaries or other tools 31 based on it, without redistributing the source?</li> 32 </ol></li> 33 34 <li><a href="#source">Source code</a> 35 <ol> 36 <li>In what language is LLVM written?</li> 37 38 <li>How portable is the LLVM source code?</li> 39 </ol></li> 40 41 <li><a href="#build">Build Problems</a> 42 <ol> 43 <li>When I run configure, it finds the wrong C compiler.</li> 44 45 <li>The <tt>configure</tt> script finds the right C compiler, but it uses 46 the LLVM linker from a previous build. What do I do?</li> 47 48 <li>When creating a dynamic library, I get a strange GLIBC error.</li> 49 50 <li>I've updated my source tree from Subversion, and now my build is trying 51 to use a file/directory that doesn't exist.</li> 52 53 <li>I've modified a Makefile in my source tree, but my build tree keeps 54 using the old version. What do I do?</li> 55 56 <li>I've upgraded to a new version of LLVM, and I get strange build 57 errors.</li> 58 59 <li>I've built LLVM and am testing it, but the tests freeze.</li> 60 61 <li>Why do test results differ when I perform different types of 62 builds?</li> 63 64 <li>Compiling LLVM with GCC 3.3.2 fails, what should I do?</li> 65 66 <li>Compiling LLVM with GCC succeeds, but the resulting tools do not work, 67 what can be wrong?</li> 68 69 <li>When I use the test suite, all of the C Backend tests fail. What is 70 wrong?</li> 71 72 <li>After Subversion update, rebuilding gives the error "No rule to make 73 target".</li> 74 75 <li><a href="#srcdir-objdir">When I compile LLVM-GCC with srcdir == objdir, 76 it fails. Why?</a></li> 77 </ol></li> 78 79 <li><a href="#felangs">Source Languages</a> 80 <ol> 81 <li><a href="#langs">What source languages are supported?</a></li> 82 83 <li><a href="#langirgen">I'd like to write a self-hosting LLVM compiler. How 84 should I interface with the LLVM middle-end optimizers and back-end code 85 generators?</a></li> 86 87 <li><a href="#langhlsupp">What support is there for higher level source 88 language constructs for building a compiler?</a></li> 89 90 <li><a href="GetElementPtr.html">I don't understand the GetElementPtr 91 instruction. Help!</a></li> 92 </ol> 93 94 <li><a href="#cfe">Using the GCC Front End</a> 95 <ol> 96 <li>When I compile software that uses a configure script, the configure 97 script thinks my system has all of the header files and libraries it is 98 testing for. How do I get configure to work correctly?</li> 99 100 <li>When I compile code using the LLVM GCC front end, it complains that it 101 cannot find libcrtend.a?</li> 102 103 <li>How can I disable all optimizations when compiling code using the LLVM 104 GCC front end?</li> 105 106 <li><a href="#translatecxx">Can I use LLVM to convert C++ code to C 107 code?</a></li> 108 109 <li><a href="#platformindependent">Can I compile C or C++ code to 110 platform-independent LLVM bitcode?</a></li> 111 </ol> 112 </li> 113 114 <li><a href="#cfe_code">Questions about code generated by the GCC front-end</a> 115 <ol> 116 <li><a href="#iosinit">What is this <tt>llvm.global_ctors</tt> and 117 <tt>_GLOBAL__I__tmp_webcompile...</tt> stuff that happens when I 118 #include <iostream>?</a></li> 119 120 <li><a href="#codedce">Where did all of my code go??</a></li> 121 122 <li><a href="#undef">What is this "<tt>undef</tt>" thing that shows up in 123 my code?</a></li> 124 125 <li><a href="#callconvwrong">Why does instcombine + simplifycfg turn 126 a call to a function with a mismatched calling convention into "unreachable"? 127 Why not make the verifier reject it?</a></li> 128 </ol> 129 </li> 130 </ol> 131 132 <div class="doc_author"> 133 <p>Written by <a href="http://llvm.org/">The LLVM Team</a></p> 134 </div> 135 136 137 <!-- *********************************************************************** --> 138 <h2> 139 <a name="license">License</a> 140 </h2> 141 <!-- *********************************************************************** --> 142 143 <div> 144 145 <div class="question"> 146 <p>Why are the LLVM source code and the front-end distributed under different 147 licenses?</p> 148 </div> 149 150 <div class="answer"> 151 <p>The C/C++ front-ends are based on GCC and must be distributed under the GPL. 152 Our aim is to distribute LLVM source code under a <em>much less 153 restrictive</em> license, in particular one that does not compel users who 154 distribute tools based on modifying the source to redistribute the modified 155 source code as well.</p> 156 </div> 157 158 <div class="question"> 159 <p>Does the University of Illinois Open Source License really qualify as an 160 "open source" license?</p> 161 </div> 162 163 <div class="answer"> 164 <p>Yes, the license 165 is <a href="http://www.opensource.org/licenses/UoI-NCSA.php">certified</a> by 166 the Open Source Initiative (OSI).</p> 167 </div> 168 169 <div class="question"> 170 <p>Can I modify LLVM source code and redistribute the modified source?</p> 171 </div> 172 173 <div class="answer"> 174 <p>Yes. The modified source distribution must retain the copyright notice and 175 follow the three bulletted conditions listed in 176 the <a href="http://llvm.org/svn/llvm-project/llvm/trunk/LICENSE.TXT">LLVM 177 license</a>.</p> 178 </div> 179 180 <div class="question"> 181 <p>Can I modify LLVM source code and redistribute binaries or other tools based 182 on it, without redistributing the source?</p> 183 </div> 184 185 <div class="answer"> 186 <p>Yes. This is why we distribute LLVM under a less restrictive license than 187 GPL, as explained in the first question above.</p> 188 </div> 189 190 </div> 191 192 <!-- *********************************************************************** --> 193 <h2> 194 <a name="source">Source Code</a> 195 </h2> 196 <!-- *********************************************************************** --> 197 198 <div> 199 200 <div class="question"> 201 <p>In what language is LLVM written?</p> 202 </div> 203 204 <div class="answer"> 205 <p>All of the LLVM tools and libraries are written in C++ with extensive use of 206 the STL.</p> 207 </div> 208 209 <div class="question"> 210 <p>How portable is the LLVM source code?</p> 211 </div> 212 213 <div class="answer"> 214 <p>The LLVM source code should be portable to most modern UNIX-like operating 215 systems. Most of the code is written in standard C++ with operating system 216 services abstracted to a support library. The tools required to build and test 217 LLVM have been ported to a plethora of platforms.</p> 218 219 <p>Some porting problems may exist in the following areas:</p> 220 221 <ul> 222 <li>The GCC front end code is not as portable as the LLVM suite, so it may not 223 compile as well on unsupported platforms.</li> 224 225 <li>The LLVM build system relies heavily on UNIX shell tools, like the Bourne 226 Shell and sed. Porting to systems without these tools (MacOS 9, Plan 9) 227 will require more effort.</li> 228 </ul> 229 230 </div> 231 232 </div> 233 234 <!-- *********************************************************************** --> 235 <h2> 236 <a name="build">Build Problems</a> 237 </h2> 238 <!-- *********************************************************************** --> 239 240 <div> 241 242 <div class="question"> 243 <p>When I run configure, it finds the wrong C compiler.</p> 244 </div> 245 246 <div class="answer"> 247 <p>The <tt>configure</tt> script attempts to locate first <tt>gcc</tt> and then 248 <tt>cc</tt>, unless it finds compiler paths set in <tt>CC</tt> 249 and <tt>CXX</tt> for the C and C++ compiler, respectively.</p> 250 251 <p>If <tt>configure</tt> finds the wrong compiler, either adjust your 252 <tt>PATH</tt> environment variable or set <tt>CC</tt> and <tt>CXX</tt> 253 explicitly.</p> 254 255 </div> 256 257 <div class="question"> 258 <p>The <tt>configure</tt> script finds the right C compiler, but it uses the 259 LLVM linker from a previous build. What do I do?</p> 260 </div> 261 262 <div class="answer"> 263 <p>The <tt>configure</tt> script uses the <tt>PATH</tt> to find executables, so 264 if it's grabbing the wrong linker/assembler/etc, there are two ways to fix 265 it:</p> 266 267 <ol> 268 <li><p>Adjust your <tt>PATH</tt> environment variable so that the correct 269 program appears first in the <tt>PATH</tt>. This may work, but may not be 270 convenient when you want them <i>first</i> in your path for other 271 work.</p></li> 272 273 <li><p>Run <tt>configure</tt> with an alternative <tt>PATH</tt> that is 274 correct. In a Borne compatible shell, the syntax would be:</p> 275 276 <pre class="doc_code"> 277 % PATH=[the path without the bad program] ./configure ... 278 </pre> 279 280 <p>This is still somewhat inconvenient, but it allows <tt>configure</tt> 281 to do its work without having to adjust your <tt>PATH</tt> 282 permanently.</p></li> 283 </ol> 284 </div> 285 286 <div class="question"> 287 <p>When creating a dynamic library, I get a strange GLIBC error.</p> 288 </div> 289 290 <div class="answer"> 291 <p>Under some operating systems (i.e. Linux), libtool does not work correctly if 292 GCC was compiled with the --disable-shared option. To work around this, 293 install your own version of GCC that has shared libraries enabled by 294 default.</p> 295 </div> 296 297 <div class="question"> 298 <p>I've updated my source tree from Subversion, and now my build is trying to 299 use a file/directory that doesn't exist.</p> 300 </div> 301 302 <div class="answer"> 303 <p>You need to re-run configure in your object directory. When new Makefiles 304 are added to the source tree, they have to be copied over to the object tree 305 in order to be used by the build.</p> 306 </div> 307 308 <div class="question"> 309 <p>I've modified a Makefile in my source tree, but my build tree keeps using the 310 old version. What do I do?</p> 311 </div> 312 313 <div class="answer"> 314 <p>If the Makefile already exists in your object tree, you can just run the 315 following command in the top level directory of your object tree:</p> 316 317 <pre class="doc_code"> 318 % ./config.status <relative path to Makefile> 319 </pre> 320 321 <p>If the Makefile is new, you will have to modify the configure script to copy 322 it over.</p> 323 </div> 324 325 <div class="question"> 326 <p>I've upgraded to a new version of LLVM, and I get strange build errors.</p> 327 </div> 328 329 <div class="answer"> 330 331 <p>Sometimes, changes to the LLVM source code alters how the build system works. 332 Changes in libtool, autoconf, or header file dependencies are especially 333 prone to this sort of problem.</p> 334 335 <p>The best thing to try is to remove the old files and re-build. In most 336 cases, this takes care of the problem. To do this, just type <tt>make 337 clean</tt> and then <tt>make</tt> in the directory that fails to build.</p> 338 </div> 339 340 <div class="question"> 341 <p>I've built LLVM and am testing it, but the tests freeze.</p> 342 </div> 343 344 <div class="answer"> 345 <p>This is most likely occurring because you built a profile or release 346 (optimized) build of LLVM and have not specified the same information on the 347 <tt>gmake</tt> command line.</p> 348 349 <p>For example, if you built LLVM with the command:</p> 350 351 <pre class="doc_code"> 352 % gmake ENABLE_PROFILING=1 353 </pre> 354 355 <p>...then you must run the tests with the following commands:</p> 356 357 <pre class="doc_code"> 358 % cd llvm/test 359 % gmake ENABLE_PROFILING=1 360 </pre> 361 </div> 362 363 <div class="question"> 364 <p>Why do test results differ when I perform different types of builds?</p> 365 </div> 366 367 <div class="answer"> 368 <p>The LLVM test suite is dependent upon several features of the LLVM tools and 369 libraries.</p> 370 371 <p>First, the debugging assertions in code are not enabled in optimized or 372 profiling builds. Hence, tests that used to fail may pass.</p> 373 374 <p>Second, some tests may rely upon debugging options or behavior that is only 375 available in the debug build. These tests will fail in an optimized or 376 profile build.</p> 377 </div> 378 379 <div class="question"> 380 <p>Compiling LLVM with GCC 3.3.2 fails, what should I do?</p> 381 </div> 382 383 <div class="answer"> 384 <p>This is <a href="http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13392">a bug in 385 GCC</a>, and affects projects other than LLVM. Try upgrading or downgrading 386 your GCC.</p> 387 </div> 388 389 <div class="question"> 390 <p>Compiling LLVM with GCC succeeds, but the resulting tools do not work, what 391 can be wrong?</p> 392 </div> 393 394 <div class="answer"> 395 <p>Several versions of GCC have shown a weakness in miscompiling the LLVM 396 codebase. Please consult your compiler version (<tt>gcc --version</tt>) to 397 find out whether it is <a href="GettingStarted.html#brokengcc">broken</a>. 398 If so, your only option is to upgrade GCC to a known good version.</p> 399 </div> 400 401 <div class="question"> 402 <p>After Subversion update, rebuilding gives the error "No rule to make 403 target".</p> 404 </div> 405 406 <div class="answer"> 407 <p>If the error is of the form:</p> 408 409 <pre class="doc_code"> 410 gmake[2]: *** No rule to make target `/path/to/somefile', needed by 411 `/path/to/another/file.d'.<br> 412 Stop. 413 </pre> 414 415 <p>This may occur anytime files are moved within the Subversion repository or 416 removed entirely. In this case, the best solution is to erase all 417 <tt>.d</tt> files, which list dependencies for source files, and rebuild:</p> 418 419 <pre class="doc_code"> 420 % cd $LLVM_OBJ_DIR 421 % rm -f `find . -name \*\.d` 422 % gmake 423 </pre> 424 425 <p>In other cases, it may be necessary to run <tt>make clean</tt> before 426 rebuilding.</p> 427 </div> 428 429 <div class="question"> 430 <p><a name="srcdir-objdir">When I compile LLVM-GCC with srcdir == objdir, it 431 fails. Why?</a></p> 432 </div> 433 434 <div class="answer"> 435 <p>The <tt>GNUmakefile</tt> in the top-level directory of LLVM-GCC is a special 436 <tt>Makefile</tt> used by Apple to invoke the <tt>build_gcc</tt> script after 437 setting up a special environment. This has the unfortunate side-effect that 438 trying to build LLVM-GCC with srcdir == objdir in a "non-Apple way" invokes 439 the <tt>GNUmakefile</tt> instead of <tt>Makefile</tt>. Because the 440 environment isn't set up correctly to do this, the build fails.</p> 441 442 <p>People not building LLVM-GCC the "Apple way" need to build LLVM-GCC with 443 srcdir != objdir, or simply remove the GNUmakefile entirely.</p> 444 445 <p>We regret the inconvenience.</p> 446 </div> 447 448 </div> 449 450 <!-- *********************************************************************** --> 451 <h2> 452 <a name="felangs">Source Languages</a> 453 </h2> 454 455 <div> 456 457 <div class="question"> 458 <p><a name="langs">What source languages are supported?</a></p> 459 </div> 460 461 <div class="answer"> 462 <p>LLVM currently has full support for C and C++ source languages. These are 463 available through a special version of GCC that LLVM calls the 464 <a href="#cfe">C Front End</a></p> 465 466 <p>There is an incomplete version of a Java front end available in the 467 <tt>java</tt> module. There is no documentation on this yet so you'll need to 468 download the code, compile it, and try it.</p> 469 470 <p>The PyPy developers are working on integrating LLVM into the PyPy backend so 471 that PyPy language can translate to LLVM.</p> 472 </div> 473 474 <div class="question"> 475 <p><a name="langirgen">I'd like to write a self-hosting LLVM compiler. How 476 should I interface with the LLVM middle-end optimizers and back-end code 477 generators?</a></p> 478 </div> 479 480 <div class="answer"> 481 <p>Your compiler front-end will communicate with LLVM by creating a module in 482 the LLVM intermediate representation (IR) format. Assuming you want to write 483 your language's compiler in the language itself (rather than C++), there are 484 3 major ways to tackle generating LLVM IR from a front-end:</p> 485 486 <ul> 487 <li><strong>Call into the LLVM libraries code using your language's FFI 488 (foreign function interface).</strong> 489 490 <ul> 491 <li><em>for:</em> best tracks changes to the LLVM IR, .ll syntax, and .bc 492 format</li> 493 494 <li><em>for:</em> enables running LLVM optimization passes without a 495 emit/parse overhead</li> 496 497 <li><em>for:</em> adapts well to a JIT context</li> 498 499 <li><em>against:</em> lots of ugly glue code to write</li> 500 </ul></li> 501 502 <li> <strong>Emit LLVM assembly from your compiler's native language.</strong> 503 <ul> 504 <li><em>for:</em> very straightforward to get started</li> 505 506 <li><em>against:</em> the .ll parser is slower than the bitcode reader 507 when interfacing to the middle end</li> 508 509 <li><em>against:</em> you'll have to re-engineer the LLVM IR object model 510 and asm writer in your language</li> 511 512 <li><em>against:</em> it may be harder to track changes to the IR</li> 513 </ul></li> 514 515 <li><strong>Emit LLVM bitcode from your compiler's native language.</strong> 516 517 <ul> 518 <li><em>for:</em> can use the more-efficient bitcode reader when 519 interfacing to the middle end</li> 520 521 <li><em>against:</em> you'll have to re-engineer the LLVM IR object 522 model and bitcode writer in your language</li> 523 524 <li><em>against:</em> it may be harder to track changes to the IR</li> 525 </ul></li> 526 </ul> 527 528 <p>If you go with the first option, the C bindings in include/llvm-c should help 529 a lot, since most languages have strong support for interfacing with C. The 530 most common hurdle with calling C from managed code is interfacing with the 531 garbage collector. The C interface was designed to require very little memory 532 management, and so is straightforward in this regard.</p> 533 </div> 534 535 <div class="question"> 536 <p><a name="langhlsupp">What support is there for a higher level source language 537 constructs for building a compiler?</a></p> 538 </div> 539 540 <div class="answer"> 541 <p>Currently, there isn't much. LLVM supports an intermediate representation 542 which is useful for code representation but will not support the high level 543 (abstract syntax tree) representation needed by most compilers. There are no 544 facilities for lexical nor semantic analysis.</p> 545 </div> 546 547 <div class="question"> 548 <p><a name="getelementptr">I don't understand the GetElementPtr 549 instruction. Help!</a></p> 550 </div> 551 552 <div class="answer"> 553 <p>See <a href="GetElementPtr.html">The Often Misunderstood GEP 554 Instruction</a>.</p> 555 </div> 556 557 </div> 558 559 <!-- *********************************************************************** --> 560 <h2> 561 <a name="cfe">Using the GCC Front End</a> 562 </h2> 563 564 <div> 565 566 <div class="question"> 567 <p>When I compile software that uses a configure script, the configure script 568 thinks my system has all of the header files and libraries it is testing for. 569 How do I get configure to work correctly?</p> 570 </div> 571 572 <div class="answer"> 573 <p>The configure script is getting things wrong because the LLVM linker allows 574 symbols to be undefined at link time (so that they can be resolved during JIT 575 or translation to the C back end). That is why configure thinks your system 576 "has everything."</p> 577 578 <p>To work around this, perform the following steps:</p> 579 580 <ol> 581 <li>Make sure the CC and CXX environment variables contains the full path to 582 the LLVM GCC front end.</li> 583 584 <li>Make sure that the regular C compiler is first in your PATH. </li> 585 586 <li>Add the string "-Wl,-native" to your CFLAGS environment variable.</li> 587 </ol> 588 589 <p>This will allow the <tt>llvm-ld</tt> linker to create a native code 590 executable instead of shell script that runs the JIT. Creating native code 591 requires standard linkage, which in turn will allow the configure script to 592 find out if code is not linking on your system because the feature isn't 593 available on your system.</p> 594 </div> 595 596 <div class="question"> 597 <p>When I compile code using the LLVM GCC front end, it complains that it cannot 598 find libcrtend.a. 599 </p> 600 </div> 601 602 <div class="answer"> 603 <p>The only way this can happen is if you haven't installed the runtime 604 library. To correct this, do:</p> 605 606 <pre class="doc_code"> 607 % cd llvm/runtime 608 % make clean ; make install-bytecode 609 </pre> 610 </div> 611 612 <div class="question"> 613 <p>How can I disable all optimizations when compiling code using the LLVM GCC 614 front end?</p> 615 </div> 616 617 <div class="answer"> 618 <p>Passing "-Wa,-disable-opt -Wl,-disable-opt" will disable *all* cleanup and 619 optimizations done at the llvm level, leaving you with the truly horrible 620 code that you desire.</p> 621 </div> 622 623 624 <div class="question"> 625 <p><a name="translatecxx">Can I use LLVM to convert C++ code to C code?</a></p> 626 </div> 627 628 <div class="answer"> 629 <p>Yes, you can use LLVM to convert code from any language LLVM supports to C. 630 Note that the generated C code will be very low level (all loops are lowered 631 to gotos, etc) and not very pretty (comments are stripped, original source 632 formatting is totally lost, variables are renamed, expressions are 633 regrouped), so this may not be what you're looking for. Also, there are 634 several limitations noted below.<p> 635 636 <p>Use commands like this:</p> 637 638 <ol> 639 <li><p>Compile your program with llvm-g++:</p> 640 641 <pre class="doc_code"> 642 % llvm-g++ -emit-llvm x.cpp -o program.bc -c 643 </pre> 644 645 <p>or:</p> 646 647 <pre class="doc_code"> 648 % llvm-g++ a.cpp -c -emit-llvm 649 % llvm-g++ b.cpp -c -emit-llvm 650 % llvm-ld a.o b.o -o program 651 </pre> 652 653 <p>This will generate program and program.bc. The .bc 654 file is the LLVM version of the program all linked together.</p></li> 655 656 <li><p>Convert the LLVM code to C code, using the LLC tool with the C 657 backend:</p> 658 659 <pre class="doc_code"> 660 % llc -march=c program.bc -o program.c 661 </pre></li> 662 663 <li><p>Finally, compile the C file:</p> 664 665 <pre class="doc_code"> 666 % cc x.c -lstdc++ 667 </pre></li> 668 669 </ol> 670 671 <p>Using LLVM does not eliminate the need for C++ library support. If you use 672 the llvm-g++ front-end, the generated code will depend on g++'s C++ support 673 libraries in the same way that code generated from g++ would. If you use 674 another C++ front-end, the generated code will depend on whatever library 675 that front-end would normally require.</p> 676 677 <p>If you are working on a platform that does not provide any C++ libraries, you 678 may be able to manually compile libstdc++ to LLVM bitcode, statically link it 679 into your program, then use the commands above to convert the whole result 680 into C code. Alternatively, you might compile the libraries and your 681 application into two different chunks of C code and link them.</p> 682 683 <p>Note that, by default, the C back end does not support exception handling. 684 If you want/need it for a certain program, you can enable it by passing 685 "-enable-correct-eh-support" to the llc program. The resultant code will use 686 setjmp/longjmp to implement exception support that is relatively slow, and 687 not C++-ABI-conforming on most platforms, but otherwise correct.</p> 688 689 <p>Also, there are a number of other limitations of the C backend that cause it 690 to produce code that does not fully conform to the C++ ABI on most 691 platforms. Some of the C++ programs in LLVM's test suite are known to fail 692 when compiled with the C back end because of ABI incompatibilities with 693 standard C++ libraries.</p> 694 </div> 695 696 <div class="question"> 697 <p><a name="platformindependent">Can I compile C or C++ code to 698 platform-independent LLVM bitcode?</a></p> 699 </div> 700 701 <div class="answer"> 702 <p>No. C and C++ are inherently platform-dependent languages. The most obvious 703 example of this is the preprocessor. A very common way that C code is made 704 portable is by using the preprocessor to include platform-specific code. In 705 practice, information about other platforms is lost after preprocessing, so 706 the result is inherently dependent on the platform that the preprocessing was 707 targeting.</p> 708 709 <p>Another example is <tt>sizeof</tt>. It's common for <tt>sizeof(long)</tt> to 710 vary between platforms. In most C front-ends, <tt>sizeof</tt> is expanded to 711 a constant immediately, thus hard-wiring a platform-specific detail.</p> 712 713 <p>Also, since many platforms define their ABIs in terms of C, and since LLVM is 714 lower-level than C, front-ends currently must emit platform-specific IR in 715 order to have the result conform to the platform ABI.</p> 716 </div> 717 718 </div> 719 720 <!-- *********************************************************************** --> 721 <h2> 722 <a name="cfe_code">Questions about code generated by the GCC front-end</a> 723 </h2> 724 725 <div> 726 727 <div class="question"> 728 <p><a name="iosinit">What is this <tt>llvm.global_ctors</tt> and 729 <tt>_GLOBAL__I__tmp_webcompile...</tt> stuff that happens when I <tt>#include 730 <iostream></tt>?</a></p> 731 </div> 732 733 <div class="answer"> 734 <p>If you <tt>#include</tt> the <tt><iostream></tt> header into a C++ 735 translation unit, the file will probably use 736 the <tt>std::cin</tt>/<tt>std::cout</tt>/... global objects. However, C++ 737 does not guarantee an order of initialization between static objects in 738 different translation units, so if a static ctor/dtor in your .cpp file 739 used <tt>std::cout</tt>, for example, the object would not necessarily be 740 automatically initialized before your use.</p> 741 742 <p>To make <tt>std::cout</tt> and friends work correctly in these scenarios, the 743 STL that we use declares a static object that gets created in every 744 translation unit that includes <tt><iostream></tt>. This object has a 745 static constructor and destructor that initializes and destroys the global 746 iostream objects before they could possibly be used in the file. The code 747 that you see in the .ll file corresponds to the constructor and destructor 748 registration code. 749 </p> 750 751 <p>If you would like to make it easier to <b>understand</b> the LLVM code 752 generated by the compiler in the demo page, consider using <tt>printf()</tt> 753 instead of <tt>iostream</tt>s to print values.</p> 754 </div> 755 756 <!--=========================================================================--> 757 758 <div class="question"> 759 <p><a name="codedce">Where did all of my code go??</a></p> 760 </div> 761 762 <div class="answer"> 763 <p>If you are using the LLVM demo page, you may often wonder what happened to 764 all of the code that you typed in. Remember that the demo script is running 765 the code through the LLVM optimizers, so if your code doesn't actually do 766 anything useful, it might all be deleted.</p> 767 768 <p>To prevent this, make sure that the code is actually needed. For example, if 769 you are computing some expression, return the value from the function instead 770 of leaving it in a local variable. If you really want to constrain the 771 optimizer, you can read from and assign to <tt>volatile</tt> global 772 variables.</p> 773 </div> 774 775 <!--=========================================================================--> 776 777 <div class="question"> 778 <p><a name="undef">What is this "<tt>undef</tt>" thing that shows up in my 779 code?</a></p> 780 </div> 781 782 <div class="answer"> 783 <p><a href="LangRef.html#undef"><tt>undef</tt></a> is the LLVM way of 784 representing a value that is not defined. You can get these if you do not 785 initialize a variable before you use it. For example, the C function:</p> 786 787 <pre class="doc_code"> 788 int X() { int i; return i; } 789 </pre> 790 791 <p>Is compiled to "<tt>ret i32 undef</tt>" because "<tt>i</tt>" never has a 792 value specified for it.</p> 793 </div> 794 795 <!--=========================================================================--> 796 797 <div class="question"> 798 <p><a name="callconvwrong">Why does instcombine + simplifycfg turn 799 a call to a function with a mismatched calling convention into "unreachable"? 800 Why not make the verifier reject it?</a></p> 801 </div> 802 803 <div class="answer"> 804 <p>This is a common problem run into by authors of front-ends that are using 805 custom calling conventions: you need to make sure to set the right calling 806 convention on both the function and on each call to the function. For example, 807 this code:</p> 808 809 <pre class="doc_code"> 810 define fastcc void @foo() { 811 ret void 812 } 813 define void @bar() { 814 call void @foo() 815 ret void 816 } 817 </pre> 818 819 <p>Is optimized to:</p> 820 821 <pre class="doc_code"> 822 define fastcc void @foo() { 823 ret void 824 } 825 define void @bar() { 826 unreachable 827 } 828 </pre> 829 830 <p>... with "opt -instcombine -simplifycfg". This often bites people because 831 "all their code disappears". Setting the calling convention on the caller and 832 callee is required for indirect calls to work, so people often ask why not make 833 the verifier reject this sort of thing.</p> 834 835 <p>The answer is that this code has undefined behavior, but it is not illegal. 836 If we made it illegal, then every transformation that could potentially create 837 this would have to ensure that it doesn't, and there is valid code that can 838 create this sort of construct (in dead code). The sorts of things that can 839 cause this to happen are fairly contrived, but we still need to accept them. 840 Here's an example:</p> 841 842 <pre class="doc_code"> 843 define fastcc void @foo() { 844 ret void 845 } 846 define internal void @bar(void()* %FP, i1 %cond) { 847 br i1 %cond, label %T, label %F 848 T: 849 call void %FP() 850 ret void 851 F: 852 call fastcc void %FP() 853 ret void 854 } 855 define void @test() { 856 %X = or i1 false, false 857 call void @bar(void()* @foo, i1 %X) 858 ret void 859 } 860 </pre> 861 862 <p>In this example, "test" always passes @foo/false into bar, which ensures that 863 it is dynamically called with the right calling conv (thus, the code is 864 perfectly well defined). If you run this through the inliner, you get this 865 (the explicit "or" is there so that the inliner doesn't dead code eliminate 866 a bunch of stuff): 867 </p> 868 869 <pre class="doc_code"> 870 define fastcc void @foo() { 871 ret void 872 } 873 define void @test() { 874 %X = or i1 false, false 875 br i1 %X, label %T.i, label %F.i 876 T.i: 877 call void @foo() 878 br label %bar.exit 879 F.i: 880 call fastcc void @foo() 881 br label %bar.exit 882 bar.exit: 883 ret void 884 } 885 </pre> 886 887 <p>Here you can see that the inlining pass made an undefined call to @foo with 888 the wrong calling convention. We really don't want to make the inliner have 889 to know about this sort of thing, so it needs to be valid code. In this case, 890 dead code elimination can trivially remove the undefined code. However, if %X 891 was an input argument to @test, the inliner would produce this: 892 </p> 893 894 <pre class="doc_code"> 895 define fastcc void @foo() { 896 ret void 897 } 898 899 define void @test(i1 %X) { 900 br i1 %X, label %T.i, label %F.i 901 T.i: 902 call void @foo() 903 br label %bar.exit 904 F.i: 905 call fastcc void @foo() 906 br label %bar.exit 907 bar.exit: 908 ret void 909 } 910 </pre> 911 912 <p>The interesting thing about this is that %X <em>must</em> be false for the 913 code to be well-defined, but no amount of dead code elimination will be able to 914 delete the broken call as unreachable. However, since instcombine/simplifycfg 915 turns the undefined call into unreachable, we end up with a branch on a 916 condition that goes to unreachable: a branch to unreachable can never happen, so 917 "-inline -instcombine -simplifycfg" is able to produce:</p> 918 919 <pre class="doc_code"> 920 define fastcc void @foo() { 921 ret void 922 } 923 define void @test(i1 %X) { 924 F.i: 925 call fastcc void @foo() 926 ret void 927 } 928 </pre> 929 930 </div> 931 932 </div> 933 934 <!-- *********************************************************************** --> 935 936 <hr> 937 <address> 938 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img 939 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a> 940 <a href="http://validator.w3.org/check/referer"><img 941 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a> 942 943 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br> 944 Last modified: $Date$ 945 </address> 946 947 </body> 948 </html> 949
