1============================ 2Clang Compiler User's Manual 3============================ 4 5.. contents:: 6 :local: 7 8Introduction 9============ 10 11The Clang Compiler is an open-source compiler for the C family of 12programming languages, aiming to be the best in class implementation of 13these languages. Clang builds on the LLVM optimizer and code generator, 14allowing it to provide high-quality optimization and code generation 15support for many targets. For more general information, please see the 16`Clang Web Site <http://clang.llvm.org>`_ or the `LLVM Web 17Site <http://llvm.org>`_. 18 19This document describes important notes about using Clang as a compiler 20for an end-user, documenting the supported features, command line 21options, etc. If you are interested in using Clang to build a tool that 22processes code, please see :doc:`InternalsManual`. If you are interested in the 23`Clang Static Analyzer <http://clang-analyzer.llvm.org>`_, please see its web 24page. 25 26Clang is designed to support the C family of programming languages, 27which includes :ref:`C <c>`, :ref:`Objective-C <objc>`, :ref:`C++ <cxx>`, and 28:ref:`Objective-C++ <objcxx>` as well as many dialects of those. For 29language-specific information, please see the corresponding language 30specific section: 31 32- :ref:`C Language <c>`: K&R C, ANSI C89, ISO C90, ISO C94 (C89+AMD1), ISO 33 C99 (+TC1, TC2, TC3). 34- :ref:`Objective-C Language <objc>`: ObjC 1, ObjC 2, ObjC 2.1, plus 35 variants depending on base language. 36- :ref:`C++ Language <cxx>` 37- :ref:`Objective C++ Language <objcxx>` 38 39In addition to these base languages and their dialects, Clang supports a 40broad variety of language extensions, which are documented in the 41corresponding language section. These extensions are provided to be 42compatible with the GCC, Microsoft, and other popular compilers as well 43as to improve functionality through Clang-specific features. The Clang 44driver and language features are intentionally designed to be as 45compatible with the GNU GCC compiler as reasonably possible, easing 46migration from GCC to Clang. In most cases, code "just works". 47Clang also provides an alternative driver, :ref:`clang-cl`, that is designed 48to be compatible with the Visual C++ compiler, cl.exe. 49 50In addition to language specific features, Clang has a variety of 51features that depend on what CPU architecture or operating system is 52being compiled for. Please see the :ref:`Target-Specific Features and 53Limitations <target_features>` section for more details. 54 55The rest of the introduction introduces some basic :ref:`compiler 56terminology <terminology>` that is used throughout this manual and 57contains a basic :ref:`introduction to using Clang <basicusage>` as a 58command line compiler. 59 60.. _terminology: 61 62Terminology 63----------- 64 65Front end, parser, backend, preprocessor, undefined behavior, 66diagnostic, optimizer 67 68.. _basicusage: 69 70Basic Usage 71----------- 72 73Intro to how to use a C compiler for newbies. 74 75compile + link compile then link debug info enabling optimizations 76picking a language to use, defaults to C11 by default. Autosenses based 77on extension. using a makefile 78 79Command Line Options 80==================== 81 82This section is generally an index into other sections. It does not go 83into depth on the ones that are covered by other sections. However, the 84first part introduces the language selection and other high level 85options like :option:`-c`, :option:`-g`, etc. 86 87Options to Control Error and Warning Messages 88--------------------------------------------- 89 90.. option:: -Werror 91 92 Turn warnings into errors. 93 94.. This is in plain monospaced font because it generates the same label as 95.. -Werror, and Sphinx complains. 96 97``-Werror=foo`` 98 99 Turn warning "foo" into an error. 100 101.. option:: -Wno-error=foo 102 103 Turn warning "foo" into an warning even if :option:`-Werror` is specified. 104 105.. option:: -Wfoo 106 107 Enable warning "foo". 108 109.. option:: -Wno-foo 110 111 Disable warning "foo". 112 113.. option:: -w 114 115 Disable all diagnostics. 116 117.. option:: -Weverything 118 119 :ref:`Enable all diagnostics. <diagnostics_enable_everything>` 120 121.. option:: -pedantic 122 123 Warn on language extensions. 124 125.. option:: -pedantic-errors 126 127 Error on language extensions. 128 129.. option:: -Wsystem-headers 130 131 Enable warnings from system headers. 132 133.. option:: -ferror-limit=123 134 135 Stop emitting diagnostics after 123 errors have been produced. The default is 136 20, and the error limit can be disabled with :option:`-ferror-limit=0`. 137 138.. option:: -ftemplate-backtrace-limit=123 139 140 Only emit up to 123 template instantiation notes within the template 141 instantiation backtrace for a single warning or error. The default is 10, and 142 the limit can be disabled with :option:`-ftemplate-backtrace-limit=0`. 143 144.. _cl_diag_formatting: 145 146Formatting of Diagnostics 147^^^^^^^^^^^^^^^^^^^^^^^^^ 148 149Clang aims to produce beautiful diagnostics by default, particularly for 150new users that first come to Clang. However, different people have 151different preferences, and sometimes Clang is driven not by a human, 152but by a program that wants consistent and easily parsable output. For 153these cases, Clang provides a wide range of options to control the exact 154output format of the diagnostics that it generates. 155 156.. _opt_fshow-column: 157 158**-f[no-]show-column** 159 Print column number in diagnostic. 160 161 This option, which defaults to on, controls whether or not Clang 162 prints the column number of a diagnostic. For example, when this is 163 enabled, Clang will print something like: 164 165 :: 166 167 test.c:28:8: warning: extra tokens at end of #endif directive [-Wextra-tokens] 168 #endif bad 169 ^ 170 // 171 172 When this is disabled, Clang will print "test.c:28: warning..." with 173 no column number. 174 175 The printed column numbers count bytes from the beginning of the 176 line; take care if your source contains multibyte characters. 177 178.. _opt_fshow-source-location: 179 180**-f[no-]show-source-location** 181 Print source file/line/column information in diagnostic. 182 183 This option, which defaults to on, controls whether or not Clang 184 prints the filename, line number and column number of a diagnostic. 185 For example, when this is enabled, Clang will print something like: 186 187 :: 188 189 test.c:28:8: warning: extra tokens at end of #endif directive [-Wextra-tokens] 190 #endif bad 191 ^ 192 // 193 194 When this is disabled, Clang will not print the "test.c:28:8: " 195 part. 196 197.. _opt_fcaret-diagnostics: 198 199**-f[no-]caret-diagnostics** 200 Print source line and ranges from source code in diagnostic. 201 This option, which defaults to on, controls whether or not Clang 202 prints the source line, source ranges, and caret when emitting a 203 diagnostic. For example, when this is enabled, Clang will print 204 something like: 205 206 :: 207 208 test.c:28:8: warning: extra tokens at end of #endif directive [-Wextra-tokens] 209 #endif bad 210 ^ 211 // 212 213**-f[no-]color-diagnostics** 214 This option, which defaults to on when a color-capable terminal is 215 detected, controls whether or not Clang prints diagnostics in color. 216 217 When this option is enabled, Clang will use colors to highlight 218 specific parts of the diagnostic, e.g., 219 220 .. nasty hack to not lose our dignity 221 222 .. raw:: html 223 224 <pre> 225 <b><span style="color:black">test.c:28:8: <span style="color:magenta">warning</span>: extra tokens at end of #endif directive [-Wextra-tokens]</span></b> 226 #endif bad 227 <span style="color:green">^</span> 228 <span style="color:green">//</span> 229 </pre> 230 231 When this is disabled, Clang will just print: 232 233 :: 234 235 test.c:2:8: warning: extra tokens at end of #endif directive [-Wextra-tokens] 236 #endif bad 237 ^ 238 // 239 240**-fansi-escape-codes** 241 Controls whether ANSI escape codes are used instead of the Windows Console 242 API to output colored diagnostics. This option is only used on Windows and 243 defaults to off. 244 245.. option:: -fdiagnostics-format=clang/msvc/vi 246 247 Changes diagnostic output format to better match IDEs and command line tools. 248 249 This option controls the output format of the filename, line number, 250 and column printed in diagnostic messages. The options, and their 251 affect on formatting a simple conversion diagnostic, follow: 252 253 **clang** (default) 254 :: 255 256 t.c:3:11: warning: conversion specifies type 'char *' but the argument has type 'int' 257 258 **msvc** 259 :: 260 261 t.c(3,11) : warning: conversion specifies type 'char *' but the argument has type 'int' 262 263 **vi** 264 :: 265 266 t.c +3:11: warning: conversion specifies type 'char *' but the argument has type 'int' 267 268.. _opt_fdiagnostics-show-option: 269 270**-f[no-]diagnostics-show-option** 271 Enable ``[-Woption]`` information in diagnostic line. 272 273 This option, which defaults to on, controls whether or not Clang 274 prints the associated :ref:`warning group <cl_diag_warning_groups>` 275 option name when outputting a warning diagnostic. For example, in 276 this output: 277 278 :: 279 280 test.c:28:8: warning: extra tokens at end of #endif directive [-Wextra-tokens] 281 #endif bad 282 ^ 283 // 284 285 Passing **-fno-diagnostics-show-option** will prevent Clang from 286 printing the [:ref:`-Wextra-tokens <opt_Wextra-tokens>`] information in 287 the diagnostic. This information tells you the flag needed to enable 288 or disable the diagnostic, either from the command line or through 289 :ref:`#pragma GCC diagnostic <pragma_GCC_diagnostic>`. 290 291.. _opt_fdiagnostics-show-category: 292 293.. option:: -fdiagnostics-show-category=none/id/name 294 295 Enable printing category information in diagnostic line. 296 297 This option, which defaults to "none", controls whether or not Clang 298 prints the category associated with a diagnostic when emitting it. 299 Each diagnostic may or many not have an associated category, if it 300 has one, it is listed in the diagnostic categorization field of the 301 diagnostic line (in the []'s). 302 303 For example, a format string warning will produce these three 304 renditions based on the setting of this option: 305 306 :: 307 308 t.c:3:11: warning: conversion specifies type 'char *' but the argument has type 'int' [-Wformat] 309 t.c:3:11: warning: conversion specifies type 'char *' but the argument has type 'int' [-Wformat,1] 310 t.c:3:11: warning: conversion specifies type 'char *' but the argument has type 'int' [-Wformat,Format String] 311 312 This category can be used by clients that want to group diagnostics 313 by category, so it should be a high level category. We want dozens 314 of these, not hundreds or thousands of them. 315 316.. _opt_fdiagnostics-fixit-info: 317 318**-f[no-]diagnostics-fixit-info** 319 Enable "FixIt" information in the diagnostics output. 320 321 This option, which defaults to on, controls whether or not Clang 322 prints the information on how to fix a specific diagnostic 323 underneath it when it knows. For example, in this output: 324 325 :: 326 327 test.c:28:8: warning: extra tokens at end of #endif directive [-Wextra-tokens] 328 #endif bad 329 ^ 330 // 331 332 Passing **-fno-diagnostics-fixit-info** will prevent Clang from 333 printing the "//" line at the end of the message. This information 334 is useful for users who may not understand what is wrong, but can be 335 confusing for machine parsing. 336 337.. _opt_fdiagnostics-print-source-range-info: 338 339**-fdiagnostics-print-source-range-info** 340 Print machine parsable information about source ranges. 341 This option makes Clang print information about source ranges in a machine 342 parsable format after the file/line/column number information. The 343 information is a simple sequence of brace enclosed ranges, where each range 344 lists the start and end line/column locations. For example, in this output: 345 346 :: 347 348 exprs.c:47:15:{47:8-47:14}{47:17-47:24}: error: invalid operands to binary expression ('int *' and '_Complex float') 349 P = (P-42) + Gamma*4; 350 ~~~~~~ ^ ~~~~~~~ 351 352 The {}'s are generated by -fdiagnostics-print-source-range-info. 353 354 The printed column numbers count bytes from the beginning of the 355 line; take care if your source contains multibyte characters. 356 357.. option:: -fdiagnostics-parseable-fixits 358 359 Print Fix-Its in a machine parseable form. 360 361 This option makes Clang print available Fix-Its in a machine 362 parseable format at the end of diagnostics. The following example 363 illustrates the format: 364 365 :: 366 367 fix-it:"t.cpp":{7:25-7:29}:"Gamma" 368 369 The range printed is a half-open range, so in this example the 370 characters at column 25 up to but not including column 29 on line 7 371 in t.cpp should be replaced with the string "Gamma". Either the 372 range or the replacement string may be empty (representing strict 373 insertions and strict erasures, respectively). Both the file name 374 and the insertion string escape backslash (as "\\\\"), tabs (as 375 "\\t"), newlines (as "\\n"), double quotes(as "\\"") and 376 non-printable characters (as octal "\\xxx"). 377 378 The printed column numbers count bytes from the beginning of the 379 line; take care if your source contains multibyte characters. 380 381.. option:: -fno-elide-type 382 383 Turns off elision in template type printing. 384 385 The default for template type printing is to elide as many template 386 arguments as possible, removing those which are the same in both 387 template types, leaving only the differences. Adding this flag will 388 print all the template arguments. If supported by the terminal, 389 highlighting will still appear on differing arguments. 390 391 Default: 392 393 :: 394 395 t.cc:4:5: note: candidate function not viable: no known conversion from 'vector<map<[...], map<float, [...]>>>' to 'vector<map<[...], map<double, [...]>>>' for 1st argument; 396 397 -fno-elide-type: 398 399 :: 400 401 t.cc:4:5: note: candidate function not viable: no known conversion from 'vector<map<int, map<float, int>>>' to 'vector<map<int, map<double, int>>>' for 1st argument; 402 403.. option:: -fdiagnostics-show-template-tree 404 405 Template type diffing prints a text tree. 406 407 For diffing large templated types, this option will cause Clang to 408 display the templates as an indented text tree, one argument per 409 line, with differences marked inline. This is compatible with 410 -fno-elide-type. 411 412 Default: 413 414 :: 415 416 t.cc:4:5: note: candidate function not viable: no known conversion from 'vector<map<[...], map<float, [...]>>>' to 'vector<map<[...], map<double, [...]>>>' for 1st argument; 417 418 With :option:`-fdiagnostics-show-template-tree`: 419 420 :: 421 422 t.cc:4:5: note: candidate function not viable: no known conversion for 1st argument; 423 vector< 424 map< 425 [...], 426 map< 427 [float != double], 428 [...]>>> 429 430.. _cl_diag_warning_groups: 431 432Individual Warning Groups 433^^^^^^^^^^^^^^^^^^^^^^^^^ 434 435TODO: Generate this from tblgen. Define one anchor per warning group. 436 437.. _opt_wextra-tokens: 438 439.. option:: -Wextra-tokens 440 441 Warn about excess tokens at the end of a preprocessor directive. 442 443 This option, which defaults to on, enables warnings about extra 444 tokens at the end of preprocessor directives. For example: 445 446 :: 447 448 test.c:28:8: warning: extra tokens at end of #endif directive [-Wextra-tokens] 449 #endif bad 450 ^ 451 452 These extra tokens are not strictly conforming, and are usually best 453 handled by commenting them out. 454 455.. option:: -Wambiguous-member-template 456 457 Warn about unqualified uses of a member template whose name resolves to 458 another template at the location of the use. 459 460 This option, which defaults to on, enables a warning in the 461 following code: 462 463 :: 464 465 template<typename T> struct set{}; 466 template<typename T> struct trait { typedef const T& type; }; 467 struct Value { 468 template<typename T> void set(typename trait<T>::type value) {} 469 }; 470 void foo() { 471 Value v; 472 v.set<double>(3.2); 473 } 474 475 C++ [basic.lookup.classref] requires this to be an error, but, 476 because it's hard to work around, Clang downgrades it to a warning 477 as an extension. 478 479.. option:: -Wbind-to-temporary-copy 480 481 Warn about an unusable copy constructor when binding a reference to a 482 temporary. 483 484 This option enables warnings about binding a 485 reference to a temporary when the temporary doesn't have a usable 486 copy constructor. For example: 487 488 :: 489 490 struct NonCopyable { 491 NonCopyable(); 492 private: 493 NonCopyable(const NonCopyable&); 494 }; 495 void foo(const NonCopyable&); 496 void bar() { 497 foo(NonCopyable()); // Disallowed in C++98; allowed in C++11. 498 } 499 500 :: 501 502 struct NonCopyable2 { 503 NonCopyable2(); 504 NonCopyable2(NonCopyable2&); 505 }; 506 void foo(const NonCopyable2&); 507 void bar() { 508 foo(NonCopyable2()); // Disallowed in C++98; allowed in C++11. 509 } 510 511 Note that if ``NonCopyable2::NonCopyable2()`` has a default argument 512 whose instantiation produces a compile error, that error will still 513 be a hard error in C++98 mode even if this warning is turned off. 514 515Options to Control Clang Crash Diagnostics 516------------------------------------------ 517 518As unbelievable as it may sound, Clang does crash from time to time. 519Generally, this only occurs to those living on the `bleeding 520edge <http://llvm.org/releases/download.html#svn>`_. Clang goes to great 521lengths to assist you in filing a bug report. Specifically, Clang 522generates preprocessed source file(s) and associated run script(s) upon 523a crash. These files should be attached to a bug report to ease 524reproducibility of the failure. Below are the command line options to 525control the crash diagnostics. 526 527.. option:: -fno-crash-diagnostics 528 529 Disable auto-generation of preprocessed source files during a clang crash. 530 531The -fno-crash-diagnostics flag can be helpful for speeding the process 532of generating a delta reduced test case. 533 534Options to Emit Optimization Reports 535------------------------------------ 536 537Optimization reports trace, at a high-level, all the major decisions 538done by compiler transformations. For instance, when the inliner 539decides to inline function ``foo()`` into ``bar()``, or the loop unroller 540decides to unroll a loop N times, or the vectorizer decides to 541vectorize a loop body. 542 543Clang offers a family of flags which the optimizers can use to emit 544a diagnostic in three cases: 545 5461. When the pass makes a transformation (:option:`-Rpass`). 547 5482. When the pass fails to make a transformation (:option:`-Rpass-missed`). 549 5503. When the pass determines whether or not to make a transformation 551 (:option:`-Rpass-analysis`). 552 553NOTE: Although the discussion below focuses on :option:`-Rpass`, the exact 554same options apply to :option:`-Rpass-missed` and :option:`-Rpass-analysis`. 555 556Since there are dozens of passes inside the compiler, each of these flags 557take a regular expression that identifies the name of the pass which should 558emit the associated diagnostic. For example, to get a report from the inliner, 559compile the code with: 560 561.. code-block:: console 562 563 $ clang -O2 -Rpass=inline code.cc -o code 564 code.cc:4:25: remark: foo inlined into bar [-Rpass=inline] 565 int bar(int j) { return foo(j, j - 2); } 566 ^ 567 568Note that remarks from the inliner are identified with `[-Rpass=inline]`. 569To request a report from every optimization pass, you should use 570:option:`-Rpass=.*` (in fact, you can use any valid POSIX regular 571expression). However, do not expect a report from every transformation 572made by the compiler. Optimization remarks do not really make sense 573outside of the major transformations (e.g., inlining, vectorization, 574loop optimizations) and not every optimization pass supports this 575feature. 576 577Current limitations 578^^^^^^^^^^^^^^^^^^^ 579 5801. Optimization remarks that refer to function names will display the 581 mangled name of the function. Since these remarks are emitted by the 582 back end of the compiler, it does not know anything about the input 583 language, nor its mangling rules. 584 5852. Some source locations are not displayed correctly. The front end has 586 a more detailed source location tracking than the locations included 587 in the debug info (e.g., the front end can locate code inside macro 588 expansions). However, the locations used by :option:`-Rpass` are 589 translated from debug annotations. That translation can be lossy, 590 which results in some remarks having no location information. 591 592Other Options 593------------- 594Clang options that that don't fit neatly into other categories. 595 596.. option:: -MV 597 598 When emitting a dependency file, use formatting conventions appropriate 599 for NMake or Jom. Ignored unless another option causes Clang to emit a 600 dependency file. 601 602When Clang emits a dependency file (e.g., you supplied the -M option) 603most filenames can be written to the file without any special formatting. 604Different Make tools will treat different sets of characters as "special" 605and use different conventions for telling the Make tool that the character 606is actually part of the filename. Normally Clang uses backslash to "escape" 607a special character, which is the convention used by GNU Make. The -MV 608option tells Clang to put double-quotes around the entire filename, which 609is the convention used by NMake and Jom. 610 611 612Language and Target-Independent Features 613======================================== 614 615Controlling Errors and Warnings 616------------------------------- 617 618Clang provides a number of ways to control which code constructs cause 619it to emit errors and warning messages, and how they are displayed to 620the console. 621 622Controlling How Clang Displays Diagnostics 623^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 624 625When Clang emits a diagnostic, it includes rich information in the 626output, and gives you fine-grain control over which information is 627printed. Clang has the ability to print this information, and these are 628the options that control it: 629 630#. A file/line/column indicator that shows exactly where the diagnostic 631 occurs in your code [:ref:`-fshow-column <opt_fshow-column>`, 632 :ref:`-fshow-source-location <opt_fshow-source-location>`]. 633#. A categorization of the diagnostic as a note, warning, error, or 634 fatal error. 635#. A text string that describes what the problem is. 636#. An option that indicates how to control the diagnostic (for 637 diagnostics that support it) 638 [:ref:`-fdiagnostics-show-option <opt_fdiagnostics-show-option>`]. 639#. A :ref:`high-level category <diagnostics_categories>` for the diagnostic 640 for clients that want to group diagnostics by class (for diagnostics 641 that support it) 642 [:ref:`-fdiagnostics-show-category <opt_fdiagnostics-show-category>`]. 643#. The line of source code that the issue occurs on, along with a caret 644 and ranges that indicate the important locations 645 [:ref:`-fcaret-diagnostics <opt_fcaret-diagnostics>`]. 646#. "FixIt" information, which is a concise explanation of how to fix the 647 problem (when Clang is certain it knows) 648 [:ref:`-fdiagnostics-fixit-info <opt_fdiagnostics-fixit-info>`]. 649#. A machine-parsable representation of the ranges involved (off by 650 default) 651 [:ref:`-fdiagnostics-print-source-range-info <opt_fdiagnostics-print-source-range-info>`]. 652 653For more information please see :ref:`Formatting of 654Diagnostics <cl_diag_formatting>`. 655 656Diagnostic Mappings 657^^^^^^^^^^^^^^^^^^^ 658 659All diagnostics are mapped into one of these 6 classes: 660 661- Ignored 662- Note 663- Remark 664- Warning 665- Error 666- Fatal 667 668.. _diagnostics_categories: 669 670Diagnostic Categories 671^^^^^^^^^^^^^^^^^^^^^ 672 673Though not shown by default, diagnostics may each be associated with a 674high-level category. This category is intended to make it possible to 675triage builds that produce a large number of errors or warnings in a 676grouped way. 677 678Categories are not shown by default, but they can be turned on with the 679:ref:`-fdiagnostics-show-category <opt_fdiagnostics-show-category>` option. 680When set to "``name``", the category is printed textually in the 681diagnostic output. When it is set to "``id``", a category number is 682printed. The mapping of category names to category id's can be obtained 683by running '``clang --print-diagnostic-categories``'. 684 685Controlling Diagnostics via Command Line Flags 686^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 687 688TODO: -W flags, -pedantic, etc 689 690.. _pragma_gcc_diagnostic: 691 692Controlling Diagnostics via Pragmas 693^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 694 695Clang can also control what diagnostics are enabled through the use of 696pragmas in the source code. This is useful for turning off specific 697warnings in a section of source code. Clang supports GCC's pragma for 698compatibility with existing source code, as well as several extensions. 699 700The pragma may control any warning that can be used from the command 701line. Warnings may be set to ignored, warning, error, or fatal. The 702following example code will tell Clang or GCC to ignore the -Wall 703warnings: 704 705.. code-block:: c 706 707 #pragma GCC diagnostic ignored "-Wall" 708 709In addition to all of the functionality provided by GCC's pragma, Clang 710also allows you to push and pop the current warning state. This is 711particularly useful when writing a header file that will be compiled by 712other people, because you don't know what warning flags they build with. 713 714In the below example :option:`-Wmultichar` is ignored for only a single line of 715code, after which the diagnostics return to whatever state had previously 716existed. 717 718.. code-block:: c 719 720 #pragma clang diagnostic push 721 #pragma clang diagnostic ignored "-Wmultichar" 722 723 char b = 'df'; // no warning. 724 725 #pragma clang diagnostic pop 726 727The push and pop pragmas will save and restore the full diagnostic state 728of the compiler, regardless of how it was set. That means that it is 729possible to use push and pop around GCC compatible diagnostics and Clang 730will push and pop them appropriately, while GCC will ignore the pushes 731and pops as unknown pragmas. It should be noted that while Clang 732supports the GCC pragma, Clang and GCC do not support the exact same set 733of warnings, so even when using GCC compatible #pragmas there is no 734guarantee that they will have identical behaviour on both compilers. 735 736In addition to controlling warnings and errors generated by the compiler, it is 737possible to generate custom warning and error messages through the following 738pragmas: 739 740.. code-block:: c 741 742 // The following will produce warning messages 743 #pragma message "some diagnostic message" 744 #pragma GCC warning "TODO: replace deprecated feature" 745 746 // The following will produce an error message 747 #pragma GCC error "Not supported" 748 749These pragmas operate similarly to the ``#warning`` and ``#error`` preprocessor 750directives, except that they may also be embedded into preprocessor macros via 751the C99 ``_Pragma`` operator, for example: 752 753.. code-block:: c 754 755 #define STR(X) #X 756 #define DEFER(M,...) M(__VA_ARGS__) 757 #define CUSTOM_ERROR(X) _Pragma(STR(GCC error(X " at line " DEFER(STR,__LINE__)))) 758 759 CUSTOM_ERROR("Feature not available"); 760 761Controlling Diagnostics in System Headers 762^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 763 764Warnings are suppressed when they occur in system headers. By default, 765an included file is treated as a system header if it is found in an 766include path specified by ``-isystem``, but this can be overridden in 767several ways. 768 769The ``system_header`` pragma can be used to mark the current file as 770being a system header. No warnings will be produced from the location of 771the pragma onwards within the same file. 772 773.. code-block:: c 774 775 char a = 'xy'; // warning 776 777 #pragma clang system_header 778 779 char b = 'ab'; // no warning 780 781The :option:`--system-header-prefix=` and :option:`--no-system-header-prefix=` 782command-line arguments can be used to override whether subsets of an include 783path are treated as system headers. When the name in a ``#include`` directive 784is found within a header search path and starts with a system prefix, the 785header is treated as a system header. The last prefix on the 786command-line which matches the specified header name takes precedence. 787For instance: 788 789.. code-block:: console 790 791 $ clang -Ifoo -isystem bar --system-header-prefix=x/ \ 792 --no-system-header-prefix=x/y/ 793 794Here, ``#include "x/a.h"`` is treated as including a system header, even 795if the header is found in ``foo``, and ``#include "x/y/b.h"`` is treated 796as not including a system header, even if the header is found in 797``bar``. 798 799A ``#include`` directive which finds a file relative to the current 800directory is treated as including a system header if the including file 801is treated as a system header. 802 803.. _diagnostics_enable_everything: 804 805Enabling All Diagnostics 806^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 807 808In addition to the traditional ``-W`` flags, one can enable **all** 809diagnostics by passing :option:`-Weverything`. This works as expected 810with 811:option:`-Werror`, and also includes the warnings from :option:`-pedantic`. 812 813Note that when combined with :option:`-w` (which disables all warnings), that 814flag wins. 815 816Controlling Static Analyzer Diagnostics 817^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 818 819While not strictly part of the compiler, the diagnostics from Clang's 820`static analyzer <http://clang-analyzer.llvm.org>`_ can also be 821influenced by the user via changes to the source code. See the available 822`annotations <http://clang-analyzer.llvm.org/annotations.html>`_ and the 823analyzer's `FAQ 824page <http://clang-analyzer.llvm.org/faq.html#exclude_code>`_ for more 825information. 826 827.. _usersmanual-precompiled-headers: 828 829Precompiled Headers 830------------------- 831 832`Precompiled headers <http://en.wikipedia.org/wiki/Precompiled_header>`__ 833are a general approach employed by many compilers to reduce compilation 834time. The underlying motivation of the approach is that it is common for 835the same (and often large) header files to be included by multiple 836source files. Consequently, compile times can often be greatly improved 837by caching some of the (redundant) work done by a compiler to process 838headers. Precompiled header files, which represent one of many ways to 839implement this optimization, are literally files that represent an 840on-disk cache that contains the vital information necessary to reduce 841some of the work needed to process a corresponding header file. While 842details of precompiled headers vary between compilers, precompiled 843headers have been shown to be highly effective at speeding up program 844compilation on systems with very large system headers (e.g., Mac OS X). 845 846Generating a PCH File 847^^^^^^^^^^^^^^^^^^^^^ 848 849To generate a PCH file using Clang, one invokes Clang with the 850:option:`-x <language>-header` option. This mirrors the interface in GCC 851for generating PCH files: 852 853.. code-block:: console 854 855 $ gcc -x c-header test.h -o test.h.gch 856 $ clang -x c-header test.h -o test.h.pch 857 858Using a PCH File 859^^^^^^^^^^^^^^^^ 860 861A PCH file can then be used as a prefix header when a :option:`-include` 862option is passed to ``clang``: 863 864.. code-block:: console 865 866 $ clang -include test.h test.c -o test 867 868The ``clang`` driver will first check if a PCH file for ``test.h`` is 869available; if so, the contents of ``test.h`` (and the files it includes) 870will be processed from the PCH file. Otherwise, Clang falls back to 871directly processing the content of ``test.h``. This mirrors the behavior 872of GCC. 873 874.. note:: 875 876 Clang does *not* automatically use PCH files for headers that are directly 877 included within a source file. For example: 878 879 .. code-block:: console 880 881 $ clang -x c-header test.h -o test.h.pch 882 $ cat test.c 883 #include "test.h" 884 $ clang test.c -o test 885 886 In this example, ``clang`` will not automatically use the PCH file for 887 ``test.h`` since ``test.h`` was included directly in the source file and not 888 specified on the command line using :option:`-include`. 889 890Relocatable PCH Files 891^^^^^^^^^^^^^^^^^^^^^ 892 893It is sometimes necessary to build a precompiled header from headers 894that are not yet in their final, installed locations. For example, one 895might build a precompiled header within the build tree that is then 896meant to be installed alongside the headers. Clang permits the creation 897of "relocatable" precompiled headers, which are built with a given path 898(into the build directory) and can later be used from an installed 899location. 900 901To build a relocatable precompiled header, place your headers into a 902subdirectory whose structure mimics the installed location. For example, 903if you want to build a precompiled header for the header ``mylib.h`` 904that will be installed into ``/usr/include``, create a subdirectory 905``build/usr/include`` and place the header ``mylib.h`` into that 906subdirectory. If ``mylib.h`` depends on other headers, then they can be 907stored within ``build/usr/include`` in a way that mimics the installed 908location. 909 910Building a relocatable precompiled header requires two additional 911arguments. First, pass the ``--relocatable-pch`` flag to indicate that 912the resulting PCH file should be relocatable. Second, pass 913:option:`-isysroot /path/to/build`, which makes all includes for your library 914relative to the build directory. For example: 915 916.. code-block:: console 917 918 # clang -x c-header --relocatable-pch -isysroot /path/to/build /path/to/build/mylib.h mylib.h.pch 919 920When loading the relocatable PCH file, the various headers used in the 921PCH file are found from the system header root. For example, ``mylib.h`` 922can be found in ``/usr/include/mylib.h``. If the headers are installed 923in some other system root, the :option:`-isysroot` option can be used provide 924a different system root from which the headers will be based. For 925example, :option:`-isysroot /Developer/SDKs/MacOSX10.4u.sdk` will look for 926``mylib.h`` in ``/Developer/SDKs/MacOSX10.4u.sdk/usr/include/mylib.h``. 927 928Relocatable precompiled headers are intended to be used in a limited 929number of cases where the compilation environment is tightly controlled 930and the precompiled header cannot be generated after headers have been 931installed. 932 933.. _controlling-code-generation: 934 935Controlling Code Generation 936--------------------------- 937 938Clang provides a number of ways to control code generation. The options 939are listed below. 940 941**-f[no-]sanitize=check1,check2,...** 942 Turn on runtime checks for various forms of undefined or suspicious 943 behavior. 944 945 This option controls whether Clang adds runtime checks for various 946 forms of undefined or suspicious behavior, and is disabled by 947 default. If a check fails, a diagnostic message is produced at 948 runtime explaining the problem. The main checks are: 949 950 - .. _opt_fsanitize_address: 951 952 ``-fsanitize=address``: 953 :doc:`AddressSanitizer`, a memory error 954 detector. 955 - .. _opt_fsanitize_thread: 956 957 ``-fsanitize=thread``: :doc:`ThreadSanitizer`, a data race detector. 958 - .. _opt_fsanitize_memory: 959 960 ``-fsanitize=memory``: :doc:`MemorySanitizer`, 961 a detector of uninitialized reads. Requires instrumentation of all 962 program code. 963 - .. _opt_fsanitize_undefined: 964 965 ``-fsanitize=undefined``: :doc:`UndefinedBehaviorSanitizer`, 966 a fast and compatible undefined behavior checker. 967 968 - ``-fsanitize=dataflow``: :doc:`DataFlowSanitizer`, a general data 969 flow analysis. 970 - ``-fsanitize=cfi``: :doc:`control flow integrity <ControlFlowIntegrity>` 971 checks. Requires ``-flto``. 972 - ``-fsanitize=safe-stack``: :doc:`safe stack <SafeStack>` 973 protection against stack-based memory corruption errors. 974 975 There are more fine-grained checks available: see 976 the :ref:`list <ubsan-checks>` of specific kinds of 977 undefined behavior that can be detected and the :ref:`list <cfi-schemes>` 978 of control flow integrity schemes. 979 980 The ``-fsanitize=`` argument must also be provided when linking, in 981 order to link to the appropriate runtime library. 982 983 It is not possible to combine more than one of the ``-fsanitize=address``, 984 ``-fsanitize=thread``, and ``-fsanitize=memory`` checkers in the same 985 program. 986 987**-f[no-]sanitize-recover=check1,check2,...** 988 989 Controls which checks enabled by ``-fsanitize=`` flag are non-fatal. 990 If the check is fatal, program will halt after the first error 991 of this kind is detected and error report is printed. 992 993 By default, non-fatal checks are those enabled by 994 :doc:`UndefinedBehaviorSanitizer`, 995 except for ``-fsanitize=return`` and ``-fsanitize=unreachable``. Some 996 sanitizers may not support recovery (or not support it by default 997 e.g. :doc:`AddressSanitizer`), and always crash the program after the issue 998 is detected. 999 1000 Note that the ``-fsanitize-trap`` flag has precedence over this flag. 1001 This means that if a check has been configured to trap elsewhere on the 1002 command line, or if the check traps by default, this flag will not have 1003 any effect unless that sanitizer's trapping behavior is disabled with 1004 ``-fno-sanitize-trap``. 1005 1006 For example, if a command line contains the flags ``-fsanitize=undefined 1007 -fsanitize-trap=undefined``, the flag ``-fsanitize-recover=alignment`` 1008 will have no effect on its own; it will need to be accompanied by 1009 ``-fno-sanitize-trap=alignment``. 1010 1011**-f[no-]sanitize-trap=check1,check2,...** 1012 1013 Controls which checks enabled by the ``-fsanitize=`` flag trap. This 1014 option is intended for use in cases where the sanitizer runtime cannot 1015 be used (for instance, when building libc or a kernel module), or where 1016 the binary size increase caused by the sanitizer runtime is a concern. 1017 1018 This flag is only compatible with :doc:`control flow integrity 1019 <ControlFlowIntegrity>` schemes and :doc:`UndefinedBehaviorSanitizer` 1020 checks other than ``vptr``. If this flag 1021 is supplied together with ``-fsanitize=undefined``, the ``vptr`` sanitizer 1022 will be implicitly disabled. 1023 1024 This flag is enabled by default for sanitizers in the ``cfi`` group. 1025 1026.. option:: -fsanitize-blacklist=/path/to/blacklist/file 1027 1028 Disable or modify sanitizer checks for objects (source files, functions, 1029 variables, types) listed in the file. See 1030 :doc:`SanitizerSpecialCaseList` for file format description. 1031 1032.. option:: -fno-sanitize-blacklist 1033 1034 Don't use blacklist file, if it was specified earlier in the command line. 1035 1036**-f[no-]sanitize-coverage=[type,features,...]** 1037 1038 Enable simple code coverage in addition to certain sanitizers. 1039 See :doc:`SanitizerCoverage` for more details. 1040 1041.. option:: -fsanitize-undefined-trap-on-error 1042 1043 Deprecated alias for ``-fsanitize-trap=undefined``. 1044 1045.. option:: -fsanitize-cfi-cross-dso 1046 1047 Enable cross-DSO control flow integrity checks. This flag modifies 1048 the behavior of sanitizers in the ``cfi`` group to allow checking 1049 of cross-DSO virtual and indirect calls. 1050 1051.. option:: -fno-assume-sane-operator-new 1052 1053 Don't assume that the C++'s new operator is sane. 1054 1055 This option tells the compiler to do not assume that C++'s global 1056 new operator will always return a pointer that does not alias any 1057 other pointer when the function returns. 1058 1059.. option:: -ftrap-function=[name] 1060 1061 Instruct code generator to emit a function call to the specified 1062 function name for ``__builtin_trap()``. 1063 1064 LLVM code generator translates ``__builtin_trap()`` to a trap 1065 instruction if it is supported by the target ISA. Otherwise, the 1066 builtin is translated into a call to ``abort``. If this option is 1067 set, then the code generator will always lower the builtin to a call 1068 to the specified function regardless of whether the target ISA has a 1069 trap instruction. This option is useful for environments (e.g. 1070 deeply embedded) where a trap cannot be properly handled, or when 1071 some custom behavior is desired. 1072 1073.. option:: -ftls-model=[model] 1074 1075 Select which TLS model to use. 1076 1077 Valid values are: ``global-dynamic``, ``local-dynamic``, 1078 ``initial-exec`` and ``local-exec``. The default value is 1079 ``global-dynamic``. The compiler may use a different model if the 1080 selected model is not supported by the target, or if a more 1081 efficient model can be used. The TLS model can be overridden per 1082 variable using the ``tls_model`` attribute. 1083 1084.. option:: -femulated-tls 1085 1086 Select emulated TLS model, which overrides all -ftls-model choices. 1087 1088 In emulated TLS mode, all access to TLS variables are converted to 1089 calls to __emutls_get_address in the runtime library. 1090 1091.. option:: -mhwdiv=[values] 1092 1093 Select the ARM modes (arm or thumb) that support hardware division 1094 instructions. 1095 1096 Valid values are: ``arm``, ``thumb`` and ``arm,thumb``. 1097 This option is used to indicate which mode (arm or thumb) supports 1098 hardware division instructions. This only applies to the ARM 1099 architecture. 1100 1101.. option:: -m[no-]crc 1102 1103 Enable or disable CRC instructions. 1104 1105 This option is used to indicate whether CRC instructions are to 1106 be generated. This only applies to the ARM architecture. 1107 1108 CRC instructions are enabled by default on ARMv8. 1109 1110.. option:: -mgeneral-regs-only 1111 1112 Generate code which only uses the general purpose registers. 1113 1114 This option restricts the generated code to use general registers 1115 only. This only applies to the AArch64 architecture. 1116 1117**-f[no-]max-type-align=[number]** 1118 Instruct the code generator to not enforce a higher alignment than the given 1119 number (of bytes) when accessing memory via an opaque pointer or reference. 1120 This cap is ignored when directly accessing a variable or when the pointee 1121 type has an explicit “aligned” attribute. 1122 1123 The value should usually be determined by the properties of the system allocator. 1124 Some builtin types, especially vector types, have very high natural alignments; 1125 when working with values of those types, Clang usually wants to use instructions 1126 that take advantage of that alignment. However, many system allocators do 1127 not promise to return memory that is more than 8-byte or 16-byte-aligned. Use 1128 this option to limit the alignment that the compiler can assume for an arbitrary 1129 pointer, which may point onto the heap. 1130 1131 This option does not affect the ABI alignment of types; the layout of structs and 1132 unions and the value returned by the alignof operator remain the same. 1133 1134 This option can be overridden on a case-by-case basis by putting an explicit 1135 “aligned” alignment on a struct, union, or typedef. For example: 1136 1137 .. code-block:: console 1138 1139 #include <immintrin.h> 1140 // Make an aligned typedef of the AVX-512 16-int vector type. 1141 typedef __v16si __aligned_v16si __attribute__((aligned(64))); 1142 1143 void initialize_vector(__aligned_v16si *v) { 1144 // The compiler may assume that ‘v’ is 64-byte aligned, regardless of the 1145 // value of -fmax-type-align. 1146 } 1147 1148 1149Profile Guided Optimization 1150--------------------------- 1151 1152Profile information enables better optimization. For example, knowing that a 1153branch is taken very frequently helps the compiler make better decisions when 1154ordering basic blocks. Knowing that a function ``foo`` is called more 1155frequently than another function ``bar`` helps the inliner. 1156 1157Clang supports profile guided optimization with two different kinds of 1158profiling. A sampling profiler can generate a profile with very low runtime 1159overhead, or you can build an instrumented version of the code that collects 1160more detailed profile information. Both kinds of profiles can provide execution 1161counts for instructions in the code and information on branches taken and 1162function invocation. 1163 1164Regardless of which kind of profiling you use, be careful to collect profiles 1165by running your code with inputs that are representative of the typical 1166behavior. Code that is not exercised in the profile will be optimized as if it 1167is unimportant, and the compiler may make poor optimization choices for code 1168that is disproportionately used while profiling. 1169 1170Differences Between Sampling and Instrumentation 1171^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1172 1173Although both techniques are used for similar purposes, there are important 1174differences between the two: 1175 11761. Profile data generated with one cannot be used by the other, and there is no 1177 conversion tool that can convert one to the other. So, a profile generated 1178 via ``-fprofile-instr-generate`` must be used with ``-fprofile-instr-use``. 1179 Similarly, sampling profiles generated by external profilers must be 1180 converted and used with ``-fprofile-sample-use``. 1181 11822. Instrumentation profile data can be used for code coverage analysis and 1183 optimization. 1184 11853. Sampling profiles can only be used for optimization. They cannot be used for 1186 code coverage analysis. Although it would be technically possible to use 1187 sampling profiles for code coverage, sample-based profiles are too 1188 coarse-grained for code coverage purposes; it would yield poor results. 1189 11904. Sampling profiles must be generated by an external tool. The profile 1191 generated by that tool must then be converted into a format that can be read 1192 by LLVM. The section on sampling profilers describes one of the supported 1193 sampling profile formats. 1194 1195 1196Using Sampling Profilers 1197^^^^^^^^^^^^^^^^^^^^^^^^ 1198 1199Sampling profilers are used to collect runtime information, such as 1200hardware counters, while your application executes. They are typically 1201very efficient and do not incur a large runtime overhead. The 1202sample data collected by the profiler can be used during compilation 1203to determine what the most executed areas of the code are. 1204 1205Using the data from a sample profiler requires some changes in the way 1206a program is built. Before the compiler can use profiling information, 1207the code needs to execute under the profiler. The following is the 1208usual build cycle when using sample profilers for optimization: 1209 12101. Build the code with source line table information. You can use all the 1211 usual build flags that you always build your application with. The only 1212 requirement is that you add ``-gline-tables-only`` or ``-g`` to the 1213 command line. This is important for the profiler to be able to map 1214 instructions back to source line locations. 1215 1216 .. code-block:: console 1217 1218 $ clang++ -O2 -gline-tables-only code.cc -o code 1219 12202. Run the executable under a sampling profiler. The specific profiler 1221 you use does not really matter, as long as its output can be converted 1222 into the format that the LLVM optimizer understands. Currently, there 1223 exists a conversion tool for the Linux Perf profiler 1224 (https://perf.wiki.kernel.org/), so these examples assume that you 1225 are using Linux Perf to profile your code. 1226 1227 .. code-block:: console 1228 1229 $ perf record -b ./code 1230 1231 Note the use of the ``-b`` flag. This tells Perf to use the Last Branch 1232 Record (LBR) to record call chains. While this is not strictly required, 1233 it provides better call information, which improves the accuracy of 1234 the profile data. 1235 12363. Convert the collected profile data to LLVM's sample profile format. 1237 This is currently supported via the AutoFDO converter ``create_llvm_prof``. 1238 It is available at http://github.com/google/autofdo. Once built and 1239 installed, you can convert the ``perf.data`` file to LLVM using 1240 the command: 1241 1242 .. code-block:: console 1243 1244 $ create_llvm_prof --binary=./code --out=code.prof 1245 1246 This will read ``perf.data`` and the binary file ``./code`` and emit 1247 the profile data in ``code.prof``. Note that if you ran ``perf`` 1248 without the ``-b`` flag, you need to use ``--use_lbr=false`` when 1249 calling ``create_llvm_prof``. 1250 12514. Build the code again using the collected profile. This step feeds 1252 the profile back to the optimizers. This should result in a binary 1253 that executes faster than the original one. Note that you are not 1254 required to build the code with the exact same arguments that you 1255 used in the first step. The only requirement is that you build the code 1256 with ``-gline-tables-only`` and ``-fprofile-sample-use``. 1257 1258 .. code-block:: console 1259 1260 $ clang++ -O2 -gline-tables-only -fprofile-sample-use=code.prof code.cc -o code 1261 1262 1263Sample Profile Formats 1264"""""""""""""""""""""" 1265 1266Since external profilers generate profile data in a variety of custom formats, 1267the data generated by the profiler must be converted into a format that can be 1268read by the backend. LLVM supports three different sample profile formats: 1269 12701. ASCII text. This is the easiest one to generate. The file is divided into 1271 sections, which correspond to each of the functions with profile 1272 information. The format is described below. It can also be generated from 1273 the binary or gcov formats using the ``llvm-profdata`` tool. 1274 12752. Binary encoding. This uses a more efficient encoding that yields smaller 1276 profile files. This is the format generated by the ``create_llvm_prof`` tool 1277 in http://github.com/google/autofdo. 1278 12793. GCC encoding. This is based on the gcov format, which is accepted by GCC. It 1280 is only interesting in environments where GCC and Clang co-exist. This 1281 encoding is only generated by the ``create_gcov`` tool in 1282 http://github.com/google/autofdo. It can be read by LLVM and 1283 ``llvm-profdata``, but it cannot be generated by either. 1284 1285If you are using Linux Perf to generate sampling profiles, you can use the 1286conversion tool ``create_llvm_prof`` described in the previous section. 1287Otherwise, you will need to write a conversion tool that converts your 1288profiler's native format into one of these three. 1289 1290 1291Sample Profile Text Format 1292"""""""""""""""""""""""""" 1293 1294This section describes the ASCII text format for sampling profiles. It is, 1295arguably, the easiest one to generate. If you are interested in generating any 1296of the other two, consult the ``ProfileData`` library in in LLVM's source tree 1297(specifically, ``include/llvm/ProfileData/SampleProfReader.h``). 1298 1299.. code-block:: console 1300 1301 function1:total_samples:total_head_samples 1302 offset1[.discriminator]: number_of_samples [fn1:num fn2:num ... ] 1303 offset2[.discriminator]: number_of_samples [fn3:num fn4:num ... ] 1304 ... 1305 offsetN[.discriminator]: number_of_samples [fn5:num fn6:num ... ] 1306 offsetA[.discriminator]: fnA:num_of_total_samples 1307 offsetA1[.discriminator]: number_of_samples [fn7:num fn8:num ... ] 1308 offsetA1[.discriminator]: number_of_samples [fn9:num fn10:num ... ] 1309 offsetB[.discriminator]: fnB:num_of_total_samples 1310 offsetB1[.discriminator]: number_of_samples [fn11:num fn12:num ... ] 1311 1312This is a nested tree in which the identation represents the nesting level 1313of the inline stack. There are no blank lines in the file. And the spacing 1314within a single line is fixed. Additional spaces will result in an error 1315while reading the file. 1316 1317Any line starting with the '#' character is completely ignored. 1318 1319Inlined calls are represented with indentation. The Inline stack is a 1320stack of source locations in which the top of the stack represents the 1321leaf function, and the bottom of the stack represents the actual 1322symbol to which the instruction belongs. 1323 1324Function names must be mangled in order for the profile loader to 1325match them in the current translation unit. The two numbers in the 1326function header specify how many total samples were accumulated in the 1327function (first number), and the total number of samples accumulated 1328in the prologue of the function (second number). This head sample 1329count provides an indicator of how frequently the function is invoked. 1330 1331There are two types of lines in the function body. 1332 1333- Sampled line represents the profile information of a source location. 1334 ``offsetN[.discriminator]: number_of_samples [fn5:num fn6:num ... ]`` 1335 1336- Callsite line represents the profile information of an inlined callsite. 1337 ``offsetA[.discriminator]: fnA:num_of_total_samples`` 1338 1339Each sampled line may contain several items. Some are optional (marked 1340below): 1341 1342a. Source line offset. This number represents the line number 1343 in the function where the sample was collected. The line number is 1344 always relative to the line where symbol of the function is 1345 defined. So, if the function has its header at line 280, the offset 1346 13 is at line 293 in the file. 1347 1348 Note that this offset should never be a negative number. This could 1349 happen in cases like macros. The debug machinery will register the 1350 line number at the point of macro expansion. So, if the macro was 1351 expanded in a line before the start of the function, the profile 1352 converter should emit a 0 as the offset (this means that the optimizers 1353 will not be able to associate a meaningful weight to the instructions 1354 in the macro). 1355 1356b. [OPTIONAL] Discriminator. This is used if the sampled program 1357 was compiled with DWARF discriminator support 1358 (http://wiki.dwarfstd.org/index.php?title=Path_Discriminators). 1359 DWARF discriminators are unsigned integer values that allow the 1360 compiler to distinguish between multiple execution paths on the 1361 same source line location. 1362 1363 For example, consider the line of code ``if (cond) foo(); else bar();``. 1364 If the predicate ``cond`` is true 80% of the time, then the edge 1365 into function ``foo`` should be considered to be taken most of the 1366 time. But both calls to ``foo`` and ``bar`` are at the same source 1367 line, so a sample count at that line is not sufficient. The 1368 compiler needs to know which part of that line is taken more 1369 frequently. 1370 1371 This is what discriminators provide. In this case, the calls to 1372 ``foo`` and ``bar`` will be at the same line, but will have 1373 different discriminator values. This allows the compiler to correctly 1374 set edge weights into ``foo`` and ``bar``. 1375 1376c. Number of samples. This is an integer quantity representing the 1377 number of samples collected by the profiler at this source 1378 location. 1379 1380d. [OPTIONAL] Potential call targets and samples. If present, this 1381 line contains a call instruction. This models both direct and 1382 number of samples. For example, 1383 1384 .. code-block:: console 1385 1386 130: 7 foo:3 bar:2 baz:7 1387 1388 The above means that at relative line offset 130 there is a call 1389 instruction that calls one of ``foo()``, ``bar()`` and ``baz()``, 1390 with ``baz()`` being the relatively more frequently called target. 1391 1392As an example, consider a program with the call chain ``main -> foo -> bar``. 1393When built with optimizations enabled, the compiler may inline the 1394calls to ``bar`` and ``foo`` inside ``main``. The generated profile 1395could then be something like this: 1396 1397.. code-block:: console 1398 1399 main:35504:0 1400 1: _Z3foov:35504 1401 2: _Z32bari:31977 1402 1.1: 31977 1403 2: 0 1404 1405This profile indicates that there were a total of 35,504 samples 1406collected in main. All of those were at line 1 (the call to ``foo``). 1407Of those, 31,977 were spent inside the body of ``bar``. The last line 1408of the profile (``2: 0``) corresponds to line 2 inside ``main``. No 1409samples were collected there. 1410 1411Profiling with Instrumentation 1412^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1413 1414Clang also supports profiling via instrumentation. This requires building a 1415special instrumented version of the code and has some runtime 1416overhead during the profiling, but it provides more detailed results than a 1417sampling profiler. It also provides reproducible results, at least to the 1418extent that the code behaves consistently across runs. 1419 1420Here are the steps for using profile guided optimization with 1421instrumentation: 1422 14231. Build an instrumented version of the code by compiling and linking with the 1424 ``-fprofile-instr-generate`` option. 1425 1426 .. code-block:: console 1427 1428 $ clang++ -O2 -fprofile-instr-generate code.cc -o code 1429 14302. Run the instrumented executable with inputs that reflect the typical usage. 1431 By default, the profile data will be written to a ``default.profraw`` file 1432 in the current directory. You can override that default by setting the 1433 ``LLVM_PROFILE_FILE`` environment variable to specify an alternate file. 1434 Any instance of ``%p`` in that file name will be replaced by the process 1435 ID, so that you can easily distinguish the profile output from multiple 1436 runs. 1437 1438 .. code-block:: console 1439 1440 $ LLVM_PROFILE_FILE="code-%p.profraw" ./code 1441 14423. Combine profiles from multiple runs and convert the "raw" profile format to 1443 the input expected by clang. Use the ``merge`` command of the 1444 ``llvm-profdata`` tool to do this. 1445 1446 .. code-block:: console 1447 1448 $ llvm-profdata merge -output=code.profdata code-*.profraw 1449 1450 Note that this step is necessary even when there is only one "raw" profile, 1451 since the merge operation also changes the file format. 1452 14534. Build the code again using the ``-fprofile-instr-use`` option to specify the 1454 collected profile data. 1455 1456 .. code-block:: console 1457 1458 $ clang++ -O2 -fprofile-instr-use=code.profdata code.cc -o code 1459 1460 You can repeat step 4 as often as you like without regenerating the 1461 profile. As you make changes to your code, clang may no longer be able to 1462 use the profile data. It will warn you when this happens. 1463 1464Profile generation and use can also be controlled by the GCC-compatible flags 1465``-fprofile-generate`` and ``-fprofile-use``. Although these flags are 1466semantically equivalent to their GCC counterparts, they *do not* handle 1467GCC-compatible profiles. They are only meant to implement GCC's semantics 1468with respect to profile creation and use. 1469 1470.. option:: -fprofile-generate[=<dirname>] 1471 1472 Without any other arguments, ``-fprofile-generate`` behaves identically to 1473 ``-fprofile-instr-generate``. When given a directory name, it generates the 1474 profile file ``default.profraw`` in the directory named ``dirname``. If 1475 ``dirname`` does not exist, it will be created at runtime. The environment 1476 variable ``LLVM_PROFILE_FILE`` can be used to override the directory and 1477 filename for the profile file at runtime. For example, 1478 1479 .. code-block:: console 1480 1481 $ clang++ -O2 -fprofile-generate=yyy/zzz code.cc -o code 1482 1483 When ``code`` is executed, the profile will be written to the file 1484 ``yyy/zzz/default.profraw``. This can be altered at runtime via the 1485 ``LLVM_PROFILE_FILE`` environment variable: 1486 1487 .. code-block:: console 1488 1489 $ LLVM_PROFILE_FILE=/tmp/myprofile/code.profraw ./code 1490 1491 The above invocation will produce the profile file 1492 ``/tmp/myprofile/code.profraw`` instead of ``yyy/zzz/default.profraw``. 1493 Notice that ``LLVM_PROFILE_FILE`` overrides the directory *and* the file 1494 name for the profile file. 1495 1496.. option:: -fprofile-use[=<pathname>] 1497 1498 Without any other arguments, ``-fprofile-use`` behaves identically to 1499 ``-fprofile-instr-use``. Otherwise, if ``pathname`` is the full path to a 1500 profile file, it reads from that file. If ``pathname`` is a directory name, 1501 it reads from ``pathname/default.profdata``. 1502 1503Disabling Instrumentation 1504^^^^^^^^^^^^^^^^^^^^^^^^^ 1505 1506In certain situations, it may be useful to disable profile generation or use 1507for specific files in a build, without affecting the main compilation flags 1508used for the other files in the project. 1509 1510In these cases, you can use the flag ``-fno-profile-instr-generate`` (or 1511``-fno-profile-generate``) to disable profile generation, and 1512``-fno-profile-instr-use`` (or ``-fno-profile-use``) to disable profile use. 1513 1514Note that these flags should appear after the corresponding profile 1515flags to have an effect. 1516 1517Controlling Debug Information 1518----------------------------- 1519 1520Controlling Size of Debug Information 1521^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1522 1523Debug info kind generated by Clang can be set by one of the flags listed 1524below. If multiple flags are present, the last one is used. 1525 1526.. option:: -g0 1527 1528 Don't generate any debug info (default). 1529 1530.. option:: -gline-tables-only 1531 1532 Generate line number tables only. 1533 1534 This kind of debug info allows to obtain stack traces with function names, 1535 file names and line numbers (by such tools as ``gdb`` or ``addr2line``). It 1536 doesn't contain any other data (e.g. description of local variables or 1537 function parameters). 1538 1539.. option:: -fstandalone-debug 1540 1541 Clang supports a number of optimizations to reduce the size of debug 1542 information in the binary. They work based on the assumption that 1543 the debug type information can be spread out over multiple 1544 compilation units. For instance, Clang will not emit type 1545 definitions for types that are not needed by a module and could be 1546 replaced with a forward declaration. Further, Clang will only emit 1547 type info for a dynamic C++ class in the module that contains the 1548 vtable for the class. 1549 1550 The **-fstandalone-debug** option turns off these optimizations. 1551 This is useful when working with 3rd-party libraries that don't come 1552 with debug information. Note that Clang will never emit type 1553 information for types that are not referenced at all by the program. 1554 1555.. option:: -fno-standalone-debug 1556 1557 On Darwin **-fstandalone-debug** is enabled by default. The 1558 **-fno-standalone-debug** option can be used to get to turn on the 1559 vtable-based optimization described above. 1560 1561.. option:: -g 1562 1563 Generate complete debug info. 1564 1565Controlling Debugger "Tuning" 1566^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1567 1568While Clang generally emits standard DWARF debug info (http://dwarfstd.org), 1569different debuggers may know how to take advantage of different specific DWARF 1570features. You can "tune" the debug info for one of several different debuggers. 1571 1572.. option:: -ggdb, -glldb, -gsce 1573 1574 Tune the debug info for the ``gdb``, ``lldb``, or Sony Computer Entertainment 1575 debugger, respectively. Each of these options implies **-g**. (Therefore, if 1576 you want both **-gline-tables-only** and debugger tuning, the tuning option 1577 must come first.) 1578 1579 1580Comment Parsing Options 1581----------------------- 1582 1583Clang parses Doxygen and non-Doxygen style documentation comments and attaches 1584them to the appropriate declaration nodes. By default, it only parses 1585Doxygen-style comments and ignores ordinary comments starting with ``//`` and 1586``/*``. 1587 1588.. option:: -Wdocumentation 1589 1590 Emit warnings about use of documentation comments. This warning group is off 1591 by default. 1592 1593 This includes checking that ``\param`` commands name parameters that actually 1594 present in the function signature, checking that ``\returns`` is used only on 1595 functions that actually return a value etc. 1596 1597.. option:: -Wno-documentation-unknown-command 1598 1599 Don't warn when encountering an unknown Doxygen command. 1600 1601.. option:: -fparse-all-comments 1602 1603 Parse all comments as documentation comments (including ordinary comments 1604 starting with ``//`` and ``/*``). 1605 1606.. option:: -fcomment-block-commands=[commands] 1607 1608 Define custom documentation commands as block commands. This allows Clang to 1609 construct the correct AST for these custom commands, and silences warnings 1610 about unknown commands. Several commands must be separated by a comma 1611 *without trailing space*; e.g. ``-fcomment-block-commands=foo,bar`` defines 1612 custom commands ``\foo`` and ``\bar``. 1613 1614 It is also possible to use ``-fcomment-block-commands`` several times; e.g. 1615 ``-fcomment-block-commands=foo -fcomment-block-commands=bar`` does the same 1616 as above. 1617 1618.. _c: 1619 1620C Language Features 1621=================== 1622 1623The support for standard C in clang is feature-complete except for the 1624C99 floating-point pragmas. 1625 1626Extensions supported by clang 1627----------------------------- 1628 1629See :doc:`LanguageExtensions`. 1630 1631Differences between various standard modes 1632------------------------------------------ 1633 1634clang supports the -std option, which changes what language mode clang 1635uses. The supported modes for C are c89, gnu89, c94, c99, gnu99, c11, 1636gnu11, and various aliases for those modes. If no -std option is 1637specified, clang defaults to gnu11 mode. Many C99 and C11 features are 1638supported in earlier modes as a conforming extension, with a warning. Use 1639``-pedantic-errors`` to request an error if a feature from a later standard 1640revision is used in an earlier mode. 1641 1642Differences between all ``c*`` and ``gnu*`` modes: 1643 1644- ``c*`` modes define "``__STRICT_ANSI__``". 1645- Target-specific defines not prefixed by underscores, like "linux", 1646 are defined in ``gnu*`` modes. 1647- Trigraphs default to being off in ``gnu*`` modes; they can be enabled by 1648 the -trigraphs option. 1649- The parser recognizes "asm" and "typeof" as keywords in ``gnu*`` modes; 1650 the variants "``__asm__``" and "``__typeof__``" are recognized in all 1651 modes. 1652- The Apple "blocks" extension is recognized by default in ``gnu*`` modes 1653 on some platforms; it can be enabled in any mode with the "-fblocks" 1654 option. 1655- Arrays that are VLA's according to the standard, but which can be 1656 constant folded by the frontend are treated as fixed size arrays. 1657 This occurs for things like "int X[(1, 2)];", which is technically a 1658 VLA. ``c*`` modes are strictly compliant and treat these as VLAs. 1659 1660Differences between ``*89`` and ``*99`` modes: 1661 1662- The ``*99`` modes default to implementing "inline" as specified in C99, 1663 while the ``*89`` modes implement the GNU version. This can be 1664 overridden for individual functions with the ``__gnu_inline__`` 1665 attribute. 1666- Digraphs are not recognized in c89 mode. 1667- The scope of names defined inside a "for", "if", "switch", "while", 1668 or "do" statement is different. (example: "``if ((struct x {int 1669 x;}*)0) {}``".) 1670- ``__STDC_VERSION__`` is not defined in ``*89`` modes. 1671- "inline" is not recognized as a keyword in c89 mode. 1672- "restrict" is not recognized as a keyword in ``*89`` modes. 1673- Commas are allowed in integer constant expressions in ``*99`` modes. 1674- Arrays which are not lvalues are not implicitly promoted to pointers 1675 in ``*89`` modes. 1676- Some warnings are different. 1677 1678Differences between ``*99`` and ``*11`` modes: 1679 1680- Warnings for use of C11 features are disabled. 1681- ``__STDC_VERSION__`` is defined to ``201112L`` rather than ``199901L``. 1682 1683c94 mode is identical to c89 mode except that digraphs are enabled in 1684c94 mode (FIXME: And ``__STDC_VERSION__`` should be defined!). 1685 1686GCC extensions not implemented yet 1687---------------------------------- 1688 1689clang tries to be compatible with gcc as much as possible, but some gcc 1690extensions are not implemented yet: 1691 1692- clang does not support #pragma weak (`bug 1693 3679 <http://llvm.org/bugs/show_bug.cgi?id=3679>`_). Due to the uses 1694 described in the bug, this is likely to be implemented at some point, 1695 at least partially. 1696- clang does not support decimal floating point types (``_Decimal32`` and 1697 friends) or fixed-point types (``_Fract`` and friends); nobody has 1698 expressed interest in these features yet, so it's hard to say when 1699 they will be implemented. 1700- clang does not support nested functions; this is a complex feature 1701 which is infrequently used, so it is unlikely to be implemented 1702 anytime soon. In C++11 it can be emulated by assigning lambda 1703 functions to local variables, e.g: 1704 1705 .. code-block:: cpp 1706 1707 auto const local_function = [&](int parameter) { 1708 // Do something 1709 }; 1710 ... 1711 local_function(1); 1712 1713- clang does not support global register variables; this is unlikely to 1714 be implemented soon because it requires additional LLVM backend 1715 support. 1716- clang does not support static initialization of flexible array 1717 members. This appears to be a rarely used extension, but could be 1718 implemented pending user demand. 1719- clang does not support 1720 ``__builtin_va_arg_pack``/``__builtin_va_arg_pack_len``. This is 1721 used rarely, but in some potentially interesting places, like the 1722 glibc headers, so it may be implemented pending user demand. Note 1723 that because clang pretends to be like GCC 4.2, and this extension 1724 was introduced in 4.3, the glibc headers will not try to use this 1725 extension with clang at the moment. 1726- clang does not support the gcc extension for forward-declaring 1727 function parameters; this has not shown up in any real-world code 1728 yet, though, so it might never be implemented. 1729 1730This is not a complete list; if you find an unsupported extension 1731missing from this list, please send an e-mail to cfe-dev. This list 1732currently excludes C++; see :ref:`C++ Language Features <cxx>`. Also, this 1733list does not include bugs in mostly-implemented features; please see 1734the `bug 1735tracker <http://llvm.org/bugs/buglist.cgi?quicksearch=product%3Aclang+component%3A-New%2BBugs%2CAST%2CBasic%2CDriver%2CHeaders%2CLLVM%2BCodeGen%2Cparser%2Cpreprocessor%2CSemantic%2BAnalyzer>`_ 1736for known existing bugs (FIXME: Is there a section for bug-reporting 1737guidelines somewhere?). 1738 1739Intentionally unsupported GCC extensions 1740---------------------------------------- 1741 1742- clang does not support the gcc extension that allows variable-length 1743 arrays in structures. This is for a few reasons: one, it is tricky to 1744 implement, two, the extension is completely undocumented, and three, 1745 the extension appears to be rarely used. Note that clang *does* 1746 support flexible array members (arrays with a zero or unspecified 1747 size at the end of a structure). 1748- clang does not have an equivalent to gcc's "fold"; this means that 1749 clang doesn't accept some constructs gcc might accept in contexts 1750 where a constant expression is required, like "x-x" where x is a 1751 variable. 1752- clang does not support ``__builtin_apply`` and friends; this extension 1753 is extremely obscure and difficult to implement reliably. 1754 1755.. _c_ms: 1756 1757Microsoft extensions 1758-------------------- 1759 1760clang has some experimental support for extensions from Microsoft Visual 1761C++; to enable it, use the ``-fms-extensions`` command-line option. This is 1762the default for Windows targets. Note that the support is incomplete. 1763Some constructs such as ``dllexport`` on classes are ignored with a warning, 1764and others such as `Microsoft IDL annotations 1765<http://msdn.microsoft.com/en-us/library/8tesw2eh.aspx>`_ are silently 1766ignored. 1767 1768clang has a ``-fms-compatibility`` flag that makes clang accept enough 1769invalid C++ to be able to parse most Microsoft headers. For example, it 1770allows `unqualified lookup of dependent base class members 1771<http://clang.llvm.org/compatibility.html#dep_lookup_bases>`_, which is 1772a common compatibility issue with clang. This flag is enabled by default 1773for Windows targets. 1774 1775``-fdelayed-template-parsing`` lets clang delay parsing of function template 1776definitions until the end of a translation unit. This flag is enabled by 1777default for Windows targets. 1778 1779- clang allows setting ``_MSC_VER`` with ``-fmsc-version=``. It defaults to 1780 1700 which is the same as Visual C/C++ 2012. Any number is supported 1781 and can greatly affect what Windows SDK and c++stdlib headers clang 1782 can compile. 1783- clang does not support the Microsoft extension where anonymous record 1784 members can be declared using user defined typedefs. 1785- clang supports the Microsoft ``#pragma pack`` feature for controlling 1786 record layout. GCC also contains support for this feature, however 1787 where MSVC and GCC are incompatible clang follows the MSVC 1788 definition. 1789- clang supports the Microsoft ``#pragma comment(lib, "foo.lib")`` feature for 1790 automatically linking against the specified library. Currently this feature 1791 only works with the Visual C++ linker. 1792- clang supports the Microsoft ``#pragma comment(linker, "/flag:foo")`` feature 1793 for adding linker flags to COFF object files. The user is responsible for 1794 ensuring that the linker understands the flags. 1795- clang defaults to C++11 for Windows targets. 1796 1797.. _cxx: 1798 1799C++ Language Features 1800===================== 1801 1802clang fully implements all of standard C++98 except for exported 1803templates (which were removed in C++11), and all of standard C++11 1804and the current draft standard for C++1y. 1805 1806Controlling implementation limits 1807--------------------------------- 1808 1809.. option:: -fbracket-depth=N 1810 1811 Sets the limit for nested parentheses, brackets, and braces to N. The 1812 default is 256. 1813 1814.. option:: -fconstexpr-depth=N 1815 1816 Sets the limit for recursive constexpr function invocations to N. The 1817 default is 512. 1818 1819.. option:: -ftemplate-depth=N 1820 1821 Sets the limit for recursively nested template instantiations to N. The 1822 default is 256. 1823 1824.. option:: -foperator-arrow-depth=N 1825 1826 Sets the limit for iterative calls to 'operator->' functions to N. The 1827 default is 256. 1828 1829.. _objc: 1830 1831Objective-C Language Features 1832============================= 1833 1834.. _objcxx: 1835 1836Objective-C++ Language Features 1837=============================== 1838 1839.. _openmp: 1840 1841OpenMP Features 1842=============== 1843 1844Clang supports all OpenMP 3.1 directives and clauses. In addition, some 1845features of OpenMP 4.0 are supported. For example, ``#pragma omp simd``, 1846``#pragma omp for simd``, ``#pragma omp parallel for simd`` directives, extended 1847set of atomic constructs, ``proc_bind`` clause for all parallel-based 1848directives, ``depend`` clause for ``#pragma omp task`` directive (except for 1849array sections), ``#pragma omp cancel`` and ``#pragma omp cancellation point`` 1850directives, and ``#pragma omp taskgroup`` directive. 1851 1852Use :option:`-fopenmp` to enable OpenMP. Support for OpenMP can be disabled with 1853:option:`-fno-openmp`. 1854 1855Controlling implementation limits 1856--------------------------------- 1857 1858.. option:: -fopenmp-use-tls 1859 1860 Controls code generation for OpenMP threadprivate variables. In presence of 1861 this option all threadprivate variables are generated the same way as thread 1862 local variables, using TLS support. If :option:`-fno-openmp-use-tls` 1863 is provided or target does not support TLS, code generation for threadprivate 1864 variables relies on OpenMP runtime library. 1865 1866.. _target_features: 1867 1868Target-Specific Features and Limitations 1869======================================== 1870 1871CPU Architectures Features and Limitations 1872------------------------------------------ 1873 1874X86 1875^^^ 1876 1877The support for X86 (both 32-bit and 64-bit) is considered stable on 1878Darwin (Mac OS X), Linux, FreeBSD, and Dragonfly BSD: it has been tested 1879to correctly compile many large C, C++, Objective-C, and Objective-C++ 1880codebases. 1881 1882On ``x86_64-mingw32``, passing i128(by value) is incompatible with the 1883Microsoft x64 calling convention. You might need to tweak 1884``WinX86_64ABIInfo::classify()`` in lib/CodeGen/TargetInfo.cpp. 1885 1886For the X86 target, clang supports the :option:`-m16` command line 1887argument which enables 16-bit code output. This is broadly similar to 1888using ``asm(".code16gcc")`` with the GNU toolchain. The generated code 1889and the ABI remains 32-bit but the assembler emits instructions 1890appropriate for a CPU running in 16-bit mode, with address-size and 1891operand-size prefixes to enable 32-bit addressing and operations. 1892 1893ARM 1894^^^ 1895 1896The support for ARM (specifically ARMv6 and ARMv7) is considered stable 1897on Darwin (iOS): it has been tested to correctly compile many large C, 1898C++, Objective-C, and Objective-C++ codebases. Clang only supports a 1899limited number of ARM architectures. It does not yet fully support 1900ARMv5, for example. 1901 1902PowerPC 1903^^^^^^^ 1904 1905The support for PowerPC (especially PowerPC64) is considered stable 1906on Linux and FreeBSD: it has been tested to correctly compile many 1907large C and C++ codebases. PowerPC (32bit) is still missing certain 1908features (e.g. PIC code on ELF platforms). 1909 1910Other platforms 1911^^^^^^^^^^^^^^^ 1912 1913clang currently contains some support for other architectures (e.g. Sparc); 1914however, significant pieces of code generation are still missing, and they 1915haven't undergone significant testing. 1916 1917clang contains limited support for the MSP430 embedded processor, but 1918both the clang support and the LLVM backend support are highly 1919experimental. 1920 1921Other platforms are completely unsupported at the moment. Adding the 1922minimal support needed for parsing and semantic analysis on a new 1923platform is quite easy; see ``lib/Basic/Targets.cpp`` in the clang source 1924tree. This level of support is also sufficient for conversion to LLVM IR 1925for simple programs. Proper support for conversion to LLVM IR requires 1926adding code to ``lib/CodeGen/CGCall.cpp`` at the moment; this is likely to 1927change soon, though. Generating assembly requires a suitable LLVM 1928backend. 1929 1930Operating System Features and Limitations 1931----------------------------------------- 1932 1933Darwin (Mac OS X) 1934^^^^^^^^^^^^^^^^^ 1935 1936Thread Sanitizer is not supported. 1937 1938Windows 1939^^^^^^^ 1940 1941Clang has experimental support for targeting "Cygming" (Cygwin / MinGW) 1942platforms. 1943 1944See also :ref:`Microsoft Extensions <c_ms>`. 1945 1946Cygwin 1947"""""" 1948 1949Clang works on Cygwin-1.7. 1950 1951MinGW32 1952""""""" 1953 1954Clang works on some mingw32 distributions. Clang assumes directories as 1955below; 1956 1957- ``C:/mingw/include`` 1958- ``C:/mingw/lib`` 1959- ``C:/mingw/lib/gcc/mingw32/4.[3-5].0/include/c++`` 1960 1961On MSYS, a few tests might fail. 1962 1963MinGW-w64 1964""""""""" 1965 1966For 32-bit (i686-w64-mingw32), and 64-bit (x86\_64-w64-mingw32), Clang 1967assumes as below; 1968 1969- ``GCC versions 4.5.0 to 4.5.3, 4.6.0 to 4.6.2, or 4.7.0 (for the C++ header search path)`` 1970- ``some_directory/bin/gcc.exe`` 1971- ``some_directory/bin/clang.exe`` 1972- ``some_directory/bin/clang++.exe`` 1973- ``some_directory/bin/../include/c++/GCC_version`` 1974- ``some_directory/bin/../include/c++/GCC_version/x86_64-w64-mingw32`` 1975- ``some_directory/bin/../include/c++/GCC_version/i686-w64-mingw32`` 1976- ``some_directory/bin/../include/c++/GCC_version/backward`` 1977- ``some_directory/bin/../x86_64-w64-mingw32/include`` 1978- ``some_directory/bin/../i686-w64-mingw32/include`` 1979- ``some_directory/bin/../include`` 1980 1981This directory layout is standard for any toolchain you will find on the 1982official `MinGW-w64 website <http://mingw-w64.sourceforge.net>`_. 1983 1984Clang expects the GCC executable "gcc.exe" compiled for 1985``i686-w64-mingw32`` (or ``x86_64-w64-mingw32``) to be present on PATH. 1986 1987`Some tests might fail <http://llvm.org/bugs/show_bug.cgi?id=9072>`_ on 1988``x86_64-w64-mingw32``. 1989 1990.. _clang-cl: 1991 1992clang-cl 1993======== 1994 1995clang-cl is an alternative command-line interface to Clang driver, designed for 1996compatibility with the Visual C++ compiler, cl.exe. 1997 1998To enable clang-cl to find system headers, libraries, and the linker when run 1999from the command-line, it should be executed inside a Visual Studio Native Tools 2000Command Prompt or a regular Command Prompt where the environment has been set 2001up using e.g. `vcvars32.bat <http://msdn.microsoft.com/en-us/library/f2ccy3wt.aspx>`_. 2002 2003clang-cl can also be used from inside Visual Studio by using an LLVM Platform 2004Toolset. 2005 2006Command-Line Options 2007-------------------- 2008 2009To be compatible with cl.exe, clang-cl supports most of the same command-line 2010options. Those options can start with either ``/`` or ``-``. It also supports 2011some of Clang's core options, such as the ``-W`` options. 2012 2013Options that are known to clang-cl, but not currently supported, are ignored 2014with a warning. For example: 2015 2016 :: 2017 2018 clang-cl.exe: warning: argument unused during compilation: '/AI' 2019 2020To suppress warnings about unused arguments, use the ``-Qunused-arguments`` option. 2021 2022Options that are not known to clang-cl will cause errors. If they are spelled with a 2023leading ``/``, they will be mistaken for a filename: 2024 2025 :: 2026 2027 clang-cl.exe: error: no such file or directory: '/foobar' 2028 2029Please `file a bug <http://llvm.org/bugs/enter_bug.cgi?product=clang&component=Driver>`_ 2030for any valid cl.exe flags that clang-cl does not understand. 2031 2032Execute ``clang-cl /?`` to see a list of supported options: 2033 2034 :: 2035 2036 CL.EXE COMPATIBILITY OPTIONS: 2037 /? Display available options 2038 /arch:<value> Set architecture for code generation 2039 /C Don't discard comments when preprocessing 2040 /c Compile only 2041 /D <macro[=value]> Define macro 2042 /EH<value> Exception handling model 2043 /EP Disable linemarker output and preprocess to stdout 2044 /E Preprocess to stdout 2045 /fallback Fall back to cl.exe if clang-cl fails to compile 2046 /FA Output assembly code file during compilation 2047 /Fa<file or directory> Output assembly code to this file during compilation (with /FA) 2048 /Fe<file or directory> Set output executable file or directory (ends in / or \) 2049 /FI <value> Include file before parsing 2050 /Fi<file> Set preprocess output file name (with /P) 2051 /Fo<file or directory> Set output object file, or directory (ends in / or \) (with /c) 2052 /fp:except- 2053 /fp:except 2054 /fp:fast 2055 /fp:precise 2056 /fp:strict 2057 /GA Assume thread-local variables are defined in the executable 2058 /GF- Disable string pooling 2059 /GR- Disable emission of RTTI data 2060 /GR Enable emission of RTTI data 2061 /Gs<value> Set stack probe size 2062 /Gw- Don't put each data item in its own section 2063 /Gw Put each data item in its own section 2064 /Gy- Don't put each function in its own section 2065 /Gy Put each function in its own section 2066 /help Display available options 2067 /I <dir> Add directory to include search path 2068 /J Make char type unsigned 2069 /LDd Create debug DLL 2070 /LD Create DLL 2071 /link <options> Forward options to the linker 2072 /MDd Use DLL debug run-time 2073 /MD Use DLL run-time 2074 /MTd Use static debug run-time 2075 /MT Use static run-time 2076 /Ob0 Disable inlining 2077 /Od Disable optimization 2078 /Oi- Disable use of builtin functions 2079 /Oi Enable use of builtin functions 2080 /Os Optimize for size 2081 /Ot Optimize for speed 2082 /Oy- Disable frame pointer omission 2083 /Oy Enable frame pointer omission 2084 /O<value> Optimization level 2085 /o <file or directory> Set output file or directory (ends in / or \) 2086 /P Preprocess to file 2087 /Qvec- Disable the loop vectorization passes 2088 /Qvec Enable the loop vectorization passes 2089 /showIncludes Print info about included files to stderr 2090 /TC Treat all source files as C 2091 /Tc <filename> Specify a C source file 2092 /TP Treat all source files as C++ 2093 /Tp <filename> Specify a C++ source file 2094 /U <macro> Undefine macro 2095 /vd<value> Control vtordisp placement 2096 /vmb Use a best-case representation method for member pointers 2097 /vmg Use a most-general representation for member pointers 2098 /vmm Set the default most-general representation to multiple inheritance 2099 /vms Set the default most-general representation to single inheritance 2100 /vmv Set the default most-general representation to virtual inheritance 2101 /volatile:iso Volatile loads and stores have standard semantics 2102 /volatile:ms Volatile loads and stores have acquire and release semantics 2103 /W0 Disable all warnings 2104 /W1 Enable -Wall 2105 /W2 Enable -Wall 2106 /W3 Enable -Wall 2107 /W4 Enable -Wall and -Wextra 2108 /Wall Enable -Wall 2109 /WX- Do not treat warnings as errors 2110 /WX Treat warnings as errors 2111 /w Disable all warnings 2112 /Z7 Enable CodeView debug information in object files 2113 /Zc:sizedDealloc- Disable C++14 sized global deallocation functions 2114 /Zc:sizedDealloc Enable C++14 sized global deallocation functions 2115 /Zc:strictStrings Treat string literals as const 2116 /Zc:threadSafeInit- Disable thread-safe initialization of static variables 2117 /Zc:threadSafeInit Enable thread-safe initialization of static variables 2118 /Zc:trigraphs- Disable trigraphs (default) 2119 /Zc:trigraphs Enable trigraphs 2120 /Zi Alias for /Z7. Does not produce PDBs. 2121 /Zl Don't mention any default libraries in the object file 2122 /Zp Set the default maximum struct packing alignment to 1 2123 /Zp<value> Specify the default maximum struct packing alignment 2124 /Zs Syntax-check only 2125 2126 OPTIONS: 2127 -### Print (but do not run) the commands to run for this compilation 2128 --analyze Run the static analyzer 2129 -fansi-escape-codes Use ANSI escape codes for diagnostics 2130 -fcolor-diagnostics Use colors in diagnostics 2131 -fdiagnostics-parseable-fixits 2132 Print fix-its in machine parseable form 2133 -fms-compatibility-version=<value> 2134 Dot-separated value representing the Microsoft compiler version 2135 number to report in _MSC_VER (0 = don't define it (default)) 2136 -fmsc-version=<value> Microsoft compiler version number to report in _MSC_VER (0 = don't 2137 define it (default)) 2138 -fno-sanitize-coverage=<value> 2139 Disable specified features of coverage instrumentation for Sanitizers 2140 -fno-sanitize-recover=<value> 2141 Disable recovery for specified sanitizers 2142 -fno-sanitize-trap=<value> 2143 Disable trapping for specified sanitizers 2144 -fsanitize-blacklist=<value> 2145 Path to blacklist file for sanitizers 2146 -fsanitize-coverage=<value> 2147 Specify the type of coverage instrumentation for Sanitizers 2148 -fsanitize-recover=<value> 2149 Enable recovery for specified sanitizers 2150 -fsanitize-trap=<value> Enable trapping for specified sanitizers 2151 -fsanitize=<check> Turn on runtime checks for various forms of undefined or suspicious 2152 behavior. See user manual for available checks 2153 -gcodeview Generate CodeView debug information 2154 -mllvm <value> Additional arguments to forward to LLVM's option processing 2155 -Qunused-arguments Don't emit warning for unused driver arguments 2156 -R<remark> Enable the specified remark 2157 --target=<value> Generate code for the given target 2158 -v Show commands to run and use verbose output 2159 -W<warning> Enable the specified warning 2160 -Xclang <arg> Pass <arg> to the clang compiler 2161 2162The /fallback Option 2163^^^^^^^^^^^^^^^^^^^^ 2164 2165When clang-cl is run with the ``/fallback`` option, it will first try to 2166compile files itself. For any file that it fails to compile, it will fall back 2167and try to compile the file by invoking cl.exe. 2168 2169This option is intended to be used as a temporary means to build projects where 2170clang-cl cannot successfully compile all the files. clang-cl may fail to compile 2171a file either because it cannot generate code for some C++ feature, or because 2172it cannot parse some Microsoft language extension. 2173