1 ===================================
2 Customizing LLVMC: Reference Manual
3 ===================================
5 This file was automatically generated by rst2html.
6 Please do not edit directly!
7 The ReST source lives in the directory 'tools/llvmc/doc'.
13 <div class="doc_author">
14 <p>Written by <a href="mailto:foldr@codedgers.com">Mikhail Glushenkov</a></p>
20 LLVMC is a generic compiler driver, designed to be customizable and
21 extensible. It plays the same role for LLVM as the ``gcc`` program
22 does for GCC - LLVMC's job is essentially to transform a set of input
23 files into a set of targets depending on configuration rules and user
24 options. What makes LLVMC different is that these transformation rules
25 are completely customizable - in fact, LLVMC knows nothing about the
26 specifics of transformation (even the command-line options are mostly
27 not hard-coded) and regards the transformation structure as an
28 abstract graph. The structure of this graph is completely determined
29 by plugins, which can be either statically or dynamically linked. This
30 makes it possible to easily adapt LLVMC for other purposes - for
31 example, as a build tool for game resources.
33 Because LLVMC employs TableGen_ as its configuration language, you
34 need to be familiar with it to customize LLVMC.
36 .. _TableGen: http://llvm.org/docs/TableGenFundamentals.html
42 LLVMC tries hard to be as compatible with ``gcc`` as possible,
43 although there are some small differences. Most of the time, however,
44 you shouldn't be able to notice them::
46 $ # This works as expected:
47 $ llvmc -O3 -Wall hello.cpp
51 One nice feature of LLVMC is that one doesn't have to distinguish between
52 different compilers for different languages (think ``g++`` vs. ``gcc``) - the
53 right toolchain is chosen automatically based on input language names (which
54 are, in turn, determined from file extensions). If you want to force files
55 ending with ".c" to compile as C++, use the ``-x`` option, just like you would
58 $ # hello.c is really a C++ file
59 $ llvmc -x c++ hello.c
63 On the other hand, when using LLVMC as a linker to combine several C++
64 object files you should provide the ``--linker`` option since it's
65 impossible for LLVMC to choose the right linker in that case::
69 [A lot of link-time errors skipped]
70 $ llvmc --linker=c++ hello.o
74 By default, LLVMC uses ``llvm-gcc`` to compile the source code. It is
75 also possible to choose the work-in-progress ``clang`` compiler with
76 the ``-clang`` option.
82 LLVMC has some built-in options that can't be overridden in the
83 configuration libraries:
85 * ``-o FILE`` - Output file name.
87 * ``-x LANGUAGE`` - Specify the language of the following input files
88 until the next -x option.
90 * ``-load PLUGIN_NAME`` - Load the specified plugin DLL. Example:
91 ``-load $LLVM_DIR/Release/lib/LLVMCSimple.so``.
93 * ``-v`` - Enable verbose mode, i.e. print out all executed commands.
95 * ``--save-temps`` - Write temporary files to the current directory and do not
96 delete them on exit. This option can also take an argument: the
97 ``--save-temps=obj`` switch will write files into the directory specified with
98 the ``-o`` option. The ``--save-temps=cwd`` and ``--save-temps`` switches are
99 both synonyms for the default behaviour.
101 * ``--check-graph`` - Check the compilation for common errors like mismatched
102 output/input language names, multiple default edges and cycles. Because of
103 plugins, these checks can't be performed at compile-time. Exit with code zero
104 if no errors were found, and return the number of found errors
105 otherwise. Hidden option, useful for debugging LLVMC plugins.
107 * ``--view-graph`` - Show a graphical representation of the compilation graph
108 and exit. Requires that you have ``dot`` and ``gv`` programs installed. Hidden
109 option, useful for debugging LLVMC plugins.
111 * ``--write-graph`` - Write a ``compilation-graph.dot`` file in the current
112 directory with the compilation graph description in Graphviz format (identical
113 to the file used by the ``--view-graph`` option). The ``-o`` option can be
114 used to set the output file name. Hidden option, useful for debugging LLVMC
117 * ``--help``, ``--help-hidden``, ``--version`` - These options have
118 their standard meaning.
120 Compiling LLVMC plugins
121 =======================
123 It's easiest to start working on your own LLVMC plugin by copying the
124 skeleton project which lives under ``$LLVMC_DIR/plugins/Simple``::
126 $ cd $LLVMC_DIR/plugins
127 $ cp -r Simple MyPlugin
130 Makefile PluginMain.cpp Simple.td
132 As you can see, our basic plugin consists of only two files (not
133 counting the build script). ``Simple.td`` contains TableGen
134 description of the compilation graph; its format is documented in the
135 following sections. ``PluginMain.cpp`` is just a helper file used to
136 compile the auto-generated C++ code produced from TableGen source. It
137 can also contain hook definitions (see `below`__).
141 The first thing that you should do is to change the ``LLVMC_PLUGIN``
142 variable in the ``Makefile`` to avoid conflicts (since this variable
143 is used to name the resulting library)::
145 LLVMC_PLUGIN=MyPlugin
147 It is also a good idea to rename ``Simple.td`` to something less
150 $ mv Simple.td MyPlugin.td
152 To build your plugin as a dynamic library, just ``cd`` to its source
153 directory and run ``make``. The resulting file will be called
154 ``plugin_llvmc_$(LLVMC_PLUGIN).$(DLL_EXTENSION)`` (in our case,
155 ``plugin_llvmc_MyPlugin.so``). This library can be then loaded in with the
156 ``-load`` option. Example::
158 $ cd $LLVMC_DIR/plugins/Simple
160 $ llvmc -load $LLVM_DIR/Release/lib/plugin_llvmc_Simple.so
162 Compiling standalone LLVMC-based drivers
163 ========================================
165 By default, the ``llvmc`` executable consists of a driver core plus several
166 statically linked plugins (``Base`` and ``Clang`` at the moment). You can
167 produce a standalone LLVMC-based driver executable by linking the core with your
168 own plugins. The recommended way to do this is by starting with the provided
169 ``Skeleton`` example (``$LLVMC_DIR/example/Skeleton``)::
171 $ cd $LLVMC_DIR/example/
172 $ cp -r Skeleton mydriver
178 If you're compiling LLVM with different source and object directories, then you
179 must perform the following additional steps before running ``make``::
181 # LLVMC_SRC_DIR = $LLVM_SRC_DIR/tools/llvmc/
182 # LLVMC_OBJ_DIR = $LLVM_OBJ_DIR/tools/llvmc/
183 $ cp $LLVMC_SRC_DIR/example/mydriver/Makefile \
184 $LLVMC_OBJ_DIR/example/mydriver/
185 $ cd $LLVMC_OBJ_DIR/example/mydriver
188 Another way to do the same thing is by using the following command::
191 $ make LLVMC_BUILTIN_PLUGINS=MyPlugin LLVMC_BASED_DRIVER_NAME=mydriver
193 This works with both srcdir == objdir and srcdir != objdir, but assumes that the
194 plugin source directory was placed under ``$LLVMC_DIR/plugins``.
196 Sometimes, you will want a 'bare-bones' version of LLVMC that has no
197 built-in plugins. It can be compiled with the following command::
200 $ make LLVMC_BUILTIN_PLUGINS=""
203 Customizing LLVMC: the compilation graph
204 ========================================
206 Each TableGen configuration file should include the common
209 include "llvm/CompilerDriver/Common.td"
211 Internally, LLVMC stores information about possible source
212 transformations in form of a graph. Nodes in this graph represent
213 tools, and edges between two nodes represent a transformation path. A
214 special "root" node is used to mark entry points for the
215 transformations. LLVMC also assigns a weight to each edge (more on
216 this later) to choose between several alternative edges.
218 The definition of the compilation graph (see file
219 ``plugins/Base/Base.td`` for an example) is just a list of edges::
221 def CompilationGraph : CompilationGraph<[
222 Edge<"root", "llvm_gcc_c">,
223 Edge<"root", "llvm_gcc_assembler">,
226 Edge<"llvm_gcc_c", "llc">,
227 Edge<"llvm_gcc_cpp", "llc">,
230 OptionalEdge<"llvm_gcc_c", "opt", (case (switch_on "opt"),
232 OptionalEdge<"llvm_gcc_cpp", "opt", (case (switch_on "opt"),
236 OptionalEdge<"llvm_gcc_assembler", "llvm_gcc_cpp_linker",
237 (case (input_languages_contain "c++"), (inc_weight),
238 (or (parameter_equals "linker", "g++"),
239 (parameter_equals "linker", "c++")), (inc_weight))>,
244 As you can see, the edges can be either default or optional, where
245 optional edges are differentiated by an additional ``case`` expression
246 used to calculate the weight of this edge. Notice also that we refer
247 to tools via their names (as strings). This makes it possible to add
248 edges to an existing compilation graph in plugins without having to
249 know about all tool definitions used in the graph.
251 The default edges are assigned a weight of 1, and optional edges get a
252 weight of 0 + 2*N where N is the number of tests that evaluated to
253 true in the ``case`` expression. It is also possible to provide an
254 integer parameter to ``inc_weight`` and ``dec_weight`` - in this case,
255 the weight is increased (or decreased) by the provided value instead
256 of the default 2. It is also possible to change the default weight of
257 an optional edge by using the ``default`` clause of the ``case``
260 When passing an input file through the graph, LLVMC picks the edge
261 with the maximum weight. To avoid ambiguity, there should be only one
262 default edge between two nodes (with the exception of the root node,
263 which gets a special treatment - there you are allowed to specify one
264 default edge *per language*).
266 When multiple plugins are loaded, their compilation graphs are merged
267 together. Since multiple edges that have the same end nodes are not
268 allowed (i.e. the graph is not a multigraph), an edge defined in
269 several plugins will be replaced by the definition from the plugin
270 that was loaded last. Plugin load order can be controlled by using the
271 plugin priority feature described above.
273 To get a visual representation of the compilation graph (useful for
274 debugging), run ``llvmc --view-graph``. You will need ``dot`` and
275 ``gsview`` installed for this to work properly.
280 Command-line options that the plugin supports are defined by using an
283 def Options : OptionList<[
284 (switch_option "E", (help "Help string")),
285 (alias_option "quiet", "q")
289 As you can see, the option list is just a list of DAGs, where each DAG
290 is an option description consisting of the option name and some
291 properties. A plugin can define more than one option list (they are
292 all merged together in the end), which can be handy if one wants to
293 separate option groups syntactically.
295 * Possible option types:
297 - ``switch_option`` - a simple boolean switch without arguments, for example
298 ``-O2`` or ``-time``. At most one occurrence is allowed.
300 - ``parameter_option`` - option that takes one argument, for example
301 ``-std=c99``. It is also allowed to use spaces instead of the equality
302 sign: ``-std c99``. At most one occurrence is allowed.
304 - ``parameter_list_option`` - same as the above, but more than one option
305 occurence is allowed.
307 - ``prefix_option`` - same as the parameter_option, but the option name and
308 argument do not have to be separated. Example: ``-ofile``. This can be also
309 specified as ``-o file``; however, ``-o=file`` will be parsed incorrectly
310 (``=file`` will be interpreted as option value). At most one occurrence is
313 - ``prefix_list_option`` - same as the above, but more than one occurence of
314 the option is allowed; example: ``-lm -lpthread``.
316 - ``alias_option`` - a special option type for creating aliases. Unlike other
317 option types, aliases are not allowed to have any properties besides the
318 aliased option name. Usage example: ``(alias_option "preprocess", "E")``
321 * Possible option properties:
323 - ``help`` - help string associated with this option. Used for ``--help``
326 - ``required`` - this option must be specified exactly once (or, in case of
327 the list options without the ``multi_val`` property, at least
328 once). Incompatible with ``zero_or_one`` and ``one_or_more``.
330 - ``one_or_more`` - the option must be specified at least one time. Useful
331 only for list options in conjunction with ``multi_val``; for ordinary lists
332 it is synonymous with ``required``. Incompatible with ``required`` and
335 - ``zero_or_one`` - the option can be specified zero or one times. Useful
336 only for list options in conjunction with ``multi_val``. Incompatible with
337 ``required`` and ``one_or_more``.
339 - ``hidden`` - the description of this option will not appear in
340 the ``--help`` output (but will appear in the ``--help-hidden``
343 - ``really_hidden`` - the option will not be mentioned in any help
346 - ``multi_val n`` - this option takes *n* arguments (can be useful in some
347 special cases). Usage example: ``(parameter_list_option "foo", (multi_val
348 3))``. Only list options can have this attribute; you can, however, use
349 the ``one_or_more`` and ``zero_or_one`` properties.
351 - ``extern`` - this option is defined in some other plugin, see below.
356 Sometimes, when linking several plugins together, one plugin needs to
357 access options defined in some other plugin. Because of the way
358 options are implemented, such options must be marked as
359 ``extern``. This is what the ``extern`` option property is
363 (switch_option "E", (extern))
366 See also the section on plugin `priorities`__.
372 Conditional evaluation
373 ======================
375 The 'case' construct is the main means by which programmability is
376 achieved in LLVMC. It can be used to calculate edge weights, program
377 actions and modify the shell commands to be executed. The 'case'
378 expression is designed after the similarly-named construct in
379 functional languages and takes the form ``(case (test_1), statement_1,
380 (test_2), statement_2, ... (test_N), statement_N)``. The statements
381 are evaluated only if the corresponding tests evaluate to true.
385 // Edge weight calculation
387 // Increases edge weight by 5 if "-A" is provided on the
388 // command-line, and by 5 more if "-B" is also provided.
390 (switch_on "A"), (inc_weight 5),
391 (switch_on "B"), (inc_weight 5))
394 // Tool command line specification
396 // Evaluates to "cmdline1" if the option "-A" is provided on the
397 // command line; to "cmdline2" if "-B" is provided;
398 // otherwise to "cmdline3".
401 (switch_on "A"), "cmdline1",
402 (switch_on "B"), "cmdline2",
403 (default), "cmdline3")
405 Note the slight difference in 'case' expression handling in contexts
406 of edge weights and command line specification - in the second example
407 the value of the ``"B"`` switch is never checked when switch ``"A"`` is
408 enabled, and the whole expression always evaluates to ``"cmdline1"`` in
411 Case expressions can also be nested, i.e. the following is legal::
413 (case (switch_on "E"), (case (switch_on "o"), ..., (default), ...)
416 You should, however, try to avoid doing that because it hurts
417 readability. It is usually better to split tool descriptions and/or
418 use TableGen inheritance instead.
420 * Possible tests are:
422 - ``switch_on`` - Returns true if a given command-line switch is
423 provided by the user. Example: ``(switch_on "opt")``.
425 - ``parameter_equals`` - Returns true if a command-line parameter equals
427 Example: ``(parameter_equals "W", "all")``.
429 - ``element_in_list`` - Returns true if a command-line parameter
430 list contains a given value.
431 Example: ``(parameter_in_list "l", "pthread")``.
433 - ``input_languages_contain`` - Returns true if a given language
434 belongs to the current input language set.
435 Example: ``(input_languages_contain "c++")``.
437 - ``in_language`` - Evaluates to true if the input file language
438 equals to the argument. At the moment works only with ``cmd_line``
439 and ``actions`` (on non-join nodes).
440 Example: ``(in_language "c++")``.
442 - ``not_empty`` - Returns true if a given option (which should be
443 either a parameter or a parameter list) is set by the
445 Example: ``(not_empty "o")``.
447 - ``empty`` - The opposite of ``not_empty``. Equivalent to ``(not (not_empty
448 X))``. Provided for convenience.
450 - ``default`` - Always evaluates to true. Should always be the last
451 test in the ``case`` expression.
453 - ``and`` - A standard logical combinator that returns true iff all
454 of its arguments return true. Used like this: ``(and (test1),
455 (test2), ... (testN))``. Nesting of ``and`` and ``or`` is allowed,
458 - ``or`` - Another logical combinator that returns true only if any
459 one of its arguments returns true. Example: ``(or (test1),
460 (test2), ... (testN))``.
463 Writing a tool description
464 ==========================
466 As was said earlier, nodes in the compilation graph represent tools,
467 which are described separately. A tool definition looks like this
468 (taken from the ``include/llvm/CompilerDriver/Tools.td`` file)::
470 def llvm_gcc_cpp : Tool<[
472 (out_language "llvm-assembler"),
473 (output_suffix "bc"),
474 (cmd_line "llvm-g++ -c $INFILE -o $OUTFILE -emit-llvm"),
478 This defines a new tool called ``llvm_gcc_cpp``, which is an alias for
479 ``llvm-g++``. As you can see, a tool definition is just a list of
480 properties; most of them should be self-explanatory. The ``sink``
481 property means that this tool should be passed all command-line
482 options that aren't mentioned in the option list.
484 The complete list of all currently implemented tool properties follows.
486 * Possible tool properties:
488 - ``in_language`` - input language name. Can be either a string or a
489 list, in case the tool supports multiple input languages.
491 - ``out_language`` - output language name. Tools are not allowed to
492 have multiple output languages.
494 - ``output_suffix`` - output file suffix. Can also be changed
495 dynamically, see documentation on actions.
497 - ``cmd_line`` - the actual command used to run the tool. You can
498 use ``$INFILE`` and ``$OUTFILE`` variables, output redirection
499 with ``>``, hook invocations (``$CALL``), environment variables
500 (via ``$ENV``) and the ``case`` construct.
502 - ``join`` - this tool is a "join node" in the graph, i.e. it gets a
503 list of input files and joins them together. Used for linkers.
505 - ``sink`` - all command-line options that are not handled by other
506 tools are passed to this tool.
508 - ``actions`` - A single big ``case`` expression that specifies how
509 this tool reacts on command-line options (described in more detail
515 A tool often needs to react to command-line options, and this is
516 precisely what the ``actions`` property is for. The next example
517 illustrates this feature::
519 def llvm_gcc_linker : Tool<[
520 (in_language "object-code"),
521 (out_language "executable"),
522 (output_suffix "out"),
523 (cmd_line "llvm-gcc $INFILE -o $OUTFILE"),
525 (actions (case (not_empty "L"), (forward "L"),
526 (not_empty "l"), (forward "l"),
528 [(append_cmd "-dummy1"), (append_cmd "-dummy2")])
531 The ``actions`` tool property is implemented on top of the omnipresent
532 ``case`` expression. It associates one or more different *actions*
533 with given conditions - in the example, the actions are ``forward``,
534 which forwards a given option unchanged, and ``append_cmd``, which
535 appends a given string to the tool execution command. Multiple actions
536 can be associated with a single condition by using a list of actions
537 (used in the example to append some dummy options). The same ``case``
538 construct can also be used in the ``cmd_line`` property to modify the
541 The "join" property used in the example means that this tool behaves
544 The list of all possible actions follows.
548 - ``append_cmd`` - append a string to the tool invocation
550 Example: ``(case (switch_on "pthread"), (append_cmd
553 - ``error` - exit with error.
554 Example: ``(error "Mixing -c and -S is not allowed!")``.
556 - ``forward`` - forward an option unchanged.
557 Example: ``(forward "Wall")``.
559 - ``forward_as`` - Change the name of an option, but forward the
561 Example: ``(forward_as "O0", "--disable-optimization")``.
563 - ``output_suffix`` - modify the output suffix of this
565 Example: ``(output_suffix "i")``.
567 - ``stop_compilation`` - stop compilation after this tool processes
568 its input. Used without arguments.
570 - ``unpack_values`` - used for for splitting and forwarding
571 comma-separated lists of options, e.g. ``-Wa,-foo=bar,-baz`` is
572 converted to ``-foo=bar -baz`` and appended to the tool invocation
574 Example: ``(unpack_values "Wa,")``.
579 If you are adding support for a new language to LLVMC, you'll need to
580 modify the language map, which defines mappings from file extensions
581 to language names. It is used to choose the proper toolchain(s) for a
582 given input file set. Language map definition looks like this::
584 def LanguageMap : LanguageMap<
585 [LangToSuffixes<"c++", ["cc", "cp", "cxx", "cpp", "CPP", "c++", "C"]>,
586 LangToSuffixes<"c", ["c"]>,
590 For example, without those definitions the following command wouldn't work::
593 llvmc: Unknown suffix: cpp
595 The language map entries should be added only for tools that are
596 linked with the root node. Since tools are not allowed to have
597 multiple output languages, for nodes "inside" the graph the input and
598 output languages should match. This is enforced at compile-time.
606 Hooks and environment variables
607 -------------------------------
609 Normally, LLVMC executes programs from the system ``PATH``. Sometimes,
610 this is not sufficient: for example, we may want to specify tool paths
611 or names in the configuration file. This can be easily achieved via
612 the hooks mechanism. To write your own hooks, just add their
613 definitions to the ``PluginMain.cpp`` or drop a ``.cpp`` file into the
614 your plugin directory. Hooks should live in the ``hooks`` namespace
615 and have the signature ``std::string hooks::MyHookName ([const char*
616 Arg0 [ const char* Arg2 [, ...]]])``. They can be used from the
617 ``cmd_line`` tool property::
619 (cmd_line "$CALL(MyHook)/path/to/file -o $CALL(AnotherHook)")
621 To pass arguments to hooks, use the following syntax::
623 (cmd_line "$CALL(MyHook, 'Arg1', 'Arg2', 'Arg # 3')/path/to/file -o1 -o2")
625 It is also possible to use environment variables in the same manner::
627 (cmd_line "$ENV(VAR1)/path/to/file -o $ENV(VAR2)")
629 To change the command line string based on user-provided options use
630 the ``case`` expression (documented `above`__)::
635 "llvm-g++ -E -x c $INFILE -o $OUTFILE",
637 "llvm-g++ -c -x c $INFILE -o $OUTFILE -emit-llvm"))
643 How plugins are loaded
644 ----------------------
646 It is possible for LLVMC plugins to depend on each other. For example,
647 one can create edges between nodes defined in some other plugin. To
648 make this work, however, that plugin should be loaded first. To
649 achieve this, the concept of plugin priority was introduced. By
650 default, every plugin has priority zero; to specify the priority
651 explicitly, put the following line in your plugin's TableGen file::
653 def Priority : PluginPriority<$PRIORITY_VALUE>;
654 # Where PRIORITY_VALUE is some integer > 0
656 Plugins are loaded in order of their (increasing) priority, starting
657 with 0. Therefore, the plugin with the highest priority value will be
663 When writing LLVMC plugins, it can be useful to get a visual view of
664 the resulting compilation graph. This can be achieved via the command
665 line option ``--view-graph``. This command assumes that Graphviz_ and
666 Ghostview_ are installed. There is also a ``--write-graph`` option that
667 creates a Graphviz source file (``compilation-graph.dot``) in the
670 Another useful ``llvmc`` option is ``--check-graph``. It checks the
671 compilation graph for common errors like mismatched output/input
672 language names, multiple default edges and cycles. These checks can't
673 be performed at compile-time because the plugins can load code
674 dynamically. When invoked with ``--check-graph``, ``llvmc`` doesn't
675 perform any compilation tasks and returns the number of encountered
676 errors as its status code.
678 .. _Graphviz: http://www.graphviz.org/
679 .. _Ghostview: http://pages.cs.wisc.edu/~ghost/
681 Conditioning on the executable name
682 -----------------------------------
684 For now, the executable name (the value passed to the driver in ``argv[0]``) is
685 accessible only in the C++ code (i.e. hooks). Use the following code::
688 extern const char* ProgramName;
691 std::string MyHook() {
693 if (strcmp(ProgramName, "mydriver") == 0) {
698 In general, you're encouraged not to make the behaviour dependent on the
699 executable file name, and use command-line switches instead. See for example how
700 the ``Base`` plugin behaves when it needs to choose the correct linker options
701 (think ``g++`` vs. ``gcc``).
707 <a href="http://jigsaw.w3.org/css-validator/check/referer">
708 <img src="http://jigsaw.w3.org/css-validator/images/vcss-blue"
709 alt="Valid CSS" /></a>
710 <a href="http://validator.w3.org/check?uri=referer">
711 <img src="http://www.w3.org/Icons/valid-xhtml10-blue"
712 alt="Valid XHTML 1.0 Transitional"/></a>
714 <a href="mailto:foldr@codedgers.com">Mikhail Glushenkov</a><br />
715 <a href="http://llvm.org">LLVM Compiler Infrastructure</a><br />
717 Last modified: $Date: 2008-12-11 11:34:48 -0600 (Thu, 11 Dec 2008) $