1 @c Copyright (C) 1988,1989,1992,1993,1994,1995,1996,1997,1998,1999,2000,2001,
2 @c 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
3 @c Free Software Foundation, Inc.
4 @c This is part of the GCC manual.
5 @c For copying conditions, see the file gcc.texi.
8 @chapter Target Description Macros and Functions
9 @cindex machine description macros
10 @cindex target description macros
11 @cindex macros, target description
12 @cindex @file{tm.h} macros
14 In addition to the file @file{@var{machine}.md}, a machine description
15 includes a C header file conventionally given the name
16 @file{@var{machine}.h} and a C source file named @file{@var{machine}.c}.
17 The header file defines numerous macros that convey the information
18 about the target machine that does not fit into the scheme of the
19 @file{.md} file. The file @file{tm.h} should be a link to
20 @file{@var{machine}.h}. The header file @file{config.h} includes
21 @file{tm.h} and most compiler source files include @file{config.h}. The
22 source file defines a variable @code{targetm}, which is a structure
23 containing pointers to functions and data relating to the target
24 machine. @file{@var{machine}.c} should also contain their definitions,
25 if they are not defined elsewhere in GCC, and other functions called
26 through the macros defined in the @file{.h} file.
29 * Target Structure:: The @code{targetm} variable.
30 * Driver:: Controlling how the driver runs the compilation passes.
31 * Run-time Target:: Defining @samp{-m} options like @option{-m68000} and @option{-m68020}.
32 * Per-Function Data:: Defining data structures for per-function information.
33 * Storage Layout:: Defining sizes and alignments of data.
34 * Type Layout:: Defining sizes and properties of basic user data types.
35 * Registers:: Naming and describing the hardware registers.
36 * Register Classes:: Defining the classes of hardware registers.
37 * Old Constraints:: The old way to define machine-specific constraints.
38 * Stack and Calling:: Defining which way the stack grows and by how much.
39 * Varargs:: Defining the varargs macros.
40 * Trampolines:: Code set up at run time to enter a nested function.
41 * Library Calls:: Controlling how library routines are implicitly called.
42 * Addressing Modes:: Defining addressing modes valid for memory operands.
43 * Anchored Addresses:: Defining how @option{-fsection-anchors} should work.
44 * Condition Code:: Defining how insns update the condition code.
45 * Costs:: Defining relative costs of different operations.
46 * Scheduling:: Adjusting the behavior of the instruction scheduler.
47 * Sections:: Dividing storage into text, data, and other sections.
48 * PIC:: Macros for position independent code.
49 * Assembler Format:: Defining how to write insns and pseudo-ops to output.
50 * Debugging Info:: Defining the format of debugging output.
51 * Floating Point:: Handling floating point for cross-compilers.
52 * Mode Switching:: Insertion of mode-switching instructions.
53 * Target Attributes:: Defining target-specific uses of @code{__attribute__}.
54 * Emulated TLS:: Emulated TLS support.
55 * MIPS Coprocessors:: MIPS coprocessor support and how to customize it.
56 * PCH Target:: Validity checking for precompiled headers.
57 * C++ ABI:: Controlling C++ ABI changes.
58 * Named Address Spaces:: Adding support for named address spaces
59 * Misc:: Everything else.
62 @node Target Structure
63 @section The Global @code{targetm} Variable
65 @cindex target functions
67 @deftypevar {struct gcc_target} targetm
68 The target @file{.c} file must define the global @code{targetm} variable
69 which contains pointers to functions and data relating to the target
70 machine. The variable is declared in @file{target.h};
71 @file{target-def.h} defines the macro @code{TARGET_INITIALIZER} which is
72 used to initialize the variable, and macros for the default initializers
73 for elements of the structure. The @file{.c} file should override those
74 macros for which the default definition is inappropriate. For example:
77 #include "target-def.h"
79 /* @r{Initialize the GCC target structure.} */
81 #undef TARGET_COMP_TYPE_ATTRIBUTES
82 #define TARGET_COMP_TYPE_ATTRIBUTES @var{machine}_comp_type_attributes
84 struct gcc_target targetm = TARGET_INITIALIZER;
88 Where a macro should be defined in the @file{.c} file in this manner to
89 form part of the @code{targetm} structure, it is documented below as a
90 ``Target Hook'' with a prototype. Many macros will change in future
91 from being defined in the @file{.h} file to being part of the
92 @code{targetm} structure.
95 @section Controlling the Compilation Driver, @file{gcc}
97 @cindex controlling the compilation driver
99 @c prevent bad page break with this line
100 You can control the compilation driver.
102 @defmac SWITCH_TAKES_ARG (@var{char})
103 A C expression which determines whether the option @option{-@var{char}}
104 takes arguments. The value should be the number of arguments that
105 option takes--zero, for many options.
107 By default, this macro is defined as
108 @code{DEFAULT_SWITCH_TAKES_ARG}, which handles the standard options
109 properly. You need not define @code{SWITCH_TAKES_ARG} unless you
110 wish to add additional options which take arguments. Any redefinition
111 should call @code{DEFAULT_SWITCH_TAKES_ARG} and then check for
115 @defmac WORD_SWITCH_TAKES_ARG (@var{name})
116 A C expression which determines whether the option @option{-@var{name}}
117 takes arguments. The value should be the number of arguments that
118 option takes--zero, for many options. This macro rather than
119 @code{SWITCH_TAKES_ARG} is used for multi-character option names.
121 By default, this macro is defined as
122 @code{DEFAULT_WORD_SWITCH_TAKES_ARG}, which handles the standard options
123 properly. You need not define @code{WORD_SWITCH_TAKES_ARG} unless you
124 wish to add additional options which take arguments. Any redefinition
125 should call @code{DEFAULT_WORD_SWITCH_TAKES_ARG} and then check for
129 @defmac SWITCH_CURTAILS_COMPILATION (@var{char})
130 A C expression which determines whether the option @option{-@var{char}}
131 stops compilation before the generation of an executable. The value is
132 boolean, nonzero if the option does stop an executable from being
133 generated, zero otherwise.
135 By default, this macro is defined as
136 @code{DEFAULT_SWITCH_CURTAILS_COMPILATION}, which handles the standard
137 options properly. You need not define
138 @code{SWITCH_CURTAILS_COMPILATION} unless you wish to add additional
139 options which affect the generation of an executable. Any redefinition
140 should call @code{DEFAULT_SWITCH_CURTAILS_COMPILATION} and then check
141 for additional options.
144 @defmac TARGET_OPTION_TRANSLATE_TABLE
145 If defined, a list of pairs of strings, the first of which is a
146 potential command line target to the @file{gcc} driver program, and the
147 second of which is a space-separated (tabs and other whitespace are not
148 supported) list of options with which to replace the first option. The
149 target defining this list is responsible for assuring that the results
150 are valid. Replacement options may not be the @code{--opt} style, they
151 must be the @code{-opt} style. It is the intention of this macro to
152 provide a mechanism for substitution that affects the multilibs chosen,
153 such as one option that enables many options, some of which select
154 multilibs. Example nonsensical definition, where @option{-malt-abi},
155 @option{-EB}, and @option{-mspoo} cause different multilibs to be chosen:
158 #define TARGET_OPTION_TRANSLATE_TABLE \
159 @{ "-fast", "-march=fast-foo -malt-abi -I/usr/fast-foo" @}, \
160 @{ "-compat", "-EB -malign=4 -mspoo" @}
164 @defmac DRIVER_SELF_SPECS
165 A list of specs for the driver itself. It should be a suitable
166 initializer for an array of strings, with no surrounding braces.
168 The driver applies these specs to its own command line between loading
169 default @file{specs} files (but not command-line specified ones) and
170 choosing the multilib directory or running any subcommands. It
171 applies them in the order given, so each spec can depend on the
172 options added by earlier ones. It is also possible to remove options
173 using @samp{%<@var{option}} in the usual way.
175 This macro can be useful when a port has several interdependent target
176 options. It provides a way of standardizing the command line so
177 that the other specs are easier to write.
179 Do not define this macro if it does not need to do anything.
182 @defmac OPTION_DEFAULT_SPECS
183 A list of specs used to support configure-time default options (i.e.@:
184 @option{--with} options) in the driver. It should be a suitable initializer
185 for an array of structures, each containing two strings, without the
186 outermost pair of surrounding braces.
188 The first item in the pair is the name of the default. This must match
189 the code in @file{config.gcc} for the target. The second item is a spec
190 to apply if a default with this name was specified. The string
191 @samp{%(VALUE)} in the spec will be replaced by the value of the default
192 everywhere it occurs.
194 The driver will apply these specs to its own command line between loading
195 default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using
196 the same mechanism as @code{DRIVER_SELF_SPECS}.
198 Do not define this macro if it does not need to do anything.
202 A C string constant that tells the GCC driver program options to
203 pass to CPP@. It can also specify how to translate options you
204 give to GCC into options for GCC to pass to the CPP@.
206 Do not define this macro if it does not need to do anything.
209 @defmac CPLUSPLUS_CPP_SPEC
210 This macro is just like @code{CPP_SPEC}, but is used for C++, rather
211 than C@. If you do not define this macro, then the value of
212 @code{CPP_SPEC} (if any) will be used instead.
216 A C string constant that tells the GCC driver program options to
217 pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language
219 It can also specify how to translate options you give to GCC into options
220 for GCC to pass to front ends.
222 Do not define this macro if it does not need to do anything.
226 A C string constant that tells the GCC driver program options to
227 pass to @code{cc1plus}. It can also specify how to translate options you
228 give to GCC into options for GCC to pass to the @code{cc1plus}.
230 Do not define this macro if it does not need to do anything.
231 Note that everything defined in CC1_SPEC is already passed to
232 @code{cc1plus} so there is no need to duplicate the contents of
233 CC1_SPEC in CC1PLUS_SPEC@.
237 A C string constant that tells the GCC driver program options to
238 pass to the assembler. It can also specify how to translate options
239 you give to GCC into options for GCC to pass to the assembler.
240 See the file @file{sun3.h} for an example of this.
242 Do not define this macro if it does not need to do anything.
245 @defmac ASM_FINAL_SPEC
246 A C string constant that tells the GCC driver program how to
247 run any programs which cleanup after the normal assembler.
248 Normally, this is not needed. See the file @file{mips.h} for
251 Do not define this macro if it does not need to do anything.
254 @defmac AS_NEEDS_DASH_FOR_PIPED_INPUT
255 Define this macro, with no value, if the driver should give the assembler
256 an argument consisting of a single dash, @option{-}, to instruct it to
257 read from its standard input (which will be a pipe connected to the
258 output of the compiler proper). This argument is given after any
259 @option{-o} option specifying the name of the output file.
261 If you do not define this macro, the assembler is assumed to read its
262 standard input if given no non-option arguments. If your assembler
263 cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct;
264 see @file{mips.h} for instance.
268 A C string constant that tells the GCC driver program options to
269 pass to the linker. It can also specify how to translate options you
270 give to GCC into options for GCC to pass to the linker.
272 Do not define this macro if it does not need to do anything.
276 Another C string constant used much like @code{LINK_SPEC}. The difference
277 between the two is that @code{LIB_SPEC} is used at the end of the
278 command given to the linker.
280 If this macro is not defined, a default is provided that
281 loads the standard C library from the usual place. See @file{gcc.c}.
285 Another C string constant that tells the GCC driver program
286 how and when to place a reference to @file{libgcc.a} into the
287 linker command line. This constant is placed both before and after
288 the value of @code{LIB_SPEC}.
290 If this macro is not defined, the GCC driver provides a default that
291 passes the string @option{-lgcc} to the linker.
294 @defmac REAL_LIBGCC_SPEC
295 By default, if @code{ENABLE_SHARED_LIBGCC} is defined, the
296 @code{LIBGCC_SPEC} is not directly used by the driver program but is
297 instead modified to refer to different versions of @file{libgcc.a}
298 depending on the values of the command line flags @option{-static},
299 @option{-shared}, @option{-static-libgcc}, and @option{-shared-libgcc}. On
300 targets where these modifications are inappropriate, define
301 @code{REAL_LIBGCC_SPEC} instead. @code{REAL_LIBGCC_SPEC} tells the
302 driver how to place a reference to @file{libgcc} on the link command
303 line, but, unlike @code{LIBGCC_SPEC}, it is used unmodified.
306 @defmac USE_LD_AS_NEEDED
307 A macro that controls the modifications to @code{LIBGCC_SPEC}
308 mentioned in @code{REAL_LIBGCC_SPEC}. If nonzero, a spec will be
309 generated that uses --as-needed and the shared libgcc in place of the
310 static exception handler library, when linking without any of
311 @code{-static}, @code{-static-libgcc}, or @code{-shared-libgcc}.
315 If defined, this C string constant is added to @code{LINK_SPEC}.
316 When @code{USE_LD_AS_NEEDED} is zero or undefined, it also affects
317 the modifications to @code{LIBGCC_SPEC} mentioned in
318 @code{REAL_LIBGCC_SPEC}.
321 @defmac STARTFILE_SPEC
322 Another C string constant used much like @code{LINK_SPEC}. The
323 difference between the two is that @code{STARTFILE_SPEC} is used at
324 the very beginning of the command given to the linker.
326 If this macro is not defined, a default is provided that loads the
327 standard C startup file from the usual place. See @file{gcc.c}.
331 Another C string constant used much like @code{LINK_SPEC}. The
332 difference between the two is that @code{ENDFILE_SPEC} is used at
333 the very end of the command given to the linker.
335 Do not define this macro if it does not need to do anything.
338 @defmac THREAD_MODEL_SPEC
339 GCC @code{-v} will print the thread model GCC was configured to use.
340 However, this doesn't work on platforms that are multilibbed on thread
341 models, such as AIX 4.3. On such platforms, define
342 @code{THREAD_MODEL_SPEC} such that it evaluates to a string without
343 blanks that names one of the recognized thread models. @code{%*}, the
344 default value of this macro, will expand to the value of
345 @code{thread_file} set in @file{config.gcc}.
348 @defmac SYSROOT_SUFFIX_SPEC
349 Define this macro to add a suffix to the target sysroot when GCC is
350 configured with a sysroot. This will cause GCC to search for usr/lib,
351 et al, within sysroot+suffix.
354 @defmac SYSROOT_HEADERS_SUFFIX_SPEC
355 Define this macro to add a headers_suffix to the target sysroot when
356 GCC is configured with a sysroot. This will cause GCC to pass the
357 updated sysroot+headers_suffix to CPP, causing it to search for
358 usr/include, et al, within sysroot+headers_suffix.
362 Define this macro to provide additional specifications to put in the
363 @file{specs} file that can be used in various specifications like
366 The definition should be an initializer for an array of structures,
367 containing a string constant, that defines the specification name, and a
368 string constant that provides the specification.
370 Do not define this macro if it does not need to do anything.
372 @code{EXTRA_SPECS} is useful when an architecture contains several
373 related targets, which have various @code{@dots{}_SPECS} which are similar
374 to each other, and the maintainer would like one central place to keep
377 For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to
378 define either @code{_CALL_SYSV} when the System V calling sequence is
379 used or @code{_CALL_AIX} when the older AIX-based calling sequence is
382 The @file{config/rs6000/rs6000.h} target file defines:
385 #define EXTRA_SPECS \
386 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @},
388 #define CPP_SYS_DEFAULT ""
391 The @file{config/rs6000/sysv.h} target file defines:
395 "%@{posix: -D_POSIX_SOURCE @} \
396 %@{mcall-sysv: -D_CALL_SYSV @} \
397 %@{!mcall-sysv: %(cpp_sysv_default) @} \
398 %@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}"
400 #undef CPP_SYSV_DEFAULT
401 #define CPP_SYSV_DEFAULT "-D_CALL_SYSV"
404 while the @file{config/rs6000/eabiaix.h} target file defines
405 @code{CPP_SYSV_DEFAULT} as:
408 #undef CPP_SYSV_DEFAULT
409 #define CPP_SYSV_DEFAULT "-D_CALL_AIX"
413 @defmac LINK_LIBGCC_SPECIAL_1
414 Define this macro if the driver program should find the library
415 @file{libgcc.a}. If you do not define this macro, the driver program will pass
416 the argument @option{-lgcc} to tell the linker to do the search.
419 @defmac LINK_GCC_C_SEQUENCE_SPEC
420 The sequence in which libgcc and libc are specified to the linker.
421 By default this is @code{%G %L %G}.
424 @defmac LINK_COMMAND_SPEC
425 A C string constant giving the complete command line need to execute the
426 linker. When you do this, you will need to update your port each time a
427 change is made to the link command line within @file{gcc.c}. Therefore,
428 define this macro only if you need to completely redefine the command
429 line for invoking the linker and there is no other way to accomplish
430 the effect you need. Overriding this macro may be avoidable by overriding
431 @code{LINK_GCC_C_SEQUENCE_SPEC} instead.
434 @defmac LINK_ELIMINATE_DUPLICATE_LDIRECTORIES
435 A nonzero value causes @command{collect2} to remove duplicate @option{-L@var{directory}} search
436 directories from linking commands. Do not give it a nonzero value if
437 removing duplicate search directories changes the linker's semantics.
440 @defmac MULTILIB_DEFAULTS
441 Define this macro as a C expression for the initializer of an array of
442 string to tell the driver program which options are defaults for this
443 target and thus do not need to be handled specially when using
444 @code{MULTILIB_OPTIONS}.
446 Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in
447 the target makefile fragment or if none of the options listed in
448 @code{MULTILIB_OPTIONS} are set by default.
449 @xref{Target Fragment}.
452 @defmac RELATIVE_PREFIX_NOT_LINKDIR
453 Define this macro to tell @command{gcc} that it should only translate
454 a @option{-B} prefix into a @option{-L} linker option if the prefix
455 indicates an absolute file name.
458 @defmac MD_EXEC_PREFIX
459 If defined, this macro is an additional prefix to try after
460 @code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched
461 when the compiler is built as a cross
462 compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it
463 to the list of directories used to find the assembler in @file{configure.in}.
466 @defmac STANDARD_STARTFILE_PREFIX
467 Define this macro as a C string constant if you wish to override the
468 standard choice of @code{libdir} as the default prefix to
469 try when searching for startup files such as @file{crt0.o}.
470 @code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler
471 is built as a cross compiler.
474 @defmac STANDARD_STARTFILE_PREFIX_1
475 Define this macro as a C string constant if you wish to override the
476 standard choice of @code{/lib} as a prefix to try after the default prefix
477 when searching for startup files such as @file{crt0.o}.
478 @code{STANDARD_STARTFILE_PREFIX_1} is not searched when the compiler
479 is built as a cross compiler.
482 @defmac STANDARD_STARTFILE_PREFIX_2
483 Define this macro as a C string constant if you wish to override the
484 standard choice of @code{/lib} as yet another prefix to try after the
485 default prefix when searching for startup files such as @file{crt0.o}.
486 @code{STANDARD_STARTFILE_PREFIX_2} is not searched when the compiler
487 is built as a cross compiler.
490 @defmac MD_STARTFILE_PREFIX
491 If defined, this macro supplies an additional prefix to try after the
492 standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the
493 compiler is built as a cross compiler.
496 @defmac MD_STARTFILE_PREFIX_1
497 If defined, this macro supplies yet another prefix to try after the
498 standard prefixes. It is not searched when the compiler is built as a
502 @defmac INIT_ENVIRONMENT
503 Define this macro as a C string constant if you wish to set environment
504 variables for programs called by the driver, such as the assembler and
505 loader. The driver passes the value of this macro to @code{putenv} to
506 initialize the necessary environment variables.
509 @defmac LOCAL_INCLUDE_DIR
510 Define this macro as a C string constant if you wish to override the
511 standard choice of @file{/usr/local/include} as the default prefix to
512 try when searching for local header files. @code{LOCAL_INCLUDE_DIR}
513 comes before @code{SYSTEM_INCLUDE_DIR} in the search order.
515 Cross compilers do not search either @file{/usr/local/include} or its
519 @defmac SYSTEM_INCLUDE_DIR
520 Define this macro as a C string constant if you wish to specify a
521 system-specific directory to search for header files before the standard
522 directory. @code{SYSTEM_INCLUDE_DIR} comes before
523 @code{STANDARD_INCLUDE_DIR} in the search order.
525 Cross compilers do not use this macro and do not search the directory
529 @defmac STANDARD_INCLUDE_DIR
530 Define this macro as a C string constant if you wish to override the
531 standard choice of @file{/usr/include} as the default prefix to
532 try when searching for header files.
534 Cross compilers ignore this macro and do not search either
535 @file{/usr/include} or its replacement.
538 @defmac STANDARD_INCLUDE_COMPONENT
539 The ``component'' corresponding to @code{STANDARD_INCLUDE_DIR}.
540 See @code{INCLUDE_DEFAULTS}, below, for the description of components.
541 If you do not define this macro, no component is used.
544 @defmac INCLUDE_DEFAULTS
545 Define this macro if you wish to override the entire default search path
546 for include files. For a native compiler, the default search path
547 usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR},
548 @code{SYSTEM_INCLUDE_DIR}, @code{GPLUSPLUS_INCLUDE_DIR}, and
549 @code{STANDARD_INCLUDE_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR}
550 and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile},
551 and specify private search areas for GCC@. The directory
552 @code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs.
554 The definition should be an initializer for an array of structures.
555 Each array element should have four elements: the directory name (a
556 string constant), the component name (also a string constant), a flag
557 for C++-only directories,
558 and a flag showing that the includes in the directory don't need to be
559 wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of
560 the array with a null element.
562 The component name denotes what GNU package the include file is part of,
563 if any, in all uppercase letters. For example, it might be @samp{GCC}
564 or @samp{BINUTILS}. If the package is part of a vendor-supplied
565 operating system, code the component name as @samp{0}.
567 For example, here is the definition used for VAX/VMS:
570 #define INCLUDE_DEFAULTS \
572 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \
573 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \
574 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \
581 Here is the order of prefixes tried for exec files:
585 Any prefixes specified by the user with @option{-B}.
588 The environment variable @code{GCC_EXEC_PREFIX} or, if @code{GCC_EXEC_PREFIX}
589 is not set and the compiler has not been installed in the configure-time
590 @var{prefix}, the location in which the compiler has actually been installed.
593 The directories specified by the environment variable @code{COMPILER_PATH}.
596 The macro @code{STANDARD_EXEC_PREFIX}, if the compiler has been installed
597 in the configured-time @var{prefix}.
600 The location @file{/usr/libexec/gcc/}, but only if this is a native compiler.
603 The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
606 The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
610 Here is the order of prefixes tried for startfiles:
614 Any prefixes specified by the user with @option{-B}.
617 The environment variable @code{GCC_EXEC_PREFIX} or its automatically determined
618 value based on the installed toolchain location.
621 The directories specified by the environment variable @code{LIBRARY_PATH}
622 (or port-specific name; native only, cross compilers do not use this).
625 The macro @code{STANDARD_EXEC_PREFIX}, but only if the toolchain is installed
626 in the configured @var{prefix} or this is a native compiler.
629 The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
632 The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
636 The macro @code{MD_STARTFILE_PREFIX}, if defined, but only if this is a
637 native compiler, or we have a target system root.
640 The macro @code{MD_STARTFILE_PREFIX_1}, if defined, but only if this is a
641 native compiler, or we have a target system root.
644 The macro @code{STANDARD_STARTFILE_PREFIX}, with any sysroot modifications.
645 If this path is relative it will be prefixed by @code{GCC_EXEC_PREFIX} and
646 the machine suffix or @code{STANDARD_EXEC_PREFIX} and the machine suffix.
649 The macro @code{STANDARD_STARTFILE_PREFIX_1}, but only if this is a native
650 compiler, or we have a target system root. The default for this macro is
654 The macro @code{STANDARD_STARTFILE_PREFIX_2}, but only if this is a native
655 compiler, or we have a target system root. The default for this macro is
659 @node Run-time Target
660 @section Run-time Target Specification
661 @cindex run-time target specification
662 @cindex predefined macros
663 @cindex target specifications
665 @c prevent bad page break with this line
666 Here are run-time target specifications.
668 @defmac TARGET_CPU_CPP_BUILTINS ()
669 This function-like macro expands to a block of code that defines
670 built-in preprocessor macros and assertions for the target CPU, using
671 the functions @code{builtin_define}, @code{builtin_define_std} and
672 @code{builtin_assert}. When the front end
673 calls this macro it provides a trailing semicolon, and since it has
674 finished command line option processing your code can use those
677 @code{builtin_assert} takes a string in the form you pass to the
678 command-line option @option{-A}, such as @code{cpu=mips}, and creates
679 the assertion. @code{builtin_define} takes a string in the form
680 accepted by option @option{-D} and unconditionally defines the macro.
682 @code{builtin_define_std} takes a string representing the name of an
683 object-like macro. If it doesn't lie in the user's namespace,
684 @code{builtin_define_std} defines it unconditionally. Otherwise, it
685 defines a version with two leading underscores, and another version
686 with two leading and trailing underscores, and defines the original
687 only if an ISO standard was not requested on the command line. For
688 example, passing @code{unix} defines @code{__unix}, @code{__unix__}
689 and possibly @code{unix}; passing @code{_mips} defines @code{__mips},
690 @code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64}
691 defines only @code{_ABI64}.
693 You can also test for the C dialect being compiled. The variable
694 @code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus}
695 or @code{clk_objective_c}. Note that if we are preprocessing
696 assembler, this variable will be @code{clk_c} but the function-like
697 macro @code{preprocessing_asm_p()} will return true, so you might want
698 to check for that first. If you need to check for strict ANSI, the
699 variable @code{flag_iso} can be used. The function-like macro
700 @code{preprocessing_trad_p()} can be used to check for traditional
704 @defmac TARGET_OS_CPP_BUILTINS ()
705 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
706 and is used for the target operating system instead.
709 @defmac TARGET_OBJFMT_CPP_BUILTINS ()
710 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
711 and is used for the target object format. @file{elfos.h} uses this
712 macro to define @code{__ELF__}, so you probably do not need to define
716 @deftypevar {extern int} target_flags
717 This variable is declared in @file{options.h}, which is included before
718 any target-specific headers.
721 @hook TARGET_DEFAULT_TARGET_FLAGS
722 This variable specifies the initial value of @code{target_flags}.
723 Its default setting is 0.
726 @cindex optional hardware or system features
727 @cindex features, optional, in system conventions
729 @hook TARGET_HANDLE_OPTION
730 This hook is called whenever the user specifies one of the
731 target-specific options described by the @file{.opt} definition files
732 (@pxref{Options}). It has the opportunity to do some option-specific
733 processing and should return true if the option is valid. The default
734 definition does nothing but return true.
736 @var{code} specifies the @code{OPT_@var{name}} enumeration value
737 associated with the selected option; @var{name} is just a rendering of
738 the option name in which non-alphanumeric characters are replaced by
739 underscores. @var{arg} specifies the string argument and is null if
740 no argument was given. If the option is flagged as a @code{UInteger}
741 (@pxref{Option properties}), @var{value} is the numeric value of the
742 argument. Otherwise @var{value} is 1 if the positive form of the
743 option was used and 0 if the ``no-'' form was.
746 @hook TARGET_HANDLE_C_OPTION
747 This target hook is called whenever the user specifies one of the
748 target-specific C language family options described by the @file{.opt}
749 definition files(@pxref{Options}). It has the opportunity to do some
750 option-specific processing and should return true if the option is
751 valid. The arguments are like for @code{TARGET_HANDLE_OPTION}. The
752 default definition does nothing but return false.
754 In general, you should use @code{TARGET_HANDLE_OPTION} to handle
755 options. However, if processing an option requires routines that are
756 only available in the C (and related language) front ends, then you
757 should use @code{TARGET_HANDLE_C_OPTION} instead.
760 @defmac TARGET_VERSION
761 This macro is a C statement to print on @code{stderr} a string
762 describing the particular machine description choice. Every machine
763 description should define @code{TARGET_VERSION}. For example:
767 #define TARGET_VERSION \
768 fprintf (stderr, " (68k, Motorola syntax)");
770 #define TARGET_VERSION \
771 fprintf (stderr, " (68k, MIT syntax)");
776 @hook TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE
777 This target function is similar to the hook @code{TARGET_OPTION_OVERRIDE}
778 but is called when the optimize level is changed via an attribute or
779 pragma or when it is reset at the end of the code affected by the
780 attribute or pragma. It is not called at the beginning of compilation
781 when @code{TARGET_OPTION_OVERRIDE} is called so if you want to perform these
782 actions then, you should have @code{TARGET_OPTION_OVERRIDE} call
783 @code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}.
786 @defmac C_COMMON_OVERRIDE_OPTIONS
787 This is similar to the @code{TARGET_OPTION_OVERRIDE} hook
788 but is only used in the C
789 language frontends (C, Objective-C, C++, Objective-C++) and so can be
790 used to alter option flag variables which only exist in those
794 @hook TARGET_OPTION_OPTIMIZATION
795 Some machines may desire to change what optimizations are performed for
796 various optimization levels. This hook, if defined, is executed once
797 just after the optimization level is determined and before the remainder
798 of the command options have been parsed. Values set in this macro are
799 used as the default values for the other command line options.
801 @var{level} is the optimization level specified; 2 if @option{-O2} is
802 specified, 1 if @option{-O} is specified, and 0 if neither is specified.
804 @var{size} is nonzero if @option{-Os} is specified and zero otherwise.
806 This macro is run once at program startup and when the optimization
807 options are changed via @code{#pragma GCC optimize} or by using the
808 @code{optimize} attribute.
810 @strong{Do not examine @code{write_symbols} in
811 this hook!} The debugging options are not supposed to alter the
815 @hook TARGET_OPTION_DEFAULT_PARAMS
818 This hook is called in response to the user invoking
819 @option{--target-help} on the command line. It gives the target a
820 chance to display extra information on the target specific command
821 line options found in its @file{.opt} file.
824 @defmac CAN_DEBUG_WITHOUT_FP
825 Define this macro if debugging can be performed even without a frame
826 pointer. If this macro is defined, GCC will turn on the
827 @option{-fomit-frame-pointer} option whenever @option{-O} is specified.
830 @defmac SWITCHABLE_TARGET
831 Some targets need to switch between substantially different subtargets
832 during compilation. For example, the MIPS target has one subtarget for
833 the traditional MIPS architecture and another for MIPS16. Source code
834 can switch between these two subarchitectures using the @code{mips16}
835 and @code{nomips16} attributes.
837 Such subtargets can differ in things like the set of available
838 registers, the set of available instructions, the costs of various
839 operations, and so on. GCC caches a lot of this type of information
840 in global variables, and recomputing them for each subtarget takes a
841 significant amount of time. The compiler therefore provides a facility
842 for maintaining several versions of the global variables and quickly
843 switching between them; see @file{target-globals.h} for details.
845 Define this macro to 1 if your target needs this facility. The default
849 @node Per-Function Data
850 @section Defining data structures for per-function information.
851 @cindex per-function data
852 @cindex data structures
854 If the target needs to store information on a per-function basis, GCC
855 provides a macro and a couple of variables to allow this. Note, just
856 using statics to store the information is a bad idea, since GCC supports
857 nested functions, so you can be halfway through encoding one function
858 when another one comes along.
860 GCC defines a data structure called @code{struct function} which
861 contains all of the data specific to an individual function. This
862 structure contains a field called @code{machine} whose type is
863 @code{struct machine_function *}, which can be used by targets to point
864 to their own specific data.
866 If a target needs per-function specific data it should define the type
867 @code{struct machine_function} and also the macro @code{INIT_EXPANDERS}.
868 This macro should be used to initialize the function pointer
869 @code{init_machine_status}. This pointer is explained below.
871 One typical use of per-function, target specific data is to create an
872 RTX to hold the register containing the function's return address. This
873 RTX can then be used to implement the @code{__builtin_return_address}
874 function, for level 0.
876 Note---earlier implementations of GCC used a single data area to hold
877 all of the per-function information. Thus when processing of a nested
878 function began the old per-function data had to be pushed onto a
879 stack, and when the processing was finished, it had to be popped off the
880 stack. GCC used to provide function pointers called
881 @code{save_machine_status} and @code{restore_machine_status} to handle
882 the saving and restoring of the target specific information. Since the
883 single data area approach is no longer used, these pointers are no
886 @defmac INIT_EXPANDERS
887 Macro called to initialize any target specific information. This macro
888 is called once per function, before generation of any RTL has begun.
889 The intention of this macro is to allow the initialization of the
890 function pointer @code{init_machine_status}.
893 @deftypevar {void (*)(struct function *)} init_machine_status
894 If this function pointer is non-@code{NULL} it will be called once per
895 function, before function compilation starts, in order to allow the
896 target to perform any target specific initialization of the
897 @code{struct function} structure. It is intended that this would be
898 used to initialize the @code{machine} of that structure.
900 @code{struct machine_function} structures are expected to be freed by GC@.
901 Generally, any memory that they reference must be allocated by using
902 GC allocation, including the structure itself.
906 @section Storage Layout
907 @cindex storage layout
909 Note that the definitions of the macros in this table which are sizes or
910 alignments measured in bits do not need to be constant. They can be C
911 expressions that refer to static variables, such as the @code{target_flags}.
912 @xref{Run-time Target}.
914 @defmac BITS_BIG_ENDIAN
915 Define this macro to have the value 1 if the most significant bit in a
916 byte has the lowest number; otherwise define it to have the value zero.
917 This means that bit-field instructions count from the most significant
918 bit. If the machine has no bit-field instructions, then this must still
919 be defined, but it doesn't matter which value it is defined to. This
920 macro need not be a constant.
922 This macro does not affect the way structure fields are packed into
923 bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
926 @defmac BYTES_BIG_ENDIAN
927 Define this macro to have the value 1 if the most significant byte in a
928 word has the lowest number. This macro need not be a constant.
931 @defmac WORDS_BIG_ENDIAN
932 Define this macro to have the value 1 if, in a multiword object, the
933 most significant word has the lowest number. This applies to both
934 memory locations and registers; GCC fundamentally assumes that the
935 order of words in memory is the same as the order in registers. This
936 macro need not be a constant.
939 @defmac LIBGCC2_WORDS_BIG_ENDIAN
940 Define this macro if @code{WORDS_BIG_ENDIAN} is not constant. This must be a
941 constant value with the same meaning as @code{WORDS_BIG_ENDIAN}, which will be
942 used only when compiling @file{libgcc2.c}. Typically the value will be set
943 based on preprocessor defines.
946 @defmac FLOAT_WORDS_BIG_ENDIAN
947 Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or
948 @code{TFmode} floating point numbers are stored in memory with the word
949 containing the sign bit at the lowest address; otherwise define it to
950 have the value 0. This macro need not be a constant.
952 You need not define this macro if the ordering is the same as for
956 @defmac BITS_PER_UNIT
957 Define this macro to be the number of bits in an addressable storage
958 unit (byte). If you do not define this macro the default is 8.
961 @defmac BITS_PER_WORD
962 Number of bits in a word. If you do not define this macro, the default
963 is @code{BITS_PER_UNIT * UNITS_PER_WORD}.
966 @defmac MAX_BITS_PER_WORD
967 Maximum number of bits in a word. If this is undefined, the default is
968 @code{BITS_PER_WORD}. Otherwise, it is the constant value that is the
969 largest value that @code{BITS_PER_WORD} can have at run-time.
972 @defmac UNITS_PER_WORD
973 Number of storage units in a word; normally the size of a general-purpose
974 register, a power of two from 1 or 8.
977 @defmac MIN_UNITS_PER_WORD
978 Minimum number of units in a word. If this is undefined, the default is
979 @code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the
980 smallest value that @code{UNITS_PER_WORD} can have at run-time.
984 Width of a pointer, in bits. You must specify a value no wider than the
985 width of @code{Pmode}. If it is not equal to the width of @code{Pmode},
986 you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify
987 a value the default is @code{BITS_PER_WORD}.
990 @defmac POINTERS_EXTEND_UNSIGNED
991 A C expression that determines how pointers should be extended from
992 @code{ptr_mode} to either @code{Pmode} or @code{word_mode}. It is
993 greater than zero if pointers should be zero-extended, zero if they
994 should be sign-extended, and negative if some other sort of conversion
995 is needed. In the last case, the extension is done by the target's
996 @code{ptr_extend} instruction.
998 You need not define this macro if the @code{ptr_mode}, @code{Pmode}
999 and @code{word_mode} are all the same width.
1002 @defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type})
1003 A macro to update @var{m} and @var{unsignedp} when an object whose type
1004 is @var{type} and which has the specified mode and signedness is to be
1005 stored in a register. This macro is only called when @var{type} is a
1008 On most RISC machines, which only have operations that operate on a full
1009 register, define this macro to set @var{m} to @code{word_mode} if
1010 @var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most
1011 cases, only integer modes should be widened because wider-precision
1012 floating-point operations are usually more expensive than their narrower
1015 For most machines, the macro definition does not change @var{unsignedp}.
1016 However, some machines, have instructions that preferentially handle
1017 either signed or unsigned quantities of certain modes. For example, on
1018 the DEC Alpha, 32-bit loads from memory and 32-bit add instructions
1019 sign-extend the result to 64 bits. On such machines, set
1020 @var{unsignedp} according to which kind of extension is more efficient.
1022 Do not define this macro if it would never modify @var{m}.
1025 @hook TARGET_PROMOTE_FUNCTION_MODE
1026 Like @code{PROMOTE_MODE}, but it is applied to outgoing function arguments or
1027 function return values. The target hook should return the new mode
1028 and possibly change @code{*@var{punsignedp}} if the promotion should
1029 change signedness. This function is called only for scalar @emph{or
1032 @var{for_return} allows to distinguish the promotion of arguments and
1033 return values. If it is @code{1}, a return value is being promoted and
1034 @code{TARGET_FUNCTION_VALUE} must perform the same promotions done here.
1035 If it is @code{2}, the returned mode should be that of the register in
1036 which an incoming parameter is copied, or the outgoing result is computed;
1037 then the hook should return the same mode as @code{promote_mode}, though
1038 the signedness may be different.
1040 The default is to not promote arguments and return values. You can
1041 also define the hook to @code{default_promote_function_mode_always_promote}
1042 if you would like to apply the same rules given by @code{PROMOTE_MODE}.
1045 @defmac PARM_BOUNDARY
1046 Normal alignment required for function parameters on the stack, in
1047 bits. All stack parameters receive at least this much alignment
1048 regardless of data type. On most machines, this is the same as the
1052 @defmac STACK_BOUNDARY
1053 Define this macro to the minimum alignment enforced by hardware for the
1054 stack pointer on this machine. The definition is a C expression for the
1055 desired alignment (measured in bits). This value is used as a default
1056 if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines,
1057 this should be the same as @code{PARM_BOUNDARY}.
1060 @defmac PREFERRED_STACK_BOUNDARY
1061 Define this macro if you wish to preserve a certain alignment for the
1062 stack pointer, greater than what the hardware enforces. The definition
1063 is a C expression for the desired alignment (measured in bits). This
1064 macro must evaluate to a value equal to or larger than
1065 @code{STACK_BOUNDARY}.
1068 @defmac INCOMING_STACK_BOUNDARY
1069 Define this macro if the incoming stack boundary may be different
1070 from @code{PREFERRED_STACK_BOUNDARY}. This macro must evaluate
1071 to a value equal to or larger than @code{STACK_BOUNDARY}.
1074 @defmac FUNCTION_BOUNDARY
1075 Alignment required for a function entry point, in bits.
1078 @defmac BIGGEST_ALIGNMENT
1079 Biggest alignment that any data type can require on this machine, in
1080 bits. Note that this is not the biggest alignment that is supported,
1081 just the biggest alignment that, when violated, may cause a fault.
1084 @defmac MALLOC_ABI_ALIGNMENT
1085 Alignment, in bits, a C conformant malloc implementation has to
1086 provide. If not defined, the default value is @code{BITS_PER_WORD}.
1089 @defmac ATTRIBUTE_ALIGNED_VALUE
1090 Alignment used by the @code{__attribute__ ((aligned))} construct. If
1091 not defined, the default value is @code{BIGGEST_ALIGNMENT}.
1094 @defmac MINIMUM_ATOMIC_ALIGNMENT
1095 If defined, the smallest alignment, in bits, that can be given to an
1096 object that can be referenced in one operation, without disturbing any
1097 nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger
1098 on machines that don't have byte or half-word store operations.
1101 @defmac BIGGEST_FIELD_ALIGNMENT
1102 Biggest alignment that any structure or union field can require on this
1103 machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for
1104 structure and union fields only, unless the field alignment has been set
1105 by the @code{__attribute__ ((aligned (@var{n})))} construct.
1108 @defmac ADJUST_FIELD_ALIGN (@var{field}, @var{computed})
1109 An expression for the alignment of a structure field @var{field} if the
1110 alignment computed in the usual way (including applying of
1111 @code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the
1112 alignment) is @var{computed}. It overrides alignment only if the
1113 field alignment has not been set by the
1114 @code{__attribute__ ((aligned (@var{n})))} construct.
1117 @defmac MAX_STACK_ALIGNMENT
1118 Biggest stack alignment guaranteed by the backend. Use this macro
1119 to specify the maximum alignment of a variable on stack.
1121 If not defined, the default value is @code{STACK_BOUNDARY}.
1123 @c FIXME: The default should be @code{PREFERRED_STACK_BOUNDARY}.
1124 @c But the fix for PR 32893 indicates that we can only guarantee
1125 @c maximum stack alignment on stack up to @code{STACK_BOUNDARY}, not
1126 @c @code{PREFERRED_STACK_BOUNDARY}, if stack alignment isn't supported.
1129 @defmac MAX_OFILE_ALIGNMENT
1130 Biggest alignment supported by the object file format of this machine.
1131 Use this macro to limit the alignment which can be specified using the
1132 @code{__attribute__ ((aligned (@var{n})))} construct. If not defined,
1133 the default value is @code{BIGGEST_ALIGNMENT}.
1135 On systems that use ELF, the default (in @file{config/elfos.h}) is
1136 the largest supported 32-bit ELF section alignment representable on
1137 a 32-bit host e.g. @samp{(((unsigned HOST_WIDEST_INT) 1 << 28) * 8)}.
1138 On 32-bit ELF the largest supported section alignment in bits is
1139 @samp{(0x80000000 * 8)}, but this is not representable on 32-bit hosts.
1142 @defmac DATA_ALIGNMENT (@var{type}, @var{basic-align})
1143 If defined, a C expression to compute the alignment for a variable in
1144 the static store. @var{type} is the data type, and @var{basic-align} is
1145 the alignment that the object would ordinarily have. The value of this
1146 macro is used instead of that alignment to align the object.
1148 If this macro is not defined, then @var{basic-align} is used.
1151 One use of this macro is to increase alignment of medium-size data to
1152 make it all fit in fewer cache lines. Another is to cause character
1153 arrays to be word-aligned so that @code{strcpy} calls that copy
1154 constants to character arrays can be done inline.
1157 @defmac CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align})
1158 If defined, a C expression to compute the alignment given to a constant
1159 that is being placed in memory. @var{constant} is the constant and
1160 @var{basic-align} is the alignment that the object would ordinarily
1161 have. The value of this macro is used instead of that alignment to
1164 If this macro is not defined, then @var{basic-align} is used.
1166 The typical use of this macro is to increase alignment for string
1167 constants to be word aligned so that @code{strcpy} calls that copy
1168 constants can be done inline.
1171 @defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align})
1172 If defined, a C expression to compute the alignment for a variable in
1173 the local store. @var{type} is the data type, and @var{basic-align} is
1174 the alignment that the object would ordinarily have. The value of this
1175 macro is used instead of that alignment to align the object.
1177 If this macro is not defined, then @var{basic-align} is used.
1179 One use of this macro is to increase alignment of medium-size data to
1180 make it all fit in fewer cache lines.
1183 @defmac STACK_SLOT_ALIGNMENT (@var{type}, @var{mode}, @var{basic-align})
1184 If defined, a C expression to compute the alignment for stack slot.
1185 @var{type} is the data type, @var{mode} is the widest mode available,
1186 and @var{basic-align} is the alignment that the slot would ordinarily
1187 have. The value of this macro is used instead of that alignment to
1190 If this macro is not defined, then @var{basic-align} is used when
1191 @var{type} is @code{NULL}. Otherwise, @code{LOCAL_ALIGNMENT} will
1194 This macro is to set alignment of stack slot to the maximum alignment
1195 of all possible modes which the slot may have.
1198 @defmac LOCAL_DECL_ALIGNMENT (@var{decl})
1199 If defined, a C expression to compute the alignment for a local
1200 variable @var{decl}.
1202 If this macro is not defined, then
1203 @code{LOCAL_ALIGNMENT (TREE_TYPE (@var{decl}), DECL_ALIGN (@var{decl}))}
1206 One use of this macro is to increase alignment of medium-size data to
1207 make it all fit in fewer cache lines.
1210 @defmac MINIMUM_ALIGNMENT (@var{exp}, @var{mode}, @var{align})
1211 If defined, a C expression to compute the minimum required alignment
1212 for dynamic stack realignment purposes for @var{exp} (a type or decl),
1213 @var{mode}, assuming normal alignment @var{align}.
1215 If this macro is not defined, then @var{align} will be used.
1218 @defmac EMPTY_FIELD_BOUNDARY
1219 Alignment in bits to be given to a structure bit-field that follows an
1220 empty field such as @code{int : 0;}.
1222 If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro.
1225 @defmac STRUCTURE_SIZE_BOUNDARY
1226 Number of bits which any structure or union's size must be a multiple of.
1227 Each structure or union's size is rounded up to a multiple of this.
1229 If you do not define this macro, the default is the same as
1230 @code{BITS_PER_UNIT}.
1233 @defmac STRICT_ALIGNMENT
1234 Define this macro to be the value 1 if instructions will fail to work
1235 if given data not on the nominal alignment. If instructions will merely
1236 go slower in that case, define this macro as 0.
1239 @defmac PCC_BITFIELD_TYPE_MATTERS
1240 Define this if you wish to imitate the way many other C compilers handle
1241 alignment of bit-fields and the structures that contain them.
1243 The behavior is that the type written for a named bit-field (@code{int},
1244 @code{short}, or other integer type) imposes an alignment for the entire
1245 structure, as if the structure really did contain an ordinary field of
1246 that type. In addition, the bit-field is placed within the structure so
1247 that it would fit within such a field, not crossing a boundary for it.
1249 Thus, on most machines, a named bit-field whose type is written as
1250 @code{int} would not cross a four-byte boundary, and would force
1251 four-byte alignment for the whole structure. (The alignment used may
1252 not be four bytes; it is controlled by the other alignment parameters.)
1254 An unnamed bit-field will not affect the alignment of the containing
1257 If the macro is defined, its definition should be a C expression;
1258 a nonzero value for the expression enables this behavior.
1260 Note that if this macro is not defined, or its value is zero, some
1261 bit-fields may cross more than one alignment boundary. The compiler can
1262 support such references if there are @samp{insv}, @samp{extv}, and
1263 @samp{extzv} insns that can directly reference memory.
1265 The other known way of making bit-fields work is to define
1266 @code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}.
1267 Then every structure can be accessed with fullwords.
1269 Unless the machine has bit-field instructions or you define
1270 @code{STRUCTURE_SIZE_BOUNDARY} that way, you must define
1271 @code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value.
1273 If your aim is to make GCC use the same conventions for laying out
1274 bit-fields as are used by another compiler, here is how to investigate
1275 what the other compiler does. Compile and run this program:
1294 printf ("Size of foo1 is %d\n",
1295 sizeof (struct foo1));
1296 printf ("Size of foo2 is %d\n",
1297 sizeof (struct foo2));
1302 If this prints 2 and 5, then the compiler's behavior is what you would
1303 get from @code{PCC_BITFIELD_TYPE_MATTERS}.
1306 @defmac BITFIELD_NBYTES_LIMITED
1307 Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited
1308 to aligning a bit-field within the structure.
1311 @hook TARGET_ALIGN_ANON_BITFIELD
1312 When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine
1313 whether unnamed bitfields affect the alignment of the containing
1314 structure. The hook should return true if the structure should inherit
1315 the alignment requirements of an unnamed bitfield's type.
1318 @hook TARGET_NARROW_VOLATILE_BITFIELD
1319 This target hook should return @code{true} if accesses to volatile bitfields
1320 should use the narrowest mode possible. It should return @code{false} if
1321 these accesses should use the bitfield container type.
1323 The default is @code{!TARGET_STRICT_ALIGN}.
1326 @defmac MEMBER_TYPE_FORCES_BLK (@var{field}, @var{mode})
1327 Return 1 if a structure or array containing @var{field} should be accessed using
1330 If @var{field} is the only field in the structure, @var{mode} is its
1331 mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the
1332 case where structures of one field would require the structure's mode to
1333 retain the field's mode.
1335 Normally, this is not needed.
1338 @defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified})
1339 Define this macro as an expression for the alignment of a type (given
1340 by @var{type} as a tree node) if the alignment computed in the usual
1341 way is @var{computed} and the alignment explicitly specified was
1344 The default is to use @var{specified} if it is larger; otherwise, use
1345 the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT}
1348 @defmac MAX_FIXED_MODE_SIZE
1349 An integer expression for the size in bits of the largest integer
1350 machine mode that should actually be used. All integer machine modes of
1351 this size or smaller can be used for structures and unions with the
1352 appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE
1353 (DImode)} is assumed.
1356 @defmac STACK_SAVEAREA_MODE (@var{save_level})
1357 If defined, an expression of type @code{enum machine_mode} that
1358 specifies the mode of the save area operand of a
1359 @code{save_stack_@var{level}} named pattern (@pxref{Standard Names}).
1360 @var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or
1361 @code{SAVE_NONLOCAL} and selects which of the three named patterns is
1362 having its mode specified.
1364 You need not define this macro if it always returns @code{Pmode}. You
1365 would most commonly define this macro if the
1366 @code{save_stack_@var{level}} patterns need to support both a 32- and a
1370 @defmac STACK_SIZE_MODE
1371 If defined, an expression of type @code{enum machine_mode} that
1372 specifies the mode of the size increment operand of an
1373 @code{allocate_stack} named pattern (@pxref{Standard Names}).
1375 You need not define this macro if it always returns @code{word_mode}.
1376 You would most commonly define this macro if the @code{allocate_stack}
1377 pattern needs to support both a 32- and a 64-bit mode.
1380 @hook TARGET_LIBGCC_CMP_RETURN_MODE
1381 This target hook should return the mode to be used for the return value
1382 of compare instructions expanded to libgcc calls. If not defined
1383 @code{word_mode} is returned which is the right choice for a majority of
1387 @hook TARGET_LIBGCC_SHIFT_COUNT_MODE
1388 This target hook should return the mode to be used for the shift count operand
1389 of shift instructions expanded to libgcc calls. If not defined
1390 @code{word_mode} is returned which is the right choice for a majority of
1394 @hook TARGET_UNWIND_WORD_MODE
1395 Return machine mode to be used for @code{_Unwind_Word} type.
1396 The default is to use @code{word_mode}.
1399 @defmac ROUND_TOWARDS_ZERO
1400 If defined, this macro should be true if the prevailing rounding
1401 mode is towards zero.
1403 Defining this macro only affects the way @file{libgcc.a} emulates
1404 floating-point arithmetic.
1406 Not defining this macro is equivalent to returning zero.
1409 @defmac LARGEST_EXPONENT_IS_NORMAL (@var{size})
1410 This macro should return true if floats with @var{size}
1411 bits do not have a NaN or infinity representation, but use the largest
1412 exponent for normal numbers instead.
1414 Defining this macro only affects the way @file{libgcc.a} emulates
1415 floating-point arithmetic.
1417 The default definition of this macro returns false for all sizes.
1420 @hook TARGET_MS_BITFIELD_LAYOUT_P
1421 This target hook returns @code{true} if bit-fields in the given
1422 @var{record_type} are to be laid out following the rules of Microsoft
1423 Visual C/C++, namely: (i) a bit-field won't share the same storage
1424 unit with the previous bit-field if their underlying types have
1425 different sizes, and the bit-field will be aligned to the highest
1426 alignment of the underlying types of itself and of the previous
1427 bit-field; (ii) a zero-sized bit-field will affect the alignment of
1428 the whole enclosing structure, even if it is unnamed; except that
1429 (iii) a zero-sized bit-field will be disregarded unless it follows
1430 another bit-field of nonzero size. If this hook returns @code{true},
1431 other macros that control bit-field layout are ignored.
1433 When a bit-field is inserted into a packed record, the whole size
1434 of the underlying type is used by one or more same-size adjacent
1435 bit-fields (that is, if its long:3, 32 bits is used in the record,
1436 and any additional adjacent long bit-fields are packed into the same
1437 chunk of 32 bits. However, if the size changes, a new field of that
1438 size is allocated). In an unpacked record, this is the same as using
1439 alignment, but not equivalent when packing.
1441 If both MS bit-fields and @samp{__attribute__((packed))} are used,
1442 the latter will take precedence. If @samp{__attribute__((packed))} is
1443 used on a single field when MS bit-fields are in use, it will take
1444 precedence for that field, but the alignment of the rest of the structure
1445 may affect its placement.
1448 @hook TARGET_DECIMAL_FLOAT_SUPPORTED_P
1449 Returns true if the target supports decimal floating point.
1452 @hook TARGET_FIXED_POINT_SUPPORTED_P
1453 Returns true if the target supports fixed-point arithmetic.
1456 @hook TARGET_EXPAND_TO_RTL_HOOK
1457 This hook is called just before expansion into rtl, allowing the target
1458 to perform additional initializations or analysis before the expansion.
1459 For example, the rs6000 port uses it to allocate a scratch stack slot
1460 for use in copying SDmode values between memory and floating point
1461 registers whenever the function being expanded has any SDmode
1465 @hook TARGET_INSTANTIATE_DECLS
1466 This hook allows the backend to perform additional instantiations on rtl
1467 that are not actually in any insns yet, but will be later.
1470 @hook TARGET_MANGLE_TYPE
1471 If your target defines any fundamental types, or any types your target
1472 uses should be mangled differently from the default, define this hook
1473 to return the appropriate encoding for these types as part of a C++
1474 mangled name. The @var{type} argument is the tree structure representing
1475 the type to be mangled. The hook may be applied to trees which are
1476 not target-specific fundamental types; it should return @code{NULL}
1477 for all such types, as well as arguments it does not recognize. If the
1478 return value is not @code{NULL}, it must point to a statically-allocated
1481 Target-specific fundamental types might be new fundamental types or
1482 qualified versions of ordinary fundamental types. Encode new
1483 fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name}
1484 is the name used for the type in source code, and @var{n} is the
1485 length of @var{name} in decimal. Encode qualified versions of
1486 ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where
1487 @var{name} is the name used for the type qualifier in source code,
1488 @var{n} is the length of @var{name} as above, and @var{code} is the
1489 code used to represent the unqualified version of this type. (See
1490 @code{write_builtin_type} in @file{cp/mangle.c} for the list of
1491 codes.) In both cases the spaces are for clarity; do not include any
1492 spaces in your string.
1494 This hook is applied to types prior to typedef resolution. If the mangled
1495 name for a particular type depends only on that type's main variant, you
1496 can perform typedef resolution yourself using @code{TYPE_MAIN_VARIANT}
1499 The default version of this hook always returns @code{NULL}, which is
1500 appropriate for a target that does not define any new fundamental
1505 @section Layout of Source Language Data Types
1507 These macros define the sizes and other characteristics of the standard
1508 basic data types used in programs being compiled. Unlike the macros in
1509 the previous section, these apply to specific features of C and related
1510 languages, rather than to fundamental aspects of storage layout.
1512 @defmac INT_TYPE_SIZE
1513 A C expression for the size in bits of the type @code{int} on the
1514 target machine. If you don't define this, the default is one word.
1517 @defmac SHORT_TYPE_SIZE
1518 A C expression for the size in bits of the type @code{short} on the
1519 target machine. If you don't define this, the default is half a word.
1520 (If this would be less than one storage unit, it is rounded up to one
1524 @defmac LONG_TYPE_SIZE
1525 A C expression for the size in bits of the type @code{long} on the
1526 target machine. If you don't define this, the default is one word.
1529 @defmac ADA_LONG_TYPE_SIZE
1530 On some machines, the size used for the Ada equivalent of the type
1531 @code{long} by a native Ada compiler differs from that used by C@. In
1532 that situation, define this macro to be a C expression to be used for
1533 the size of that type. If you don't define this, the default is the
1534 value of @code{LONG_TYPE_SIZE}.
1537 @defmac LONG_LONG_TYPE_SIZE
1538 A C expression for the size in bits of the type @code{long long} on the
1539 target machine. If you don't define this, the default is two
1540 words. If you want to support GNU Ada on your machine, the value of this
1541 macro must be at least 64.
1544 @defmac CHAR_TYPE_SIZE
1545 A C expression for the size in bits of the type @code{char} on the
1546 target machine. If you don't define this, the default is
1547 @code{BITS_PER_UNIT}.
1550 @defmac BOOL_TYPE_SIZE
1551 A C expression for the size in bits of the C++ type @code{bool} and
1552 C99 type @code{_Bool} on the target machine. If you don't define
1553 this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}.
1556 @defmac FLOAT_TYPE_SIZE
1557 A C expression for the size in bits of the type @code{float} on the
1558 target machine. If you don't define this, the default is one word.
1561 @defmac DOUBLE_TYPE_SIZE
1562 A C expression for the size in bits of the type @code{double} on the
1563 target machine. If you don't define this, the default is two
1567 @defmac LONG_DOUBLE_TYPE_SIZE
1568 A C expression for the size in bits of the type @code{long double} on
1569 the target machine. If you don't define this, the default is two
1573 @defmac SHORT_FRACT_TYPE_SIZE
1574 A C expression for the size in bits of the type @code{short _Fract} on
1575 the target machine. If you don't define this, the default is
1576 @code{BITS_PER_UNIT}.
1579 @defmac FRACT_TYPE_SIZE
1580 A C expression for the size in bits of the type @code{_Fract} on
1581 the target machine. If you don't define this, the default is
1582 @code{BITS_PER_UNIT * 2}.
1585 @defmac LONG_FRACT_TYPE_SIZE
1586 A C expression for the size in bits of the type @code{long _Fract} on
1587 the target machine. If you don't define this, the default is
1588 @code{BITS_PER_UNIT * 4}.
1591 @defmac LONG_LONG_FRACT_TYPE_SIZE
1592 A C expression for the size in bits of the type @code{long long _Fract} on
1593 the target machine. If you don't define this, the default is
1594 @code{BITS_PER_UNIT * 8}.
1597 @defmac SHORT_ACCUM_TYPE_SIZE
1598 A C expression for the size in bits of the type @code{short _Accum} on
1599 the target machine. If you don't define this, the default is
1600 @code{BITS_PER_UNIT * 2}.
1603 @defmac ACCUM_TYPE_SIZE
1604 A C expression for the size in bits of the type @code{_Accum} on
1605 the target machine. If you don't define this, the default is
1606 @code{BITS_PER_UNIT * 4}.
1609 @defmac LONG_ACCUM_TYPE_SIZE
1610 A C expression for the size in bits of the type @code{long _Accum} on
1611 the target machine. If you don't define this, the default is
1612 @code{BITS_PER_UNIT * 8}.
1615 @defmac LONG_LONG_ACCUM_TYPE_SIZE
1616 A C expression for the size in bits of the type @code{long long _Accum} on
1617 the target machine. If you don't define this, the default is
1618 @code{BITS_PER_UNIT * 16}.
1621 @defmac LIBGCC2_LONG_DOUBLE_TYPE_SIZE
1622 Define this macro if @code{LONG_DOUBLE_TYPE_SIZE} is not constant or
1623 if you want routines in @file{libgcc2.a} for a size other than
1624 @code{LONG_DOUBLE_TYPE_SIZE}. If you don't define this, the
1625 default is @code{LONG_DOUBLE_TYPE_SIZE}.
1628 @defmac LIBGCC2_HAS_DF_MODE
1629 Define this macro if neither @code{LIBGCC2_DOUBLE_TYPE_SIZE} nor
1630 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is
1631 @code{DFmode} but you want @code{DFmode} routines in @file{libgcc2.a}
1632 anyway. If you don't define this and either @code{LIBGCC2_DOUBLE_TYPE_SIZE}
1633 or @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64 then the default is 1,
1637 @defmac LIBGCC2_HAS_XF_MODE
1638 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1639 @code{XFmode} but you want @code{XFmode} routines in @file{libgcc2.a}
1640 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1641 is 80 then the default is 1, otherwise it is 0.
1644 @defmac LIBGCC2_HAS_TF_MODE
1645 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1646 @code{TFmode} but you want @code{TFmode} routines in @file{libgcc2.a}
1647 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1648 is 128 then the default is 1, otherwise it is 0.
1655 Define these macros to be the size in bits of the mantissa of
1656 @code{SFmode}, @code{DFmode}, @code{XFmode} and @code{TFmode} values,
1657 if the defaults in @file{libgcc2.h} are inappropriate. By default,
1658 @code{FLT_MANT_DIG} is used for @code{SF_SIZE}, @code{LDBL_MANT_DIG}
1659 for @code{XF_SIZE} and @code{TF_SIZE}, and @code{DBL_MANT_DIG} or
1660 @code{LDBL_MANT_DIG} for @code{DF_SIZE} according to whether
1661 @code{LIBGCC2_DOUBLE_TYPE_SIZE} or
1662 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64.
1665 @defmac TARGET_FLT_EVAL_METHOD
1666 A C expression for the value for @code{FLT_EVAL_METHOD} in @file{float.h},
1667 assuming, if applicable, that the floating-point control word is in its
1668 default state. If you do not define this macro the value of
1669 @code{FLT_EVAL_METHOD} will be zero.
1672 @defmac WIDEST_HARDWARE_FP_SIZE
1673 A C expression for the size in bits of the widest floating-point format
1674 supported by the hardware. If you define this macro, you must specify a
1675 value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}.
1676 If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE}
1680 @defmac DEFAULT_SIGNED_CHAR
1681 An expression whose value is 1 or 0, according to whether the type
1682 @code{char} should be signed or unsigned by default. The user can
1683 always override this default with the options @option{-fsigned-char}
1684 and @option{-funsigned-char}.
1687 @hook TARGET_DEFAULT_SHORT_ENUMS
1688 This target hook should return true if the compiler should give an
1689 @code{enum} type only as many bytes as it takes to represent the range
1690 of possible values of that type. It should return false if all
1691 @code{enum} types should be allocated like @code{int}.
1693 The default is to return false.
1697 A C expression for a string describing the name of the data type to use
1698 for size values. The typedef name @code{size_t} is defined using the
1699 contents of the string.
1701 The string can contain more than one keyword. If so, separate them with
1702 spaces, and write first any length keyword, then @code{unsigned} if
1703 appropriate, and finally @code{int}. The string must exactly match one
1704 of the data type names defined in the function
1705 @code{init_decl_processing} in the file @file{c-decl.c}. You may not
1706 omit @code{int} or change the order---that would cause the compiler to
1709 If you don't define this macro, the default is @code{"long unsigned
1713 @defmac PTRDIFF_TYPE
1714 A C expression for a string describing the name of the data type to use
1715 for the result of subtracting two pointers. The typedef name
1716 @code{ptrdiff_t} is defined using the contents of the string. See
1717 @code{SIZE_TYPE} above for more information.
1719 If you don't define this macro, the default is @code{"long int"}.
1723 A C expression for a string describing the name of the data type to use
1724 for wide characters. The typedef name @code{wchar_t} is defined using
1725 the contents of the string. See @code{SIZE_TYPE} above for more
1728 If you don't define this macro, the default is @code{"int"}.
1731 @defmac WCHAR_TYPE_SIZE
1732 A C expression for the size in bits of the data type for wide
1733 characters. This is used in @code{cpp}, which cannot make use of
1738 A C expression for a string describing the name of the data type to
1739 use for wide characters passed to @code{printf} and returned from
1740 @code{getwc}. The typedef name @code{wint_t} is defined using the
1741 contents of the string. See @code{SIZE_TYPE} above for more
1744 If you don't define this macro, the default is @code{"unsigned int"}.
1748 A C expression for a string describing the name of the data type that
1749 can represent any value of any standard or extended signed integer type.
1750 The typedef name @code{intmax_t} is defined using the contents of the
1751 string. See @code{SIZE_TYPE} above for more information.
1753 If you don't define this macro, the default is the first of
1754 @code{"int"}, @code{"long int"}, or @code{"long long int"} that has as
1755 much precision as @code{long long int}.
1758 @defmac UINTMAX_TYPE
1759 A C expression for a string describing the name of the data type that
1760 can represent any value of any standard or extended unsigned integer
1761 type. The typedef name @code{uintmax_t} is defined using the contents
1762 of the string. See @code{SIZE_TYPE} above for more information.
1764 If you don't define this macro, the default is the first of
1765 @code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long
1766 unsigned int"} that has as much precision as @code{long long unsigned
1770 @defmac SIG_ATOMIC_TYPE
1776 @defmacx UINT16_TYPE
1777 @defmacx UINT32_TYPE
1778 @defmacx UINT64_TYPE
1779 @defmacx INT_LEAST8_TYPE
1780 @defmacx INT_LEAST16_TYPE
1781 @defmacx INT_LEAST32_TYPE
1782 @defmacx INT_LEAST64_TYPE
1783 @defmacx UINT_LEAST8_TYPE
1784 @defmacx UINT_LEAST16_TYPE
1785 @defmacx UINT_LEAST32_TYPE
1786 @defmacx UINT_LEAST64_TYPE
1787 @defmacx INT_FAST8_TYPE
1788 @defmacx INT_FAST16_TYPE
1789 @defmacx INT_FAST32_TYPE
1790 @defmacx INT_FAST64_TYPE
1791 @defmacx UINT_FAST8_TYPE
1792 @defmacx UINT_FAST16_TYPE
1793 @defmacx UINT_FAST32_TYPE
1794 @defmacx UINT_FAST64_TYPE
1795 @defmacx INTPTR_TYPE
1796 @defmacx UINTPTR_TYPE
1797 C expressions for the standard types @code{sig_atomic_t},
1798 @code{int8_t}, @code{int16_t}, @code{int32_t}, @code{int64_t},
1799 @code{uint8_t}, @code{uint16_t}, @code{uint32_t}, @code{uint64_t},
1800 @code{int_least8_t}, @code{int_least16_t}, @code{int_least32_t},
1801 @code{int_least64_t}, @code{uint_least8_t}, @code{uint_least16_t},
1802 @code{uint_least32_t}, @code{uint_least64_t}, @code{int_fast8_t},
1803 @code{int_fast16_t}, @code{int_fast32_t}, @code{int_fast64_t},
1804 @code{uint_fast8_t}, @code{uint_fast16_t}, @code{uint_fast32_t},
1805 @code{uint_fast64_t}, @code{intptr_t}, and @code{uintptr_t}. See
1806 @code{SIZE_TYPE} above for more information.
1808 If any of these macros evaluates to a null pointer, the corresponding
1809 type is not supported; if GCC is configured to provide
1810 @code{<stdint.h>} in such a case, the header provided may not conform
1811 to C99, depending on the type in question. The defaults for all of
1812 these macros are null pointers.
1815 @defmac TARGET_PTRMEMFUNC_VBIT_LOCATION
1816 The C++ compiler represents a pointer-to-member-function with a struct
1823 ptrdiff_t vtable_index;
1830 The C++ compiler must use one bit to indicate whether the function that
1831 will be called through a pointer-to-member-function is virtual.
1832 Normally, we assume that the low-order bit of a function pointer must
1833 always be zero. Then, by ensuring that the vtable_index is odd, we can
1834 distinguish which variant of the union is in use. But, on some
1835 platforms function pointers can be odd, and so this doesn't work. In
1836 that case, we use the low-order bit of the @code{delta} field, and shift
1837 the remainder of the @code{delta} field to the left.
1839 GCC will automatically make the right selection about where to store
1840 this bit using the @code{FUNCTION_BOUNDARY} setting for your platform.
1841 However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY}
1842 set such that functions always start at even addresses, but the lowest
1843 bit of pointers to functions indicate whether the function at that
1844 address is in ARM or Thumb mode. If this is the case of your
1845 architecture, you should define this macro to
1846 @code{ptrmemfunc_vbit_in_delta}.
1848 In general, you should not have to define this macro. On architectures
1849 in which function addresses are always even, according to
1850 @code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to
1851 @code{ptrmemfunc_vbit_in_pfn}.
1854 @defmac TARGET_VTABLE_USES_DESCRIPTORS
1855 Normally, the C++ compiler uses function pointers in vtables. This
1856 macro allows the target to change to use ``function descriptors''
1857 instead. Function descriptors are found on targets for whom a
1858 function pointer is actually a small data structure. Normally the
1859 data structure consists of the actual code address plus a data
1860 pointer to which the function's data is relative.
1862 If vtables are used, the value of this macro should be the number
1863 of words that the function descriptor occupies.
1866 @defmac TARGET_VTABLE_ENTRY_ALIGN
1867 By default, the vtable entries are void pointers, the so the alignment
1868 is the same as pointer alignment. The value of this macro specifies
1869 the alignment of the vtable entry in bits. It should be defined only
1870 when special alignment is necessary. */
1873 @defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE
1874 There are a few non-descriptor entries in the vtable at offsets below
1875 zero. If these entries must be padded (say, to preserve the alignment
1876 specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number
1877 of words in each data entry.
1881 @section Register Usage
1882 @cindex register usage
1884 This section explains how to describe what registers the target machine
1885 has, and how (in general) they can be used.
1887 The description of which registers a specific instruction can use is
1888 done with register classes; see @ref{Register Classes}. For information
1889 on using registers to access a stack frame, see @ref{Frame Registers}.
1890 For passing values in registers, see @ref{Register Arguments}.
1891 For returning values in registers, see @ref{Scalar Return}.
1894 * Register Basics:: Number and kinds of registers.
1895 * Allocation Order:: Order in which registers are allocated.
1896 * Values in Registers:: What kinds of values each reg can hold.
1897 * Leaf Functions:: Renumbering registers for leaf functions.
1898 * Stack Registers:: Handling a register stack such as 80387.
1901 @node Register Basics
1902 @subsection Basic Characteristics of Registers
1904 @c prevent bad page break with this line
1905 Registers have various characteristics.
1907 @defmac FIRST_PSEUDO_REGISTER
1908 Number of hardware registers known to the compiler. They receive
1909 numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1910 pseudo register's number really is assigned the number
1911 @code{FIRST_PSEUDO_REGISTER}.
1914 @defmac FIXED_REGISTERS
1915 @cindex fixed register
1916 An initializer that says which registers are used for fixed purposes
1917 all throughout the compiled code and are therefore not available for
1918 general allocation. These would include the stack pointer, the frame
1919 pointer (except on machines where that can be used as a general
1920 register when no frame pointer is needed), the program counter on
1921 machines where that is considered one of the addressable registers,
1922 and any other numbered register with a standard use.
1924 This information is expressed as a sequence of numbers, separated by
1925 commas and surrounded by braces. The @var{n}th number is 1 if
1926 register @var{n} is fixed, 0 otherwise.
1928 The table initialized from this macro, and the table initialized by
1929 the following one, may be overridden at run time either automatically,
1930 by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
1931 the user with the command options @option{-ffixed-@var{reg}},
1932 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}.
1935 @defmac CALL_USED_REGISTERS
1936 @cindex call-used register
1937 @cindex call-clobbered register
1938 @cindex call-saved register
1939 Like @code{FIXED_REGISTERS} but has 1 for each register that is
1940 clobbered (in general) by function calls as well as for fixed
1941 registers. This macro therefore identifies the registers that are not
1942 available for general allocation of values that must live across
1945 If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1946 automatically saves it on function entry and restores it on function
1947 exit, if the register is used within the function.
1950 @defmac CALL_REALLY_USED_REGISTERS
1951 @cindex call-used register
1952 @cindex call-clobbered register
1953 @cindex call-saved register
1954 Like @code{CALL_USED_REGISTERS} except this macro doesn't require
1955 that the entire set of @code{FIXED_REGISTERS} be included.
1956 (@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}).
1957 This macro is optional. If not specified, it defaults to the value
1958 of @code{CALL_USED_REGISTERS}.
1961 @defmac HARD_REGNO_CALL_PART_CLOBBERED (@var{regno}, @var{mode})
1962 @cindex call-used register
1963 @cindex call-clobbered register
1964 @cindex call-saved register
1965 A C expression that is nonzero if it is not permissible to store a
1966 value of mode @var{mode} in hard register number @var{regno} across a
1967 call without some part of it being clobbered. For most machines this
1968 macro need not be defined. It is only required for machines that do not
1969 preserve the entire contents of a register across a call.
1973 @findex call_used_regs
1976 @findex reg_class_contents
1977 @defmac CONDITIONAL_REGISTER_USAGE
1978 Zero or more C statements that may conditionally modify five variables
1979 @code{fixed_regs}, @code{call_used_regs}, @code{global_regs},
1980 @code{reg_names}, and @code{reg_class_contents}, to take into account
1981 any dependence of these register sets on target flags. The first three
1982 of these are of type @code{char []} (interpreted as Boolean vectors).
1983 @code{global_regs} is a @code{const char *[]}, and
1984 @code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is
1985 called, @code{fixed_regs}, @code{call_used_regs},
1986 @code{reg_class_contents}, and @code{reg_names} have been initialized
1987 from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS},
1988 @code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively.
1989 @code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}},
1990 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}
1991 command options have been applied.
1993 You need not define this macro if it has no work to do.
1995 @cindex disabling certain registers
1996 @cindex controlling register usage
1997 If the usage of an entire class of registers depends on the target
1998 flags, you may indicate this to GCC by using this macro to modify
1999 @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
2000 registers in the classes which should not be used by GCC@. Also define
2001 the macro @code{REG_CLASS_FROM_LETTER} / @code{REG_CLASS_FROM_CONSTRAINT}
2002 to return @code{NO_REGS} if it
2003 is called with a letter for a class that shouldn't be used.
2005 (However, if this class is not included in @code{GENERAL_REGS} and all
2006 of the insn patterns whose constraints permit this class are
2007 controlled by target switches, then GCC will automatically avoid using
2008 these registers when the target switches are opposed to them.)
2011 @defmac INCOMING_REGNO (@var{out})
2012 Define this macro if the target machine has register windows. This C
2013 expression returns the register number as seen by the called function
2014 corresponding to the register number @var{out} as seen by the calling
2015 function. Return @var{out} if register number @var{out} is not an
2019 @defmac OUTGOING_REGNO (@var{in})
2020 Define this macro if the target machine has register windows. This C
2021 expression returns the register number as seen by the calling function
2022 corresponding to the register number @var{in} as seen by the called
2023 function. Return @var{in} if register number @var{in} is not an inbound
2027 @defmac LOCAL_REGNO (@var{regno})
2028 Define this macro if the target machine has register windows. This C
2029 expression returns true if the register is call-saved but is in the
2030 register window. Unlike most call-saved registers, such registers
2031 need not be explicitly restored on function exit or during non-local
2036 If the program counter has a register number, define this as that
2037 register number. Otherwise, do not define it.
2040 @node Allocation Order
2041 @subsection Order of Allocation of Registers
2042 @cindex order of register allocation
2043 @cindex register allocation order
2045 @c prevent bad page break with this line
2046 Registers are allocated in order.
2048 @defmac REG_ALLOC_ORDER
2049 If defined, an initializer for a vector of integers, containing the
2050 numbers of hard registers in the order in which GCC should prefer
2051 to use them (from most preferred to least).
2053 If this macro is not defined, registers are used lowest numbered first
2054 (all else being equal).
2056 One use of this macro is on machines where the highest numbered
2057 registers must always be saved and the save-multiple-registers
2058 instruction supports only sequences of consecutive registers. On such
2059 machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists
2060 the highest numbered allocable register first.
2063 @defmac ADJUST_REG_ALLOC_ORDER
2064 A C statement (sans semicolon) to choose the order in which to allocate
2065 hard registers for pseudo-registers local to a basic block.
2067 Store the desired register order in the array @code{reg_alloc_order}.
2068 Element 0 should be the register to allocate first; element 1, the next
2069 register; and so on.
2071 The macro body should not assume anything about the contents of
2072 @code{reg_alloc_order} before execution of the macro.
2074 On most machines, it is not necessary to define this macro.
2077 @defmac HONOR_REG_ALLOC_ORDER
2078 Normally, IRA tries to estimate the costs for saving a register in the
2079 prologue and restoring it in the epilogue. This discourages it from
2080 using call-saved registers. If a machine wants to ensure that IRA
2081 allocates registers in the order given by REG_ALLOC_ORDER even if some
2082 call-saved registers appear earlier than call-used ones, this macro
2086 @defmac IRA_HARD_REGNO_ADD_COST_MULTIPLIER (@var{regno})
2087 In some case register allocation order is not enough for the
2088 Integrated Register Allocator (@acronym{IRA}) to generate a good code.
2089 If this macro is defined, it should return a floating point value
2090 based on @var{regno}. The cost of using @var{regno} for a pseudo will
2091 be increased by approximately the pseudo's usage frequency times the
2092 value returned by this macro. Not defining this macro is equivalent
2093 to having it always return @code{0.0}.
2095 On most machines, it is not necessary to define this macro.
2098 @node Values in Registers
2099 @subsection How Values Fit in Registers
2101 This section discusses the macros that describe which kinds of values
2102 (specifically, which machine modes) each register can hold, and how many
2103 consecutive registers are needed for a given mode.
2105 @defmac HARD_REGNO_NREGS (@var{regno}, @var{mode})
2106 A C expression for the number of consecutive hard registers, starting
2107 at register number @var{regno}, required to hold a value of mode
2108 @var{mode}. This macro must never return zero, even if a register
2109 cannot hold the requested mode - indicate that with HARD_REGNO_MODE_OK
2110 and/or CANNOT_CHANGE_MODE_CLASS instead.
2112 On a machine where all registers are exactly one word, a suitable
2113 definition of this macro is
2116 #define HARD_REGNO_NREGS(REGNO, MODE) \
2117 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
2122 @defmac HARD_REGNO_NREGS_HAS_PADDING (@var{regno}, @var{mode})
2123 A C expression that is nonzero if a value of mode @var{mode}, stored
2124 in memory, ends with padding that causes it to take up more space than
2125 in registers starting at register number @var{regno} (as determined by
2126 multiplying GCC's notion of the size of the register when containing
2127 this mode by the number of registers returned by
2128 @code{HARD_REGNO_NREGS}). By default this is zero.
2130 For example, if a floating-point value is stored in three 32-bit
2131 registers but takes up 128 bits in memory, then this would be
2134 This macros only needs to be defined if there are cases where
2135 @code{subreg_get_info}
2136 would otherwise wrongly determine that a @code{subreg} can be
2137 represented by an offset to the register number, when in fact such a
2138 @code{subreg} would contain some of the padding not stored in
2139 registers and so not be representable.
2142 @defmac HARD_REGNO_NREGS_WITH_PADDING (@var{regno}, @var{mode})
2143 For values of @var{regno} and @var{mode} for which
2144 @code{HARD_REGNO_NREGS_HAS_PADDING} returns nonzero, a C expression
2145 returning the greater number of registers required to hold the value
2146 including any padding. In the example above, the value would be four.
2149 @defmac REGMODE_NATURAL_SIZE (@var{mode})
2150 Define this macro if the natural size of registers that hold values
2151 of mode @var{mode} is not the word size. It is a C expression that
2152 should give the natural size in bytes for the specified mode. It is
2153 used by the register allocator to try to optimize its results. This
2154 happens for example on SPARC 64-bit where the natural size of
2155 floating-point registers is still 32-bit.
2158 @defmac HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
2159 A C expression that is nonzero if it is permissible to store a value
2160 of mode @var{mode} in hard register number @var{regno} (or in several
2161 registers starting with that one). For a machine where all registers
2162 are equivalent, a suitable definition is
2165 #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
2168 You need not include code to check for the numbers of fixed registers,
2169 because the allocation mechanism considers them to be always occupied.
2171 @cindex register pairs
2172 On some machines, double-precision values must be kept in even/odd
2173 register pairs. You can implement that by defining this macro to reject
2174 odd register numbers for such modes.
2176 The minimum requirement for a mode to be OK in a register is that the
2177 @samp{mov@var{mode}} instruction pattern support moves between the
2178 register and other hard register in the same class and that moving a
2179 value into the register and back out not alter it.
2181 Since the same instruction used to move @code{word_mode} will work for
2182 all narrower integer modes, it is not necessary on any machine for
2183 @code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided
2184 you define patterns @samp{movhi}, etc., to take advantage of this. This
2185 is useful because of the interaction between @code{HARD_REGNO_MODE_OK}
2186 and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes
2189 Many machines have special registers for floating point arithmetic.
2190 Often people assume that floating point machine modes are allowed only
2191 in floating point registers. This is not true. Any registers that
2192 can hold integers can safely @emph{hold} a floating point machine
2193 mode, whether or not floating arithmetic can be done on it in those
2194 registers. Integer move instructions can be used to move the values.
2196 On some machines, though, the converse is true: fixed-point machine
2197 modes may not go in floating registers. This is true if the floating
2198 registers normalize any value stored in them, because storing a
2199 non-floating value there would garble it. In this case,
2200 @code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
2201 floating registers. But if the floating registers do not automatically
2202 normalize, if you can store any bit pattern in one and retrieve it
2203 unchanged without a trap, then any machine mode may go in a floating
2204 register, so you can define this macro to say so.
2206 The primary significance of special floating registers is rather that
2207 they are the registers acceptable in floating point arithmetic
2208 instructions. However, this is of no concern to
2209 @code{HARD_REGNO_MODE_OK}. You handle it by writing the proper
2210 constraints for those instructions.
2212 On some machines, the floating registers are especially slow to access,
2213 so that it is better to store a value in a stack frame than in such a
2214 register if floating point arithmetic is not being done. As long as the
2215 floating registers are not in class @code{GENERAL_REGS}, they will not
2216 be used unless some pattern's constraint asks for one.
2219 @defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to})
2220 A C expression that is nonzero if it is OK to rename a hard register
2221 @var{from} to another hard register @var{to}.
2223 One common use of this macro is to prevent renaming of a register to
2224 another register that is not saved by a prologue in an interrupt
2227 The default is always nonzero.
2230 @defmac MODES_TIEABLE_P (@var{mode1}, @var{mode2})
2231 A C expression that is nonzero if a value of mode
2232 @var{mode1} is accessible in mode @var{mode2} without copying.
2234 If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
2235 @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for
2236 any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})}
2237 should be nonzero. If they differ for any @var{r}, you should define
2238 this macro to return zero unless some other mechanism ensures the
2239 accessibility of the value in a narrower mode.
2241 You should define this macro to return nonzero in as many cases as
2242 possible since doing so will allow GCC to perform better register
2246 @hook TARGET_HARD_REGNO_SCRATCH_OK
2247 This target hook should return @code{true} if it is OK to use a hard register
2248 @var{regno} as scratch reg in peephole2.
2250 One common use of this macro is to prevent using of a register that
2251 is not saved by a prologue in an interrupt handler.
2253 The default version of this hook always returns @code{true}.
2256 @defmac AVOID_CCMODE_COPIES
2257 Define this macro if the compiler should avoid copies to/from @code{CCmode}
2258 registers. You should only define this macro if support for copying to/from
2259 @code{CCmode} is incomplete.
2262 @node Leaf Functions
2263 @subsection Handling Leaf Functions
2265 @cindex leaf functions
2266 @cindex functions, leaf
2267 On some machines, a leaf function (i.e., one which makes no calls) can run
2268 more efficiently if it does not make its own register window. Often this
2269 means it is required to receive its arguments in the registers where they
2270 are passed by the caller, instead of the registers where they would
2273 The special treatment for leaf functions generally applies only when
2274 other conditions are met; for example, often they may use only those
2275 registers for its own variables and temporaries. We use the term ``leaf
2276 function'' to mean a function that is suitable for this special
2277 handling, so that functions with no calls are not necessarily ``leaf
2280 GCC assigns register numbers before it knows whether the function is
2281 suitable for leaf function treatment. So it needs to renumber the
2282 registers in order to output a leaf function. The following macros
2285 @defmac LEAF_REGISTERS
2286 Name of a char vector, indexed by hard register number, which
2287 contains 1 for a register that is allowable in a candidate for leaf
2290 If leaf function treatment involves renumbering the registers, then the
2291 registers marked here should be the ones before renumbering---those that
2292 GCC would ordinarily allocate. The registers which will actually be
2293 used in the assembler code, after renumbering, should not be marked with 1
2296 Define this macro only if the target machine offers a way to optimize
2297 the treatment of leaf functions.
2300 @defmac LEAF_REG_REMAP (@var{regno})
2301 A C expression whose value is the register number to which @var{regno}
2302 should be renumbered, when a function is treated as a leaf function.
2304 If @var{regno} is a register number which should not appear in a leaf
2305 function before renumbering, then the expression should yield @minus{}1, which
2306 will cause the compiler to abort.
2308 Define this macro only if the target machine offers a way to optimize the
2309 treatment of leaf functions, and registers need to be renumbered to do
2313 @findex current_function_is_leaf
2314 @findex current_function_uses_only_leaf_regs
2315 @code{TARGET_ASM_FUNCTION_PROLOGUE} and
2316 @code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions
2317 specially. They can test the C variable @code{current_function_is_leaf}
2318 which is nonzero for leaf functions. @code{current_function_is_leaf} is
2319 set prior to local register allocation and is valid for the remaining
2320 compiler passes. They can also test the C variable
2321 @code{current_function_uses_only_leaf_regs} which is nonzero for leaf
2322 functions which only use leaf registers.
2323 @code{current_function_uses_only_leaf_regs} is valid after all passes
2324 that modify the instructions have been run and is only useful if
2325 @code{LEAF_REGISTERS} is defined.
2326 @c changed this to fix overfull. ALSO: why the "it" at the beginning
2327 @c of the next paragraph?! --mew 2feb93
2329 @node Stack Registers
2330 @subsection Registers That Form a Stack
2332 There are special features to handle computers where some of the
2333 ``registers'' form a stack. Stack registers are normally written by
2334 pushing onto the stack, and are numbered relative to the top of the
2337 Currently, GCC can only handle one group of stack-like registers, and
2338 they must be consecutively numbered. Furthermore, the existing
2339 support for stack-like registers is specific to the 80387 floating
2340 point coprocessor. If you have a new architecture that uses
2341 stack-like registers, you will need to do substantial work on
2342 @file{reg-stack.c} and write your machine description to cooperate
2343 with it, as well as defining these macros.
2346 Define this if the machine has any stack-like registers.
2349 @defmac STACK_REG_COVER_CLASS
2350 This is a cover class containing the stack registers. Define this if
2351 the machine has any stack-like registers.
2354 @defmac FIRST_STACK_REG
2355 The number of the first stack-like register. This one is the top
2359 @defmac LAST_STACK_REG
2360 The number of the last stack-like register. This one is the bottom of
2364 @node Register Classes
2365 @section Register Classes
2366 @cindex register class definitions
2367 @cindex class definitions, register
2369 On many machines, the numbered registers are not all equivalent.
2370 For example, certain registers may not be allowed for indexed addressing;
2371 certain registers may not be allowed in some instructions. These machine
2372 restrictions are described to the compiler using @dfn{register classes}.
2374 You define a number of register classes, giving each one a name and saying
2375 which of the registers belong to it. Then you can specify register classes
2376 that are allowed as operands to particular instruction patterns.
2380 In general, each register will belong to several classes. In fact, one
2381 class must be named @code{ALL_REGS} and contain all the registers. Another
2382 class must be named @code{NO_REGS} and contain no registers. Often the
2383 union of two classes will be another class; however, this is not required.
2385 @findex GENERAL_REGS
2386 One of the classes must be named @code{GENERAL_REGS}. There is nothing
2387 terribly special about the name, but the operand constraint letters
2388 @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
2389 the same as @code{ALL_REGS}, just define it as a macro which expands
2392 Order the classes so that if class @var{x} is contained in class @var{y}
2393 then @var{x} has a lower class number than @var{y}.
2395 The way classes other than @code{GENERAL_REGS} are specified in operand
2396 constraints is through machine-dependent operand constraint letters.
2397 You can define such letters to correspond to various classes, then use
2398 them in operand constraints.
2400 You should define a class for the union of two classes whenever some
2401 instruction allows both classes. For example, if an instruction allows
2402 either a floating point (coprocessor) register or a general register for a
2403 certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
2404 which includes both of them. Otherwise you will get suboptimal code.
2406 You must also specify certain redundant information about the register
2407 classes: for each class, which classes contain it and which ones are
2408 contained in it; for each pair of classes, the largest class contained
2411 When a value occupying several consecutive registers is expected in a
2412 certain class, all the registers used must belong to that class.
2413 Therefore, register classes cannot be used to enforce a requirement for
2414 a register pair to start with an even-numbered register. The way to
2415 specify this requirement is with @code{HARD_REGNO_MODE_OK}.
2417 Register classes used for input-operands of bitwise-and or shift
2418 instructions have a special requirement: each such class must have, for
2419 each fixed-point machine mode, a subclass whose registers can transfer that
2420 mode to or from memory. For example, on some machines, the operations for
2421 single-byte values (@code{QImode}) are limited to certain registers. When
2422 this is so, each register class that is used in a bitwise-and or shift
2423 instruction must have a subclass consisting of registers from which
2424 single-byte values can be loaded or stored. This is so that
2425 @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
2427 @deftp {Data type} {enum reg_class}
2428 An enumerated type that must be defined with all the register class names
2429 as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS}
2430 must be the last register class, followed by one more enumerated value,
2431 @code{LIM_REG_CLASSES}, which is not a register class but rather
2432 tells how many classes there are.
2434 Each register class has a number, which is the value of casting
2435 the class name to type @code{int}. The number serves as an index
2436 in many of the tables described below.
2439 @defmac N_REG_CLASSES
2440 The number of distinct register classes, defined as follows:
2443 #define N_REG_CLASSES (int) LIM_REG_CLASSES
2447 @defmac REG_CLASS_NAMES
2448 An initializer containing the names of the register classes as C string
2449 constants. These names are used in writing some of the debugging dumps.
2452 @defmac REG_CLASS_CONTENTS
2453 An initializer containing the contents of the register classes, as integers
2454 which are bit masks. The @var{n}th integer specifies the contents of class
2455 @var{n}. The way the integer @var{mask} is interpreted is that
2456 register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
2458 When the machine has more than 32 registers, an integer does not suffice.
2459 Then the integers are replaced by sub-initializers, braced groupings containing
2460 several integers. Each sub-initializer must be suitable as an initializer
2461 for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
2462 In this situation, the first integer in each sub-initializer corresponds to
2463 registers 0 through 31, the second integer to registers 32 through 63, and
2467 @defmac REGNO_REG_CLASS (@var{regno})
2468 A C expression whose value is a register class containing hard register
2469 @var{regno}. In general there is more than one such class; choose a class
2470 which is @dfn{minimal}, meaning that no smaller class also contains the
2474 @defmac BASE_REG_CLASS
2475 A macro whose definition is the name of the class to which a valid
2476 base register must belong. A base register is one used in an address
2477 which is the register value plus a displacement.
2480 @defmac MODE_BASE_REG_CLASS (@var{mode})
2481 This is a variation of the @code{BASE_REG_CLASS} macro which allows
2482 the selection of a base register in a mode dependent manner. If
2483 @var{mode} is VOIDmode then it should return the same value as
2484 @code{BASE_REG_CLASS}.
2487 @defmac MODE_BASE_REG_REG_CLASS (@var{mode})
2488 A C expression whose value is the register class to which a valid
2489 base register must belong in order to be used in a base plus index
2490 register address. You should define this macro if base plus index
2491 addresses have different requirements than other base register uses.
2494 @defmac MODE_CODE_BASE_REG_CLASS (@var{mode}, @var{outer_code}, @var{index_code})
2495 A C expression whose value is the register class to which a valid
2496 base register must belong. @var{outer_code} and @var{index_code} define the
2497 context in which the base register occurs. @var{outer_code} is the code of
2498 the immediately enclosing expression (@code{MEM} for the top level of an
2499 address, @code{ADDRESS} for something that occurs in an
2500 @code{address_operand}). @var{index_code} is the code of the corresponding
2501 index expression if @var{outer_code} is @code{PLUS}; @code{SCRATCH} otherwise.
2504 @defmac INDEX_REG_CLASS
2505 A macro whose definition is the name of the class to which a valid
2506 index register must belong. An index register is one used in an
2507 address where its value is either multiplied by a scale factor or
2508 added to another register (as well as added to a displacement).
2511 @defmac REGNO_OK_FOR_BASE_P (@var{num})
2512 A C expression which is nonzero if register number @var{num} is
2513 suitable for use as a base register in operand addresses.
2514 Like @code{TARGET_LEGITIMATE_ADDRESS_P}, this macro should also
2515 define a strict and a non-strict variant. Both variants behave
2516 the same for hard register; for pseudos, the strict variant will
2517 pass only those that have been allocated to a valid hard registers,
2518 while the non-strict variant will pass all pseudos.
2520 @findex REG_OK_STRICT
2521 Compiler source files that want to use the strict variant of this and
2522 other macros define the macro @code{REG_OK_STRICT}. You should use an
2523 @code{#ifdef REG_OK_STRICT} conditional to define the strict variant in
2524 that case and the non-strict variant otherwise.
2527 @defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode})
2528 A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that
2529 that expression may examine the mode of the memory reference in
2530 @var{mode}. You should define this macro if the mode of the memory
2531 reference affects whether a register may be used as a base register. If
2532 you define this macro, the compiler will use it instead of
2533 @code{REGNO_OK_FOR_BASE_P}. The mode may be @code{VOIDmode} for
2534 addresses that appear outside a @code{MEM}, i.e., as an
2535 @code{address_operand}.
2537 This macro also has strict and non-strict variants.
2540 @defmac REGNO_MODE_OK_FOR_REG_BASE_P (@var{num}, @var{mode})
2541 A C expression which is nonzero if register number @var{num} is suitable for
2542 use as a base register in base plus index operand addresses, accessing
2543 memory in mode @var{mode}. It may be either a suitable hard register or a
2544 pseudo register that has been allocated such a hard register. You should
2545 define this macro if base plus index addresses have different requirements
2546 than other base register uses.
2548 Use of this macro is deprecated; please use the more general
2549 @code{REGNO_MODE_CODE_OK_FOR_BASE_P}.
2551 This macro also has strict and non-strict variants.
2554 @defmac REGNO_MODE_CODE_OK_FOR_BASE_P (@var{num}, @var{mode}, @var{outer_code}, @var{index_code})
2555 A C expression that is just like @code{REGNO_MODE_OK_FOR_BASE_P}, except
2556 that that expression may examine the context in which the register
2557 appears in the memory reference. @var{outer_code} is the code of the
2558 immediately enclosing expression (@code{MEM} if at the top level of the
2559 address, @code{ADDRESS} for something that occurs in an
2560 @code{address_operand}). @var{index_code} is the code of the
2561 corresponding index expression if @var{outer_code} is @code{PLUS};
2562 @code{SCRATCH} otherwise. The mode may be @code{VOIDmode} for addresses
2563 that appear outside a @code{MEM}, i.e., as an @code{address_operand}.
2565 This macro also has strict and non-strict variants.
2568 @defmac REGNO_OK_FOR_INDEX_P (@var{num})
2569 A C expression which is nonzero if register number @var{num} is
2570 suitable for use as an index register in operand addresses. It may be
2571 either a suitable hard register or a pseudo register that has been
2572 allocated such a hard register.
2574 The difference between an index register and a base register is that
2575 the index register may be scaled. If an address involves the sum of
2576 two registers, neither one of them scaled, then either one may be
2577 labeled the ``base'' and the other the ``index''; but whichever
2578 labeling is used must fit the machine's constraints of which registers
2579 may serve in each capacity. The compiler will try both labelings,
2580 looking for one that is valid, and will reload one or both registers
2581 only if neither labeling works.
2583 This macro also has strict and non-strict variants.
2586 @hook TARGET_PREFERRED_RELOAD_CLASS
2587 A target hook that places additional restrictions on the register class
2588 to use when it is necessary to copy value @var{x} into a register in class
2589 @var{rclass}. The value is a register class; perhaps @var{rclass}, or perhaps
2590 another, smaller class.
2592 The default version of this hook always returns value of @code{rclass} argument.
2594 Sometimes returning a more restrictive class makes better code. For
2595 example, on the 68000, when @var{x} is an integer constant that is in range
2596 for a @samp{moveq} instruction, the value of this macro is always
2597 @code{DATA_REGS} as long as @var{rclass} includes the data registers.
2598 Requiring a data register guarantees that a @samp{moveq} will be used.
2600 One case where @code{TARGET_PREFERRED_RELOAD_CLASS} must not return
2601 @var{rclass} is if @var{x} is a legitimate constant which cannot be
2602 loaded into some register class. By returning @code{NO_REGS} you can
2603 force @var{x} into a memory location. For example, rs6000 can load
2604 immediate values into general-purpose registers, but does not have an
2605 instruction for loading an immediate value into a floating-point
2606 register, so @code{TARGET_PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2607 @var{x} is a floating-point constant. If the constant can't be loaded
2608 into any kind of register, code generation will be better if
2609 @code{LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2610 of using @code{TARGET_PREFERRED_RELOAD_CLASS}.
2612 If an insn has pseudos in it after register allocation, reload will go
2613 through the alternatives and call repeatedly @code{TARGET_PREFERRED_RELOAD_CLASS}
2614 to find the best one. Returning @code{NO_REGS}, in this case, makes
2615 reload add a @code{!} in front of the constraint: the x86 back-end uses
2616 this feature to discourage usage of 387 registers when math is done in
2617 the SSE registers (and vice versa).
2620 @defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
2621 A C expression that places additional restrictions on the register class
2622 to use when it is necessary to copy value @var{x} into a register in class
2623 @var{class}. The value is a register class; perhaps @var{class}, or perhaps
2624 another, smaller class. On many machines, the following definition is
2628 #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
2631 Sometimes returning a more restrictive class makes better code. For
2632 example, on the 68000, when @var{x} is an integer constant that is in range
2633 for a @samp{moveq} instruction, the value of this macro is always
2634 @code{DATA_REGS} as long as @var{class} includes the data registers.
2635 Requiring a data register guarantees that a @samp{moveq} will be used.
2637 One case where @code{PREFERRED_RELOAD_CLASS} must not return
2638 @var{class} is if @var{x} is a legitimate constant which cannot be
2639 loaded into some register class. By returning @code{NO_REGS} you can
2640 force @var{x} into a memory location. For example, rs6000 can load
2641 immediate values into general-purpose registers, but does not have an
2642 instruction for loading an immediate value into a floating-point
2643 register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2644 @var{x} is a floating-point constant. If the constant can't be loaded
2645 into any kind of register, code generation will be better if
2646 @code{LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2647 of using @code{PREFERRED_RELOAD_CLASS}.
2649 If an insn has pseudos in it after register allocation, reload will go
2650 through the alternatives and call repeatedly @code{PREFERRED_RELOAD_CLASS}
2651 to find the best one. Returning @code{NO_REGS}, in this case, makes
2652 reload add a @code{!} in front of the constraint: the x86 back-end uses
2653 this feature to discourage usage of 387 registers when math is done in
2654 the SSE registers (and vice versa).
2657 @defmac PREFERRED_OUTPUT_RELOAD_CLASS (@var{x}, @var{class})
2658 Like @code{PREFERRED_RELOAD_CLASS}, but for output reloads instead of
2659 input reloads. If you don't define this macro, the default is to use
2660 @var{class}, unchanged.
2662 You can also use @code{PREFERRED_OUTPUT_RELOAD_CLASS} to discourage
2663 reload from using some alternatives, like @code{PREFERRED_RELOAD_CLASS}.
2666 @defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class})
2667 A C expression that places additional restrictions on the register class
2668 to use when it is necessary to be able to hold a value of mode
2669 @var{mode} in a reload register for which class @var{class} would
2672 Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when
2673 there are certain modes that simply can't go in certain reload classes.
2675 The value is a register class; perhaps @var{class}, or perhaps another,
2678 Don't define this macro unless the target machine has limitations which
2679 require the macro to do something nontrivial.
2682 @hook TARGET_SECONDARY_RELOAD
2683 Many machines have some registers that cannot be copied directly to or
2684 from memory or even from other types of registers. An example is the
2685 @samp{MQ} register, which on most machines, can only be copied to or
2686 from general registers, but not memory. Below, we shall be using the
2687 term 'intermediate register' when a move operation cannot be performed
2688 directly, but has to be done by copying the source into the intermediate
2689 register first, and then copying the intermediate register to the
2690 destination. An intermediate register always has the same mode as
2691 source and destination. Since it holds the actual value being copied,
2692 reload might apply optimizations to re-use an intermediate register
2693 and eliding the copy from the source when it can determine that the
2694 intermediate register still holds the required value.
2696 Another kind of secondary reload is required on some machines which
2697 allow copying all registers to and from memory, but require a scratch
2698 register for stores to some memory locations (e.g., those with symbolic
2699 address on the RT, and those with certain symbolic address on the SPARC
2700 when compiling PIC)@. Scratch registers need not have the same mode
2701 as the value being copied, and usually hold a different value than
2702 that being copied. Special patterns in the md file are needed to
2703 describe how the copy is performed with the help of the scratch register;
2704 these patterns also describe the number, register class(es) and mode(s)
2705 of the scratch register(s).
2707 In some cases, both an intermediate and a scratch register are required.
2709 For input reloads, this target hook is called with nonzero @var{in_p},
2710 and @var{x} is an rtx that needs to be copied to a register of class
2711 @var{reload_class} in @var{reload_mode}. For output reloads, this target
2712 hook is called with zero @var{in_p}, and a register of class @var{reload_class}
2713 needs to be copied to rtx @var{x} in @var{reload_mode}.
2715 If copying a register of @var{reload_class} from/to @var{x} requires
2716 an intermediate register, the hook @code{secondary_reload} should
2717 return the register class required for this intermediate register.
2718 If no intermediate register is required, it should return NO_REGS.
2719 If more than one intermediate register is required, describe the one
2720 that is closest in the copy chain to the reload register.
2722 If scratch registers are needed, you also have to describe how to
2723 perform the copy from/to the reload register to/from this
2724 closest intermediate register. Or if no intermediate register is
2725 required, but still a scratch register is needed, describe the
2726 copy from/to the reload register to/from the reload operand @var{x}.
2728 You do this by setting @code{sri->icode} to the instruction code of a pattern
2729 in the md file which performs the move. Operands 0 and 1 are the output
2730 and input of this copy, respectively. Operands from operand 2 onward are
2731 for scratch operands. These scratch operands must have a mode, and a
2732 single-register-class
2733 @c [later: or memory]
2736 When an intermediate register is used, the @code{secondary_reload}
2737 hook will be called again to determine how to copy the intermediate
2738 register to/from the reload operand @var{x}, so your hook must also
2739 have code to handle the register class of the intermediate operand.
2741 @c [For later: maybe we'll allow multi-alternative reload patterns -
2742 @c the port maintainer could name a mov<mode> pattern that has clobbers -
2743 @c and match the constraints of input and output to determine the required
2744 @c alternative. A restriction would be that constraints used to match
2745 @c against reloads registers would have to be written as register class
2746 @c constraints, or we need a new target macro / hook that tells us if an
2747 @c arbitrary constraint can match an unknown register of a given class.
2748 @c Such a macro / hook would also be useful in other places.]
2751 @var{x} might be a pseudo-register or a @code{subreg} of a
2752 pseudo-register, which could either be in a hard register or in memory.
2753 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2754 in memory and the hard register number if it is in a register.
2756 Scratch operands in memory (constraint @code{"=m"} / @code{"=&m"}) are
2757 currently not supported. For the time being, you will have to continue
2758 to use @code{SECONDARY_MEMORY_NEEDED} for that purpose.
2760 @code{copy_cost} also uses this target hook to find out how values are
2761 copied. If you want it to include some extra cost for the need to allocate
2762 (a) scratch register(s), set @code{sri->extra_cost} to the additional cost.
2763 Or if two dependent moves are supposed to have a lower cost than the sum
2764 of the individual moves due to expected fortuitous scheduling and/or special
2765 forwarding logic, you can set @code{sri->extra_cost} to a negative amount.
2768 @defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2769 @defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2770 @defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2771 These macros are obsolete, new ports should use the target hook
2772 @code{TARGET_SECONDARY_RELOAD} instead.
2774 These are obsolete macros, replaced by the @code{TARGET_SECONDARY_RELOAD}
2775 target hook. Older ports still define these macros to indicate to the
2776 reload phase that it may
2777 need to allocate at least one register for a reload in addition to the
2778 register to contain the data. Specifically, if copying @var{x} to a
2779 register @var{class} in @var{mode} requires an intermediate register,
2780 you were supposed to define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the
2781 largest register class all of whose registers can be used as
2782 intermediate registers or scratch registers.
2784 If copying a register @var{class} in @var{mode} to @var{x} requires an
2785 intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS}
2786 was supposed to be defined be defined to return the largest register
2787 class required. If the
2788 requirements for input and output reloads were the same, the macro
2789 @code{SECONDARY_RELOAD_CLASS} should have been used instead of defining both
2792 The values returned by these macros are often @code{GENERAL_REGS}.
2793 Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x}
2794 can be directly copied to or from a register of @var{class} in
2795 @var{mode} without requiring a scratch register. Do not define this
2796 macro if it would always return @code{NO_REGS}.
2798 If a scratch register is required (either with or without an
2799 intermediate register), you were supposed to define patterns for
2800 @samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required
2801 (@pxref{Standard Names}. These patterns, which were normally
2802 implemented with a @code{define_expand}, should be similar to the
2803 @samp{mov@var{m}} patterns, except that operand 2 is the scratch
2806 These patterns need constraints for the reload register and scratch
2808 contain a single register class. If the original reload register (whose
2809 class is @var{class}) can meet the constraint given in the pattern, the
2810 value returned by these macros is used for the class of the scratch
2811 register. Otherwise, two additional reload registers are required.
2812 Their classes are obtained from the constraints in the insn pattern.
2814 @var{x} might be a pseudo-register or a @code{subreg} of a
2815 pseudo-register, which could either be in a hard register or in memory.
2816 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2817 in memory and the hard register number if it is in a register.
2819 These macros should not be used in the case where a particular class of
2820 registers can only be copied to memory and not to another class of
2821 registers. In that case, secondary reload registers are not needed and
2822 would not be helpful. Instead, a stack location must be used to perform
2823 the copy and the @code{mov@var{m}} pattern should use memory as an
2824 intermediate storage. This case often occurs between floating-point and
2828 @defmac SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m})
2829 Certain machines have the property that some registers cannot be copied
2830 to some other registers without using memory. Define this macro on
2831 those machines to be a C expression that is nonzero if objects of mode
2832 @var{m} in registers of @var{class1} can only be copied to registers of
2833 class @var{class2} by storing a register of @var{class1} into memory
2834 and loading that memory location into a register of @var{class2}.
2836 Do not define this macro if its value would always be zero.
2839 @defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode})
2840 Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler
2841 allocates a stack slot for a memory location needed for register copies.
2842 If this macro is defined, the compiler instead uses the memory location
2843 defined by this macro.
2845 Do not define this macro if you do not define
2846 @code{SECONDARY_MEMORY_NEEDED}.
2849 @defmac SECONDARY_MEMORY_NEEDED_MODE (@var{mode})
2850 When the compiler needs a secondary memory location to copy between two
2851 registers of mode @var{mode}, it normally allocates sufficient memory to
2852 hold a quantity of @code{BITS_PER_WORD} bits and performs the store and
2853 load operations in a mode that many bits wide and whose class is the
2854 same as that of @var{mode}.
2856 This is right thing to do on most machines because it ensures that all
2857 bits of the register are copied and prevents accesses to the registers
2858 in a narrower mode, which some machines prohibit for floating-point
2861 However, this default behavior is not correct on some machines, such as
2862 the DEC Alpha, that store short integers in floating-point registers
2863 differently than in integer registers. On those machines, the default
2864 widening will not work correctly and you must define this macro to
2865 suppress that widening in some cases. See the file @file{alpha.h} for
2868 Do not define this macro if you do not define
2869 @code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that
2870 is @code{BITS_PER_WORD} bits wide is correct for your machine.
2873 @hook TARGET_CLASS_LIKELY_SPILLED_P
2874 A target hook which returns @code{true} if pseudos that have been assigned
2875 to registers of class @var{rclass} would likely be spilled because
2876 registers of @var{rclass} are needed for spill registers.
2878 The default version of this target hook returns @code{true} if @var{rclass}
2879 has exactly one register and @code{false} otherwise. On most machines, this
2880 default should be used. Only use this target hook to some other expression
2881 if pseudos allocated by @file{local-alloc.c} end up in memory because their
2882 hard registers were needed for spill registers. If this target hook returns
2883 @code{false} for those classes, those pseudos will only be allocated by
2884 @file{global.c}, which knows how to reallocate the pseudo to another
2885 register. If there would not be another register available for reallocation,
2886 you should not change the implementation of this target hook since
2887 the only effect of such implementation would be to slow down register
2891 @defmac CLASS_MAX_NREGS (@var{class}, @var{mode})
2892 A C expression for the maximum number of consecutive registers
2893 of class @var{class} needed to hold a value of mode @var{mode}.
2895 This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact,
2896 the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
2897 should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno},
2898 @var{mode})} for all @var{regno} values in the class @var{class}.
2900 This macro helps control the handling of multiple-word values
2904 @defmac CANNOT_CHANGE_MODE_CLASS (@var{from}, @var{to}, @var{class})
2905 If defined, a C expression that returns nonzero for a @var{class} for which
2906 a change from mode @var{from} to mode @var{to} is invalid.
2908 For the example, loading 32-bit integer or floating-point objects into
2909 floating-point registers on the Alpha extends them to 64 bits.
2910 Therefore loading a 64-bit object and then storing it as a 32-bit object
2911 does not store the low-order 32 bits, as would be the case for a normal
2912 register. Therefore, @file{alpha.h} defines @code{CANNOT_CHANGE_MODE_CLASS}
2916 #define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \
2917 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \
2918 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0)
2922 @hook TARGET_IRA_COVER_CLASSES
2923 Return an array of cover classes for the Integrated Register Allocator
2924 (@acronym{IRA}). Cover classes are a set of non-intersecting register
2925 classes covering all hard registers used for register allocation
2926 purposes. If a move between two registers in the same cover class is
2927 possible, it should be cheaper than a load or store of the registers.
2928 The array is terminated by a @code{LIM_REG_CLASSES} element.
2930 The order of cover classes in the array is important. If two classes
2931 have the same cost of usage for a pseudo, the class occurred first in
2932 the array is chosen for the pseudo.
2934 This hook is called once at compiler startup, after the command-line
2935 options have been processed. It is then re-examined by every call to
2936 @code{target_reinit}.
2938 The default implementation returns @code{IRA_COVER_CLASSES}, if defined,
2939 otherwise there is no default implementation. You must define either this
2940 macro or @code{IRA_COVER_CLASSES} in order to use the integrated register
2941 allocator with Chaitin-Briggs coloring. If the macro is not defined,
2942 the only available coloring algorithm is Chow's priority coloring.
2945 @defmac IRA_COVER_CLASSES
2946 See the documentation for @code{TARGET_IRA_COVER_CLASSES}.
2949 @node Old Constraints
2950 @section Obsolete Macros for Defining Constraints
2951 @cindex defining constraints, obsolete method
2952 @cindex constraints, defining, obsolete method
2954 Machine-specific constraints can be defined with these macros instead
2955 of the machine description constructs described in @ref{Define
2956 Constraints}. This mechanism is obsolete. New ports should not use
2957 it; old ports should convert to the new mechanism.
2959 @defmac CONSTRAINT_LEN (@var{char}, @var{str})
2960 For the constraint at the start of @var{str}, which starts with the letter
2961 @var{c}, return the length. This allows you to have register class /
2962 constant / extra constraints that are longer than a single letter;
2963 you don't need to define this macro if you can do with single-letter
2964 constraints only. The definition of this macro should use
2965 DEFAULT_CONSTRAINT_LEN for all the characters that you don't want
2966 to handle specially.
2967 There are some sanity checks in genoutput.c that check the constraint lengths
2968 for the md file, so you can also use this macro to help you while you are
2969 transitioning from a byzantine single-letter-constraint scheme: when you
2970 return a negative length for a constraint you want to re-use, genoutput
2971 will complain about every instance where it is used in the md file.
2974 @defmac REG_CLASS_FROM_LETTER (@var{char})
2975 A C expression which defines the machine-dependent operand constraint
2976 letters for register classes. If @var{char} is such a letter, the
2977 value should be the register class corresponding to it. Otherwise,
2978 the value should be @code{NO_REGS}. The register letter @samp{r},
2979 corresponding to class @code{GENERAL_REGS}, will not be passed
2980 to this macro; you do not need to handle it.
2983 @defmac REG_CLASS_FROM_CONSTRAINT (@var{char}, @var{str})
2984 Like @code{REG_CLASS_FROM_LETTER}, but you also get the constraint string
2985 passed in @var{str}, so that you can use suffixes to distinguish between
2989 @defmac CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
2990 A C expression that defines the machine-dependent operand constraint
2991 letters (@samp{I}, @samp{J}, @samp{K}, @dots{} @samp{P}) that specify
2992 particular ranges of integer values. If @var{c} is one of those
2993 letters, the expression should check that @var{value}, an integer, is in
2994 the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
2995 not one of those letters, the value should be 0 regardless of
2999 @defmac CONST_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
3000 Like @code{CONST_OK_FOR_LETTER_P}, but you also get the constraint
3001 string passed in @var{str}, so that you can use suffixes to distinguish
3002 between different variants.
3005 @defmac CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
3006 A C expression that defines the machine-dependent operand constraint
3007 letters that specify particular ranges of @code{const_double} values
3008 (@samp{G} or @samp{H}).
3010 If @var{c} is one of those letters, the expression should check that
3011 @var{value}, an RTX of code @code{const_double}, is in the appropriate
3012 range and return 1 if so, 0 otherwise. If @var{c} is not one of those
3013 letters, the value should be 0 regardless of @var{value}.
3015 @code{const_double} is used for all floating-point constants and for
3016 @code{DImode} fixed-point constants. A given letter can accept either
3017 or both kinds of values. It can use @code{GET_MODE} to distinguish
3018 between these kinds.
3021 @defmac CONST_DOUBLE_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
3022 Like @code{CONST_DOUBLE_OK_FOR_LETTER_P}, but you also get the constraint
3023 string passed in @var{str}, so that you can use suffixes to distinguish
3024 between different variants.
3027 @defmac EXTRA_CONSTRAINT (@var{value}, @var{c})
3028 A C expression that defines the optional machine-dependent constraint
3029 letters that can be used to segregate specific types of operands, usually
3030 memory references, for the target machine. Any letter that is not
3031 elsewhere defined and not matched by @code{REG_CLASS_FROM_LETTER} /
3032 @code{REG_CLASS_FROM_CONSTRAINT}
3033 may be used. Normally this macro will not be defined.
3035 If it is required for a particular target machine, it should return 1
3036 if @var{value} corresponds to the operand type represented by the
3037 constraint letter @var{c}. If @var{c} is not defined as an extra
3038 constraint, the value returned should be 0 regardless of @var{value}.
3040 For example, on the ROMP, load instructions cannot have their output
3041 in r0 if the memory reference contains a symbolic address. Constraint
3042 letter @samp{Q} is defined as representing a memory address that does
3043 @emph{not} contain a symbolic address. An alternative is specified with
3044 a @samp{Q} constraint on the input and @samp{r} on the output. The next
3045 alternative specifies @samp{m} on the input and a register class that
3046 does not include r0 on the output.
3049 @defmac EXTRA_CONSTRAINT_STR (@var{value}, @var{c}, @var{str})
3050 Like @code{EXTRA_CONSTRAINT}, but you also get the constraint string passed
3051 in @var{str}, so that you can use suffixes to distinguish between different
3055 @defmac EXTRA_MEMORY_CONSTRAINT (@var{c}, @var{str})
3056 A C expression that defines the optional machine-dependent constraint
3057 letters, amongst those accepted by @code{EXTRA_CONSTRAINT}, that should
3058 be treated like memory constraints by the reload pass.
3060 It should return 1 if the operand type represented by the constraint
3061 at the start of @var{str}, the first letter of which is the letter @var{c},
3062 comprises a subset of all memory references including
3063 all those whose address is simply a base register. This allows the reload
3064 pass to reload an operand, if it does not directly correspond to the operand
3065 type of @var{c}, by copying its address into a base register.
3067 For example, on the S/390, some instructions do not accept arbitrary
3068 memory references, but only those that do not make use of an index
3069 register. The constraint letter @samp{Q} is defined via
3070 @code{EXTRA_CONSTRAINT} as representing a memory address of this type.
3071 If the letter @samp{Q} is marked as @code{EXTRA_MEMORY_CONSTRAINT},
3072 a @samp{Q} constraint can handle any memory operand, because the
3073 reload pass knows it can be reloaded by copying the memory address
3074 into a base register if required. This is analogous to the way
3075 an @samp{o} constraint can handle any memory operand.
3078 @defmac EXTRA_ADDRESS_CONSTRAINT (@var{c}, @var{str})
3079 A C expression that defines the optional machine-dependent constraint
3080 letters, amongst those accepted by @code{EXTRA_CONSTRAINT} /
3081 @code{EXTRA_CONSTRAINT_STR}, that should
3082 be treated like address constraints by the reload pass.
3084 It should return 1 if the operand type represented by the constraint
3085 at the start of @var{str}, which starts with the letter @var{c}, comprises
3086 a subset of all memory addresses including
3087 all those that consist of just a base register. This allows the reload
3088 pass to reload an operand, if it does not directly correspond to the operand
3089 type of @var{str}, by copying it into a base register.
3091 Any constraint marked as @code{EXTRA_ADDRESS_CONSTRAINT} can only
3092 be used with the @code{address_operand} predicate. It is treated
3093 analogously to the @samp{p} constraint.
3096 @node Stack and Calling
3097 @section Stack Layout and Calling Conventions
3098 @cindex calling conventions
3100 @c prevent bad page break with this line
3101 This describes the stack layout and calling conventions.
3105 * Exception Handling::
3110 * Register Arguments::
3112 * Aggregate Return::
3117 * Stack Smashing Protection::
3121 @subsection Basic Stack Layout
3122 @cindex stack frame layout
3123 @cindex frame layout
3125 @c prevent bad page break with this line
3126 Here is the basic stack layout.
3128 @defmac STACK_GROWS_DOWNWARD
3129 Define this macro if pushing a word onto the stack moves the stack
3130 pointer to a smaller address.
3132 When we say, ``define this macro if @dots{}'', it means that the
3133 compiler checks this macro only with @code{#ifdef} so the precise
3134 definition used does not matter.
3137 @defmac STACK_PUSH_CODE
3138 This macro defines the operation used when something is pushed
3139 on the stack. In RTL, a push operation will be
3140 @code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})}
3142 The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC},
3143 and @code{POST_INC}. Which of these is correct depends on
3144 the stack direction and on whether the stack pointer points
3145 to the last item on the stack or whether it points to the
3146 space for the next item on the stack.
3148 The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is
3149 defined, which is almost always right, and @code{PRE_INC} otherwise,
3150 which is often wrong.
3153 @defmac FRAME_GROWS_DOWNWARD
3154 Define this macro to nonzero value if the addresses of local variable slots
3155 are at negative offsets from the frame pointer.
3158 @defmac ARGS_GROW_DOWNWARD
3159 Define this macro if successive arguments to a function occupy decreasing
3160 addresses on the stack.
3163 @defmac STARTING_FRAME_OFFSET
3164 Offset from the frame pointer to the first local variable slot to be allocated.
3166 If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by
3167 subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}.
3168 Otherwise, it is found by adding the length of the first slot to the
3169 value @code{STARTING_FRAME_OFFSET}.
3170 @c i'm not sure if the above is still correct.. had to change it to get
3171 @c rid of an overfull. --mew 2feb93
3174 @defmac STACK_ALIGNMENT_NEEDED
3175 Define to zero to disable final alignment of the stack during reload.
3176 The nonzero default for this macro is suitable for most ports.
3178 On ports where @code{STARTING_FRAME_OFFSET} is nonzero or where there
3179 is a register save block following the local block that doesn't require
3180 alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable
3181 stack alignment and do it in the backend.
3184 @defmac STACK_POINTER_OFFSET
3185 Offset from the stack pointer register to the first location at which
3186 outgoing arguments are placed. If not specified, the default value of
3187 zero is used. This is the proper value for most machines.
3189 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
3190 the first location at which outgoing arguments are placed.
3193 @defmac FIRST_PARM_OFFSET (@var{fundecl})
3194 Offset from the argument pointer register to the first argument's
3195 address. On some machines it may depend on the data type of the
3198 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
3199 the first argument's address.
3202 @defmac STACK_DYNAMIC_OFFSET (@var{fundecl})
3203 Offset from the stack pointer register to an item dynamically allocated
3204 on the stack, e.g., by @code{alloca}.
3206 The default value for this macro is @code{STACK_POINTER_OFFSET} plus the
3207 length of the outgoing arguments. The default is correct for most
3208 machines. See @file{function.c} for details.
3211 @defmac INITIAL_FRAME_ADDRESS_RTX
3212 A C expression whose value is RTL representing the address of the initial
3213 stack frame. This address is passed to @code{RETURN_ADDR_RTX} and
3214 @code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable
3215 default value will be used. Define this macro in order to make frame pointer
3216 elimination work in the presence of @code{__builtin_frame_address (count)} and
3217 @code{__builtin_return_address (count)} for @code{count} not equal to zero.
3220 @defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr})
3221 A C expression whose value is RTL representing the address in a stack
3222 frame where the pointer to the caller's frame is stored. Assume that
3223 @var{frameaddr} is an RTL expression for the address of the stack frame
3226 If you don't define this macro, the default is to return the value
3227 of @var{frameaddr}---that is, the stack frame address is also the
3228 address of the stack word that points to the previous frame.
3231 @defmac SETUP_FRAME_ADDRESSES
3232 If defined, a C expression that produces the machine-specific code to
3233 setup the stack so that arbitrary frames can be accessed. For example,
3234 on the SPARC, we must flush all of the register windows to the stack
3235 before we can access arbitrary stack frames. You will seldom need to
3239 @hook TARGET_BUILTIN_SETJMP_FRAME_VALUE
3240 This target hook should return an rtx that is used to store
3241 the address of the current frame into the built in @code{setjmp} buffer.
3242 The default value, @code{virtual_stack_vars_rtx}, is correct for most
3243 machines. One reason you may need to define this target hook is if
3244 @code{hard_frame_pointer_rtx} is the appropriate value on your machine.
3247 @defmac FRAME_ADDR_RTX (@var{frameaddr})
3248 A C expression whose value is RTL representing the value of the frame
3249 address for the current frame. @var{frameaddr} is the frame pointer
3250 of the current frame. This is used for __builtin_frame_address.
3251 You need only define this macro if the frame address is not the same
3252 as the frame pointer. Most machines do not need to define it.
3255 @defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr})
3256 A C expression whose value is RTL representing the value of the return
3257 address for the frame @var{count} steps up from the current frame, after
3258 the prologue. @var{frameaddr} is the frame pointer of the @var{count}
3259 frame, or the frame pointer of the @var{count} @minus{} 1 frame if
3260 @code{RETURN_ADDR_IN_PREVIOUS_FRAME} is defined.
3262 The value of the expression must always be the correct address when
3263 @var{count} is zero, but may be @code{NULL_RTX} if there is no way to
3264 determine the return address of other frames.
3267 @defmac RETURN_ADDR_IN_PREVIOUS_FRAME
3268 Define this if the return address of a particular stack frame is accessed
3269 from the frame pointer of the previous stack frame.
3272 @defmac INCOMING_RETURN_ADDR_RTX
3273 A C expression whose value is RTL representing the location of the
3274 incoming return address at the beginning of any function, before the
3275 prologue. This RTL is either a @code{REG}, indicating that the return
3276 value is saved in @samp{REG}, or a @code{MEM} representing a location in
3279 You only need to define this macro if you want to support call frame
3280 debugging information like that provided by DWARF 2.
3282 If this RTL is a @code{REG}, you should also define
3283 @code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}.
3286 @defmac DWARF_ALT_FRAME_RETURN_COLUMN
3287 A C expression whose value is an integer giving a DWARF 2 column
3288 number that may be used as an alternative return column. The column
3289 must not correspond to any gcc hard register (that is, it must not
3290 be in the range of @code{DWARF_FRAME_REGNUM}).
3292 This macro can be useful if @code{DWARF_FRAME_RETURN_COLUMN} is set to a
3293 general register, but an alternative column needs to be used for signal
3294 frames. Some targets have also used different frame return columns
3298 @defmac DWARF_ZERO_REG
3299 A C expression whose value is an integer giving a DWARF 2 register
3300 number that is considered to always have the value zero. This should
3301 only be defined if the target has an architected zero register, and
3302 someone decided it was a good idea to use that register number to
3303 terminate the stack backtrace. New ports should avoid this.
3306 @hook TARGET_DWARF_HANDLE_FRAME_UNSPEC
3307 This target hook allows the backend to emit frame-related insns that
3308 contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging
3309 info engine will invoke it on insns of the form
3311 (set (reg) (unspec [@dots{}] UNSPEC_INDEX))
3315 (set (reg) (unspec_volatile [@dots{}] UNSPECV_INDEX)).
3317 to let the backend emit the call frame instructions. @var{label} is
3318 the CFI label attached to the insn, @var{pattern} is the pattern of
3319 the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}.
3322 @defmac INCOMING_FRAME_SP_OFFSET
3323 A C expression whose value is an integer giving the offset, in bytes,
3324 from the value of the stack pointer register to the top of the stack
3325 frame at the beginning of any function, before the prologue. The top of
3326 the frame is defined to be the value of the stack pointer in the
3327 previous frame, just before the call instruction.
3329 You only need to define this macro if you want to support call frame
3330 debugging information like that provided by DWARF 2.
3333 @defmac ARG_POINTER_CFA_OFFSET (@var{fundecl})
3334 A C expression whose value is an integer giving the offset, in bytes,
3335 from the argument pointer to the canonical frame address (cfa). The
3336 final value should coincide with that calculated by
3337 @code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable
3338 during virtual register instantiation.
3340 The default value for this macro is
3341 @code{FIRST_PARM_OFFSET (fundecl) + crtl->args.pretend_args_size},
3342 which is correct for most machines; in general, the arguments are found
3343 immediately before the stack frame. Note that this is not the case on
3344 some targets that save registers into the caller's frame, such as SPARC
3345 and rs6000, and so such targets need to define this macro.
3347 You only need to define this macro if the default is incorrect, and you
3348 want to support call frame debugging information like that provided by
3352 @defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl})
3353 If defined, a C expression whose value is an integer giving the offset
3354 in bytes from the frame pointer to the canonical frame address (cfa).
3355 The final value should coincide with that calculated by
3356 @code{INCOMING_FRAME_SP_OFFSET}.
3358 Normally the CFA is calculated as an offset from the argument pointer,
3359 via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is
3360 variable due to the ABI, this may not be possible. If this macro is
3361 defined, it implies that the virtual register instantiation should be
3362 based on the frame pointer instead of the argument pointer. Only one
3363 of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET}
3367 @defmac CFA_FRAME_BASE_OFFSET (@var{fundecl})
3368 If defined, a C expression whose value is an integer giving the offset
3369 in bytes from the canonical frame address (cfa) to the frame base used
3370 in DWARF 2 debug information. The default is zero. A different value
3371 may reduce the size of debug information on some ports.
3374 @node Exception Handling
3375 @subsection Exception Handling Support
3376 @cindex exception handling
3378 @defmac EH_RETURN_DATA_REGNO (@var{N})
3379 A C expression whose value is the @var{N}th register number used for
3380 data by exception handlers, or @code{INVALID_REGNUM} if fewer than
3381 @var{N} registers are usable.
3383 The exception handling library routines communicate with the exception
3384 handlers via a set of agreed upon registers. Ideally these registers
3385 should be call-clobbered; it is possible to use call-saved registers,
3386 but may negatively impact code size. The target must support at least
3387 2 data registers, but should define 4 if there are enough free registers.
3389 You must define this macro if you want to support call frame exception
3390 handling like that provided by DWARF 2.
3393 @defmac EH_RETURN_STACKADJ_RTX
3394 A C expression whose value is RTL representing a location in which
3395 to store a stack adjustment to be applied before function return.
3396 This is used to unwind the stack to an exception handler's call frame.
3397 It will be assigned zero on code paths that return normally.
3399 Typically this is a call-clobbered hard register that is otherwise
3400 untouched by the epilogue, but could also be a stack slot.
3402 Do not define this macro if the stack pointer is saved and restored
3403 by the regular prolog and epilog code in the call frame itself; in
3404 this case, the exception handling library routines will update the
3405 stack location to be restored in place. Otherwise, you must define
3406 this macro if you want to support call frame exception handling like
3407 that provided by DWARF 2.
3410 @defmac EH_RETURN_HANDLER_RTX
3411 A C expression whose value is RTL representing a location in which
3412 to store the address of an exception handler to which we should
3413 return. It will not be assigned on code paths that return normally.
3415 Typically this is the location in the call frame at which the normal
3416 return address is stored. For targets that return by popping an
3417 address off the stack, this might be a memory address just below
3418 the @emph{target} call frame rather than inside the current call
3419 frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already
3420 been assigned, so it may be used to calculate the location of the
3423 Some targets have more complex requirements than storing to an
3424 address calculable during initial code generation. In that case
3425 the @code{eh_return} instruction pattern should be used instead.
3427 If you want to support call frame exception handling, you must
3428 define either this macro or the @code{eh_return} instruction pattern.
3431 @defmac RETURN_ADDR_OFFSET
3432 If defined, an integer-valued C expression for which rtl will be generated
3433 to add it to the exception handler address before it is searched in the
3434 exception handling tables, and to subtract it again from the address before
3435 using it to return to the exception handler.
3438 @defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global})
3439 This macro chooses the encoding of pointers embedded in the exception
3440 handling sections. If at all possible, this should be defined such
3441 that the exception handling section will not require dynamic relocations,
3442 and so may be read-only.
3444 @var{code} is 0 for data, 1 for code labels, 2 for function pointers.
3445 @var{global} is true if the symbol may be affected by dynamic relocations.
3446 The macro should return a combination of the @code{DW_EH_PE_*} defines
3447 as found in @file{dwarf2.h}.
3449 If this macro is not defined, pointers will not be encoded but
3450 represented directly.
3453 @defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done})
3454 This macro allows the target to emit whatever special magic is required
3455 to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}.
3456 Generic code takes care of pc-relative and indirect encodings; this must
3457 be defined if the target uses text-relative or data-relative encodings.
3459 This is a C statement that branches to @var{done} if the format was
3460 handled. @var{encoding} is the format chosen, @var{size} is the number
3461 of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF}
3465 @defmac MD_UNWIND_SUPPORT
3466 A string specifying a file to be #include'd in unwind-dw2.c. The file
3467 so included typically defines @code{MD_FALLBACK_FRAME_STATE_FOR}.
3470 @defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs})
3471 This macro allows the target to add CPU and operating system specific
3472 code to the call-frame unwinder for use when there is no unwind data
3473 available. The most common reason to implement this macro is to unwind
3474 through signal frames.
3476 This macro is called from @code{uw_frame_state_for} in
3477 @file{unwind-dw2.c}, @file{unwind-dw2-xtensa.c} and
3478 @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context};
3479 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra}
3480 for the address of the code being executed and @code{context->cfa} for
3481 the stack pointer value. If the frame can be decoded, the register
3482 save addresses should be updated in @var{fs} and the macro should
3483 evaluate to @code{_URC_NO_REASON}. If the frame cannot be decoded,
3484 the macro should evaluate to @code{_URC_END_OF_STACK}.
3486 For proper signal handling in Java this macro is accompanied by
3487 @code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers.
3490 @defmac MD_HANDLE_UNWABI (@var{context}, @var{fs})
3491 This macro allows the target to add operating system specific code to the
3492 call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive,
3493 usually used for signal or interrupt frames.
3495 This macro is called from @code{uw_update_context} in @file{unwind-ia64.c}.
3496 @var{context} is an @code{_Unwind_Context};
3497 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi}
3498 for the abi and context in the @code{.unwabi} directive. If the
3499 @code{.unwabi} directive can be handled, the register save addresses should
3500 be updated in @var{fs}.
3503 @defmac TARGET_USES_WEAK_UNWIND_INFO
3504 A C expression that evaluates to true if the target requires unwind
3505 info to be given comdat linkage. Define it to be @code{1} if comdat
3506 linkage is necessary. The default is @code{0}.
3509 @node Stack Checking
3510 @subsection Specifying How Stack Checking is Done
3512 GCC will check that stack references are within the boundaries of the
3513 stack, if the option @option{-fstack-check} is specified, in one of
3518 If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC
3519 will assume that you have arranged for full stack checking to be done
3520 at appropriate places in the configuration files. GCC will not do
3521 other special processing.
3524 If @code{STACK_CHECK_BUILTIN} is zero and the value of the
3525 @code{STACK_CHECK_STATIC_BUILTIN} macro is nonzero, GCC will assume
3526 that you have arranged for static stack checking (checking of the
3527 static stack frame of functions) to be done at appropriate places
3528 in the configuration files. GCC will only emit code to do dynamic
3529 stack checking (checking on dynamic stack allocations) using the third
3533 If neither of the above are true, GCC will generate code to periodically
3534 ``probe'' the stack pointer using the values of the macros defined below.
3537 If neither STACK_CHECK_BUILTIN nor STACK_CHECK_STATIC_BUILTIN is defined,
3538 GCC will change its allocation strategy for large objects if the option
3539 @option{-fstack-check} is specified: they will always be allocated
3540 dynamically if their size exceeds @code{STACK_CHECK_MAX_VAR_SIZE} bytes.
3542 @defmac STACK_CHECK_BUILTIN
3543 A nonzero value if stack checking is done by the configuration files in a
3544 machine-dependent manner. You should define this macro if stack checking
3545 is required by the ABI of your machine or if you would like to do stack
3546 checking in some more efficient way than the generic approach. The default
3547 value of this macro is zero.
3550 @defmac STACK_CHECK_STATIC_BUILTIN
3551 A nonzero value if static stack checking is done by the configuration files
3552 in a machine-dependent manner. You should define this macro if you would
3553 like to do static stack checking in some more efficient way than the generic
3554 approach. The default value of this macro is zero.
3557 @defmac STACK_CHECK_PROBE_INTERVAL_EXP
3558 An integer specifying the interval at which GCC must generate stack probe
3559 instructions, defined as 2 raised to this integer. You will normally
3560 define this macro so that the interval be no larger than the size of
3561 the ``guard pages'' at the end of a stack area. The default value
3562 of 12 (4096-byte interval) is suitable for most systems.
3565 @defmac STACK_CHECK_MOVING_SP
3566 An integer which is nonzero if GCC should move the stack pointer page by page
3567 when doing probes. This can be necessary on systems where the stack pointer
3568 contains the bottom address of the memory area accessible to the executing
3569 thread at any point in time. In this situation an alternate signal stack
3570 is required in order to be able to recover from a stack overflow. The
3571 default value of this macro is zero.
3574 @defmac STACK_CHECK_PROTECT
3575 The number of bytes of stack needed to recover from a stack overflow, for
3576 languages where such a recovery is supported. The default value of 75 words
3577 with the @code{setjmp}/@code{longjmp}-based exception handling mechanism and
3578 8192 bytes with other exception handling mechanisms should be adequate for
3582 The following macros are relevant only if neither STACK_CHECK_BUILTIN
3583 nor STACK_CHECK_STATIC_BUILTIN is defined; you can omit them altogether
3584 in the opposite case.
3586 @defmac STACK_CHECK_MAX_FRAME_SIZE
3587 The maximum size of a stack frame, in bytes. GCC will generate probe
3588 instructions in non-leaf functions to ensure at least this many bytes of
3589 stack are available. If a stack frame is larger than this size, stack
3590 checking will not be reliable and GCC will issue a warning. The
3591 default is chosen so that GCC only generates one instruction on most
3592 systems. You should normally not change the default value of this macro.
3595 @defmac STACK_CHECK_FIXED_FRAME_SIZE
3596 GCC uses this value to generate the above warning message. It
3597 represents the amount of fixed frame used by a function, not including
3598 space for any callee-saved registers, temporaries and user variables.
3599 You need only specify an upper bound for this amount and will normally
3600 use the default of four words.
3603 @defmac STACK_CHECK_MAX_VAR_SIZE
3604 The maximum size, in bytes, of an object that GCC will place in the
3605 fixed area of the stack frame when the user specifies
3606 @option{-fstack-check}.
3607 GCC computed the default from the values of the above macros and you will
3608 normally not need to override that default.
3612 @node Frame Registers
3613 @subsection Registers That Address the Stack Frame
3615 @c prevent bad page break with this line
3616 This discusses registers that address the stack frame.
3618 @defmac STACK_POINTER_REGNUM
3619 The register number of the stack pointer register, which must also be a
3620 fixed register according to @code{FIXED_REGISTERS}. On most machines,
3621 the hardware determines which register this is.
3624 @defmac FRAME_POINTER_REGNUM
3625 The register number of the frame pointer register, which is used to
3626 access automatic variables in the stack frame. On some machines, the
3627 hardware determines which register this is. On other machines, you can
3628 choose any register you wish for this purpose.
3631 @defmac HARD_FRAME_POINTER_REGNUM
3632 On some machines the offset between the frame pointer and starting
3633 offset of the automatic variables is not known until after register
3634 allocation has been done (for example, because the saved registers are
3635 between these two locations). On those machines, define
3636 @code{FRAME_POINTER_REGNUM} the number of a special, fixed register to
3637 be used internally until the offset is known, and define
3638 @code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number
3639 used for the frame pointer.
3641 You should define this macro only in the very rare circumstances when it
3642 is not possible to calculate the offset between the frame pointer and
3643 the automatic variables until after register allocation has been
3644 completed. When this macro is defined, you must also indicate in your
3645 definition of @code{ELIMINABLE_REGS} how to eliminate
3646 @code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM}
3647 or @code{STACK_POINTER_REGNUM}.
3649 Do not define this macro if it would be the same as
3650 @code{FRAME_POINTER_REGNUM}.
3653 @defmac ARG_POINTER_REGNUM
3654 The register number of the arg pointer register, which is used to access
3655 the function's argument list. On some machines, this is the same as the
3656 frame pointer register. On some machines, the hardware determines which
3657 register this is. On other machines, you can choose any register you
3658 wish for this purpose. If this is not the same register as the frame
3659 pointer register, then you must mark it as a fixed register according to
3660 @code{FIXED_REGISTERS}, or arrange to be able to eliminate it
3661 (@pxref{Elimination}).
3664 @defmac HARD_FRAME_POINTER_IS_FRAME_POINTER
3665 Define this to a preprocessor constant that is nonzero if
3666 @code{hard_frame_pointer_rtx} and @code{frame_pointer_rtx} should be
3667 the same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM
3668 == FRAME_POINTER_REGNUM)}; you only need to define this macro if that
3669 definition is not suitable for use in preprocessor conditionals.
3672 @defmac HARD_FRAME_POINTER_IS_ARG_POINTER
3673 Define this to a preprocessor constant that is nonzero if
3674 @code{hard_frame_pointer_rtx} and @code{arg_pointer_rtx} should be the
3675 same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM ==
3676 ARG_POINTER_REGNUM)}; you only need to define this macro if that
3677 definition is not suitable for use in preprocessor conditionals.
3680 @defmac RETURN_ADDRESS_POINTER_REGNUM
3681 The register number of the return address pointer register, which is used to
3682 access the current function's return address from the stack. On some
3683 machines, the return address is not at a fixed offset from the frame
3684 pointer or stack pointer or argument pointer. This register can be defined
3685 to point to the return address on the stack, and then be converted by
3686 @code{ELIMINABLE_REGS} into either the frame pointer or stack pointer.
3688 Do not define this macro unless there is no other way to get the return
3689 address from the stack.
3692 @defmac STATIC_CHAIN_REGNUM
3693 @defmacx STATIC_CHAIN_INCOMING_REGNUM
3694 Register numbers used for passing a function's static chain pointer. If
3695 register windows are used, the register number as seen by the called
3696 function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register
3697 number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If
3698 these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need
3701 The static chain register need not be a fixed register.
3703 If the static chain is passed in memory, these macros should not be
3704 defined; instead, the @code{TARGET_STATIC_CHAIN} hook should be used.
3707 @hook TARGET_STATIC_CHAIN
3708 This hook replaces the use of @code{STATIC_CHAIN_REGNUM} et al for
3709 targets that may use different static chain locations for different
3710 nested functions. This may be required if the target has function
3711 attributes that affect the calling conventions of the function and
3712 those calling conventions use different static chain locations.
3714 The default version of this hook uses @code{STATIC_CHAIN_REGNUM} et al.
3716 If the static chain is passed in memory, this hook should be used to
3717 provide rtx giving @code{mem} expressions that denote where they are stored.
3718 Often the @code{mem} expression as seen by the caller will be at an offset
3719 from the stack pointer and the @code{mem} expression as seen by the callee
3720 will be at an offset from the frame pointer.
3721 @findex stack_pointer_rtx
3722 @findex frame_pointer_rtx
3723 @findex arg_pointer_rtx
3724 The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and
3725 @code{arg_pointer_rtx} will have been initialized and should be used
3726 to refer to those items.
3729 @defmac DWARF_FRAME_REGISTERS
3730 This macro specifies the maximum number of hard registers that can be
3731 saved in a call frame. This is used to size data structures used in
3732 DWARF2 exception handling.
3734 Prior to GCC 3.0, this macro was needed in order to establish a stable
3735 exception handling ABI in the face of adding new hard registers for ISA
3736 extensions. In GCC 3.0 and later, the EH ABI is insulated from changes
3737 in the number of hard registers. Nevertheless, this macro can still be
3738 used to reduce the runtime memory requirements of the exception handling
3739 routines, which can be substantial if the ISA contains a lot of
3740 registers that are not call-saved.
3742 If this macro is not defined, it defaults to
3743 @code{FIRST_PSEUDO_REGISTER}.
3746 @defmac PRE_GCC3_DWARF_FRAME_REGISTERS
3748 This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided
3749 for backward compatibility in pre GCC 3.0 compiled code.
3751 If this macro is not defined, it defaults to
3752 @code{DWARF_FRAME_REGISTERS}.
3755 @defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno})
3757 Define this macro if the target's representation for dwarf registers
3758 is different than the internal representation for unwind column.
3759 Given a dwarf register, this macro should return the internal unwind
3760 column number to use instead.
3762 See the PowerPC's SPE target for an example.
3765 @defmac DWARF_FRAME_REGNUM (@var{regno})
3767 Define this macro if the target's representation for dwarf registers
3768 used in .eh_frame or .debug_frame is different from that used in other
3769 debug info sections. Given a GCC hard register number, this macro
3770 should return the .eh_frame register number. The default is
3771 @code{DBX_REGISTER_NUMBER (@var{regno})}.
3775 @defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh})
3777 Define this macro to map register numbers held in the call frame info
3778 that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that
3779 should be output in .debug_frame (@code{@var{for_eh}} is zero) and
3780 .eh_frame (@code{@var{for_eh}} is nonzero). The default is to
3781 return @code{@var{regno}}.
3786 @subsection Eliminating Frame Pointer and Arg Pointer
3788 @c prevent bad page break with this line
3789 This is about eliminating the frame pointer and arg pointer.
3791 @hook TARGET_FRAME_POINTER_REQUIRED
3792 This target hook should return @code{true} if a function must have and use
3793 a frame pointer. This target hook is called in the reload pass. If its return
3794 value is @code{true} the function will have a frame pointer.
3796 This target hook can in principle examine the current function and decide
3797 according to the facts, but on most machines the constant @code{false} or the
3798 constant @code{true} suffices. Use @code{false} when the machine allows code
3799 to be generated with no frame pointer, and doing so saves some time or space.
3800 Use @code{true} when there is no possible advantage to avoiding a frame
3803 In certain cases, the compiler does not know how to produce valid code
3804 without a frame pointer. The compiler recognizes those cases and
3805 automatically gives the function a frame pointer regardless of what
3806 @code{TARGET_FRAME_POINTER_REQUIRED} returns. You don't need to worry about
3809 In a function that does not require a frame pointer, the frame pointer
3810 register can be allocated for ordinary usage, unless you mark it as a
3811 fixed register. See @code{FIXED_REGISTERS} for more information.
3813 Default return value is @code{false}.
3816 @findex get_frame_size
3817 @defmac INITIAL_FRAME_POINTER_OFFSET (@var{depth-var})
3818 A C statement to store in the variable @var{depth-var} the difference
3819 between the frame pointer and the stack pointer values immediately after
3820 the function prologue. The value would be computed from information
3821 such as the result of @code{get_frame_size ()} and the tables of
3822 registers @code{regs_ever_live} and @code{call_used_regs}.
3824 If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and
3825 need not be defined. Otherwise, it must be defined even if
3826 @code{TARGET_FRAME_POINTER_REQUIRED} always returns true; in that
3827 case, you may set @var{depth-var} to anything.
3830 @defmac ELIMINABLE_REGS
3831 If defined, this macro specifies a table of register pairs used to
3832 eliminate unneeded registers that point into the stack frame. If it is not
3833 defined, the only elimination attempted by the compiler is to replace
3834 references to the frame pointer with references to the stack pointer.
3836 The definition of this macro is a list of structure initializations, each
3837 of which specifies an original and replacement register.
3839 On some machines, the position of the argument pointer is not known until
3840 the compilation is completed. In such a case, a separate hard register
3841 must be used for the argument pointer. This register can be eliminated by
3842 replacing it with either the frame pointer or the argument pointer,
3843 depending on whether or not the frame pointer has been eliminated.
3845 In this case, you might specify:
3847 #define ELIMINABLE_REGS \
3848 @{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \
3849 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \
3850 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@}
3853 Note that the elimination of the argument pointer with the stack pointer is
3854 specified first since that is the preferred elimination.
3857 @hook TARGET_CAN_ELIMINATE
3858 This target hook should returns @code{true} if the compiler is allowed to
3859 try to replace register number @var{from_reg} with register number
3860 @var{to_reg}. This target hook need only be defined if @code{ELIMINABLE_REGS}
3861 is defined, and will usually be @code{true}, since most of the cases
3862 preventing register elimination are things that the compiler already
3865 Default return value is @code{true}.
3868 @defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var})
3869 This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It
3870 specifies the initial difference between the specified pair of
3871 registers. This macro must be defined if @code{ELIMINABLE_REGS} is
3875 @node Stack Arguments
3876 @subsection Passing Function Arguments on the Stack
3877 @cindex arguments on stack
3878 @cindex stack arguments
3880 The macros in this section control how arguments are passed
3881 on the stack. See the following section for other macros that
3882 control passing certain arguments in registers.
3884 @hook TARGET_PROMOTE_PROTOTYPES
3885 This target hook returns @code{true} if an argument declared in a
3886 prototype as an integral type smaller than @code{int} should actually be
3887 passed as an @code{int}. In addition to avoiding errors in certain
3888 cases of mismatch, it also makes for better code on certain machines.
3889 The default is to not promote prototypes.
3893 A C expression. If nonzero, push insns will be used to pass
3895 If the target machine does not have a push instruction, set it to zero.
3896 That directs GCC to use an alternate strategy: to
3897 allocate the entire argument block and then store the arguments into
3898 it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too.
3901 @defmac PUSH_ARGS_REVERSED
3902 A C expression. If nonzero, function arguments will be evaluated from
3903 last to first, rather than from first to last. If this macro is not
3904 defined, it defaults to @code{PUSH_ARGS} on targets where the stack
3905 and args grow in opposite directions, and 0 otherwise.
3908 @defmac PUSH_ROUNDING (@var{npushed})
3909 A C expression that is the number of bytes actually pushed onto the
3910 stack when an instruction attempts to push @var{npushed} bytes.
3912 On some machines, the definition
3915 #define PUSH_ROUNDING(BYTES) (BYTES)
3919 will suffice. But on other machines, instructions that appear
3920 to push one byte actually push two bytes in an attempt to maintain
3921 alignment. Then the definition should be
3924 #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
3928 @findex current_function_outgoing_args_size
3929 @defmac ACCUMULATE_OUTGOING_ARGS
3930 A C expression. If nonzero, the maximum amount of space required for outgoing arguments
3931 will be computed and placed into the variable
3932 @code{current_function_outgoing_args_size}. No space will be pushed
3933 onto the stack for each call; instead, the function prologue should
3934 increase the stack frame size by this amount.
3936 Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS}
3940 @defmac REG_PARM_STACK_SPACE (@var{fndecl})
3941 Define this macro if functions should assume that stack space has been
3942 allocated for arguments even when their values are passed in
3945 The value of this macro is the size, in bytes, of the area reserved for
3946 arguments passed in registers for the function represented by @var{fndecl},
3947 which can be zero if GCC is calling a library function.
3948 The argument @var{fndecl} can be the FUNCTION_DECL, or the type itself
3951 This space can be allocated by the caller, or be a part of the
3952 machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says
3955 @c above is overfull. not sure what to do. --mew 5feb93 did
3956 @c something, not sure if it looks good. --mew 10feb93
3958 @defmac OUTGOING_REG_PARM_STACK_SPACE (@var{fntype})
3959 Define this to a nonzero value if it is the responsibility of the
3960 caller to allocate the area reserved for arguments passed in registers
3961 when calling a function of @var{fntype}. @var{fntype} may be NULL
3962 if the function called is a library function.
3964 If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls
3965 whether the space for these arguments counts in the value of
3966 @code{current_function_outgoing_args_size}.
3969 @defmac STACK_PARMS_IN_REG_PARM_AREA
3970 Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the
3971 stack parameters don't skip the area specified by it.
3972 @c i changed this, makes more sens and it should have taken care of the
3973 @c overfull.. not as specific, tho. --mew 5feb93
3975 Normally, when a parameter is not passed in registers, it is placed on the
3976 stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro
3977 suppresses this behavior and causes the parameter to be passed on the
3978 stack in its natural location.
3981 @hook TARGET_RETURN_POPS_ARGS
3982 This target hook returns the number of bytes of its own arguments that
3983 a function pops on returning, or 0 if the function pops no arguments
3984 and the caller must therefore pop them all after the function returns.
3986 @var{fundecl} is a C variable whose value is a tree node that describes
3987 the function in question. Normally it is a node of type
3988 @code{FUNCTION_DECL} that describes the declaration of the function.
3989 From this you can obtain the @code{DECL_ATTRIBUTES} of the function.
3991 @var{funtype} is a C variable whose value is a tree node that
3992 describes the function in question. Normally it is a node of type
3993 @code{FUNCTION_TYPE} that describes the data type of the function.
3994 From this it is possible to obtain the data types of the value and
3995 arguments (if known).
3997 When a call to a library function is being considered, @var{fundecl}
3998 will contain an identifier node for the library function. Thus, if
3999 you need to distinguish among various library functions, you can do so
4000 by their names. Note that ``library function'' in this context means
4001 a function used to perform arithmetic, whose name is known specially
4002 in the compiler and was not mentioned in the C code being compiled.
4004 @var{size} is the number of bytes of arguments passed on the
4005 stack. If a variable number of bytes is passed, it is zero, and
4006 argument popping will always be the responsibility of the calling function.
4008 On the VAX, all functions always pop their arguments, so the definition
4009 of this macro is @var{size}. On the 68000, using the standard
4010 calling convention, no functions pop their arguments, so the value of
4011 the macro is always 0 in this case. But an alternative calling
4012 convention is available in which functions that take a fixed number of
4013 arguments pop them but other functions (such as @code{printf}) pop
4014 nothing (the caller pops all). When this convention is in use,
4015 @var{funtype} is examined to determine whether a function takes a fixed
4016 number of arguments.
4019 @defmac CALL_POPS_ARGS (@var{cum})
4020 A C expression that should indicate the number of bytes a call sequence
4021 pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS}
4022 when compiling a function call.
4024 @var{cum} is the variable in which all arguments to the called function
4025 have been accumulated.
4027 On certain architectures, such as the SH5, a call trampoline is used
4028 that pops certain registers off the stack, depending on the arguments
4029 that have been passed to the function. Since this is a property of the
4030 call site, not of the called function, @code{RETURN_POPS_ARGS} is not
4034 @node Register Arguments
4035 @subsection Passing Arguments in Registers
4036 @cindex arguments in registers
4037 @cindex registers arguments
4039 This section describes the macros which let you control how various
4040 types of arguments are passed in registers or how they are arranged in
4043 @defmac FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
4044 A C expression that controls whether a function argument is passed
4045 in a register, and which register.
4047 The arguments are @var{cum}, which summarizes all the previous
4048 arguments; @var{mode}, the machine mode of the argument; @var{type},
4049 the data type of the argument as a tree node or 0 if that is not known
4050 (which happens for C support library functions); and @var{named},
4051 which is 1 for an ordinary argument and 0 for nameless arguments that
4052 correspond to @samp{@dots{}} in the called function's prototype.
4053 @var{type} can be an incomplete type if a syntax error has previously
4056 The value of the expression is usually either a @code{reg} RTX for the
4057 hard register in which to pass the argument, or zero to pass the
4058 argument on the stack.
4060 For machines like the VAX and 68000, where normally all arguments are
4061 pushed, zero suffices as a definition.
4063 The value of the expression can also be a @code{parallel} RTX@. This is
4064 used when an argument is passed in multiple locations. The mode of the
4065 @code{parallel} should be the mode of the entire argument. The
4066 @code{parallel} holds any number of @code{expr_list} pairs; each one
4067 describes where part of the argument is passed. In each
4068 @code{expr_list} the first operand must be a @code{reg} RTX for the hard
4069 register in which to pass this part of the argument, and the mode of the
4070 register RTX indicates how large this part of the argument is. The
4071 second operand of the @code{expr_list} is a @code{const_int} which gives
4072 the offset in bytes into the entire argument of where this part starts.
4073 As a special exception the first @code{expr_list} in the @code{parallel}
4074 RTX may have a first operand of zero. This indicates that the entire
4075 argument is also stored on the stack.
4077 The last time this macro is called, it is called with @code{MODE ==
4078 VOIDmode}, and its result is passed to the @code{call} or @code{call_value}
4079 pattern as operands 2 and 3 respectively.
4081 @cindex @file{stdarg.h} and register arguments
4082 The usual way to make the ISO library @file{stdarg.h} work on a machine
4083 where some arguments are usually passed in registers, is to cause
4084 nameless arguments to be passed on the stack instead. This is done
4085 by making @code{FUNCTION_ARG} return 0 whenever @var{named} is 0.
4087 @cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{FUNCTION_ARG}
4088 @cindex @code{REG_PARM_STACK_SPACE}, and @code{FUNCTION_ARG}
4089 You may use the hook @code{targetm.calls.must_pass_in_stack}
4090 in the definition of this macro to determine if this argument is of a
4091 type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE}
4092 is not defined and @code{FUNCTION_ARG} returns nonzero for such an
4093 argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is
4094 defined, the argument will be computed in the stack and then loaded into
4098 @hook TARGET_MUST_PASS_IN_STACK
4099 This target hook should return @code{true} if we should not pass @var{type}
4100 solely in registers. The file @file{expr.h} defines a
4101 definition that is usually appropriate, refer to @file{expr.h} for additional
4105 @defmac FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
4106 Define this macro if the target machine has ``register windows'', so
4107 that the register in which a function sees an arguments is not
4108 necessarily the same as the one in which the caller passed the
4111 For such machines, @code{FUNCTION_ARG} computes the register in which
4112 the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
4113 be defined in a similar fashion to tell the function being called
4114 where the arguments will arrive.
4116 If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
4117 serves both purposes.
4120 @hook TARGET_ARG_PARTIAL_BYTES
4121 This target hook returns the number of bytes at the beginning of an
4122 argument that must be put in registers. The value must be zero for
4123 arguments that are passed entirely in registers or that are entirely
4124 pushed on the stack.
4126 On some machines, certain arguments must be passed partially in
4127 registers and partially in memory. On these machines, typically the
4128 first few words of arguments are passed in registers, and the rest
4129 on the stack. If a multi-word argument (a @code{double} or a
4130 structure) crosses that boundary, its first few words must be passed
4131 in registers and the rest must be pushed. This macro tells the
4132 compiler when this occurs, and how many bytes should go in registers.
4134 @code{FUNCTION_ARG} for these arguments should return the first
4135 register to be used by the caller for this argument; likewise
4136 @code{FUNCTION_INCOMING_ARG}, for the called function.
4139 @hook TARGET_PASS_BY_REFERENCE
4140 This target hook should return @code{true} if an argument at the
4141 position indicated by @var{cum} should be passed by reference. This
4142 predicate is queried after target independent reasons for being
4143 passed by reference, such as @code{TREE_ADDRESSABLE (type)}.
4145 If the hook returns true, a copy of that argument is made in memory and a
4146 pointer to the argument is passed instead of the argument itself.
4147 The pointer is passed in whatever way is appropriate for passing a pointer
4151 @hook TARGET_CALLEE_COPIES
4152 The function argument described by the parameters to this hook is
4153 known to be passed by reference. The hook should return true if the
4154 function argument should be copied by the callee instead of copied
4157 For any argument for which the hook returns true, if it can be
4158 determined that the argument is not modified, then a copy need
4161 The default version of this hook always returns false.
4164 @defmac CUMULATIVE_ARGS
4165 A C type for declaring a variable that is used as the first argument of
4166 @code{FUNCTION_ARG} and other related values. For some target machines,
4167 the type @code{int} suffices and can hold the number of bytes of
4170 There is no need to record in @code{CUMULATIVE_ARGS} anything about the
4171 arguments that have been passed on the stack. The compiler has other
4172 variables to keep track of that. For target machines on which all
4173 arguments are passed on the stack, there is no need to store anything in
4174 @code{CUMULATIVE_ARGS}; however, the data structure must exist and
4175 should not be empty, so use @code{int}.
4178 @defmac OVERRIDE_ABI_FORMAT (@var{fndecl})
4179 If defined, this macro is called before generating any code for a
4180 function, but after the @var{cfun} descriptor for the function has been
4181 created. The back end may use this macro to update @var{cfun} to
4182 reflect an ABI other than that which would normally be used by default.
4183 If the compiler is generating code for a compiler-generated function,
4184 @var{fndecl} may be @code{NULL}.
4187 @defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args})
4188 A C statement (sans semicolon) for initializing the variable
4189 @var{cum} for the state at the beginning of the argument list. The
4190 variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype}
4191 is the tree node for the data type of the function which will receive
4192 the args, or 0 if the args are to a compiler support library function.
4193 For direct calls that are not libcalls, @var{fndecl} contain the
4194 declaration node of the function. @var{fndecl} is also set when
4195 @code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function
4196 being compiled. @var{n_named_args} is set to the number of named
4197 arguments, including a structure return address if it is passed as a
4198 parameter, when making a call. When processing incoming arguments,
4199 @var{n_named_args} is set to @minus{}1.
4201 When processing a call to a compiler support library function,
4202 @var{libname} identifies which one. It is a @code{symbol_ref} rtx which
4203 contains the name of the function, as a string. @var{libname} is 0 when
4204 an ordinary C function call is being processed. Thus, each time this
4205 macro is called, either @var{libname} or @var{fntype} is nonzero, but
4206 never both of them at once.
4209 @defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname})
4210 Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls,
4211 it gets a @code{MODE} argument instead of @var{fntype}, that would be
4212 @code{NULL}. @var{indirect} would always be zero, too. If this macro
4213 is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname,
4214 0)} is used instead.
4217 @defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname})
4218 Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of
4219 finding the arguments for the function being compiled. If this macro is
4220 undefined, @code{INIT_CUMULATIVE_ARGS} is used instead.
4222 The value passed for @var{libname} is always 0, since library routines
4223 with special calling conventions are never compiled with GCC@. The
4224 argument @var{libname} exists for symmetry with
4225 @code{INIT_CUMULATIVE_ARGS}.
4226 @c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe.
4227 @c --mew 5feb93 i switched the order of the sentences. --mew 10feb93
4230 @defmac FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
4231 A C statement (sans semicolon) to update the summarizer variable
4232 @var{cum} to advance past an argument in the argument list. The
4233 values @var{mode}, @var{type} and @var{named} describe that argument.
4234 Once this is done, the variable @var{cum} is suitable for analyzing
4235 the @emph{following} argument with @code{FUNCTION_ARG}, etc.
4237 This macro need not do anything if the argument in question was passed
4238 on the stack. The compiler knows how to track the amount of stack space
4239 used for arguments without any special help.
4242 @defmac FUNCTION_ARG_OFFSET (@var{mode}, @var{type})
4243 If defined, a C expression that is the number of bytes to add to the
4244 offset of the argument passed in memory. This is needed for the SPU,
4245 which passes @code{char} and @code{short} arguments in the preferred
4246 slot that is in the middle of the quad word instead of starting at the
4250 @defmac FUNCTION_ARG_PADDING (@var{mode}, @var{type})
4251 If defined, a C expression which determines whether, and in which direction,
4252 to pad out an argument with extra space. The value should be of type
4253 @code{enum direction}: either @code{upward} to pad above the argument,
4254 @code{downward} to pad below, or @code{none} to inhibit padding.
4256 The @emph{amount} of padding is always just enough to reach the next
4257 multiple of @code{FUNCTION_ARG_BOUNDARY}; this macro does not control
4260 This macro has a default definition which is right for most systems.
4261 For little-endian machines, the default is to pad upward. For
4262 big-endian machines, the default is to pad downward for an argument of
4263 constant size shorter than an @code{int}, and upward otherwise.
4266 @defmac PAD_VARARGS_DOWN
4267 If defined, a C expression which determines whether the default
4268 implementation of va_arg will attempt to pad down before reading the
4269 next argument, if that argument is smaller than its aligned space as
4270 controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such
4271 arguments are padded down if @code{BYTES_BIG_ENDIAN} is true.
4274 @defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first})
4275 Specify padding for the last element of a block move between registers and
4276 memory. @var{first} is nonzero if this is the only element. Defining this
4277 macro allows better control of register function parameters on big-endian
4278 machines, without using @code{PARALLEL} rtl. In particular,
4279 @code{MUST_PASS_IN_STACK} need not test padding and mode of types in
4280 registers, as there is no longer a "wrong" part of a register; For example,
4281 a three byte aggregate may be passed in the high part of a register if so
4285 @defmac FUNCTION_ARG_BOUNDARY (@var{mode}, @var{type})
4286 If defined, a C expression that gives the alignment boundary, in bits,
4287 of an argument with the specified mode and type. If it is not defined,
4288 @code{PARM_BOUNDARY} is used for all arguments.
4291 @defmac FUNCTION_ARG_REGNO_P (@var{regno})
4292 A C expression that is nonzero if @var{regno} is the number of a hard
4293 register in which function arguments are sometimes passed. This does
4294 @emph{not} include implicit arguments such as the static chain and
4295 the structure-value address. On many machines, no registers can be
4296 used for this purpose since all function arguments are pushed on the
4300 @hook TARGET_SPLIT_COMPLEX_ARG
4301 This hook should return true if parameter of type @var{type} are passed
4302 as two scalar parameters. By default, GCC will attempt to pack complex
4303 arguments into the target's word size. Some ABIs require complex arguments
4304 to be split and treated as their individual components. For example, on
4305 AIX64, complex floats should be passed in a pair of floating point
4306 registers, even though a complex float would fit in one 64-bit floating
4309 The default value of this hook is @code{NULL}, which is treated as always
4313 @hook TARGET_BUILD_BUILTIN_VA_LIST
4314 This hook returns a type node for @code{va_list} for the target.
4315 The default version of the hook returns @code{void*}.
4318 @hook TARGET_ENUM_VA_LIST_P
4319 This target hook is used in function @code{c_common_nodes_and_builtins}
4320 to iterate through the target specific builtin types for va_list. The
4321 variable @var{idx} is used as iterator. @var{pname} has to be a pointer
4322 to a @code{const char *} and @var{ptree} a pointer to a @code{tree} typed
4324 The arguments @var{pname} and @var{ptree} are used to store the result of
4325 this macro and are set to the name of the va_list builtin type and its
4327 If the return value of this macro is zero, then there is no more element.
4328 Otherwise the @var{IDX} should be increased for the next call of this
4329 macro to iterate through all types.
4332 @hook TARGET_FN_ABI_VA_LIST
4333 This hook returns the va_list type of the calling convention specified by
4335 The default version of this hook returns @code{va_list_type_node}.
4338 @hook TARGET_CANONICAL_VA_LIST_TYPE
4339 This hook returns the va_list type of the calling convention specified by the
4340 type of @var{type}. If @var{type} is not a valid va_list type, it returns
4344 @hook TARGET_GIMPLIFY_VA_ARG_EXPR
4345 This hook performs target-specific gimplification of
4346 @code{VA_ARG_EXPR}. The first two parameters correspond to the
4347 arguments to @code{va_arg}; the latter two are as in
4348 @code{gimplify.c:gimplify_expr}.
4351 @hook TARGET_VALID_POINTER_MODE
4352 Define this to return nonzero if the port can handle pointers
4353 with machine mode @var{mode}. The default version of this
4354 hook returns true for both @code{ptr_mode} and @code{Pmode}.
4357 @hook TARGET_SCALAR_MODE_SUPPORTED_P
4358 Define this to return nonzero if the port is prepared to handle
4359 insns involving scalar mode @var{mode}. For a scalar mode to be
4360 considered supported, all the basic arithmetic and comparisons
4363 The default version of this hook returns true for any mode
4364 required to handle the basic C types (as defined by the port).
4365 Included here are the double-word arithmetic supported by the
4366 code in @file{optabs.c}.
4369 @hook TARGET_VECTOR_MODE_SUPPORTED_P
4370 Define this to return nonzero if the port is prepared to handle
4371 insns involving vector mode @var{mode}. At the very least, it
4372 must have move patterns for this mode.
4375 @hook TARGET_SMALL_REGISTER_CLASSES_FOR_MODE_P
4376 Define this to return nonzero for machine modes for which the port has
4377 small register classes. If this target hook returns nonzero for a given
4378 @var{mode}, the compiler will try to minimize the lifetime of registers
4379 in @var{mode}. The hook may be called with @code{VOIDmode} as argument.
4380 In this case, the hook is expected to return nonzero if it returns nonzero
4383 On some machines, it is risky to let hard registers live across arbitrary
4384 insns. Typically, these machines have instructions that require values
4385 to be in specific registers (like an accumulator), and reload will fail
4386 if the required hard register is used for another purpose across such an
4389 Passes before reload do not know which hard registers will be used
4390 in an instruction, but the machine modes of the registers set or used in
4391 the instruction are already known. And for some machines, register
4392 classes are small for, say, integer registers but not for floating point
4393 registers. For example, the AMD x86-64 architecture requires specific
4394 registers for the legacy x86 integer instructions, but there are many
4395 SSE registers for floating point operations. On such targets, a good
4396 strategy may be to return nonzero from this hook for @code{INTEGRAL_MODE_P}
4397 machine modes but zero for the SSE register classes.
4399 The default version of this hook retuns false for any mode. It is always
4400 safe to redefine this hook to return with a nonzero value. But if you
4401 unnecessarily define it, you will reduce the amount of optimizations
4402 that can be performed in some cases. If you do not define this hook
4403 to return a nonzero value when it is required, the compiler will run out
4404 of spill registers and print a fatal error message.
4408 @subsection How Scalar Function Values Are Returned
4409 @cindex return values in registers
4410 @cindex values, returned by functions
4411 @cindex scalars, returned as values
4413 This section discusses the macros that control returning scalars as
4414 values---values that can fit in registers.
4416 @hook TARGET_FUNCTION_VALUE
4418 Define this to return an RTX representing the place where a function
4419 returns or receives a value of data type @var{ret_type}, a tree node
4420 representing a data type. @var{fn_decl_or_type} is a tree node
4421 representing @code{FUNCTION_DECL} or @code{FUNCTION_TYPE} of a
4422 function being called. If @var{outgoing} is false, the hook should
4423 compute the register in which the caller will see the return value.
4424 Otherwise, the hook should return an RTX representing the place where
4425 a function returns a value.
4427 On many machines, only @code{TYPE_MODE (@var{ret_type})} is relevant.
4428 (Actually, on most machines, scalar values are returned in the same
4429 place regardless of mode.) The value of the expression is usually a
4430 @code{reg} RTX for the hard register where the return value is stored.
4431 The value can also be a @code{parallel} RTX, if the return value is in
4432 multiple places. See @code{FUNCTION_ARG} for an explanation of the
4433 @code{parallel} form. Note that the callee will populate every
4434 location specified in the @code{parallel}, but if the first element of
4435 the @code{parallel} contains the whole return value, callers will use
4436 that element as the canonical location and ignore the others. The m68k
4437 port uses this type of @code{parallel} to return pointers in both
4438 @samp{%a0} (the canonical location) and @samp{%d0}.
4440 If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply
4441 the same promotion rules specified in @code{PROMOTE_MODE} if
4442 @var{valtype} is a scalar type.
4444 If the precise function being called is known, @var{func} is a tree
4445 node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
4446 pointer. This makes it possible to use a different value-returning
4447 convention for specific functions when all their calls are
4450 Some target machines have ``register windows'' so that the register in
4451 which a function returns its value is not the same as the one in which
4452 the caller sees the value. For such machines, you should return
4453 different RTX depending on @var{outgoing}.
4455 @code{TARGET_FUNCTION_VALUE} is not used for return values with
4456 aggregate data types, because these are returned in another way. See
4457 @code{TARGET_STRUCT_VALUE_RTX} and related macros, below.
4460 @defmac FUNCTION_VALUE (@var{valtype}, @var{func})
4461 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for
4462 a new target instead.
4465 @defmac LIBCALL_VALUE (@var{mode})
4466 A C expression to create an RTX representing the place where a library
4467 function returns a value of mode @var{mode}.
4469 Note that ``library function'' in this context means a compiler
4470 support routine, used to perform arithmetic, whose name is known
4471 specially by the compiler and was not mentioned in the C code being
4475 @hook TARGET_LIBCALL_VALUE
4476 Define this hook if the back-end needs to know the name of the libcall
4477 function in order to determine where the result should be returned.
4479 The mode of the result is given by @var{mode} and the name of the called
4480 library function is given by @var{fun}. The hook should return an RTX
4481 representing the place where the library function result will be returned.
4483 If this hook is not defined, then LIBCALL_VALUE will be used.
4486 @defmac FUNCTION_VALUE_REGNO_P (@var{regno})
4487 A C expression that is nonzero if @var{regno} is the number of a hard
4488 register in which the values of called function may come back.
4490 A register whose use for returning values is limited to serving as the
4491 second of a pair (for a value of type @code{double}, say) need not be
4492 recognized by this macro. So for most machines, this definition
4496 #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
4499 If the machine has register windows, so that the caller and the called
4500 function use different registers for the return value, this macro
4501 should recognize only the caller's register numbers.
4503 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE_REGNO_P}
4504 for a new target instead.
4507 @hook TARGET_FUNCTION_VALUE_REGNO_P
4508 A target hook that return @code{true} if @var{regno} is the number of a hard
4509 register in which the values of called function may come back.
4511 A register whose use for returning values is limited to serving as the
4512 second of a pair (for a value of type @code{double}, say) need not be
4513 recognized by this target hook.
4515 If the machine has register windows, so that the caller and the called
4516 function use different registers for the return value, this target hook
4517 should recognize only the caller's register numbers.
4519 If this hook is not defined, then FUNCTION_VALUE_REGNO_P will be used.
4522 @defmac APPLY_RESULT_SIZE
4523 Define this macro if @samp{untyped_call} and @samp{untyped_return}
4524 need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for
4525 saving and restoring an arbitrary return value.
4528 @hook TARGET_RETURN_IN_MSB
4529 This hook should return true if values of type @var{type} are returned
4530 at the most significant end of a register (in other words, if they are
4531 padded at the least significant end). You can assume that @var{type}
4532 is returned in a register; the caller is required to check this.
4534 Note that the register provided by @code{TARGET_FUNCTION_VALUE} must
4535 be able to hold the complete return value. For example, if a 1-, 2-
4536 or 3-byte structure is returned at the most significant end of a
4537 4-byte register, @code{TARGET_FUNCTION_VALUE} should provide an
4541 @node Aggregate Return
4542 @subsection How Large Values Are Returned
4543 @cindex aggregates as return values
4544 @cindex large return values
4545 @cindex returning aggregate values
4546 @cindex structure value address
4548 When a function value's mode is @code{BLKmode} (and in some other
4549 cases), the value is not returned according to
4550 @code{TARGET_FUNCTION_VALUE} (@pxref{Scalar Return}). Instead, the
4551 caller passes the address of a block of memory in which the value
4552 should be stored. This address is called the @dfn{structure value
4555 This section describes how to control returning structure values in
4558 @hook TARGET_RETURN_IN_MEMORY
4559 This target hook should return a nonzero value to say to return the
4560 function value in memory, just as large structures are always returned.
4561 Here @var{type} will be the data type of the value, and @var{fntype}
4562 will be the type of the function doing the returning, or @code{NULL} for
4565 Note that values of mode @code{BLKmode} must be explicitly handled
4566 by this function. Also, the option @option{-fpcc-struct-return}
4567 takes effect regardless of this macro. On most systems, it is
4568 possible to leave the hook undefined; this causes a default
4569 definition to be used, whose value is the constant 1 for @code{BLKmode}
4570 values, and 0 otherwise.
4572 Do not use this hook to indicate that structures and unions should always
4573 be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN}
4577 @defmac DEFAULT_PCC_STRUCT_RETURN
4578 Define this macro to be 1 if all structure and union return values must be
4579 in memory. Since this results in slower code, this should be defined
4580 only if needed for compatibility with other compilers or with an ABI@.
4581 If you define this macro to be 0, then the conventions used for structure
4582 and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY}
4585 If not defined, this defaults to the value 1.
4588 @hook TARGET_STRUCT_VALUE_RTX
4589 This target hook should return the location of the structure value
4590 address (normally a @code{mem} or @code{reg}), or 0 if the address is
4591 passed as an ``invisible'' first argument. Note that @var{fndecl} may
4592 be @code{NULL}, for libcalls. You do not need to define this target
4593 hook if the address is always passed as an ``invisible'' first
4596 On some architectures the place where the structure value address
4597 is found by the called function is not the same place that the
4598 caller put it. This can be due to register windows, or it could
4599 be because the function prologue moves it to a different place.
4600 @var{incoming} is @code{1} or @code{2} when the location is needed in
4601 the context of the called function, and @code{0} in the context of
4604 If @var{incoming} is nonzero and the address is to be found on the
4605 stack, return a @code{mem} which refers to the frame pointer. If
4606 @var{incoming} is @code{2}, the result is being used to fetch the
4607 structure value address at the beginning of a function. If you need
4608 to emit adjusting code, you should do it at this point.
4611 @defmac PCC_STATIC_STRUCT_RETURN
4612 Define this macro if the usual system convention on the target machine
4613 for returning structures and unions is for the called function to return
4614 the address of a static variable containing the value.
4616 Do not define this if the usual system convention is for the caller to
4617 pass an address to the subroutine.
4619 This macro has effect in @option{-fpcc-struct-return} mode, but it does
4620 nothing when you use @option{-freg-struct-return} mode.
4624 @subsection Caller-Saves Register Allocation
4626 If you enable it, GCC can save registers around function calls. This
4627 makes it possible to use call-clobbered registers to hold variables that
4628 must live across calls.
4630 @defmac CALLER_SAVE_PROFITABLE (@var{refs}, @var{calls})
4631 A C expression to determine whether it is worthwhile to consider placing
4632 a pseudo-register in a call-clobbered hard register and saving and
4633 restoring it around each function call. The expression should be 1 when
4634 this is worth doing, and 0 otherwise.
4636 If you don't define this macro, a default is used which is good on most
4637 machines: @code{4 * @var{calls} < @var{refs}}.
4640 @defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs})
4641 A C expression specifying which mode is required for saving @var{nregs}
4642 of a pseudo-register in call-clobbered hard register @var{regno}. If
4643 @var{regno} is unsuitable for caller save, @code{VOIDmode} should be
4644 returned. For most machines this macro need not be defined since GCC
4645 will select the smallest suitable mode.
4648 @node Function Entry
4649 @subsection Function Entry and Exit
4650 @cindex function entry and exit
4654 This section describes the macros that output function entry
4655 (@dfn{prologue}) and exit (@dfn{epilogue}) code.
4657 @hook TARGET_ASM_FUNCTION_PROLOGUE
4658 If defined, a function that outputs the assembler code for entry to a
4659 function. The prologue is responsible for setting up the stack frame,
4660 initializing the frame pointer register, saving registers that must be
4661 saved, and allocating @var{size} additional bytes of storage for the
4662 local variables. @var{size} is an integer. @var{file} is a stdio
4663 stream to which the assembler code should be output.
4665 The label for the beginning of the function need not be output by this
4666 macro. That has already been done when the macro is run.
4668 @findex regs_ever_live
4669 To determine which registers to save, the macro can refer to the array
4670 @code{regs_ever_live}: element @var{r} is nonzero if hard register
4671 @var{r} is used anywhere within the function. This implies the function
4672 prologue should save register @var{r}, provided it is not one of the
4673 call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use
4674 @code{regs_ever_live}.)
4676 On machines that have ``register windows'', the function entry code does
4677 not save on the stack the registers that are in the windows, even if
4678 they are supposed to be preserved by function calls; instead it takes
4679 appropriate steps to ``push'' the register stack, if any non-call-used
4680 registers are used in the function.
4682 @findex frame_pointer_needed
4683 On machines where functions may or may not have frame-pointers, the
4684 function entry code must vary accordingly; it must set up the frame
4685 pointer if one is wanted, and not otherwise. To determine whether a
4686 frame pointer is in wanted, the macro can refer to the variable
4687 @code{frame_pointer_needed}. The variable's value will be 1 at run
4688 time in a function that needs a frame pointer. @xref{Elimination}.
4690 The function entry code is responsible for allocating any stack space
4691 required for the function. This stack space consists of the regions
4692 listed below. In most cases, these regions are allocated in the
4693 order listed, with the last listed region closest to the top of the
4694 stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and
4695 the highest address if it is not defined). You can use a different order
4696 for a machine if doing so is more convenient or required for
4697 compatibility reasons. Except in cases where required by standard
4698 or by a debugger, there is no reason why the stack layout used by GCC
4699 need agree with that used by other compilers for a machine.
4702 @hook TARGET_ASM_FUNCTION_END_PROLOGUE
4703 If defined, a function that outputs assembler code at the end of a
4704 prologue. This should be used when the function prologue is being
4705 emitted as RTL, and you have some extra assembler that needs to be
4706 emitted. @xref{prologue instruction pattern}.
4709 @hook TARGET_ASM_FUNCTION_BEGIN_EPILOGUE
4710 If defined, a function that outputs assembler code at the start of an
4711 epilogue. This should be used when the function epilogue is being
4712 emitted as RTL, and you have some extra assembler that needs to be
4713 emitted. @xref{epilogue instruction pattern}.
4716 @hook TARGET_ASM_FUNCTION_EPILOGUE
4717 If defined, a function that outputs the assembler code for exit from a
4718 function. The epilogue is responsible for restoring the saved
4719 registers and stack pointer to their values when the function was
4720 called, and returning control to the caller. This macro takes the
4721 same arguments as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the
4722 registers to restore are determined from @code{regs_ever_live} and
4723 @code{CALL_USED_REGISTERS} in the same way.
4725 On some machines, there is a single instruction that does all the work
4726 of returning from the function. On these machines, give that
4727 instruction the name @samp{return} and do not define the macro
4728 @code{TARGET_ASM_FUNCTION_EPILOGUE} at all.
4730 Do not define a pattern named @samp{return} if you want the
4731 @code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target
4732 switches to control whether return instructions or epilogues are used,
4733 define a @samp{return} pattern with a validity condition that tests the
4734 target switches appropriately. If the @samp{return} pattern's validity
4735 condition is false, epilogues will be used.
4737 On machines where functions may or may not have frame-pointers, the
4738 function exit code must vary accordingly. Sometimes the code for these
4739 two cases is completely different. To determine whether a frame pointer
4740 is wanted, the macro can refer to the variable
4741 @code{frame_pointer_needed}. The variable's value will be 1 when compiling
4742 a function that needs a frame pointer.
4744 Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and
4745 @code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially.
4746 The C variable @code{current_function_is_leaf} is nonzero for such a
4747 function. @xref{Leaf Functions}.
4749 On some machines, some functions pop their arguments on exit while
4750 others leave that for the caller to do. For example, the 68020 when
4751 given @option{-mrtd} pops arguments in functions that take a fixed
4752 number of arguments.
4754 @findex current_function_pops_args
4755 Your definition of the macro @code{RETURN_POPS_ARGS} decides which
4756 functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE}
4757 needs to know what was decided. The number of bytes of the current
4758 function's arguments that this function should pop is available in
4759 @code{crtl->args.pops_args}. @xref{Scalar Return}.
4764 @findex current_function_pretend_args_size
4765 A region of @code{current_function_pretend_args_size} bytes of
4766 uninitialized space just underneath the first argument arriving on the
4767 stack. (This may not be at the very start of the allocated stack region
4768 if the calling sequence has pushed anything else since pushing the stack
4769 arguments. But usually, on such machines, nothing else has been pushed
4770 yet, because the function prologue itself does all the pushing.) This
4771 region is used on machines where an argument may be passed partly in
4772 registers and partly in memory, and, in some cases to support the
4773 features in @code{<stdarg.h>}.
4776 An area of memory used to save certain registers used by the function.
4777 The size of this area, which may also include space for such things as
4778 the return address and pointers to previous stack frames, is
4779 machine-specific and usually depends on which registers have been used
4780 in the function. Machines with register windows often do not require
4784 A region of at least @var{size} bytes, possibly rounded up to an allocation
4785 boundary, to contain the local variables of the function. On some machines,
4786 this region and the save area may occur in the opposite order, with the
4787 save area closer to the top of the stack.
4790 @cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames
4791 Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of
4792 @code{current_function_outgoing_args_size} bytes to be used for outgoing
4793 argument lists of the function. @xref{Stack Arguments}.
4796 @defmac EXIT_IGNORE_STACK
4797 Define this macro as a C expression that is nonzero if the return
4798 instruction or the function epilogue ignores the value of the stack
4799 pointer; in other words, if it is safe to delete an instruction to
4800 adjust the stack pointer before a return from the function. The
4803 Note that this macro's value is relevant only for functions for which
4804 frame pointers are maintained. It is never safe to delete a final
4805 stack adjustment in a function that has no frame pointer, and the
4806 compiler knows this regardless of @code{EXIT_IGNORE_STACK}.
4809 @defmac EPILOGUE_USES (@var{regno})
4810 Define this macro as a C expression that is nonzero for registers that are
4811 used by the epilogue or the @samp{return} pattern. The stack and frame
4812 pointer registers are already assumed to be used as needed.
4815 @defmac EH_USES (@var{regno})
4816 Define this macro as a C expression that is nonzero for registers that are
4817 used by the exception handling mechanism, and so should be considered live
4818 on entry to an exception edge.
4821 @defmac DELAY_SLOTS_FOR_EPILOGUE
4822 Define this macro if the function epilogue contains delay slots to which
4823 instructions from the rest of the function can be ``moved''. The
4824 definition should be a C expression whose value is an integer
4825 representing the number of delay slots there.
4828 @defmac ELIGIBLE_FOR_EPILOGUE_DELAY (@var{insn}, @var{n})
4829 A C expression that returns 1 if @var{insn} can be placed in delay
4830 slot number @var{n} of the epilogue.
4832 The argument @var{n} is an integer which identifies the delay slot now
4833 being considered (since different slots may have different rules of
4834 eligibility). It is never negative and is always less than the number
4835 of epilogue delay slots (what @code{DELAY_SLOTS_FOR_EPILOGUE} returns).
4836 If you reject a particular insn for a given delay slot, in principle, it
4837 may be reconsidered for a subsequent delay slot. Also, other insns may
4838 (at least in principle) be considered for the so far unfilled delay
4841 @findex current_function_epilogue_delay_list
4842 @findex final_scan_insn
4843 The insns accepted to fill the epilogue delay slots are put in an RTL
4844 list made with @code{insn_list} objects, stored in the variable
4845 @code{current_function_epilogue_delay_list}. The insn for the first
4846 delay slot comes first in the list. Your definition of the macro
4847 @code{TARGET_ASM_FUNCTION_EPILOGUE} should fill the delay slots by
4848 outputting the insns in this list, usually by calling
4849 @code{final_scan_insn}.
4851 You need not define this macro if you did not define
4852 @code{DELAY_SLOTS_FOR_EPILOGUE}.
4855 @hook TARGET_ASM_OUTPUT_MI_THUNK
4856 A function that outputs the assembler code for a thunk
4857 function, used to implement C++ virtual function calls with multiple
4858 inheritance. The thunk acts as a wrapper around a virtual function,
4859 adjusting the implicit object parameter before handing control off to
4862 First, emit code to add the integer @var{delta} to the location that
4863 contains the incoming first argument. Assume that this argument
4864 contains a pointer, and is the one used to pass the @code{this} pointer
4865 in C++. This is the incoming argument @emph{before} the function prologue,
4866 e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of
4867 all other incoming arguments.
4869 Then, if @var{vcall_offset} is nonzero, an additional adjustment should be
4870 made after adding @code{delta}. In particular, if @var{p} is the
4871 adjusted pointer, the following adjustment should be made:
4874 p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)]
4877 After the additions, emit code to jump to @var{function}, which is a
4878 @code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does
4879 not touch the return address. Hence returning from @var{FUNCTION} will
4880 return to whoever called the current @samp{thunk}.
4882 The effect must be as if @var{function} had been called directly with
4883 the adjusted first argument. This macro is responsible for emitting all
4884 of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE}
4885 and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked.
4887 The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function}
4888 have already been extracted from it.) It might possibly be useful on
4889 some targets, but probably not.
4891 If you do not define this macro, the target-independent code in the C++
4892 front end will generate a less efficient heavyweight thunk that calls
4893 @var{function} instead of jumping to it. The generic approach does
4894 not support varargs.
4897 @hook TARGET_ASM_CAN_OUTPUT_MI_THUNK
4898 A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able
4899 to output the assembler code for the thunk function specified by the
4900 arguments it is passed, and false otherwise. In the latter case, the
4901 generic approach will be used by the C++ front end, with the limitations
4906 @subsection Generating Code for Profiling
4907 @cindex profiling, code generation
4909 These macros will help you generate code for profiling.
4911 @defmac FUNCTION_PROFILER (@var{file}, @var{labelno})
4912 A C statement or compound statement to output to @var{file} some
4913 assembler code to call the profiling subroutine @code{mcount}.
4916 The details of how @code{mcount} expects to be called are determined by
4917 your operating system environment, not by GCC@. To figure them out,
4918 compile a small program for profiling using the system's installed C
4919 compiler and look at the assembler code that results.
4921 Older implementations of @code{mcount} expect the address of a counter
4922 variable to be loaded into some register. The name of this variable is
4923 @samp{LP} followed by the number @var{labelno}, so you would generate
4924 the name using @samp{LP%d} in a @code{fprintf}.
4927 @defmac PROFILE_HOOK
4928 A C statement or compound statement to output to @var{file} some assembly
4929 code to call the profiling subroutine @code{mcount} even the target does
4930 not support profiling.
4933 @defmac NO_PROFILE_COUNTERS
4934 Define this macro to be an expression with a nonzero value if the
4935 @code{mcount} subroutine on your system does not need a counter variable
4936 allocated for each function. This is true for almost all modern
4937 implementations. If you define this macro, you must not use the
4938 @var{labelno} argument to @code{FUNCTION_PROFILER}.
4941 @defmac PROFILE_BEFORE_PROLOGUE
4942 Define this macro if the code for function profiling should come before
4943 the function prologue. Normally, the profiling code comes after.
4947 @subsection Permitting tail calls
4950 @hook TARGET_FUNCTION_OK_FOR_SIBCALL
4951 True if it is ok to do sibling call optimization for the specified
4952 call expression @var{exp}. @var{decl} will be the called function,
4953 or @code{NULL} if this is an indirect call.
4955 It is not uncommon for limitations of calling conventions to prevent
4956 tail calls to functions outside the current unit of translation, or
4957 during PIC compilation. The hook is used to enforce these restrictions,
4958 as the @code{sibcall} md pattern can not fail, or fall over to a
4959 ``normal'' call. The criteria for successful sibling call optimization
4960 may vary greatly between different architectures.
4963 @hook TARGET_EXTRA_LIVE_ON_ENTRY
4964 Add any hard registers to @var{regs} that are live on entry to the
4965 function. This hook only needs to be defined to provide registers that
4966 cannot be found by examination of FUNCTION_ARG_REGNO_P, the callee saved
4967 registers, STATIC_CHAIN_INCOMING_REGNUM, STATIC_CHAIN_REGNUM,
4968 TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM, EH_USES,
4969 FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM.
4972 @node Stack Smashing Protection
4973 @subsection Stack smashing protection
4974 @cindex stack smashing protection
4976 @hook TARGET_STACK_PROTECT_GUARD
4977 This hook returns a @code{DECL} node for the external variable to use
4978 for the stack protection guard. This variable is initialized by the
4979 runtime to some random value and is used to initialize the guard value
4980 that is placed at the top of the local stack frame. The type of this
4981 variable must be @code{ptr_type_node}.
4983 The default version of this hook creates a variable called
4984 @samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}.
4987 @hook TARGET_STACK_PROTECT_FAIL
4988 This hook returns a tree expression that alerts the runtime that the
4989 stack protect guard variable has been modified. This expression should
4990 involve a call to a @code{noreturn} function.
4992 The default version of this hook invokes a function called
4993 @samp{__stack_chk_fail}, taking no arguments. This function is
4994 normally defined in @file{libgcc2.c}.
4997 @hook TARGET_SUPPORTS_SPLIT_STACK
5000 @section Implementing the Varargs Macros
5001 @cindex varargs implementation
5003 GCC comes with an implementation of @code{<varargs.h>} and
5004 @code{<stdarg.h>} that work without change on machines that pass arguments
5005 on the stack. Other machines require their own implementations of
5006 varargs, and the two machine independent header files must have
5007 conditionals to include it.
5009 ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in
5010 the calling convention for @code{va_start}. The traditional
5011 implementation takes just one argument, which is the variable in which
5012 to store the argument pointer. The ISO implementation of
5013 @code{va_start} takes an additional second argument. The user is
5014 supposed to write the last named argument of the function here.
5016 However, @code{va_start} should not use this argument. The way to find
5017 the end of the named arguments is with the built-in functions described
5020 @defmac __builtin_saveregs ()
5021 Use this built-in function to save the argument registers in memory so
5022 that the varargs mechanism can access them. Both ISO and traditional
5023 versions of @code{va_start} must use @code{__builtin_saveregs}, unless
5024 you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead.
5026 On some machines, @code{__builtin_saveregs} is open-coded under the
5027 control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On
5028 other machines, it calls a routine written in assembler language,
5029 found in @file{libgcc2.c}.
5031 Code generated for the call to @code{__builtin_saveregs} appears at the
5032 beginning of the function, as opposed to where the call to
5033 @code{__builtin_saveregs} is written, regardless of what the code is.
5034 This is because the registers must be saved before the function starts
5035 to use them for its own purposes.
5036 @c i rewrote the first sentence above to fix an overfull hbox. --mew
5040 @defmac __builtin_next_arg (@var{lastarg})
5041 This builtin returns the address of the first anonymous stack
5042 argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it
5043 returns the address of the location above the first anonymous stack
5044 argument. Use it in @code{va_start} to initialize the pointer for
5045 fetching arguments from the stack. Also use it in @code{va_start} to
5046 verify that the second parameter @var{lastarg} is the last named argument
5047 of the current function.
5050 @defmac __builtin_classify_type (@var{object})
5051 Since each machine has its own conventions for which data types are
5052 passed in which kind of register, your implementation of @code{va_arg}
5053 has to embody these conventions. The easiest way to categorize the
5054 specified data type is to use @code{__builtin_classify_type} together
5055 with @code{sizeof} and @code{__alignof__}.
5057 @code{__builtin_classify_type} ignores the value of @var{object},
5058 considering only its data type. It returns an integer describing what
5059 kind of type that is---integer, floating, pointer, structure, and so on.
5061 The file @file{typeclass.h} defines an enumeration that you can use to
5062 interpret the values of @code{__builtin_classify_type}.
5065 These machine description macros help implement varargs:
5067 @hook TARGET_EXPAND_BUILTIN_SAVEREGS
5068 If defined, this hook produces the machine-specific code for a call to
5069 @code{__builtin_saveregs}. This code will be moved to the very
5070 beginning of the function, before any parameter access are made. The
5071 return value of this function should be an RTX that contains the value
5072 to use as the return of @code{__builtin_saveregs}.
5075 @hook TARGET_SETUP_INCOMING_VARARGS
5076 This target hook offers an alternative to using
5077 @code{__builtin_saveregs} and defining the hook
5078 @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous
5079 register arguments into the stack so that all the arguments appear to
5080 have been passed consecutively on the stack. Once this is done, you can
5081 use the standard implementation of varargs that works for machines that
5082 pass all their arguments on the stack.
5084 The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data
5085 structure, containing the values that are obtained after processing the
5086 named arguments. The arguments @var{mode} and @var{type} describe the
5087 last named argument---its machine mode and its data type as a tree node.
5089 The target hook should do two things: first, push onto the stack all the
5090 argument registers @emph{not} used for the named arguments, and second,
5091 store the size of the data thus pushed into the @code{int}-valued
5092 variable pointed to by @var{pretend_args_size}. The value that you
5093 store here will serve as additional offset for setting up the stack
5096 Because you must generate code to push the anonymous arguments at
5097 compile time without knowing their data types,
5098 @code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that
5099 have just a single category of argument register and use it uniformly
5102 If the argument @var{second_time} is nonzero, it means that the
5103 arguments of the function are being analyzed for the second time. This
5104 happens for an inline function, which is not actually compiled until the
5105 end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should
5106 not generate any instructions in this case.
5109 @hook TARGET_STRICT_ARGUMENT_NAMING
5110 Define this hook to return @code{true} if the location where a function
5111 argument is passed depends on whether or not it is a named argument.
5113 This hook controls how the @var{named} argument to @code{FUNCTION_ARG}
5114 is set for varargs and stdarg functions. If this hook returns
5115 @code{true}, the @var{named} argument is always true for named
5116 arguments, and false for unnamed arguments. If it returns @code{false},
5117 but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true},
5118 then all arguments are treated as named. Otherwise, all named arguments
5119 except the last are treated as named.
5121 You need not define this hook if it always returns @code{false}.
5124 @hook TARGET_PRETEND_OUTGOING_VARARGS_NAMED
5125 If you need to conditionally change ABIs so that one works with
5126 @code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither
5127 @code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was
5128 defined, then define this hook to return @code{true} if
5129 @code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise.
5130 Otherwise, you should not define this hook.
5134 @section Trampolines for Nested Functions
5135 @cindex trampolines for nested functions
5136 @cindex nested functions, trampolines for
5138 A @dfn{trampoline} is a small piece of code that is created at run time
5139 when the address of a nested function is taken. It normally resides on
5140 the stack, in the stack frame of the containing function. These macros
5141 tell GCC how to generate code to allocate and initialize a
5144 The instructions in the trampoline must do two things: load a constant
5145 address into the static chain register, and jump to the real address of
5146 the nested function. On CISC machines such as the m68k, this requires
5147 two instructions, a move immediate and a jump. Then the two addresses
5148 exist in the trampoline as word-long immediate operands. On RISC
5149 machines, it is often necessary to load each address into a register in
5150 two parts. Then pieces of each address form separate immediate
5153 The code generated to initialize the trampoline must store the variable
5154 parts---the static chain value and the function address---into the
5155 immediate operands of the instructions. On a CISC machine, this is
5156 simply a matter of copying each address to a memory reference at the
5157 proper offset from the start of the trampoline. On a RISC machine, it
5158 may be necessary to take out pieces of the address and store them
5161 @hook TARGET_ASM_TRAMPOLINE_TEMPLATE
5162 This hook is called by @code{assemble_trampoline_template} to output,
5163 on the stream @var{f}, assembler code for a block of data that contains
5164 the constant parts of a trampoline. This code should not include a
5165 label---the label is taken care of automatically.
5167 If you do not define this hook, it means no template is needed
5168 for the target. Do not define this hook on systems where the block move
5169 code to copy the trampoline into place would be larger than the code
5170 to generate it on the spot.
5173 @defmac TRAMPOLINE_SECTION
5174 Return the section into which the trampoline template is to be placed
5175 (@pxref{Sections}). The default value is @code{readonly_data_section}.
5178 @defmac TRAMPOLINE_SIZE
5179 A C expression for the size in bytes of the trampoline, as an integer.
5182 @defmac TRAMPOLINE_ALIGNMENT
5183 Alignment required for trampolines, in bits.
5185 If you don't define this macro, the value of @code{FUNCTION_ALIGNMENT}
5186 is used for aligning trampolines.
5189 @hook TARGET_TRAMPOLINE_INIT
5190 This hook is called to initialize a trampoline.
5191 @var{m_tramp} is an RTX for the memory block for the trampoline; @var{fndecl}
5192 is the @code{FUNCTION_DECL} for the nested function; @var{static_chain} is an
5193 RTX for the static chain value that should be passed to the function
5196 If the target defines @code{TARGET_ASM_TRAMPOLINE_TEMPLATE}, then the
5197 first thing this hook should do is emit a block move into @var{m_tramp}
5198 from the memory block returned by @code{assemble_trampoline_template}.
5199 Note that the block move need only cover the constant parts of the
5200 trampoline. If the target isolates the variable parts of the trampoline
5201 to the end, not all @code{TRAMPOLINE_SIZE} bytes need be copied.
5203 If the target requires any other actions, such as flushing caches or
5204 enabling stack execution, these actions should be performed after
5205 initializing the trampoline proper.
5208 @hook TARGET_TRAMPOLINE_ADJUST_ADDRESS
5209 This hook should perform any machine-specific adjustment in
5210 the address of the trampoline. Its argument contains the address of the
5211 memory block that was passed to @code{TARGET_TRAMPOLINE_INIT}. In case
5212 the address to be used for a function call should be different from the
5213 address at which the template was stored, the different address should
5214 be returned; otherwise @var{addr} should be returned unchanged.
5215 If this hook is not defined, @var{addr} will be used for function calls.
5218 Implementing trampolines is difficult on many machines because they have
5219 separate instruction and data caches. Writing into a stack location
5220 fails to clear the memory in the instruction cache, so when the program
5221 jumps to that location, it executes the old contents.
5223 Here are two possible solutions. One is to clear the relevant parts of
5224 the instruction cache whenever a trampoline is set up. The other is to
5225 make all trampolines identical, by having them jump to a standard
5226 subroutine. The former technique makes trampoline execution faster; the
5227 latter makes initialization faster.
5229 To clear the instruction cache when a trampoline is initialized, define
5230 the following macro.
5232 @defmac CLEAR_INSN_CACHE (@var{beg}, @var{end})
5233 If defined, expands to a C expression clearing the @emph{instruction
5234 cache} in the specified interval. The definition of this macro would
5235 typically be a series of @code{asm} statements. Both @var{beg} and
5236 @var{end} are both pointer expressions.
5239 The operating system may also require the stack to be made executable
5240 before calling the trampoline. To implement this requirement, define
5241 the following macro.
5243 @defmac ENABLE_EXECUTE_STACK
5244 Define this macro if certain operations must be performed before executing
5245 code located on the stack. The macro should expand to a series of C
5246 file-scope constructs (e.g.@: functions) and provide a unique entry point
5247 named @code{__enable_execute_stack}. The target is responsible for
5248 emitting calls to the entry point in the code, for example from the
5249 @code{TARGET_TRAMPOLINE_INIT} hook.
5252 To use a standard subroutine, define the following macro. In addition,
5253 you must make sure that the instructions in a trampoline fill an entire
5254 cache line with identical instructions, or else ensure that the
5255 beginning of the trampoline code is always aligned at the same point in
5256 its cache line. Look in @file{m68k.h} as a guide.
5258 @defmac TRANSFER_FROM_TRAMPOLINE
5259 Define this macro if trampolines need a special subroutine to do their
5260 work. The macro should expand to a series of @code{asm} statements
5261 which will be compiled with GCC@. They go in a library function named
5262 @code{__transfer_from_trampoline}.
5264 If you need to avoid executing the ordinary prologue code of a compiled
5265 C function when you jump to the subroutine, you can do so by placing a
5266 special label of your own in the assembler code. Use one @code{asm}
5267 statement to generate an assembler label, and another to make the label
5268 global. Then trampolines can use that label to jump directly to your
5269 special assembler code.
5273 @section Implicit Calls to Library Routines
5274 @cindex library subroutine names
5275 @cindex @file{libgcc.a}
5277 @c prevent bad page break with this line
5278 Here is an explanation of implicit calls to library routines.
5280 @defmac DECLARE_LIBRARY_RENAMES
5281 This macro, if defined, should expand to a piece of C code that will get
5282 expanded when compiling functions for libgcc.a. It can be used to
5283 provide alternate names for GCC's internal library functions if there
5284 are ABI-mandated names that the compiler should provide.
5287 @findex set_optab_libfunc
5288 @findex init_one_libfunc
5289 @hook TARGET_INIT_LIBFUNCS
5290 This hook should declare additional library routines or rename
5291 existing ones, using the functions @code{set_optab_libfunc} and
5292 @code{init_one_libfunc} defined in @file{optabs.c}.
5293 @code{init_optabs} calls this macro after initializing all the normal
5296 The default is to do nothing. Most ports don't need to define this hook.
5299 @defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison})
5300 This macro should return @code{true} if the library routine that
5301 implements the floating point comparison operator @var{comparison} in
5302 mode @var{mode} will return a boolean, and @var{false} if it will
5305 GCC's own floating point libraries return tristates from the
5306 comparison operators, so the default returns false always. Most ports
5307 don't need to define this macro.
5310 @defmac TARGET_LIB_INT_CMP_BIASED
5311 This macro should evaluate to @code{true} if the integer comparison
5312 functions (like @code{__cmpdi2}) return 0 to indicate that the first
5313 operand is smaller than the second, 1 to indicate that they are equal,
5314 and 2 to indicate that the first operand is greater than the second.
5315 If this macro evaluates to @code{false} the comparison functions return
5316 @minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines
5317 in @file{libgcc.a}, you do not need to define this macro.
5320 @cindex US Software GOFAST, floating point emulation library
5321 @cindex floating point emulation library, US Software GOFAST
5322 @cindex GOFAST, floating point emulation library
5323 @findex gofast_maybe_init_libfuncs
5324 @defmac US_SOFTWARE_GOFAST
5325 Define this macro if your system C library uses the US Software GOFAST
5326 library to provide floating point emulation.
5328 In addition to defining this macro, your architecture must set
5329 @code{TARGET_INIT_LIBFUNCS} to @code{gofast_maybe_init_libfuncs}, or
5330 else call that function from its version of that hook. It is defined
5331 in @file{config/gofast.h}, which must be included by your
5332 architecture's @file{@var{cpu}.c} file. See @file{sparc/sparc.c} for
5335 If this macro is defined, the
5336 @code{TARGET_FLOAT_LIB_COMPARE_RETURNS_BOOL} target hook must return
5337 false for @code{SFmode} and @code{DFmode} comparisons.
5340 @cindex @code{EDOM}, implicit usage
5343 The value of @code{EDOM} on the target machine, as a C integer constant
5344 expression. If you don't define this macro, GCC does not attempt to
5345 deposit the value of @code{EDOM} into @code{errno} directly. Look in
5346 @file{/usr/include/errno.h} to find the value of @code{EDOM} on your
5349 If you do not define @code{TARGET_EDOM}, then compiled code reports
5350 domain errors by calling the library function and letting it report the
5351 error. If mathematical functions on your system use @code{matherr} when
5352 there is an error, then you should leave @code{TARGET_EDOM} undefined so
5353 that @code{matherr} is used normally.
5356 @cindex @code{errno}, implicit usage
5357 @defmac GEN_ERRNO_RTX
5358 Define this macro as a C expression to create an rtl expression that
5359 refers to the global ``variable'' @code{errno}. (On certain systems,
5360 @code{errno} may not actually be a variable.) If you don't define this
5361 macro, a reasonable default is used.
5364 @cindex C99 math functions, implicit usage
5365 @defmac TARGET_C99_FUNCTIONS
5366 When this macro is nonzero, GCC will implicitly optimize @code{sin} calls into
5367 @code{sinf} and similarly for other functions defined by C99 standard. The
5368 default is zero because a number of existing systems lack support for these
5369 functions in their runtime so this macro needs to be redefined to one on
5370 systems that do support the C99 runtime.
5373 @cindex sincos math function, implicit usage
5374 @defmac TARGET_HAS_SINCOS
5375 When this macro is nonzero, GCC will implicitly optimize calls to @code{sin}
5376 and @code{cos} with the same argument to a call to @code{sincos}. The
5377 default is zero. The target has to provide the following functions:
5379 void sincos(double x, double *sin, double *cos);
5380 void sincosf(float x, float *sin, float *cos);
5381 void sincosl(long double x, long double *sin, long double *cos);
5385 @defmac NEXT_OBJC_RUNTIME
5386 Define this macro to generate code for Objective-C message sending using
5387 the calling convention of the NeXT system. This calling convention
5388 involves passing the object, the selector and the method arguments all
5389 at once to the method-lookup library function.
5391 The default calling convention passes just the object and the selector
5392 to the lookup function, which returns a pointer to the method.
5395 @node Addressing Modes
5396 @section Addressing Modes
5397 @cindex addressing modes
5399 @c prevent bad page break with this line
5400 This is about addressing modes.
5402 @defmac HAVE_PRE_INCREMENT
5403 @defmacx HAVE_PRE_DECREMENT
5404 @defmacx HAVE_POST_INCREMENT
5405 @defmacx HAVE_POST_DECREMENT
5406 A C expression that is nonzero if the machine supports pre-increment,
5407 pre-decrement, post-increment, or post-decrement addressing respectively.
5410 @defmac HAVE_PRE_MODIFY_DISP
5411 @defmacx HAVE_POST_MODIFY_DISP
5412 A C expression that is nonzero if the machine supports pre- or
5413 post-address side-effect generation involving constants other than
5414 the size of the memory operand.
5417 @defmac HAVE_PRE_MODIFY_REG
5418 @defmacx HAVE_POST_MODIFY_REG
5419 A C expression that is nonzero if the machine supports pre- or
5420 post-address side-effect generation involving a register displacement.
5423 @defmac CONSTANT_ADDRESS_P (@var{x})
5424 A C expression that is 1 if the RTX @var{x} is a constant which
5425 is a valid address. On most machines the default definition of
5426 @code{(CONSTANT_P (@var{x}) && GET_CODE (@var{x}) != CONST_DOUBLE)}
5427 is acceptable, but a few machines are more restrictive as to which
5428 constant addresses are supported.
5431 @defmac CONSTANT_P (@var{x})
5432 @code{CONSTANT_P}, which is defined by target-independent code,
5433 accepts integer-values expressions whose values are not explicitly
5434 known, such as @code{symbol_ref}, @code{label_ref}, and @code{high}
5435 expressions and @code{const} arithmetic expressions, in addition to
5436 @code{const_int} and @code{const_double} expressions.
5439 @defmac MAX_REGS_PER_ADDRESS
5440 A number, the maximum number of registers that can appear in a valid
5441 memory address. Note that it is up to you to specify a value equal to
5442 the maximum number that @code{TARGET_LEGITIMATE_ADDRESS_P} would ever
5446 @hook TARGET_LEGITIMATE_ADDRESS_P
5447 A function that returns whether @var{x} (an RTX) is a legitimate memory
5448 address on the target machine for a memory operand of mode @var{mode}.
5450 Legitimate addresses are defined in two variants: a strict variant and a
5451 non-strict one. The @var{strict} parameter chooses which variant is
5452 desired by the caller.
5454 The strict variant is used in the reload pass. It must be defined so
5455 that any pseudo-register that has not been allocated a hard register is
5456 considered a memory reference. This is because in contexts where some
5457 kind of register is required, a pseudo-register with no hard register
5458 must be rejected. For non-hard registers, the strict variant should look
5459 up the @code{reg_renumber} array; it should then proceed using the hard
5460 register number in the array, or treat the pseudo as a memory reference
5461 if the array holds @code{-1}.
5463 The non-strict variant is used in other passes. It must be defined to
5464 accept all pseudo-registers in every context where some kind of
5465 register is required.
5467 Normally, constant addresses which are the sum of a @code{symbol_ref}
5468 and an integer are stored inside a @code{const} RTX to mark them as
5469 constant. Therefore, there is no need to recognize such sums
5470 specifically as legitimate addresses. Normally you would simply
5471 recognize any @code{const} as legitimate.
5473 Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
5474 sums that are not marked with @code{const}. It assumes that a naked
5475 @code{plus} indicates indexing. If so, then you @emph{must} reject such
5476 naked constant sums as illegitimate addresses, so that none of them will
5477 be given to @code{PRINT_OPERAND_ADDRESS}.
5479 @cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation
5480 On some machines, whether a symbolic address is legitimate depends on
5481 the section that the address refers to. On these machines, define the
5482 target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information
5483 into the @code{symbol_ref}, and then check for it here. When you see a
5484 @code{const}, you will have to look inside it to find the
5485 @code{symbol_ref} in order to determine the section. @xref{Assembler
5488 @cindex @code{GO_IF_LEGITIMATE_ADDRESS}
5489 Some ports are still using a deprecated legacy substitute for
5490 this hook, the @code{GO_IF_LEGITIMATE_ADDRESS} macro. This macro
5494 #define GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
5498 and should @code{goto @var{label}} if the address @var{x} is a valid
5499 address on the target machine for a memory operand of mode @var{mode}.
5500 Whether the strict or non-strict variants are desired is defined by
5501 the @code{REG_OK_STRICT} macro introduced earlier in this section.
5502 Using the hook is usually simpler because it limits the number of
5503 files that are recompiled when changes are made.
5506 @defmac TARGET_MEM_CONSTRAINT
5507 A single character to be used instead of the default @code{'m'}
5508 character for general memory addresses. This defines the constraint
5509 letter which matches the memory addresses accepted by
5510 @code{TARGET_LEGITIMATE_ADDRESS_P}. Define this macro if you want to
5511 support new address formats in your back end without changing the
5512 semantics of the @code{'m'} constraint. This is necessary in order to
5513 preserve functionality of inline assembly constructs using the
5514 @code{'m'} constraint.
5517 @defmac FIND_BASE_TERM (@var{x})
5518 A C expression to determine the base term of address @var{x},
5519 or to provide a simplified version of @var{x} from which @file{alias.c}
5520 can easily find the base term. This macro is used in only two places:
5521 @code{find_base_value} and @code{find_base_term} in @file{alias.c}.
5523 It is always safe for this macro to not be defined. It exists so
5524 that alias analysis can understand machine-dependent addresses.
5526 The typical use of this macro is to handle addresses containing
5527 a label_ref or symbol_ref within an UNSPEC@.
5530 @hook TARGET_LEGITIMIZE_ADDRESS
5531 This hook is given an invalid memory address @var{x} for an
5532 operand of mode @var{mode} and should try to return a valid memory
5535 @findex break_out_memory_refs
5536 @var{x} will always be the result of a call to @code{break_out_memory_refs},
5537 and @var{oldx} will be the operand that was given to that function to produce
5540 The code of the hook should not alter the substructure of
5541 @var{x}. If it transforms @var{x} into a more legitimate form, it
5542 should return the new @var{x}.
5544 It is not necessary for this hook to come up with a legitimate address.
5545 The compiler has standard ways of doing so in all cases. In fact, it
5546 is safe to omit this hook or make it return @var{x} if it cannot find
5547 a valid way to legitimize the address. But often a machine-dependent
5548 strategy can generate better code.
5551 @defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win})
5552 A C compound statement that attempts to replace @var{x}, which is an address
5553 that needs reloading, with a valid memory address for an operand of mode
5554 @var{mode}. @var{win} will be a C statement label elsewhere in the code.
5555 It is not necessary to define this macro, but it might be useful for
5556 performance reasons.
5558 For example, on the i386, it is sometimes possible to use a single
5559 reload register instead of two by reloading a sum of two pseudo
5560 registers into a register. On the other hand, for number of RISC
5561 processors offsets are limited so that often an intermediate address
5562 needs to be generated in order to address a stack slot. By defining
5563 @code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses
5564 generated for adjacent some stack slots can be made identical, and thus
5567 @emph{Note}: This macro should be used with caution. It is necessary
5568 to know something of how reload works in order to effectively use this,
5569 and it is quite easy to produce macros that build in too much knowledge
5570 of reload internals.
5572 @emph{Note}: This macro must be able to reload an address created by a
5573 previous invocation of this macro. If it fails to handle such addresses
5574 then the compiler may generate incorrect code or abort.
5577 The macro definition should use @code{push_reload} to indicate parts that
5578 need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually
5579 suitable to be passed unaltered to @code{push_reload}.
5581 The code generated by this macro must not alter the substructure of
5582 @var{x}. If it transforms @var{x} into a more legitimate form, it
5583 should assign @var{x} (which will always be a C variable) a new value.
5584 This also applies to parts that you change indirectly by calling
5587 @findex strict_memory_address_p
5588 The macro definition may use @code{strict_memory_address_p} to test if
5589 the address has become legitimate.
5592 If you want to change only a part of @var{x}, one standard way of doing
5593 this is to use @code{copy_rtx}. Note, however, that it unshares only a
5594 single level of rtl. Thus, if the part to be changed is not at the
5595 top level, you'll need to replace first the top level.
5596 It is not necessary for this macro to come up with a legitimate
5597 address; but often a machine-dependent strategy can generate better code.
5600 @hook TARGET_MODE_DEPENDENT_ADDRESS_P
5601 This hook returns @code{true} if memory address @var{addr} can have
5602 different meanings depending on the machine mode of the memory
5603 reference it is used for or if the address is valid for some modes
5606 Autoincrement and autodecrement addresses typically have mode-dependent
5607 effects because the amount of the increment or decrement is the size
5608 of the operand being addressed. Some machines have other mode-dependent
5609 addresses. Many RISC machines have no mode-dependent addresses.
5611 You may assume that @var{addr} is a valid address for the machine.
5613 The default version of this hook returns @code{false}.
5616 @defmac GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
5617 A C statement or compound statement with a conditional @code{goto
5618 @var{label};} executed if memory address @var{x} (an RTX) can have
5619 different meanings depending on the machine mode of the memory
5620 reference it is used for or if the address is valid for some modes
5623 Autoincrement and autodecrement addresses typically have mode-dependent
5624 effects because the amount of the increment or decrement is the size
5625 of the operand being addressed. Some machines have other mode-dependent
5626 addresses. Many RISC machines have no mode-dependent addresses.
5628 You may assume that @var{addr} is a valid address for the machine.
5630 These are obsolete macros, replaced by the
5631 @code{TARGET_MODE_DEPENDENT_ADDRESS_P} target hook.
5634 @defmac LEGITIMATE_CONSTANT_P (@var{x})
5635 A C expression that is nonzero if @var{x} is a legitimate constant for
5636 an immediate operand on the target machine. You can assume that
5637 @var{x} satisfies @code{CONSTANT_P}, so you need not check this. In fact,
5638 @samp{1} is a suitable definition for this macro on machines where
5639 anything @code{CONSTANT_P} is valid.
5642 @hook TARGET_DELEGITIMIZE_ADDRESS
5643 This hook is used to undo the possibly obfuscating effects of the
5644 @code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target
5645 macros. Some backend implementations of these macros wrap symbol
5646 references inside an @code{UNSPEC} rtx to represent PIC or similar
5647 addressing modes. This target hook allows GCC's optimizers to understand
5648 the semantics of these opaque @code{UNSPEC}s by converting them back
5649 into their original form.
5652 @hook TARGET_CANNOT_FORCE_CONST_MEM
5653 This hook should return true if @var{x} is of a form that cannot (or
5654 should not) be spilled to the constant pool. The default version of
5655 this hook returns false.
5657 The primary reason to define this hook is to prevent reload from
5658 deciding that a non-legitimate constant would be better reloaded
5659 from the constant pool instead of spilling and reloading a register
5660 holding the constant. This restriction is often true of addresses
5661 of TLS symbols for various targets.
5664 @hook TARGET_USE_BLOCKS_FOR_CONSTANT_P
5665 This hook should return true if pool entries for constant @var{x} can
5666 be placed in an @code{object_block} structure. @var{mode} is the mode
5669 The default version returns false for all constants.
5672 @hook TARGET_BUILTIN_RECIPROCAL
5673 This hook should return the DECL of a function that implements reciprocal of
5674 the builtin function with builtin function code @var{fn}, or
5675 @code{NULL_TREE} if such a function is not available. @var{md_fn} is true
5676 when @var{fn} is a code of a machine-dependent builtin function. When
5677 @var{sqrt} is true, additional optimizations that apply only to the reciprocal
5678 of a square root function are performed, and only reciprocals of @code{sqrt}
5682 @hook TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD
5683 This hook should return the DECL of a function @var{f} that given an
5684 address @var{addr} as an argument returns a mask @var{m} that can be
5685 used to extract from two vectors the relevant data that resides in
5686 @var{addr} in case @var{addr} is not properly aligned.
5688 The autovectorizer, when vectorizing a load operation from an address
5689 @var{addr} that may be unaligned, will generate two vector loads from
5690 the two aligned addresses around @var{addr}. It then generates a
5691 @code{REALIGN_LOAD} operation to extract the relevant data from the
5692 two loaded vectors. The first two arguments to @code{REALIGN_LOAD},
5693 @var{v1} and @var{v2}, are the two vectors, each of size @var{VS}, and
5694 the third argument, @var{OFF}, defines how the data will be extracted
5695 from these two vectors: if @var{OFF} is 0, then the returned vector is
5696 @var{v2}; otherwise, the returned vector is composed from the last
5697 @var{VS}-@var{OFF} elements of @var{v1} concatenated to the first
5698 @var{OFF} elements of @var{v2}.
5700 If this hook is defined, the autovectorizer will generate a call
5701 to @var{f} (using the DECL tree that this hook returns) and will
5702 use the return value of @var{f} as the argument @var{OFF} to
5703 @code{REALIGN_LOAD}. Therefore, the mask @var{m} returned by @var{f}
5704 should comply with the semantics expected by @code{REALIGN_LOAD}
5706 If this hook is not defined, then @var{addr} will be used as
5707 the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low
5708 log2(@var{VS}) @minus{} 1 bits of @var{addr} will be considered.
5711 @hook TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN
5712 This hook should return the DECL of a function @var{f} that implements
5713 widening multiplication of the even elements of two input vectors of type @var{x}.
5715 If this hook is defined, the autovectorizer will use it along with the
5716 @code{TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD} target hook when vectorizing
5717 widening multiplication in cases that the order of the results does not have to be
5718 preserved (e.g.@: used only by a reduction computation). Otherwise, the
5719 @code{widen_mult_hi/lo} idioms will be used.
5722 @hook TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD
5723 This hook should return the DECL of a function @var{f} that implements
5724 widening multiplication of the odd elements of two input vectors of type @var{x}.
5726 If this hook is defined, the autovectorizer will use it along with the
5727 @code{TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN} target hook when vectorizing
5728 widening multiplication in cases that the order of the results does not have to be
5729 preserved (e.g.@: used only by a reduction computation). Otherwise, the
5730 @code{widen_mult_hi/lo} idioms will be used.
5733 @hook TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST
5734 Returns cost of different scalar or vector statements for vectorization cost model.
5735 For vector memory operations the cost may depend on type (@var{vectype}) and
5736 misalignment value (@var{misalign}).
5739 @hook TARGET_VECTORIZE_VECTOR_ALIGNMENT_REACHABLE
5740 Return true if vector alignment is reachable (by peeling N iterations) for the given type.
5743 @hook TARGET_VECTORIZE_BUILTIN_VEC_PERM
5744 Target builtin that implements vector permute.
5747 @hook TARGET_VECTORIZE_BUILTIN_VEC_PERM_OK
5748 Return true if a vector created for @code{builtin_vec_perm} is valid.
5751 @hook TARGET_VECTORIZE_BUILTIN_CONVERSION
5752 This hook should return the DECL of a function that implements conversion of the
5753 input vector of type @var{src_type} to type @var{dest_type}.
5754 The value of @var{code} is one of the enumerators in @code{enum tree_code} and
5755 specifies how the conversion is to be applied
5756 (truncation, rounding, etc.).
5758 If this hook is defined, the autovectorizer will use the
5759 @code{TARGET_VECTORIZE_BUILTIN_CONVERSION} target hook when vectorizing
5760 conversion. Otherwise, it will return @code{NULL_TREE}.
5763 @hook TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION
5764 This hook should return the decl of a function that implements the
5765 vectorized variant of the builtin function with builtin function code
5766 @var{code} or @code{NULL_TREE} if such a function is not available.
5767 The value of @var{fndecl} is the builtin function declaration. The
5768 return type of the vectorized function shall be of vector type
5769 @var{vec_type_out} and the argument types should be @var{vec_type_in}.
5772 @hook TARGET_VECTORIZE_SUPPORT_VECTOR_MISALIGNMENT
5773 This hook should return true if the target supports misaligned vector
5774 store/load of a specific factor denoted in the @var{misalignment}
5775 parameter. The vector store/load should be of machine mode @var{mode} and
5776 the elements in the vectors should be of type @var{type}. @var{is_packed}
5777 parameter is true if the memory access is defined in a packed struct.
5780 @hook TARGET_VECTORIZE_PREFERRED_SIMD_MODE
5781 This hook should return the preferred mode for vectorizing scalar
5782 mode @var{mode}. The default is
5783 equal to @code{word_mode}, because the vectorizer can do some
5784 transformations even in absence of specialized @acronym{SIMD} hardware.
5787 @hook TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_SIZES
5788 This hook should return a mask of sizes that should be iterated over
5789 after trying to autovectorize using the vector size derived from the
5790 mode returned by @code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE}.
5791 The default is zero which means to not iterate over other vector sizes.
5794 @node Anchored Addresses
5795 @section Anchored Addresses
5796 @cindex anchored addresses
5797 @cindex @option{-fsection-anchors}
5799 GCC usually addresses every static object as a separate entity.
5800 For example, if we have:
5804 int foo (void) @{ return a + b + c; @}
5807 the code for @code{foo} will usually calculate three separate symbolic
5808 addresses: those of @code{a}, @code{b} and @code{c}. On some targets,
5809 it would be better to calculate just one symbolic address and access
5810 the three variables relative to it. The equivalent pseudocode would
5816 register int *xr = &x;
5817 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x];
5821 (which isn't valid C). We refer to shared addresses like @code{x} as
5822 ``section anchors''. Their use is controlled by @option{-fsection-anchors}.
5824 The hooks below describe the target properties that GCC needs to know
5825 in order to make effective use of section anchors. It won't use
5826 section anchors at all unless either @code{TARGET_MIN_ANCHOR_OFFSET}
5827 or @code{TARGET_MAX_ANCHOR_OFFSET} is set to a nonzero value.
5829 @hook TARGET_MIN_ANCHOR_OFFSET
5830 The minimum offset that should be applied to a section anchor.
5831 On most targets, it should be the smallest offset that can be
5832 applied to a base register while still giving a legitimate address
5833 for every mode. The default value is 0.
5836 @hook TARGET_MAX_ANCHOR_OFFSET
5837 Like @code{TARGET_MIN_ANCHOR_OFFSET}, but the maximum (inclusive)
5838 offset that should be applied to section anchors. The default
5842 @hook TARGET_ASM_OUTPUT_ANCHOR
5843 Write the assembly code to define section anchor @var{x}, which is a
5844 @code{SYMBOL_REF} for which @samp{SYMBOL_REF_ANCHOR_P (@var{x})} is true.
5845 The hook is called with the assembly output position set to the beginning
5846 of @code{SYMBOL_REF_BLOCK (@var{x})}.
5848 If @code{ASM_OUTPUT_DEF} is available, the hook's default definition uses
5849 it to define the symbol as @samp{. + SYMBOL_REF_BLOCK_OFFSET (@var{x})}.
5850 If @code{ASM_OUTPUT_DEF} is not available, the hook's default definition
5851 is @code{NULL}, which disables the use of section anchors altogether.
5854 @hook TARGET_USE_ANCHORS_FOR_SYMBOL_P
5855 Return true if GCC should attempt to use anchors to access @code{SYMBOL_REF}
5856 @var{x}. You can assume @samp{SYMBOL_REF_HAS_BLOCK_INFO_P (@var{x})} and
5857 @samp{!SYMBOL_REF_ANCHOR_P (@var{x})}.
5859 The default version is correct for most targets, but you might need to
5860 intercept this hook to handle things like target-specific attributes
5861 or target-specific sections.
5864 @node Condition Code
5865 @section Condition Code Status
5866 @cindex condition code status
5868 The macros in this section can be split in two families, according to the
5869 two ways of representing condition codes in GCC.
5871 The first representation is the so called @code{(cc0)} representation
5872 (@pxref{Jump Patterns}), where all instructions can have an implicit
5873 clobber of the condition codes. The second is the condition code
5874 register representation, which provides better schedulability for
5875 architectures that do have a condition code register, but on which
5876 most instructions do not affect it. The latter category includes
5879 The implicit clobbering poses a strong restriction on the placement of
5880 the definition and use of the condition code, which need to be in adjacent
5881 insns for machines using @code{(cc0)}. This can prevent important
5882 optimizations on some machines. For example, on the IBM RS/6000, there
5883 is a delay for taken branches unless the condition code register is set
5884 three instructions earlier than the conditional branch. The instruction
5885 scheduler cannot perform this optimization if it is not permitted to
5886 separate the definition and use of the condition code register.
5888 For this reason, it is possible and suggested to use a register to
5889 represent the condition code for new ports. If there is a specific
5890 condition code register in the machine, use a hard register. If the
5891 condition code or comparison result can be placed in any general register,
5892 or if there are multiple condition registers, use a pseudo register.
5893 Registers used to store the condition code value will usually have a mode
5894 that is in class @code{MODE_CC}.
5896 Alternatively, you can use @code{BImode} if the comparison operator is
5897 specified already in the compare instruction. In this case, you are not
5898 interested in most macros in this section.
5901 * CC0 Condition Codes:: Old style representation of condition codes.
5902 * MODE_CC Condition Codes:: Modern representation of condition codes.
5903 * Cond. Exec. Macros:: Macros to control conditional execution.
5906 @node CC0 Condition Codes
5907 @subsection Representation of condition codes using @code{(cc0)}
5911 The file @file{conditions.h} defines a variable @code{cc_status} to
5912 describe how the condition code was computed (in case the interpretation of
5913 the condition code depends on the instruction that it was set by). This
5914 variable contains the RTL expressions on which the condition code is
5915 currently based, and several standard flags.
5917 Sometimes additional machine-specific flags must be defined in the machine
5918 description header file. It can also add additional machine-specific
5919 information by defining @code{CC_STATUS_MDEP}.
5921 @defmac CC_STATUS_MDEP
5922 C code for a data type which is used for declaring the @code{mdep}
5923 component of @code{cc_status}. It defaults to @code{int}.
5925 This macro is not used on machines that do not use @code{cc0}.
5928 @defmac CC_STATUS_MDEP_INIT
5929 A C expression to initialize the @code{mdep} field to ``empty''.
5930 The default definition does nothing, since most machines don't use
5931 the field anyway. If you want to use the field, you should probably
5932 define this macro to initialize it.
5934 This macro is not used on machines that do not use @code{cc0}.
5937 @defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn})
5938 A C compound statement to set the components of @code{cc_status}
5939 appropriately for an insn @var{insn} whose body is @var{exp}. It is
5940 this macro's responsibility to recognize insns that set the condition
5941 code as a byproduct of other activity as well as those that explicitly
5944 This macro is not used on machines that do not use @code{cc0}.
5946 If there are insns that do not set the condition code but do alter
5947 other machine registers, this macro must check to see whether they
5948 invalidate the expressions that the condition code is recorded as
5949 reflecting. For example, on the 68000, insns that store in address
5950 registers do not set the condition code, which means that usually
5951 @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
5952 insns. But suppose that the previous insn set the condition code
5953 based on location @samp{a4@@(102)} and the current insn stores a new
5954 value in @samp{a4}. Although the condition code is not changed by
5955 this, it will no longer be true that it reflects the contents of
5956 @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
5957 @code{cc_status} in this case to say that nothing is known about the
5958 condition code value.
5960 The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
5961 with the results of peephole optimization: insns whose patterns are
5962 @code{parallel} RTXs containing various @code{reg}, @code{mem} or
5963 constants which are just the operands. The RTL structure of these
5964 insns is not sufficient to indicate what the insns actually do. What
5965 @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
5966 @code{CC_STATUS_INIT}.
5968 A possible definition of @code{NOTICE_UPDATE_CC} is to call a function
5969 that looks at an attribute (@pxref{Insn Attributes}) named, for example,
5970 @samp{cc}. This avoids having detailed information about patterns in
5971 two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}.
5974 @node MODE_CC Condition Codes
5975 @subsection Representation of condition codes using registers
5979 @defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y})
5980 On many machines, the condition code may be produced by other instructions
5981 than compares, for example the branch can use directly the condition
5982 code set by a subtract instruction. However, on some machines
5983 when the condition code is set this way some bits (such as the overflow
5984 bit) are not set in the same way as a test instruction, so that a different
5985 branch instruction must be used for some conditional branches. When
5986 this happens, use the machine mode of the condition code register to
5987 record different formats of the condition code register. Modes can
5988 also be used to record which compare instruction (e.g. a signed or an
5989 unsigned comparison) produced the condition codes.
5991 If other modes than @code{CCmode} are required, add them to
5992 @file{@var{machine}-modes.def} and define @code{SELECT_CC_MODE} to choose
5993 a mode given an operand of a compare. This is needed because the modes
5994 have to be chosen not only during RTL generation but also, for example,
5995 by instruction combination. The result of @code{SELECT_CC_MODE} should
5996 be consistent with the mode used in the patterns; for example to support
5997 the case of the add on the SPARC discussed above, we have the pattern
6001 [(set (reg:CC_NOOV 0)
6003 (plus:SI (match_operand:SI 0 "register_operand" "%r")
6004 (match_operand:SI 1 "arith_operand" "rI"))
6011 together with a @code{SELECT_CC_MODE} that returns @code{CC_NOOVmode}
6012 for comparisons whose argument is a @code{plus}:
6015 #define SELECT_CC_MODE(OP,X,Y) \
6016 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \
6017 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \
6018 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \
6019 || GET_CODE (X) == NEG) \
6020 ? CC_NOOVmode : CCmode))
6023 Another reason to use modes is to retain information on which operands
6024 were used by the comparison; see @code{REVERSIBLE_CC_MODE} later in
6027 You should define this macro if and only if you define extra CC modes
6028 in @file{@var{machine}-modes.def}.
6031 @defmac CANONICALIZE_COMPARISON (@var{code}, @var{op0}, @var{op1})
6032 On some machines not all possible comparisons are defined, but you can
6033 convert an invalid comparison into a valid one. For example, the Alpha
6034 does not have a @code{GT} comparison, but you can use an @code{LT}
6035 comparison instead and swap the order of the operands.
6037 On such machines, define this macro to be a C statement to do any
6038 required conversions. @var{code} is the initial comparison code
6039 and @var{op0} and @var{op1} are the left and right operands of the
6040 comparison, respectively. You should modify @var{code}, @var{op0}, and
6041 @var{op1} as required.
6043 GCC will not assume that the comparison resulting from this macro is
6044 valid but will see if the resulting insn matches a pattern in the
6047 You need not define this macro if it would never change the comparison
6051 @defmac REVERSIBLE_CC_MODE (@var{mode})
6052 A C expression whose value is one if it is always safe to reverse a
6053 comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE}
6054 can ever return @var{mode} for a floating-point inequality comparison,
6055 then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero.
6057 You need not define this macro if it would always returns zero or if the
6058 floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}.
6059 For example, here is the definition used on the SPARC, where floating-point
6060 inequality comparisons are always given @code{CCFPEmode}:
6063 #define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode)
6067 @defmac REVERSE_CONDITION (@var{code}, @var{mode})
6068 A C expression whose value is reversed condition code of the @var{code} for
6069 comparison done in CC_MODE @var{mode}. The macro is used only in case
6070 @code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case
6071 machine has some non-standard way how to reverse certain conditionals. For
6072 instance in case all floating point conditions are non-trapping, compiler may
6073 freely convert unordered compares to ordered one. Then definition may look
6077 #define REVERSE_CONDITION(CODE, MODE) \
6078 ((MODE) != CCFPmode ? reverse_condition (CODE) \
6079 : reverse_condition_maybe_unordered (CODE))
6083 @hook TARGET_FIXED_CONDITION_CODE_REGS
6084 On targets which do not use @code{(cc0)}, and which use a hard
6085 register rather than a pseudo-register to hold condition codes, the
6086 regular CSE passes are often not able to identify cases in which the
6087 hard register is set to a common value. Use this hook to enable a
6088 small pass which optimizes such cases. This hook should return true
6089 to enable this pass, and it should set the integers to which its
6090 arguments point to the hard register numbers used for condition codes.
6091 When there is only one such register, as is true on most systems, the
6092 integer pointed to by @var{p2} should be set to
6093 @code{INVALID_REGNUM}.
6095 The default version of this hook returns false.
6098 @hook TARGET_CC_MODES_COMPATIBLE
6099 On targets which use multiple condition code modes in class
6100 @code{MODE_CC}, it is sometimes the case that a comparison can be
6101 validly done in more than one mode. On such a system, define this
6102 target hook to take two mode arguments and to return a mode in which
6103 both comparisons may be validly done. If there is no such mode,
6104 return @code{VOIDmode}.
6106 The default version of this hook checks whether the modes are the
6107 same. If they are, it returns that mode. If they are different, it
6108 returns @code{VOIDmode}.
6111 @node Cond. Exec. Macros
6112 @subsection Macros to control conditional execution
6113 @findex conditional execution
6116 There is one macro that may need to be defined for targets
6117 supporting conditional execution, independent of how they
6118 represent conditional branches.
6120 @defmac REVERSE_CONDEXEC_PREDICATES_P (@var{op1}, @var{op2})
6121 A C expression that returns true if the conditional execution predicate
6122 @var{op1}, a comparison operation, is the inverse of @var{op2} and vice
6123 versa. Define this to return 0 if the target has conditional execution
6124 predicates that cannot be reversed safely. There is no need to validate
6125 that the arguments of op1 and op2 are the same, this is done separately.
6126 If no expansion is specified, this macro is defined as follows:
6129 #define REVERSE_CONDEXEC_PREDICATES_P (x, y) \
6130 (GET_CODE ((x)) == reversed_comparison_code ((y), NULL))
6135 @section Describing Relative Costs of Operations
6136 @cindex costs of instructions
6137 @cindex relative costs
6138 @cindex speed of instructions
6140 These macros let you describe the relative speed of various operations
6141 on the target machine.
6143 @defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to})
6144 A C expression for the cost of moving data of mode @var{mode} from a
6145 register in class @var{from} to one in class @var{to}. The classes are
6146 expressed using the enumeration values such as @code{GENERAL_REGS}. A
6147 value of 2 is the default; other values are interpreted relative to
6150 It is not required that the cost always equal 2 when @var{from} is the
6151 same as @var{to}; on some machines it is expensive to move between
6152 registers if they are not general registers.
6154 If reload sees an insn consisting of a single @code{set} between two
6155 hard registers, and if @code{REGISTER_MOVE_COST} applied to their
6156 classes returns a value of 2, reload does not check to ensure that the
6157 constraints of the insn are met. Setting a cost of other than 2 will
6158 allow reload to verify that the constraints are met. You should do this
6159 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
6161 These macros are obsolete, new ports should use the target hook
6162 @code{TARGET_REGISTER_MOVE_COST} instead.
6165 @hook TARGET_REGISTER_MOVE_COST
6166 This target hook should return the cost of moving data of mode @var{mode}
6167 from a register in class @var{from} to one in class @var{to}. The classes
6168 are expressed using the enumeration values such as @code{GENERAL_REGS}.
6169 A value of 2 is the default; other values are interpreted relative to
6172 It is not required that the cost always equal 2 when @var{from} is the
6173 same as @var{to}; on some machines it is expensive to move between
6174 registers if they are not general registers.
6176 If reload sees an insn consisting of a single @code{set} between two
6177 hard registers, and if @code{TARGET_REGISTER_MOVE_COST} applied to their
6178 classes returns a value of 2, reload does not check to ensure that the
6179 constraints of the insn are met. Setting a cost of other than 2 will
6180 allow reload to verify that the constraints are met. You should do this
6181 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
6183 The default version of this function returns 2.
6186 @defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in})
6187 A C expression for the cost of moving data of mode @var{mode} between a
6188 register of class @var{class} and memory; @var{in} is zero if the value
6189 is to be written to memory, nonzero if it is to be read in. This cost
6190 is relative to those in @code{REGISTER_MOVE_COST}. If moving between
6191 registers and memory is more expensive than between two registers, you
6192 should define this macro to express the relative cost.
6194 If you do not define this macro, GCC uses a default cost of 4 plus
6195 the cost of copying via a secondary reload register, if one is
6196 needed. If your machine requires a secondary reload register to copy
6197 between memory and a register of @var{class} but the reload mechanism is
6198 more complex than copying via an intermediate, define this macro to
6199 reflect the actual cost of the move.
6201 GCC defines the function @code{memory_move_secondary_cost} if
6202 secondary reloads are needed. It computes the costs due to copying via
6203 a secondary register. If your machine copies from memory using a
6204 secondary register in the conventional way but the default base value of
6205 4 is not correct for your machine, define this macro to add some other
6206 value to the result of that function. The arguments to that function
6207 are the same as to this macro.
6209 These macros are obsolete, new ports should use the target hook
6210 @code{TARGET_MEMORY_MOVE_COST} instead.
6213 @hook TARGET_MEMORY_MOVE_COST
6214 This target hook should return the cost of moving data of mode @var{mode}
6215 between a register of class @var{rclass} and memory; @var{in} is @code{false}
6216 if the value is to be written to memory, @code{true} if it is to be read in.
6217 This cost is relative to those in @code{TARGET_REGISTER_MOVE_COST}.
6218 If moving between registers and memory is more expensive than between two
6219 registers, you should add this target hook to express the relative cost.
6221 If you do not add this target hook, GCC uses a default cost of 4 plus
6222 the cost of copying via a secondary reload register, if one is
6223 needed. If your machine requires a secondary reload register to copy
6224 between memory and a register of @var{rclass} but the reload mechanism is
6225 more complex than copying via an intermediate, use this target hook to
6226 reflect the actual cost of the move.
6228 GCC defines the function @code{memory_move_secondary_cost} if
6229 secondary reloads are needed. It computes the costs due to copying via
6230 a secondary register. If your machine copies from memory using a
6231 secondary register in the conventional way but the default base value of
6232 4 is not correct for your machine, use this target hook to add some other
6233 value to the result of that function. The arguments to that function
6234 are the same as to this target hook.
6237 @defmac BRANCH_COST (@var{speed_p}, @var{predictable_p})
6238 A C expression for the cost of a branch instruction. A value of 1 is the
6239 default; other values are interpreted relative to that. Parameter @var{speed_p}
6240 is true when the branch in question should be optimized for speed. When
6241 it is false, @code{BRANCH_COST} should be returning value optimal for code size
6242 rather then performance considerations. @var{predictable_p} is true for well
6243 predictable branches. On many architectures the @code{BRANCH_COST} can be
6247 Here are additional macros which do not specify precise relative costs,
6248 but only that certain actions are more expensive than GCC would
6251 @defmac SLOW_BYTE_ACCESS
6252 Define this macro as a C expression which is nonzero if accessing less
6253 than a word of memory (i.e.@: a @code{char} or a @code{short}) is no
6254 faster than accessing a word of memory, i.e., if such access
6255 require more than one instruction or if there is no difference in cost
6256 between byte and (aligned) word loads.
6258 When this macro is not defined, the compiler will access a field by
6259 finding the smallest containing object; when it is defined, a fullword
6260 load will be used if alignment permits. Unless bytes accesses are
6261 faster than word accesses, using word accesses is preferable since it
6262 may eliminate subsequent memory access if subsequent accesses occur to
6263 other fields in the same word of the structure, but to different bytes.
6266 @defmac SLOW_UNALIGNED_ACCESS (@var{mode}, @var{alignment})
6267 Define this macro to be the value 1 if memory accesses described by the
6268 @var{mode} and @var{alignment} parameters have a cost many times greater
6269 than aligned accesses, for example if they are emulated in a trap
6272 When this macro is nonzero, the compiler will act as if
6273 @code{STRICT_ALIGNMENT} were nonzero when generating code for block
6274 moves. This can cause significantly more instructions to be produced.
6275 Therefore, do not set this macro nonzero if unaligned accesses only add a
6276 cycle or two to the time for a memory access.
6278 If the value of this macro is always zero, it need not be defined. If
6279 this macro is defined, it should produce a nonzero value when
6280 @code{STRICT_ALIGNMENT} is nonzero.
6283 @defmac MOVE_RATIO (@var{speed})
6284 The threshold of number of scalar memory-to-memory move insns, @emph{below}
6285 which a sequence of insns should be generated instead of a
6286 string move insn or a library call. Increasing the value will always
6287 make code faster, but eventually incurs high cost in increased code size.
6289 Note that on machines where the corresponding move insn is a
6290 @code{define_expand} that emits a sequence of insns, this macro counts
6291 the number of such sequences.
6293 The parameter @var{speed} is true if the code is currently being
6294 optimized for speed rather than size.
6296 If you don't define this, a reasonable default is used.
6299 @defmac MOVE_BY_PIECES_P (@var{size}, @var{alignment})
6300 A C expression used to determine whether @code{move_by_pieces} will be used to
6301 copy a chunk of memory, or whether some other block move mechanism
6302 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6303 than @code{MOVE_RATIO}.
6306 @defmac MOVE_MAX_PIECES
6307 A C expression used by @code{move_by_pieces} to determine the largest unit
6308 a load or store used to copy memory is. Defaults to @code{MOVE_MAX}.
6311 @defmac CLEAR_RATIO (@var{speed})
6312 The threshold of number of scalar move insns, @emph{below} which a sequence
6313 of insns should be generated to clear memory instead of a string clear insn
6314 or a library call. Increasing the value will always make code faster, but
6315 eventually incurs high cost in increased code size.
6317 The parameter @var{speed} is true if the code is currently being
6318 optimized for speed rather than size.
6320 If you don't define this, a reasonable default is used.
6323 @defmac CLEAR_BY_PIECES_P (@var{size}, @var{alignment})
6324 A C expression used to determine whether @code{clear_by_pieces} will be used
6325 to clear a chunk of memory, or whether some other block clear mechanism
6326 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6327 than @code{CLEAR_RATIO}.
6330 @defmac SET_RATIO (@var{speed})
6331 The threshold of number of scalar move insns, @emph{below} which a sequence
6332 of insns should be generated to set memory to a constant value, instead of
6333 a block set insn or a library call.
6334 Increasing the value will always make code faster, but
6335 eventually incurs high cost in increased code size.
6337 The parameter @var{speed} is true if the code is currently being
6338 optimized for speed rather than size.
6340 If you don't define this, it defaults to the value of @code{MOVE_RATIO}.
6343 @defmac SET_BY_PIECES_P (@var{size}, @var{alignment})
6344 A C expression used to determine whether @code{store_by_pieces} will be
6345 used to set a chunk of memory to a constant value, or whether some
6346 other mechanism will be used. Used by @code{__builtin_memset} when
6347 storing values other than constant zero.
6348 Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6349 than @code{SET_RATIO}.
6352 @defmac STORE_BY_PIECES_P (@var{size}, @var{alignment})
6353 A C expression used to determine whether @code{store_by_pieces} will be
6354 used to set a chunk of memory to a constant string value, or whether some
6355 other mechanism will be used. Used by @code{__builtin_strcpy} when
6356 called with a constant source string.
6357 Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6358 than @code{MOVE_RATIO}.
6361 @defmac USE_LOAD_POST_INCREMENT (@var{mode})
6362 A C expression used to determine whether a load postincrement is a good
6363 thing to use for a given mode. Defaults to the value of
6364 @code{HAVE_POST_INCREMENT}.
6367 @defmac USE_LOAD_POST_DECREMENT (@var{mode})
6368 A C expression used to determine whether a load postdecrement is a good
6369 thing to use for a given mode. Defaults to the value of
6370 @code{HAVE_POST_DECREMENT}.
6373 @defmac USE_LOAD_PRE_INCREMENT (@var{mode})
6374 A C expression used to determine whether a load preincrement is a good
6375 thing to use for a given mode. Defaults to the value of
6376 @code{HAVE_PRE_INCREMENT}.
6379 @defmac USE_LOAD_PRE_DECREMENT (@var{mode})
6380 A C expression used to determine whether a load predecrement is a good
6381 thing to use for a given mode. Defaults to the value of
6382 @code{HAVE_PRE_DECREMENT}.
6385 @defmac USE_STORE_POST_INCREMENT (@var{mode})
6386 A C expression used to determine whether a store postincrement is a good
6387 thing to use for a given mode. Defaults to the value of
6388 @code{HAVE_POST_INCREMENT}.
6391 @defmac USE_STORE_POST_DECREMENT (@var{mode})
6392 A C expression used to determine whether a store postdecrement is a good
6393 thing to use for a given mode. Defaults to the value of
6394 @code{HAVE_POST_DECREMENT}.
6397 @defmac USE_STORE_PRE_INCREMENT (@var{mode})
6398 This macro is used to determine whether a store preincrement is a good
6399 thing to use for a given mode. Defaults to the value of
6400 @code{HAVE_PRE_INCREMENT}.
6403 @defmac USE_STORE_PRE_DECREMENT (@var{mode})
6404 This macro is used to determine whether a store predecrement is a good
6405 thing to use for a given mode. Defaults to the value of
6406 @code{HAVE_PRE_DECREMENT}.
6409 @defmac NO_FUNCTION_CSE
6410 Define this macro if it is as good or better to call a constant
6411 function address than to call an address kept in a register.
6414 @defmac RANGE_TEST_NON_SHORT_CIRCUIT
6415 Define this macro if a non-short-circuit operation produced by
6416 @samp{fold_range_test ()} is optimal. This macro defaults to true if
6417 @code{BRANCH_COST} is greater than or equal to the value 2.
6420 @hook TARGET_RTX_COSTS
6421 This target hook describes the relative costs of RTL expressions.
6423 The cost may depend on the precise form of the expression, which is
6424 available for examination in @var{x}, and the rtx code of the expression
6425 in which it is contained, found in @var{outer_code}. @var{code} is the
6426 expression code---redundant, since it can be obtained with
6427 @code{GET_CODE (@var{x})}.
6429 In implementing this hook, you can use the construct
6430 @code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast
6433 On entry to the hook, @code{*@var{total}} contains a default estimate
6434 for the cost of the expression. The hook should modify this value as
6435 necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)}
6436 for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus
6437 operations, and @code{COSTS_N_INSNS (1)} for all other operations.
6439 When optimizing for code size, i.e.@: when @code{speed} is
6440 false, this target hook should be used to estimate the relative
6441 size cost of an expression, again relative to @code{COSTS_N_INSNS}.
6443 The hook returns true when all subexpressions of @var{x} have been
6444 processed, and false when @code{rtx_cost} should recurse.
6447 @hook TARGET_ADDRESS_COST
6448 This hook computes the cost of an addressing mode that contains
6449 @var{address}. If not defined, the cost is computed from
6450 the @var{address} expression and the @code{TARGET_RTX_COST} hook.
6452 For most CISC machines, the default cost is a good approximation of the
6453 true cost of the addressing mode. However, on RISC machines, all
6454 instructions normally have the same length and execution time. Hence
6455 all addresses will have equal costs.
6457 In cases where more than one form of an address is known, the form with
6458 the lowest cost will be used. If multiple forms have the same, lowest,
6459 cost, the one that is the most complex will be used.
6461 For example, suppose an address that is equal to the sum of a register
6462 and a constant is used twice in the same basic block. When this macro
6463 is not defined, the address will be computed in a register and memory
6464 references will be indirect through that register. On machines where
6465 the cost of the addressing mode containing the sum is no higher than
6466 that of a simple indirect reference, this will produce an additional
6467 instruction and possibly require an additional register. Proper
6468 specification of this macro eliminates this overhead for such machines.
6470 This hook is never called with an invalid address.
6472 On machines where an address involving more than one register is as
6473 cheap as an address computation involving only one register, defining
6474 @code{TARGET_ADDRESS_COST} to reflect this can cause two registers to
6475 be live over a region of code where only one would have been if
6476 @code{TARGET_ADDRESS_COST} were not defined in that manner. This effect
6477 should be considered in the definition of this macro. Equivalent costs
6478 should probably only be given to addresses with different numbers of
6479 registers on machines with lots of registers.
6483 @section Adjusting the Instruction Scheduler
6485 The instruction scheduler may need a fair amount of machine-specific
6486 adjustment in order to produce good code. GCC provides several target
6487 hooks for this purpose. It is usually enough to define just a few of
6488 them: try the first ones in this list first.
6490 @hook TARGET_SCHED_ISSUE_RATE
6491 This hook returns the maximum number of instructions that can ever
6492 issue at the same time on the target machine. The default is one.
6493 Although the insn scheduler can define itself the possibility of issue
6494 an insn on the same cycle, the value can serve as an additional
6495 constraint to issue insns on the same simulated processor cycle (see
6496 hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}).
6497 This value must be constant over the entire compilation. If you need
6498 it to vary depending on what the instructions are, you must use
6499 @samp{TARGET_SCHED_VARIABLE_ISSUE}.
6502 @hook TARGET_SCHED_VARIABLE_ISSUE
6503 This hook is executed by the scheduler after it has scheduled an insn
6504 from the ready list. It should return the number of insns which can
6505 still be issued in the current cycle. The default is
6506 @samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and
6507 @code{USE}, which normally are not counted against the issue rate.
6508 You should define this hook if some insns take more machine resources
6509 than others, so that fewer insns can follow them in the same cycle.
6510 @var{file} is either a null pointer, or a stdio stream to write any
6511 debug output to. @var{verbose} is the verbose level provided by
6512 @option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that
6516 @hook TARGET_SCHED_ADJUST_COST
6517 This function corrects the value of @var{cost} based on the
6518 relationship between @var{insn} and @var{dep_insn} through the
6519 dependence @var{link}. It should return the new value. The default
6520 is to make no adjustment to @var{cost}. This can be used for example
6521 to specify to the scheduler using the traditional pipeline description
6522 that an output- or anti-dependence does not incur the same cost as a
6523 data-dependence. If the scheduler using the automaton based pipeline
6524 description, the cost of anti-dependence is zero and the cost of
6525 output-dependence is maximum of one and the difference of latency
6526 times of the first and the second insns. If these values are not
6527 acceptable, you could use the hook to modify them too. See also
6528 @pxref{Processor pipeline description}.
6531 @hook TARGET_SCHED_ADJUST_PRIORITY
6532 This hook adjusts the integer scheduling priority @var{priority} of
6533 @var{insn}. It should return the new priority. Increase the priority to
6534 execute @var{insn} earlier, reduce the priority to execute @var{insn}
6535 later. Do not define this hook if you do not need to adjust the
6536 scheduling priorities of insns.
6539 @hook TARGET_SCHED_REORDER
6540 This hook is executed by the scheduler after it has scheduled the ready
6541 list, to allow the machine description to reorder it (for example to
6542 combine two small instructions together on @samp{VLIW} machines).
6543 @var{file} is either a null pointer, or a stdio stream to write any
6544 debug output to. @var{verbose} is the verbose level provided by
6545 @option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready
6546 list of instructions that are ready to be scheduled. @var{n_readyp} is
6547 a pointer to the number of elements in the ready list. The scheduler
6548 reads the ready list in reverse order, starting with
6549 @var{ready}[@var{*n_readyp} @minus{} 1] and going to @var{ready}[0]. @var{clock}
6550 is the timer tick of the scheduler. You may modify the ready list and
6551 the number of ready insns. The return value is the number of insns that
6552 can issue this cycle; normally this is just @code{issue_rate}. See also
6553 @samp{TARGET_SCHED_REORDER2}.
6556 @hook TARGET_SCHED_REORDER2
6557 Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That
6558 function is called whenever the scheduler starts a new cycle. This one
6559 is called once per iteration over a cycle, immediately after
6560 @samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and
6561 return the number of insns to be scheduled in the same cycle. Defining
6562 this hook can be useful if there are frequent situations where
6563 scheduling one insn causes other insns to become ready in the same
6564 cycle. These other insns can then be taken into account properly.
6567 @hook TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK
6568 This hook is called after evaluation forward dependencies of insns in
6569 chain given by two parameter values (@var{head} and @var{tail}
6570 correspondingly) but before insns scheduling of the insn chain. For
6571 example, it can be used for better insn classification if it requires
6572 analysis of dependencies. This hook can use backward and forward
6573 dependencies of the insn scheduler because they are already
6577 @hook TARGET_SCHED_INIT
6578 This hook is executed by the scheduler at the beginning of each block of
6579 instructions that are to be scheduled. @var{file} is either a null
6580 pointer, or a stdio stream to write any debug output to. @var{verbose}
6581 is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6582 @var{max_ready} is the maximum number of insns in the current scheduling
6583 region that can be live at the same time. This can be used to allocate
6584 scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}.
6587 @hook TARGET_SCHED_FINISH
6588 This hook is executed by the scheduler at the end of each block of
6589 instructions that are to be scheduled. It can be used to perform
6590 cleanup of any actions done by the other scheduling hooks. @var{file}
6591 is either a null pointer, or a stdio stream to write any debug output
6592 to. @var{verbose} is the verbose level provided by
6593 @option{-fsched-verbose-@var{n}}.
6596 @hook TARGET_SCHED_INIT_GLOBAL
6597 This hook is executed by the scheduler after function level initializations.
6598 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
6599 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6600 @var{old_max_uid} is the maximum insn uid when scheduling begins.
6603 @hook TARGET_SCHED_FINISH_GLOBAL
6604 This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}.
6605 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
6606 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6609 @hook TARGET_SCHED_DFA_PRE_CYCLE_INSN
6610 The hook returns an RTL insn. The automaton state used in the
6611 pipeline hazard recognizer is changed as if the insn were scheduled
6612 when the new simulated processor cycle starts. Usage of the hook may
6613 simplify the automaton pipeline description for some @acronym{VLIW}
6614 processors. If the hook is defined, it is used only for the automaton
6615 based pipeline description. The default is not to change the state
6616 when the new simulated processor cycle starts.
6619 @hook TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN
6620 The hook can be used to initialize data used by the previous hook.
6623 @hook TARGET_SCHED_DFA_POST_CYCLE_INSN
6624 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
6625 to changed the state as if the insn were scheduled when the new
6626 simulated processor cycle finishes.
6629 @hook TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN
6630 The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but
6631 used to initialize data used by the previous hook.
6634 @hook TARGET_SCHED_DFA_PRE_ADVANCE_CYCLE
6635 The hook to notify target that the current simulated cycle is about to finish.
6636 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
6637 to change the state in more complicated situations - e.g., when advancing
6638 state on a single insn is not enough.
6641 @hook TARGET_SCHED_DFA_POST_ADVANCE_CYCLE
6642 The hook to notify target that new simulated cycle has just started.
6643 The hook is analogous to @samp{TARGET_SCHED_DFA_POST_CYCLE_INSN} but used
6644 to change the state in more complicated situations - e.g., when advancing
6645 state on a single insn is not enough.
6648 @hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD
6649 This hook controls better choosing an insn from the ready insn queue
6650 for the @acronym{DFA}-based insn scheduler. Usually the scheduler
6651 chooses the first insn from the queue. If the hook returns a positive
6652 value, an additional scheduler code tries all permutations of
6653 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()}
6654 subsequent ready insns to choose an insn whose issue will result in
6655 maximal number of issued insns on the same cycle. For the
6656 @acronym{VLIW} processor, the code could actually solve the problem of
6657 packing simple insns into the @acronym{VLIW} insn. Of course, if the
6658 rules of @acronym{VLIW} packing are described in the automaton.
6660 This code also could be used for superscalar @acronym{RISC}
6661 processors. Let us consider a superscalar @acronym{RISC} processor
6662 with 3 pipelines. Some insns can be executed in pipelines @var{A} or
6663 @var{B}, some insns can be executed only in pipelines @var{B} or
6664 @var{C}, and one insn can be executed in pipeline @var{B}. The
6665 processor may issue the 1st insn into @var{A} and the 2nd one into
6666 @var{B}. In this case, the 3rd insn will wait for freeing @var{B}
6667 until the next cycle. If the scheduler issues the 3rd insn the first,
6668 the processor could issue all 3 insns per cycle.
6670 Actually this code demonstrates advantages of the automaton based
6671 pipeline hazard recognizer. We try quickly and easy many insn
6672 schedules to choose the best one.
6674 The default is no multipass scheduling.
6677 @hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD
6679 This hook controls what insns from the ready insn queue will be
6680 considered for the multipass insn scheduling. If the hook returns
6681 zero for @var{insn}, the insn will be not chosen to
6684 The default is that any ready insns can be chosen to be issued.
6687 @hook TARGET_SCHED_DFA_NEW_CYCLE
6688 This hook is called by the insn scheduler before issuing @var{insn}
6689 on cycle @var{clock}. If the hook returns nonzero,
6690 @var{insn} is not issued on this processor cycle. Instead,
6691 the processor cycle is advanced. If *@var{sort_p}
6692 is zero, the insn ready queue is not sorted on the new cycle
6693 start as usually. @var{dump} and @var{verbose} specify the file and
6694 verbosity level to use for debugging output.
6695 @var{last_clock} and @var{clock} are, respectively, the
6696 processor cycle on which the previous insn has been issued,
6697 and the current processor cycle.
6700 @hook TARGET_SCHED_IS_COSTLY_DEPENDENCE
6701 This hook is used to define which dependences are considered costly by
6702 the target, so costly that it is not advisable to schedule the insns that
6703 are involved in the dependence too close to one another. The parameters
6704 to this hook are as follows: The first parameter @var{_dep} is the dependence
6705 being evaluated. The second parameter @var{cost} is the cost of the
6706 dependence as estimated by the scheduler, and the third
6707 parameter @var{distance} is the distance in cycles between the two insns.
6708 The hook returns @code{true} if considering the distance between the two
6709 insns the dependence between them is considered costly by the target,
6710 and @code{false} otherwise.
6712 Defining this hook can be useful in multiple-issue out-of-order machines,
6713 where (a) it's practically hopeless to predict the actual data/resource
6714 delays, however: (b) there's a better chance to predict the actual grouping
6715 that will be formed, and (c) correctly emulating the grouping can be very
6716 important. In such targets one may want to allow issuing dependent insns
6717 closer to one another---i.e., closer than the dependence distance; however,
6718 not in cases of ``costly dependences'', which this hooks allows to define.
6721 @hook TARGET_SCHED_H_I_D_EXTENDED
6722 This hook is called by the insn scheduler after emitting a new instruction to
6723 the instruction stream. The hook notifies a target backend to extend its
6724 per instruction data structures.
6727 @hook TARGET_SCHED_ALLOC_SCHED_CONTEXT
6728 Return a pointer to a store large enough to hold target scheduling context.
6731 @hook TARGET_SCHED_INIT_SCHED_CONTEXT
6732 Initialize store pointed to by @var{tc} to hold target scheduling context.
6733 It @var{clean_p} is true then initialize @var{tc} as if scheduler is at the
6734 beginning of the block. Otherwise, copy the current context into @var{tc}.
6737 @hook TARGET_SCHED_SET_SCHED_CONTEXT
6738 Copy target scheduling context pointed to by @var{tc} to the current context.
6741 @hook TARGET_SCHED_CLEAR_SCHED_CONTEXT
6742 Deallocate internal data in target scheduling context pointed to by @var{tc}.
6745 @hook TARGET_SCHED_FREE_SCHED_CONTEXT
6746 Deallocate a store for target scheduling context pointed to by @var{tc}.
6749 @hook TARGET_SCHED_SPECULATE_INSN
6750 This hook is called by the insn scheduler when @var{insn} has only
6751 speculative dependencies and therefore can be scheduled speculatively.
6752 The hook is used to check if the pattern of @var{insn} has a speculative
6753 version and, in case of successful check, to generate that speculative
6754 pattern. The hook should return 1, if the instruction has a speculative form,
6755 or @minus{}1, if it doesn't. @var{request} describes the type of requested
6756 speculation. If the return value equals 1 then @var{new_pat} is assigned
6757 the generated speculative pattern.
6760 @hook TARGET_SCHED_NEEDS_BLOCK_P
6761 This hook is called by the insn scheduler during generation of recovery code
6762 for @var{insn}. It should return @code{true}, if the corresponding check
6763 instruction should branch to recovery code, or @code{false} otherwise.
6766 @hook TARGET_SCHED_GEN_SPEC_CHECK
6767 This hook is called by the insn scheduler to generate a pattern for recovery
6768 check instruction. If @var{mutate_p} is zero, then @var{insn} is a
6769 speculative instruction for which the check should be generated.
6770 @var{label} is either a label of a basic block, where recovery code should
6771 be emitted, or a null pointer, when requested check doesn't branch to
6772 recovery code (a simple check). If @var{mutate_p} is nonzero, then
6773 a pattern for a branchy check corresponding to a simple check denoted by
6774 @var{insn} should be generated. In this case @var{label} can't be null.
6777 @hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD_SPEC
6778 This hook is used as a workaround for
6779 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD} not being
6780 called on the first instruction of the ready list. The hook is used to
6781 discard speculative instructions that stand first in the ready list from
6782 being scheduled on the current cycle. If the hook returns @code{false},
6783 @var{insn} will not be chosen to be issued.
6784 For non-speculative instructions,
6785 the hook should always return @code{true}. For example, in the ia64 backend
6786 the hook is used to cancel data speculative insns when the ALAT table
6790 @hook TARGET_SCHED_SET_SCHED_FLAGS
6791 This hook is used by the insn scheduler to find out what features should be
6793 The structure *@var{spec_info} should be filled in by the target.
6794 The structure describes speculation types that can be used in the scheduler.
6797 @hook TARGET_SCHED_SMS_RES_MII
6798 This hook is called by the swing modulo scheduler to calculate a
6799 resource-based lower bound which is based on the resources available in
6800 the machine and the resources required by each instruction. The target
6801 backend can use @var{g} to calculate such bound. A very simple lower
6802 bound will be used in case this hook is not implemented: the total number
6803 of instructions divided by the issue rate.
6806 @hook TARGET_SCHED_DISPATCH
6807 This hook is called by Haifa Scheduler. It returns true if dispatch scheduling
6808 is supported in hardware and the condition specified in the parameter is true.
6811 @hook TARGET_SCHED_DISPATCH_DO
6812 This hook is called by Haifa Scheduler. It performs the operation specified
6813 in its second parameter.
6817 @section Dividing the Output into Sections (Texts, Data, @dots{})
6818 @c the above section title is WAY too long. maybe cut the part between
6819 @c the (...)? --mew 10feb93
6821 An object file is divided into sections containing different types of
6822 data. In the most common case, there are three sections: the @dfn{text
6823 section}, which holds instructions and read-only data; the @dfn{data
6824 section}, which holds initialized writable data; and the @dfn{bss
6825 section}, which holds uninitialized data. Some systems have other kinds
6828 @file{varasm.c} provides several well-known sections, such as
6829 @code{text_section}, @code{data_section} and @code{bss_section}.
6830 The normal way of controlling a @code{@var{foo}_section} variable
6831 is to define the associated @code{@var{FOO}_SECTION_ASM_OP} macro,
6832 as described below. The macros are only read once, when @file{varasm.c}
6833 initializes itself, so their values must be run-time constants.
6834 They may however depend on command-line flags.
6836 @emph{Note:} Some run-time files, such @file{crtstuff.c}, also make
6837 use of the @code{@var{FOO}_SECTION_ASM_OP} macros, and expect them
6838 to be string literals.
6840 Some assemblers require a different string to be written every time a
6841 section is selected. If your assembler falls into this category, you
6842 should define the @code{TARGET_ASM_INIT_SECTIONS} hook and use
6843 @code{get_unnamed_section} to set up the sections.
6845 You must always create a @code{text_section}, either by defining
6846 @code{TEXT_SECTION_ASM_OP} or by initializing @code{text_section}
6847 in @code{TARGET_ASM_INIT_SECTIONS}. The same is true of
6848 @code{data_section} and @code{DATA_SECTION_ASM_OP}. If you do not
6849 create a distinct @code{readonly_data_section}, the default is to
6850 reuse @code{text_section}.
6852 All the other @file{varasm.c} sections are optional, and are null
6853 if the target does not provide them.
6855 @defmac TEXT_SECTION_ASM_OP
6856 A C expression whose value is a string, including spacing, containing the
6857 assembler operation that should precede instructions and read-only data.
6858 Normally @code{"\t.text"} is right.
6861 @defmac HOT_TEXT_SECTION_NAME
6862 If defined, a C string constant for the name of the section containing most
6863 frequently executed functions of the program. If not defined, GCC will provide
6864 a default definition if the target supports named sections.
6867 @defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME
6868 If defined, a C string constant for the name of the section containing unlikely
6869 executed functions in the program.
6872 @defmac DATA_SECTION_ASM_OP
6873 A C expression whose value is a string, including spacing, containing the
6874 assembler operation to identify the following data as writable initialized
6875 data. Normally @code{"\t.data"} is right.
6878 @defmac SDATA_SECTION_ASM_OP
6879 If defined, a C expression whose value is a string, including spacing,
6880 containing the assembler operation to identify the following data as
6881 initialized, writable small data.
6884 @defmac READONLY_DATA_SECTION_ASM_OP
6885 A C expression whose value is a string, including spacing, containing the
6886 assembler operation to identify the following data as read-only initialized
6890 @defmac BSS_SECTION_ASM_OP
6891 If defined, a C expression whose value is a string, including spacing,
6892 containing the assembler operation to identify the following data as
6893 uninitialized global data. If not defined, and neither
6894 @code{ASM_OUTPUT_BSS} nor @code{ASM_OUTPUT_ALIGNED_BSS} are defined,
6895 uninitialized global data will be output in the data section if
6896 @option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be
6900 @defmac SBSS_SECTION_ASM_OP
6901 If defined, a C expression whose value is a string, including spacing,
6902 containing the assembler operation to identify the following data as
6903 uninitialized, writable small data.
6906 @defmac TLS_COMMON_ASM_OP
6907 If defined, a C expression whose value is a string containing the
6908 assembler operation to identify the following data as thread-local
6909 common data. The default is @code{".tls_common"}.
6912 @defmac TLS_SECTION_ASM_FLAG
6913 If defined, a C expression whose value is a character constant
6914 containing the flag used to mark a section as a TLS section. The
6915 default is @code{'T'}.
6918 @defmac INIT_SECTION_ASM_OP
6919 If defined, a C expression whose value is a string, including spacing,
6920 containing the assembler operation to identify the following data as
6921 initialization code. If not defined, GCC will assume such a section does
6922 not exist. This section has no corresponding @code{init_section}
6923 variable; it is used entirely in runtime code.
6926 @defmac FINI_SECTION_ASM_OP
6927 If defined, a C expression whose value is a string, including spacing,
6928 containing the assembler operation to identify the following data as
6929 finalization code. If not defined, GCC will assume such a section does
6930 not exist. This section has no corresponding @code{fini_section}
6931 variable; it is used entirely in runtime code.
6934 @defmac INIT_ARRAY_SECTION_ASM_OP
6935 If defined, a C expression whose value is a string, including spacing,
6936 containing the assembler operation to identify the following data as
6937 part of the @code{.init_array} (or equivalent) section. If not
6938 defined, GCC will assume such a section does not exist. Do not define
6939 both this macro and @code{INIT_SECTION_ASM_OP}.
6942 @defmac FINI_ARRAY_SECTION_ASM_OP
6943 If defined, a C expression whose value is a string, including spacing,
6944 containing the assembler operation to identify the following data as
6945 part of the @code{.fini_array} (or equivalent) section. If not
6946 defined, GCC will assume such a section does not exist. Do not define
6947 both this macro and @code{FINI_SECTION_ASM_OP}.
6950 @defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function})
6951 If defined, an ASM statement that switches to a different section
6952 via @var{section_op}, calls @var{function}, and switches back to
6953 the text section. This is used in @file{crtstuff.c} if
6954 @code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls
6955 to initialization and finalization functions from the init and fini
6956 sections. By default, this macro uses a simple function call. Some
6957 ports need hand-crafted assembly code to avoid dependencies on
6958 registers initialized in the function prologue or to ensure that
6959 constant pools don't end up too far way in the text section.
6962 @defmac TARGET_LIBGCC_SDATA_SECTION
6963 If defined, a string which names the section into which small
6964 variables defined in crtstuff and libgcc should go. This is useful
6965 when the target has options for optimizing access to small data, and
6966 you want the crtstuff and libgcc routines to be conservative in what
6967 they expect of your application yet liberal in what your application
6968 expects. For example, for targets with a @code{.sdata} section (like
6969 MIPS), you could compile crtstuff with @code{-G 0} so that it doesn't
6970 require small data support from your application, but use this macro
6971 to put small data into @code{.sdata} so that your application can
6972 access these variables whether it uses small data or not.
6975 @defmac FORCE_CODE_SECTION_ALIGN
6976 If defined, an ASM statement that aligns a code section to some
6977 arbitrary boundary. This is used to force all fragments of the
6978 @code{.init} and @code{.fini} sections to have to same alignment
6979 and thus prevent the linker from having to add any padding.
6982 @defmac JUMP_TABLES_IN_TEXT_SECTION
6983 Define this macro to be an expression with a nonzero value if jump
6984 tables (for @code{tablejump} insns) should be output in the text
6985 section, along with the assembler instructions. Otherwise, the
6986 readonly data section is used.
6988 This macro is irrelevant if there is no separate readonly data section.
6991 @hook TARGET_ASM_INIT_SECTIONS
6992 Define this hook if you need to do something special to set up the
6993 @file{varasm.c} sections, or if your target has some special sections
6994 of its own that you need to create.
6996 GCC calls this hook after processing the command line, but before writing
6997 any assembly code, and before calling any of the section-returning hooks
7001 @hook TARGET_ASM_RELOC_RW_MASK
7002 Return a mask describing how relocations should be treated when
7003 selecting sections. Bit 1 should be set if global relocations
7004 should be placed in a read-write section; bit 0 should be set if
7005 local relocations should be placed in a read-write section.
7007 The default version of this function returns 3 when @option{-fpic}
7008 is in effect, and 0 otherwise. The hook is typically redefined
7009 when the target cannot support (some kinds of) dynamic relocations
7010 in read-only sections even in executables.
7013 @hook TARGET_ASM_SELECT_SECTION
7014 Return the section into which @var{exp} should be placed. You can
7015 assume that @var{exp} is either a @code{VAR_DECL} node or a constant of
7016 some sort. @var{reloc} indicates whether the initial value of @var{exp}
7017 requires link-time relocations. Bit 0 is set when variable contains
7018 local relocations only, while bit 1 is set for global relocations.
7019 @var{align} is the constant alignment in bits.
7021 The default version of this function takes care of putting read-only
7022 variables in @code{readonly_data_section}.
7024 See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}.
7027 @defmac USE_SELECT_SECTION_FOR_FUNCTIONS
7028 Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called
7029 for @code{FUNCTION_DECL}s as well as for variables and constants.
7031 In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the
7032 function has been determined to be likely to be called, and nonzero if
7033 it is unlikely to be called.
7036 @hook TARGET_ASM_UNIQUE_SECTION
7037 Build up a unique section name, expressed as a @code{STRING_CST} node,
7038 and assign it to @samp{DECL_SECTION_NAME (@var{decl})}.
7039 As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether
7040 the initial value of @var{exp} requires link-time relocations.
7042 The default version of this function appends the symbol name to the
7043 ELF section name that would normally be used for the symbol. For
7044 example, the function @code{foo} would be placed in @code{.text.foo}.
7045 Whatever the actual target object format, this is often good enough.
7048 @hook TARGET_ASM_FUNCTION_RODATA_SECTION
7049 Return the readonly data section associated with
7050 @samp{DECL_SECTION_NAME (@var{decl})}.
7051 The default version of this function selects @code{.gnu.linkonce.r.name} if
7052 the function's section is @code{.gnu.linkonce.t.name}, @code{.rodata.name}
7053 if function is in @code{.text.name}, and the normal readonly-data section
7057 @hook TARGET_ASM_SELECT_RTX_SECTION
7058 Return the section into which a constant @var{x}, of mode @var{mode},
7059 should be placed. You can assume that @var{x} is some kind of
7060 constant in RTL@. The argument @var{mode} is redundant except in the
7061 case of a @code{const_int} rtx. @var{align} is the constant alignment
7064 The default version of this function takes care of putting symbolic
7065 constants in @code{flag_pic} mode in @code{data_section} and everything
7066 else in @code{readonly_data_section}.
7069 @hook TARGET_MANGLE_DECL_ASSEMBLER_NAME
7070 Define this hook if you need to postprocess the assembler name generated
7071 by target-independent code. The @var{id} provided to this hook will be
7072 the computed name (e.g., the macro @code{DECL_NAME} of the @var{decl} in C,
7073 or the mangled name of the @var{decl} in C++). The return value of the
7074 hook is an @code{IDENTIFIER_NODE} for the appropriate mangled name on
7075 your target system. The default implementation of this hook just
7076 returns the @var{id} provided.
7079 @hook TARGET_ENCODE_SECTION_INFO
7080 Define this hook if references to a symbol or a constant must be
7081 treated differently depending on something about the variable or
7082 function named by the symbol (such as what section it is in).
7084 The hook is executed immediately after rtl has been created for
7085 @var{decl}, which may be a variable or function declaration or
7086 an entry in the constant pool. In either case, @var{rtl} is the
7087 rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})}
7088 in this hook; that field may not have been initialized yet.
7090 In the case of a constant, it is safe to assume that the rtl is
7091 a @code{mem} whose address is a @code{symbol_ref}. Most decls
7092 will also have this form, but that is not guaranteed. Global
7093 register variables, for instance, will have a @code{reg} for their
7094 rtl. (Normally the right thing to do with such unusual rtl is
7097 The @var{new_decl_p} argument will be true if this is the first time
7098 that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will
7099 be false for subsequent invocations, which will happen for duplicate
7100 declarations. Whether or not anything must be done for the duplicate
7101 declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}.
7102 @var{new_decl_p} is always true when the hook is called for a constant.
7104 @cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO}
7105 The usual thing for this hook to do is to record flags in the
7106 @code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}.
7107 Historically, the name string was modified if it was necessary to
7108 encode more than one bit of information, but this practice is now
7109 discouraged; use @code{SYMBOL_REF_FLAGS}.
7111 The default definition of this hook, @code{default_encode_section_info}
7112 in @file{varasm.c}, sets a number of commonly-useful bits in
7113 @code{SYMBOL_REF_FLAGS}. Check whether the default does what you need
7114 before overriding it.
7117 @hook TARGET_STRIP_NAME_ENCODING
7118 Decode @var{name} and return the real name part, sans
7119 the characters that @code{TARGET_ENCODE_SECTION_INFO}
7123 @hook TARGET_IN_SMALL_DATA_P
7124 Returns true if @var{exp} should be placed into a ``small data'' section.
7125 The default version of this hook always returns false.
7128 @hook TARGET_HAVE_SRODATA_SECTION
7129 Contains the value true if the target places read-only
7130 ``small data'' into a separate section. The default value is false.
7133 @hook TARGET_PROFILE_BEFORE_PROLOGUE
7135 @hook TARGET_BINDS_LOCAL_P
7136 Returns true if @var{exp} names an object for which name resolution
7137 rules must resolve to the current ``module'' (dynamic shared library
7138 or executable image).
7140 The default version of this hook implements the name resolution rules
7141 for ELF, which has a looser model of global name binding than other
7142 currently supported object file formats.
7145 @hook TARGET_HAVE_TLS
7146 Contains the value true if the target supports thread-local storage.
7147 The default value is false.
7152 @section Position Independent Code
7153 @cindex position independent code
7156 This section describes macros that help implement generation of position
7157 independent code. Simply defining these macros is not enough to
7158 generate valid PIC; you must also add support to the hook
7159 @code{TARGET_LEGITIMATE_ADDRESS_P} and to the macro
7160 @code{PRINT_OPERAND_ADDRESS}, as well as @code{LEGITIMIZE_ADDRESS}. You
7161 must modify the definition of @samp{movsi} to do something appropriate
7162 when the source operand contains a symbolic address. You may also
7163 need to alter the handling of switch statements so that they use
7165 @c i rearranged the order of the macros above to try to force one of
7166 @c them to the next line, to eliminate an overfull hbox. --mew 10feb93
7168 @defmac PIC_OFFSET_TABLE_REGNUM
7169 The register number of the register used to address a table of static
7170 data addresses in memory. In some cases this register is defined by a
7171 processor's ``application binary interface'' (ABI)@. When this macro
7172 is defined, RTL is generated for this register once, as with the stack
7173 pointer and frame pointer registers. If this macro is not defined, it
7174 is up to the machine-dependent files to allocate such a register (if
7175 necessary). Note that this register must be fixed when in use (e.g.@:
7176 when @code{flag_pic} is true).
7179 @defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED
7180 A C expression that is nonzero if the register defined by
7181 @code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. If not defined,
7182 the default is zero. Do not define
7183 this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined.
7186 @defmac LEGITIMATE_PIC_OPERAND_P (@var{x})
7187 A C expression that is nonzero if @var{x} is a legitimate immediate
7188 operand on the target machine when generating position independent code.
7189 You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not
7190 check this. You can also assume @var{flag_pic} is true, so you need not
7191 check it either. You need not define this macro if all constants
7192 (including @code{SYMBOL_REF}) can be immediate operands when generating
7193 position independent code.
7196 @node Assembler Format
7197 @section Defining the Output Assembler Language
7199 This section describes macros whose principal purpose is to describe how
7200 to write instructions in assembler language---rather than what the
7204 * File Framework:: Structural information for the assembler file.
7205 * Data Output:: Output of constants (numbers, strings, addresses).
7206 * Uninitialized Data:: Output of uninitialized variables.
7207 * Label Output:: Output and generation of labels.
7208 * Initialization:: General principles of initialization
7209 and termination routines.
7210 * Macros for Initialization::
7211 Specific macros that control the handling of
7212 initialization and termination routines.
7213 * Instruction Output:: Output of actual instructions.
7214 * Dispatch Tables:: Output of jump tables.
7215 * Exception Region Output:: Output of exception region code.
7216 * Alignment Output:: Pseudo ops for alignment and skipping data.
7219 @node File Framework
7220 @subsection The Overall Framework of an Assembler File
7221 @cindex assembler format
7222 @cindex output of assembler code
7224 @c prevent bad page break with this line
7225 This describes the overall framework of an assembly file.
7227 @findex default_file_start
7228 @hook TARGET_ASM_FILE_START
7229 Output to @code{asm_out_file} any text which the assembler expects to
7230 find at the beginning of a file. The default behavior is controlled
7231 by two flags, documented below. Unless your target's assembler is
7232 quite unusual, if you override the default, you should call
7233 @code{default_file_start} at some point in your target hook. This
7234 lets other target files rely on these variables.
7237 @hook TARGET_ASM_FILE_START_APP_OFF
7238 If this flag is true, the text of the macro @code{ASM_APP_OFF} will be
7239 printed as the very first line in the assembly file, unless
7240 @option{-fverbose-asm} is in effect. (If that macro has been defined
7241 to the empty string, this variable has no effect.) With the normal
7242 definition of @code{ASM_APP_OFF}, the effect is to notify the GNU
7243 assembler that it need not bother stripping comments or extra
7244 whitespace from its input. This allows it to work a bit faster.
7246 The default is false. You should not set it to true unless you have
7247 verified that your port does not generate any extra whitespace or
7248 comments that will cause GAS to issue errors in NO_APP mode.
7251 @hook TARGET_ASM_FILE_START_FILE_DIRECTIVE
7252 If this flag is true, @code{output_file_directive} will be called
7253 for the primary source file, immediately after printing
7254 @code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect
7255 this to be done. The default is false.
7258 @hook TARGET_ASM_FILE_END
7259 Output to @code{asm_out_file} any text which the assembler expects
7260 to find at the end of a file. The default is to output nothing.
7263 @deftypefun void file_end_indicate_exec_stack ()
7264 Some systems use a common convention, the @samp{.note.GNU-stack}
7265 special section, to indicate whether or not an object file relies on
7266 the stack being executable. If your system uses this convention, you
7267 should define @code{TARGET_ASM_FILE_END} to this function. If you
7268 need to do other things in that hook, have your hook function call
7272 @hook TARGET_ASM_LTO_START
7273 Output to @code{asm_out_file} any text which the assembler expects
7274 to find at the start of an LTO section. The default is to output
7278 @hook TARGET_ASM_LTO_END
7279 Output to @code{asm_out_file} any text which the assembler expects
7280 to find at the end of an LTO section. The default is to output
7284 @hook TARGET_ASM_CODE_END
7285 Output to @code{asm_out_file} any text which is needed before emitting
7286 unwind info and debug info at the end of a file. Some targets emit
7287 here PIC setup thunks that cannot be emitted at the end of file,
7288 because they couldn't have unwind info then. The default is to output
7292 @defmac ASM_COMMENT_START
7293 A C string constant describing how to begin a comment in the target
7294 assembler language. The compiler assumes that the comment will end at
7295 the end of the line.
7299 A C string constant for text to be output before each @code{asm}
7300 statement or group of consecutive ones. Normally this is
7301 @code{"#APP"}, which is a comment that has no effect on most
7302 assemblers but tells the GNU assembler that it must check the lines
7303 that follow for all valid assembler constructs.
7307 A C string constant for text to be output after each @code{asm}
7308 statement or group of consecutive ones. Normally this is
7309 @code{"#NO_APP"}, which tells the GNU assembler to resume making the
7310 time-saving assumptions that are valid for ordinary compiler output.
7313 @defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
7314 A C statement to output COFF information or DWARF debugging information
7315 which indicates that filename @var{name} is the current source file to
7316 the stdio stream @var{stream}.
7318 This macro need not be defined if the standard form of output
7319 for the file format in use is appropriate.
7322 @hook TARGET_ASM_OUTPUT_SOURCE_FILENAME
7324 @defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string})
7325 A C statement to output the string @var{string} to the stdio stream
7326 @var{stream}. If you do not call the function @code{output_quoted_string}
7327 in your config files, GCC will only call it to output filenames to
7328 the assembler source. So you can use it to canonicalize the format
7329 of the filename using this macro.
7332 @defmac ASM_OUTPUT_IDENT (@var{stream}, @var{string})
7333 A C statement to output something to the assembler file to handle a
7334 @samp{#ident} directive containing the text @var{string}. If this
7335 macro is not defined, nothing is output for a @samp{#ident} directive.
7338 @hook TARGET_ASM_NAMED_SECTION
7339 Output assembly directives to switch to section @var{name}. The section
7340 should have attributes as specified by @var{flags}, which is a bit mask
7341 of the @code{SECTION_*} flags defined in @file{output.h}. If @var{decl}
7342 is non-NULL, it is the @code{VAR_DECL} or @code{FUNCTION_DECL} with which
7343 this section is associated.
7346 @hook TARGET_HAVE_NAMED_SECTIONS
7347 This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}.
7350 @anchor{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}
7351 @hook TARGET_HAVE_SWITCHABLE_BSS_SECTIONS
7352 This flag is true if we can create zeroed data by switching to a BSS
7353 section and then using @code{ASM_OUTPUT_SKIP} to allocate the space.
7354 This is true on most ELF targets.
7357 @hook TARGET_SECTION_TYPE_FLAGS
7358 Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION}
7359 based on a variable or function decl, a section name, and whether or not the
7360 declaration's initializer may contain runtime relocations. @var{decl} may be
7361 null, in which case read-write data should be assumed.
7363 The default version of this function handles choosing code vs data,
7364 read-only vs read-write data, and @code{flag_pic}. You should only
7365 need to override this if your target has special flags that might be
7366 set via @code{__attribute__}.
7369 @hook TARGET_ASM_RECORD_GCC_SWITCHES
7370 Provides the target with the ability to record the gcc command line
7371 switches that have been passed to the compiler, and options that are
7372 enabled. The @var{type} argument specifies what is being recorded.
7373 It can take the following values:
7376 @item SWITCH_TYPE_PASSED
7377 @var{text} is a command line switch that has been set by the user.
7379 @item SWITCH_TYPE_ENABLED
7380 @var{text} is an option which has been enabled. This might be as a
7381 direct result of a command line switch, or because it is enabled by
7382 default or because it has been enabled as a side effect of a different
7383 command line switch. For example, the @option{-O2} switch enables
7384 various different individual optimization passes.
7386 @item SWITCH_TYPE_DESCRIPTIVE
7387 @var{text} is either NULL or some descriptive text which should be
7388 ignored. If @var{text} is NULL then it is being used to warn the
7389 target hook that either recording is starting or ending. The first
7390 time @var{type} is SWITCH_TYPE_DESCRIPTIVE and @var{text} is NULL, the
7391 warning is for start up and the second time the warning is for
7392 wind down. This feature is to allow the target hook to make any
7393 necessary preparations before it starts to record switches and to
7394 perform any necessary tidying up after it has finished recording
7397 @item SWITCH_TYPE_LINE_START
7398 This option can be ignored by this target hook.
7400 @item SWITCH_TYPE_LINE_END
7401 This option can be ignored by this target hook.
7404 The hook's return value must be zero. Other return values may be
7405 supported in the future.
7407 By default this hook is set to NULL, but an example implementation is
7408 provided for ELF based targets. Called @var{elf_record_gcc_switches},
7409 it records the switches as ASCII text inside a new, string mergeable
7410 section in the assembler output file. The name of the new section is
7411 provided by the @code{TARGET_ASM_RECORD_GCC_SWITCHES_SECTION} target
7415 @hook TARGET_ASM_RECORD_GCC_SWITCHES_SECTION
7416 This is the name of the section that will be created by the example
7417 ELF implementation of the @code{TARGET_ASM_RECORD_GCC_SWITCHES} target
7423 @subsection Output of Data
7426 @hook TARGET_ASM_BYTE_OP
7427 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP
7428 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP
7429 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP
7430 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP
7431 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP
7432 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP
7433 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP
7434 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP
7435 These hooks specify assembly directives for creating certain kinds
7436 of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a
7437 byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an
7438 aligned two-byte object, and so on. Any of the hooks may be
7439 @code{NULL}, indicating that no suitable directive is available.
7441 The compiler will print these strings at the start of a new line,
7442 followed immediately by the object's initial value. In most cases,
7443 the string should contain a tab, a pseudo-op, and then another tab.
7446 @hook TARGET_ASM_INTEGER
7447 The @code{assemble_integer} function uses this hook to output an
7448 integer object. @var{x} is the object's value, @var{size} is its size
7449 in bytes and @var{aligned_p} indicates whether it is aligned. The
7450 function should return @code{true} if it was able to output the
7451 object. If it returns false, @code{assemble_integer} will try to
7452 split the object into smaller parts.
7454 The default implementation of this hook will use the
7455 @code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false}
7456 when the relevant string is @code{NULL}.
7459 @hook TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA
7460 A target hook to recognize @var{rtx} patterns that @code{output_addr_const}
7461 can't deal with, and output assembly code to @var{file} corresponding to
7462 the pattern @var{x}. This may be used to allow machine-dependent
7463 @code{UNSPEC}s to appear within constants.
7465 If target hook fails to recognize a pattern, it must return @code{false},
7466 so that a standard error message is printed. If it prints an error message
7467 itself, by calling, for example, @code{output_operand_lossage}, it may just
7471 @defmac OUTPUT_ADDR_CONST_EXTRA (@var{stream}, @var{x}, @var{fail})
7472 A C statement to recognize @var{rtx} patterns that
7473 @code{output_addr_const} can't deal with, and output assembly code to
7474 @var{stream} corresponding to the pattern @var{x}. This may be used to
7475 allow machine-dependent @code{UNSPEC}s to appear within constants.
7477 If @code{OUTPUT_ADDR_CONST_EXTRA} fails to recognize a pattern, it must
7478 @code{goto fail}, so that a standard error message is printed. If it
7479 prints an error message itself, by calling, for example,
7480 @code{output_operand_lossage}, it may just complete normally.
7483 @defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
7484 A C statement to output to the stdio stream @var{stream} an assembler
7485 instruction to assemble a string constant containing the @var{len}
7486 bytes at @var{ptr}. @var{ptr} will be a C expression of type
7487 @code{char *} and @var{len} a C expression of type @code{int}.
7489 If the assembler has a @code{.ascii} pseudo-op as found in the
7490 Berkeley Unix assembler, do not define the macro
7491 @code{ASM_OUTPUT_ASCII}.
7494 @defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n})
7495 A C statement to output word @var{n} of a function descriptor for
7496 @var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS}
7497 is defined, and is otherwise unused.
7500 @defmac CONSTANT_POOL_BEFORE_FUNCTION
7501 You may define this macro as a C expression. You should define the
7502 expression to have a nonzero value if GCC should output the constant
7503 pool for a function before the code for the function, or a zero value if
7504 GCC should output the constant pool after the function. If you do
7505 not define this macro, the usual case, GCC will output the constant
7506 pool before the function.
7509 @defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size})
7510 A C statement to output assembler commands to define the start of the
7511 constant pool for a function. @var{funname} is a string giving
7512 the name of the function. Should the return type of the function
7513 be required, it can be obtained via @var{fundecl}. @var{size}
7514 is the size, in bytes, of the constant pool that will be written
7515 immediately after this call.
7517 If no constant-pool prefix is required, the usual case, this macro need
7521 @defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto})
7522 A C statement (with or without semicolon) to output a constant in the
7523 constant pool, if it needs special treatment. (This macro need not do
7524 anything for RTL expressions that can be output normally.)
7526 The argument @var{file} is the standard I/O stream to output the
7527 assembler code on. @var{x} is the RTL expression for the constant to
7528 output, and @var{mode} is the machine mode (in case @var{x} is a
7529 @samp{const_int}). @var{align} is the required alignment for the value
7530 @var{x}; you should output an assembler directive to force this much
7533 The argument @var{labelno} is a number to use in an internal label for
7534 the address of this pool entry. The definition of this macro is
7535 responsible for outputting the label definition at the proper place.
7536 Here is how to do this:
7539 @code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno});
7542 When you output a pool entry specially, you should end with a
7543 @code{goto} to the label @var{jumpto}. This will prevent the same pool
7544 entry from being output a second time in the usual manner.
7546 You need not define this macro if it would do nothing.
7549 @defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size})
7550 A C statement to output assembler commands to at the end of the constant
7551 pool for a function. @var{funname} is a string giving the name of the
7552 function. Should the return type of the function be required, you can
7553 obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the
7554 constant pool that GCC wrote immediately before this call.
7556 If no constant-pool epilogue is required, the usual case, you need not
7560 @defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C}, @var{STR})
7561 Define this macro as a C expression which is nonzero if @var{C} is
7562 used as a logical line separator by the assembler. @var{STR} points
7563 to the position in the string where @var{C} was found; this can be used if
7564 a line separator uses multiple characters.
7566 If you do not define this macro, the default is that only
7567 the character @samp{;} is treated as a logical line separator.
7570 @hook TARGET_ASM_OPEN_PAREN
7571 These target hooks are C string constants, describing the syntax in the
7572 assembler for grouping arithmetic expressions. If not overridden, they
7573 default to normal parentheses, which is correct for most assemblers.
7576 These macros are provided by @file{real.h} for writing the definitions
7577 of @code{ASM_OUTPUT_DOUBLE} and the like:
7579 @defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l})
7580 @defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l})
7581 @defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l})
7582 @defmacx REAL_VALUE_TO_TARGET_DECIMAL32 (@var{x}, @var{l})
7583 @defmacx REAL_VALUE_TO_TARGET_DECIMAL64 (@var{x}, @var{l})
7584 @defmacx REAL_VALUE_TO_TARGET_DECIMAL128 (@var{x}, @var{l})
7585 These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the
7586 target's floating point representation, and store its bit pattern in
7587 the variable @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE} and
7588 @code{REAL_VALUE_TO_TARGET_DECIMAL32}, this variable should be a
7589 simple @code{long int}. For the others, it should be an array of
7590 @code{long int}. The number of elements in this array is determined
7591 by the size of the desired target floating point data type: 32 bits of
7592 it go in each @code{long int} array element. Each array element holds
7593 32 bits of the result, even if @code{long int} is wider than 32 bits
7594 on the host machine.
7596 The array element values are designed so that you can print them out
7597 using @code{fprintf} in the order they should appear in the target
7601 @node Uninitialized Data
7602 @subsection Output of Uninitialized Variables
7604 Each of the macros in this section is used to do the whole job of
7605 outputting a single uninitialized variable.
7607 @defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
7608 A C statement (sans semicolon) to output to the stdio stream
7609 @var{stream} the assembler definition of a common-label named
7610 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7611 is the size rounded up to whatever alignment the caller wants. It is
7612 possible that @var{size} may be zero, for instance if a struct with no
7613 other member than a zero-length array is defined. In this case, the
7614 backend must output a symbol definition that allocates at least one
7615 byte, both so that the address of the resulting object does not compare
7616 equal to any other, and because some object formats cannot even express
7617 the concept of a zero-sized common symbol, as that is how they represent
7618 an ordinary undefined external.
7620 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7621 output the name itself; before and after that, output the additional
7622 assembler syntax for defining the name, and a newline.
7624 This macro controls how the assembler definitions of uninitialized
7625 common global variables are output.
7628 @defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment})
7629 Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a
7630 separate, explicit argument. If you define this macro, it is used in
7631 place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in
7632 handling the required alignment of the variable. The alignment is specified
7633 as the number of bits.
7636 @defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7637 Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the
7638 variable to be output, if there is one, or @code{NULL_TREE} if there
7639 is no corresponding variable. If you define this macro, GCC will use it
7640 in place of both @code{ASM_OUTPUT_COMMON} and
7641 @code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see
7642 the variable's decl in order to chose what to output.
7645 @defmac ASM_OUTPUT_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded})
7646 A C statement (sans semicolon) to output to the stdio stream
7647 @var{stream} the assembler definition of uninitialized global @var{decl} named
7648 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7649 is the size rounded up to whatever alignment the caller wants.
7651 Try to use function @code{asm_output_bss} defined in @file{varasm.c} when
7652 defining this macro. If unable, use the expression
7653 @code{assemble_name (@var{stream}, @var{name})} to output the name itself;
7654 before and after that, output the additional assembler syntax for defining
7655 the name, and a newline.
7657 There are two ways of handling global BSS@. One is to define either
7658 this macro or its aligned counterpart, @code{ASM_OUTPUT_ALIGNED_BSS}.
7659 The other is to have @code{TARGET_ASM_SELECT_SECTION} return a
7660 switchable BSS section (@pxref{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}).
7661 You do not need to do both.
7663 Some languages do not have @code{common} data, and require a
7664 non-common form of global BSS in order to handle uninitialized globals
7665 efficiently. C++ is one example of this. However, if the target does
7666 not support global BSS, the front end may choose to make globals
7667 common in order to save space in the object file.
7670 @defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7671 Like @code{ASM_OUTPUT_BSS} except takes the required alignment as a
7672 separate, explicit argument. If you define this macro, it is used in
7673 place of @code{ASM_OUTPUT_BSS}, and gives you more flexibility in
7674 handling the required alignment of the variable. The alignment is specified
7675 as the number of bits.
7677 Try to use function @code{asm_output_aligned_bss} defined in file
7678 @file{varasm.c} when defining this macro.
7681 @defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
7682 A C statement (sans semicolon) to output to the stdio stream
7683 @var{stream} the assembler definition of a local-common-label named
7684 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7685 is the size rounded up to whatever alignment the caller wants.
7687 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7688 output the name itself; before and after that, output the additional
7689 assembler syntax for defining the name, and a newline.
7691 This macro controls how the assembler definitions of uninitialized
7692 static variables are output.
7695 @defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment})
7696 Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a
7697 separate, explicit argument. If you define this macro, it is used in
7698 place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in
7699 handling the required alignment of the variable. The alignment is specified
7700 as the number of bits.
7703 @defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7704 Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the
7705 variable to be output, if there is one, or @code{NULL_TREE} if there
7706 is no corresponding variable. If you define this macro, GCC will use it
7707 in place of both @code{ASM_OUTPUT_DECL} and
7708 @code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see
7709 the variable's decl in order to chose what to output.
7713 @subsection Output and Generation of Labels
7715 @c prevent bad page break with this line
7716 This is about outputting labels.
7718 @findex assemble_name
7719 @defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name})
7720 A C statement (sans semicolon) to output to the stdio stream
7721 @var{stream} the assembler definition of a label named @var{name}.
7722 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7723 output the name itself; before and after that, output the additional
7724 assembler syntax for defining the name, and a newline. A default
7725 definition of this macro is provided which is correct for most systems.
7728 @defmac ASM_OUTPUT_FUNCTION_LABEL (@var{stream}, @var{name}, @var{decl})
7729 A C statement (sans semicolon) to output to the stdio stream
7730 @var{stream} the assembler definition of a label named @var{name} of
7732 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7733 output the name itself; before and after that, output the additional
7734 assembler syntax for defining the name, and a newline. A default
7735 definition of this macro is provided which is correct for most systems.
7737 If this macro is not defined, then the function name is defined in the
7738 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7741 @findex assemble_name_raw
7742 @defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name})
7743 Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known
7744 to refer to a compiler-generated label. The default definition uses
7745 @code{assemble_name_raw}, which is like @code{assemble_name} except
7746 that it is more efficient.
7750 A C string containing the appropriate assembler directive to specify the
7751 size of a symbol, without any arguments. On systems that use ELF, the
7752 default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other
7753 systems, the default is not to define this macro.
7755 Define this macro only if it is correct to use the default definitions
7756 of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE}
7757 for your system. If you need your own custom definitions of those
7758 macros, or if you do not need explicit symbol sizes at all, do not
7762 @defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size})
7763 A C statement (sans semicolon) to output to the stdio stream
7764 @var{stream} a directive telling the assembler that the size of the
7765 symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}.
7766 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
7770 @defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name})
7771 A C statement (sans semicolon) to output to the stdio stream
7772 @var{stream} a directive telling the assembler to calculate the size of
7773 the symbol @var{name} by subtracting its address from the current
7776 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
7777 provided. The default assumes that the assembler recognizes a special
7778 @samp{.} symbol as referring to the current address, and can calculate
7779 the difference between this and another symbol. If your assembler does
7780 not recognize @samp{.} or cannot do calculations with it, you will need
7781 to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique.
7785 A C string containing the appropriate assembler directive to specify the
7786 type of a symbol, without any arguments. On systems that use ELF, the
7787 default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other
7788 systems, the default is not to define this macro.
7790 Define this macro only if it is correct to use the default definition of
7791 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
7792 custom definition of this macro, or if you do not need explicit symbol
7793 types at all, do not define this macro.
7796 @defmac TYPE_OPERAND_FMT
7797 A C string which specifies (using @code{printf} syntax) the format of
7798 the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the
7799 default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems,
7800 the default is not to define this macro.
7802 Define this macro only if it is correct to use the default definition of
7803 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
7804 custom definition of this macro, or if you do not need explicit symbol
7805 types at all, do not define this macro.
7808 @defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type})
7809 A C statement (sans semicolon) to output to the stdio stream
7810 @var{stream} a directive telling the assembler that the type of the
7811 symbol @var{name} is @var{type}. @var{type} is a C string; currently,
7812 that string is always either @samp{"function"} or @samp{"object"}, but
7813 you should not count on this.
7815 If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default
7816 definition of this macro is provided.
7819 @defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
7820 A C statement (sans semicolon) to output to the stdio stream
7821 @var{stream} any text necessary for declaring the name @var{name} of a
7822 function which is being defined. This macro is responsible for
7823 outputting the label definition (perhaps using
7824 @code{ASM_OUTPUT_FUNCTION_LABEL}). The argument @var{decl} is the
7825 @code{FUNCTION_DECL} tree node representing the function.
7827 If this macro is not defined, then the function name is defined in the
7828 usual manner as a label (by means of @code{ASM_OUTPUT_FUNCTION_LABEL}).
7830 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
7834 @defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl})
7835 A C statement (sans semicolon) to output to the stdio stream
7836 @var{stream} any text necessary for declaring the size of a function
7837 which is being defined. The argument @var{name} is the name of the
7838 function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node
7839 representing the function.
7841 If this macro is not defined, then the function size is not defined.
7843 You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition
7847 @defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl})
7848 A C statement (sans semicolon) to output to the stdio stream
7849 @var{stream} any text necessary for declaring the name @var{name} of an
7850 initialized variable which is being defined. This macro must output the
7851 label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument
7852 @var{decl} is the @code{VAR_DECL} tree node representing the variable.
7854 If this macro is not defined, then the variable name is defined in the
7855 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7857 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or
7858 @code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro.
7861 @hook TARGET_ASM_DECLARE_CONSTANT_NAME
7862 A target hook to output to the stdio stream @var{file} any text necessary
7863 for declaring the name @var{name} of a constant which is being defined. This
7864 target hook is responsible for outputting the label definition (perhaps using
7865 @code{assemble_label}). The argument @var{exp} is the value of the constant,
7866 and @var{size} is the size of the constant in bytes. The @var{name}
7867 will be an internal label.
7869 The default version of this target hook, define the @var{name} in the
7870 usual manner as a label (by means of @code{assemble_label}).
7872 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in this target hook.
7875 @defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name})
7876 A C statement (sans semicolon) to output to the stdio stream
7877 @var{stream} any text necessary for claiming a register @var{regno}
7878 for a global variable @var{decl} with name @var{name}.
7880 If you don't define this macro, that is equivalent to defining it to do
7884 @defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend})
7885 A C statement (sans semicolon) to finish up declaring a variable name
7886 once the compiler has processed its initializer fully and thus has had a
7887 chance to determine the size of an array when controlled by an
7888 initializer. This is used on systems where it's necessary to declare
7889 something about the size of the object.
7891 If you don't define this macro, that is equivalent to defining it to do
7894 You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or
7895 @code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro.
7898 @hook TARGET_ASM_GLOBALIZE_LABEL
7899 This target hook is a function to output to the stdio stream
7900 @var{stream} some commands that will make the label @var{name} global;
7901 that is, available for reference from other files.
7903 The default implementation relies on a proper definition of
7904 @code{GLOBAL_ASM_OP}.
7907 @hook TARGET_ASM_GLOBALIZE_DECL_NAME
7908 This target hook is a function to output to the stdio stream
7909 @var{stream} some commands that will make the name associated with @var{decl}
7910 global; that is, available for reference from other files.
7912 The default implementation uses the TARGET_ASM_GLOBALIZE_LABEL target hook.
7915 @defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name})
7916 A C statement (sans semicolon) to output to the stdio stream
7917 @var{stream} some commands that will make the label @var{name} weak;
7918 that is, available for reference from other files but only used if
7919 no other definition is available. Use the expression
7920 @code{assemble_name (@var{stream}, @var{name})} to output the name
7921 itself; before and after that, output the additional assembler syntax
7922 for making that name weak, and a newline.
7924 If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not
7925 support weak symbols and you should not define the @code{SUPPORTS_WEAK}
7929 @defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value})
7930 Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and
7931 @code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function
7932 or variable decl. If @var{value} is not @code{NULL}, this C statement
7933 should output to the stdio stream @var{stream} assembler code which
7934 defines (equates) the weak symbol @var{name} to have the value
7935 @var{value}. If @var{value} is @code{NULL}, it should output commands
7936 to make @var{name} weak.
7939 @defmac ASM_OUTPUT_WEAKREF (@var{stream}, @var{decl}, @var{name}, @var{value})
7940 Outputs a directive that enables @var{name} to be used to refer to
7941 symbol @var{value} with weak-symbol semantics. @code{decl} is the
7942 declaration of @code{name}.
7945 @defmac SUPPORTS_WEAK
7946 A preprocessor constant expression which evaluates to true if the target
7947 supports weak symbols.
7949 If you don't define this macro, @file{defaults.h} provides a default
7950 definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL}
7951 is defined, the default definition is @samp{1}; otherwise, it is @samp{0}.
7954 @defmac TARGET_SUPPORTS_WEAK
7955 A C expression which evaluates to true if the target supports weak symbols.
7957 If you don't define this macro, @file{defaults.h} provides a default
7958 definition. The default definition is @samp{(SUPPORTS_WEAK)}. Define
7959 this macro if you want to control weak symbol support with a compiler
7960 flag such as @option{-melf}.
7963 @defmac MAKE_DECL_ONE_ONLY (@var{decl})
7964 A C statement (sans semicolon) to mark @var{decl} to be emitted as a
7965 public symbol such that extra copies in multiple translation units will
7966 be discarded by the linker. Define this macro if your object file
7967 format provides support for this concept, such as the @samp{COMDAT}
7968 section flags in the Microsoft Windows PE/COFF format, and this support
7969 requires changes to @var{decl}, such as putting it in a separate section.
7972 @defmac SUPPORTS_ONE_ONLY
7973 A C expression which evaluates to true if the target supports one-only
7976 If you don't define this macro, @file{varasm.c} provides a default
7977 definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default
7978 definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if
7979 you want to control one-only symbol support with a compiler flag, or if
7980 setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to
7981 be emitted as one-only.
7984 @hook TARGET_ASM_ASSEMBLE_VISIBILITY
7985 This target hook is a function to output to @var{asm_out_file} some
7986 commands that will make the symbol(s) associated with @var{decl} have
7987 hidden, protected or internal visibility as specified by @var{visibility}.
7990 @defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC
7991 A C expression that evaluates to true if the target's linker expects
7992 that weak symbols do not appear in a static archive's table of contents.
7993 The default is @code{0}.
7995 Leaving weak symbols out of an archive's table of contents means that,
7996 if a symbol will only have a definition in one translation unit and
7997 will have undefined references from other translation units, that
7998 symbol should not be weak. Defining this macro to be nonzero will
7999 thus have the effect that certain symbols that would normally be weak
8000 (explicit template instantiations, and vtables for polymorphic classes
8001 with noninline key methods) will instead be nonweak.
8003 The C++ ABI requires this macro to be zero. Define this macro for
8004 targets where full C++ ABI compliance is impossible and where linker
8005 restrictions require weak symbols to be left out of a static archive's
8009 @defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
8010 A C statement (sans semicolon) to output to the stdio stream
8011 @var{stream} any text necessary for declaring the name of an external
8012 symbol named @var{name} which is referenced in this compilation but
8013 not defined. The value of @var{decl} is the tree node for the
8016 This macro need not be defined if it does not need to output anything.
8017 The GNU assembler and most Unix assemblers don't require anything.
8020 @hook TARGET_ASM_EXTERNAL_LIBCALL
8021 This target hook is a function to output to @var{asm_out_file} an assembler
8022 pseudo-op to declare a library function name external. The name of the
8023 library function is given by @var{symref}, which is a @code{symbol_ref}.
8026 @hook TARGET_ASM_MARK_DECL_PRESERVED
8027 This target hook is a function to output to @var{asm_out_file} an assembler
8028 directive to annotate @var{symbol} as used. The Darwin target uses the
8029 .no_dead_code_strip directive.
8032 @defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
8033 A C statement (sans semicolon) to output to the stdio stream
8034 @var{stream} a reference in assembler syntax to a label named
8035 @var{name}. This should add @samp{_} to the front of the name, if that
8036 is customary on your operating system, as it is in most Berkeley Unix
8037 systems. This macro is used in @code{assemble_name}.
8040 @defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym})
8041 A C statement (sans semicolon) to output a reference to
8042 @code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name}
8043 will be used to output the name of the symbol. This macro may be used
8044 to modify the way a symbol is referenced depending on information
8045 encoded by @code{TARGET_ENCODE_SECTION_INFO}.
8048 @defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf})
8049 A C statement (sans semicolon) to output a reference to @var{buf}, the
8050 result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined,
8051 @code{assemble_name} will be used to output the name of the symbol.
8052 This macro is not used by @code{output_asm_label}, or the @code{%l}
8053 specifier that calls it; the intention is that this macro should be set
8054 when it is necessary to output a label differently when its address is
8058 @hook TARGET_ASM_INTERNAL_LABEL
8059 A function to output to the stdio stream @var{stream} a label whose
8060 name is made from the string @var{prefix} and the number @var{labelno}.
8062 It is absolutely essential that these labels be distinct from the labels
8063 used for user-level functions and variables. Otherwise, certain programs
8064 will have name conflicts with internal labels.
8066 It is desirable to exclude internal labels from the symbol table of the
8067 object file. Most assemblers have a naming convention for labels that
8068 should be excluded; on many systems, the letter @samp{L} at the
8069 beginning of a label has this effect. You should find out what
8070 convention your system uses, and follow it.
8072 The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}.
8075 @defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num})
8076 A C statement to output to the stdio stream @var{stream} a debug info
8077 label whose name is made from the string @var{prefix} and the number
8078 @var{num}. This is useful for VLIW targets, where debug info labels
8079 may need to be treated differently than branch target labels. On some
8080 systems, branch target labels must be at the beginning of instruction
8081 bundles, but debug info labels can occur in the middle of instruction
8084 If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be
8088 @defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
8089 A C statement to store into the string @var{string} a label whose name
8090 is made from the string @var{prefix} and the number @var{num}.
8092 This string, when output subsequently by @code{assemble_name}, should
8093 produce the output that @code{(*targetm.asm_out.internal_label)} would produce
8094 with the same @var{prefix} and @var{num}.
8096 If the string begins with @samp{*}, then @code{assemble_name} will
8097 output the rest of the string unchanged. It is often convenient for
8098 @code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the
8099 string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets
8100 to output the string, and may change it. (Of course,
8101 @code{ASM_OUTPUT_LABELREF} is also part of your machine description, so
8102 you should know what it does on your machine.)
8105 @defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
8106 A C expression to assign to @var{outvar} (which is a variable of type
8107 @code{char *}) a newly allocated string made from the string
8108 @var{name} and the number @var{number}, with some suitable punctuation
8109 added. Use @code{alloca} to get space for the string.
8111 The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to
8112 produce an assembler label for an internal static variable whose name is
8113 @var{name}. Therefore, the string must be such as to result in valid
8114 assembler code. The argument @var{number} is different each time this
8115 macro is executed; it prevents conflicts between similarly-named
8116 internal static variables in different scopes.
8118 Ideally this string should not be a valid C identifier, to prevent any
8119 conflict with the user's own symbols. Most assemblers allow periods
8120 or percent signs in assembler symbols; putting at least one of these
8121 between the name and the number will suffice.
8123 If this macro is not defined, a default definition will be provided
8124 which is correct for most systems.
8127 @defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value})
8128 A C statement to output to the stdio stream @var{stream} assembler code
8129 which defines (equates) the symbol @var{name} to have the value @var{value}.
8132 If @code{SET_ASM_OP} is defined, a default definition is provided which is
8133 correct for most systems.
8136 @defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value})
8137 A C statement to output to the stdio stream @var{stream} assembler code
8138 which defines (equates) the symbol whose tree node is @var{decl_of_name}
8139 to have the value of the tree node @var{decl_of_value}. This macro will
8140 be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if
8141 the tree nodes are available.
8144 If @code{SET_ASM_OP} is defined, a default definition is provided which is
8145 correct for most systems.
8148 @defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value})
8149 A C statement that evaluates to true if the assembler code which defines
8150 (equates) the symbol whose tree node is @var{decl_of_name} to have the value
8151 of the tree node @var{decl_of_value} should be emitted near the end of the
8152 current compilation unit. The default is to not defer output of defines.
8153 This macro affects defines output by @samp{ASM_OUTPUT_DEF} and
8154 @samp{ASM_OUTPUT_DEF_FROM_DECLS}.
8157 @defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value})
8158 A C statement to output to the stdio stream @var{stream} assembler code
8159 which defines (equates) the weak symbol @var{name} to have the value
8160 @var{value}. If @var{value} is @code{NULL}, it defines @var{name} as
8161 an undefined weak symbol.
8163 Define this macro if the target only supports weak aliases; define
8164 @code{ASM_OUTPUT_DEF} instead if possible.
8167 @defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name})
8168 Define this macro to override the default assembler names used for
8169 Objective-C methods.
8171 The default name is a unique method number followed by the name of the
8172 class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of
8173 the category is also included in the assembler name (e.g.@:
8176 These names are safe on most systems, but make debugging difficult since
8177 the method's selector is not present in the name. Therefore, particular
8178 systems define other ways of computing names.
8180 @var{buf} is an expression of type @code{char *} which gives you a
8181 buffer in which to store the name; its length is as long as
8182 @var{class_name}, @var{cat_name} and @var{sel_name} put together, plus
8183 50 characters extra.
8185 The argument @var{is_inst} specifies whether the method is an instance
8186 method or a class method; @var{class_name} is the name of the class;
8187 @var{cat_name} is the name of the category (or @code{NULL} if the method is not
8188 in a category); and @var{sel_name} is the name of the selector.
8190 On systems where the assembler can handle quoted names, you can use this
8191 macro to provide more human-readable names.
8194 @defmac ASM_DECLARE_CLASS_REFERENCE (@var{stream}, @var{name})
8195 A C statement (sans semicolon) to output to the stdio stream
8196 @var{stream} commands to declare that the label @var{name} is an
8197 Objective-C class reference. This is only needed for targets whose
8198 linkers have special support for NeXT-style runtimes.
8201 @defmac ASM_DECLARE_UNRESOLVED_REFERENCE (@var{stream}, @var{name})
8202 A C statement (sans semicolon) to output to the stdio stream
8203 @var{stream} commands to declare that the label @var{name} is an
8204 unresolved Objective-C class reference. This is only needed for targets
8205 whose linkers have special support for NeXT-style runtimes.
8208 @node Initialization
8209 @subsection How Initialization Functions Are Handled
8210 @cindex initialization routines
8211 @cindex termination routines
8212 @cindex constructors, output of
8213 @cindex destructors, output of
8215 The compiled code for certain languages includes @dfn{constructors}
8216 (also called @dfn{initialization routines})---functions to initialize
8217 data in the program when the program is started. These functions need
8218 to be called before the program is ``started''---that is to say, before
8219 @code{main} is called.
8221 Compiling some languages generates @dfn{destructors} (also called
8222 @dfn{termination routines}) that should be called when the program
8225 To make the initialization and termination functions work, the compiler
8226 must output something in the assembler code to cause those functions to
8227 be called at the appropriate time. When you port the compiler to a new
8228 system, you need to specify how to do this.
8230 There are two major ways that GCC currently supports the execution of
8231 initialization and termination functions. Each way has two variants.
8232 Much of the structure is common to all four variations.
8234 @findex __CTOR_LIST__
8235 @findex __DTOR_LIST__
8236 The linker must build two lists of these functions---a list of
8237 initialization functions, called @code{__CTOR_LIST__}, and a list of
8238 termination functions, called @code{__DTOR_LIST__}.
8240 Each list always begins with an ignored function pointer (which may hold
8241 0, @minus{}1, or a count of the function pointers after it, depending on
8242 the environment). This is followed by a series of zero or more function
8243 pointers to constructors (or destructors), followed by a function
8244 pointer containing zero.
8246 Depending on the operating system and its executable file format, either
8247 @file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup
8248 time and exit time. Constructors are called in reverse order of the
8249 list; destructors in forward order.
8251 The best way to handle static constructors works only for object file
8252 formats which provide arbitrarily-named sections. A section is set
8253 aside for a list of constructors, and another for a list of destructors.
8254 Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each
8255 object file that defines an initialization function also puts a word in
8256 the constructor section to point to that function. The linker
8257 accumulates all these words into one contiguous @samp{.ctors} section.
8258 Termination functions are handled similarly.
8260 This method will be chosen as the default by @file{target-def.h} if
8261 @code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not
8262 support arbitrary sections, but does support special designated
8263 constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP}
8264 and @code{DTORS_SECTION_ASM_OP} to achieve the same effect.
8266 When arbitrary sections are available, there are two variants, depending
8267 upon how the code in @file{crtstuff.c} is called. On systems that
8268 support a @dfn{.init} section which is executed at program startup,
8269 parts of @file{crtstuff.c} are compiled into that section. The
8270 program is linked by the @command{gcc} driver like this:
8273 ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o
8276 The prologue of a function (@code{__init}) appears in the @code{.init}
8277 section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise
8278 for the function @code{__fini} in the @dfn{.fini} section. Normally these
8279 files are provided by the operating system or by the GNU C library, but
8280 are provided by GCC for a few targets.
8282 The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets)
8283 compiled from @file{crtstuff.c}. They contain, among other things, code
8284 fragments within the @code{.init} and @code{.fini} sections that branch
8285 to routines in the @code{.text} section. The linker will pull all parts
8286 of a section together, which results in a complete @code{__init} function
8287 that invokes the routines we need at startup.
8289 To use this variant, you must define the @code{INIT_SECTION_ASM_OP}
8292 If no init section is available, when GCC compiles any function called
8293 @code{main} (or more accurately, any function designated as a program
8294 entry point by the language front end calling @code{expand_main_function}),
8295 it inserts a procedure call to @code{__main} as the first executable code
8296 after the function prologue. The @code{__main} function is defined
8297 in @file{libgcc2.c} and runs the global constructors.
8299 In file formats that don't support arbitrary sections, there are again
8300 two variants. In the simplest variant, the GNU linker (GNU @code{ld})
8301 and an `a.out' format must be used. In this case,
8302 @code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs}
8303 entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__},
8304 and with the address of the void function containing the initialization
8305 code as its value. The GNU linker recognizes this as a request to add
8306 the value to a @dfn{set}; the values are accumulated, and are eventually
8307 placed in the executable as a vector in the format described above, with
8308 a leading (ignored) count and a trailing zero element.
8309 @code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init
8310 section is available, the absence of @code{INIT_SECTION_ASM_OP} causes
8311 the compilation of @code{main} to call @code{__main} as above, starting
8312 the initialization process.
8314 The last variant uses neither arbitrary sections nor the GNU linker.
8315 This is preferable when you want to do dynamic linking and when using
8316 file formats which the GNU linker does not support, such as `ECOFF'@. In
8317 this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and
8318 termination functions are recognized simply by their names. This requires
8319 an extra program in the linkage step, called @command{collect2}. This program
8320 pretends to be the linker, for use with GCC; it does its job by running
8321 the ordinary linker, but also arranges to include the vectors of
8322 initialization and termination functions. These functions are called
8323 via @code{__main} as described above. In order to use this method,
8324 @code{use_collect2} must be defined in the target in @file{config.gcc}.
8327 The following section describes the specific macros that control and
8328 customize the handling of initialization and termination functions.
8331 @node Macros for Initialization
8332 @subsection Macros Controlling Initialization Routines
8334 Here are the macros that control how the compiler handles initialization
8335 and termination functions:
8337 @defmac INIT_SECTION_ASM_OP
8338 If defined, a C string constant, including spacing, for the assembler
8339 operation to identify the following data as initialization code. If not
8340 defined, GCC will assume such a section does not exist. When you are
8341 using special sections for initialization and termination functions, this
8342 macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to
8343 run the initialization functions.
8346 @defmac HAS_INIT_SECTION
8347 If defined, @code{main} will not call @code{__main} as described above.
8348 This macro should be defined for systems that control start-up code
8349 on a symbol-by-symbol basis, such as OSF/1, and should not
8350 be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}.
8353 @defmac LD_INIT_SWITCH
8354 If defined, a C string constant for a switch that tells the linker that
8355 the following symbol is an initialization routine.
8358 @defmac LD_FINI_SWITCH
8359 If defined, a C string constant for a switch that tells the linker that
8360 the following symbol is a finalization routine.
8363 @defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func})
8364 If defined, a C statement that will write a function that can be
8365 automatically called when a shared library is loaded. The function
8366 should call @var{func}, which takes no arguments. If not defined, and
8367 the object format requires an explicit initialization function, then a
8368 function called @code{_GLOBAL__DI} will be generated.
8370 This function and the following one are used by collect2 when linking a
8371 shared library that needs constructors or destructors, or has DWARF2
8372 exception tables embedded in the code.
8375 @defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func})
8376 If defined, a C statement that will write a function that can be
8377 automatically called when a shared library is unloaded. The function
8378 should call @var{func}, which takes no arguments. If not defined, and
8379 the object format requires an explicit finalization function, then a
8380 function called @code{_GLOBAL__DD} will be generated.
8383 @defmac INVOKE__main
8384 If defined, @code{main} will call @code{__main} despite the presence of
8385 @code{INIT_SECTION_ASM_OP}. This macro should be defined for systems
8386 where the init section is not actually run automatically, but is still
8387 useful for collecting the lists of constructors and destructors.
8390 @defmac SUPPORTS_INIT_PRIORITY
8391 If nonzero, the C++ @code{init_priority} attribute is supported and the
8392 compiler should emit instructions to control the order of initialization
8393 of objects. If zero, the compiler will issue an error message upon
8394 encountering an @code{init_priority} attribute.
8397 @hook TARGET_HAVE_CTORS_DTORS
8398 This value is true if the target supports some ``native'' method of
8399 collecting constructors and destructors to be run at startup and exit.
8400 It is false if we must use @command{collect2}.
8403 @hook TARGET_ASM_CONSTRUCTOR
8404 If defined, a function that outputs assembler code to arrange to call
8405 the function referenced by @var{symbol} at initialization time.
8407 Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking
8408 no arguments and with no return value. If the target supports initialization
8409 priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY};
8410 otherwise it must be @code{DEFAULT_INIT_PRIORITY}.
8412 If this macro is not defined by the target, a suitable default will
8413 be chosen if (1) the target supports arbitrary section names, (2) the
8414 target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2}
8418 @hook TARGET_ASM_DESTRUCTOR
8419 This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination
8420 functions rather than initialization functions.
8423 If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine
8424 generated for the generated object file will have static linkage.
8426 If your system uses @command{collect2} as the means of processing
8427 constructors, then that program normally uses @command{nm} to scan
8428 an object file for constructor functions to be called.
8430 On certain kinds of systems, you can define this macro to make
8431 @command{collect2} work faster (and, in some cases, make it work at all):
8433 @defmac OBJECT_FORMAT_COFF
8434 Define this macro if the system uses COFF (Common Object File Format)
8435 object files, so that @command{collect2} can assume this format and scan
8436 object files directly for dynamic constructor/destructor functions.
8438 This macro is effective only in a native compiler; @command{collect2} as
8439 part of a cross compiler always uses @command{nm} for the target machine.
8442 @defmac REAL_NM_FILE_NAME
8443 Define this macro as a C string constant containing the file name to use
8444 to execute @command{nm}. The default is to search the path normally for
8447 If your system supports shared libraries and has a program to list the
8448 dynamic dependencies of a given library or executable, you can define
8449 these macros to enable support for running initialization and
8450 termination functions in shared libraries:
8454 Define this macro to a C string constant containing the name of the program
8455 which lists dynamic dependencies, like @command{"ldd"} under SunOS 4.
8458 @defmac PARSE_LDD_OUTPUT (@var{ptr})
8459 Define this macro to be C code that extracts filenames from the output
8460 of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable
8461 of type @code{char *} that points to the beginning of a line of output
8462 from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the
8463 code must advance @var{ptr} to the beginning of the filename on that
8464 line. Otherwise, it must set @var{ptr} to @code{NULL}.
8467 @defmac SHLIB_SUFFIX
8468 Define this macro to a C string constant containing the default shared
8469 library extension of the target (e.g., @samp{".so"}). @command{collect2}
8470 strips version information after this suffix when generating global
8471 constructor and destructor names. This define is only needed on targets
8472 that use @command{collect2} to process constructors and destructors.
8475 @node Instruction Output
8476 @subsection Output of Assembler Instructions
8478 @c prevent bad page break with this line
8479 This describes assembler instruction output.
8481 @defmac REGISTER_NAMES
8482 A C initializer containing the assembler's names for the machine
8483 registers, each one as a C string constant. This is what translates
8484 register numbers in the compiler into assembler language.
8487 @defmac ADDITIONAL_REGISTER_NAMES
8488 If defined, a C initializer for an array of structures containing a name
8489 and a register number. This macro defines additional names for hard
8490 registers, thus allowing the @code{asm} option in declarations to refer
8491 to registers using alternate names.
8494 @defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
8495 Define this macro if you are using an unusual assembler that
8496 requires different names for the machine instructions.
8498 The definition is a C statement or statements which output an
8499 assembler instruction opcode to the stdio stream @var{stream}. The
8500 macro-operand @var{ptr} is a variable of type @code{char *} which
8501 points to the opcode name in its ``internal'' form---the form that is
8502 written in the machine description. The definition should output the
8503 opcode name to @var{stream}, performing any translation you desire, and
8504 increment the variable @var{ptr} to point at the end of the opcode
8505 so that it will not be output twice.
8507 In fact, your macro definition may process less than the entire opcode
8508 name, or more than the opcode name; but if you want to process text
8509 that includes @samp{%}-sequences to substitute operands, you must take
8510 care of the substitution yourself. Just be sure to increment
8511 @var{ptr} over whatever text should not be output normally.
8513 @findex recog_data.operand
8514 If you need to look at the operand values, they can be found as the
8515 elements of @code{recog_data.operand}.
8517 If the macro definition does nothing, the instruction is output
8521 @defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
8522 If defined, a C statement to be executed just prior to the output of
8523 assembler code for @var{insn}, to modify the extracted operands so
8524 they will be output differently.
8526 Here the argument @var{opvec} is the vector containing the operands
8527 extracted from @var{insn}, and @var{noperands} is the number of
8528 elements of the vector which contain meaningful data for this insn.
8529 The contents of this vector are what will be used to convert the insn
8530 template into assembler code, so you can change the assembler output
8531 by changing the contents of the vector.
8533 This macro is useful when various assembler syntaxes share a single
8534 file of instruction patterns; by defining this macro differently, you
8535 can cause a large class of instructions to be output differently (such
8536 as with rearranged operands). Naturally, variations in assembler
8537 syntax affecting individual insn patterns ought to be handled by
8538 writing conditional output routines in those patterns.
8540 If this macro is not defined, it is equivalent to a null statement.
8543 @hook TARGET_ASM_FINAL_POSTSCAN_INSN
8544 If defined, this target hook is a function which is executed just after the
8545 output of assembler code for @var{insn}, to change the mode of the assembler
8548 Here the argument @var{opvec} is the vector containing the operands
8549 extracted from @var{insn}, and @var{noperands} is the number of
8550 elements of the vector which contain meaningful data for this insn.
8551 The contents of this vector are what was used to convert the insn
8552 template into assembler code, so you can change the assembler mode
8553 by checking the contents of the vector.
8556 @defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
8557 A C compound statement to output to stdio stream @var{stream} the
8558 assembler syntax for an instruction operand @var{x}. @var{x} is an
8561 @var{code} is a value that can be used to specify one of several ways
8562 of printing the operand. It is used when identical operands must be
8563 printed differently depending on the context. @var{code} comes from
8564 the @samp{%} specification that was used to request printing of the
8565 operand. If the specification was just @samp{%@var{digit}} then
8566 @var{code} is 0; if the specification was @samp{%@var{ltr}
8567 @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
8570 If @var{x} is a register, this macro should print the register's name.
8571 The names can be found in an array @code{reg_names} whose type is
8572 @code{char *[]}. @code{reg_names} is initialized from
8573 @code{REGISTER_NAMES}.
8575 When the machine description has a specification @samp{%@var{punct}}
8576 (a @samp{%} followed by a punctuation character), this macro is called
8577 with a null pointer for @var{x} and the punctuation character for
8581 @defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code})
8582 A C expression which evaluates to true if @var{code} is a valid
8583 punctuation character for use in the @code{PRINT_OPERAND} macro. If
8584 @code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
8585 punctuation characters (except for the standard one, @samp{%}) are used
8589 @defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
8590 A C compound statement to output to stdio stream @var{stream} the
8591 assembler syntax for an instruction operand that is a memory reference
8592 whose address is @var{x}. @var{x} is an RTL expression.
8594 @cindex @code{TARGET_ENCODE_SECTION_INFO} usage
8595 On some machines, the syntax for a symbolic address depends on the
8596 section that the address refers to. On these machines, define the hook
8597 @code{TARGET_ENCODE_SECTION_INFO} to store the information into the
8598 @code{symbol_ref}, and then check for it here. @xref{Assembler
8602 @findex dbr_sequence_length
8603 @defmac DBR_OUTPUT_SEQEND (@var{file})
8604 A C statement, to be executed after all slot-filler instructions have
8605 been output. If necessary, call @code{dbr_sequence_length} to
8606 determine the number of slots filled in a sequence (zero if not
8607 currently outputting a sequence), to decide how many no-ops to output,
8610 Don't define this macro if it has nothing to do, but it is helpful in
8611 reading assembly output if the extent of the delay sequence is made
8612 explicit (e.g.@: with white space).
8615 @findex final_sequence
8616 Note that output routines for instructions with delay slots must be
8617 prepared to deal with not being output as part of a sequence
8618 (i.e.@: when the scheduling pass is not run, or when no slot fillers could be
8619 found.) The variable @code{final_sequence} is null when not
8620 processing a sequence, otherwise it contains the @code{sequence} rtx
8624 @defmac REGISTER_PREFIX
8625 @defmacx LOCAL_LABEL_PREFIX
8626 @defmacx USER_LABEL_PREFIX
8627 @defmacx IMMEDIATE_PREFIX
8628 If defined, C string expressions to be used for the @samp{%R}, @samp{%L},
8629 @samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see
8630 @file{final.c}). These are useful when a single @file{md} file must
8631 support multiple assembler formats. In that case, the various @file{tm.h}
8632 files can define these macros differently.
8635 @defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format})
8636 If defined this macro should expand to a series of @code{case}
8637 statements which will be parsed inside the @code{switch} statement of
8638 the @code{asm_fprintf} function. This allows targets to define extra
8639 printf formats which may useful when generating their assembler
8640 statements. Note that uppercase letters are reserved for future
8641 generic extensions to asm_fprintf, and so are not available to target
8642 specific code. The output file is given by the parameter @var{file}.
8643 The varargs input pointer is @var{argptr} and the rest of the format
8644 string, starting the character after the one that is being switched
8645 upon, is pointed to by @var{format}.
8648 @defmac ASSEMBLER_DIALECT
8649 If your target supports multiple dialects of assembler language (such as
8650 different opcodes), define this macro as a C expression that gives the
8651 numeric index of the assembler language dialect to use, with zero as the
8654 If this macro is defined, you may use constructs of the form
8656 @samp{@{option0|option1|option2@dots{}@}}
8659 in the output templates of patterns (@pxref{Output Template}) or in the
8660 first argument of @code{asm_fprintf}. This construct outputs
8661 @samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of
8662 @code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters
8663 within these strings retain their usual meaning. If there are fewer
8664 alternatives within the braces than the value of
8665 @code{ASSEMBLER_DIALECT}, the construct outputs nothing.
8667 If you do not define this macro, the characters @samp{@{}, @samp{|} and
8668 @samp{@}} do not have any special meaning when used in templates or
8669 operands to @code{asm_fprintf}.
8671 Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX},
8672 @code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express
8673 the variations in assembler language syntax with that mechanism. Define
8674 @code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax
8675 if the syntax variant are larger and involve such things as different
8676 opcodes or operand order.
8679 @defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
8680 A C expression to output to @var{stream} some assembler code
8681 which will push hard register number @var{regno} onto the stack.
8682 The code need not be optimal, since this macro is used only when
8686 @defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
8687 A C expression to output to @var{stream} some assembler code
8688 which will pop hard register number @var{regno} off of the stack.
8689 The code need not be optimal, since this macro is used only when
8693 @node Dispatch Tables
8694 @subsection Output of Dispatch Tables
8696 @c prevent bad page break with this line
8697 This concerns dispatch tables.
8699 @cindex dispatch table
8700 @defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel})
8701 A C statement to output to the stdio stream @var{stream} an assembler
8702 pseudo-instruction to generate a difference between two labels.
8703 @var{value} and @var{rel} are the numbers of two internal labels. The
8704 definitions of these labels are output using
8705 @code{(*targetm.asm_out.internal_label)}, and they must be printed in the same
8706 way here. For example,
8709 fprintf (@var{stream}, "\t.word L%d-L%d\n",
8710 @var{value}, @var{rel})
8713 You must provide this macro on machines where the addresses in a
8714 dispatch table are relative to the table's own address. If defined, GCC
8715 will also use this macro on all machines when producing PIC@.
8716 @var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the
8717 mode and flags can be read.
8720 @defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
8721 This macro should be provided on machines where the addresses
8722 in a dispatch table are absolute.
8724 The definition should be a C statement to output to the stdio stream
8725 @var{stream} an assembler pseudo-instruction to generate a reference to
8726 a label. @var{value} is the number of an internal label whose
8727 definition is output using @code{(*targetm.asm_out.internal_label)}.
8731 fprintf (@var{stream}, "\t.word L%d\n", @var{value})
8735 @defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
8736 Define this if the label before a jump-table needs to be output
8737 specially. The first three arguments are the same as for
8738 @code{(*targetm.asm_out.internal_label)}; the fourth argument is the
8739 jump-table which follows (a @code{jump_insn} containing an
8740 @code{addr_vec} or @code{addr_diff_vec}).
8742 This feature is used on system V to output a @code{swbeg} statement
8745 If this macro is not defined, these labels are output with
8746 @code{(*targetm.asm_out.internal_label)}.
8749 @defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
8750 Define this if something special must be output at the end of a
8751 jump-table. The definition should be a C statement to be executed
8752 after the assembler code for the table is written. It should write
8753 the appropriate code to stdio stream @var{stream}. The argument
8754 @var{table} is the jump-table insn, and @var{num} is the label-number
8755 of the preceding label.
8757 If this macro is not defined, nothing special is output at the end of
8761 @hook TARGET_ASM_EMIT_UNWIND_LABEL
8762 This target hook emits a label at the beginning of each FDE@. It
8763 should be defined on targets where FDEs need special labels, and it
8764 should write the appropriate label, for the FDE associated with the
8765 function declaration @var{decl}, to the stdio stream @var{stream}.
8766 The third argument, @var{for_eh}, is a boolean: true if this is for an
8767 exception table. The fourth argument, @var{empty}, is a boolean:
8768 true if this is a placeholder label for an omitted FDE@.
8770 The default is that FDEs are not given nonlocal labels.
8773 @hook TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL
8774 This target hook emits a label at the beginning of the exception table.
8775 It should be defined on targets where it is desirable for the table
8776 to be broken up according to function.
8778 The default is that no label is emitted.
8781 @hook TARGET_ASM_EMIT_EXCEPT_PERSONALITY
8783 @hook TARGET_ASM_UNWIND_EMIT
8784 This target hook emits assembly directives required to unwind the
8785 given instruction. This is only used when @code{TARGET_EXCEPT_UNWIND_INFO}
8786 returns @code{UI_TARGET}.
8789 @hook TARGET_ASM_UNWIND_EMIT_BEFORE_INSN
8791 @node Exception Region Output
8792 @subsection Assembler Commands for Exception Regions
8794 @c prevent bad page break with this line
8796 This describes commands marking the start and the end of an exception
8799 @defmac EH_FRAME_SECTION_NAME
8800 If defined, a C string constant for the name of the section containing
8801 exception handling frame unwind information. If not defined, GCC will
8802 provide a default definition if the target supports named sections.
8803 @file{crtstuff.c} uses this macro to switch to the appropriate section.
8805 You should define this symbol if your target supports DWARF 2 frame
8806 unwind information and the default definition does not work.
8809 @defmac EH_FRAME_IN_DATA_SECTION
8810 If defined, DWARF 2 frame unwind information will be placed in the
8811 data section even though the target supports named sections. This
8812 might be necessary, for instance, if the system linker does garbage
8813 collection and sections cannot be marked as not to be collected.
8815 Do not define this macro unless @code{TARGET_ASM_NAMED_SECTION} is
8819 @defmac EH_TABLES_CAN_BE_READ_ONLY
8820 Define this macro to 1 if your target is such that no frame unwind
8821 information encoding used with non-PIC code will ever require a
8822 runtime relocation, but the linker may not support merging read-only
8823 and read-write sections into a single read-write section.
8826 @defmac MASK_RETURN_ADDR
8827 An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so
8828 that it does not contain any extraneous set bits in it.
8831 @defmac DWARF2_UNWIND_INFO
8832 Define this macro to 0 if your target supports DWARF 2 frame unwind
8833 information, but it does not yet work with exception handling.
8834 Otherwise, if your target supports this information (if it defines
8835 @code{INCOMING_RETURN_ADDR_RTX} and either @code{UNALIGNED_INT_ASM_OP}
8836 or @code{OBJECT_FORMAT_ELF}), GCC will provide a default definition of 1.
8839 @hook TARGET_EXCEPT_UNWIND_INFO
8840 This hook defines the mechanism that will be used for exception handling
8841 by the target. If the target has ABI specified unwind tables, the hook
8842 should return @code{UI_TARGET}. If the target is to use the
8843 @code{setjmp}/@code{longjmp}-based exception handling scheme, the hook
8844 should return @code{UI_SJLJ}. If the target supports DWARF 2 frame unwind
8845 information, the hook should return @code{UI_DWARF2}.
8847 A target may, if exceptions are disabled, choose to return @code{UI_NONE}.
8848 This may end up simplifying other parts of target-specific code. The
8849 default implementation of this hook never returns @code{UI_NONE}.
8851 Note that the value returned by this hook should be constant. It should
8852 not depend on anything except command-line switches. In particular, the
8853 setting @code{UI_SJLJ} must be fixed at compiler start-up as C pre-processor
8854 macros and builtin functions related to exception handling are set up
8855 depending on this setting.
8857 The default implementation of the hook first honors the
8858 @option{--enable-sjlj-exceptions} configure option, then
8859 @code{DWARF2_UNWIND_INFO}, and finally defaults to @code{UI_SJLJ}.
8862 @hook TARGET_UNWIND_TABLES_DEFAULT
8863 This variable should be set to @code{true} if the target ABI requires unwinding
8864 tables even when exceptions are not used.
8867 @defmac MUST_USE_SJLJ_EXCEPTIONS
8868 This macro need only be defined if @code{DWARF2_UNWIND_INFO} is
8869 runtime-variable. In that case, @file{except.h} cannot correctly
8870 determine the corresponding definition of @code{MUST_USE_SJLJ_EXCEPTIONS},
8871 so the target must provide it directly.
8874 @defmac DONT_USE_BUILTIN_SETJMP
8875 Define this macro to 1 if the @code{setjmp}/@code{longjmp}-based scheme
8876 should use the @code{setjmp}/@code{longjmp} functions from the C library
8877 instead of the @code{__builtin_setjmp}/@code{__builtin_longjmp} machinery.
8880 @defmac DWARF_CIE_DATA_ALIGNMENT
8881 This macro need only be defined if the target might save registers in the
8882 function prologue at an offset to the stack pointer that is not aligned to
8883 @code{UNITS_PER_WORD}. The definition should be the negative minimum
8884 alignment if @code{STACK_GROWS_DOWNWARD} is defined, and the positive
8885 minimum alignment otherwise. @xref{SDB and DWARF}. Only applicable if
8886 the target supports DWARF 2 frame unwind information.
8889 @hook TARGET_TERMINATE_DW2_EH_FRAME_INFO
8890 Contains the value true if the target should add a zero word onto the
8891 end of a Dwarf-2 frame info section when used for exception handling.
8892 Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and
8896 @hook TARGET_DWARF_REGISTER_SPAN
8897 Given a register, this hook should return a parallel of registers to
8898 represent where to find the register pieces. Define this hook if the
8899 register and its mode are represented in Dwarf in non-contiguous
8900 locations, or if the register should be represented in more than one
8901 register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}.
8902 If not defined, the default is to return @code{NULL_RTX}.
8905 @hook TARGET_INIT_DWARF_REG_SIZES_EXTRA
8906 If some registers are represented in Dwarf-2 unwind information in
8907 multiple pieces, define this hook to fill in information about the
8908 sizes of those pieces in the table used by the unwinder at runtime.
8909 It will be called by @code{expand_builtin_init_dwarf_reg_sizes} after
8910 filling in a single size corresponding to each hard register;
8911 @var{address} is the address of the table.
8914 @hook TARGET_ASM_TTYPE
8915 This hook is used to output a reference from a frame unwinding table to
8916 the type_info object identified by @var{sym}. It should return @code{true}
8917 if the reference was output. Returning @code{false} will cause the
8918 reference to be output using the normal Dwarf2 routines.
8921 @hook TARGET_ARM_EABI_UNWINDER
8922 This flag should be set to @code{true} on targets that use an ARM EABI
8923 based unwinding library, and @code{false} on other targets. This effects
8924 the format of unwinding tables, and how the unwinder in entered after
8925 running a cleanup. The default is @code{false}.
8928 @node Alignment Output
8929 @subsection Assembler Commands for Alignment
8931 @c prevent bad page break with this line
8932 This describes commands for alignment.
8934 @defmac JUMP_ALIGN (@var{label})
8935 The alignment (log base 2) to put in front of @var{label}, which is
8936 a common destination of jumps and has no fallthru incoming edge.
8938 This macro need not be defined if you don't want any special alignment
8939 to be done at such a time. Most machine descriptions do not currently
8942 Unless it's necessary to inspect the @var{label} parameter, it is better
8943 to set the variable @var{align_jumps} in the target's
8944 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
8945 selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation.
8948 @defmac LABEL_ALIGN_AFTER_BARRIER (@var{label})
8949 The alignment (log base 2) to put in front of @var{label}, which follows
8952 This macro need not be defined if you don't want any special alignment
8953 to be done at such a time. Most machine descriptions do not currently
8957 @defmac LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP
8958 The maximum number of bytes to skip when applying
8959 @code{LABEL_ALIGN_AFTER_BARRIER}. This works only if
8960 @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
8963 @defmac LOOP_ALIGN (@var{label})
8964 The alignment (log base 2) to put in front of @var{label}, which follows
8965 a @code{NOTE_INSN_LOOP_BEG} note.
8967 This macro need not be defined if you don't want any special alignment
8968 to be done at such a time. Most machine descriptions do not currently
8971 Unless it's necessary to inspect the @var{label} parameter, it is better
8972 to set the variable @code{align_loops} in the target's
8973 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
8974 selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation.
8977 @defmac LOOP_ALIGN_MAX_SKIP
8978 The maximum number of bytes to skip when applying @code{LOOP_ALIGN}.
8979 This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
8982 @defmac LABEL_ALIGN (@var{label})
8983 The alignment (log base 2) to put in front of @var{label}.
8984 If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment,
8985 the maximum of the specified values is used.
8987 Unless it's necessary to inspect the @var{label} parameter, it is better
8988 to set the variable @code{align_labels} in the target's
8989 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
8990 selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation.
8993 @defmac LABEL_ALIGN_MAX_SKIP
8994 The maximum number of bytes to skip when applying @code{LABEL_ALIGN}.
8995 This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
8998 @defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
8999 A C statement to output to the stdio stream @var{stream} an assembler
9000 instruction to advance the location counter by @var{nbytes} bytes.
9001 Those bytes should be zero when loaded. @var{nbytes} will be a C
9002 expression of type @code{unsigned HOST_WIDE_INT}.
9005 @defmac ASM_NO_SKIP_IN_TEXT
9006 Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the
9007 text section because it fails to put zeros in the bytes that are skipped.
9008 This is true on many Unix systems, where the pseudo--op to skip bytes
9009 produces no-op instructions rather than zeros when used in the text
9013 @defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
9014 A C statement to output to the stdio stream @var{stream} an assembler
9015 command to advance the location counter to a multiple of 2 to the
9016 @var{power} bytes. @var{power} will be a C expression of type @code{int}.
9019 @defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power})
9020 Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used
9021 for padding, if necessary.
9024 @defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip})
9025 A C statement to output to the stdio stream @var{stream} an assembler
9026 command to advance the location counter to a multiple of 2 to the
9027 @var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to
9028 satisfy the alignment request. @var{power} and @var{max_skip} will be
9029 a C expression of type @code{int}.
9033 @node Debugging Info
9034 @section Controlling Debugging Information Format
9036 @c prevent bad page break with this line
9037 This describes how to specify debugging information.
9040 * All Debuggers:: Macros that affect all debugging formats uniformly.
9041 * DBX Options:: Macros enabling specific options in DBX format.
9042 * DBX Hooks:: Hook macros for varying DBX format.
9043 * File Names and DBX:: Macros controlling output of file names in DBX format.
9044 * SDB and DWARF:: Macros for SDB (COFF) and DWARF formats.
9045 * VMS Debug:: Macros for VMS debug format.
9049 @subsection Macros Affecting All Debugging Formats
9051 @c prevent bad page break with this line
9052 These macros affect all debugging formats.
9054 @defmac DBX_REGISTER_NUMBER (@var{regno})
9055 A C expression that returns the DBX register number for the compiler
9056 register number @var{regno}. In the default macro provided, the value
9057 of this expression will be @var{regno} itself. But sometimes there are
9058 some registers that the compiler knows about and DBX does not, or vice
9059 versa. In such cases, some register may need to have one number in the
9060 compiler and another for DBX@.
9062 If two registers have consecutive numbers inside GCC, and they can be
9063 used as a pair to hold a multiword value, then they @emph{must} have
9064 consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}.
9065 Otherwise, debuggers will be unable to access such a pair, because they
9066 expect register pairs to be consecutive in their own numbering scheme.
9068 If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that
9069 does not preserve register pairs, then what you must do instead is
9070 redefine the actual register numbering scheme.
9073 @defmac DEBUGGER_AUTO_OFFSET (@var{x})
9074 A C expression that returns the integer offset value for an automatic
9075 variable having address @var{x} (an RTL expression). The default
9076 computation assumes that @var{x} is based on the frame-pointer and
9077 gives the offset from the frame-pointer. This is required for targets
9078 that produce debugging output for DBX or COFF-style debugging output
9079 for SDB and allow the frame-pointer to be eliminated when the
9080 @option{-g} options is used.
9083 @defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x})
9084 A C expression that returns the integer offset value for an argument
9085 having address @var{x} (an RTL expression). The nominal offset is
9089 @defmac PREFERRED_DEBUGGING_TYPE
9090 A C expression that returns the type of debugging output GCC should
9091 produce when the user specifies just @option{-g}. Define
9092 this if you have arranged for GCC to support more than one format of
9093 debugging output. Currently, the allowable values are @code{DBX_DEBUG},
9094 @code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG},
9095 @code{XCOFF_DEBUG}, @code{VMS_DEBUG}, and @code{VMS_AND_DWARF2_DEBUG}.
9097 When the user specifies @option{-ggdb}, GCC normally also uses the
9098 value of this macro to select the debugging output format, but with two
9099 exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the
9100 value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is
9101 defined, GCC uses @code{DBX_DEBUG}.
9103 The value of this macro only affects the default debugging output; the
9104 user can always get a specific type of output by using @option{-gstabs},
9105 @option{-gcoff}, @option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}.
9109 @subsection Specific Options for DBX Output
9111 @c prevent bad page break with this line
9112 These are specific options for DBX output.
9114 @defmac DBX_DEBUGGING_INFO
9115 Define this macro if GCC should produce debugging output for DBX
9116 in response to the @option{-g} option.
9119 @defmac XCOFF_DEBUGGING_INFO
9120 Define this macro if GCC should produce XCOFF format debugging output
9121 in response to the @option{-g} option. This is a variant of DBX format.
9124 @defmac DEFAULT_GDB_EXTENSIONS
9125 Define this macro to control whether GCC should by default generate
9126 GDB's extended version of DBX debugging information (assuming DBX-format
9127 debugging information is enabled at all). If you don't define the
9128 macro, the default is 1: always generate the extended information
9129 if there is any occasion to.
9132 @defmac DEBUG_SYMS_TEXT
9133 Define this macro if all @code{.stabs} commands should be output while
9134 in the text section.
9137 @defmac ASM_STABS_OP
9138 A C string constant, including spacing, naming the assembler pseudo op to
9139 use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol.
9140 If you don't define this macro, @code{"\t.stabs\t"} is used. This macro
9141 applies only to DBX debugging information format.
9144 @defmac ASM_STABD_OP
9145 A C string constant, including spacing, naming the assembler pseudo op to
9146 use instead of @code{"\t.stabd\t"} to define a debugging symbol whose
9147 value is the current location. If you don't define this macro,
9148 @code{"\t.stabd\t"} is used. This macro applies only to DBX debugging
9152 @defmac ASM_STABN_OP
9153 A C string constant, including spacing, naming the assembler pseudo op to
9154 use instead of @code{"\t.stabn\t"} to define a debugging symbol with no
9155 name. If you don't define this macro, @code{"\t.stabn\t"} is used. This
9156 macro applies only to DBX debugging information format.
9159 @defmac DBX_NO_XREFS
9160 Define this macro if DBX on your system does not support the construct
9161 @samp{xs@var{tagname}}. On some systems, this construct is used to
9162 describe a forward reference to a structure named @var{tagname}.
9163 On other systems, this construct is not supported at all.
9166 @defmac DBX_CONTIN_LENGTH
9167 A symbol name in DBX-format debugging information is normally
9168 continued (split into two separate @code{.stabs} directives) when it
9169 exceeds a certain length (by default, 80 characters). On some
9170 operating systems, DBX requires this splitting; on others, splitting
9171 must not be done. You can inhibit splitting by defining this macro
9172 with the value zero. You can override the default splitting-length by
9173 defining this macro as an expression for the length you desire.
9176 @defmac DBX_CONTIN_CHAR
9177 Normally continuation is indicated by adding a @samp{\} character to
9178 the end of a @code{.stabs} string when a continuation follows. To use
9179 a different character instead, define this macro as a character
9180 constant for the character you want to use. Do not define this macro
9181 if backslash is correct for your system.
9184 @defmac DBX_STATIC_STAB_DATA_SECTION
9185 Define this macro if it is necessary to go to the data section before
9186 outputting the @samp{.stabs} pseudo-op for a non-global static
9190 @defmac DBX_TYPE_DECL_STABS_CODE
9191 The value to use in the ``code'' field of the @code{.stabs} directive
9192 for a typedef. The default is @code{N_LSYM}.
9195 @defmac DBX_STATIC_CONST_VAR_CODE
9196 The value to use in the ``code'' field of the @code{.stabs} directive
9197 for a static variable located in the text section. DBX format does not
9198 provide any ``right'' way to do this. The default is @code{N_FUN}.
9201 @defmac DBX_REGPARM_STABS_CODE
9202 The value to use in the ``code'' field of the @code{.stabs} directive
9203 for a parameter passed in registers. DBX format does not provide any
9204 ``right'' way to do this. The default is @code{N_RSYM}.
9207 @defmac DBX_REGPARM_STABS_LETTER
9208 The letter to use in DBX symbol data to identify a symbol as a parameter
9209 passed in registers. DBX format does not customarily provide any way to
9210 do this. The default is @code{'P'}.
9213 @defmac DBX_FUNCTION_FIRST
9214 Define this macro if the DBX information for a function and its
9215 arguments should precede the assembler code for the function. Normally,
9216 in DBX format, the debugging information entirely follows the assembler
9220 @defmac DBX_BLOCKS_FUNCTION_RELATIVE
9221 Define this macro, with value 1, if the value of a symbol describing
9222 the scope of a block (@code{N_LBRAC} or @code{N_RBRAC}) should be
9223 relative to the start of the enclosing function. Normally, GCC uses
9224 an absolute address.
9227 @defmac DBX_LINES_FUNCTION_RELATIVE
9228 Define this macro, with value 1, if the value of a symbol indicating
9229 the current line number (@code{N_SLINE}) should be relative to the
9230 start of the enclosing function. Normally, GCC uses an absolute address.
9233 @defmac DBX_USE_BINCL
9234 Define this macro if GCC should generate @code{N_BINCL} and
9235 @code{N_EINCL} stabs for included header files, as on Sun systems. This
9236 macro also directs GCC to output a type number as a pair of a file
9237 number and a type number within the file. Normally, GCC does not
9238 generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single
9239 number for a type number.
9243 @subsection Open-Ended Hooks for DBX Format
9245 @c prevent bad page break with this line
9246 These are hooks for DBX format.
9248 @defmac DBX_OUTPUT_LBRAC (@var{stream}, @var{name})
9249 Define this macro to say how to output to @var{stream} the debugging
9250 information for the start of a scope level for variable names. The
9251 argument @var{name} is the name of an assembler symbol (for use with
9252 @code{assemble_name}) whose value is the address where the scope begins.
9255 @defmac DBX_OUTPUT_RBRAC (@var{stream}, @var{name})
9256 Like @code{DBX_OUTPUT_LBRAC}, but for the end of a scope level.
9259 @defmac DBX_OUTPUT_NFUN (@var{stream}, @var{lscope_label}, @var{decl})
9260 Define this macro if the target machine requires special handling to
9261 output an @code{N_FUN} entry for the function @var{decl}.
9264 @defmac DBX_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter})
9265 A C statement to output DBX debugging information before code for line
9266 number @var{line} of the current source file to the stdio stream
9267 @var{stream}. @var{counter} is the number of time the macro was
9268 invoked, including the current invocation; it is intended to generate
9269 unique labels in the assembly output.
9271 This macro should not be defined if the default output is correct, or
9272 if it can be made correct by defining @code{DBX_LINES_FUNCTION_RELATIVE}.
9275 @defmac NO_DBX_FUNCTION_END
9276 Some stabs encapsulation formats (in particular ECOFF), cannot handle the
9277 @code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct.
9278 On those machines, define this macro to turn this feature off without
9279 disturbing the rest of the gdb extensions.
9282 @defmac NO_DBX_BNSYM_ENSYM
9283 Some assemblers cannot handle the @code{.stabd BNSYM/ENSYM,0,0} gdb dbx
9284 extension construct. On those machines, define this macro to turn this
9285 feature off without disturbing the rest of the gdb extensions.
9288 @node File Names and DBX
9289 @subsection File Names in DBX Format
9291 @c prevent bad page break with this line
9292 This describes file names in DBX format.
9294 @defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name})
9295 A C statement to output DBX debugging information to the stdio stream
9296 @var{stream}, which indicates that file @var{name} is the main source
9297 file---the file specified as the input file for compilation.
9298 This macro is called only once, at the beginning of compilation.
9300 This macro need not be defined if the standard form of output
9301 for DBX debugging information is appropriate.
9303 It may be necessary to refer to a label equal to the beginning of the
9304 text section. You can use @samp{assemble_name (stream, ltext_label_name)}
9305 to do so. If you do this, you must also set the variable
9306 @var{used_ltext_label_name} to @code{true}.
9309 @defmac NO_DBX_MAIN_SOURCE_DIRECTORY
9310 Define this macro, with value 1, if GCC should not emit an indication
9311 of the current directory for compilation and current source language at
9312 the beginning of the file.
9315 @defmac NO_DBX_GCC_MARKER
9316 Define this macro, with value 1, if GCC should not emit an indication
9317 that this object file was compiled by GCC@. The default is to emit
9318 an @code{N_OPT} stab at the beginning of every source file, with
9319 @samp{gcc2_compiled.} for the string and value 0.
9322 @defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name})
9323 A C statement to output DBX debugging information at the end of
9324 compilation of the main source file @var{name}. Output should be
9325 written to the stdio stream @var{stream}.
9327 If you don't define this macro, nothing special is output at the end
9328 of compilation, which is correct for most machines.
9331 @defmac DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END
9332 Define this macro @emph{instead of} defining
9333 @code{DBX_OUTPUT_MAIN_SOURCE_FILE_END}, if what needs to be output at
9334 the end of compilation is an @code{N_SO} stab with an empty string,
9335 whose value is the highest absolute text address in the file.
9340 @subsection Macros for SDB and DWARF Output
9342 @c prevent bad page break with this line
9343 Here are macros for SDB and DWARF output.
9345 @defmac SDB_DEBUGGING_INFO
9346 Define this macro if GCC should produce COFF-style debugging output
9347 for SDB in response to the @option{-g} option.
9350 @defmac DWARF2_DEBUGGING_INFO
9351 Define this macro if GCC should produce dwarf version 2 format
9352 debugging output in response to the @option{-g} option.
9354 @hook TARGET_DWARF_CALLING_CONVENTION
9355 Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to
9356 be emitted for each function. Instead of an integer return the enum
9357 value for the @code{DW_CC_} tag.
9360 To support optional call frame debugging information, you must also
9361 define @code{INCOMING_RETURN_ADDR_RTX} and either set
9362 @code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the
9363 prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save}
9364 as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't.
9367 @defmac DWARF2_FRAME_INFO
9368 Define this macro to a nonzero value if GCC should always output
9369 Dwarf 2 frame information. If @code{TARGET_EXCEPT_UNWIND_INFO}
9370 (@pxref{Exception Region Output}) returns @code{UI_DWARF2}, and
9371 exceptions are enabled, GCC will output this information not matter
9372 how you define @code{DWARF2_FRAME_INFO}.
9375 @hook TARGET_DEBUG_UNWIND_INFO
9376 This hook defines the mechanism that will be used for describing frame
9377 unwind information to the debugger. Normally the hook will return
9378 @code{UI_DWARF2} if DWARF 2 debug information is enabled, and
9379 return @code{UI_NONE} otherwise.
9381 A target may return @code{UI_DWARF2} even when DWARF 2 debug information
9382 is disabled in order to always output DWARF 2 frame information.
9384 A target may return @code{UI_TARGET} if it has ABI specified unwind tables.
9385 This will suppress generation of the normal debug frame unwind information.
9388 @defmac DWARF2_ASM_LINE_DEBUG_INFO
9389 Define this macro to be a nonzero value if the assembler can generate Dwarf 2
9390 line debug info sections. This will result in much more compact line number
9391 tables, and hence is desirable if it works.
9394 @hook TARGET_WANT_DEBUG_PUB_SECTIONS
9396 @defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
9397 A C statement to issue assembly directives that create a difference
9398 @var{lab1} minus @var{lab2}, using an integer of the given @var{size}.
9401 @defmac ASM_OUTPUT_DWARF_VMS_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
9402 A C statement to issue assembly directives that create a difference
9403 between the two given labels in system defined units, e.g. instruction
9404 slots on IA64 VMS, using an integer of the given size.
9407 @defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label}, @var{section})
9408 A C statement to issue assembly directives that create a
9409 section-relative reference to the given @var{label}, using an integer of the
9410 given @var{size}. The label is known to be defined in the given @var{section}.
9413 @defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label})
9414 A C statement to issue assembly directives that create a self-relative
9415 reference to the given @var{label}, using an integer of the given @var{size}.
9418 @defmac ASM_OUTPUT_DWARF_TABLE_REF (@var{label})
9419 A C statement to issue assembly directives that create a reference to
9420 the DWARF table identifier @var{label} from the current section. This
9421 is used on some systems to avoid garbage collecting a DWARF table which
9422 is referenced by a function.
9425 @hook TARGET_ASM_OUTPUT_DWARF_DTPREL
9426 If defined, this target hook is a function which outputs a DTP-relative
9427 reference to the given TLS symbol of the specified size.
9430 @defmac PUT_SDB_@dots{}
9431 Define these macros to override the assembler syntax for the special
9432 SDB assembler directives. See @file{sdbout.c} for a list of these
9433 macros and their arguments. If the standard syntax is used, you need
9434 not define them yourself.
9438 Some assemblers do not support a semicolon as a delimiter, even between
9439 SDB assembler directives. In that case, define this macro to be the
9440 delimiter to use (usually @samp{\n}). It is not necessary to define
9441 a new set of @code{PUT_SDB_@var{op}} macros if this is the only change
9445 @defmac SDB_ALLOW_UNKNOWN_REFERENCES
9446 Define this macro to allow references to unknown structure,
9447 union, or enumeration tags to be emitted. Standard COFF does not
9448 allow handling of unknown references, MIPS ECOFF has support for
9452 @defmac SDB_ALLOW_FORWARD_REFERENCES
9453 Define this macro to allow references to structure, union, or
9454 enumeration tags that have not yet been seen to be handled. Some
9455 assemblers choke if forward tags are used, while some require it.
9458 @defmac SDB_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
9459 A C statement to output SDB debugging information before code for line
9460 number @var{line} of the current source file to the stdio stream
9461 @var{stream}. The default is to emit an @code{.ln} directive.
9466 @subsection Macros for VMS Debug Format
9468 @c prevent bad page break with this line
9469 Here are macros for VMS debug format.
9471 @defmac VMS_DEBUGGING_INFO
9472 Define this macro if GCC should produce debugging output for VMS
9473 in response to the @option{-g} option. The default behavior for VMS
9474 is to generate minimal debug info for a traceback in the absence of
9475 @option{-g} unless explicitly overridden with @option{-g0}. This
9476 behavior is controlled by @code{TARGET_OPTION_OPTIMIZATION} and
9477 @code{TARGET_OPTION_OVERRIDE}.
9480 @node Floating Point
9481 @section Cross Compilation and Floating Point
9482 @cindex cross compilation and floating point
9483 @cindex floating point and cross compilation
9485 While all modern machines use twos-complement representation for integers,
9486 there are a variety of representations for floating point numbers. This
9487 means that in a cross-compiler the representation of floating point numbers
9488 in the compiled program may be different from that used in the machine
9489 doing the compilation.
9491 Because different representation systems may offer different amounts of
9492 range and precision, all floating point constants must be represented in
9493 the target machine's format. Therefore, the cross compiler cannot
9494 safely use the host machine's floating point arithmetic; it must emulate
9495 the target's arithmetic. To ensure consistency, GCC always uses
9496 emulation to work with floating point values, even when the host and
9497 target floating point formats are identical.
9499 The following macros are provided by @file{real.h} for the compiler to
9500 use. All parts of the compiler which generate or optimize
9501 floating-point calculations must use these macros. They may evaluate
9502 their operands more than once, so operands must not have side effects.
9504 @defmac REAL_VALUE_TYPE
9505 The C data type to be used to hold a floating point value in the target
9506 machine's format. Typically this is a @code{struct} containing an
9507 array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque
9511 @deftypefn Macro int REAL_VALUES_EQUAL (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9512 Compares for equality the two values, @var{x} and @var{y}. If the target
9513 floating point format supports negative zeroes and/or NaNs,
9514 @samp{REAL_VALUES_EQUAL (-0.0, 0.0)} is true, and
9515 @samp{REAL_VALUES_EQUAL (NaN, NaN)} is false.
9518 @deftypefn Macro int REAL_VALUES_LESS (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9519 Tests whether @var{x} is less than @var{y}.
9522 @deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x})
9523 Truncates @var{x} to a signed integer, rounding toward zero.
9526 @deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x})
9527 Truncates @var{x} to an unsigned integer, rounding toward zero. If
9528 @var{x} is negative, returns zero.
9531 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, enum machine_mode @var{mode})
9532 Converts @var{string} into a floating point number in the target machine's
9533 representation for mode @var{mode}. This routine can handle both
9534 decimal and hexadecimal floating point constants, using the syntax
9535 defined by the C language for both.
9538 @deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x})
9539 Returns 1 if @var{x} is negative (including negative zero), 0 otherwise.
9542 @deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x})
9543 Determines whether @var{x} represents infinity (positive or negative).
9546 @deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x})
9547 Determines whether @var{x} represents a ``NaN'' (not-a-number).
9550 @deftypefn Macro void REAL_ARITHMETIC (REAL_VALUE_TYPE @var{output}, enum tree_code @var{code}, REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9551 Calculates an arithmetic operation on the two floating point values
9552 @var{x} and @var{y}, storing the result in @var{output} (which must be a
9555 The operation to be performed is specified by @var{code}. Only the
9556 following codes are supported: @code{PLUS_EXPR}, @code{MINUS_EXPR},
9557 @code{MULT_EXPR}, @code{RDIV_EXPR}, @code{MAX_EXPR}, @code{MIN_EXPR}.
9559 If @code{REAL_ARITHMETIC} is asked to evaluate division by zero and the
9560 target's floating point format cannot represent infinity, it will call
9561 @code{abort}. Callers should check for this situation first, using
9562 @code{MODE_HAS_INFINITIES}. @xref{Storage Layout}.
9565 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x})
9566 Returns the negative of the floating point value @var{x}.
9569 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x})
9570 Returns the absolute value of @var{x}.
9573 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_TRUNCATE (REAL_VALUE_TYPE @var{mode}, enum machine_mode @var{x})
9574 Truncates the floating point value @var{x} to fit in @var{mode}. The
9575 return value is still a full-size @code{REAL_VALUE_TYPE}, but it has an
9576 appropriate bit pattern to be output as a floating constant whose
9577 precision accords with mode @var{mode}.
9580 @deftypefn Macro void REAL_VALUE_TO_INT (HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, REAL_VALUE_TYPE @var{x})
9581 Converts a floating point value @var{x} into a double-precision integer
9582 which is then stored into @var{low} and @var{high}. If the value is not
9583 integral, it is truncated.
9586 @deftypefn Macro void REAL_VALUE_FROM_INT (REAL_VALUE_TYPE @var{x}, HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, enum machine_mode @var{mode})
9587 Converts a double-precision integer found in @var{low} and @var{high},
9588 into a floating point value which is then stored into @var{x}. The
9589 value is truncated to fit in mode @var{mode}.
9592 @node Mode Switching
9593 @section Mode Switching Instructions
9594 @cindex mode switching
9595 The following macros control mode switching optimizations:
9597 @defmac OPTIMIZE_MODE_SWITCHING (@var{entity})
9598 Define this macro if the port needs extra instructions inserted for mode
9599 switching in an optimizing compilation.
9601 For an example, the SH4 can perform both single and double precision
9602 floating point operations, but to perform a single precision operation,
9603 the FPSCR PR bit has to be cleared, while for a double precision
9604 operation, this bit has to be set. Changing the PR bit requires a general
9605 purpose register as a scratch register, hence these FPSCR sets have to
9606 be inserted before reload, i.e.@: you can't put this into instruction emitting
9607 or @code{TARGET_MACHINE_DEPENDENT_REORG}.
9609 You can have multiple entities that are mode-switched, and select at run time
9610 which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should
9611 return nonzero for any @var{entity} that needs mode-switching.
9612 If you define this macro, you also have to define
9613 @code{NUM_MODES_FOR_MODE_SWITCHING}, @code{MODE_NEEDED},
9614 @code{MODE_PRIORITY_TO_MODE} and @code{EMIT_MODE_SET}.
9615 @code{MODE_AFTER}, @code{MODE_ENTRY}, and @code{MODE_EXIT}
9619 @defmac NUM_MODES_FOR_MODE_SWITCHING
9620 If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as
9621 initializer for an array of integers. Each initializer element
9622 N refers to an entity that needs mode switching, and specifies the number
9623 of different modes that might need to be set for this entity.
9624 The position of the initializer in the initializer---starting counting at
9625 zero---determines the integer that is used to refer to the mode-switched
9627 In macros that take mode arguments / yield a mode result, modes are
9628 represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode
9629 switch is needed / supplied.
9632 @defmac MODE_NEEDED (@var{entity}, @var{insn})
9633 @var{entity} is an integer specifying a mode-switched entity. If
9634 @code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to
9635 return an integer value not larger than the corresponding element in
9636 @code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must
9637 be switched into prior to the execution of @var{insn}.
9640 @defmac MODE_AFTER (@var{mode}, @var{insn})
9641 If this macro is defined, it is evaluated for every @var{insn} during
9642 mode switching. It determines the mode that an insn results in (if
9643 different from the incoming mode).
9646 @defmac MODE_ENTRY (@var{entity})
9647 If this macro is defined, it is evaluated for every @var{entity} that needs
9648 mode switching. It should evaluate to an integer, which is a mode that
9649 @var{entity} is assumed to be switched to at function entry. If @code{MODE_ENTRY}
9650 is defined then @code{MODE_EXIT} must be defined.
9653 @defmac MODE_EXIT (@var{entity})
9654 If this macro is defined, it is evaluated for every @var{entity} that needs
9655 mode switching. It should evaluate to an integer, which is a mode that
9656 @var{entity} is assumed to be switched to at function exit. If @code{MODE_EXIT}
9657 is defined then @code{MODE_ENTRY} must be defined.
9660 @defmac MODE_PRIORITY_TO_MODE (@var{entity}, @var{n})
9661 This macro specifies the order in which modes for @var{entity} are processed.
9662 0 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the
9663 lowest. The value of the macro should be an integer designating a mode
9664 for @var{entity}. For any fixed @var{entity}, @code{mode_priority_to_mode}
9665 (@var{entity}, @var{n}) shall be a bijection in 0 @dots{}
9666 @code{num_modes_for_mode_switching[@var{entity}] - 1}.
9669 @defmac EMIT_MODE_SET (@var{entity}, @var{mode}, @var{hard_regs_live})
9670 Generate one or more insns to set @var{entity} to @var{mode}.
9671 @var{hard_reg_live} is the set of hard registers live at the point where
9672 the insn(s) are to be inserted.
9675 @node Target Attributes
9676 @section Defining target-specific uses of @code{__attribute__}
9677 @cindex target attributes
9678 @cindex machine attributes
9679 @cindex attributes, target-specific
9681 Target-specific attributes may be defined for functions, data and types.
9682 These are described using the following target hooks; they also need to
9683 be documented in @file{extend.texi}.
9685 @hook TARGET_ATTRIBUTE_TABLE
9686 If defined, this target hook points to an array of @samp{struct
9687 attribute_spec} (defined in @file{tree.h}) specifying the machine
9688 specific attributes for this target and some of the restrictions on the
9689 entities to which these attributes are applied and the arguments they
9693 @hook TARGET_ATTRIBUTE_TAKES_IDENTIFIER_P
9694 If defined, this target hook is a function which returns true if the
9695 machine-specific attribute named @var{name} expects an identifier
9696 given as its first argument to be passed on as a plain identifier, not
9697 subjected to name lookup. If this is not defined, the default is
9698 false for all machine-specific attributes.
9701 @hook TARGET_COMP_TYPE_ATTRIBUTES
9702 If defined, this target hook is a function which returns zero if the attributes on
9703 @var{type1} and @var{type2} are incompatible, one if they are compatible,
9704 and two if they are nearly compatible (which causes a warning to be
9705 generated). If this is not defined, machine-specific attributes are
9706 supposed always to be compatible.
9709 @hook TARGET_SET_DEFAULT_TYPE_ATTRIBUTES
9710 If defined, this target hook is a function which assigns default attributes to
9711 the newly defined @var{type}.
9714 @hook TARGET_MERGE_TYPE_ATTRIBUTES
9715 Define this target hook if the merging of type attributes needs special
9716 handling. If defined, the result is a list of the combined
9717 @code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed
9718 that @code{comptypes} has already been called and returned 1. This
9719 function may call @code{merge_attributes} to handle machine-independent
9723 @hook TARGET_MERGE_DECL_ATTRIBUTES
9724 Define this target hook if the merging of decl attributes needs special
9725 handling. If defined, the result is a list of the combined
9726 @code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}.
9727 @var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of
9728 when this is needed are when one attribute overrides another, or when an
9729 attribute is nullified by a subsequent definition. This function may
9730 call @code{merge_attributes} to handle machine-independent merging.
9732 @findex TARGET_DLLIMPORT_DECL_ATTRIBUTES
9733 If the only target-specific handling you require is @samp{dllimport}
9734 for Microsoft Windows targets, you should define the macro
9735 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler
9736 will then define a function called
9737 @code{merge_dllimport_decl_attributes} which can then be defined as
9738 the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also
9739 add @code{handle_dll_attribute} in the attribute table for your port
9740 to perform initial processing of the @samp{dllimport} and
9741 @samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and
9742 @file{i386/i386.c}, for example.
9745 @hook TARGET_VALID_DLLIMPORT_ATTRIBUTE_P
9747 @defmac TARGET_DECLSPEC
9748 Define this macro to a nonzero value if you want to treat
9749 @code{__declspec(X)} as equivalent to @code{__attribute((X))}. By
9750 default, this behavior is enabled only for targets that define
9751 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation
9752 of @code{__declspec} is via a built-in macro, but you should not rely
9753 on this implementation detail.
9756 @hook TARGET_INSERT_ATTRIBUTES
9757 Define this target hook if you want to be able to add attributes to a decl
9758 when it is being created. This is normally useful for back ends which
9759 wish to implement a pragma by using the attributes which correspond to
9760 the pragma's effect. The @var{node} argument is the decl which is being
9761 created. The @var{attr_ptr} argument is a pointer to the attribute list
9762 for this decl. The list itself should not be modified, since it may be
9763 shared with other decls, but attributes may be chained on the head of
9764 the list and @code{*@var{attr_ptr}} modified to point to the new
9765 attributes, or a copy of the list may be made if further changes are
9769 @hook TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P
9771 This target hook returns @code{true} if it is ok to inline @var{fndecl}
9772 into the current function, despite its having target-specific
9773 attributes, @code{false} otherwise. By default, if a function has a
9774 target specific attribute attached to it, it will not be inlined.
9777 @hook TARGET_OPTION_VALID_ATTRIBUTE_P
9778 This hook is called to parse the @code{attribute(option("..."))}, and
9779 it allows the function to set different target machine compile time
9780 options for the current function that might be different than the
9781 options specified on the command line. The hook should return
9782 @code{true} if the options are valid.
9784 The hook should set the @var{DECL_FUNCTION_SPECIFIC_TARGET} field in
9785 the function declaration to hold a pointer to a target specific
9786 @var{struct cl_target_option} structure.
9789 @hook TARGET_OPTION_SAVE
9790 This hook is called to save any additional target specific information
9791 in the @var{struct cl_target_option} structure for function specific
9793 @xref{Option file format}.
9796 @hook TARGET_OPTION_RESTORE
9797 This hook is called to restore any additional target specific
9798 information in the @var{struct cl_target_option} structure for
9799 function specific options.
9802 @hook TARGET_OPTION_PRINT
9803 This hook is called to print any additional target specific
9804 information in the @var{struct cl_target_option} structure for
9805 function specific options.
9808 @hook TARGET_OPTION_PRAGMA_PARSE
9809 This target hook parses the options for @code{#pragma GCC option} to
9810 set the machine specific options for functions that occur later in the
9811 input stream. The options should be the same as handled by the
9812 @code{TARGET_OPTION_VALID_ATTRIBUTE_P} hook.
9815 @hook TARGET_OPTION_OVERRIDE
9816 Sometimes certain combinations of command options do not make sense on
9817 a particular target machine. You can override the hook
9818 @code{TARGET_OPTION_OVERRIDE} to take account of this. This hooks is called
9819 once just after all the command options have been parsed.
9821 Don't use this hook to turn on various extra optimizations for
9822 @option{-O}. That is what @code{TARGET_OPTION_OPTIMIZATION} is for.
9824 If you need to do something whenever the optimization level is
9825 changed via the optimize attribute or pragma, see
9826 @code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}
9829 @hook TARGET_CAN_INLINE_P
9830 This target hook returns @code{false} if the @var{caller} function
9831 cannot inline @var{callee}, based on target specific information. By
9832 default, inlining is not allowed if the callee function has function
9833 specific target options and the caller does not use the same options.
9837 @section Emulating TLS
9838 @cindex Emulated TLS
9840 For targets whose psABI does not provide Thread Local Storage via
9841 specific relocations and instruction sequences, an emulation layer is
9842 used. A set of target hooks allows this emulation layer to be
9843 configured for the requirements of a particular target. For instance
9844 the psABI may in fact specify TLS support in terms of an emulation
9847 The emulation layer works by creating a control object for every TLS
9848 object. To access the TLS object, a lookup function is provided
9849 which, when given the address of the control object, will return the
9850 address of the current thread's instance of the TLS object.
9852 @hook TARGET_EMUTLS_GET_ADDRESS
9853 Contains the name of the helper function that uses a TLS control
9854 object to locate a TLS instance. The default causes libgcc's
9855 emulated TLS helper function to be used.
9858 @hook TARGET_EMUTLS_REGISTER_COMMON
9859 Contains the name of the helper function that should be used at
9860 program startup to register TLS objects that are implicitly
9861 initialized to zero. If this is @code{NULL}, all TLS objects will
9862 have explicit initializers. The default causes libgcc's emulated TLS
9863 registration function to be used.
9866 @hook TARGET_EMUTLS_VAR_SECTION
9867 Contains the name of the section in which TLS control variables should
9868 be placed. The default of @code{NULL} allows these to be placed in
9872 @hook TARGET_EMUTLS_TMPL_SECTION
9873 Contains the name of the section in which TLS initializers should be
9874 placed. The default of @code{NULL} allows these to be placed in any
9878 @hook TARGET_EMUTLS_VAR_PREFIX
9879 Contains the prefix to be prepended to TLS control variable names.
9880 The default of @code{NULL} uses a target-specific prefix.
9883 @hook TARGET_EMUTLS_TMPL_PREFIX
9884 Contains the prefix to be prepended to TLS initializer objects. The
9885 default of @code{NULL} uses a target-specific prefix.
9888 @hook TARGET_EMUTLS_VAR_FIELDS
9889 Specifies a function that generates the FIELD_DECLs for a TLS control
9890 object type. @var{type} is the RECORD_TYPE the fields are for and
9891 @var{name} should be filled with the structure tag, if the default of
9892 @code{__emutls_object} is unsuitable. The default creates a type suitable
9893 for libgcc's emulated TLS function.
9896 @hook TARGET_EMUTLS_VAR_INIT
9897 Specifies a function that generates the CONSTRUCTOR to initialize a
9898 TLS control object. @var{var} is the TLS control object, @var{decl}
9899 is the TLS object and @var{tmpl_addr} is the address of the
9900 initializer. The default initializes libgcc's emulated TLS control object.
9903 @hook TARGET_EMUTLS_VAR_ALIGN_FIXED
9904 Specifies whether the alignment of TLS control variable objects is
9905 fixed and should not be increased as some backends may do to optimize
9906 single objects. The default is false.
9909 @hook TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS
9910 Specifies whether a DWARF @code{DW_OP_form_tls_address} location descriptor
9911 may be used to describe emulated TLS control objects.
9914 @node MIPS Coprocessors
9915 @section Defining coprocessor specifics for MIPS targets.
9916 @cindex MIPS coprocessor-definition macros
9918 The MIPS specification allows MIPS implementations to have as many as 4
9919 coprocessors, each with as many as 32 private registers. GCC supports
9920 accessing these registers and transferring values between the registers
9921 and memory using asm-ized variables. For example:
9924 register unsigned int cp0count asm ("c0r1");
9930 (``c0r1'' is the default name of register 1 in coprocessor 0; alternate
9931 names may be added as described below, or the default names may be
9932 overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.)
9934 Coprocessor registers are assumed to be epilogue-used; sets to them will
9935 be preserved even if it does not appear that the register is used again
9936 later in the function.
9938 Another note: according to the MIPS spec, coprocessor 1 (if present) is
9939 the FPU@. One accesses COP1 registers through standard mips
9940 floating-point support; they are not included in this mechanism.
9942 There is one macro used in defining the MIPS coprocessor interface which
9943 you may want to override in subtargets; it is described below.
9945 @defmac ALL_COP_ADDITIONAL_REGISTER_NAMES
9946 A comma-separated list (with leading comma) of pairs describing the
9947 alternate names of coprocessor registers. The format of each entry should be
9949 @{ @var{alternatename}, @var{register_number}@}
9955 @section Parameters for Precompiled Header Validity Checking
9956 @cindex parameters, precompiled headers
9958 @hook TARGET_GET_PCH_VALIDITY
9959 This hook returns a pointer to the data needed by
9960 @code{TARGET_PCH_VALID_P} and sets
9961 @samp{*@var{sz}} to the size of the data in bytes.
9964 @hook TARGET_PCH_VALID_P
9965 This hook checks whether the options used to create a PCH file are
9966 compatible with the current settings. It returns @code{NULL}
9967 if so and a suitable error message if not. Error messages will
9968 be presented to the user and must be localized using @samp{_(@var{msg})}.
9970 @var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY}
9971 when the PCH file was created and @var{sz} is the size of that data in bytes.
9972 It's safe to assume that the data was created by the same version of the
9973 compiler, so no format checking is needed.
9975 The default definition of @code{default_pch_valid_p} should be
9976 suitable for most targets.
9979 @hook TARGET_CHECK_PCH_TARGET_FLAGS
9980 If this hook is nonnull, the default implementation of
9981 @code{TARGET_PCH_VALID_P} will use it to check for compatible values
9982 of @code{target_flags}. @var{pch_flags} specifies the value that
9983 @code{target_flags} had when the PCH file was created. The return
9984 value is the same as for @code{TARGET_PCH_VALID_P}.
9988 @section C++ ABI parameters
9989 @cindex parameters, c++ abi
9991 @hook TARGET_CXX_GUARD_TYPE
9992 Define this hook to override the integer type used for guard variables.
9993 These are used to implement one-time construction of static objects. The
9994 default is long_long_integer_type_node.
9997 @hook TARGET_CXX_GUARD_MASK_BIT
9998 This hook determines how guard variables are used. It should return
9999 @code{false} (the default) if the first byte should be used. A return value of
10000 @code{true} indicates that only the least significant bit should be used.
10003 @hook TARGET_CXX_GET_COOKIE_SIZE
10004 This hook returns the size of the cookie to use when allocating an array
10005 whose elements have the indicated @var{type}. Assumes that it is already
10006 known that a cookie is needed. The default is
10007 @code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the
10008 IA64/Generic C++ ABI@.
10011 @hook TARGET_CXX_COOKIE_HAS_SIZE
10012 This hook should return @code{true} if the element size should be stored in
10013 array cookies. The default is to return @code{false}.
10016 @hook TARGET_CXX_IMPORT_EXPORT_CLASS
10017 If defined by a backend this hook allows the decision made to export
10018 class @var{type} to be overruled. Upon entry @var{import_export}
10019 will contain 1 if the class is going to be exported, @minus{}1 if it is going
10020 to be imported and 0 otherwise. This function should return the
10021 modified value and perform any other actions necessary to support the
10022 backend's targeted operating system.
10025 @hook TARGET_CXX_CDTOR_RETURNS_THIS
10026 This hook should return @code{true} if constructors and destructors return
10027 the address of the object created/destroyed. The default is to return
10031 @hook TARGET_CXX_KEY_METHOD_MAY_BE_INLINE
10032 This hook returns true if the key method for a class (i.e., the method
10033 which, if defined in the current translation unit, causes the virtual
10034 table to be emitted) may be an inline function. Under the standard
10035 Itanium C++ ABI the key method may be an inline function so long as
10036 the function is not declared inline in the class definition. Under
10037 some variants of the ABI, an inline function can never be the key
10038 method. The default is to return @code{true}.
10041 @hook TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY
10043 @hook TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT
10044 This hook returns true (the default) if virtual tables and other
10045 similar implicit class data objects are always COMDAT if they have
10046 external linkage. If this hook returns false, then class data for
10047 classes whose virtual table will be emitted in only one translation
10048 unit will not be COMDAT.
10051 @hook TARGET_CXX_LIBRARY_RTTI_COMDAT
10052 This hook returns true (the default) if the RTTI information for
10053 the basic types which is defined in the C++ runtime should always
10054 be COMDAT, false if it should not be COMDAT.
10057 @hook TARGET_CXX_USE_AEABI_ATEXIT
10058 This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI)
10059 should be used to register static destructors when @option{-fuse-cxa-atexit}
10060 is in effect. The default is to return false to use @code{__cxa_atexit}.
10063 @hook TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT
10064 This hook returns true if the target @code{atexit} function can be used
10065 in the same manner as @code{__cxa_atexit} to register C++ static
10066 destructors. This requires that @code{atexit}-registered functions in
10067 shared libraries are run in the correct order when the libraries are
10068 unloaded. The default is to return false.
10071 @hook TARGET_CXX_ADJUST_CLASS_AT_DEFINITION
10073 @node Named Address Spaces
10074 @section Adding support for named address spaces
10075 @cindex named address spaces
10077 The draft technical report of the ISO/IEC JTC1 S22 WG14 N1275
10078 standards committee, @cite{Programming Languages - C - Extensions to
10079 support embedded processors}, specifies a syntax for embedded
10080 processors to specify alternate address spaces. You can configure a
10081 GCC port to support section 5.1 of the draft report to add support for
10082 address spaces other than the default address space. These address
10083 spaces are new keywords that are similar to the @code{volatile} and
10084 @code{const} type attributes.
10086 Pointers to named address spaces can have a different size than
10087 pointers to the generic address space.
10089 For example, the SPU port uses the @code{__ea} address space to refer
10090 to memory in the host processor, rather than memory local to the SPU
10091 processor. Access to memory in the @code{__ea} address space involves
10092 issuing DMA operations to move data between the host processor and the
10093 local processor memory address space. Pointers in the @code{__ea}
10094 address space are either 32 bits or 64 bits based on the
10095 @option{-mea32} or @option{-mea64} switches (native SPU pointers are
10098 Internally, address spaces are represented as a small integer in the
10099 range 0 to 15 with address space 0 being reserved for the generic
10102 To register a named address space qualifier keyword with the C front end,
10103 the target may call the @code{c_register_addr_space} routine. For example,
10104 the SPU port uses the following to declare @code{__ea} as the keyword for
10105 named address space #1:
10107 #define ADDR_SPACE_EA 1
10108 c_register_addr_space ("__ea", ADDR_SPACE_EA);
10111 @hook TARGET_ADDR_SPACE_POINTER_MODE
10112 Define this to return the machine mode to use for pointers to
10113 @var{address_space} if the target supports named address spaces.
10114 The default version of this hook returns @code{ptr_mode} for the
10115 generic address space only.
10118 @hook TARGET_ADDR_SPACE_ADDRESS_MODE
10119 Define this to return the machine mode to use for addresses in
10120 @var{address_space} if the target supports named address spaces.
10121 The default version of this hook returns @code{Pmode} for the
10122 generic address space only.
10125 @hook TARGET_ADDR_SPACE_VALID_POINTER_MODE
10126 Define this to return nonzero if the port can handle pointers
10127 with machine mode @var{mode} to address space @var{as}. This target
10128 hook is the same as the @code{TARGET_VALID_POINTER_MODE} target hook,
10129 except that it includes explicit named address space support. The default
10130 version of this hook returns true for the modes returned by either the
10131 @code{TARGET_ADDR_SPACE_POINTER_MODE} or @code{TARGET_ADDR_SPACE_ADDRESS_MODE}
10132 target hooks for the given address space.
10135 @hook TARGET_ADDR_SPACE_LEGITIMATE_ADDRESS_P
10136 Define this to return true if @var{exp} is a valid address for mode
10137 @var{mode} in the named address space @var{as}. The @var{strict}
10138 parameter says whether strict addressing is in effect after reload has
10139 finished. This target hook is the same as the
10140 @code{TARGET_LEGITIMATE_ADDRESS_P} target hook, except that it includes
10141 explicit named address space support.
10144 @hook TARGET_ADDR_SPACE_LEGITIMIZE_ADDRESS
10145 Define this to modify an invalid address @var{x} to be a valid address
10146 with mode @var{mode} in the named address space @var{as}. This target
10147 hook is the same as the @code{TARGET_LEGITIMIZE_ADDRESS} target hook,
10148 except that it includes explicit named address space support.
10151 @hook TARGET_ADDR_SPACE_SUBSET_P
10152 Define this to return whether the @var{subset} named address space is
10153 contained within the @var{superset} named address space. Pointers to
10154 a named address space that is a subset of another named address space
10155 will be converted automatically without a cast if used together in
10156 arithmetic operations. Pointers to a superset address space can be
10157 converted to pointers to a subset address space via explicit casts.
10160 @hook TARGET_ADDR_SPACE_CONVERT
10161 Define this to convert the pointer expression represented by the RTL
10162 @var{op} with type @var{from_type} that points to a named address
10163 space to a new pointer expression with type @var{to_type} that points
10164 to a different named address space. When this hook it called, it is
10165 guaranteed that one of the two address spaces is a subset of the other,
10166 as determined by the @code{TARGET_ADDR_SPACE_SUBSET_P} target hook.
10170 @section Miscellaneous Parameters
10171 @cindex parameters, miscellaneous
10173 @c prevent bad page break with this line
10174 Here are several miscellaneous parameters.
10176 @defmac HAS_LONG_COND_BRANCH
10177 Define this boolean macro to indicate whether or not your architecture
10178 has conditional branches that can span all of memory. It is used in
10179 conjunction with an optimization that partitions hot and cold basic
10180 blocks into separate sections of the executable. If this macro is
10181 set to false, gcc will convert any conditional branches that attempt
10182 to cross between sections into unconditional branches or indirect jumps.
10185 @defmac HAS_LONG_UNCOND_BRANCH
10186 Define this boolean macro to indicate whether or not your architecture
10187 has unconditional branches that can span all of memory. It is used in
10188 conjunction with an optimization that partitions hot and cold basic
10189 blocks into separate sections of the executable. If this macro is
10190 set to false, gcc will convert any unconditional branches that attempt
10191 to cross between sections into indirect jumps.
10194 @defmac CASE_VECTOR_MODE
10195 An alias for a machine mode name. This is the machine mode that
10196 elements of a jump-table should have.
10199 @defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body})
10200 Optional: return the preferred mode for an @code{addr_diff_vec}
10201 when the minimum and maximum offset are known. If you define this,
10202 it enables extra code in branch shortening to deal with @code{addr_diff_vec}.
10203 To make this work, you also have to define @code{INSN_ALIGN} and
10204 make the alignment for @code{addr_diff_vec} explicit.
10205 The @var{body} argument is provided so that the offset_unsigned and scale
10206 flags can be updated.
10209 @defmac CASE_VECTOR_PC_RELATIVE
10210 Define this macro to be a C expression to indicate when jump-tables
10211 should contain relative addresses. You need not define this macro if
10212 jump-tables never contain relative addresses, or jump-tables should
10213 contain relative addresses only when @option{-fPIC} or @option{-fPIC}
10217 @hook TARGET_CASE_VALUES_THRESHOLD
10218 This function return the smallest number of different values for which it
10219 is best to use a jump-table instead of a tree of conditional branches.
10220 The default is four for machines with a @code{casesi} instruction and
10221 five otherwise. This is best for most machines.
10224 @defmac CASE_USE_BIT_TESTS
10225 Define this macro to be a C expression to indicate whether C switch
10226 statements may be implemented by a sequence of bit tests. This is
10227 advantageous on processors that can efficiently implement left shift
10228 of 1 by the number of bits held in a register, but inappropriate on
10229 targets that would require a loop. By default, this macro returns
10230 @code{true} if the target defines an @code{ashlsi3} pattern, and
10231 @code{false} otherwise.
10234 @defmac WORD_REGISTER_OPERATIONS
10235 Define this macro if operations between registers with integral mode
10236 smaller than a word are always performed on the entire register.
10237 Most RISC machines have this property and most CISC machines do not.
10240 @defmac LOAD_EXTEND_OP (@var{mem_mode})
10241 Define this macro to be a C expression indicating when insns that read
10242 memory in @var{mem_mode}, an integral mode narrower than a word, set the
10243 bits outside of @var{mem_mode} to be either the sign-extension or the
10244 zero-extension of the data read. Return @code{SIGN_EXTEND} for values
10245 of @var{mem_mode} for which the
10246 insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and
10247 @code{UNKNOWN} for other modes.
10249 This macro is not called with @var{mem_mode} non-integral or with a width
10250 greater than or equal to @code{BITS_PER_WORD}, so you may return any
10251 value in this case. Do not define this macro if it would always return
10252 @code{UNKNOWN}. On machines where this macro is defined, you will normally
10253 define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}.
10255 You may return a non-@code{UNKNOWN} value even if for some hard registers
10256 the sign extension is not performed, if for the @code{REGNO_REG_CLASS}
10257 of these hard registers @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero
10258 when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any
10259 integral mode larger than this but not larger than @code{word_mode}.
10261 You must return @code{UNKNOWN} if for some hard registers that allow this
10262 mode, @code{CANNOT_CHANGE_MODE_CLASS} says that they cannot change to
10263 @code{word_mode}, but that they can change to another integral mode that
10264 is larger then @var{mem_mode} but still smaller than @code{word_mode}.
10267 @defmac SHORT_IMMEDIATES_SIGN_EXTEND
10268 Define this macro if loading short immediate values into registers sign
10272 @defmac FIXUNS_TRUNC_LIKE_FIX_TRUNC
10273 Define this macro if the same instructions that convert a floating
10274 point number to a signed fixed point number also convert validly to an
10278 @hook TARGET_MIN_DIVISIONS_FOR_RECIP_MUL
10279 When @option{-ffast-math} is in effect, GCC tries to optimize
10280 divisions by the same divisor, by turning them into multiplications by
10281 the reciprocal. This target hook specifies the minimum number of divisions
10282 that should be there for GCC to perform the optimization for a variable
10283 of mode @var{mode}. The default implementation returns 3 if the machine
10284 has an instruction for the division, and 2 if it does not.
10288 The maximum number of bytes that a single instruction can move quickly
10289 between memory and registers or between two memory locations.
10292 @defmac MAX_MOVE_MAX
10293 The maximum number of bytes that a single instruction can move quickly
10294 between memory and registers or between two memory locations. If this
10295 is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the
10296 constant value that is the largest value that @code{MOVE_MAX} can have
10300 @defmac SHIFT_COUNT_TRUNCATED
10301 A C expression that is nonzero if on this machine the number of bits
10302 actually used for the count of a shift operation is equal to the number
10303 of bits needed to represent the size of the object being shifted. When
10304 this macro is nonzero, the compiler will assume that it is safe to omit
10305 a sign-extend, zero-extend, and certain bitwise `and' instructions that
10306 truncates the count of a shift operation. On machines that have
10307 instructions that act on bit-fields at variable positions, which may
10308 include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED}
10309 also enables deletion of truncations of the values that serve as
10310 arguments to bit-field instructions.
10312 If both types of instructions truncate the count (for shifts) and
10313 position (for bit-field operations), or if no variable-position bit-field
10314 instructions exist, you should define this macro.
10316 However, on some machines, such as the 80386 and the 680x0, truncation
10317 only applies to shift operations and not the (real or pretended)
10318 bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on
10319 such machines. Instead, add patterns to the @file{md} file that include
10320 the implied truncation of the shift instructions.
10322 You need not define this macro if it would always have the value of zero.
10325 @anchor{TARGET_SHIFT_TRUNCATION_MASK}
10326 @hook TARGET_SHIFT_TRUNCATION_MASK
10327 This function describes how the standard shift patterns for @var{mode}
10328 deal with shifts by negative amounts or by more than the width of the mode.
10329 @xref{shift patterns}.
10331 On many machines, the shift patterns will apply a mask @var{m} to the
10332 shift count, meaning that a fixed-width shift of @var{x} by @var{y} is
10333 equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If
10334 this is true for mode @var{mode}, the function should return @var{m},
10335 otherwise it should return 0. A return value of 0 indicates that no
10336 particular behavior is guaranteed.
10338 Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does
10339 @emph{not} apply to general shift rtxes; it applies only to instructions
10340 that are generated by the named shift patterns.
10342 The default implementation of this function returns
10343 @code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED}
10344 and 0 otherwise. This definition is always safe, but if
10345 @code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns
10346 nevertheless truncate the shift count, you may get better code
10350 @defmac TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
10351 A C expression which is nonzero if on this machine it is safe to
10352 ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
10353 bits (where @var{outprec} is smaller than @var{inprec}) by merely
10354 operating on it as if it had only @var{outprec} bits.
10356 On many machines, this expression can be 1.
10358 @c rearranged this, removed the phrase "it is reported that". this was
10359 @c to fix an overfull hbox. --mew 10feb93
10360 When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for
10361 modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result.
10362 If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in
10363 such cases may improve things.
10366 @hook TARGET_MODE_REP_EXTENDED
10367 The representation of an integral mode can be such that the values
10368 are always extended to a wider integral mode. Return
10369 @code{SIGN_EXTEND} if values of @var{mode} are represented in
10370 sign-extended form to @var{rep_mode}. Return @code{UNKNOWN}
10371 otherwise. (Currently, none of the targets use zero-extended
10372 representation this way so unlike @code{LOAD_EXTEND_OP},
10373 @code{TARGET_MODE_REP_EXTENDED} is expected to return either
10374 @code{SIGN_EXTEND} or @code{UNKNOWN}. Also no target extends
10375 @var{mode} to @var{rep_mode} so that @var{rep_mode} is not the next
10376 widest integral mode and currently we take advantage of this fact.)
10378 Similarly to @code{LOAD_EXTEND_OP} you may return a non-@code{UNKNOWN}
10379 value even if the extension is not performed on certain hard registers
10380 as long as for the @code{REGNO_REG_CLASS} of these hard registers
10381 @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero.
10383 Note that @code{TARGET_MODE_REP_EXTENDED} and @code{LOAD_EXTEND_OP}
10384 describe two related properties. If you define
10385 @code{TARGET_MODE_REP_EXTENDED (mode, word_mode)} you probably also want
10386 to define @code{LOAD_EXTEND_OP (mode)} to return the same type of
10389 In order to enforce the representation of @code{mode},
10390 @code{TRULY_NOOP_TRUNCATION} should return false when truncating to
10394 @defmac STORE_FLAG_VALUE
10395 A C expression describing the value returned by a comparison operator
10396 with an integral mode and stored by a store-flag instruction
10397 (@samp{cstore@var{mode}4}) when the condition is true. This description must
10398 apply to @emph{all} the @samp{cstore@var{mode}4} patterns and all the
10399 comparison operators whose results have a @code{MODE_INT} mode.
10401 A value of 1 or @minus{}1 means that the instruction implementing the
10402 comparison operator returns exactly 1 or @minus{}1 when the comparison is true
10403 and 0 when the comparison is false. Otherwise, the value indicates
10404 which bits of the result are guaranteed to be 1 when the comparison is
10405 true. This value is interpreted in the mode of the comparison
10406 operation, which is given by the mode of the first operand in the
10407 @samp{cstore@var{mode}4} pattern. Either the low bit or the sign bit of
10408 @code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by
10411 If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will
10412 generate code that depends only on the specified bits. It can also
10413 replace comparison operators with equivalent operations if they cause
10414 the required bits to be set, even if the remaining bits are undefined.
10415 For example, on a machine whose comparison operators return an
10416 @code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as
10417 @samp{0x80000000}, saying that just the sign bit is relevant, the
10421 (ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0))
10425 can be converted to
10428 (ashift:SI @var{x} (const_int @var{n}))
10432 where @var{n} is the appropriate shift count to move the bit being
10433 tested into the sign bit.
10435 There is no way to describe a machine that always sets the low-order bit
10436 for a true value, but does not guarantee the value of any other bits,
10437 but we do not know of any machine that has such an instruction. If you
10438 are trying to port GCC to such a machine, include an instruction to
10439 perform a logical-and of the result with 1 in the pattern for the
10440 comparison operators and let us know at @email{gcc@@gcc.gnu.org}.
10442 Often, a machine will have multiple instructions that obtain a value
10443 from a comparison (or the condition codes). Here are rules to guide the
10444 choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions
10449 Use the shortest sequence that yields a valid definition for
10450 @code{STORE_FLAG_VALUE}. It is more efficient for the compiler to
10451 ``normalize'' the value (convert it to, e.g., 1 or 0) than for the
10452 comparison operators to do so because there may be opportunities to
10453 combine the normalization with other operations.
10456 For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being
10457 slightly preferred on machines with expensive jumps and 1 preferred on
10461 As a second choice, choose a value of @samp{0x80000001} if instructions
10462 exist that set both the sign and low-order bits but do not define the
10466 Otherwise, use a value of @samp{0x80000000}.
10469 Many machines can produce both the value chosen for
10470 @code{STORE_FLAG_VALUE} and its negation in the same number of
10471 instructions. On those machines, you should also define a pattern for
10472 those cases, e.g., one matching
10475 (set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C})))
10478 Some machines can also perform @code{and} or @code{plus} operations on
10479 condition code values with less instructions than the corresponding
10480 @samp{cstore@var{mode}4} insn followed by @code{and} or @code{plus}. On those
10481 machines, define the appropriate patterns. Use the names @code{incscc}
10482 and @code{decscc}, respectively, for the patterns which perform
10483 @code{plus} or @code{minus} operations on condition code values. See
10484 @file{rs6000.md} for some examples. The GNU Superoptizer can be used to
10485 find such instruction sequences on other machines.
10487 If this macro is not defined, the default value, 1, is used. You need
10488 not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
10489 instructions, or if the value generated by these instructions is 1.
10492 @defmac FLOAT_STORE_FLAG_VALUE (@var{mode})
10493 A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is
10494 returned when comparison operators with floating-point results are true.
10495 Define this macro on machines that have comparison operations that return
10496 floating-point values. If there are no such operations, do not define
10500 @defmac VECTOR_STORE_FLAG_VALUE (@var{mode})
10501 A C expression that gives a rtx representing the nonzero true element
10502 for vector comparisons. The returned rtx should be valid for the inner
10503 mode of @var{mode} which is guaranteed to be a vector mode. Define
10504 this macro on machines that have vector comparison operations that
10505 return a vector result. If there are no such operations, do not define
10506 this macro. Typically, this macro is defined as @code{const1_rtx} or
10507 @code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent
10508 the compiler optimizing such vector comparison operations for the
10512 @defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
10513 @defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
10514 A C expression that indicates whether the architecture defines a value
10515 for @code{clz} or @code{ctz} with a zero operand.
10516 A result of @code{0} indicates the value is undefined.
10517 If the value is defined for only the RTL expression, the macro should
10518 evaluate to @code{1}; if the value applies also to the corresponding optab
10519 entry (which is normally the case if it expands directly into
10520 the corresponding RTL), then the macro should evaluate to @code{2}.
10521 In the cases where the value is defined, @var{value} should be set to
10524 If this macro is not defined, the value of @code{clz} or
10525 @code{ctz} at zero is assumed to be undefined.
10527 This macro must be defined if the target's expansion for @code{ffs}
10528 relies on a particular value to get correct results. Otherwise it
10529 is not necessary, though it may be used to optimize some corner cases, and
10530 to provide a default expansion for the @code{ffs} optab.
10532 Note that regardless of this macro the ``definedness'' of @code{clz}
10533 and @code{ctz} at zero do @emph{not} extend to the builtin functions
10534 visible to the user. Thus one may be free to adjust the value at will
10535 to match the target expansion of these operations without fear of
10540 An alias for the machine mode for pointers. On most machines, define
10541 this to be the integer mode corresponding to the width of a hardware
10542 pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines.
10543 On some machines you must define this to be one of the partial integer
10544 modes, such as @code{PSImode}.
10546 The width of @code{Pmode} must be at least as large as the value of
10547 @code{POINTER_SIZE}. If it is not equal, you must define the macro
10548 @code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended
10552 @defmac FUNCTION_MODE
10553 An alias for the machine mode used for memory references to functions
10554 being called, in @code{call} RTL expressions. On most CISC machines,
10555 where an instruction can begin at any byte address, this should be
10556 @code{QImode}. On most RISC machines, where all instructions have fixed
10557 size and alignment, this should be a mode with the same size and alignment
10558 as the machine instruction words - typically @code{SImode} or @code{HImode}.
10561 @defmac STDC_0_IN_SYSTEM_HEADERS
10562 In normal operation, the preprocessor expands @code{__STDC__} to the
10563 constant 1, to signify that GCC conforms to ISO Standard C@. On some
10564 hosts, like Solaris, the system compiler uses a different convention,
10565 where @code{__STDC__} is normally 0, but is 1 if the user specifies
10566 strict conformance to the C Standard.
10568 Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host
10569 convention when processing system header files, but when processing user
10570 files @code{__STDC__} will always expand to 1.
10573 @defmac NO_IMPLICIT_EXTERN_C
10574 Define this macro if the system header files support C++ as well as C@.
10575 This macro inhibits the usual method of using system header files in
10576 C++, which is to pretend that the file's contents are enclosed in
10577 @samp{extern "C" @{@dots{}@}}.
10582 @defmac REGISTER_TARGET_PRAGMAS ()
10583 Define this macro if you want to implement any target-specific pragmas.
10584 If defined, it is a C expression which makes a series of calls to
10585 @code{c_register_pragma} or @code{c_register_pragma_with_expansion}
10586 for each pragma. The macro may also do any
10587 setup required for the pragmas.
10589 The primary reason to define this macro is to provide compatibility with
10590 other compilers for the same target. In general, we discourage
10591 definition of target-specific pragmas for GCC@.
10593 If the pragma can be implemented by attributes then you should consider
10594 defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well.
10596 Preprocessor macros that appear on pragma lines are not expanded. All
10597 @samp{#pragma} directives that do not match any registered pragma are
10598 silently ignored, unless the user specifies @option{-Wunknown-pragmas}.
10601 @deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
10602 @deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
10604 Each call to @code{c_register_pragma} or
10605 @code{c_register_pragma_with_expansion} establishes one pragma. The
10606 @var{callback} routine will be called when the preprocessor encounters a
10610 #pragma [@var{space}] @var{name} @dots{}
10613 @var{space} is the case-sensitive namespace of the pragma, or
10614 @code{NULL} to put the pragma in the global namespace. The callback
10615 routine receives @var{pfile} as its first argument, which can be passed
10616 on to cpplib's functions if necessary. You can lex tokens after the
10617 @var{name} by calling @code{pragma_lex}. Tokens that are not read by the
10618 callback will be silently ignored. The end of the line is indicated by
10619 a token of type @code{CPP_EOF}. Macro expansion occurs on the
10620 arguments of pragmas registered with
10621 @code{c_register_pragma_with_expansion} but not on the arguments of
10622 pragmas registered with @code{c_register_pragma}.
10624 Note that the use of @code{pragma_lex} is specific to the C and C++
10625 compilers. It will not work in the Java or Fortran compilers, or any
10626 other language compilers for that matter. Thus if @code{pragma_lex} is going
10627 to be called from target-specific code, it must only be done so when
10628 building the C and C++ compilers. This can be done by defining the
10629 variables @code{c_target_objs} and @code{cxx_target_objs} in the
10630 target entry in the @file{config.gcc} file. These variables should name
10631 the target-specific, language-specific object file which contains the
10632 code that uses @code{pragma_lex}. Note it will also be necessary to add a
10633 rule to the makefile fragment pointed to by @code{tmake_file} that shows
10634 how to build this object file.
10639 @defmac HANDLE_SYSV_PRAGMA
10640 Define this macro (to a value of 1) if you want the System V style
10641 pragmas @samp{#pragma pack(<n>)} and @samp{#pragma weak <name>
10642 [=<value>]} to be supported by gcc.
10644 The pack pragma specifies the maximum alignment (in bytes) of fields
10645 within a structure, in much the same way as the @samp{__aligned__} and
10646 @samp{__packed__} @code{__attribute__}s do. A pack value of zero resets
10647 the behavior to the default.
10649 A subtlety for Microsoft Visual C/C++ style bit-field packing
10650 (e.g.@: -mms-bitfields) for targets that support it:
10651 When a bit-field is inserted into a packed record, the whole size
10652 of the underlying type is used by one or more same-size adjacent
10653 bit-fields (that is, if its long:3, 32 bits is used in the record,
10654 and any additional adjacent long bit-fields are packed into the same
10655 chunk of 32 bits. However, if the size changes, a new field of that
10656 size is allocated).
10658 If both MS bit-fields and @samp{__attribute__((packed))} are used,
10659 the latter will take precedence. If @samp{__attribute__((packed))} is
10660 used on a single field when MS bit-fields are in use, it will take
10661 precedence for that field, but the alignment of the rest of the structure
10662 may affect its placement.
10664 The weak pragma only works if @code{SUPPORTS_WEAK} and
10665 @code{ASM_WEAKEN_LABEL} are defined. If enabled it allows the creation
10666 of specifically named weak labels, optionally with a value.
10671 @defmac HANDLE_PRAGMA_PACK_PUSH_POP
10672 Define this macro (to a value of 1) if you want to support the Win32
10673 style pragmas @samp{#pragma pack(push[,@var{n}])} and @samp{#pragma
10674 pack(pop)}. The @samp{pack(push,[@var{n}])} pragma specifies the maximum
10675 alignment (in bytes) of fields within a structure, in much the same way as
10676 the @samp{__aligned__} and @samp{__packed__} @code{__attribute__}s do. A
10677 pack value of zero resets the behavior to the default. Successive
10678 invocations of this pragma cause the previous values to be stacked, so
10679 that invocations of @samp{#pragma pack(pop)} will return to the previous
10683 @defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION
10684 Define this macro, as well as
10685 @code{HANDLE_SYSV_PRAGMA}, if macros should be expanded in the
10686 arguments of @samp{#pragma pack}.
10689 @hook TARGET_HANDLE_PRAGMA_EXTERN_PREFIX
10691 @defmac TARGET_DEFAULT_PACK_STRUCT
10692 If your target requires a structure packing default other than 0 (meaning
10693 the machine default), define this macro to the necessary value (in bytes).
10694 This must be a value that would also be valid to use with
10695 @samp{#pragma pack()} (that is, a small power of two).
10698 @defmac DOLLARS_IN_IDENTIFIERS
10699 Define this macro to control use of the character @samp{$} in
10700 identifier names for the C family of languages. 0 means @samp{$} is
10701 not allowed by default; 1 means it is allowed. 1 is the default;
10702 there is no need to define this macro in that case.
10705 @defmac NO_DOLLAR_IN_LABEL
10706 Define this macro if the assembler does not accept the character
10707 @samp{$} in label names. By default constructors and destructors in
10708 G++ have @samp{$} in the identifiers. If this macro is defined,
10709 @samp{.} is used instead.
10712 @defmac NO_DOT_IN_LABEL
10713 Define this macro if the assembler does not accept the character
10714 @samp{.} in label names. By default constructors and destructors in G++
10715 have names that use @samp{.}. If this macro is defined, these names
10716 are rewritten to avoid @samp{.}.
10719 @defmac INSN_SETS_ARE_DELAYED (@var{insn})
10720 Define this macro as a C expression that is nonzero if it is safe for the
10721 delay slot scheduler to place instructions in the delay slot of @var{insn},
10722 even if they appear to use a resource set or clobbered in @var{insn}.
10723 @var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that
10724 every @code{call_insn} has this behavior. On machines where some @code{insn}
10725 or @code{jump_insn} is really a function call and hence has this behavior,
10726 you should define this macro.
10728 You need not define this macro if it would always return zero.
10731 @defmac INSN_REFERENCES_ARE_DELAYED (@var{insn})
10732 Define this macro as a C expression that is nonzero if it is safe for the
10733 delay slot scheduler to place instructions in the delay slot of @var{insn},
10734 even if they appear to set or clobber a resource referenced in @var{insn}.
10735 @var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where
10736 some @code{insn} or @code{jump_insn} is really a function call and its operands
10737 are registers whose use is actually in the subroutine it calls, you should
10738 define this macro. Doing so allows the delay slot scheduler to move
10739 instructions which copy arguments into the argument registers into the delay
10740 slot of @var{insn}.
10742 You need not define this macro if it would always return zero.
10745 @defmac MULTIPLE_SYMBOL_SPACES
10746 Define this macro as a C expression that is nonzero if, in some cases,
10747 global symbols from one translation unit may not be bound to undefined
10748 symbols in another translation unit without user intervention. For
10749 instance, under Microsoft Windows symbols must be explicitly imported
10750 from shared libraries (DLLs).
10752 You need not define this macro if it would always evaluate to zero.
10755 @hook TARGET_MD_ASM_CLOBBERS
10756 This target hook should add to @var{clobbers} @code{STRING_CST} trees for
10757 any hard regs the port wishes to automatically clobber for an asm.
10758 It should return the result of the last @code{tree_cons} used to add a
10759 clobber. The @var{outputs}, @var{inputs} and @var{clobber} lists are the
10760 corresponding parameters to the asm and may be inspected to avoid
10761 clobbering a register that is an input or output of the asm. You can use
10762 @code{tree_overlaps_hard_reg_set}, declared in @file{tree.h}, to test
10763 for overlap with regards to asm-declared registers.
10766 @defmac MATH_LIBRARY
10767 Define this macro as a C string constant for the linker argument to link
10768 in the system math library, minus the initial @samp{"-l"}, or
10769 @samp{""} if the target does not have a
10770 separate math library.
10772 You need only define this macro if the default of @samp{"m"} is wrong.
10775 @defmac LIBRARY_PATH_ENV
10776 Define this macro as a C string constant for the environment variable that
10777 specifies where the linker should look for libraries.
10779 You need only define this macro if the default of @samp{"LIBRARY_PATH"}
10783 @defmac TARGET_POSIX_IO
10784 Define this macro if the target supports the following POSIX@ file
10785 functions, access, mkdir and file locking with fcntl / F_SETLKW@.
10786 Defining @code{TARGET_POSIX_IO} will enable the test coverage code
10787 to use file locking when exiting a program, which avoids race conditions
10788 if the program has forked. It will also create directories at run-time
10789 for cross-profiling.
10792 @defmac MAX_CONDITIONAL_EXECUTE
10794 A C expression for the maximum number of instructions to execute via
10795 conditional execution instructions instead of a branch. A value of
10796 @code{BRANCH_COST}+1 is the default if the machine does not use cc0, and
10797 1 if it does use cc0.
10800 @defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr})
10801 Used if the target needs to perform machine-dependent modifications on the
10802 conditionals used for turning basic blocks into conditionally executed code.
10803 @var{ce_info} points to a data structure, @code{struct ce_if_block}, which
10804 contains information about the currently processed blocks. @var{true_expr}
10805 and @var{false_expr} are the tests that are used for converting the
10806 then-block and the else-block, respectively. Set either @var{true_expr} or
10807 @var{false_expr} to a null pointer if the tests cannot be converted.
10810 @defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr})
10811 Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated
10812 if-statements into conditions combined by @code{and} and @code{or} operations.
10813 @var{bb} contains the basic block that contains the test that is currently
10814 being processed and about to be turned into a condition.
10817 @defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn})
10818 A C expression to modify the @var{PATTERN} of an @var{INSN} that is to
10819 be converted to conditional execution format. @var{ce_info} points to
10820 a data structure, @code{struct ce_if_block}, which contains information
10821 about the currently processed blocks.
10824 @defmac IFCVT_MODIFY_FINAL (@var{ce_info})
10825 A C expression to perform any final machine dependent modifications in
10826 converting code to conditional execution. The involved basic blocks
10827 can be found in the @code{struct ce_if_block} structure that is pointed
10828 to by @var{ce_info}.
10831 @defmac IFCVT_MODIFY_CANCEL (@var{ce_info})
10832 A C expression to cancel any machine dependent modifications in
10833 converting code to conditional execution. The involved basic blocks
10834 can be found in the @code{struct ce_if_block} structure that is pointed
10835 to by @var{ce_info}.
10838 @defmac IFCVT_INIT_EXTRA_FIELDS (@var{ce_info})
10839 A C expression to initialize any extra fields in a @code{struct ce_if_block}
10840 structure, which are defined by the @code{IFCVT_EXTRA_FIELDS} macro.
10843 @defmac IFCVT_EXTRA_FIELDS
10844 If defined, it should expand to a set of field declarations that will be
10845 added to the @code{struct ce_if_block} structure. These should be initialized
10846 by the @code{IFCVT_INIT_EXTRA_FIELDS} macro.
10849 @hook TARGET_MACHINE_DEPENDENT_REORG
10850 If non-null, this hook performs a target-specific pass over the
10851 instruction stream. The compiler will run it at all optimization levels,
10852 just before the point at which it normally does delayed-branch scheduling.
10854 The exact purpose of the hook varies from target to target. Some use
10855 it to do transformations that are necessary for correctness, such as
10856 laying out in-function constant pools or avoiding hardware hazards.
10857 Others use it as an opportunity to do some machine-dependent optimizations.
10859 You need not implement the hook if it has nothing to do. The default
10860 definition is null.
10863 @hook TARGET_INIT_BUILTINS
10864 Define this hook if you have any machine-specific built-in functions
10865 that need to be defined. It should be a function that performs the
10868 Machine specific built-in functions can be useful to expand special machine
10869 instructions that would otherwise not normally be generated because
10870 they have no equivalent in the source language (for example, SIMD vector
10871 instructions or prefetch instructions).
10873 To create a built-in function, call the function
10874 @code{lang_hooks.builtin_function}
10875 which is defined by the language front end. You can use any type nodes set
10876 up by @code{build_common_tree_nodes} and @code{build_common_tree_nodes_2};
10877 only language front ends that use those two functions will call
10878 @samp{TARGET_INIT_BUILTINS}.
10881 @hook TARGET_BUILTIN_DECL
10882 Define this hook if you have any machine-specific built-in functions
10883 that need to be defined. It should be a function that returns the
10884 builtin function declaration for the builtin function code @var{code}.
10885 If there is no such builtin and it cannot be initialized at this time
10886 if @var{initialize_p} is true the function should return @code{NULL_TREE}.
10887 If @var{code} is out of range the function should return
10888 @code{error_mark_node}.
10891 @hook TARGET_EXPAND_BUILTIN
10893 Expand a call to a machine specific built-in function that was set up by
10894 @samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the
10895 function call; the result should go to @var{target} if that is
10896 convenient, and have mode @var{mode} if that is convenient.
10897 @var{subtarget} may be used as the target for computing one of
10898 @var{exp}'s operands. @var{ignore} is nonzero if the value is to be
10899 ignored. This function should return the result of the call to the
10903 @hook TARGET_RESOLVE_OVERLOADED_BUILTIN
10904 Select a replacement for a machine specific built-in function that
10905 was set up by @samp{TARGET_INIT_BUILTINS}. This is done
10906 @emph{before} regular type checking, and so allows the target to
10907 implement a crude form of function overloading. @var{fndecl} is the
10908 declaration of the built-in function. @var{arglist} is the list of
10909 arguments passed to the built-in function. The result is a
10910 complete expression that implements the operation, usually
10911 another @code{CALL_EXPR}.
10912 @var{arglist} really has type @samp{VEC(tree,gc)*}
10915 @hook TARGET_FOLD_BUILTIN
10916 Fold a call to a machine specific built-in function that was set up by
10917 @samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the
10918 built-in function. @var{n_args} is the number of arguments passed to
10919 the function; the arguments themselves are pointed to by @var{argp}.
10920 The result is another tree containing a simplified expression for the
10921 call's result. If @var{ignore} is true the value will be ignored.
10924 @hook TARGET_INVALID_WITHIN_DOLOOP
10926 Take an instruction in @var{insn} and return NULL if it is valid within a
10927 low-overhead loop, otherwise return a string explaining why doloop
10928 could not be applied.
10930 Many targets use special registers for low-overhead looping. For any
10931 instruction that clobbers these this function should return a string indicating
10932 the reason why the doloop could not be applied.
10933 By default, the RTL loop optimizer does not use a present doloop pattern for
10934 loops containing function calls or branch on table instructions.
10937 @defmac MD_CAN_REDIRECT_BRANCH (@var{branch1}, @var{branch2})
10939 Take a branch insn in @var{branch1} and another in @var{branch2}.
10940 Return true if redirecting @var{branch1} to the destination of
10941 @var{branch2} is possible.
10943 On some targets, branches may have a limited range. Optimizing the
10944 filling of delay slots can result in branches being redirected, and this
10945 may in turn cause a branch offset to overflow.
10948 @hook TARGET_COMMUTATIVE_P
10949 This target hook returns @code{true} if @var{x} is considered to be commutative.
10950 Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider
10951 PLUS to be commutative inside a MEM@. @var{outer_code} is the rtx code
10952 of the enclosing rtl, if known, otherwise it is UNKNOWN.
10955 @hook TARGET_ALLOCATE_INITIAL_VALUE
10957 When the initial value of a hard register has been copied in a pseudo
10958 register, it is often not necessary to actually allocate another register
10959 to this pseudo register, because the original hard register or a stack slot
10960 it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE}
10961 is called at the start of register allocation once for each hard register
10962 that had its initial value copied by using
10963 @code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}.
10964 Possible values are @code{NULL_RTX}, if you don't want
10965 to do any special allocation, a @code{REG} rtx---that would typically be
10966 the hard register itself, if it is known not to be clobbered---or a
10968 If you are returning a @code{MEM}, this is only a hint for the allocator;
10969 it might decide to use another register anyways.
10970 You may use @code{current_function_leaf_function} in the hook, functions
10971 that use @code{REG_N_SETS}, to determine if the hard
10972 register in question will not be clobbered.
10973 The default value of this hook is @code{NULL}, which disables any special
10977 @hook TARGET_UNSPEC_MAY_TRAP_P
10978 This target hook returns nonzero if @var{x}, an @code{unspec} or
10979 @code{unspec_volatile} operation, might cause a trap. Targets can use
10980 this hook to enhance precision of analysis for @code{unspec} and
10981 @code{unspec_volatile} operations. You may call @code{may_trap_p_1}
10982 to analyze inner elements of @var{x} in which case @var{flags} should be
10986 @hook TARGET_SET_CURRENT_FUNCTION
10987 The compiler invokes this hook whenever it changes its current function
10988 context (@code{cfun}). You can define this function if
10989 the back end needs to perform any initialization or reset actions on a
10990 per-function basis. For example, it may be used to implement function
10991 attributes that affect register usage or code generation patterns.
10992 The argument @var{decl} is the declaration for the new function context,
10993 and may be null to indicate that the compiler has left a function context
10994 and is returning to processing at the top level.
10995 The default hook function does nothing.
10997 GCC sets @code{cfun} to a dummy function context during initialization of
10998 some parts of the back end. The hook function is not invoked in this
10999 situation; you need not worry about the hook being invoked recursively,
11000 or when the back end is in a partially-initialized state.
11001 @code{cfun} might be @code{NULL} to indicate processing at top level,
11002 outside of any function scope.
11005 @defmac TARGET_OBJECT_SUFFIX
11006 Define this macro to be a C string representing the suffix for object
11007 files on your target machine. If you do not define this macro, GCC will
11008 use @samp{.o} as the suffix for object files.
11011 @defmac TARGET_EXECUTABLE_SUFFIX
11012 Define this macro to be a C string representing the suffix to be
11013 automatically added to executable files on your target machine. If you
11014 do not define this macro, GCC will use the null string as the suffix for
11018 @defmac COLLECT_EXPORT_LIST
11019 If defined, @code{collect2} will scan the individual object files
11020 specified on its command line and create an export list for the linker.
11021 Define this macro for systems like AIX, where the linker discards
11022 object files that are not referenced from @code{main} and uses export
11026 @defmac MODIFY_JNI_METHOD_CALL (@var{mdecl})
11027 Define this macro to a C expression representing a variant of the
11028 method call @var{mdecl}, if Java Native Interface (JNI) methods
11029 must be invoked differently from other methods on your target.
11030 For example, on 32-bit Microsoft Windows, JNI methods must be invoked using
11031 the @code{stdcall} calling convention and this macro is then
11032 defined as this expression:
11035 build_type_attribute_variant (@var{mdecl},
11037 (get_identifier ("stdcall"),
11042 @hook TARGET_CANNOT_MODIFY_JUMPS_P
11043 This target hook returns @code{true} past the point in which new jump
11044 instructions could be created. On machines that require a register for
11045 every jump such as the SHmedia ISA of SH5, this point would typically be
11046 reload, so this target hook should be defined to a function such as:
11050 cannot_modify_jumps_past_reload_p ()
11052 return (reload_completed || reload_in_progress);
11057 @hook TARGET_BRANCH_TARGET_REGISTER_CLASS
11058 This target hook returns a register class for which branch target register
11059 optimizations should be applied. All registers in this class should be
11060 usable interchangeably. After reload, registers in this class will be
11061 re-allocated and loads will be hoisted out of loops and be subjected
11062 to inter-block scheduling.
11065 @hook TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED
11066 Branch target register optimization will by default exclude callee-saved
11068 that are not already live during the current function; if this target hook
11069 returns true, they will be included. The target code must than make sure
11070 that all target registers in the class returned by
11071 @samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are
11072 saved. @var{after_prologue_epilogue_gen} indicates if prologues and
11073 epilogues have already been generated. Note, even if you only return
11074 true when @var{after_prologue_epilogue_gen} is false, you still are likely
11075 to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET}
11076 to reserve space for caller-saved target registers.
11079 @hook TARGET_HAVE_CONDITIONAL_EXECUTION
11080 This target hook returns true if the target supports conditional execution.
11081 This target hook is required only when the target has several different
11082 modes and they have different conditional execution capability, such as ARM.
11085 @hook TARGET_LOOP_UNROLL_ADJUST
11086 This target hook returns a new value for the number of times @var{loop}
11087 should be unrolled. The parameter @var{nunroll} is the number of times
11088 the loop is to be unrolled. The parameter @var{loop} is a pointer to
11089 the loop, which is going to be checked for unrolling. This target hook
11090 is required only when the target has special constraints like maximum
11091 number of memory accesses.
11094 @defmac POWI_MAX_MULTS
11095 If defined, this macro is interpreted as a signed integer C expression
11096 that specifies the maximum number of floating point multiplications
11097 that should be emitted when expanding exponentiation by an integer
11098 constant inline. When this value is defined, exponentiation requiring
11099 more than this number of multiplications is implemented by calling the
11100 system library's @code{pow}, @code{powf} or @code{powl} routines.
11101 The default value places no upper bound on the multiplication count.
11104 @deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
11105 This target hook should register any extra include files for the
11106 target. The parameter @var{stdinc} indicates if normal include files
11107 are present. The parameter @var{sysroot} is the system root directory.
11108 The parameter @var{iprefix} is the prefix for the gcc directory.
11111 @deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
11112 This target hook should register any extra include files for the
11113 target before any standard headers. The parameter @var{stdinc}
11114 indicates if normal include files are present. The parameter
11115 @var{sysroot} is the system root directory. The parameter
11116 @var{iprefix} is the prefix for the gcc directory.
11119 @deftypefn Macro void TARGET_OPTF (char *@var{path})
11120 This target hook should register special include paths for the target.
11121 The parameter @var{path} is the include to register. On Darwin
11122 systems, this is used for Framework includes, which have semantics
11123 that are different from @option{-I}.
11126 @defmac bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl})
11127 This target macro returns @code{true} if it is safe to use a local alias
11128 for a virtual function @var{fndecl} when constructing thunks,
11129 @code{false} otherwise. By default, the macro returns @code{true} for all
11130 functions, if a target supports aliases (i.e.@: defines
11131 @code{ASM_OUTPUT_DEF}), @code{false} otherwise,
11134 @defmac TARGET_FORMAT_TYPES
11135 If defined, this macro is the name of a global variable containing
11136 target-specific format checking information for the @option{-Wformat}
11137 option. The default is to have no target-specific format checks.
11140 @defmac TARGET_N_FORMAT_TYPES
11141 If defined, this macro is the number of entries in
11142 @code{TARGET_FORMAT_TYPES}.
11145 @defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES
11146 If defined, this macro is the name of a global variable containing
11147 target-specific format overrides for the @option{-Wformat} option. The
11148 default is to have no target-specific format overrides. If defined,
11149 @code{TARGET_FORMAT_TYPES} must be defined, too.
11152 @defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT
11153 If defined, this macro specifies the number of entries in
11154 @code{TARGET_OVERRIDES_FORMAT_ATTRIBUTES}.
11157 @defmac TARGET_OVERRIDES_FORMAT_INIT
11158 If defined, this macro specifies the optional initialization
11159 routine for target specific customizations of the system printf
11160 and scanf formatter settings.
11163 @hook TARGET_RELAXED_ORDERING
11164 If set to @code{true}, means that the target's memory model does not
11165 guarantee that loads which do not depend on one another will access
11166 main memory in the order of the instruction stream; if ordering is
11167 important, an explicit memory barrier must be used. This is true of
11168 many recent processors which implement a policy of ``relaxed,''
11169 ``weak,'' or ``release'' memory consistency, such as Alpha, PowerPC,
11170 and ia64. The default is @code{false}.
11173 @hook TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN
11174 If defined, this macro returns the diagnostic message when it is
11175 illegal to pass argument @var{val} to function @var{funcdecl}
11176 with prototype @var{typelist}.
11179 @hook TARGET_INVALID_CONVERSION
11180 If defined, this macro returns the diagnostic message when it is
11181 invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL}
11182 if validity should be determined by the front end.
11185 @hook TARGET_INVALID_UNARY_OP
11186 If defined, this macro returns the diagnostic message when it is
11187 invalid to apply operation @var{op} (where unary plus is denoted by
11188 @code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL}
11189 if validity should be determined by the front end.
11192 @hook TARGET_INVALID_BINARY_OP
11193 If defined, this macro returns the diagnostic message when it is
11194 invalid to apply operation @var{op} to operands of types @var{type1}
11195 and @var{type2}, or @code{NULL} if validity should be determined by
11199 @hook TARGET_INVALID_PARAMETER_TYPE
11200 If defined, this macro returns the diagnostic message when it is
11201 invalid for functions to include parameters of type @var{type},
11202 or @code{NULL} if validity should be determined by
11203 the front end. This is currently used only by the C and C++ front ends.
11206 @hook TARGET_INVALID_RETURN_TYPE
11207 If defined, this macro returns the diagnostic message when it is
11208 invalid for functions to have return type @var{type},
11209 or @code{NULL} if validity should be determined by
11210 the front end. This is currently used only by the C and C++ front ends.
11213 @hook TARGET_PROMOTED_TYPE
11214 If defined, this target hook returns the type to which values of
11215 @var{type} should be promoted when they appear in expressions,
11216 analogous to the integer promotions, or @code{NULL_TREE} to use the
11217 front end's normal promotion rules. This hook is useful when there are
11218 target-specific types with special promotion rules.
11219 This is currently used only by the C and C++ front ends.
11222 @hook TARGET_CONVERT_TO_TYPE
11223 If defined, this hook returns the result of converting @var{expr} to
11224 @var{type}. It should return the converted expression,
11225 or @code{NULL_TREE} to apply the front end's normal conversion rules.
11226 This hook is useful when there are target-specific types with special
11228 This is currently used only by the C and C++ front ends.
11231 @defmac TARGET_USE_JCR_SECTION
11232 This macro determines whether to use the JCR section to register Java
11233 classes. By default, TARGET_USE_JCR_SECTION is defined to 1 if both
11234 SUPPORTS_WEAK and TARGET_HAVE_NAMED_SECTIONS are true, else 0.
11238 This macro determines the size of the objective C jump buffer for the
11239 NeXT runtime. By default, OBJC_JBLEN is defined to an innocuous value.
11242 @defmac LIBGCC2_UNWIND_ATTRIBUTE
11243 Define this macro if any target-specific attributes need to be attached
11244 to the functions in @file{libgcc} that provide low-level support for
11245 call stack unwinding. It is used in declarations in @file{unwind-generic.h}
11246 and the associated definitions of those functions.
11249 @hook TARGET_UPDATE_STACK_BOUNDARY
11250 Define this macro to update the current function stack boundary if
11254 @hook TARGET_GET_DRAP_RTX
11255 This hook should return an rtx for Dynamic Realign Argument Pointer (DRAP) if a
11256 different argument pointer register is needed to access the function's
11257 argument list due to stack realignment. Return @code{NULL} if no DRAP
11261 @hook TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS
11262 When optimization is disabled, this hook indicates whether or not
11263 arguments should be allocated to stack slots. Normally, GCC allocates
11264 stacks slots for arguments when not optimizing in order to make
11265 debugging easier. However, when a function is declared with
11266 @code{__attribute__((naked))}, there is no stack frame, and the compiler
11267 cannot safely move arguments from the registers in which they are passed
11268 to the stack. Therefore, this hook should return true in general, but
11269 false for naked functions. The default implementation always returns true.
11272 @hook TARGET_CONST_ANCHOR
11273 On some architectures it can take multiple instructions to synthesize
11274 a constant. If there is another constant already in a register that
11275 is close enough in value then it is preferable that the new constant
11276 is computed from this register using immediate addition or
11277 subtraction. We accomplish this through CSE. Besides the value of
11278 the constant we also add a lower and an upper constant anchor to the
11279 available expressions. These are then queried when encountering new
11280 constants. The anchors are computed by rounding the constant up and
11281 down to a multiple of the value of @code{TARGET_CONST_ANCHOR}.
11282 @code{TARGET_CONST_ANCHOR} should be the maximum positive value
11283 accepted by immediate-add plus one. We currently assume that the
11284 value of @code{TARGET_CONST_ANCHOR} is a power of 2. For example, on
11285 MIPS, where add-immediate takes a 16-bit signed value,
11286 @code{TARGET_CONST_ANCHOR} is set to @samp{0x8000}. The default value
11287 is zero, which disables this optimization. @end deftypevr