1 @c Copyright (C) 1988,1989,1992,1993,1994,1995,1996,1997,1998,1999,2000,2001,
2 @c 2002, 2003 Free Software Foundation, Inc.
3 @c This is part of the GCC manual.
4 @c For copying conditions, see the file gcc.texi.
7 @chapter Target Description Macros and Functions
8 @cindex machine description macros
9 @cindex target description macros
10 @cindex macros, target description
11 @cindex @file{tm.h} macros
13 In addition to the file @file{@var{machine}.md}, a machine description
14 includes a C header file conventionally given the name
15 @file{@var{machine}.h} and a C source file named @file{@var{machine}.c}.
16 The header file defines numerous macros that convey the information
17 about the target machine that does not fit into the scheme of the
18 @file{.md} file. The file @file{tm.h} should be a link to
19 @file{@var{machine}.h}. The header file @file{config.h} includes
20 @file{tm.h} and most compiler source files include @file{config.h}. The
21 source file defines a variable @code{targetm}, which is a structure
22 containing pointers to functions and data relating to the target
23 machine. @file{@var{machine}.c} should also contain their definitions,
24 if they are not defined elsewhere in GCC, and other functions called
25 through the macros defined in the @file{.h} file.
28 * Target Structure:: The @code{targetm} variable.
29 * Driver:: Controlling how the driver runs the compilation passes.
30 * Run-time Target:: Defining @samp{-m} options like @option{-m68000} and @option{-m68020}.
31 * Per-Function Data:: Defining data structures for per-function information.
32 * Storage Layout:: Defining sizes and alignments of data.
33 * Type Layout:: Defining sizes and properties of basic user data types.
34 * Escape Sequences:: Defining the value of target character escape sequences
35 * Registers:: Naming and describing the hardware registers.
36 * Register Classes:: Defining the classes of hardware registers.
37 * Stack and Calling:: Defining which way the stack grows and by how much.
38 * Varargs:: Defining the varargs macros.
39 * Trampolines:: Code set up at run time to enter a nested function.
40 * Library Calls:: Controlling how library routines are implicitly called.
41 * Addressing Modes:: Defining addressing modes valid for memory operands.
42 * Condition Code:: Defining how insns update the condition code.
43 * Costs:: Defining relative costs of different operations.
44 * Scheduling:: Adjusting the behavior of the instruction scheduler.
45 * Sections:: Dividing storage into text, data, and other sections.
46 * PIC:: Macros for position independent code.
47 * Assembler Format:: Defining how to write insns and pseudo-ops to output.
48 * Debugging Info:: Defining the format of debugging output.
49 * Floating Point:: Handling floating point for cross-compilers.
50 * Mode Switching:: Insertion of mode-switching instructions.
51 * Target Attributes:: Defining target-specific uses of @code{__attribute__}.
52 * MIPS Coprocessors:: MIPS coprocessor support and how to customize it.
53 * Misc:: Everything else.
56 @node Target Structure
57 @section The Global @code{targetm} Variable
59 @cindex target functions
61 @deftypevar {struct gcc_target} targetm
62 The target @file{.c} file must define the global @code{targetm} variable
63 which contains pointers to functions and data relating to the target
64 machine. The variable is declared in @file{target.h};
65 @file{target-def.h} defines the macro @code{TARGET_INITIALIZER} which is
66 used to initialize the variable, and macros for the default initializers
67 for elements of the structure. The @file{.c} file should override those
68 macros for which the default definition is inappropriate. For example:
71 #include "target-def.h"
73 /* @r{Initialize the GCC target structure.} */
75 #undef TARGET_COMP_TYPE_ATTRIBUTES
76 #define TARGET_COMP_TYPE_ATTRIBUTES @var{machine}_comp_type_attributes
78 struct gcc_target targetm = TARGET_INITIALIZER;
82 Where a macro should be defined in the @file{.c} file in this manner to
83 form part of the @code{targetm} structure, it is documented below as a
84 ``Target Hook'' with a prototype. Many macros will change in future
85 from being defined in the @file{.h} file to being part of the
86 @code{targetm} structure.
89 @section Controlling the Compilation Driver, @file{gcc}
91 @cindex controlling the compilation driver
93 @c prevent bad page break with this line
94 You can control the compilation driver.
96 @defmac SWITCH_TAKES_ARG (@var{char})
97 A C expression which determines whether the option @option{-@var{char}}
98 takes arguments. The value should be the number of arguments that
99 option takes--zero, for many options.
101 By default, this macro is defined as
102 @code{DEFAULT_SWITCH_TAKES_ARG}, which handles the standard options
103 properly. You need not define @code{SWITCH_TAKES_ARG} unless you
104 wish to add additional options which take arguments. Any redefinition
105 should call @code{DEFAULT_SWITCH_TAKES_ARG} and then check for
109 @defmac WORD_SWITCH_TAKES_ARG (@var{name})
110 A C expression which determines whether the option @option{-@var{name}}
111 takes arguments. The value should be the number of arguments that
112 option takes--zero, for many options. This macro rather than
113 @code{SWITCH_TAKES_ARG} is used for multi-character option names.
115 By default, this macro is defined as
116 @code{DEFAULT_WORD_SWITCH_TAKES_ARG}, which handles the standard options
117 properly. You need not define @code{WORD_SWITCH_TAKES_ARG} unless you
118 wish to add additional options which take arguments. Any redefinition
119 should call @code{DEFAULT_WORD_SWITCH_TAKES_ARG} and then check for
123 @defmac SWITCH_CURTAILS_COMPILATION (@var{char})
124 A C expression which determines whether the option @option{-@var{char}}
125 stops compilation before the generation of an executable. The value is
126 boolean, nonzero if the option does stop an executable from being
127 generated, zero otherwise.
129 By default, this macro is defined as
130 @code{DEFAULT_SWITCH_CURTAILS_COMPILATION}, which handles the standard
131 options properly. You need not define
132 @code{SWITCH_CURTAILS_COMPILATION} unless you wish to add additional
133 options which affect the generation of an executable. Any redefinition
134 should call @code{DEFAULT_SWITCH_CURTAILS_COMPILATION} and then check
135 for additional options.
138 @defmac SWITCHES_NEED_SPACES
139 A string-valued C expression which enumerates the options for which
140 the linker needs a space between the option and its argument.
142 If this macro is not defined, the default value is @code{""}.
145 @defmac TARGET_OPTION_TRANSLATE_TABLE
146 If defined, a list of pairs of strings, the first of which is a
147 potential command line target to the @file{gcc} driver program, and the
148 second of which is a space-separated (tabs and other whitespace are not
149 supported) list of options with which to replace the first option. The
150 target defining this list is responsible for assuring that the results
151 are valid. Replacement options may not be the @code{--opt} style, they
152 must be the @code{-opt} style. It is the intention of this macro to
153 provide a mechanism for substitution that affects the multilibs chosen,
154 such as one option that enables many options, some of which select
155 multilibs. Example nonsensical definition, where @code{-malt-abi},
156 @code{-EB}, and @code{-mspoo} cause different multilibs to be chosen:
159 #define TARGET_OPTION_TRANSLATE_TABLE \
160 @{ "-fast", "-march=fast-foo -malt-abi -I/usr/fast-foo" @}, \
161 @{ "-compat", "-EB -malign=4 -mspoo" @}
165 @defmac DRIVER_SELF_SPECS
166 A list of specs for the driver itself. It should be a suitable
167 initializer for an array of strings, with no surrounding braces.
169 The driver applies these specs to its own command line between loading
170 default @file{specs} files (but not command-line specified ones) and
171 choosing the multilib directory or running any subcommands. It
172 applies them in the order given, so each spec can depend on the
173 options added by earlier ones. It is also possible to remove options
174 using @samp{%<@var{option}} in the usual way.
176 This macro can be useful when a port has several interdependent target
177 options. It provides a way of standardizing the command line so
178 that the other specs are easier to write.
180 Do not define this macro if it does not need to do anything.
183 @defmac OPTION_DEFAULT_SPECS
184 A list of specs used to support configure-time default options (i.e.@:
185 @option{--with} options) in the driver. It should be a suitable initializer
186 for an array of structures, each containing two strings, without the
187 outermost pair of surrounding braces.
189 The first item in the pair is the name of the default. This must match
190 the code in @file{config.gcc} for the target. The second item is a spec
191 to apply if a default with this name was specified. The string
192 @samp{%(VALUE)} in the spec will be replaced by the value of the default
193 everywhere it occurs.
195 The driver will apply these specs to its own command line between loading
196 default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using
197 the same mechanism as @code{DRIVER_SELF_SPECS}.
199 Do not define this macro if it does not need to do anything.
203 A C string constant that tells the GCC driver program options to
204 pass to CPP@. It can also specify how to translate options you
205 give to GCC into options for GCC to pass to the CPP@.
207 Do not define this macro if it does not need to do anything.
210 @defmac CPLUSPLUS_CPP_SPEC
211 This macro is just like @code{CPP_SPEC}, but is used for C++, rather
212 than C@. If you do not define this macro, then the value of
213 @code{CPP_SPEC} (if any) will be used instead.
217 A C string constant that tells the GCC driver program options to
218 pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language
220 It can also specify how to translate options you give to GCC into options
221 for GCC to pass to front ends.
223 Do not define this macro if it does not need to do anything.
227 A C string constant that tells the GCC driver program options to
228 pass to @code{cc1plus}. It can also specify how to translate options you
229 give to GCC into options for GCC to pass to the @code{cc1plus}.
231 Do not define this macro if it does not need to do anything.
232 Note that everything defined in CC1_SPEC is already passed to
233 @code{cc1plus} so there is no need to duplicate the contents of
234 CC1_SPEC in CC1PLUS_SPEC@.
238 A C string constant that tells the GCC driver program options to
239 pass to the assembler. It can also specify how to translate options
240 you give to GCC into options for GCC to pass to the assembler.
241 See the file @file{sun3.h} for an example of this.
243 Do not define this macro if it does not need to do anything.
246 @defmac ASM_FINAL_SPEC
247 A C string constant that tells the GCC driver program how to
248 run any programs which cleanup after the normal assembler.
249 Normally, this is not needed. See the file @file{mips.h} for
252 Do not define this macro if it does not need to do anything.
255 @defmac AS_NEEDS_DASH_FOR_PIPED_INPUT
256 Define this macro, with no value, if the driver should give the assembler
257 an argument consisting of a single dash, @option{-}, to instruct it to
258 read from its standard input (which will be a pipe connected to the
259 output of the compiler proper). This argument is given after any
260 @option{-o} option specifying the name of the output file.
262 If you do not define this macro, the assembler is assumed to read its
263 standard input if given no non-option arguments. If your assembler
264 cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct;
265 see @file{mips.h} for instance.
269 A C string constant that tells the GCC driver program options to
270 pass to the linker. It can also specify how to translate options you
271 give to GCC into options for GCC to pass to the linker.
273 Do not define this macro if it does not need to do anything.
277 Another C string constant used much like @code{LINK_SPEC}. The difference
278 between the two is that @code{LIB_SPEC} is used at the end of the
279 command given to the linker.
281 If this macro is not defined, a default is provided that
282 loads the standard C library from the usual place. See @file{gcc.c}.
286 Another C string constant that tells the GCC driver program
287 how and when to place a reference to @file{libgcc.a} into the
288 linker command line. This constant is placed both before and after
289 the value of @code{LIB_SPEC}.
291 If this macro is not defined, the GCC driver provides a default that
292 passes the string @option{-lgcc} to the linker.
295 @defmac STARTFILE_SPEC
296 Another C string constant used much like @code{LINK_SPEC}. The
297 difference between the two is that @code{STARTFILE_SPEC} is used at
298 the very beginning of the command given to the linker.
300 If this macro is not defined, a default is provided that loads the
301 standard C startup file from the usual place. See @file{gcc.c}.
305 Another C string constant used much like @code{LINK_SPEC}. The
306 difference between the two is that @code{ENDFILE_SPEC} is used at
307 the very end of the command given to the linker.
309 Do not define this macro if it does not need to do anything.
312 @defmac THREAD_MODEL_SPEC
313 GCC @code{-v} will print the thread model GCC was configured to use.
314 However, this doesn't work on platforms that are multilibbed on thread
315 models, such as AIX 4.3. On such platforms, define
316 @code{THREAD_MODEL_SPEC} such that it evaluates to a string without
317 blanks that names one of the recognized thread models. @code{%*}, the
318 default value of this macro, will expand to the value of
319 @code{thread_file} set in @file{config.gcc}.
322 @defmac SYSROOT_SUFFIX_SPEC
323 Define this macro to add a suffix to the target sysroot when GCC is
324 configured with a sysroot. This will cause GCC to search for usr/lib,
325 et al, within sysroot+suffix.
328 @defmac SYSROOT_HEADERS_SUFFIX_SPEC
329 Define this macro to add a headers_suffix to the target sysroot when
330 GCC is configured with a sysroot. This will cause GCC to pass the
331 updated sysroot+headers_suffix to CPP@, causing it to search for
332 usr/include, et al, within sysroot+headers_suffix.
336 Define this macro to provide additional specifications to put in the
337 @file{specs} file that can be used in various specifications like
340 The definition should be an initializer for an array of structures,
341 containing a string constant, that defines the specification name, and a
342 string constant that provides the specification.
344 Do not define this macro if it does not need to do anything.
346 @code{EXTRA_SPECS} is useful when an architecture contains several
347 related targets, which have various @code{@dots{}_SPECS} which are similar
348 to each other, and the maintainer would like one central place to keep
351 For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to
352 define either @code{_CALL_SYSV} when the System V calling sequence is
353 used or @code{_CALL_AIX} when the older AIX-based calling sequence is
356 The @file{config/rs6000/rs6000.h} target file defines:
359 #define EXTRA_SPECS \
360 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @},
362 #define CPP_SYS_DEFAULT ""
365 The @file{config/rs6000/sysv.h} target file defines:
369 "%@{posix: -D_POSIX_SOURCE @} \
370 %@{mcall-sysv: -D_CALL_SYSV @} \
371 %@{!mcall-sysv: %(cpp_sysv_default) @} \
372 %@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}"
374 #undef CPP_SYSV_DEFAULT
375 #define CPP_SYSV_DEFAULT "-D_CALL_SYSV"
378 while the @file{config/rs6000/eabiaix.h} target file defines
379 @code{CPP_SYSV_DEFAULT} as:
382 #undef CPP_SYSV_DEFAULT
383 #define CPP_SYSV_DEFAULT "-D_CALL_AIX"
387 @defmac LINK_LIBGCC_SPECIAL
388 Define this macro if the driver program should find the library
389 @file{libgcc.a} itself and should not pass @option{-L} options to the
390 linker. If you do not define this macro, the driver program will pass
391 the argument @option{-lgcc} to tell the linker to do the search and will
392 pass @option{-L} options to it.
395 @defmac LINK_LIBGCC_SPECIAL_1
396 Define this macro if the driver program should find the library
397 @file{libgcc.a}. If you do not define this macro, the driver program will pass
398 the argument @option{-lgcc} to tell the linker to do the search.
399 This macro is similar to @code{LINK_LIBGCC_SPECIAL}, except that it does
400 not affect @option{-L} options.
403 @defmac LINK_GCC_C_SEQUENCE_SPEC
404 The sequence in which libgcc and libc are specified to the linker.
405 By default this is @code{%G %L %G}.
408 @defmac LINK_COMMAND_SPEC
409 A C string constant giving the complete command line need to execute the
410 linker. When you do this, you will need to update your port each time a
411 change is made to the link command line within @file{gcc.c}. Therefore,
412 define this macro only if you need to completely redefine the command
413 line for invoking the linker and there is no other way to accomplish
414 the effect you need. Overriding this macro may be avoidable by overriding
415 @code{LINK_GCC_C_SEQUENCE_SPEC} instead.
418 @defmac LINK_ELIMINATE_DUPLICATE_LDIRECTORIES
419 A nonzero value causes @command{collect2} to remove duplicate @option{-L@var{directory}} search
420 directories from linking commands. Do not give it a nonzero value if
421 removing duplicate search directories changes the linker's semantics.
424 @defmac MULTILIB_DEFAULTS
425 Define this macro as a C expression for the initializer of an array of
426 string to tell the driver program which options are defaults for this
427 target and thus do not need to be handled specially when using
428 @code{MULTILIB_OPTIONS}.
430 Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in
431 the target makefile fragment or if none of the options listed in
432 @code{MULTILIB_OPTIONS} are set by default.
433 @xref{Target Fragment}.
436 @defmac RELATIVE_PREFIX_NOT_LINKDIR
437 Define this macro to tell @command{gcc} that it should only translate
438 a @option{-B} prefix into a @option{-L} linker option if the prefix
439 indicates an absolute file name.
442 @defmac STANDARD_EXEC_PREFIX
443 Define this macro as a C string constant if you wish to override the
444 standard choice of @file{/usr/local/lib/gcc-lib/} as the default prefix to
445 try when searching for the executable files of the compiler.
448 @defmac MD_EXEC_PREFIX
449 If defined, this macro is an additional prefix to try after
450 @code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched
451 when the @option{-b} option is used, or the compiler is built as a cross
452 compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it
453 to the list of directories used to find the assembler in @file{configure.in}.
456 @defmac STANDARD_STARTFILE_PREFIX
457 Define this macro as a C string constant if you wish to override the
458 standard choice of @file{/usr/local/lib/} as the default prefix to
459 try when searching for startup files such as @file{crt0.o}.
462 @defmac MD_STARTFILE_PREFIX
463 If defined, this macro supplies an additional prefix to try after the
464 standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the
465 @option{-b} option is used, or when the compiler is built as a cross
469 @defmac MD_STARTFILE_PREFIX_1
470 If defined, this macro supplies yet another prefix to try after the
471 standard prefixes. It is not searched when the @option{-b} option is
472 used, or when the compiler is built as a cross compiler.
475 @defmac INIT_ENVIRONMENT
476 Define this macro as a C string constant if you wish to set environment
477 variables for programs called by the driver, such as the assembler and
478 loader. The driver passes the value of this macro to @code{putenv} to
479 initialize the necessary environment variables.
482 @defmac LOCAL_INCLUDE_DIR
483 Define this macro as a C string constant if you wish to override the
484 standard choice of @file{/usr/local/include} as the default prefix to
485 try when searching for local header files. @code{LOCAL_INCLUDE_DIR}
486 comes before @code{SYSTEM_INCLUDE_DIR} in the search order.
488 Cross compilers do not search either @file{/usr/local/include} or its
492 @defmac MODIFY_TARGET_NAME
493 Define this macro if you wish to define command-line switches that
494 modify the default target name.
496 For each switch, you can include a string to be appended to the first
497 part of the configuration name or a string to be deleted from the
498 configuration name, if present. The definition should be an initializer
499 for an array of structures. Each array element should have three
500 elements: the switch name (a string constant, including the initial
501 dash), one of the enumeration codes @code{ADD} or @code{DELETE} to
502 indicate whether the string should be inserted or deleted, and the string
503 to be inserted or deleted (a string constant).
505 For example, on a machine where @samp{64} at the end of the
506 configuration name denotes a 64-bit target and you want the @option{-32}
507 and @option{-64} switches to select between 32- and 64-bit targets, you would
511 #define MODIFY_TARGET_NAME \
512 @{ @{ "-32", DELETE, "64"@}, \
513 @{"-64", ADD, "64"@}@}
517 @defmac SYSTEM_INCLUDE_DIR
518 Define this macro as a C string constant if you wish to specify a
519 system-specific directory to search for header files before the standard
520 directory. @code{SYSTEM_INCLUDE_DIR} comes before
521 @code{STANDARD_INCLUDE_DIR} in the search order.
523 Cross compilers do not use this macro and do not search the directory
527 @defmac STANDARD_INCLUDE_DIR
528 Define this macro as a C string constant if you wish to override the
529 standard choice of @file{/usr/include} as the default prefix to
530 try when searching for header files.
532 Cross compilers ignore this macro and do not search either
533 @file{/usr/include} or its replacement.
536 @defmac STANDARD_INCLUDE_COMPONENT
537 The ``component'' corresponding to @code{STANDARD_INCLUDE_DIR}.
538 See @code{INCLUDE_DEFAULTS}, below, for the description of components.
539 If you do not define this macro, no component is used.
542 @defmac INCLUDE_DEFAULTS
543 Define this macro if you wish to override the entire default search path
544 for include files. For a native compiler, the default search path
545 usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR},
546 @code{SYSTEM_INCLUDE_DIR}, @code{GPLUSPLUS_INCLUDE_DIR}, and
547 @code{STANDARD_INCLUDE_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR}
548 and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile},
549 and specify private search areas for GCC@. The directory
550 @code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs.
552 The definition should be an initializer for an array of structures.
553 Each array element should have four elements: the directory name (a
554 string constant), the component name (also a string constant), a flag
555 for C++-only directories,
556 and a flag showing that the includes in the directory don't need to be
557 wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of
558 the array with a null element.
560 The component name denotes what GNU package the include file is part of,
561 if any, in all upper-case letters. For example, it might be @samp{GCC}
562 or @samp{BINUTILS}. If the package is part of a vendor-supplied
563 operating system, code the component name as @samp{0}.
565 For example, here is the definition used for VAX/VMS:
568 #define INCLUDE_DEFAULTS \
570 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \
571 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \
572 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \
579 Here is the order of prefixes tried for exec files:
583 Any prefixes specified by the user with @option{-B}.
586 The environment variable @code{GCC_EXEC_PREFIX}, if any.
589 The directories specified by the environment variable @code{COMPILER_PATH}.
592 The macro @code{STANDARD_EXEC_PREFIX}.
595 @file{/usr/lib/gcc/}.
598 The macro @code{MD_EXEC_PREFIX}, if any.
601 Here is the order of prefixes tried for startfiles:
605 Any prefixes specified by the user with @option{-B}.
608 The environment variable @code{GCC_EXEC_PREFIX}, if any.
611 The directories specified by the environment variable @code{LIBRARY_PATH}
612 (or port-specific name; native only, cross compilers do not use this).
615 The macro @code{STANDARD_EXEC_PREFIX}.
618 @file{/usr/lib/gcc/}.
621 The macro @code{MD_EXEC_PREFIX}, if any.
624 The macro @code{MD_STARTFILE_PREFIX}, if any.
627 The macro @code{STANDARD_STARTFILE_PREFIX}.
636 @node Run-time Target
637 @section Run-time Target Specification
638 @cindex run-time target specification
639 @cindex predefined macros
640 @cindex target specifications
642 @c prevent bad page break with this line
643 Here are run-time target specifications.
645 @defmac TARGET_CPU_CPP_BUILTINS ()
646 This function-like macro expands to a block of code that defines
647 built-in preprocessor macros and assertions for the target cpu, using
648 the functions @code{builtin_define}, @code{builtin_define_std} and
649 @code{builtin_assert}. When the front end
650 calls this macro it provides a trailing semicolon, and since it has
651 finished command line option processing your code can use those
654 @code{builtin_assert} takes a string in the form you pass to the
655 command-line option @option{-A}, such as @code{cpu=mips}, and creates
656 the assertion. @code{builtin_define} takes a string in the form
657 accepted by option @option{-D} and unconditionally defines the macro.
659 @code{builtin_define_std} takes a string representing the name of an
660 object-like macro. If it doesn't lie in the user's namespace,
661 @code{builtin_define_std} defines it unconditionally. Otherwise, it
662 defines a version with two leading underscores, and another version
663 with two leading and trailing underscores, and defines the original
664 only if an ISO standard was not requested on the command line. For
665 example, passing @code{unix} defines @code{__unix}, @code{__unix__}
666 and possibly @code{unix}; passing @code{_mips} defines @code{__mips},
667 @code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64}
668 defines only @code{_ABI64}.
670 You can also test for the C dialect being compiled. The variable
671 @code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus}
672 or @code{clk_objective_c}. Note that if we are preprocessing
673 assembler, this variable will be @code{clk_c} but the function-like
674 macro @code{preprocessing_asm_p()} will return true, so you might want
675 to check for that first. If you need to check for strict ANSI, the
676 variable @code{flag_iso} can be used. The function-like macro
677 @code{preprocessing_trad_p()} can be used to check for traditional
681 @defmac TARGET_OS_CPP_BUILTINS ()
682 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
683 and is used for the target operating system instead.
686 @defmac TARGET_OBJFMT_CPP_BUILTINS ()
687 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
688 and is used for the target object format. @file{elfos.h} uses this
689 macro to define @code{__ELF__}, so you probably do not need to define
693 @deftypevar {extern int} target_flags
694 This declaration should be present.
697 @cindex optional hardware or system features
698 @cindex features, optional, in system conventions
700 @defmac TARGET_@var{featurename}
701 This series of macros is to allow compiler command arguments to
702 enable or disable the use of optional features of the target machine.
703 For example, one machine description serves both the 68000 and
704 the 68020; a command argument tells the compiler whether it should
705 use 68020-only instructions or not. This command argument works
706 by means of a macro @code{TARGET_68020} that tests a bit in
709 Define a macro @code{TARGET_@var{featurename}} for each such option.
710 Its definition should test a bit in @code{target_flags}. It is
711 recommended that a helper macro @code{MASK_@var{featurename}}
712 is defined for each bit-value to test, and used in
713 @code{TARGET_@var{featurename}} and @code{TARGET_SWITCHES}. For
717 #define TARGET_MASK_68020 1
718 #define TARGET_68020 (target_flags & MASK_68020)
721 One place where these macros are used is in the condition-expressions
722 of instruction patterns. Note how @code{TARGET_68020} appears
723 frequently in the 68000 machine description file, @file{m68k.md}.
724 Another place they are used is in the definitions of the other
725 macros in the @file{@var{machine}.h} file.
728 @defmac TARGET_SWITCHES
729 This macro defines names of command options to set and clear
730 bits in @code{target_flags}. Its definition is an initializer
731 with a subgrouping for each command option.
733 Each subgrouping contains a string constant, that defines the option
734 name, a number, which contains the bits to set in
735 @code{target_flags}, and a second string which is the description
736 displayed by @option{--help}. If the number is negative then the bits specified
737 by the number are cleared instead of being set. If the description
738 string is present but empty, then no help information will be displayed
739 for that option, but it will not count as an undocumented option. The
740 actual option name is made by appending @samp{-m} to the specified name.
741 Non-empty description strings should be marked with @code{N_(@dots{})} for
742 @command{xgettext}. Please do not mark empty strings because the empty
743 string is reserved by GNU gettext. @code{gettext("")} returns the header entry
744 of the message catalog with meta information, not the empty string.
746 In addition to the description for @option{--help},
747 more detailed documentation for each option should be added to
750 One of the subgroupings should have a null string. The number in
751 this grouping is the default value for @code{target_flags}. Any
752 target options act starting with that value.
754 Here is an example which defines @option{-m68000} and @option{-m68020}
755 with opposite meanings, and picks the latter as the default:
758 #define TARGET_SWITCHES \
759 @{ @{ "68020", MASK_68020, "" @}, \
760 @{ "68000", -MASK_68020, \
761 N_("Compile for the 68000") @}, \
762 @{ "", MASK_68020, "" @}, \
767 @defmac TARGET_OPTIONS
768 This macro is similar to @code{TARGET_SWITCHES} but defines names of command
769 options that have values. Its definition is an initializer with a
770 subgrouping for each command option.
772 Each subgrouping contains a string constant, that defines the option
773 name, the address of a variable, a description string, and a value.
774 Non-empty description strings should be marked with @code{N_(@dots{})}
775 for @command{xgettext}. Please do not mark empty strings because the
776 empty string is reserved by GNU gettext. @code{gettext("")} returns the
777 header entry of the message catalog with meta information, not the empty
780 If the value listed in the table is @code{NULL}, then the variable, type
781 @code{char *}, is set to the variable part of the given option if the
782 fixed part matches. In other words, if the first part of the option
783 matches what's in the table, the variable will be set to point to the
784 rest of the option. This allows the user to specify a value for that
785 option. The actual option name is made by appending @samp{-m} to the
786 specified name. Again, each option should also be documented in
789 If the value listed in the table is non-@code{NULL}, then the option
790 must match the option in the table exactly (with @samp{-m}), and the
791 variable is set to point to the value listed in the table.
793 Here is an example which defines @option{-mshort-data-@var{number}}. If the
794 given option is @option{-mshort-data-512}, the variable @code{m88k_short_data}
795 will be set to the string @code{"512"}.
798 extern char *m88k_short_data;
799 #define TARGET_OPTIONS \
800 @{ @{ "short-data-", &m88k_short_data, \
801 N_("Specify the size of the short data section"), 0 @} @}
804 Here is a variant of the above that allows the user to also specify
805 just @option{-mshort-data} where a default of @code{"64"} is used.
808 extern char *m88k_short_data;
809 #define TARGET_OPTIONS \
810 @{ @{ "short-data-", &m88k_short_data, \
811 N_("Specify the size of the short data section"), 0 @} \
812 @{ "short-data", &m88k_short_data, "", "64" @},
816 Here is an example which defines @option{-mno-alu}, @option{-malu1}, and
817 @option{-malu2} as a three-state switch, along with suitable macros for
818 checking the state of the option (documentation is elided for brevity).
822 char *chip_alu = ""; /* Specify default here. */
825 extern char *chip_alu;
826 #define TARGET_OPTIONS \
827 @{ @{ "no-alu", &chip_alu, "", "" @}, \
828 @{ "alu1", &chip_alu, "", "1" @}, \
829 @{ "alu2", &chip_alu, "", "2" @}, @}
830 #define TARGET_ALU (chip_alu[0] != '\0')
831 #define TARGET_ALU1 (chip_alu[0] == '1')
832 #define TARGET_ALU2 (chip_alu[0] == '2')
836 @defmac TARGET_VERSION
837 This macro is a C statement to print on @code{stderr} a string
838 describing the particular machine description choice. Every machine
839 description should define @code{TARGET_VERSION}. For example:
843 #define TARGET_VERSION \
844 fprintf (stderr, " (68k, Motorola syntax)");
846 #define TARGET_VERSION \
847 fprintf (stderr, " (68k, MIT syntax)");
852 @defmac OVERRIDE_OPTIONS
853 Sometimes certain combinations of command options do not make sense on
854 a particular target machine. You can define a macro
855 @code{OVERRIDE_OPTIONS} to take account of this. This macro, if
856 defined, is executed once just after all the command options have been
859 Don't use this macro to turn on various extra optimizations for
860 @option{-O}. That is what @code{OPTIMIZATION_OPTIONS} is for.
863 @defmac OPTIMIZATION_OPTIONS (@var{level}, @var{size})
864 Some machines may desire to change what optimizations are performed for
865 various optimization levels. This macro, if defined, is executed once
866 just after the optimization level is determined and before the remainder
867 of the command options have been parsed. Values set in this macro are
868 used as the default values for the other command line options.
870 @var{level} is the optimization level specified; 2 if @option{-O2} is
871 specified, 1 if @option{-O} is specified, and 0 if neither is specified.
873 @var{size} is nonzero if @option{-Os} is specified and zero otherwise.
875 You should not use this macro to change options that are not
876 machine-specific. These should uniformly selected by the same
877 optimization level on all supported machines. Use this macro to enable
878 machine-specific optimizations.
880 @strong{Do not examine @code{write_symbols} in
881 this macro!} The debugging options are not supposed to alter the
885 @defmac CAN_DEBUG_WITHOUT_FP
886 Define this macro if debugging can be performed even without a frame
887 pointer. If this macro is defined, GCC will turn on the
888 @option{-fomit-frame-pointer} option whenever @option{-O} is specified.
891 @node Per-Function Data
892 @section Defining data structures for per-function information.
893 @cindex per-function data
894 @cindex data structures
896 If the target needs to store information on a per-function basis, GCC
897 provides a macro and a couple of variables to allow this. Note, just
898 using statics to store the information is a bad idea, since GCC supports
899 nested functions, so you can be halfway through encoding one function
900 when another one comes along.
902 GCC defines a data structure called @code{struct function} which
903 contains all of the data specific to an individual function. This
904 structure contains a field called @code{machine} whose type is
905 @code{struct machine_function *}, which can be used by targets to point
906 to their own specific data.
908 If a target needs per-function specific data it should define the type
909 @code{struct machine_function} and also the macro @code{INIT_EXPANDERS}.
910 This macro should be used to initialize the function pointer
911 @code{init_machine_status}. This pointer is explained below.
913 One typical use of per-function, target specific data is to create an
914 RTX to hold the register containing the function's return address. This
915 RTX can then be used to implement the @code{__builtin_return_address}
916 function, for level 0.
918 Note---earlier implementations of GCC used a single data area to hold
919 all of the per-function information. Thus when processing of a nested
920 function began the old per-function data had to be pushed onto a
921 stack, and when the processing was finished, it had to be popped off the
922 stack. GCC used to provide function pointers called
923 @code{save_machine_status} and @code{restore_machine_status} to handle
924 the saving and restoring of the target specific information. Since the
925 single data area approach is no longer used, these pointers are no
928 @defmac INIT_EXPANDERS
929 Macro called to initialize any target specific information. This macro
930 is called once per function, before generation of any RTL has begun.
931 The intention of this macro is to allow the initialization of the
932 function pointer @code{init_machine_status}.
935 @deftypevar {void (*)(struct function *)} init_machine_status
936 If this function pointer is non-@code{NULL} it will be called once per
937 function, before function compilation starts, in order to allow the
938 target to perform any target specific initialization of the
939 @code{struct function} structure. It is intended that this would be
940 used to initialize the @code{machine} of that structure.
942 @code{struct machine_function} structures are expected to be freed by GC.
943 Generally, any memory that they reference must be allocated by using
944 @code{ggc_alloc}, including the structure itself.
948 @section Storage Layout
949 @cindex storage layout
951 Note that the definitions of the macros in this table which are sizes or
952 alignments measured in bits do not need to be constant. They can be C
953 expressions that refer to static variables, such as the @code{target_flags}.
954 @xref{Run-time Target}.
956 @defmac BITS_BIG_ENDIAN
957 Define this macro to have the value 1 if the most significant bit in a
958 byte has the lowest number; otherwise define it to have the value zero.
959 This means that bit-field instructions count from the most significant
960 bit. If the machine has no bit-field instructions, then this must still
961 be defined, but it doesn't matter which value it is defined to. This
962 macro need not be a constant.
964 This macro does not affect the way structure fields are packed into
965 bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
968 @defmac BYTES_BIG_ENDIAN
969 Define this macro to have the value 1 if the most significant byte in a
970 word has the lowest number. This macro need not be a constant.
973 @defmac WORDS_BIG_ENDIAN
974 Define this macro to have the value 1 if, in a multiword object, the
975 most significant word has the lowest number. This applies to both
976 memory locations and registers; GCC fundamentally assumes that the
977 order of words in memory is the same as the order in registers. This
978 macro need not be a constant.
981 @defmac LIBGCC2_WORDS_BIG_ENDIAN
982 Define this macro if @code{WORDS_BIG_ENDIAN} is not constant. This must be a
983 constant value with the same meaning as @code{WORDS_BIG_ENDIAN}, which will be
984 used only when compiling @file{libgcc2.c}. Typically the value will be set
985 based on preprocessor defines.
988 @defmac FLOAT_WORDS_BIG_ENDIAN
989 Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or
990 @code{TFmode} floating point numbers are stored in memory with the word
991 containing the sign bit at the lowest address; otherwise define it to
992 have the value 0. This macro need not be a constant.
994 You need not define this macro if the ordering is the same as for
998 @defmac BITS_PER_UNIT
999 Define this macro to be the number of bits in an addressable storage
1000 unit (byte). If you do not define this macro the default is 8.
1003 @defmac BITS_PER_WORD
1004 Number of bits in a word. If you do not define this macro, the default
1005 is @code{BITS_PER_UNIT * UNITS_PER_WORD}.
1008 @defmac MAX_BITS_PER_WORD
1009 Maximum number of bits in a word. If this is undefined, the default is
1010 @code{BITS_PER_WORD}. Otherwise, it is the constant value that is the
1011 largest value that @code{BITS_PER_WORD} can have at run-time.
1014 @defmac UNITS_PER_WORD
1015 Number of storage units in a word; normally 4.
1018 @defmac MIN_UNITS_PER_WORD
1019 Minimum number of units in a word. If this is undefined, the default is
1020 @code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the
1021 smallest value that @code{UNITS_PER_WORD} can have at run-time.
1024 @defmac POINTER_SIZE
1025 Width of a pointer, in bits. You must specify a value no wider than the
1026 width of @code{Pmode}. If it is not equal to the width of @code{Pmode},
1027 you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify
1028 a value the default is @code{BITS_PER_WORD}.
1031 @defmac POINTERS_EXTEND_UNSIGNED
1032 A C expression whose value is greater than zero if pointers that need to be
1033 extended from being @code{POINTER_SIZE} bits wide to @code{Pmode} are to
1034 be zero-extended and zero if they are to be sign-extended. If the value
1035 is less then zero then there must be an "ptr_extend" instruction that
1036 extends a pointer from @code{POINTER_SIZE} to @code{Pmode}.
1038 You need not define this macro if the @code{POINTER_SIZE} is equal
1039 to the width of @code{Pmode}.
1042 @defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type})
1043 A macro to update @var{m} and @var{unsignedp} when an object whose type
1044 is @var{type} and which has the specified mode and signedness is to be
1045 stored in a register. This macro is only called when @var{type} is a
1048 On most RISC machines, which only have operations that operate on a full
1049 register, define this macro to set @var{m} to @code{word_mode} if
1050 @var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most
1051 cases, only integer modes should be widened because wider-precision
1052 floating-point operations are usually more expensive than their narrower
1055 For most machines, the macro definition does not change @var{unsignedp}.
1056 However, some machines, have instructions that preferentially handle
1057 either signed or unsigned quantities of certain modes. For example, on
1058 the DEC Alpha, 32-bit loads from memory and 32-bit add instructions
1059 sign-extend the result to 64 bits. On such machines, set
1060 @var{unsignedp} according to which kind of extension is more efficient.
1062 Do not define this macro if it would never modify @var{m}.
1065 @defmac PROMOTE_FUNCTION_ARGS
1066 Define this macro if the promotion described by @code{PROMOTE_MODE}
1067 should also be done for outgoing function arguments.
1070 @defmac PROMOTE_FUNCTION_RETURN
1071 Define this macro if the promotion described by @code{PROMOTE_MODE}
1072 should also be done for the return value of functions.
1074 If this macro is defined, @code{FUNCTION_VALUE} must perform the same
1075 promotions done by @code{PROMOTE_MODE}.
1078 @defmac PROMOTE_FOR_CALL_ONLY
1079 Define this macro if the promotion described by @code{PROMOTE_MODE}
1080 should @emph{only} be performed for outgoing function arguments or
1081 function return values, as specified by @code{PROMOTE_FUNCTION_ARGS}
1082 and @code{PROMOTE_FUNCTION_RETURN}, respectively.
1085 @defmac PARM_BOUNDARY
1086 Normal alignment required for function parameters on the stack, in
1087 bits. All stack parameters receive at least this much alignment
1088 regardless of data type. On most machines, this is the same as the
1092 @defmac STACK_BOUNDARY
1093 Define this macro to the minimum alignment enforced by hardware for the
1094 stack pointer on this machine. The definition is a C expression for the
1095 desired alignment (measured in bits). This value is used as a default
1096 if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines,
1097 this should be the same as @code{PARM_BOUNDARY}.
1100 @defmac PREFERRED_STACK_BOUNDARY
1101 Define this macro if you wish to preserve a certain alignment for the
1102 stack pointer, greater than what the hardware enforces. The definition
1103 is a C expression for the desired alignment (measured in bits). This
1104 macro must evaluate to a value equal to or larger than
1105 @code{STACK_BOUNDARY}.
1108 @defmac FORCE_PREFERRED_STACK_BOUNDARY_IN_MAIN
1109 A C expression that evaluates true if @code{PREFERRED_STACK_BOUNDARY} is
1110 not guaranteed by the runtime and we should emit code to align the stack
1111 at the beginning of @code{main}.
1113 @cindex @code{PUSH_ROUNDING}, interaction with @code{PREFERRED_STACK_BOUNDARY}
1114 If @code{PUSH_ROUNDING} is not defined, the stack will always be aligned
1115 to the specified boundary. If @code{PUSH_ROUNDING} is defined and specifies
1116 a less strict alignment than @code{PREFERRED_STACK_BOUNDARY}, the stack may
1117 be momentarily unaligned while pushing arguments.
1120 @defmac FUNCTION_BOUNDARY
1121 Alignment required for a function entry point, in bits.
1124 @defmac BIGGEST_ALIGNMENT
1125 Biggest alignment that any data type can require on this machine, in bits.
1128 @defmac MINIMUM_ATOMIC_ALIGNMENT
1129 If defined, the smallest alignment, in bits, that can be given to an
1130 object that can be referenced in one operation, without disturbing any
1131 nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger
1132 on machines that don't have byte or half-word store operations.
1135 @defmac BIGGEST_FIELD_ALIGNMENT
1136 Biggest alignment that any structure or union field can require on this
1137 machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for
1138 structure and union fields only, unless the field alignment has been set
1139 by the @code{__attribute__ ((aligned (@var{n})))} construct.
1142 @defmac ADJUST_FIELD_ALIGN (@var{field}, @var{computed})
1143 An expression for the alignment of a structure field @var{field} if the
1144 alignment computed in the usual way (including applying of
1145 @code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the
1146 alignment) is @var{computed}. It overrides alignment only if the
1147 field alignment has not been set by the
1148 @code{__attribute__ ((aligned (@var{n})))} construct.
1151 @defmac MAX_OFILE_ALIGNMENT
1152 Biggest alignment supported by the object file format of this machine.
1153 Use this macro to limit the alignment which can be specified using the
1154 @code{__attribute__ ((aligned (@var{n})))} construct. If not defined,
1155 the default value is @code{BIGGEST_ALIGNMENT}.
1158 @defmac DATA_ALIGNMENT (@var{type}, @var{basic-align})
1159 If defined, a C expression to compute the alignment for a variable in
1160 the static store. @var{type} is the data type, and @var{basic-align} is
1161 the alignment that the object would ordinarily have. The value of this
1162 macro is used instead of that alignment to align the object.
1164 If this macro is not defined, then @var{basic-align} is used.
1167 One use of this macro is to increase alignment of medium-size data to
1168 make it all fit in fewer cache lines. Another is to cause character
1169 arrays to be word-aligned so that @code{strcpy} calls that copy
1170 constants to character arrays can be done inline.
1173 @defmac CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align})
1174 If defined, a C expression to compute the alignment given to a constant
1175 that is being placed in memory. @var{constant} is the constant and
1176 @var{basic-align} is the alignment that the object would ordinarily
1177 have. The value of this macro is used instead of that alignment to
1180 If this macro is not defined, then @var{basic-align} is used.
1182 The typical use of this macro is to increase alignment for string
1183 constants to be word aligned so that @code{strcpy} calls that copy
1184 constants can be done inline.
1187 @defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align})
1188 If defined, a C expression to compute the alignment for a variable in
1189 the local store. @var{type} is the data type, and @var{basic-align} is
1190 the alignment that the object would ordinarily have. The value of this
1191 macro is used instead of that alignment to align the object.
1193 If this macro is not defined, then @var{basic-align} is used.
1195 One use of this macro is to increase alignment of medium-size data to
1196 make it all fit in fewer cache lines.
1199 @defmac EMPTY_FIELD_BOUNDARY
1200 Alignment in bits to be given to a structure bit-field that follows an
1201 empty field such as @code{int : 0;}.
1203 If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro.
1206 @defmac STRUCTURE_SIZE_BOUNDARY
1207 Number of bits which any structure or union's size must be a multiple of.
1208 Each structure or union's size is rounded up to a multiple of this.
1210 If you do not define this macro, the default is the same as
1211 @code{BITS_PER_UNIT}.
1214 @defmac STRICT_ALIGNMENT
1215 Define this macro to be the value 1 if instructions will fail to work
1216 if given data not on the nominal alignment. If instructions will merely
1217 go slower in that case, define this macro as 0.
1220 @defmac PCC_BITFIELD_TYPE_MATTERS
1221 Define this if you wish to imitate the way many other C compilers handle
1222 alignment of bit-fields and the structures that contain them.
1224 The behavior is that the type written for a named bit-field (@code{int},
1225 @code{short}, or other integer type) imposes an alignment for the entire
1226 structure, as if the structure really did contain an ordinary field of
1227 that type. In addition, the bit-field is placed within the structure so
1228 that it would fit within such a field, not crossing a boundary for it.
1230 Thus, on most machines, a named bit-field whose type is written as
1231 @code{int} would not cross a four-byte boundary, and would force
1232 four-byte alignment for the whole structure. (The alignment used may
1233 not be four bytes; it is controlled by the other alignment parameters.)
1235 An unnamed bit-field will not affect the alignment of the containing
1238 If the macro is defined, its definition should be a C expression;
1239 a nonzero value for the expression enables this behavior.
1241 Note that if this macro is not defined, or its value is zero, some
1242 bit-fields may cross more than one alignment boundary. The compiler can
1243 support such references if there are @samp{insv}, @samp{extv}, and
1244 @samp{extzv} insns that can directly reference memory.
1246 The other known way of making bit-fields work is to define
1247 @code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}.
1248 Then every structure can be accessed with fullwords.
1250 Unless the machine has bit-field instructions or you define
1251 @code{STRUCTURE_SIZE_BOUNDARY} that way, you must define
1252 @code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value.
1254 If your aim is to make GCC use the same conventions for laying out
1255 bit-fields as are used by another compiler, here is how to investigate
1256 what the other compiler does. Compile and run this program:
1275 printf ("Size of foo1 is %d\n",
1276 sizeof (struct foo1));
1277 printf ("Size of foo2 is %d\n",
1278 sizeof (struct foo2));
1283 If this prints 2 and 5, then the compiler's behavior is what you would
1284 get from @code{PCC_BITFIELD_TYPE_MATTERS}.
1287 @defmac BITFIELD_NBYTES_LIMITED
1288 Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited
1289 to aligning a bit-field within the structure.
1292 @defmac MEMBER_TYPE_FORCES_BLK (@var{field}, @var{mode})
1293 Return 1 if a structure or array containing @var{field} should be accessed using
1296 If @var{field} is the only field in the structure, @var{mode} is its
1297 mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the
1298 case where structures of one field would require the structure's mode to
1299 retain the field's mode.
1301 Normally, this is not needed. See the file @file{c4x.h} for an example
1302 of how to use this macro to prevent a structure having a floating point
1303 field from being accessed in an integer mode.
1306 @defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified})
1307 Define this macro as an expression for the alignment of a type (given
1308 by @var{type} as a tree node) if the alignment computed in the usual
1309 way is @var{computed} and the alignment explicitly specified was
1312 The default is to use @var{specified} if it is larger; otherwise, use
1313 the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT}
1316 @defmac MAX_FIXED_MODE_SIZE
1317 An integer expression for the size in bits of the largest integer
1318 machine mode that should actually be used. All integer machine modes of
1319 this size or smaller can be used for structures and unions with the
1320 appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE
1321 (DImode)} is assumed.
1324 @defmac VECTOR_MODE_SUPPORTED_P (@var{mode})
1325 Define this macro to be nonzero if the port is prepared to handle insns
1326 involving vector mode @var{mode}. At the very least, it must have move
1327 patterns for this mode.
1330 @defmac STACK_SAVEAREA_MODE (@var{save_level})
1331 If defined, an expression of type @code{enum machine_mode} that
1332 specifies the mode of the save area operand of a
1333 @code{save_stack_@var{level}} named pattern (@pxref{Standard Names}).
1334 @var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or
1335 @code{SAVE_NONLOCAL} and selects which of the three named patterns is
1336 having its mode specified.
1338 You need not define this macro if it always returns @code{Pmode}. You
1339 would most commonly define this macro if the
1340 @code{save_stack_@var{level}} patterns need to support both a 32- and a
1344 @defmac STACK_SIZE_MODE
1345 If defined, an expression of type @code{enum machine_mode} that
1346 specifies the mode of the size increment operand of an
1347 @code{allocate_stack} named pattern (@pxref{Standard Names}).
1349 You need not define this macro if it always returns @code{word_mode}.
1350 You would most commonly define this macro if the @code{allocate_stack}
1351 pattern needs to support both a 32- and a 64-bit mode.
1354 @defmac TARGET_FLOAT_FORMAT
1355 A code distinguishing the floating point format of the target machine.
1356 There are four defined values:
1359 @item IEEE_FLOAT_FORMAT
1360 This code indicates IEEE floating point. It is the default; there is no
1361 need to define @code{TARGET_FLOAT_FORMAT} when the format is IEEE@.
1363 @item VAX_FLOAT_FORMAT
1364 This code indicates the ``F float'' (for @code{float}) and ``D float''
1365 or ``G float'' formats (for @code{double}) used on the VAX and PDP-11@.
1367 @item IBM_FLOAT_FORMAT
1368 This code indicates the format used on the IBM System/370.
1370 @item C4X_FLOAT_FORMAT
1371 This code indicates the format used on the TMS320C3x/C4x.
1374 If your target uses a floating point format other than these, you must
1375 define a new @var{name}_FLOAT_FORMAT code for it, and add support for
1376 it to @file{real.c}.
1378 The ordering of the component words of floating point values stored in
1379 memory is controlled by @code{FLOAT_WORDS_BIG_ENDIAN}.
1382 @defmac MODE_HAS_NANS (@var{mode})
1383 When defined, this macro should be true if @var{mode} has a NaN
1384 representation. The compiler assumes that NaNs are not equal to
1385 anything (including themselves) and that addition, subtraction,
1386 multiplication and division all return NaNs when one operand is
1389 By default, this macro is true if @var{mode} is a floating-point
1390 mode and the target floating-point format is IEEE@.
1393 @defmac MODE_HAS_INFINITIES (@var{mode})
1394 This macro should be true if @var{mode} can represent infinity. At
1395 present, the compiler uses this macro to decide whether @samp{x - x}
1396 is always defined. By default, the macro is true when @var{mode}
1397 is a floating-point mode and the target format is IEEE@.
1400 @defmac MODE_HAS_SIGNED_ZEROS (@var{mode})
1401 True if @var{mode} distinguishes between positive and negative zero.
1402 The rules are expected to follow the IEEE standard:
1406 @samp{x + x} has the same sign as @samp{x}.
1409 If the sum of two values with opposite sign is zero, the result is
1410 positive for all rounding modes expect towards @minus{}infinity, for
1411 which it is negative.
1414 The sign of a product or quotient is negative when exactly one
1415 of the operands is negative.
1418 The default definition is true if @var{mode} is a floating-point
1419 mode and the target format is IEEE@.
1422 @defmac MODE_HAS_SIGN_DEPENDENT_ROUNDING (@var{mode})
1423 If defined, this macro should be true for @var{mode} if it has at
1424 least one rounding mode in which @samp{x} and @samp{-x} can be
1425 rounded to numbers of different magnitude. Two such modes are
1426 towards @minus{}infinity and towards +infinity.
1428 The default definition of this macro is true if @var{mode} is
1429 a floating-point mode and the target format is IEEE@.
1432 @defmac ROUND_TOWARDS_ZERO
1433 If defined, this macro should be true if the prevailing rounding
1434 mode is towards zero. A true value has the following effects:
1438 @code{MODE_HAS_SIGN_DEPENDENT_ROUNDING} will be false for all modes.
1441 @file{libgcc.a}'s floating-point emulator will round towards zero
1442 rather than towards nearest.
1445 The compiler's floating-point emulator will round towards zero after
1446 doing arithmetic, and when converting from the internal float format to
1450 The macro does not affect the parsing of string literals. When the
1451 primary rounding mode is towards zero, library functions like
1452 @code{strtod} might still round towards nearest, and the compiler's
1453 parser should behave like the target's @code{strtod} where possible.
1455 Not defining this macro is equivalent to returning zero.
1458 @defmac LARGEST_EXPONENT_IS_NORMAL (@var{size})
1459 This macro should return true if floats with @var{size}
1460 bits do not have a NaN or infinity representation, but use the largest
1461 exponent for normal numbers instead.
1463 Defining this macro to true for @var{size} causes @code{MODE_HAS_NANS}
1464 and @code{MODE_HAS_INFINITIES} to be false for @var{size}-bit modes.
1465 It also affects the way @file{libgcc.a} and @file{real.c} emulate
1466 floating-point arithmetic.
1468 The default definition of this macro returns false for all sizes.
1471 @deftypefn {Target Hook} bool TARGET_VECTOR_OPAQUE_P (tree @var{type})
1472 This target hook should return @code{true} a vector is opaque. That
1473 is, if no cast is needed when copying a vector value of type
1474 @var{type} into another vector lvalue of the same size. Vector opaque
1475 types cannot be initialized. The default is that there are no such
1479 @deftypefn {Target Hook} bool TARGET_MS_BITFIELD_LAYOUT_P (tree @var{record_type})
1480 This target hook returns @code{true} if bit-fields in the given
1481 @var{record_type} are to be laid out following the rules of Microsoft
1482 Visual C/C++, namely: (i) a bit-field won't share the same storage
1483 unit with the previous bit-field if their underlying types have
1484 different sizes, and the bit-field will be aligned to the highest
1485 alignment of the underlying types of itself and of the previous
1486 bit-field; (ii) a zero-sized bit-field will affect the alignment of
1487 the whole enclosing structure, even if it is unnamed; except that
1488 (iii) a zero-sized bit-field will be disregarded unless it follows
1489 another bit-field of nonzero size. If this hook returns @code{true},
1490 other macros that control bit-field layout are ignored.
1492 When a bit-field is inserted into a packed record, the whole size
1493 of the underlying type is used by one or more same-size adjacent
1494 bit-fields (that is, if its long:3, 32 bits is used in the record,
1495 and any additional adjacent long bit-fields are packed into the same
1496 chunk of 32 bits. However, if the size changes, a new field of that
1497 size is allocated). In an unpacked record, this is the same as using
1498 alignment, but not equivalent when packing.
1500 If both MS bit-fields and @samp{__attribute__((packed))} are used,
1501 the latter will take precedence. If @samp{__attribute__((packed))} is
1502 used on a single field when MS bit-fields are in use, it will take
1503 precedence for that field, but the alignment of the rest of the structure
1504 may affect its placement.
1508 @section Layout of Source Language Data Types
1510 These macros define the sizes and other characteristics of the standard
1511 basic data types used in programs being compiled. Unlike the macros in
1512 the previous section, these apply to specific features of C and related
1513 languages, rather than to fundamental aspects of storage layout.
1515 @defmac INT_TYPE_SIZE
1516 A C expression for the size in bits of the type @code{int} on the
1517 target machine. If you don't define this, the default is one word.
1520 @defmac SHORT_TYPE_SIZE
1521 A C expression for the size in bits of the type @code{short} on the
1522 target machine. If you don't define this, the default is half a word.
1523 (If this would be less than one storage unit, it is rounded up to one
1527 @defmac LONG_TYPE_SIZE
1528 A C expression for the size in bits of the type @code{long} on the
1529 target machine. If you don't define this, the default is one word.
1532 @defmac ADA_LONG_TYPE_SIZE
1533 On some machines, the size used for the Ada equivalent of the type
1534 @code{long} by a native Ada compiler differs from that used by C. In
1535 that situation, define this macro to be a C expression to be used for
1536 the size of that type. If you don't define this, the default is the
1537 value of @code{LONG_TYPE_SIZE}.
1540 @defmac MAX_LONG_TYPE_SIZE
1541 Maximum number for the size in bits of the type @code{long} on the
1542 target machine. If this is undefined, the default is
1543 @code{LONG_TYPE_SIZE}. Otherwise, it is the constant value that is the
1544 largest value that @code{LONG_TYPE_SIZE} can have at run-time. This is
1548 @defmac LONG_LONG_TYPE_SIZE
1549 A C expression for the size in bits of the type @code{long long} on the
1550 target machine. If you don't define this, the default is two
1551 words. If you want to support GNU Ada on your machine, the value of this
1552 macro must be at least 64.
1555 @defmac CHAR_TYPE_SIZE
1556 A C expression for the size in bits of the type @code{char} on the
1557 target machine. If you don't define this, the default is
1558 @code{BITS_PER_UNIT}.
1561 @defmac BOOL_TYPE_SIZE
1562 A C expression for the size in bits of the C++ type @code{bool} and
1563 C99 type @code{_Bool} on the target machine. If you don't define
1564 this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}.
1567 @defmac FLOAT_TYPE_SIZE
1568 A C expression for the size in bits of the type @code{float} on the
1569 target machine. If you don't define this, the default is one word.
1572 @defmac DOUBLE_TYPE_SIZE
1573 A C expression for the size in bits of the type @code{double} on the
1574 target machine. If you don't define this, the default is two
1578 @defmac LONG_DOUBLE_TYPE_SIZE
1579 A C expression for the size in bits of the type @code{long double} on
1580 the target machine. If you don't define this, the default is two
1584 @defmac MAX_LONG_DOUBLE_TYPE_SIZE
1585 Maximum number for the size in bits of the type @code{long double} on the
1586 target machine. If this is undefined, the default is
1587 @code{LONG_DOUBLE_TYPE_SIZE}. Otherwise, it is the constant value that is
1588 the largest value that @code{LONG_DOUBLE_TYPE_SIZE} can have at run-time.
1589 This is used in @code{cpp}.
1592 @defmac TARGET_FLT_EVAL_METHOD
1593 A C expression for the value for @code{FLT_EVAL_METHOD} in @file{float.h},
1594 assuming, if applicable, that the floating-point control word is in its
1595 default state. If you do not define this macro the value of
1596 @code{FLT_EVAL_METHOD} will be zero.
1599 @defmac WIDEST_HARDWARE_FP_SIZE
1600 A C expression for the size in bits of the widest floating-point format
1601 supported by the hardware. If you define this macro, you must specify a
1602 value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}.
1603 If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE}
1607 @defmac DEFAULT_SIGNED_CHAR
1608 An expression whose value is 1 or 0, according to whether the type
1609 @code{char} should be signed or unsigned by default. The user can
1610 always override this default with the options @option{-fsigned-char}
1611 and @option{-funsigned-char}.
1614 @defmac DEFAULT_SHORT_ENUMS
1615 A C expression to determine whether to give an @code{enum} type
1616 only as many bytes as it takes to represent the range of possible values
1617 of that type. A nonzero value means to do that; a zero value means all
1618 @code{enum} types should be allocated like @code{int}.
1620 If you don't define the macro, the default is 0.
1624 A C expression for a string describing the name of the data type to use
1625 for size values. The typedef name @code{size_t} is defined using the
1626 contents of the string.
1628 The string can contain more than one keyword. If so, separate them with
1629 spaces, and write first any length keyword, then @code{unsigned} if
1630 appropriate, and finally @code{int}. The string must exactly match one
1631 of the data type names defined in the function
1632 @code{init_decl_processing} in the file @file{c-decl.c}. You may not
1633 omit @code{int} or change the order---that would cause the compiler to
1636 If you don't define this macro, the default is @code{"long unsigned
1640 @defmac PTRDIFF_TYPE
1641 A C expression for a string describing the name of the data type to use
1642 for the result of subtracting two pointers. The typedef name
1643 @code{ptrdiff_t} is defined using the contents of the string. See
1644 @code{SIZE_TYPE} above for more information.
1646 If you don't define this macro, the default is @code{"long int"}.
1650 A C expression for a string describing the name of the data type to use
1651 for wide characters. The typedef name @code{wchar_t} is defined using
1652 the contents of the string. See @code{SIZE_TYPE} above for more
1655 If you don't define this macro, the default is @code{"int"}.
1658 @defmac WCHAR_TYPE_SIZE
1659 A C expression for the size in bits of the data type for wide
1660 characters. This is used in @code{cpp}, which cannot make use of
1664 @defmac MAX_WCHAR_TYPE_SIZE
1665 Maximum number for the size in bits of the data type for wide
1666 characters. If this is undefined, the default is
1667 @code{WCHAR_TYPE_SIZE}. Otherwise, it is the constant value that is the
1668 largest value that @code{WCHAR_TYPE_SIZE} can have at run-time. This is
1672 @defmac GCOV_TYPE_SIZE
1673 A C expression for the size in bits of the type used for gcov counters on the
1674 target machine. If you don't define this, the default is one
1675 @code{LONG_TYPE_SIZE} in case it is greater or equal to 64-bit and
1676 @code{LONG_LONG_TYPE_SIZE} otherwise. You may want to re-define the type to
1677 ensure atomicity for counters in multithreaded programs.
1681 A C expression for a string describing the name of the data type to
1682 use for wide characters passed to @code{printf} and returned from
1683 @code{getwc}. The typedef name @code{wint_t} is defined using the
1684 contents of the string. See @code{SIZE_TYPE} above for more
1687 If you don't define this macro, the default is @code{"unsigned int"}.
1691 A C expression for a string describing the name of the data type that
1692 can represent any value of any standard or extended signed integer type.
1693 The typedef name @code{intmax_t} is defined using the contents of the
1694 string. See @code{SIZE_TYPE} above for more information.
1696 If you don't define this macro, the default is the first of
1697 @code{"int"}, @code{"long int"}, or @code{"long long int"} that has as
1698 much precision as @code{long long int}.
1701 @defmac UINTMAX_TYPE
1702 A C expression for a string describing the name of the data type that
1703 can represent any value of any standard or extended unsigned integer
1704 type. The typedef name @code{uintmax_t} is defined using the contents
1705 of the string. See @code{SIZE_TYPE} above for more information.
1707 If you don't define this macro, the default is the first of
1708 @code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long
1709 unsigned int"} that has as much precision as @code{long long unsigned
1713 @defmac TARGET_PTRMEMFUNC_VBIT_LOCATION
1714 The C++ compiler represents a pointer-to-member-function with a struct
1721 ptrdiff_t vtable_index;
1728 The C++ compiler must use one bit to indicate whether the function that
1729 will be called through a pointer-to-member-function is virtual.
1730 Normally, we assume that the low-order bit of a function pointer must
1731 always be zero. Then, by ensuring that the vtable_index is odd, we can
1732 distinguish which variant of the union is in use. But, on some
1733 platforms function pointers can be odd, and so this doesn't work. In
1734 that case, we use the low-order bit of the @code{delta} field, and shift
1735 the remainder of the @code{delta} field to the left.
1737 GCC will automatically make the right selection about where to store
1738 this bit using the @code{FUNCTION_BOUNDARY} setting for your platform.
1739 However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY}
1740 set such that functions always start at even addresses, but the lowest
1741 bit of pointers to functions indicate whether the function at that
1742 address is in ARM or Thumb mode. If this is the case of your
1743 architecture, you should define this macro to
1744 @code{ptrmemfunc_vbit_in_delta}.
1746 In general, you should not have to define this macro. On architectures
1747 in which function addresses are always even, according to
1748 @code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to
1749 @code{ptrmemfunc_vbit_in_pfn}.
1752 @defmac TARGET_VTABLE_USES_DESCRIPTORS
1753 Normally, the C++ compiler uses function pointers in vtables. This
1754 macro allows the target to change to use ``function descriptors''
1755 instead. Function descriptors are found on targets for whom a
1756 function pointer is actually a small data structure. Normally the
1757 data structure consists of the actual code address plus a data
1758 pointer to which the function's data is relative.
1760 If vtables are used, the value of this macro should be the number
1761 of words that the function descriptor occupies.
1764 @defmac TARGET_VTABLE_ENTRY_ALIGN
1765 By default, the vtable entries are void pointers, the so the alignment
1766 is the same as pointer alignment. The value of this macro specifies
1767 the alignment of the vtable entry in bits. It should be defined only
1768 when special alignment is necessary. */
1771 @defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE
1772 There are a few non-descriptor entries in the vtable at offsets below
1773 zero. If these entries must be padded (say, to preserve the alignment
1774 specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number
1775 of words in each data entry.
1778 @node Escape Sequences
1779 @section Target Character Escape Sequences
1780 @cindex escape sequences
1782 By default, GCC assumes that the C character escape sequences take on
1783 their ASCII values for the target. If this is not correct, you must
1784 explicitly define all of the macros below. All of them must evaluate
1785 to constants; they are used in @code{case} statements.
1791 @findex TARGET_NEWLINE
1794 @multitable {@code{TARGET_NEWLINE}} {Escape} {ASCII character}
1795 @item Macro @tab Escape @tab ASCII character
1796 @item @code{TARGET_BELL} @tab @kbd{\a} @tab @code{07}, @code{BEL}
1797 @item @code{TARGET_CR} @tab @kbd{\r} @tab @code{0D}, @code{CR}
1798 @item @code{TARGET_ESC} @tab @kbd{\e}, @kbd{\E} @tab @code{1B}, @code{ESC}
1799 @item @code{TARGET_FF} @tab @kbd{\f} @tab @code{0C}, @code{FF}
1800 @item @code{TARGET_NEWLINE} @tab @kbd{\n} @tab @code{0A}, @code{LF}
1801 @item @code{TARGET_TAB} @tab @kbd{\t} @tab @code{09}, @code{HT}
1802 @item @code{TARGET_VT} @tab @kbd{\v} @tab @code{0B}, @code{VT}
1806 Note that the @kbd{\e} and @kbd{\E} escapes are GNU extensions, not
1807 part of the C standard.
1810 @section Register Usage
1811 @cindex register usage
1813 This section explains how to describe what registers the target machine
1814 has, and how (in general) they can be used.
1816 The description of which registers a specific instruction can use is
1817 done with register classes; see @ref{Register Classes}. For information
1818 on using registers to access a stack frame, see @ref{Frame Registers}.
1819 For passing values in registers, see @ref{Register Arguments}.
1820 For returning values in registers, see @ref{Scalar Return}.
1823 * Register Basics:: Number and kinds of registers.
1824 * Allocation Order:: Order in which registers are allocated.
1825 * Values in Registers:: What kinds of values each reg can hold.
1826 * Leaf Functions:: Renumbering registers for leaf functions.
1827 * Stack Registers:: Handling a register stack such as 80387.
1830 @node Register Basics
1831 @subsection Basic Characteristics of Registers
1833 @c prevent bad page break with this line
1834 Registers have various characteristics.
1836 @defmac FIRST_PSEUDO_REGISTER
1837 Number of hardware registers known to the compiler. They receive
1838 numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1839 pseudo register's number really is assigned the number
1840 @code{FIRST_PSEUDO_REGISTER}.
1843 @defmac FIXED_REGISTERS
1844 @cindex fixed register
1845 An initializer that says which registers are used for fixed purposes
1846 all throughout the compiled code and are therefore not available for
1847 general allocation. These would include the stack pointer, the frame
1848 pointer (except on machines where that can be used as a general
1849 register when no frame pointer is needed), the program counter on
1850 machines where that is considered one of the addressable registers,
1851 and any other numbered register with a standard use.
1853 This information is expressed as a sequence of numbers, separated by
1854 commas and surrounded by braces. The @var{n}th number is 1 if
1855 register @var{n} is fixed, 0 otherwise.
1857 The table initialized from this macro, and the table initialized by
1858 the following one, may be overridden at run time either automatically,
1859 by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
1860 the user with the command options @option{-ffixed-@var{reg}},
1861 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}.
1864 @defmac CALL_USED_REGISTERS
1865 @cindex call-used register
1866 @cindex call-clobbered register
1867 @cindex call-saved register
1868 Like @code{FIXED_REGISTERS} but has 1 for each register that is
1869 clobbered (in general) by function calls as well as for fixed
1870 registers. This macro therefore identifies the registers that are not
1871 available for general allocation of values that must live across
1874 If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1875 automatically saves it on function entry and restores it on function
1876 exit, if the register is used within the function.
1879 @defmac CALL_REALLY_USED_REGISTERS
1880 @cindex call-used register
1881 @cindex call-clobbered register
1882 @cindex call-saved register
1883 Like @code{CALL_USED_REGISTERS} except this macro doesn't require
1884 that the entire set of @code{FIXED_REGISTERS} be included.
1885 (@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}).
1886 This macro is optional. If not specified, it defaults to the value
1887 of @code{CALL_USED_REGISTERS}.
1890 @defmac HARD_REGNO_CALL_PART_CLOBBERED (@var{regno}, @var{mode})
1891 @cindex call-used register
1892 @cindex call-clobbered register
1893 @cindex call-saved register
1894 A C expression that is nonzero if it is not permissible to store a
1895 value of mode @var{mode} in hard register number @var{regno} across a
1896 call without some part of it being clobbered. For most machines this
1897 macro need not be defined. It is only required for machines that do not
1898 preserve the entire contents of a register across a call.
1902 @findex call_used_regs
1905 @findex reg_class_contents
1906 @defmac CONDITIONAL_REGISTER_USAGE
1907 Zero or more C statements that may conditionally modify five variables
1908 @code{fixed_regs}, @code{call_used_regs}, @code{global_regs},
1909 @code{reg_names}, and @code{reg_class_contents}, to take into account
1910 any dependence of these register sets on target flags. The first three
1911 of these are of type @code{char []} (interpreted as Boolean vectors).
1912 @code{global_regs} is a @code{const char *[]}, and
1913 @code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is
1914 called, @code{fixed_regs}, @code{call_used_regs},
1915 @code{reg_class_contents}, and @code{reg_names} have been initialized
1916 from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS},
1917 @code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively.
1918 @code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}},
1919 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}
1920 command options have been applied.
1922 You need not define this macro if it has no work to do.
1924 @cindex disabling certain registers
1925 @cindex controlling register usage
1926 If the usage of an entire class of registers depends on the target
1927 flags, you may indicate this to GCC by using this macro to modify
1928 @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
1929 registers in the classes which should not be used by GCC@. Also define
1930 the macro @code{REG_CLASS_FROM_LETTER} / @code{REG_CLASS_FROM_CONSTRAINT}
1931 to return @code{NO_REGS} if it
1932 is called with a letter for a class that shouldn't be used.
1934 (However, if this class is not included in @code{GENERAL_REGS} and all
1935 of the insn patterns whose constraints permit this class are
1936 controlled by target switches, then GCC will automatically avoid using
1937 these registers when the target switches are opposed to them.)
1940 @defmac NON_SAVING_SETJMP
1941 If this macro is defined and has a nonzero value, it means that
1942 @code{setjmp} and related functions fail to save the registers, or that
1943 @code{longjmp} fails to restore them. To compensate, the compiler
1944 avoids putting variables in registers in functions that use
1948 @defmac INCOMING_REGNO (@var{out})
1949 Define this macro if the target machine has register windows. This C
1950 expression returns the register number as seen by the called function
1951 corresponding to the register number @var{out} as seen by the calling
1952 function. Return @var{out} if register number @var{out} is not an
1956 @defmac OUTGOING_REGNO (@var{in})
1957 Define this macro if the target machine has register windows. This C
1958 expression returns the register number as seen by the calling function
1959 corresponding to the register number @var{in} as seen by the called
1960 function. Return @var{in} if register number @var{in} is not an inbound
1964 @defmac LOCAL_REGNO (@var{regno})
1965 Define this macro if the target machine has register windows. This C
1966 expression returns true if the register is call-saved but is in the
1967 register window. Unlike most call-saved registers, such registers
1968 need not be explicitly restored on function exit or during non-local
1973 If the program counter has a register number, define this as that
1974 register number. Otherwise, do not define it.
1977 @node Allocation Order
1978 @subsection Order of Allocation of Registers
1979 @cindex order of register allocation
1980 @cindex register allocation order
1982 @c prevent bad page break with this line
1983 Registers are allocated in order.
1985 @defmac REG_ALLOC_ORDER
1986 If defined, an initializer for a vector of integers, containing the
1987 numbers of hard registers in the order in which GCC should prefer
1988 to use them (from most preferred to least).
1990 If this macro is not defined, registers are used lowest numbered first
1991 (all else being equal).
1993 One use of this macro is on machines where the highest numbered
1994 registers must always be saved and the save-multiple-registers
1995 instruction supports only sequences of consecutive registers. On such
1996 machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists
1997 the highest numbered allocable register first.
2000 @defmac ORDER_REGS_FOR_LOCAL_ALLOC
2001 A C statement (sans semicolon) to choose the order in which to allocate
2002 hard registers for pseudo-registers local to a basic block.
2004 Store the desired register order in the array @code{reg_alloc_order}.
2005 Element 0 should be the register to allocate first; element 1, the next
2006 register; and so on.
2008 The macro body should not assume anything about the contents of
2009 @code{reg_alloc_order} before execution of the macro.
2011 On most machines, it is not necessary to define this macro.
2014 @node Values in Registers
2015 @subsection How Values Fit in Registers
2017 This section discusses the macros that describe which kinds of values
2018 (specifically, which machine modes) each register can hold, and how many
2019 consecutive registers are needed for a given mode.
2021 @defmac HARD_REGNO_NREGS (@var{regno}, @var{mode})
2022 A C expression for the number of consecutive hard registers, starting
2023 at register number @var{regno}, required to hold a value of mode
2026 On a machine where all registers are exactly one word, a suitable
2027 definition of this macro is
2030 #define HARD_REGNO_NREGS(REGNO, MODE) \
2031 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
2036 @defmac HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
2037 A C expression that is nonzero if it is permissible to store a value
2038 of mode @var{mode} in hard register number @var{regno} (or in several
2039 registers starting with that one). For a machine where all registers
2040 are equivalent, a suitable definition is
2043 #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
2046 You need not include code to check for the numbers of fixed registers,
2047 because the allocation mechanism considers them to be always occupied.
2049 @cindex register pairs
2050 On some machines, double-precision values must be kept in even/odd
2051 register pairs. You can implement that by defining this macro to reject
2052 odd register numbers for such modes.
2054 The minimum requirement for a mode to be OK in a register is that the
2055 @samp{mov@var{mode}} instruction pattern support moves between the
2056 register and other hard register in the same class and that moving a
2057 value into the register and back out not alter it.
2059 Since the same instruction used to move @code{word_mode} will work for
2060 all narrower integer modes, it is not necessary on any machine for
2061 @code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided
2062 you define patterns @samp{movhi}, etc., to take advantage of this. This
2063 is useful because of the interaction between @code{HARD_REGNO_MODE_OK}
2064 and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes
2067 Many machines have special registers for floating point arithmetic.
2068 Often people assume that floating point machine modes are allowed only
2069 in floating point registers. This is not true. Any registers that
2070 can hold integers can safely @emph{hold} a floating point machine
2071 mode, whether or not floating arithmetic can be done on it in those
2072 registers. Integer move instructions can be used to move the values.
2074 On some machines, though, the converse is true: fixed-point machine
2075 modes may not go in floating registers. This is true if the floating
2076 registers normalize any value stored in them, because storing a
2077 non-floating value there would garble it. In this case,
2078 @code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
2079 floating registers. But if the floating registers do not automatically
2080 normalize, if you can store any bit pattern in one and retrieve it
2081 unchanged without a trap, then any machine mode may go in a floating
2082 register, so you can define this macro to say so.
2084 The primary significance of special floating registers is rather that
2085 they are the registers acceptable in floating point arithmetic
2086 instructions. However, this is of no concern to
2087 @code{HARD_REGNO_MODE_OK}. You handle it by writing the proper
2088 constraints for those instructions.
2090 On some machines, the floating registers are especially slow to access,
2091 so that it is better to store a value in a stack frame than in such a
2092 register if floating point arithmetic is not being done. As long as the
2093 floating registers are not in class @code{GENERAL_REGS}, they will not
2094 be used unless some pattern's constraint asks for one.
2097 @defmac MODES_TIEABLE_P (@var{mode1}, @var{mode2})
2098 A C expression that is nonzero if a value of mode
2099 @var{mode1} is accessible in mode @var{mode2} without copying.
2101 If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
2102 @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for
2103 any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})}
2104 should be nonzero. If they differ for any @var{r}, you should define
2105 this macro to return zero unless some other mechanism ensures the
2106 accessibility of the value in a narrower mode.
2108 You should define this macro to return nonzero in as many cases as
2109 possible since doing so will allow GCC to perform better register
2113 @defmac AVOID_CCMODE_COPIES
2114 Define this macro if the compiler should avoid copies to/from @code{CCmode}
2115 registers. You should only define this macro if support for copying to/from
2116 @code{CCmode} is incomplete.
2119 @node Leaf Functions
2120 @subsection Handling Leaf Functions
2122 @cindex leaf functions
2123 @cindex functions, leaf
2124 On some machines, a leaf function (i.e., one which makes no calls) can run
2125 more efficiently if it does not make its own register window. Often this
2126 means it is required to receive its arguments in the registers where they
2127 are passed by the caller, instead of the registers where they would
2130 The special treatment for leaf functions generally applies only when
2131 other conditions are met; for example, often they may use only those
2132 registers for its own variables and temporaries. We use the term ``leaf
2133 function'' to mean a function that is suitable for this special
2134 handling, so that functions with no calls are not necessarily ``leaf
2137 GCC assigns register numbers before it knows whether the function is
2138 suitable for leaf function treatment. So it needs to renumber the
2139 registers in order to output a leaf function. The following macros
2142 @defmac LEAF_REGISTERS
2143 Name of a char vector, indexed by hard register number, which
2144 contains 1 for a register that is allowable in a candidate for leaf
2147 If leaf function treatment involves renumbering the registers, then the
2148 registers marked here should be the ones before renumbering---those that
2149 GCC would ordinarily allocate. The registers which will actually be
2150 used in the assembler code, after renumbering, should not be marked with 1
2153 Define this macro only if the target machine offers a way to optimize
2154 the treatment of leaf functions.
2157 @defmac LEAF_REG_REMAP (@var{regno})
2158 A C expression whose value is the register number to which @var{regno}
2159 should be renumbered, when a function is treated as a leaf function.
2161 If @var{regno} is a register number which should not appear in a leaf
2162 function before renumbering, then the expression should yield @minus{}1, which
2163 will cause the compiler to abort.
2165 Define this macro only if the target machine offers a way to optimize the
2166 treatment of leaf functions, and registers need to be renumbered to do
2170 @findex current_function_is_leaf
2171 @findex current_function_uses_only_leaf_regs
2172 @code{TARGET_ASM_FUNCTION_PROLOGUE} and
2173 @code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions
2174 specially. They can test the C variable @code{current_function_is_leaf}
2175 which is nonzero for leaf functions. @code{current_function_is_leaf} is
2176 set prior to local register allocation and is valid for the remaining
2177 compiler passes. They can also test the C variable
2178 @code{current_function_uses_only_leaf_regs} which is nonzero for leaf
2179 functions which only use leaf registers.
2180 @code{current_function_uses_only_leaf_regs} is valid after reload and is
2181 only useful if @code{LEAF_REGISTERS} is defined.
2182 @c changed this to fix overfull. ALSO: why the "it" at the beginning
2183 @c of the next paragraph?! --mew 2feb93
2185 @node Stack Registers
2186 @subsection Registers That Form a Stack
2188 There are special features to handle computers where some of the
2189 ``registers'' form a stack. Stack registers are normally written by
2190 pushing onto the stack, and are numbered relative to the top of the
2193 Currently, GCC can only handle one group of stack-like registers, and
2194 they must be consecutively numbered. Furthermore, the existing
2195 support for stack-like registers is specific to the 80387 floating
2196 point coprocessor. If you have a new architecture that uses
2197 stack-like registers, you will need to do substantial work on
2198 @file{reg-stack.c} and write your machine description to cooperate
2199 with it, as well as defining these macros.
2202 Define this if the machine has any stack-like registers.
2205 @defmac FIRST_STACK_REG
2206 The number of the first stack-like register. This one is the top
2210 @defmac LAST_STACK_REG
2211 The number of the last stack-like register. This one is the bottom of
2215 @node Register Classes
2216 @section Register Classes
2217 @cindex register class definitions
2218 @cindex class definitions, register
2220 On many machines, the numbered registers are not all equivalent.
2221 For example, certain registers may not be allowed for indexed addressing;
2222 certain registers may not be allowed in some instructions. These machine
2223 restrictions are described to the compiler using @dfn{register classes}.
2225 You define a number of register classes, giving each one a name and saying
2226 which of the registers belong to it. Then you can specify register classes
2227 that are allowed as operands to particular instruction patterns.
2231 In general, each register will belong to several classes. In fact, one
2232 class must be named @code{ALL_REGS} and contain all the registers. Another
2233 class must be named @code{NO_REGS} and contain no registers. Often the
2234 union of two classes will be another class; however, this is not required.
2236 @findex GENERAL_REGS
2237 One of the classes must be named @code{GENERAL_REGS}. There is nothing
2238 terribly special about the name, but the operand constraint letters
2239 @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
2240 the same as @code{ALL_REGS}, just define it as a macro which expands
2243 Order the classes so that if class @var{x} is contained in class @var{y}
2244 then @var{x} has a lower class number than @var{y}.
2246 The way classes other than @code{GENERAL_REGS} are specified in operand
2247 constraints is through machine-dependent operand constraint letters.
2248 You can define such letters to correspond to various classes, then use
2249 them in operand constraints.
2251 You should define a class for the union of two classes whenever some
2252 instruction allows both classes. For example, if an instruction allows
2253 either a floating point (coprocessor) register or a general register for a
2254 certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
2255 which includes both of them. Otherwise you will get suboptimal code.
2257 You must also specify certain redundant information about the register
2258 classes: for each class, which classes contain it and which ones are
2259 contained in it; for each pair of classes, the largest class contained
2262 When a value occupying several consecutive registers is expected in a
2263 certain class, all the registers used must belong to that class.
2264 Therefore, register classes cannot be used to enforce a requirement for
2265 a register pair to start with an even-numbered register. The way to
2266 specify this requirement is with @code{HARD_REGNO_MODE_OK}.
2268 Register classes used for input-operands of bitwise-and or shift
2269 instructions have a special requirement: each such class must have, for
2270 each fixed-point machine mode, a subclass whose registers can transfer that
2271 mode to or from memory. For example, on some machines, the operations for
2272 single-byte values (@code{QImode}) are limited to certain registers. When
2273 this is so, each register class that is used in a bitwise-and or shift
2274 instruction must have a subclass consisting of registers from which
2275 single-byte values can be loaded or stored. This is so that
2276 @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
2278 @deftp {Data type} {enum reg_class}
2279 An enumeral type that must be defined with all the register class names
2280 as enumeral values. @code{NO_REGS} must be first. @code{ALL_REGS}
2281 must be the last register class, followed by one more enumeral value,
2282 @code{LIM_REG_CLASSES}, which is not a register class but rather
2283 tells how many classes there are.
2285 Each register class has a number, which is the value of casting
2286 the class name to type @code{int}. The number serves as an index
2287 in many of the tables described below.
2290 @defmac N_REG_CLASSES
2291 The number of distinct register classes, defined as follows:
2294 #define N_REG_CLASSES (int) LIM_REG_CLASSES
2298 @defmac REG_CLASS_NAMES
2299 An initializer containing the names of the register classes as C string
2300 constants. These names are used in writing some of the debugging dumps.
2303 @defmac REG_CLASS_CONTENTS
2304 An initializer containing the contents of the register classes, as integers
2305 which are bit masks. The @var{n}th integer specifies the contents of class
2306 @var{n}. The way the integer @var{mask} is interpreted is that
2307 register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
2309 When the machine has more than 32 registers, an integer does not suffice.
2310 Then the integers are replaced by sub-initializers, braced groupings containing
2311 several integers. Each sub-initializer must be suitable as an initializer
2312 for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
2313 In this situation, the first integer in each sub-initializer corresponds to
2314 registers 0 through 31, the second integer to registers 32 through 63, and
2318 @defmac REGNO_REG_CLASS (@var{regno})
2319 A C expression whose value is a register class containing hard register
2320 @var{regno}. In general there is more than one such class; choose a class
2321 which is @dfn{minimal}, meaning that no smaller class also contains the
2325 @defmac BASE_REG_CLASS
2326 A macro whose definition is the name of the class to which a valid
2327 base register must belong. A base register is one used in an address
2328 which is the register value plus a displacement.
2331 @defmac MODE_BASE_REG_CLASS (@var{mode})
2332 This is a variation of the @code{BASE_REG_CLASS} macro which allows
2333 the selection of a base register in a mode dependent manner. If
2334 @var{mode} is VOIDmode then it should return the same value as
2335 @code{BASE_REG_CLASS}.
2338 @defmac INDEX_REG_CLASS
2339 A macro whose definition is the name of the class to which a valid
2340 index register must belong. An index register is one used in an
2341 address where its value is either multiplied by a scale factor or
2342 added to another register (as well as added to a displacement).
2345 @defmac CONSTRAINT_LEN (@var{char}, @var{str})
2346 For the constraint at the start of @var{str}, which starts with the letter
2347 @var{c}, return the length. This allows you to have register class /
2348 constant / extra constraints that are longer than a single letter;
2349 you don't need to define this macro if you can do with single-letter
2350 constraints only. The definition of this macro should use
2351 DEFAULT_CONSTRAINT_LEN for all the characters that you don't want
2352 to handle specially.
2353 There are some sanity checks in genoutput.c that check the constraint lengths
2354 for the md file, so you can also use this macro to help you while you are
2355 transitioning from a byzantine single-letter-constraint scheme: when you
2356 return a negative length for a constraint you want to re-use, genoutput
2357 will complain about every instance where it is used in the md file.
2360 @defmac REG_CLASS_FROM_LETTER (@var{char})
2361 A C expression which defines the machine-dependent operand constraint
2362 letters for register classes. If @var{char} is such a letter, the
2363 value should be the register class corresponding to it. Otherwise,
2364 the value should be @code{NO_REGS}. The register letter @samp{r},
2365 corresponding to class @code{GENERAL_REGS}, will not be passed
2366 to this macro; you do not need to handle it.
2369 @defmac REG_CLASS_FROM_CONSTRAINT (@var{char}, @var{str})
2370 Like @code{REG_CLASS_FROM_LETTER}, but you also get the constraint string
2371 passed in @var{str}, so that you can use suffixes to distinguish between
2375 @defmac REGNO_OK_FOR_BASE_P (@var{num})
2376 A C expression which is nonzero if register number @var{num} is
2377 suitable for use as a base register in operand addresses. It may be
2378 either a suitable hard register or a pseudo register that has been
2379 allocated such a hard register.
2382 @defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode})
2383 A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that
2384 that expression may examine the mode of the memory reference in
2385 @var{mode}. You should define this macro if the mode of the memory
2386 reference affects whether a register may be used as a base register. If
2387 you define this macro, the compiler will use it instead of
2388 @code{REGNO_OK_FOR_BASE_P}.
2391 @defmac REGNO_OK_FOR_INDEX_P (@var{num})
2392 A C expression which is nonzero if register number @var{num} is
2393 suitable for use as an index register in operand addresses. It may be
2394 either a suitable hard register or a pseudo register that has been
2395 allocated such a hard register.
2397 The difference between an index register and a base register is that
2398 the index register may be scaled. If an address involves the sum of
2399 two registers, neither one of them scaled, then either one may be
2400 labeled the ``base'' and the other the ``index''; but whichever
2401 labeling is used must fit the machine's constraints of which registers
2402 may serve in each capacity. The compiler will try both labelings,
2403 looking for one that is valid, and will reload one or both registers
2404 only if neither labeling works.
2407 @defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
2408 A C expression that places additional restrictions on the register class
2409 to use when it is necessary to copy value @var{x} into a register in class
2410 @var{class}. The value is a register class; perhaps @var{class}, or perhaps
2411 another, smaller class. On many machines, the following definition is
2415 #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
2418 Sometimes returning a more restrictive class makes better code. For
2419 example, on the 68000, when @var{x} is an integer constant that is in range
2420 for a @samp{moveq} instruction, the value of this macro is always
2421 @code{DATA_REGS} as long as @var{class} includes the data registers.
2422 Requiring a data register guarantees that a @samp{moveq} will be used.
2424 If @var{x} is a @code{const_double}, by returning @code{NO_REGS}
2425 you can force @var{x} into a memory constant. This is useful on
2426 certain machines where immediate floating values cannot be loaded into
2427 certain kinds of registers.
2430 @defmac PREFERRED_OUTPUT_RELOAD_CLASS (@var{x}, @var{class})
2431 Like @code{PREFERRED_RELOAD_CLASS}, but for output reloads instead of
2432 input reloads. If you don't define this macro, the default is to use
2433 @var{class}, unchanged.
2436 @defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class})
2437 A C expression that places additional restrictions on the register class
2438 to use when it is necessary to be able to hold a value of mode
2439 @var{mode} in a reload register for which class @var{class} would
2442 Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when
2443 there are certain modes that simply can't go in certain reload classes.
2445 The value is a register class; perhaps @var{class}, or perhaps another,
2448 Don't define this macro unless the target machine has limitations which
2449 require the macro to do something nontrivial.
2452 @defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2453 @defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2454 @defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2455 Many machines have some registers that cannot be copied directly to or
2456 from memory or even from other types of registers. An example is the
2457 @samp{MQ} register, which on most machines, can only be copied to or
2458 from general registers, but not memory. Some machines allow copying all
2459 registers to and from memory, but require a scratch register for stores
2460 to some memory locations (e.g., those with symbolic address on the RT,
2461 and those with certain symbolic address on the SPARC when compiling
2462 PIC)@. In some cases, both an intermediate and a scratch register are
2465 You should define these macros to indicate to the reload phase that it may
2466 need to allocate at least one register for a reload in addition to the
2467 register to contain the data. Specifically, if copying @var{x} to a
2468 register @var{class} in @var{mode} requires an intermediate register,
2469 you should define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the
2470 largest register class all of whose registers can be used as
2471 intermediate registers or scratch registers.
2473 If copying a register @var{class} in @var{mode} to @var{x} requires an
2474 intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS}
2475 should be defined to return the largest register class required. If the
2476 requirements for input and output reloads are the same, the macro
2477 @code{SECONDARY_RELOAD_CLASS} should be used instead of defining both
2480 The values returned by these macros are often @code{GENERAL_REGS}.
2481 Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x}
2482 can be directly copied to or from a register of @var{class} in
2483 @var{mode} without requiring a scratch register. Do not define this
2484 macro if it would always return @code{NO_REGS}.
2486 If a scratch register is required (either with or without an
2487 intermediate register), you should define patterns for
2488 @samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required
2489 (@pxref{Standard Names}. These patterns, which will normally be
2490 implemented with a @code{define_expand}, should be similar to the
2491 @samp{mov@var{m}} patterns, except that operand 2 is the scratch
2494 Define constraints for the reload register and scratch register that
2495 contain a single register class. If the original reload register (whose
2496 class is @var{class}) can meet the constraint given in the pattern, the
2497 value returned by these macros is used for the class of the scratch
2498 register. Otherwise, two additional reload registers are required.
2499 Their classes are obtained from the constraints in the insn pattern.
2501 @var{x} might be a pseudo-register or a @code{subreg} of a
2502 pseudo-register, which could either be in a hard register or in memory.
2503 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2504 in memory and the hard register number if it is in a register.
2506 These macros should not be used in the case where a particular class of
2507 registers can only be copied to memory and not to another class of
2508 registers. In that case, secondary reload registers are not needed and
2509 would not be helpful. Instead, a stack location must be used to perform
2510 the copy and the @code{mov@var{m}} pattern should use memory as an
2511 intermediate storage. This case often occurs between floating-point and
2515 @defmac SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m})
2516 Certain machines have the property that some registers cannot be copied
2517 to some other registers without using memory. Define this macro on
2518 those machines to be a C expression that is nonzero if objects of mode
2519 @var{m} in registers of @var{class1} can only be copied to registers of
2520 class @var{class2} by storing a register of @var{class1} into memory
2521 and loading that memory location into a register of @var{class2}.
2523 Do not define this macro if its value would always be zero.
2526 @defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode})
2527 Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler
2528 allocates a stack slot for a memory location needed for register copies.
2529 If this macro is defined, the compiler instead uses the memory location
2530 defined by this macro.
2532 Do not define this macro if you do not define
2533 @code{SECONDARY_MEMORY_NEEDED}.
2536 @defmac SECONDARY_MEMORY_NEEDED_MODE (@var{mode})
2537 When the compiler needs a secondary memory location to copy between two
2538 registers of mode @var{mode}, it normally allocates sufficient memory to
2539 hold a quantity of @code{BITS_PER_WORD} bits and performs the store and
2540 load operations in a mode that many bits wide and whose class is the
2541 same as that of @var{mode}.
2543 This is right thing to do on most machines because it ensures that all
2544 bits of the register are copied and prevents accesses to the registers
2545 in a narrower mode, which some machines prohibit for floating-point
2548 However, this default behavior is not correct on some machines, such as
2549 the DEC Alpha, that store short integers in floating-point registers
2550 differently than in integer registers. On those machines, the default
2551 widening will not work correctly and you must define this macro to
2552 suppress that widening in some cases. See the file @file{alpha.h} for
2555 Do not define this macro if you do not define
2556 @code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that
2557 is @code{BITS_PER_WORD} bits wide is correct for your machine.
2560 @defmac SMALL_REGISTER_CLASSES
2561 On some machines, it is risky to let hard registers live across arbitrary
2562 insns. Typically, these machines have instructions that require values
2563 to be in specific registers (like an accumulator), and reload will fail
2564 if the required hard register is used for another purpose across such an
2567 Define @code{SMALL_REGISTER_CLASSES} to be an expression with a nonzero
2568 value on these machines. When this macro has a nonzero value, the
2569 compiler will try to minimize the lifetime of hard registers.
2571 It is always safe to define this macro with a nonzero value, but if you
2572 unnecessarily define it, you will reduce the amount of optimizations
2573 that can be performed in some cases. If you do not define this macro
2574 with a nonzero value when it is required, the compiler will run out of
2575 spill registers and print a fatal error message. For most machines, you
2576 should not define this macro at all.
2579 @defmac CLASS_LIKELY_SPILLED_P (@var{class})
2580 A C expression whose value is nonzero if pseudos that have been assigned
2581 to registers of class @var{class} would likely be spilled because
2582 registers of @var{class} are needed for spill registers.
2584 The default value of this macro returns 1 if @var{class} has exactly one
2585 register and zero otherwise. On most machines, this default should be
2586 used. Only define this macro to some other expression if pseudos
2587 allocated by @file{local-alloc.c} end up in memory because their hard
2588 registers were needed for spill registers. If this macro returns nonzero
2589 for those classes, those pseudos will only be allocated by
2590 @file{global.c}, which knows how to reallocate the pseudo to another
2591 register. If there would not be another register available for
2592 reallocation, you should not change the definition of this macro since
2593 the only effect of such a definition would be to slow down register
2597 @defmac CLASS_MAX_NREGS (@var{class}, @var{mode})
2598 A C expression for the maximum number of consecutive registers
2599 of class @var{class} needed to hold a value of mode @var{mode}.
2601 This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact,
2602 the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
2603 should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno},
2604 @var{mode})} for all @var{regno} values in the class @var{class}.
2606 This macro helps control the handling of multiple-word values
2610 @defmac CANNOT_CHANGE_MODE_CLASS (@var{from}, @var{to}, @var{class})
2611 If defined, a C expression that returns nonzero for a @var{class} for which
2612 a change from mode @var{from} to mode @var{to} is invalid.
2614 For the example, loading 32-bit integer or floating-point objects into
2615 floating-point registers on the Alpha extends them to 64 bits.
2616 Therefore loading a 64-bit object and then storing it as a 32-bit object
2617 does not store the low-order 32 bits, as would be the case for a normal
2618 register. Therefore, @file{alpha.h} defines @code{CANNOT_CHANGE_MODE_CLASS}
2622 #define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \
2623 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \
2624 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0)
2628 Three other special macros describe which operands fit which constraint
2631 @defmac CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
2632 A C expression that defines the machine-dependent operand constraint
2633 letters (@samp{I}, @samp{J}, @samp{K}, @dots{} @samp{P}) that specify
2634 particular ranges of integer values. If @var{c} is one of those
2635 letters, the expression should check that @var{value}, an integer, is in
2636 the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
2637 not one of those letters, the value should be 0 regardless of
2641 @defmac CONST_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2642 Like @code{CONST_OK_FOR_LETTER_P}, but you also get the constraint
2643 string passed in @var{str}, so that you can use suffixes to distinguish
2644 between different variants.
2647 @defmac CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
2648 A C expression that defines the machine-dependent operand constraint
2649 letters that specify particular ranges of @code{const_double} values
2650 (@samp{G} or @samp{H}).
2652 If @var{c} is one of those letters, the expression should check that
2653 @var{value}, an RTX of code @code{const_double}, is in the appropriate
2654 range and return 1 if so, 0 otherwise. If @var{c} is not one of those
2655 letters, the value should be 0 regardless of @var{value}.
2657 @code{const_double} is used for all floating-point constants and for
2658 @code{DImode} fixed-point constants. A given letter can accept either
2659 or both kinds of values. It can use @code{GET_MODE} to distinguish
2660 between these kinds.
2663 @defmac CONST_DOUBLE_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2664 Like @code{CONST_DOUBLE_OK_FOR_LETTER_P}, but you also get the constraint
2665 string passed in @var{str}, so that you can use suffixes to distinguish
2666 between different variants.
2669 @defmac EXTRA_CONSTRAINT (@var{value}, @var{c})
2670 A C expression that defines the optional machine-dependent constraint
2671 letters that can be used to segregate specific types of operands, usually
2672 memory references, for the target machine. Any letter that is not
2673 elsewhere defined and not matched by @code{REG_CLASS_FROM_LETTER} /
2674 @code{REG_CLASS_FROM_CONSTRAINT}
2675 may be used. Normally this macro will not be defined.
2677 If it is required for a particular target machine, it should return 1
2678 if @var{value} corresponds to the operand type represented by the
2679 constraint letter @var{c}. If @var{c} is not defined as an extra
2680 constraint, the value returned should be 0 regardless of @var{value}.
2682 For example, on the ROMP, load instructions cannot have their output
2683 in r0 if the memory reference contains a symbolic address. Constraint
2684 letter @samp{Q} is defined as representing a memory address that does
2685 @emph{not} contain a symbolic address. An alternative is specified with
2686 a @samp{Q} constraint on the input and @samp{r} on the output. The next
2687 alternative specifies @samp{m} on the input and a register class that
2688 does not include r0 on the output.
2691 @defmac EXTRA_CONSTRAINT_STR (@var{value}, @var{c}, @var{str})
2692 Like @code{EXTRA_CONSTRAINT}, but you also get the constraint string passed
2693 in @var{str}, so that you can use suffixes to distinguish between different
2697 @defmac EXTRA_MEMORY_CONSTRAINT (@var{c}, @var{str})
2698 A C expression that defines the optional machine-dependent constraint
2699 letters, amongst those accepted by @code{EXTRA_CONSTRAINT}, that should
2700 be treated like memory constraints by the reload pass.
2702 It should return 1 if the operand type represented by the constraint
2703 at the start of @var{str}, the first letter of which is the letter @var{c},
2704 comprises a subset of all memory references including
2705 all those whose address is simply a base register. This allows the reload
2706 pass to reload an operand, if it does not directly correspond to the operand
2707 type of @var{c}, by copying its address into a base register.
2709 For example, on the S/390, some instructions do not accept arbitrary
2710 memory references, but only those that do not make use of an index
2711 register. The constraint letter @samp{Q} is defined via
2712 @code{EXTRA_CONSTRAINT} as representing a memory address of this type.
2713 If the letter @samp{Q} is marked as @code{EXTRA_MEMORY_CONSTRAINT},
2714 a @samp{Q} constraint can handle any memory operand, because the
2715 reload pass knows it can be reloaded by copying the memory address
2716 into a base register if required. This is analogous to the way
2717 a @samp{o} constraint can handle any memory operand.
2720 @defmac EXTRA_ADDRESS_CONSTRAINT (@var{c}, @var{str})
2721 A C expression that defines the optional machine-dependent constraint
2722 letters, amongst those accepted by @code{EXTRA_CONSTRAINT} /
2723 @code{EXTRA_CONSTRAINT_STR}, that should
2724 be treated like address constraints by the reload pass.
2726 It should return 1 if the operand type represented by the constraint
2727 at the start of @var{str}, which starts with the letter @var{c}, comprises
2728 a subset of all memory addresses including
2729 all those that consist of just a base register. This allows the reload
2730 pass to reload an operand, if it does not directly correspond to the operand
2731 type of @var{str}, by copying it into a base register.
2733 Any constraint marked as @code{EXTRA_ADDRESS_CONSTRAINT} can only
2734 be used with the @code{address_operand} predicate. It is treated
2735 analogously to the @samp{p} constraint.
2738 @node Stack and Calling
2739 @section Stack Layout and Calling Conventions
2740 @cindex calling conventions
2742 @c prevent bad page break with this line
2743 This describes the stack layout and calling conventions.
2747 * Exception Handling::
2752 * Register Arguments::
2754 * Aggregate Return::
2762 @subsection Basic Stack Layout
2763 @cindex stack frame layout
2764 @cindex frame layout
2766 @c prevent bad page break with this line
2767 Here is the basic stack layout.
2769 @defmac STACK_GROWS_DOWNWARD
2770 Define this macro if pushing a word onto the stack moves the stack
2771 pointer to a smaller address.
2773 When we say, ``define this macro if @dots{},'' it means that the
2774 compiler checks this macro only with @code{#ifdef} so the precise
2775 definition used does not matter.
2778 @defmac STACK_PUSH_CODE
2779 This macro defines the operation used when something is pushed
2780 on the stack. In RTL, a push operation will be
2781 @code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})}
2783 The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC},
2784 and @code{POST_INC}. Which of these is correct depends on
2785 the stack direction and on whether the stack pointer points
2786 to the last item on the stack or whether it points to the
2787 space for the next item on the stack.
2789 The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is
2790 defined, which is almost always right, and @code{PRE_INC} otherwise,
2791 which is often wrong.
2794 @defmac FRAME_GROWS_DOWNWARD
2795 Define this macro if the addresses of local variable slots are at negative
2796 offsets from the frame pointer.
2799 @defmac ARGS_GROW_DOWNWARD
2800 Define this macro if successive arguments to a function occupy decreasing
2801 addresses on the stack.
2804 @defmac STARTING_FRAME_OFFSET
2805 Offset from the frame pointer to the first local variable slot to be allocated.
2807 If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by
2808 subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}.
2809 Otherwise, it is found by adding the length of the first slot to the
2810 value @code{STARTING_FRAME_OFFSET}.
2811 @c i'm not sure if the above is still correct.. had to change it to get
2812 @c rid of an overfull. --mew 2feb93
2815 @defmac STACK_ALIGNMENT_NEEDED
2816 Define to zero to disable final alignment of the stack during reload.
2817 The nonzero default for this macro is suitable for most ports.
2819 On ports where @code{STARTING_FRAME_OFFSET} is nonzero or where there
2820 is a register save block following the local block that doesn't require
2821 alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable
2822 stack alignment and do it in the backend.
2825 @defmac STACK_POINTER_OFFSET
2826 Offset from the stack pointer register to the first location at which
2827 outgoing arguments are placed. If not specified, the default value of
2828 zero is used. This is the proper value for most machines.
2830 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
2831 the first location at which outgoing arguments are placed.
2834 @defmac FIRST_PARM_OFFSET (@var{fundecl})
2835 Offset from the argument pointer register to the first argument's
2836 address. On some machines it may depend on the data type of the
2839 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
2840 the first argument's address.
2843 @defmac STACK_DYNAMIC_OFFSET (@var{fundecl})
2844 Offset from the stack pointer register to an item dynamically allocated
2845 on the stack, e.g., by @code{alloca}.
2847 The default value for this macro is @code{STACK_POINTER_OFFSET} plus the
2848 length of the outgoing arguments. The default is correct for most
2849 machines. See @file{function.c} for details.
2852 @defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr})
2853 A C expression whose value is RTL representing the address in a stack
2854 frame where the pointer to the caller's frame is stored. Assume that
2855 @var{frameaddr} is an RTL expression for the address of the stack frame
2858 If you don't define this macro, the default is to return the value
2859 of @var{frameaddr}---that is, the stack frame address is also the
2860 address of the stack word that points to the previous frame.
2863 @defmac SETUP_FRAME_ADDRESSES
2864 If defined, a C expression that produces the machine-specific code to
2865 setup the stack so that arbitrary frames can be accessed. For example,
2866 on the SPARC, we must flush all of the register windows to the stack
2867 before we can access arbitrary stack frames. You will seldom need to
2871 @defmac BUILTIN_SETJMP_FRAME_VALUE
2872 If defined, a C expression that contains an rtx that is used to store
2873 the address of the current frame into the built in @code{setjmp} buffer.
2874 The default value, @code{virtual_stack_vars_rtx}, is correct for most
2875 machines. One reason you may need to define this macro is if
2876 @code{hard_frame_pointer_rtx} is the appropriate value on your machine.
2879 @defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr})
2880 A C expression whose value is RTL representing the value of the return
2881 address for the frame @var{count} steps up from the current frame, after
2882 the prologue. @var{frameaddr} is the frame pointer of the @var{count}
2883 frame, or the frame pointer of the @var{count} @minus{} 1 frame if
2884 @code{RETURN_ADDR_IN_PREVIOUS_FRAME} is defined.
2886 The value of the expression must always be the correct address when
2887 @var{count} is zero, but may be @code{NULL_RTX} if there is not way to
2888 determine the return address of other frames.
2891 @defmac RETURN_ADDR_IN_PREVIOUS_FRAME
2892 Define this if the return address of a particular stack frame is accessed
2893 from the frame pointer of the previous stack frame.
2896 @defmac INCOMING_RETURN_ADDR_RTX
2897 A C expression whose value is RTL representing the location of the
2898 incoming return address at the beginning of any function, before the
2899 prologue. This RTL is either a @code{REG}, indicating that the return
2900 value is saved in @samp{REG}, or a @code{MEM} representing a location in
2903 You only need to define this macro if you want to support call frame
2904 debugging information like that provided by DWARF 2.
2906 If this RTL is a @code{REG}, you should also define
2907 @code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}.
2910 @defmac INCOMING_FRAME_SP_OFFSET
2911 A C expression whose value is an integer giving the offset, in bytes,
2912 from the value of the stack pointer register to the top of the stack
2913 frame at the beginning of any function, before the prologue. The top of
2914 the frame is defined to be the value of the stack pointer in the
2915 previous frame, just before the call instruction.
2917 You only need to define this macro if you want to support call frame
2918 debugging information like that provided by DWARF 2.
2921 @defmac ARG_POINTER_CFA_OFFSET (@var{fundecl})
2922 A C expression whose value is an integer giving the offset, in bytes,
2923 from the argument pointer to the canonical frame address (cfa). The
2924 final value should coincide with that calculated by
2925 @code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable
2926 during virtual register instantiation.
2928 The default value for this macro is @code{FIRST_PARM_OFFSET (fundecl)},
2929 which is correct for most machines; in general, the arguments are found
2930 immediately before the stack frame. Note that this is not the case on
2931 some targets that save registers into the caller's frame, such as SPARC
2932 and rs6000, and so such targets need to define this macro.
2934 You only need to define this macro if the default is incorrect, and you
2935 want to support call frame debugging information like that provided by
2940 Define this macro if the stack size for the target is very small. This
2941 has the effect of disabling gcc's built-in @samp{alloca}, though
2942 @samp{__builtin_alloca} is not affected.
2945 @node Exception Handling
2946 @subsection Exception Handling Support
2947 @cindex exception handling
2949 @defmac EH_RETURN_DATA_REGNO (@var{N})
2950 A C expression whose value is the @var{N}th register number used for
2951 data by exception handlers, or @code{INVALID_REGNUM} if fewer than
2952 @var{N} registers are usable.
2954 The exception handling library routines communicate with the exception
2955 handlers via a set of agreed upon registers. Ideally these registers
2956 should be call-clobbered; it is possible to use call-saved registers,
2957 but may negatively impact code size. The target must support at least
2958 2 data registers, but should define 4 if there are enough free registers.
2960 You must define this macro if you want to support call frame exception
2961 handling like that provided by DWARF 2.
2964 @defmac EH_RETURN_STACKADJ_RTX
2965 A C expression whose value is RTL representing a location in which
2966 to store a stack adjustment to be applied before function return.
2967 This is used to unwind the stack to an exception handler's call frame.
2968 It will be assigned zero on code paths that return normally.
2970 Typically this is a call-clobbered hard register that is otherwise
2971 untouched by the epilogue, but could also be a stack slot.
2973 Do not define this macro if the stack pointer is saved and restored
2974 by the regular prolog and epilog code in the call frame itself; in
2975 this case, the exception handling library routines will update the
2976 stack location to be restored in place. Otherwise, you must define
2977 this macro if you want to support call frame exception handling like
2978 that provided by DWARF 2.
2981 @defmac EH_RETURN_HANDLER_RTX
2982 A C expression whose value is RTL representing a location in which
2983 to store the address of an exception handler to which we should
2984 return. It will not be assigned on code paths that return normally.
2986 Typically this is the location in the call frame at which the normal
2987 return address is stored. For targets that return by popping an
2988 address off the stack, this might be a memory address just below
2989 the @emph{target} call frame rather than inside the current call
2990 frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already
2991 been assigned, so it may be used to calculate the location of the
2994 Some targets have more complex requirements than storing to an
2995 address calculable during initial code generation. In that case
2996 the @code{eh_return} instruction pattern should be used instead.
2998 If you want to support call frame exception handling, you must
2999 define either this macro or the @code{eh_return} instruction pattern.
3002 @defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global})
3003 This macro chooses the encoding of pointers embedded in the exception
3004 handling sections. If at all possible, this should be defined such
3005 that the exception handling section will not require dynamic relocations,
3006 and so may be read-only.
3008 @var{code} is 0 for data, 1 for code labels, 2 for function pointers.
3009 @var{global} is true if the symbol may be affected by dynamic relocations.
3010 The macro should return a combination of the @code{DW_EH_PE_*} defines
3011 as found in @file{dwarf2.h}.
3013 If this macro is not defined, pointers will not be encoded but
3014 represented directly.
3017 @defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done})
3018 This macro allows the target to emit whatever special magic is required
3019 to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}.
3020 Generic code takes care of pc-relative and indirect encodings; this must
3021 be defined if the target uses text-relative or data-relative encodings.
3023 This is a C statement that branches to @var{done} if the format was
3024 handled. @var{encoding} is the format chosen, @var{size} is the number
3025 of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF}
3029 @defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs}, @var{success})
3030 This macro allows the target to add cpu and operating system specific
3031 code to the call-frame unwinder for use when there is no unwind data
3032 available. The most common reason to implement this macro is to unwind
3033 through signal frames.
3035 This macro is called from @code{uw_frame_state_for} in @file{unwind-dw2.c}
3036 and @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context};
3037 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra}
3038 for the address of the code being executed and @code{context->cfa} for
3039 the stack pointer value. If the frame can be decoded, the register save
3040 addresses should be updated in @var{fs} and the macro should branch to
3041 @var{success}. If the frame cannot be decoded, the macro should do
3044 For proper signal handling in Java this macro is accompanied by
3045 @code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers.
3048 @node Stack Checking
3049 @subsection Specifying How Stack Checking is Done
3051 GCC will check that stack references are within the boundaries of
3052 the stack, if the @option{-fstack-check} is specified, in one of three ways:
3056 If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC
3057 will assume that you have arranged for stack checking to be done at
3058 appropriate places in the configuration files, e.g., in
3059 @code{TARGET_ASM_FUNCTION_PROLOGUE}. GCC will do not other special
3063 If @code{STACK_CHECK_BUILTIN} is zero and you defined a named pattern
3064 called @code{check_stack} in your @file{md} file, GCC will call that
3065 pattern with one argument which is the address to compare the stack
3066 value against. You must arrange for this pattern to report an error if
3067 the stack pointer is out of range.
3070 If neither of the above are true, GCC will generate code to periodically
3071 ``probe'' the stack pointer using the values of the macros defined below.
3074 Normally, you will use the default values of these macros, so GCC
3075 will use the third approach.
3077 @defmac STACK_CHECK_BUILTIN
3078 A nonzero value if stack checking is done by the configuration files in a
3079 machine-dependent manner. You should define this macro if stack checking
3080 is require by the ABI of your machine or if you would like to have to stack
3081 checking in some more efficient way than GCC's portable approach.
3082 The default value of this macro is zero.
3085 @defmac STACK_CHECK_PROBE_INTERVAL
3086 An integer representing the interval at which GCC must generate stack
3087 probe instructions. You will normally define this macro to be no larger
3088 than the size of the ``guard pages'' at the end of a stack area. The
3089 default value of 4096 is suitable for most systems.
3092 @defmac STACK_CHECK_PROBE_LOAD
3093 A integer which is nonzero if GCC should perform the stack probe
3094 as a load instruction and zero if GCC should use a store instruction.
3095 The default is zero, which is the most efficient choice on most systems.
3098 @defmac STACK_CHECK_PROTECT
3099 The number of bytes of stack needed to recover from a stack overflow,
3100 for languages where such a recovery is supported. The default value of
3101 75 words should be adequate for most machines.
3104 @defmac STACK_CHECK_MAX_FRAME_SIZE
3105 The maximum size of a stack frame, in bytes. GCC will generate probe
3106 instructions in non-leaf functions to ensure at least this many bytes of
3107 stack are available. If a stack frame is larger than this size, stack
3108 checking will not be reliable and GCC will issue a warning. The
3109 default is chosen so that GCC only generates one instruction on most
3110 systems. You should normally not change the default value of this macro.
3113 @defmac STACK_CHECK_FIXED_FRAME_SIZE
3114 GCC uses this value to generate the above warning message. It
3115 represents the amount of fixed frame used by a function, not including
3116 space for any callee-saved registers, temporaries and user variables.
3117 You need only specify an upper bound for this amount and will normally
3118 use the default of four words.
3121 @defmac STACK_CHECK_MAX_VAR_SIZE
3122 The maximum size, in bytes, of an object that GCC will place in the
3123 fixed area of the stack frame when the user specifies
3124 @option{-fstack-check}.
3125 GCC computed the default from the values of the above macros and you will
3126 normally not need to override that default.
3130 @node Frame Registers
3131 @subsection Registers That Address the Stack Frame
3133 @c prevent bad page break with this line
3134 This discusses registers that address the stack frame.
3136 @defmac STACK_POINTER_REGNUM
3137 The register number of the stack pointer register, which must also be a
3138 fixed register according to @code{FIXED_REGISTERS}. On most machines,
3139 the hardware determines which register this is.
3142 @defmac FRAME_POINTER_REGNUM
3143 The register number of the frame pointer register, which is used to
3144 access automatic variables in the stack frame. On some machines, the
3145 hardware determines which register this is. On other machines, you can
3146 choose any register you wish for this purpose.
3149 @defmac HARD_FRAME_POINTER_REGNUM
3150 On some machines the offset between the frame pointer and starting
3151 offset of the automatic variables is not known until after register
3152 allocation has been done (for example, because the saved registers are
3153 between these two locations). On those machines, define
3154 @code{FRAME_POINTER_REGNUM} the number of a special, fixed register to
3155 be used internally until the offset is known, and define
3156 @code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number
3157 used for the frame pointer.
3159 You should define this macro only in the very rare circumstances when it
3160 is not possible to calculate the offset between the frame pointer and
3161 the automatic variables until after register allocation has been
3162 completed. When this macro is defined, you must also indicate in your
3163 definition of @code{ELIMINABLE_REGS} how to eliminate
3164 @code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM}
3165 or @code{STACK_POINTER_REGNUM}.
3167 Do not define this macro if it would be the same as
3168 @code{FRAME_POINTER_REGNUM}.
3171 @defmac ARG_POINTER_REGNUM
3172 The register number of the arg pointer register, which is used to access
3173 the function's argument list. On some machines, this is the same as the
3174 frame pointer register. On some machines, the hardware determines which
3175 register this is. On other machines, you can choose any register you
3176 wish for this purpose. If this is not the same register as the frame
3177 pointer register, then you must mark it as a fixed register according to
3178 @code{FIXED_REGISTERS}, or arrange to be able to eliminate it
3179 (@pxref{Elimination}).
3182 @defmac RETURN_ADDRESS_POINTER_REGNUM
3183 The register number of the return address pointer register, which is used to
3184 access the current function's return address from the stack. On some
3185 machines, the return address is not at a fixed offset from the frame
3186 pointer or stack pointer or argument pointer. This register can be defined
3187 to point to the return address on the stack, and then be converted by
3188 @code{ELIMINABLE_REGS} into either the frame pointer or stack pointer.
3190 Do not define this macro unless there is no other way to get the return
3191 address from the stack.
3194 @defmac STATIC_CHAIN_REGNUM
3195 @defmacx STATIC_CHAIN_INCOMING_REGNUM
3196 Register numbers used for passing a function's static chain pointer. If
3197 register windows are used, the register number as seen by the called
3198 function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register
3199 number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If
3200 these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need
3203 The static chain register need not be a fixed register.
3205 If the static chain is passed in memory, these macros should not be
3206 defined; instead, the next two macros should be defined.
3209 @defmac STATIC_CHAIN
3210 @defmacx STATIC_CHAIN_INCOMING
3211 If the static chain is passed in memory, these macros provide rtx giving
3212 @code{mem} expressions that denote where they are stored.
3213 @code{STATIC_CHAIN} and @code{STATIC_CHAIN_INCOMING} give the locations
3214 as seen by the calling and called functions, respectively. Often the former
3215 will be at an offset from the stack pointer and the latter at an offset from
3218 @findex stack_pointer_rtx
3219 @findex frame_pointer_rtx
3220 @findex arg_pointer_rtx
3221 The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and
3222 @code{arg_pointer_rtx} will have been initialized prior to the use of these
3223 macros and should be used to refer to those items.
3225 If the static chain is passed in a register, the two previous macros should
3229 @defmac DWARF_FRAME_REGISTERS
3230 This macro specifies the maximum number of hard registers that can be
3231 saved in a call frame. This is used to size data structures used in
3232 DWARF2 exception handling.
3234 Prior to GCC 3.0, this macro was needed in order to establish a stable
3235 exception handling ABI in the face of adding new hard registers for ISA
3236 extensions. In GCC 3.0 and later, the EH ABI is insulated from changes
3237 in the number of hard registers. Nevertheless, this macro can still be
3238 used to reduce the runtime memory requirements of the exception handling
3239 routines, which can be substantial if the ISA contains a lot of
3240 registers that are not call-saved.
3242 If this macro is not defined, it defaults to
3243 @code{FIRST_PSEUDO_REGISTER}.
3246 @defmac PRE_GCC3_DWARF_FRAME_REGISTERS
3248 This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided
3249 for backward compatibility in pre GCC 3.0 compiled code.
3251 If this macro is not defined, it defaults to
3252 @code{DWARF_FRAME_REGISTERS}.
3255 @defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno})
3257 Define this macro if the target's representation for dwarf registers
3258 is different than the internal representation for unwind column.
3259 Given a dwarf register, this macro should return the internal unwind
3260 column number to use instead.
3262 See the PowerPC's SPE target for an example.
3266 @subsection Eliminating Frame Pointer and Arg Pointer
3268 @c prevent bad page break with this line
3269 This is about eliminating the frame pointer and arg pointer.
3271 @defmac FRAME_POINTER_REQUIRED
3272 A C expression which is nonzero if a function must have and use a frame
3273 pointer. This expression is evaluated in the reload pass. If its value is
3274 nonzero the function will have a frame pointer.
3276 The expression can in principle examine the current function and decide
3277 according to the facts, but on most machines the constant 0 or the
3278 constant 1 suffices. Use 0 when the machine allows code to be generated
3279 with no frame pointer, and doing so saves some time or space. Use 1
3280 when there is no possible advantage to avoiding a frame pointer.
3282 In certain cases, the compiler does not know how to produce valid code
3283 without a frame pointer. The compiler recognizes those cases and
3284 automatically gives the function a frame pointer regardless of what
3285 @code{FRAME_POINTER_REQUIRED} says. You don't need to worry about
3288 In a function that does not require a frame pointer, the frame pointer
3289 register can be allocated for ordinary usage, unless you mark it as a
3290 fixed register. See @code{FIXED_REGISTERS} for more information.
3293 @findex get_frame_size
3294 @defmac INITIAL_FRAME_POINTER_OFFSET (@var{depth-var})
3295 A C statement to store in the variable @var{depth-var} the difference
3296 between the frame pointer and the stack pointer values immediately after
3297 the function prologue. The value would be computed from information
3298 such as the result of @code{get_frame_size ()} and the tables of
3299 registers @code{regs_ever_live} and @code{call_used_regs}.
3301 If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and
3302 need not be defined. Otherwise, it must be defined even if
3303 @code{FRAME_POINTER_REQUIRED} is defined to always be true; in that
3304 case, you may set @var{depth-var} to anything.
3307 @defmac ELIMINABLE_REGS
3308 If defined, this macro specifies a table of register pairs used to
3309 eliminate unneeded registers that point into the stack frame. If it is not
3310 defined, the only elimination attempted by the compiler is to replace
3311 references to the frame pointer with references to the stack pointer.
3313 The definition of this macro is a list of structure initializations, each
3314 of which specifies an original and replacement register.
3316 On some machines, the position of the argument pointer is not known until
3317 the compilation is completed. In such a case, a separate hard register
3318 must be used for the argument pointer. This register can be eliminated by
3319 replacing it with either the frame pointer or the argument pointer,
3320 depending on whether or not the frame pointer has been eliminated.
3322 In this case, you might specify:
3324 #define ELIMINABLE_REGS \
3325 @{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \
3326 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \
3327 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@}
3330 Note that the elimination of the argument pointer with the stack pointer is
3331 specified first since that is the preferred elimination.
3334 @defmac CAN_ELIMINATE (@var{from-reg}, @var{to-reg})
3335 A C expression that returns nonzero if the compiler is allowed to try
3336 to replace register number @var{from-reg} with register number
3337 @var{to-reg}. This macro need only be defined if @code{ELIMINABLE_REGS}
3338 is defined, and will usually be the constant 1, since most of the cases
3339 preventing register elimination are things that the compiler already
3343 @defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var})
3344 This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It
3345 specifies the initial difference between the specified pair of
3346 registers. This macro must be defined if @code{ELIMINABLE_REGS} is
3350 @node Stack Arguments
3351 @subsection Passing Function Arguments on the Stack
3352 @cindex arguments on stack
3353 @cindex stack arguments
3355 The macros in this section control how arguments are passed
3356 on the stack. See the following section for other macros that
3357 control passing certain arguments in registers.
3359 @defmac PROMOTE_PROTOTYPES
3360 A C expression whose value is nonzero if an argument declared in
3361 a prototype as an integral type smaller than @code{int} should
3362 actually be passed as an @code{int}. In addition to avoiding
3363 errors in certain cases of mismatch, it also makes for better
3364 code on certain machines. If the macro is not defined in target
3365 header files, it defaults to 0.
3369 A C expression. If nonzero, push insns will be used to pass
3371 If the target machine does not have a push instruction, set it to zero.
3372 That directs GCC to use an alternate strategy: to
3373 allocate the entire argument block and then store the arguments into
3374 it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too.
3377 @defmac PUSH_ARGS_REVERSED
3378 A C expression. If nonzero, function arguments will be evaluated from
3379 last to first, rather than from first to last. If this macro is not
3380 defined, it defaults to @code{PUSH_ARGS} on targets where the stack
3381 and args grow in opposite directions, and 0 otherwise.
3384 @defmac PUSH_ROUNDING (@var{npushed})
3385 A C expression that is the number of bytes actually pushed onto the
3386 stack when an instruction attempts to push @var{npushed} bytes.
3388 On some machines, the definition
3391 #define PUSH_ROUNDING(BYTES) (BYTES)
3395 will suffice. But on other machines, instructions that appear
3396 to push one byte actually push two bytes in an attempt to maintain
3397 alignment. Then the definition should be
3400 #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
3404 @findex current_function_outgoing_args_size
3405 @defmac ACCUMULATE_OUTGOING_ARGS
3406 A C expression. If nonzero, the maximum amount of space required for outgoing arguments
3407 will be computed and placed into the variable
3408 @code{current_function_outgoing_args_size}. No space will be pushed
3409 onto the stack for each call; instead, the function prologue should
3410 increase the stack frame size by this amount.
3412 Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS}
3416 @defmac REG_PARM_STACK_SPACE (@var{fndecl})
3417 Define this macro if functions should assume that stack space has been
3418 allocated for arguments even when their values are passed in
3421 The value of this macro is the size, in bytes, of the area reserved for
3422 arguments passed in registers for the function represented by @var{fndecl},
3423 which can be zero if GCC is calling a library function.
3425 This space can be allocated by the caller, or be a part of the
3426 machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says
3429 @c above is overfull. not sure what to do. --mew 5feb93 did
3430 @c something, not sure if it looks good. --mew 10feb93
3432 @defmac MAYBE_REG_PARM_STACK_SPACE
3433 @defmacx FINAL_REG_PARM_STACK_SPACE (@var{const_size}, @var{var_size})
3434 Define these macros in addition to the one above if functions might
3435 allocate stack space for arguments even when their values are passed
3436 in registers. These should be used when the stack space allocated
3437 for arguments in registers is not a simple constant independent of the
3438 function declaration.
3440 The value of the first macro is the size, in bytes, of the area that
3441 we should initially assume would be reserved for arguments passed in registers.
3443 The value of the second macro is the actual size, in bytes, of the area
3444 that will be reserved for arguments passed in registers. This takes two
3445 arguments: an integer representing the number of bytes of fixed sized
3446 arguments on the stack, and a tree representing the number of bytes of
3447 variable sized arguments on the stack.
3449 When these macros are defined, @code{REG_PARM_STACK_SPACE} will only be
3450 called for libcall functions, the current function, or for a function
3451 being called when it is known that such stack space must be allocated.
3452 In each case this value can be easily computed.
3454 When deciding whether a called function needs such stack space, and how
3455 much space to reserve, GCC uses these two macros instead of
3456 @code{REG_PARM_STACK_SPACE}.
3459 @defmac OUTGOING_REG_PARM_STACK_SPACE
3460 Define this if it is the responsibility of the caller to allocate the area
3461 reserved for arguments passed in registers.
3463 If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls
3464 whether the space for these arguments counts in the value of
3465 @code{current_function_outgoing_args_size}.
3468 @defmac STACK_PARMS_IN_REG_PARM_AREA
3469 Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the
3470 stack parameters don't skip the area specified by it.
3471 @c i changed this, makes more sens and it should have taken care of the
3472 @c overfull.. not as specific, tho. --mew 5feb93
3474 Normally, when a parameter is not passed in registers, it is placed on the
3475 stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro
3476 suppresses this behavior and causes the parameter to be passed on the
3477 stack in its natural location.
3480 @defmac RETURN_POPS_ARGS (@var{fundecl}, @var{funtype}, @var{stack-size})
3481 A C expression that should indicate the number of bytes of its own
3482 arguments that a function pops on returning, or 0 if the
3483 function pops no arguments and the caller must therefore pop them all
3484 after the function returns.
3486 @var{fundecl} is a C variable whose value is a tree node that describes
3487 the function in question. Normally it is a node of type
3488 @code{FUNCTION_DECL} that describes the declaration of the function.
3489 From this you can obtain the @code{DECL_ATTRIBUTES} of the function.
3491 @var{funtype} is a C variable whose value is a tree node that
3492 describes the function in question. Normally it is a node of type
3493 @code{FUNCTION_TYPE} that describes the data type of the function.
3494 From this it is possible to obtain the data types of the value and
3495 arguments (if known).
3497 When a call to a library function is being considered, @var{fundecl}
3498 will contain an identifier node for the library function. Thus, if
3499 you need to distinguish among various library functions, you can do so
3500 by their names. Note that ``library function'' in this context means
3501 a function used to perform arithmetic, whose name is known specially
3502 in the compiler and was not mentioned in the C code being compiled.
3504 @var{stack-size} is the number of bytes of arguments passed on the
3505 stack. If a variable number of bytes is passed, it is zero, and
3506 argument popping will always be the responsibility of the calling function.
3508 On the VAX, all functions always pop their arguments, so the definition
3509 of this macro is @var{stack-size}. On the 68000, using the standard
3510 calling convention, no functions pop their arguments, so the value of
3511 the macro is always 0 in this case. But an alternative calling
3512 convention is available in which functions that take a fixed number of
3513 arguments pop them but other functions (such as @code{printf}) pop
3514 nothing (the caller pops all). When this convention is in use,
3515 @var{funtype} is examined to determine whether a function takes a fixed
3516 number of arguments.
3519 @defmac CALL_POPS_ARGS (@var{cum})
3520 A C expression that should indicate the number of bytes a call sequence
3521 pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS}
3522 when compiling a function call.
3524 @var{cum} is the variable in which all arguments to the called function
3525 have been accumulated.
3527 On certain architectures, such as the SH5, a call trampoline is used
3528 that pops certain registers off the stack, depending on the arguments
3529 that have been passed to the function. Since this is a property of the
3530 call site, not of the called function, @code{RETURN_POPS_ARGS} is not
3534 @node Register Arguments
3535 @subsection Passing Arguments in Registers
3536 @cindex arguments in registers
3537 @cindex registers arguments
3539 This section describes the macros which let you control how various
3540 types of arguments are passed in registers or how they are arranged in
3543 @defmac FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
3544 A C expression that controls whether a function argument is passed
3545 in a register, and which register.
3547 The arguments are @var{cum}, which summarizes all the previous
3548 arguments; @var{mode}, the machine mode of the argument; @var{type},
3549 the data type of the argument as a tree node or 0 if that is not known
3550 (which happens for C support library functions); and @var{named},
3551 which is 1 for an ordinary argument and 0 for nameless arguments that
3552 correspond to @samp{@dots{}} in the called function's prototype.
3553 @var{type} can be an incomplete type if a syntax error has previously
3556 The value of the expression is usually either a @code{reg} RTX for the
3557 hard register in which to pass the argument, or zero to pass the
3558 argument on the stack.
3560 For machines like the VAX and 68000, where normally all arguments are
3561 pushed, zero suffices as a definition.
3563 The value of the expression can also be a @code{parallel} RTX@. This is
3564 used when an argument is passed in multiple locations. The mode of the
3565 @code{parallel} should be the mode of the entire argument. The
3566 @code{parallel} holds any number of @code{expr_list} pairs; each one
3567 describes where part of the argument is passed. In each
3568 @code{expr_list} the first operand must be a @code{reg} RTX for the hard
3569 register in which to pass this part of the argument, and the mode of the
3570 register RTX indicates how large this part of the argument is. The
3571 second operand of the @code{expr_list} is a @code{const_int} which gives
3572 the offset in bytes into the entire argument of where this part starts.
3573 As a special exception the first @code{expr_list} in the @code{parallel}
3574 RTX may have a first operand of zero. This indicates that the entire
3575 argument is also stored on the stack.
3577 The last time this macro is called, it is called with @code{MODE ==
3578 VOIDmode}, and its result is passed to the @code{call} or @code{call_value}
3579 pattern as operands 2 and 3 respectively.
3581 @cindex @file{stdarg.h} and register arguments
3582 The usual way to make the ISO library @file{stdarg.h} work on a machine
3583 where some arguments are usually passed in registers, is to cause
3584 nameless arguments to be passed on the stack instead. This is done
3585 by making @code{FUNCTION_ARG} return 0 whenever @var{named} is 0.
3587 @cindex @code{MUST_PASS_IN_STACK}, and @code{FUNCTION_ARG}
3588 @cindex @code{REG_PARM_STACK_SPACE}, and @code{FUNCTION_ARG}
3589 You may use the macro @code{MUST_PASS_IN_STACK (@var{mode}, @var{type})}
3590 in the definition of this macro to determine if this argument is of a
3591 type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE}
3592 is not defined and @code{FUNCTION_ARG} returns nonzero for such an
3593 argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is
3594 defined, the argument will be computed in the stack and then loaded into
3598 @defmac MUST_PASS_IN_STACK (@var{mode}, @var{type})
3599 Define as a C expression that evaluates to nonzero if we do not know how
3600 to pass TYPE solely in registers. The file @file{expr.h} defines a
3601 definition that is usually appropriate, refer to @file{expr.h} for additional
3605 @defmac FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
3606 Define this macro if the target machine has ``register windows'', so
3607 that the register in which a function sees an arguments is not
3608 necessarily the same as the one in which the caller passed the
3611 For such machines, @code{FUNCTION_ARG} computes the register in which
3612 the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
3613 be defined in a similar fashion to tell the function being called
3614 where the arguments will arrive.
3616 If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
3617 serves both purposes.
3620 @defmac FUNCTION_ARG_PARTIAL_NREGS (@var{cum}, @var{mode}, @var{type}, @var{named})
3621 A C expression for the number of words, at the beginning of an
3622 argument, that must be put in registers. The value must be zero for
3623 arguments that are passed entirely in registers or that are entirely
3624 pushed on the stack.
3626 On some machines, certain arguments must be passed partially in
3627 registers and partially in memory. On these machines, typically the
3628 first @var{n} words of arguments are passed in registers, and the rest
3629 on the stack. If a multi-word argument (a @code{double} or a
3630 structure) crosses that boundary, its first few words must be passed
3631 in registers and the rest must be pushed. This macro tells the
3632 compiler when this occurs, and how many of the words should go in
3635 @code{FUNCTION_ARG} for these arguments should return the first
3636 register to be used by the caller for this argument; likewise
3637 @code{FUNCTION_INCOMING_ARG}, for the called function.
3640 @defmac FUNCTION_ARG_PASS_BY_REFERENCE (@var{cum}, @var{mode}, @var{type}, @var{named})
3641 A C expression that indicates when an argument must be passed by reference.
3642 If nonzero for an argument, a copy of that argument is made in memory and a
3643 pointer to the argument is passed instead of the argument itself.
3644 The pointer is passed in whatever way is appropriate for passing a pointer
3647 On machines where @code{REG_PARM_STACK_SPACE} is not defined, a suitable
3648 definition of this macro might be
3650 #define FUNCTION_ARG_PASS_BY_REFERENCE\
3651 (CUM, MODE, TYPE, NAMED) \
3652 MUST_PASS_IN_STACK (MODE, TYPE)
3654 @c this is *still* too long. --mew 5feb93
3657 @defmac FUNCTION_ARG_CALLEE_COPIES (@var{cum}, @var{mode}, @var{type}, @var{named})
3658 If defined, a C expression that indicates when it is the called function's
3659 responsibility to make a copy of arguments passed by invisible reference.
3660 Normally, the caller makes a copy and passes the address of the copy to the
3661 routine being called. When @code{FUNCTION_ARG_CALLEE_COPIES} is defined and is
3662 nonzero, the caller does not make a copy. Instead, it passes a pointer to the
3663 ``live'' value. The called function must not modify this value. If it can be
3664 determined that the value won't be modified, it need not make a copy;
3665 otherwise a copy must be made.
3668 @defmac CUMULATIVE_ARGS
3669 A C type for declaring a variable that is used as the first argument of
3670 @code{FUNCTION_ARG} and other related values. For some target machines,
3671 the type @code{int} suffices and can hold the number of bytes of
3674 There is no need to record in @code{CUMULATIVE_ARGS} anything about the
3675 arguments that have been passed on the stack. The compiler has other
3676 variables to keep track of that. For target machines on which all
3677 arguments are passed on the stack, there is no need to store anything in
3678 @code{CUMULATIVE_ARGS}; however, the data structure must exist and
3679 should not be empty, so use @code{int}.
3682 @defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl})
3683 A C statement (sans semicolon) for initializing the variable
3684 @var{cum} for the state at the beginning of the argument list. The
3685 variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype}
3686 is the tree node for the data type of the function which will receive
3687 the args, or 0 if the args are to a compiler support library function.
3688 For direct calls that are not libcalls, @var{fndecl} contain the
3689 declaration node of the function. @var{fndecl} is also set when
3690 @code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function
3693 When processing a call to a compiler support library function,
3694 @var{libname} identifies which one. It is a @code{symbol_ref} rtx which
3695 contains the name of the function, as a string. @var{libname} is 0 when
3696 an ordinary C function call is being processed. Thus, each time this
3697 macro is called, either @var{libname} or @var{fntype} is nonzero, but
3698 never both of them at once.
3701 @defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname})
3702 Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls,
3703 it gets a @code{MODE} argument instead of @var{fntype}, that would be
3704 @code{NULL}. @var{indirect} would always be zero, too. If this macro
3705 is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname,
3706 0)} is used instead.
3709 @defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname})
3710 Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of
3711 finding the arguments for the function being compiled. If this macro is
3712 undefined, @code{INIT_CUMULATIVE_ARGS} is used instead.
3714 The value passed for @var{libname} is always 0, since library routines
3715 with special calling conventions are never compiled with GCC@. The
3716 argument @var{libname} exists for symmetry with
3717 @code{INIT_CUMULATIVE_ARGS}.
3718 @c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe.
3719 @c --mew 5feb93 i switched the order of the sentences. --mew 10feb93
3722 @defmac FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
3723 A C statement (sans semicolon) to update the summarizer variable
3724 @var{cum} to advance past an argument in the argument list. The
3725 values @var{mode}, @var{type} and @var{named} describe that argument.
3726 Once this is done, the variable @var{cum} is suitable for analyzing
3727 the @emph{following} argument with @code{FUNCTION_ARG}, etc.
3729 This macro need not do anything if the argument in question was passed
3730 on the stack. The compiler knows how to track the amount of stack space
3731 used for arguments without any special help.
3734 @defmac FUNCTION_ARG_PADDING (@var{mode}, @var{type})
3735 If defined, a C expression which determines whether, and in which direction,
3736 to pad out an argument with extra space. The value should be of type
3737 @code{enum direction}: either @code{upward} to pad above the argument,
3738 @code{downward} to pad below, or @code{none} to inhibit padding.
3740 The @emph{amount} of padding is always just enough to reach the next
3741 multiple of @code{FUNCTION_ARG_BOUNDARY}; this macro does not control
3744 This macro has a default definition which is right for most systems.
3745 For little-endian machines, the default is to pad upward. For
3746 big-endian machines, the default is to pad downward for an argument of
3747 constant size shorter than an @code{int}, and upward otherwise.
3750 @defmac PAD_VARARGS_DOWN
3751 If defined, a C expression which determines whether the default
3752 implementation of va_arg will attempt to pad down before reading the
3753 next argument, if that argument is smaller than its aligned space as
3754 controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such
3755 arguments are padded down if @code{BYTES_BIG_ENDIAN} is true.
3758 @defmac FUNCTION_ARG_BOUNDARY (@var{mode}, @var{type})
3759 If defined, a C expression that gives the alignment boundary, in bits,
3760 of an argument with the specified mode and type. If it is not defined,
3761 @code{PARM_BOUNDARY} is used for all arguments.
3764 @defmac FUNCTION_ARG_REGNO_P (@var{regno})
3765 A C expression that is nonzero if @var{regno} is the number of a hard
3766 register in which function arguments are sometimes passed. This does
3767 @emph{not} include implicit arguments such as the static chain and
3768 the structure-value address. On many machines, no registers can be
3769 used for this purpose since all function arguments are pushed on the
3773 @defmac SPLIT_COMPLEX_ARGS
3775 Define this macro to a nonzero value if complex function arguments
3776 should be split into their corresponding components. By default, GCC
3777 will attempt to pack complex arguments into the target's word size.
3778 Some ABIs require complex arguments to be split and treated as their
3779 individual components. For example, on AIX64, complex floats should
3780 be passed in a pair of floating point registers, even though a complex
3781 float would fit in one 64-bit floating point register.
3784 @defmac LOAD_ARGS_REVERSED
3785 If defined, the order in which arguments are loaded into their
3786 respective argument registers is reversed so that the last
3787 argument is loaded first. This macro only affects arguments
3788 passed in registers.
3792 @subsection How Scalar Function Values Are Returned
3793 @cindex return values in registers
3794 @cindex values, returned by functions
3795 @cindex scalars, returned as values
3797 This section discusses the macros that control returning scalars as
3798 values---values that can fit in registers.
3800 @defmac FUNCTION_VALUE (@var{valtype}, @var{func})
3801 A C expression to create an RTX representing the place where a
3802 function returns a value of data type @var{valtype}. @var{valtype} is
3803 a tree node representing a data type. Write @code{TYPE_MODE
3804 (@var{valtype})} to get the machine mode used to represent that type.
3805 On many machines, only the mode is relevant. (Actually, on most
3806 machines, scalar values are returned in the same place regardless of
3809 The value of the expression is usually a @code{reg} RTX for the hard
3810 register where the return value is stored. The value can also be a
3811 @code{parallel} RTX, if the return value is in multiple places. See
3812 @code{FUNCTION_ARG} for an explanation of the @code{parallel} form.
3814 If @code{PROMOTE_FUNCTION_RETURN} is defined, you must apply the same
3815 promotion rules specified in @code{PROMOTE_MODE} if @var{valtype} is a
3818 If the precise function being called is known, @var{func} is a tree
3819 node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
3820 pointer. This makes it possible to use a different value-returning
3821 convention for specific functions when all their calls are
3824 @code{FUNCTION_VALUE} is not used for return vales with aggregate data
3825 types, because these are returned in another way. See
3826 @code{STRUCT_VALUE_REGNUM} and related macros, below.
3829 @defmac FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
3830 Define this macro if the target machine has ``register windows''
3831 so that the register in which a function returns its value is not
3832 the same as the one in which the caller sees the value.
3834 For such machines, @code{FUNCTION_VALUE} computes the register in which
3835 the caller will see the value. @code{FUNCTION_OUTGOING_VALUE} should be
3836 defined in a similar fashion to tell the function where to put the
3839 If @code{FUNCTION_OUTGOING_VALUE} is not defined,
3840 @code{FUNCTION_VALUE} serves both purposes.
3842 @code{FUNCTION_OUTGOING_VALUE} is not used for return vales with
3843 aggregate data types, because these are returned in another way. See
3844 @code{STRUCT_VALUE_REGNUM} and related macros, below.
3847 @defmac LIBCALL_VALUE (@var{mode})
3848 A C expression to create an RTX representing the place where a library
3849 function returns a value of mode @var{mode}. If the precise function
3850 being called is known, @var{func} is a tree node
3851 (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
3852 pointer. This makes it possible to use a different value-returning
3853 convention for specific functions when all their calls are
3856 Note that ``library function'' in this context means a compiler
3857 support routine, used to perform arithmetic, whose name is known
3858 specially by the compiler and was not mentioned in the C code being
3861 The definition of @code{LIBRARY_VALUE} need not be concerned aggregate
3862 data types, because none of the library functions returns such types.
3865 @defmac FUNCTION_VALUE_REGNO_P (@var{regno})
3866 A C expression that is nonzero if @var{regno} is the number of a hard
3867 register in which the values of called function may come back.
3869 A register whose use for returning values is limited to serving as the
3870 second of a pair (for a value of type @code{double}, say) need not be
3871 recognized by this macro. So for most machines, this definition
3875 #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
3878 If the machine has register windows, so that the caller and the called
3879 function use different registers for the return value, this macro
3880 should recognize only the caller's register numbers.
3883 @defmac APPLY_RESULT_SIZE
3884 Define this macro if @samp{untyped_call} and @samp{untyped_return}
3885 need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for
3886 saving and restoring an arbitrary return value.
3889 @node Aggregate Return
3890 @subsection How Large Values Are Returned
3891 @cindex aggregates as return values
3892 @cindex large return values
3893 @cindex returning aggregate values
3894 @cindex structure value address
3896 When a function value's mode is @code{BLKmode} (and in some other
3897 cases), the value is not returned according to @code{FUNCTION_VALUE}
3898 (@pxref{Scalar Return}). Instead, the caller passes the address of a
3899 block of memory in which the value should be stored. This address
3900 is called the @dfn{structure value address}.
3902 This section describes how to control returning structure values in
3905 @defmac RETURN_IN_MEMORY (@var{type})
3906 A C expression which can inhibit the returning of certain function
3907 values in registers, based on the type of value. A nonzero value says
3908 to return the function value in memory, just as large structures are
3909 always returned. Here @var{type} will be a C expression of type
3910 @code{tree}, representing the data type of the value.
3912 Note that values of mode @code{BLKmode} must be explicitly handled
3913 by this macro. Also, the option @option{-fpcc-struct-return}
3914 takes effect regardless of this macro. On most systems, it is
3915 possible to leave the macro undefined; this causes a default
3916 definition to be used, whose value is the constant 1 for @code{BLKmode}
3917 values, and 0 otherwise.
3919 Do not use this macro to indicate that structures and unions should always
3920 be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN}
3924 @defmac DEFAULT_PCC_STRUCT_RETURN
3925 Define this macro to be 1 if all structure and union return values must be
3926 in memory. Since this results in slower code, this should be defined
3927 only if needed for compatibility with other compilers or with an ABI@.
3928 If you define this macro to be 0, then the conventions used for structure
3929 and union return values are decided by the @code{RETURN_IN_MEMORY} macro.
3931 If not defined, this defaults to the value 1.
3934 @defmac STRUCT_VALUE_REGNUM
3935 If the structure value address is passed in a register, then
3936 @code{STRUCT_VALUE_REGNUM} should be the number of that register.
3939 @defmac STRUCT_VALUE
3940 If the structure value address is not passed in a register, define
3941 @code{STRUCT_VALUE} as an expression returning an RTX for the place
3942 where the address is passed. If it returns 0, the address is passed as
3943 an ``invisible'' first argument.
3946 @defmac STRUCT_VALUE_INCOMING_REGNUM
3947 On some architectures the place where the structure value address
3948 is found by the called function is not the same place that the
3949 caller put it. This can be due to register windows, or it could
3950 be because the function prologue moves it to a different place.
3952 If the incoming location of the structure value address is in a
3953 register, define this macro as the register number.
3956 @defmac STRUCT_VALUE_INCOMING
3957 If the incoming location is not a register, then you should define
3958 @code{STRUCT_VALUE_INCOMING} as an expression for an RTX for where the
3959 called function should find the value. If it should find the value on
3960 the stack, define this to create a @code{mem} which refers to the frame
3961 pointer. A definition of 0 means that the address is passed as an
3962 ``invisible'' first argument.
3965 @defmac PCC_STATIC_STRUCT_RETURN
3966 Define this macro if the usual system convention on the target machine
3967 for returning structures and unions is for the called function to return
3968 the address of a static variable containing the value.
3970 Do not define this if the usual system convention is for the caller to
3971 pass an address to the subroutine.
3973 This macro has effect in @option{-fpcc-struct-return} mode, but it does
3974 nothing when you use @option{-freg-struct-return} mode.
3978 @subsection Caller-Saves Register Allocation
3980 If you enable it, GCC can save registers around function calls. This
3981 makes it possible to use call-clobbered registers to hold variables that
3982 must live across calls.
3984 @defmac DEFAULT_CALLER_SAVES
3985 Define this macro if function calls on the target machine do not preserve
3986 any registers; in other words, if @code{CALL_USED_REGISTERS} has 1
3987 for all registers. When defined, this macro enables @option{-fcaller-saves}
3988 by default for all optimization levels. It has no effect for optimization
3989 levels 2 and higher, where @option{-fcaller-saves} is the default.
3992 @defmac CALLER_SAVE_PROFITABLE (@var{refs}, @var{calls})
3993 A C expression to determine whether it is worthwhile to consider placing
3994 a pseudo-register in a call-clobbered hard register and saving and
3995 restoring it around each function call. The expression should be 1 when
3996 this is worth doing, and 0 otherwise.
3998 If you don't define this macro, a default is used which is good on most
3999 machines: @code{4 * @var{calls} < @var{refs}}.
4002 @defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs})
4003 A C expression specifying which mode is required for saving @var{nregs}
4004 of a pseudo-register in call-clobbered hard register @var{regno}. If
4005 @var{regno} is unsuitable for caller save, @code{VOIDmode} should be
4006 returned. For most machines this macro need not be defined since GCC
4007 will select the smallest suitable mode.
4010 @node Function Entry
4011 @subsection Function Entry and Exit
4012 @cindex function entry and exit
4016 This section describes the macros that output function entry
4017 (@dfn{prologue}) and exit (@dfn{epilogue}) code.
4019 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_PROLOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4020 If defined, a function that outputs the assembler code for entry to a
4021 function. The prologue is responsible for setting up the stack frame,
4022 initializing the frame pointer register, saving registers that must be
4023 saved, and allocating @var{size} additional bytes of storage for the
4024 local variables. @var{size} is an integer. @var{file} is a stdio
4025 stream to which the assembler code should be output.
4027 The label for the beginning of the function need not be output by this
4028 macro. That has already been done when the macro is run.
4030 @findex regs_ever_live
4031 To determine which registers to save, the macro can refer to the array
4032 @code{regs_ever_live}: element @var{r} is nonzero if hard register
4033 @var{r} is used anywhere within the function. This implies the function
4034 prologue should save register @var{r}, provided it is not one of the
4035 call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use
4036 @code{regs_ever_live}.)
4038 On machines that have ``register windows'', the function entry code does
4039 not save on the stack the registers that are in the windows, even if
4040 they are supposed to be preserved by function calls; instead it takes
4041 appropriate steps to ``push'' the register stack, if any non-call-used
4042 registers are used in the function.
4044 @findex frame_pointer_needed
4045 On machines where functions may or may not have frame-pointers, the
4046 function entry code must vary accordingly; it must set up the frame
4047 pointer if one is wanted, and not otherwise. To determine whether a
4048 frame pointer is in wanted, the macro can refer to the variable
4049 @code{frame_pointer_needed}. The variable's value will be 1 at run
4050 time in a function that needs a frame pointer. @xref{Elimination}.
4052 The function entry code is responsible for allocating any stack space
4053 required for the function. This stack space consists of the regions
4054 listed below. In most cases, these regions are allocated in the
4055 order listed, with the last listed region closest to the top of the
4056 stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and
4057 the highest address if it is not defined). You can use a different order
4058 for a machine if doing so is more convenient or required for
4059 compatibility reasons. Except in cases where required by standard
4060 or by a debugger, there is no reason why the stack layout used by GCC
4061 need agree with that used by other compilers for a machine.
4064 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_END_PROLOGUE (FILE *@var{file})
4065 If defined, a function that outputs assembler code at the end of a
4066 prologue. This should be used when the function prologue is being
4067 emitted as RTL, and you have some extra assembler that needs to be
4068 emitted. @xref{prologue instruction pattern}.
4071 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_BEGIN_EPILOGUE (FILE *@var{file})
4072 If defined, a function that outputs assembler code at the start of an
4073 epilogue. This should be used when the function epilogue is being
4074 emitted as RTL, and you have some extra assembler that needs to be
4075 emitted. @xref{epilogue instruction pattern}.
4078 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_EPILOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4079 If defined, a function that outputs the assembler code for exit from a
4080 function. The epilogue is responsible for restoring the saved
4081 registers and stack pointer to their values when the function was
4082 called, and returning control to the caller. This macro takes the
4083 same arguments as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the
4084 registers to restore are determined from @code{regs_ever_live} and
4085 @code{CALL_USED_REGISTERS} in the same way.
4087 On some machines, there is a single instruction that does all the work
4088 of returning from the function. On these machines, give that
4089 instruction the name @samp{return} and do not define the macro
4090 @code{TARGET_ASM_FUNCTION_EPILOGUE} at all.
4092 Do not define a pattern named @samp{return} if you want the
4093 @code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target
4094 switches to control whether return instructions or epilogues are used,
4095 define a @samp{return} pattern with a validity condition that tests the
4096 target switches appropriately. If the @samp{return} pattern's validity
4097 condition is false, epilogues will be used.
4099 On machines where functions may or may not have frame-pointers, the
4100 function exit code must vary accordingly. Sometimes the code for these
4101 two cases is completely different. To determine whether a frame pointer
4102 is wanted, the macro can refer to the variable
4103 @code{frame_pointer_needed}. The variable's value will be 1 when compiling
4104 a function that needs a frame pointer.
4106 Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and
4107 @code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially.
4108 The C variable @code{current_function_is_leaf} is nonzero for such a
4109 function. @xref{Leaf Functions}.
4111 On some machines, some functions pop their arguments on exit while
4112 others leave that for the caller to do. For example, the 68020 when
4113 given @option{-mrtd} pops arguments in functions that take a fixed
4114 number of arguments.
4116 @findex current_function_pops_args
4117 Your definition of the macro @code{RETURN_POPS_ARGS} decides which
4118 functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE}
4119 needs to know what was decided. The variable that is called
4120 @code{current_function_pops_args} is the number of bytes of its
4121 arguments that a function should pop. @xref{Scalar Return}.
4122 @c what is the "its arguments" in the above sentence referring to, pray
4123 @c tell? --mew 5feb93
4128 @findex current_function_pretend_args_size
4129 A region of @code{current_function_pretend_args_size} bytes of
4130 uninitialized space just underneath the first argument arriving on the
4131 stack. (This may not be at the very start of the allocated stack region
4132 if the calling sequence has pushed anything else since pushing the stack
4133 arguments. But usually, on such machines, nothing else has been pushed
4134 yet, because the function prologue itself does all the pushing.) This
4135 region is used on machines where an argument may be passed partly in
4136 registers and partly in memory, and, in some cases to support the
4137 features in @code{<stdarg.h>}.
4140 An area of memory used to save certain registers used by the function.
4141 The size of this area, which may also include space for such things as
4142 the return address and pointers to previous stack frames, is
4143 machine-specific and usually depends on which registers have been used
4144 in the function. Machines with register windows often do not require
4148 A region of at least @var{size} bytes, possibly rounded up to an allocation
4149 boundary, to contain the local variables of the function. On some machines,
4150 this region and the save area may occur in the opposite order, with the
4151 save area closer to the top of the stack.
4154 @cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames
4155 Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of
4156 @code{current_function_outgoing_args_size} bytes to be used for outgoing
4157 argument lists of the function. @xref{Stack Arguments}.
4160 Normally, it is necessary for the macros
4161 @code{TARGET_ASM_FUNCTION_PROLOGUE} and
4162 @code{TARGET_ASM_FUNCTION_EPILOGUE} to treat leaf functions specially.
4163 The C variable @code{current_function_is_leaf} is nonzero for such a
4166 @defmac EXIT_IGNORE_STACK
4167 Define this macro as a C expression that is nonzero if the return
4168 instruction or the function epilogue ignores the value of the stack
4169 pointer; in other words, if it is safe to delete an instruction to
4170 adjust the stack pointer before a return from the function.
4172 Note that this macro's value is relevant only for functions for which
4173 frame pointers are maintained. It is never safe to delete a final
4174 stack adjustment in a function that has no frame pointer, and the
4175 compiler knows this regardless of @code{EXIT_IGNORE_STACK}.
4178 @defmac EPILOGUE_USES (@var{regno})
4179 Define this macro as a C expression that is nonzero for registers that are
4180 used by the epilogue or the @samp{return} pattern. The stack and frame
4181 pointer registers are already be assumed to be used as needed.
4184 @defmac EH_USES (@var{regno})
4185 Define this macro as a C expression that is nonzero for registers that are
4186 used by the exception handling mechanism, and so should be considered live
4187 on entry to an exception edge.
4190 @defmac DELAY_SLOTS_FOR_EPILOGUE
4191 Define this macro if the function epilogue contains delay slots to which
4192 instructions from the rest of the function can be ``moved''. The
4193 definition should be a C expression whose value is an integer
4194 representing the number of delay slots there.
4197 @defmac ELIGIBLE_FOR_EPILOGUE_DELAY (@var{insn}, @var{n})
4198 A C expression that returns 1 if @var{insn} can be placed in delay
4199 slot number @var{n} of the epilogue.
4201 The argument @var{n} is an integer which identifies the delay slot now
4202 being considered (since different slots may have different rules of
4203 eligibility). It is never negative and is always less than the number
4204 of epilogue delay slots (what @code{DELAY_SLOTS_FOR_EPILOGUE} returns).
4205 If you reject a particular insn for a given delay slot, in principle, it
4206 may be reconsidered for a subsequent delay slot. Also, other insns may
4207 (at least in principle) be considered for the so far unfilled delay
4210 @findex current_function_epilogue_delay_list
4211 @findex final_scan_insn
4212 The insns accepted to fill the epilogue delay slots are put in an RTL
4213 list made with @code{insn_list} objects, stored in the variable
4214 @code{current_function_epilogue_delay_list}. The insn for the first
4215 delay slot comes first in the list. Your definition of the macro
4216 @code{TARGET_ASM_FUNCTION_EPILOGUE} should fill the delay slots by
4217 outputting the insns in this list, usually by calling
4218 @code{final_scan_insn}.
4220 You need not define this macro if you did not define
4221 @code{DELAY_SLOTS_FOR_EPILOGUE}.
4224 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_MI_THUNK (FILE *@var{file}, tree @var{thunk_fndecl}, HOST_WIDE_INT @var{delta}, tree @var{function})
4225 A function that outputs the assembler code for a thunk
4226 function, used to implement C++ virtual function calls with multiple
4227 inheritance. The thunk acts as a wrapper around a virtual function,
4228 adjusting the implicit object parameter before handing control off to
4231 First, emit code to add the integer @var{delta} to the location that
4232 contains the incoming first argument. Assume that this argument
4233 contains a pointer, and is the one used to pass the @code{this} pointer
4234 in C++. This is the incoming argument @emph{before} the function prologue,
4235 e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of
4236 all other incoming arguments.
4238 After the addition, emit code to jump to @var{function}, which is a
4239 @code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does
4240 not touch the return address. Hence returning from @var{FUNCTION} will
4241 return to whoever called the current @samp{thunk}.
4243 The effect must be as if @var{function} had been called directly with
4244 the adjusted first argument. This macro is responsible for emitting all
4245 of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE}
4246 and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked.
4248 The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function}
4249 have already been extracted from it.) It might possibly be useful on
4250 some targets, but probably not.
4252 If you do not define this macro, the target-independent code in the C++
4253 front end will generate a less efficient heavyweight thunk that calls
4254 @var{function} instead of jumping to it. The generic approach does
4255 not support varargs.
4258 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_MI_VCALL_THUNK (FILE *@var{file}, tree @var{thunk_fndecl}, HOST_WIDE_INT @var{delta}, int @var{vcall_offset}, tree @var{function})
4259 A function like @code{TARGET_ASM_OUTPUT_MI_THUNK}, except that if
4260 @var{vcall_offset} is nonzero, an additional adjustment should be made
4261 after adding @code{delta}. In particular, if @var{p} is the
4262 adjusted pointer, the following adjustment should be made:
4265 p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)]
4269 If this function is defined, it will always be used in place of
4270 @code{TARGET_ASM_OUTPUT_MI_THUNK}.
4274 @subsection Generating Code for Profiling
4275 @cindex profiling, code generation
4277 These macros will help you generate code for profiling.
4279 @defmac FUNCTION_PROFILER (@var{file}, @var{labelno})
4280 A C statement or compound statement to output to @var{file} some
4281 assembler code to call the profiling subroutine @code{mcount}.
4284 The details of how @code{mcount} expects to be called are determined by
4285 your operating system environment, not by GCC@. To figure them out,
4286 compile a small program for profiling using the system's installed C
4287 compiler and look at the assembler code that results.
4289 Older implementations of @code{mcount} expect the address of a counter
4290 variable to be loaded into some register. The name of this variable is
4291 @samp{LP} followed by the number @var{labelno}, so you would generate
4292 the name using @samp{LP%d} in a @code{fprintf}.
4295 @defmac PROFILE_HOOK
4296 A C statement or compound statement to output to @var{file} some assembly
4297 code to call the profiling subroutine @code{mcount} even the target does
4298 not support profiling.
4301 @defmac NO_PROFILE_COUNTERS
4302 Define this macro if the @code{mcount} subroutine on your system does
4303 not need a counter variable allocated for each function. This is true
4304 for almost all modern implementations. If you define this macro, you
4305 must not use the @var{labelno} argument to @code{FUNCTION_PROFILER}.
4308 @defmac PROFILE_BEFORE_PROLOGUE
4309 Define this macro if the code for function profiling should come before
4310 the function prologue. Normally, the profiling code comes after.
4314 @subsection Permitting tail calls
4317 @deftypefn {Target Hook} bool TARGET_FUNCTION_OK_FOR_SIBCALL (tree @var{decl}, tree @var{exp})
4318 True if it is ok to do sibling call optimization for the specified
4319 call expression @var{exp}. @var{decl} will be the called function,
4320 or @code{NULL} if this is an indirect call.
4322 It is not uncommon for limitations of calling conventions to prevent
4323 tail calls to functions outside the current unit of translation, or
4324 during PIC compilation. The hook is used to enforce these restrictions,
4325 as the @code{sibcall} md pattern can not fail, or fall over to a
4326 ``normal'' call. The criteria for successful sibling call optimization
4327 may vary greatly between different architectures.
4331 @section Implementing the Varargs Macros
4332 @cindex varargs implementation
4334 GCC comes with an implementation of @code{<varargs.h>} and
4335 @code{<stdarg.h>} that work without change on machines that pass arguments
4336 on the stack. Other machines require their own implementations of
4337 varargs, and the two machine independent header files must have
4338 conditionals to include it.
4340 ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in
4341 the calling convention for @code{va_start}. The traditional
4342 implementation takes just one argument, which is the variable in which
4343 to store the argument pointer. The ISO implementation of
4344 @code{va_start} takes an additional second argument. The user is
4345 supposed to write the last named argument of the function here.
4347 However, @code{va_start} should not use this argument. The way to find
4348 the end of the named arguments is with the built-in functions described
4351 @defmac __builtin_saveregs ()
4352 Use this built-in function to save the argument registers in memory so
4353 that the varargs mechanism can access them. Both ISO and traditional
4354 versions of @code{va_start} must use @code{__builtin_saveregs}, unless
4355 you use @code{SETUP_INCOMING_VARARGS} (see below) instead.
4357 On some machines, @code{__builtin_saveregs} is open-coded under the
4358 control of the macro @code{EXPAND_BUILTIN_SAVEREGS}. On other machines,
4359 it calls a routine written in assembler language, found in
4362 Code generated for the call to @code{__builtin_saveregs} appears at the
4363 beginning of the function, as opposed to where the call to
4364 @code{__builtin_saveregs} is written, regardless of what the code is.
4365 This is because the registers must be saved before the function starts
4366 to use them for its own purposes.
4367 @c i rewrote the first sentence above to fix an overfull hbox. --mew
4371 @defmac __builtin_args_info (@var{category})
4372 Use this built-in function to find the first anonymous arguments in
4375 In general, a machine may have several categories of registers used for
4376 arguments, each for a particular category of data types. (For example,
4377 on some machines, floating-point registers are used for floating-point
4378 arguments while other arguments are passed in the general registers.)
4379 To make non-varargs functions use the proper calling convention, you
4380 have defined the @code{CUMULATIVE_ARGS} data type to record how many
4381 registers in each category have been used so far
4383 @code{__builtin_args_info} accesses the same data structure of type
4384 @code{CUMULATIVE_ARGS} after the ordinary argument layout is finished
4385 with it, with @var{category} specifying which word to access. Thus, the
4386 value indicates the first unused register in a given category.
4388 Normally, you would use @code{__builtin_args_info} in the implementation
4389 of @code{va_start}, accessing each category just once and storing the
4390 value in the @code{va_list} object. This is because @code{va_list} will
4391 have to update the values, and there is no way to alter the
4392 values accessed by @code{__builtin_args_info}.
4395 @defmac __builtin_next_arg (@var{lastarg})
4396 This is the equivalent of @code{__builtin_args_info}, for stack
4397 arguments. It returns the address of the first anonymous stack
4398 argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it
4399 returns the address of the location above the first anonymous stack
4400 argument. Use it in @code{va_start} to initialize the pointer for
4401 fetching arguments from the stack. Also use it in @code{va_start} to
4402 verify that the second parameter @var{lastarg} is the last named argument
4403 of the current function.
4406 @defmac __builtin_classify_type (@var{object})
4407 Since each machine has its own conventions for which data types are
4408 passed in which kind of register, your implementation of @code{va_arg}
4409 has to embody these conventions. The easiest way to categorize the
4410 specified data type is to use @code{__builtin_classify_type} together
4411 with @code{sizeof} and @code{__alignof__}.
4413 @code{__builtin_classify_type} ignores the value of @var{object},
4414 considering only its data type. It returns an integer describing what
4415 kind of type that is---integer, floating, pointer, structure, and so on.
4417 The file @file{typeclass.h} defines an enumeration that you can use to
4418 interpret the values of @code{__builtin_classify_type}.
4421 These machine description macros help implement varargs:
4423 @defmac EXPAND_BUILTIN_SAVEREGS ()
4424 If defined, is a C expression that produces the machine-specific code
4425 for a call to @code{__builtin_saveregs}. This code will be moved to the
4426 very beginning of the function, before any parameter access are made.
4427 The return value of this function should be an RTX that contains the
4428 value to use as the return of @code{__builtin_saveregs}.
4431 @defmac SETUP_INCOMING_VARARGS (@var{args_so_far}, @var{mode}, @var{type}, @var{pretend_args_size}, @var{second_time})
4432 This macro offers an alternative to using @code{__builtin_saveregs} and
4433 defining the macro @code{EXPAND_BUILTIN_SAVEREGS}. Use it to store the
4434 anonymous register arguments into the stack so that all the arguments
4435 appear to have been passed consecutively on the stack. Once this is
4436 done, you can use the standard implementation of varargs that works for
4437 machines that pass all their arguments on the stack.
4439 The argument @var{args_so_far} is the @code{CUMULATIVE_ARGS} data
4440 structure, containing the values that are obtained after processing the
4441 named arguments. The arguments @var{mode} and @var{type} describe the
4442 last named argument---its machine mode and its data type as a tree node.
4444 The macro implementation should do two things: first, push onto the
4445 stack all the argument registers @emph{not} used for the named
4446 arguments, and second, store the size of the data thus pushed into the
4447 @code{int}-valued variable whose name is supplied as the argument
4448 @var{pretend_args_size}. The value that you store here will serve as
4449 additional offset for setting up the stack frame.
4451 Because you must generate code to push the anonymous arguments at
4452 compile time without knowing their data types,
4453 @code{SETUP_INCOMING_VARARGS} is only useful on machines that have just
4454 a single category of argument register and use it uniformly for all data
4457 If the argument @var{second_time} is nonzero, it means that the
4458 arguments of the function are being analyzed for the second time. This
4459 happens for an inline function, which is not actually compiled until the
4460 end of the source file. The macro @code{SETUP_INCOMING_VARARGS} should
4461 not generate any instructions in this case.
4464 @defmac STRICT_ARGUMENT_NAMING
4465 Define this macro to be a nonzero value if the location where a function
4466 argument is passed depends on whether or not it is a named argument.
4468 This macro controls how the @var{named} argument to @code{FUNCTION_ARG}
4469 is set for varargs and stdarg functions. If this macro returns a
4470 nonzero value, the @var{named} argument is always true for named
4471 arguments, and false for unnamed arguments. If it returns a value of
4472 zero, but @code{SETUP_INCOMING_VARARGS} is defined, then all arguments
4473 are treated as named. Otherwise, all named arguments except the last
4474 are treated as named.
4476 You need not define this macro if it always returns zero.
4479 @defmac PRETEND_OUTGOING_VARARGS_NAMED
4480 If you need to conditionally change ABIs so that one works with
4481 @code{SETUP_INCOMING_VARARGS}, but the other works like neither
4482 @code{SETUP_INCOMING_VARARGS} nor @code{STRICT_ARGUMENT_NAMING} was
4483 defined, then define this macro to return nonzero if
4484 @code{SETUP_INCOMING_VARARGS} is used, zero otherwise.
4485 Otherwise, you should not define this macro.
4489 @section Trampolines for Nested Functions
4490 @cindex trampolines for nested functions
4491 @cindex nested functions, trampolines for
4493 A @dfn{trampoline} is a small piece of code that is created at run time
4494 when the address of a nested function is taken. It normally resides on
4495 the stack, in the stack frame of the containing function. These macros
4496 tell GCC how to generate code to allocate and initialize a
4499 The instructions in the trampoline must do two things: load a constant
4500 address into the static chain register, and jump to the real address of
4501 the nested function. On CISC machines such as the m68k, this requires
4502 two instructions, a move immediate and a jump. Then the two addresses
4503 exist in the trampoline as word-long immediate operands. On RISC
4504 machines, it is often necessary to load each address into a register in
4505 two parts. Then pieces of each address form separate immediate
4508 The code generated to initialize the trampoline must store the variable
4509 parts---the static chain value and the function address---into the
4510 immediate operands of the instructions. On a CISC machine, this is
4511 simply a matter of copying each address to a memory reference at the
4512 proper offset from the start of the trampoline. On a RISC machine, it
4513 may be necessary to take out pieces of the address and store them
4516 @defmac TRAMPOLINE_TEMPLATE (@var{file})
4517 A C statement to output, on the stream @var{file}, assembler code for a
4518 block of data that contains the constant parts of a trampoline. This
4519 code should not include a label---the label is taken care of
4522 If you do not define this macro, it means no template is needed
4523 for the target. Do not define this macro on systems where the block move
4524 code to copy the trampoline into place would be larger than the code
4525 to generate it on the spot.
4528 @defmac TRAMPOLINE_SECTION
4529 The name of a subroutine to switch to the section in which the
4530 trampoline template is to be placed (@pxref{Sections}). The default is
4531 a value of @samp{readonly_data_section}, which places the trampoline in
4532 the section containing read-only data.
4535 @defmac TRAMPOLINE_SIZE
4536 A C expression for the size in bytes of the trampoline, as an integer.
4539 @defmac TRAMPOLINE_ALIGNMENT
4540 Alignment required for trampolines, in bits.
4542 If you don't define this macro, the value of @code{BIGGEST_ALIGNMENT}
4543 is used for aligning trampolines.
4546 @defmac INITIALIZE_TRAMPOLINE (@var{addr}, @var{fnaddr}, @var{static_chain})
4547 A C statement to initialize the variable parts of a trampoline.
4548 @var{addr} is an RTX for the address of the trampoline; @var{fnaddr} is
4549 an RTX for the address of the nested function; @var{static_chain} is an
4550 RTX for the static chain value that should be passed to the function
4554 @defmac TRAMPOLINE_ADJUST_ADDRESS (@var{addr})
4555 A C statement that should perform any machine-specific adjustment in
4556 the address of the trampoline. Its argument contains the address that
4557 was passed to @code{INITIALIZE_TRAMPOLINE}. In case the address to be
4558 used for a function call should be different from the address in which
4559 the template was stored, the different address should be assigned to
4560 @var{addr}. If this macro is not defined, @var{addr} will be used for
4563 @cindex @code{TARGET_ASM_FUNCTION_EPILOGUE} and trampolines
4564 @cindex @code{TARGET_ASM_FUNCTION_PROLOGUE} and trampolines
4565 If this macro is not defined, by default the trampoline is allocated as
4566 a stack slot. This default is right for most machines. The exceptions
4567 are machines where it is impossible to execute instructions in the stack
4568 area. On such machines, you may have to implement a separate stack,
4569 using this macro in conjunction with @code{TARGET_ASM_FUNCTION_PROLOGUE}
4570 and @code{TARGET_ASM_FUNCTION_EPILOGUE}.
4572 @var{fp} points to a data structure, a @code{struct function}, which
4573 describes the compilation status of the immediate containing function of
4574 the function which the trampoline is for. The stack slot for the
4575 trampoline is in the stack frame of this containing function. Other
4576 allocation strategies probably must do something analogous with this
4580 Implementing trampolines is difficult on many machines because they have
4581 separate instruction and data caches. Writing into a stack location
4582 fails to clear the memory in the instruction cache, so when the program
4583 jumps to that location, it executes the old contents.
4585 Here are two possible solutions. One is to clear the relevant parts of
4586 the instruction cache whenever a trampoline is set up. The other is to
4587 make all trampolines identical, by having them jump to a standard
4588 subroutine. The former technique makes trampoline execution faster; the
4589 latter makes initialization faster.
4591 To clear the instruction cache when a trampoline is initialized, define
4592 the following macro.
4594 @defmac CLEAR_INSN_CACHE (@var{beg}, @var{end})
4595 If defined, expands to a C expression clearing the @emph{instruction
4596 cache} in the specified interval. The definition of this macro would
4597 typically be a series of @code{asm} statements. Both @var{beg} and
4598 @var{end} are both pointer expressions.
4601 To use a standard subroutine, define the following macro. In addition,
4602 you must make sure that the instructions in a trampoline fill an entire
4603 cache line with identical instructions, or else ensure that the
4604 beginning of the trampoline code is always aligned at the same point in
4605 its cache line. Look in @file{m68k.h} as a guide.
4607 @defmac TRANSFER_FROM_TRAMPOLINE
4608 Define this macro if trampolines need a special subroutine to do their
4609 work. The macro should expand to a series of @code{asm} statements
4610 which will be compiled with GCC@. They go in a library function named
4611 @code{__transfer_from_trampoline}.
4613 If you need to avoid executing the ordinary prologue code of a compiled
4614 C function when you jump to the subroutine, you can do so by placing a
4615 special label of your own in the assembler code. Use one @code{asm}
4616 statement to generate an assembler label, and another to make the label
4617 global. Then trampolines can use that label to jump directly to your
4618 special assembler code.
4622 @section Implicit Calls to Library Routines
4623 @cindex library subroutine names
4624 @cindex @file{libgcc.a}
4626 @c prevent bad page break with this line
4627 Here is an explanation of implicit calls to library routines.
4629 @defmac MULSI3_LIBCALL
4630 A C string constant giving the name of the function to call for
4631 multiplication of one signed full-word by another. If you do not
4632 define this macro, the default name is used, which is @code{__mulsi3},
4633 a function defined in @file{libgcc.a}.
4636 @defmac DIVSI3_LIBCALL
4637 A C string constant giving the name of the function to call for
4638 division of one signed full-word by another. If you do not define
4639 this macro, the default name is used, which is @code{__divsi3}, a
4640 function defined in @file{libgcc.a}.
4643 @defmac UDIVSI3_LIBCALL
4644 A C string constant giving the name of the function to call for
4645 division of one unsigned full-word by another. If you do not define
4646 this macro, the default name is used, which is @code{__udivsi3}, a
4647 function defined in @file{libgcc.a}.
4650 @defmac MODSI3_LIBCALL
4651 A C string constant giving the name of the function to call for the
4652 remainder in division of one signed full-word by another. If you do
4653 not define this macro, the default name is used, which is
4654 @code{__modsi3}, a function defined in @file{libgcc.a}.
4657 @defmac UMODSI3_LIBCALL
4658 A C string constant giving the name of the function to call for the
4659 remainder in division of one unsigned full-word by another. If you do
4660 not define this macro, the default name is used, which is
4661 @code{__umodsi3}, a function defined in @file{libgcc.a}.
4664 @defmac MULDI3_LIBCALL
4665 A C string constant giving the name of the function to call for
4666 multiplication of one signed double-word by another. If you do not
4667 define this macro, the default name is used, which is @code{__muldi3},
4668 a function defined in @file{libgcc.a}.
4671 @defmac DIVDI3_LIBCALL
4672 A C string constant giving the name of the function to call for
4673 division of one signed double-word by another. If you do not define
4674 this macro, the default name is used, which is @code{__divdi3}, a
4675 function defined in @file{libgcc.a}.
4678 @defmac UDIVDI3_LIBCALL
4679 A C string constant giving the name of the function to call for
4680 division of one unsigned full-word by another. If you do not define
4681 this macro, the default name is used, which is @code{__udivdi3}, a
4682 function defined in @file{libgcc.a}.
4685 @defmac MODDI3_LIBCALL
4686 A C string constant giving the name of the function to call for the
4687 remainder in division of one signed double-word by another. If you do
4688 not define this macro, the default name is used, which is
4689 @code{__moddi3}, a function defined in @file{libgcc.a}.
4692 @defmac UMODDI3_LIBCALL
4693 A C string constant giving the name of the function to call for the
4694 remainder in division of one unsigned full-word by another. If you do
4695 not define this macro, the default name is used, which is
4696 @code{__umoddi3}, a function defined in @file{libgcc.a}.
4699 @defmac DECLARE_LIBRARY_RENAMES
4700 This macro, if defined, should expand to a piece of C code that will get
4701 expanded when compiling functions for libgcc.a. It can be used to
4702 provide alternate names for gcc's internal library functions if there
4703 are ABI-mandated names that the compiler should provide.
4706 @defmac INIT_TARGET_OPTABS
4707 Define this macro as a C statement that declares additional library
4708 routines renames existing ones. @code{init_optabs} calls this macro after
4709 initializing all the normal library routines.
4712 @defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison})
4713 Define this macro as a C statement that returns nonzero if a call to
4714 the floating point comparison library function will return a boolean
4715 value that indicates the result of the comparison. It should return
4716 zero if one of gcc's own libgcc functions is called.
4718 Most ports don't need to define this macro.
4721 @cindex @code{EDOM}, implicit usage
4724 The value of @code{EDOM} on the target machine, as a C integer constant
4725 expression. If you don't define this macro, GCC does not attempt to
4726 deposit the value of @code{EDOM} into @code{errno} directly. Look in
4727 @file{/usr/include/errno.h} to find the value of @code{EDOM} on your
4730 If you do not define @code{TARGET_EDOM}, then compiled code reports
4731 domain errors by calling the library function and letting it report the
4732 error. If mathematical functions on your system use @code{matherr} when
4733 there is an error, then you should leave @code{TARGET_EDOM} undefined so
4734 that @code{matherr} is used normally.
4737 @cindex @code{errno}, implicit usage
4738 @defmac GEN_ERRNO_RTX
4739 Define this macro as a C expression to create an rtl expression that
4740 refers to the global ``variable'' @code{errno}. (On certain systems,
4741 @code{errno} may not actually be a variable.) If you don't define this
4742 macro, a reasonable default is used.
4745 @cindex @code{bcopy}, implicit usage
4746 @cindex @code{memcpy}, implicit usage
4747 @cindex @code{memmove}, implicit usage
4748 @cindex @code{bzero}, implicit usage
4749 @cindex @code{memset}, implicit usage
4750 @defmac TARGET_MEM_FUNCTIONS
4751 Define this macro if GCC should generate calls to the ISO C
4752 (and System V) library functions @code{memcpy}, @code{memmove} and
4753 @code{memset} rather than the BSD functions @code{bcopy} and @code{bzero}.
4756 @cindex C99 math functions, implicit usage
4757 @defmac TARGET_C99_FUNCTIONS
4758 When this macro is nonzero, GCC will implicitly optimize @code{sin} calls into
4759 @code{sinf} and similarly for other functions defined by C99 standard. The
4760 default is nonzero that should be proper value for most modern systems, however
4761 number of existing systems lacks support for these functions in the runtime so
4762 they needs this macro to be redefined to 0.
4765 @defmac LIBGCC_NEEDS_DOUBLE
4766 Define this macro if @code{float} arguments cannot be passed to library
4767 routines (so they must be converted to @code{double}). This macro
4768 affects both how library calls are generated and how the library
4769 routines in @file{libgcc.a} accept their arguments. It is useful on
4770 machines where floating and fixed point arguments are passed
4771 differently, such as the i860.
4774 @defmac NEXT_OBJC_RUNTIME
4775 Define this macro to generate code for Objective-C message sending using
4776 the calling convention of the NeXT system. This calling convention
4777 involves passing the object, the selector and the method arguments all
4778 at once to the method-lookup library function.
4780 The default calling convention passes just the object and the selector
4781 to the lookup function, which returns a pointer to the method.
4784 @node Addressing Modes
4785 @section Addressing Modes
4786 @cindex addressing modes
4788 @c prevent bad page break with this line
4789 This is about addressing modes.
4791 @defmac HAVE_PRE_INCREMENT
4792 @defmacx HAVE_PRE_DECREMENT
4793 @defmacx HAVE_POST_INCREMENT
4794 @defmacx HAVE_POST_DECREMENT
4795 A C expression that is nonzero if the machine supports pre-increment,
4796 pre-decrement, post-increment, or post-decrement addressing respectively.
4799 @defmac HAVE_PRE_MODIFY_DISP
4800 @defmacx HAVE_POST_MODIFY_DISP
4801 A C expression that is nonzero if the machine supports pre- or
4802 post-address side-effect generation involving constants other than
4803 the size of the memory operand.
4806 @defmac HAVE_PRE_MODIFY_REG
4807 @defmacx HAVE_POST_MODIFY_REG
4808 A C expression that is nonzero if the machine supports pre- or
4809 post-address side-effect generation involving a register displacement.
4812 @defmac CONSTANT_ADDRESS_P (@var{x})
4813 A C expression that is 1 if the RTX @var{x} is a constant which
4814 is a valid address. On most machines, this can be defined as
4815 @code{CONSTANT_P (@var{x})}, but a few machines are more restrictive
4816 in which constant addresses are supported.
4819 @defmac CONSTANT_P (@var{x})
4820 @code{CONSTANT_P}, which is defined by target-independent code,
4821 accepts integer-values expressions whose values are not explicitly
4822 known, such as @code{symbol_ref}, @code{label_ref}, and @code{high}
4823 expressions and @code{const} arithmetic expressions, in addition to
4824 @code{const_int} and @code{const_double} expressions.
4827 @defmac MAX_REGS_PER_ADDRESS
4828 A number, the maximum number of registers that can appear in a valid
4829 memory address. Note that it is up to you to specify a value equal to
4830 the maximum number that @code{GO_IF_LEGITIMATE_ADDRESS} would ever
4834 @defmac GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
4835 A C compound statement with a conditional @code{goto @var{label};}
4836 executed if @var{x} (an RTX) is a legitimate memory address on the
4837 target machine for a memory operand of mode @var{mode}.
4839 It usually pays to define several simpler macros to serve as
4840 subroutines for this one. Otherwise it may be too complicated to
4843 This macro must exist in two variants: a strict variant and a
4844 non-strict one. The strict variant is used in the reload pass. It
4845 must be defined so that any pseudo-register that has not been
4846 allocated a hard register is considered a memory reference. In
4847 contexts where some kind of register is required, a pseudo-register
4848 with no hard register must be rejected.
4850 The non-strict variant is used in other passes. It must be defined to
4851 accept all pseudo-registers in every context where some kind of
4852 register is required.
4854 @findex REG_OK_STRICT
4855 Compiler source files that want to use the strict variant of this
4856 macro define the macro @code{REG_OK_STRICT}. You should use an
4857 @code{#ifdef REG_OK_STRICT} conditional to define the strict variant
4858 in that case and the non-strict variant otherwise.
4860 Subroutines to check for acceptable registers for various purposes (one
4861 for base registers, one for index registers, and so on) are typically
4862 among the subroutines used to define @code{GO_IF_LEGITIMATE_ADDRESS}.
4863 Then only these subroutine macros need have two variants; the higher
4864 levels of macros may be the same whether strict or not.
4866 Normally, constant addresses which are the sum of a @code{symbol_ref}
4867 and an integer are stored inside a @code{const} RTX to mark them as
4868 constant. Therefore, there is no need to recognize such sums
4869 specifically as legitimate addresses. Normally you would simply
4870 recognize any @code{const} as legitimate.
4872 Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
4873 sums that are not marked with @code{const}. It assumes that a naked
4874 @code{plus} indicates indexing. If so, then you @emph{must} reject such
4875 naked constant sums as illegitimate addresses, so that none of them will
4876 be given to @code{PRINT_OPERAND_ADDRESS}.
4878 @cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation
4879 On some machines, whether a symbolic address is legitimate depends on
4880 the section that the address refers to. On these machines, define the
4881 target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information
4882 into the @code{symbol_ref}, and then check for it here. When you see a
4883 @code{const}, you will have to look inside it to find the
4884 @code{symbol_ref} in order to determine the section. @xref{Assembler
4888 @defmac REG_OK_FOR_BASE_P (@var{x})
4889 A C expression that is nonzero if @var{x} (assumed to be a @code{reg}
4890 RTX) is valid for use as a base register. For hard registers, it
4891 should always accept those which the hardware permits and reject the
4892 others. Whether the macro accepts or rejects pseudo registers must be
4893 controlled by @code{REG_OK_STRICT} as described above. This usually
4894 requires two variant definitions, of which @code{REG_OK_STRICT}
4895 controls the one actually used.
4898 @defmac REG_MODE_OK_FOR_BASE_P (@var{x}, @var{mode})
4899 A C expression that is just like @code{REG_OK_FOR_BASE_P}, except that
4900 that expression may examine the mode of the memory reference in
4901 @var{mode}. You should define this macro if the mode of the memory
4902 reference affects whether a register may be used as a base register. If
4903 you define this macro, the compiler will use it instead of
4904 @code{REG_OK_FOR_BASE_P}.
4907 @defmac REG_OK_FOR_INDEX_P (@var{x})
4908 A C expression that is nonzero if @var{x} (assumed to be a @code{reg}
4909 RTX) is valid for use as an index register.
4911 The difference between an index register and a base register is that
4912 the index register may be scaled. If an address involves the sum of
4913 two registers, neither one of them scaled, then either one may be
4914 labeled the ``base'' and the other the ``index''; but whichever
4915 labeling is used must fit the machine's constraints of which registers
4916 may serve in each capacity. The compiler will try both labelings,
4917 looking for one that is valid, and will reload one or both registers
4918 only if neither labeling works.
4921 @defmac FIND_BASE_TERM (@var{x})
4922 A C expression to determine the base term of address @var{x}.
4923 This macro is used in only one place: `find_base_term' in alias.c.
4925 It is always safe for this macro to not be defined. It exists so
4926 that alias analysis can understand machine-dependent addresses.
4928 The typical use of this macro is to handle addresses containing
4929 a label_ref or symbol_ref within an UNSPEC@.
4932 @defmac LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win})
4933 A C compound statement that attempts to replace @var{x} with a valid
4934 memory address for an operand of mode @var{mode}. @var{win} will be a
4935 C statement label elsewhere in the code; the macro definition may use
4938 GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win});
4942 to avoid further processing if the address has become legitimate.
4944 @findex break_out_memory_refs
4945 @var{x} will always be the result of a call to @code{break_out_memory_refs},
4946 and @var{oldx} will be the operand that was given to that function to produce
4949 The code generated by this macro should not alter the substructure of
4950 @var{x}. If it transforms @var{x} into a more legitimate form, it
4951 should assign @var{x} (which will always be a C variable) a new value.
4953 It is not necessary for this macro to come up with a legitimate
4954 address. The compiler has standard ways of doing so in all cases. In
4955 fact, it is safe for this macro to do nothing. But often a
4956 machine-dependent strategy can generate better code.
4959 @defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win})
4960 A C compound statement that attempts to replace @var{x}, which is an address
4961 that needs reloading, with a valid memory address for an operand of mode
4962 @var{mode}. @var{win} will be a C statement label elsewhere in the code.
4963 It is not necessary to define this macro, but it might be useful for
4964 performance reasons.
4966 For example, on the i386, it is sometimes possible to use a single
4967 reload register instead of two by reloading a sum of two pseudo
4968 registers into a register. On the other hand, for number of RISC
4969 processors offsets are limited so that often an intermediate address
4970 needs to be generated in order to address a stack slot. By defining
4971 @code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses
4972 generated for adjacent some stack slots can be made identical, and thus
4975 @emph{Note}: This macro should be used with caution. It is necessary
4976 to know something of how reload works in order to effectively use this,
4977 and it is quite easy to produce macros that build in too much knowledge
4978 of reload internals.
4980 @emph{Note}: This macro must be able to reload an address created by a
4981 previous invocation of this macro. If it fails to handle such addresses
4982 then the compiler may generate incorrect code or abort.
4985 The macro definition should use @code{push_reload} to indicate parts that
4986 need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually
4987 suitable to be passed unaltered to @code{push_reload}.
4989 The code generated by this macro must not alter the substructure of
4990 @var{x}. If it transforms @var{x} into a more legitimate form, it
4991 should assign @var{x} (which will always be a C variable) a new value.
4992 This also applies to parts that you change indirectly by calling
4995 @findex strict_memory_address_p
4996 The macro definition may use @code{strict_memory_address_p} to test if
4997 the address has become legitimate.
5000 If you want to change only a part of @var{x}, one standard way of doing
5001 this is to use @code{copy_rtx}. Note, however, that is unshares only a
5002 single level of rtl. Thus, if the part to be changed is not at the
5003 top level, you'll need to replace first the top level.
5004 It is not necessary for this macro to come up with a legitimate
5005 address; but often a machine-dependent strategy can generate better code.
5008 @defmac GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
5009 A C statement or compound statement with a conditional @code{goto
5010 @var{label};} executed if memory address @var{x} (an RTX) can have
5011 different meanings depending on the machine mode of the memory
5012 reference it is used for or if the address is valid for some modes
5015 Autoincrement and autodecrement addresses typically have mode-dependent
5016 effects because the amount of the increment or decrement is the size
5017 of the operand being addressed. Some machines have other mode-dependent
5018 addresses. Many RISC machines have no mode-dependent addresses.
5020 You may assume that @var{addr} is a valid address for the machine.
5023 @defmac LEGITIMATE_CONSTANT_P (@var{x})
5024 A C expression that is nonzero if @var{x} is a legitimate constant for
5025 an immediate operand on the target machine. You can assume that
5026 @var{x} satisfies @code{CONSTANT_P}, so you need not check this. In fact,
5027 @samp{1} is a suitable definition for this macro on machines where
5028 anything @code{CONSTANT_P} is valid.
5031 @node Condition Code
5032 @section Condition Code Status
5033 @cindex condition code status
5035 @c prevent bad page break with this line
5036 This describes the condition code status.
5039 The file @file{conditions.h} defines a variable @code{cc_status} to
5040 describe how the condition code was computed (in case the interpretation of
5041 the condition code depends on the instruction that it was set by). This
5042 variable contains the RTL expressions on which the condition code is
5043 currently based, and several standard flags.
5045 Sometimes additional machine-specific flags must be defined in the machine
5046 description header file. It can also add additional machine-specific
5047 information by defining @code{CC_STATUS_MDEP}.
5049 @defmac CC_STATUS_MDEP
5050 C code for a data type which is used for declaring the @code{mdep}
5051 component of @code{cc_status}. It defaults to @code{int}.
5053 This macro is not used on machines that do not use @code{cc0}.
5056 @defmac CC_STATUS_MDEP_INIT
5057 A C expression to initialize the @code{mdep} field to ``empty''.
5058 The default definition does nothing, since most machines don't use
5059 the field anyway. If you want to use the field, you should probably
5060 define this macro to initialize it.
5062 This macro is not used on machines that do not use @code{cc0}.
5065 @defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn})
5066 A C compound statement to set the components of @code{cc_status}
5067 appropriately for an insn @var{insn} whose body is @var{exp}. It is
5068 this macro's responsibility to recognize insns that set the condition
5069 code as a byproduct of other activity as well as those that explicitly
5072 This macro is not used on machines that do not use @code{cc0}.
5074 If there are insns that do not set the condition code but do alter
5075 other machine registers, this macro must check to see whether they
5076 invalidate the expressions that the condition code is recorded as
5077 reflecting. For example, on the 68000, insns that store in address
5078 registers do not set the condition code, which means that usually
5079 @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
5080 insns. But suppose that the previous insn set the condition code
5081 based on location @samp{a4@@(102)} and the current insn stores a new
5082 value in @samp{a4}. Although the condition code is not changed by
5083 this, it will no longer be true that it reflects the contents of
5084 @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
5085 @code{cc_status} in this case to say that nothing is known about the
5086 condition code value.
5088 The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
5089 with the results of peephole optimization: insns whose patterns are
5090 @code{parallel} RTXs containing various @code{reg}, @code{mem} or
5091 constants which are just the operands. The RTL structure of these
5092 insns is not sufficient to indicate what the insns actually do. What
5093 @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
5094 @code{CC_STATUS_INIT}.
5096 A possible definition of @code{NOTICE_UPDATE_CC} is to call a function
5097 that looks at an attribute (@pxref{Insn Attributes}) named, for example,
5098 @samp{cc}. This avoids having detailed information about patterns in
5099 two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}.
5102 @defmac EXTRA_CC_MODES
5103 Condition codes are represented in registers by machine modes of class
5104 @code{MODE_CC}. By default, there is just one mode, @code{CCmode}, with
5105 this class. If you need more such modes, create a file named
5106 @file{@var{machine}-modes.def} in your @file{config/@var{machine}}
5107 directory (@pxref{Back End, , Anatomy of a Target Back End}), containing
5108 a list of these modes. Each entry in the list should be a call to the
5109 macro @code{CC}. This macro takes one argument, which is the name of
5110 the mode: it should begin with @samp{CC}. Do not put quotation marks
5111 around the name, or include the trailing @samp{mode}; these are
5112 automatically added. There should not be anything else in the file
5115 A sample @file{@var{machine}-modes.def} file might look like this:
5118 CC (CC_NOOV) /* @r{Comparison only valid if there was no overflow.} */
5119 CC (CCFP) /* @r{Floating point comparison that cannot trap.} */
5120 CC (CCFPE) /* @r{Floating point comparison that may trap.} */
5123 When you create this file, the macro @code{EXTRA_CC_MODES} is
5124 automatically defined by @command{configure}, with value @samp{1}.
5127 @defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y})
5128 Returns a mode from class @code{MODE_CC} to be used when comparison
5129 operation code @var{op} is applied to rtx @var{x} and @var{y}. For
5130 example, on the SPARC, @code{SELECT_CC_MODE} is defined as (see
5131 @pxref{Jump Patterns} for a description of the reason for this
5135 #define SELECT_CC_MODE(OP,X,Y) \
5136 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \
5137 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \
5138 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \
5139 || GET_CODE (X) == NEG) \
5140 ? CC_NOOVmode : CCmode))
5143 You need not define this macro if @code{EXTRA_CC_MODES} is not defined.
5146 @defmac CANONICALIZE_COMPARISON (@var{code}, @var{op0}, @var{op1})
5147 On some machines not all possible comparisons are defined, but you can
5148 convert an invalid comparison into a valid one. For example, the Alpha
5149 does not have a @code{GT} comparison, but you can use an @code{LT}
5150 comparison instead and swap the order of the operands.
5152 On such machines, define this macro to be a C statement to do any
5153 required conversions. @var{code} is the initial comparison code
5154 and @var{op0} and @var{op1} are the left and right operands of the
5155 comparison, respectively. You should modify @var{code}, @var{op0}, and
5156 @var{op1} as required.
5158 GCC will not assume that the comparison resulting from this macro is
5159 valid but will see if the resulting insn matches a pattern in the
5162 You need not define this macro if it would never change the comparison
5166 @defmac REVERSIBLE_CC_MODE (@var{mode})
5167 A C expression whose value is one if it is always safe to reverse a
5168 comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE}
5169 can ever return @var{mode} for a floating-point inequality comparison,
5170 then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero.
5172 You need not define this macro if it would always returns zero or if the
5173 floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}.
5174 For example, here is the definition used on the SPARC, where floating-point
5175 inequality comparisons are always given @code{CCFPEmode}:
5178 #define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode)
5182 @defmac REVERSE_CONDITION (@var{code}, @var{mode})
5183 A C expression whose value is reversed condition code of the @var{code} for
5184 comparison done in CC_MODE @var{mode}. The macro is used only in case
5185 @code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case
5186 machine has some non-standard way how to reverse certain conditionals. For
5187 instance in case all floating point conditions are non-trapping, compiler may
5188 freely convert unordered compares to ordered one. Then definition may look
5192 #define REVERSE_CONDITION(CODE, MODE) \
5193 ((MODE) != CCFPmode ? reverse_condition (CODE) \
5194 : reverse_condition_maybe_unordered (CODE))
5198 @defmac REVERSE_CONDEXEC_PREDICATES_P (@var{code1}, @var{code2})
5199 A C expression that returns true if the conditional execution predicate
5200 @var{code1} is the inverse of @var{code2} and vice versa. Define this to
5201 return 0 if the target has conditional execution predicates that cannot be
5202 reversed safely. If no expansion is specified, this macro is defined as
5206 #define REVERSE_CONDEXEC_PREDICATES_P (x, y) \
5207 ((x) == reverse_condition (y))
5212 @section Describing Relative Costs of Operations
5213 @cindex costs of instructions
5214 @cindex relative costs
5215 @cindex speed of instructions
5217 These macros let you describe the relative speed of various operations
5218 on the target machine.
5220 @defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to})
5221 A C expression for the cost of moving data of mode @var{mode} from a
5222 register in class @var{from} to one in class @var{to}. The classes are
5223 expressed using the enumeration values such as @code{GENERAL_REGS}. A
5224 value of 2 is the default; other values are interpreted relative to
5227 It is not required that the cost always equal 2 when @var{from} is the
5228 same as @var{to}; on some machines it is expensive to move between
5229 registers if they are not general registers.
5231 If reload sees an insn consisting of a single @code{set} between two
5232 hard registers, and if @code{REGISTER_MOVE_COST} applied to their
5233 classes returns a value of 2, reload does not check to ensure that the
5234 constraints of the insn are met. Setting a cost of other than 2 will
5235 allow reload to verify that the constraints are met. You should do this
5236 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
5239 @defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in})
5240 A C expression for the cost of moving data of mode @var{mode} between a
5241 register of class @var{class} and memory; @var{in} is zero if the value
5242 is to be written to memory, nonzero if it is to be read in. This cost
5243 is relative to those in @code{REGISTER_MOVE_COST}. If moving between
5244 registers and memory is more expensive than between two registers, you
5245 should define this macro to express the relative cost.
5247 If you do not define this macro, GCC uses a default cost of 4 plus
5248 the cost of copying via a secondary reload register, if one is
5249 needed. If your machine requires a secondary reload register to copy
5250 between memory and a register of @var{class} but the reload mechanism is
5251 more complex than copying via an intermediate, define this macro to
5252 reflect the actual cost of the move.
5254 GCC defines the function @code{memory_move_secondary_cost} if
5255 secondary reloads are needed. It computes the costs due to copying via
5256 a secondary register. If your machine copies from memory using a
5257 secondary register in the conventional way but the default base value of
5258 4 is not correct for your machine, define this macro to add some other
5259 value to the result of that function. The arguments to that function
5260 are the same as to this macro.
5264 A C expression for the cost of a branch instruction. A value of 1 is
5265 the default; other values are interpreted relative to that.
5268 Here are additional macros which do not specify precise relative costs,
5269 but only that certain actions are more expensive than GCC would
5272 @defmac SLOW_BYTE_ACCESS
5273 Define this macro as a C expression which is nonzero if accessing less
5274 than a word of memory (i.e.@: a @code{char} or a @code{short}) is no
5275 faster than accessing a word of memory, i.e., if such access
5276 require more than one instruction or if there is no difference in cost
5277 between byte and (aligned) word loads.
5279 When this macro is not defined, the compiler will access a field by
5280 finding the smallest containing object; when it is defined, a fullword
5281 load will be used if alignment permits. Unless bytes accesses are
5282 faster than word accesses, using word accesses is preferable since it
5283 may eliminate subsequent memory access if subsequent accesses occur to
5284 other fields in the same word of the structure, but to different bytes.
5287 @defmac SLOW_UNALIGNED_ACCESS (@var{mode}, @var{alignment})
5288 Define this macro to be the value 1 if memory accesses described by the
5289 @var{mode} and @var{alignment} parameters have a cost many times greater
5290 than aligned accesses, for example if they are emulated in a trap
5293 When this macro is nonzero, the compiler will act as if
5294 @code{STRICT_ALIGNMENT} were nonzero when generating code for block
5295 moves. This can cause significantly more instructions to be produced.
5296 Therefore, do not set this macro nonzero if unaligned accesses only add a
5297 cycle or two to the time for a memory access.
5299 If the value of this macro is always zero, it need not be defined. If
5300 this macro is defined, it should produce a nonzero value when
5301 @code{STRICT_ALIGNMENT} is nonzero.
5305 The threshold of number of scalar memory-to-memory move insns, @emph{below}
5306 which a sequence of insns should be generated instead of a
5307 string move insn or a library call. Increasing the value will always
5308 make code faster, but eventually incurs high cost in increased code size.
5310 Note that on machines where the corresponding move insn is a
5311 @code{define_expand} that emits a sequence of insns, this macro counts
5312 the number of such sequences.
5314 If you don't define this, a reasonable default is used.
5317 @defmac MOVE_BY_PIECES_P (@var{size}, @var{alignment})
5318 A C expression used to determine whether @code{move_by_pieces} will be used to
5319 copy a chunk of memory, or whether some other block move mechanism
5320 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
5321 than @code{MOVE_RATIO}.
5324 @defmac MOVE_MAX_PIECES
5325 A C expression used by @code{move_by_pieces} to determine the largest unit
5326 a load or store used to copy memory is. Defaults to @code{MOVE_MAX}.
5330 The threshold of number of scalar move insns, @emph{below} which a sequence
5331 of insns should be generated to clear memory instead of a string clear insn
5332 or a library call. Increasing the value will always make code faster, but
5333 eventually incurs high cost in increased code size.
5335 If you don't define this, a reasonable default is used.
5338 @defmac CLEAR_BY_PIECES_P (@var{size}, @var{alignment})
5339 A C expression used to determine whether @code{clear_by_pieces} will be used
5340 to clear a chunk of memory, or whether some other block clear mechanism
5341 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
5342 than @code{CLEAR_RATIO}.
5345 @defmac STORE_BY_PIECES_P (@var{size}, @var{alignment})
5346 A C expression used to determine whether @code{store_by_pieces} will be
5347 used to set a chunk of memory to a constant value, or whether some other
5348 mechanism will be used. Used by @code{__builtin_memset} when storing
5349 values other than constant zero and by @code{__builtin_strcpy} when
5350 when called with a constant source string.
5351 Defaults to @code{MOVE_BY_PIECES_P}.
5354 @defmac USE_LOAD_POST_INCREMENT (@var{mode})
5355 A C expression used to determine whether a load postincrement is a good
5356 thing to use for a given mode. Defaults to the value of
5357 @code{HAVE_POST_INCREMENT}.
5360 @defmac USE_LOAD_POST_DECREMENT (@var{mode})
5361 A C expression used to determine whether a load postdecrement is a good
5362 thing to use for a given mode. Defaults to the value of
5363 @code{HAVE_POST_DECREMENT}.
5366 @defmac USE_LOAD_PRE_INCREMENT (@var{mode})
5367 A C expression used to determine whether a load preincrement is a good
5368 thing to use for a given mode. Defaults to the value of
5369 @code{HAVE_PRE_INCREMENT}.
5372 @defmac USE_LOAD_PRE_DECREMENT (@var{mode})
5373 A C expression used to determine whether a load predecrement is a good
5374 thing to use for a given mode. Defaults to the value of
5375 @code{HAVE_PRE_DECREMENT}.
5378 @defmac USE_STORE_POST_INCREMENT (@var{mode})
5379 A C expression used to determine whether a store postincrement is a good
5380 thing to use for a given mode. Defaults to the value of
5381 @code{HAVE_POST_INCREMENT}.
5384 @defmac USE_STORE_POST_DECREMENT (@var{mode})
5385 A C expression used to determine whether a store postdecrement is a good
5386 thing to use for a given mode. Defaults to the value of
5387 @code{HAVE_POST_DECREMENT}.
5390 @defmac USE_STORE_PRE_INCREMENT (@var{mode})
5391 This macro is used to determine whether a store preincrement is a good
5392 thing to use for a given mode. Defaults to the value of
5393 @code{HAVE_PRE_INCREMENT}.
5396 @defmac USE_STORE_PRE_DECREMENT (@var{mode})
5397 This macro is used to determine whether a store predecrement is a good
5398 thing to use for a given mode. Defaults to the value of
5399 @code{HAVE_PRE_DECREMENT}.
5402 @defmac NO_FUNCTION_CSE
5403 Define this macro if it is as good or better to call a constant
5404 function address than to call an address kept in a register.
5407 @defmac NO_RECURSIVE_FUNCTION_CSE
5408 Define this macro if it is as good or better for a function to call
5409 itself with an explicit address than to call an address kept in a
5413 @defmac RANGE_TEST_NON_SHORT_CIRCUIT
5414 Define this macro if a non-short-circuit operation produced by
5415 @samp{fold_range_test ()} is optimal. This macro defaults to true if
5416 @code{BRANCH_COST} is greater than or equal to the value 2.
5419 @deftypefn {Target Hook} bool TARGET_RTX_COSTS (rtx @var{x}, int @var{code}, int @var{outer_code}, int *@var{total})
5420 This target hook describes the relative costs of RTL expressions.
5422 The cost may depend on the precise form of the expression, which is
5423 available for examination in @var{x}, and the rtx code of the expression
5424 in which it is contained, found in @var{outer_code}. @var{code} is the
5425 expression code---redundant, since it can be obtained with
5426 @code{GET_CODE (@var{x})}.
5428 In implementing this hook, you can use the construct
5429 @code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast
5432 On entry to the hook, @code{*@var{total}} contains a default estimate
5433 for the cost of the expression. The hook should modify this value as
5436 The hook returns true when all subexpressions of @var{x} have been
5437 processed, and false when @code{rtx_cost} should recurse.
5440 @deftypefn {Target Hook} int TARGET_ADDRESS_COST (rtx @var{address})
5441 This hook computes the cost of an addressing mode that contains
5442 @var{address}. If not defined, the cost is computed from
5443 the @var{address} expression and the @code{TARGET_RTX_COST} hook.
5445 For most CISC machines, the default cost is a good approximation of the
5446 true cost of the addressing mode. However, on RISC machines, all
5447 instructions normally have the same length and execution time. Hence
5448 all addresses will have equal costs.
5450 In cases where more than one form of an address is known, the form with
5451 the lowest cost will be used. If multiple forms have the same, lowest,
5452 cost, the one that is the most complex will be used.
5454 For example, suppose an address that is equal to the sum of a register
5455 and a constant is used twice in the same basic block. When this macro
5456 is not defined, the address will be computed in a register and memory
5457 references will be indirect through that register. On machines where
5458 the cost of the addressing mode containing the sum is no higher than
5459 that of a simple indirect reference, this will produce an additional
5460 instruction and possibly require an additional register. Proper
5461 specification of this macro eliminates this overhead for such machines.
5463 This hook is never called with an invalid address.
5465 On machines where an address involving more than one register is as
5466 cheap as an address computation involving only one register, defining
5467 @code{TARGET_ADDRESS_COST} to reflect this can cause two registers to
5468 be live over a region of code where only one would have been if
5469 @code{TARGET_ADDRESS_COST} were not defined in that manner. This effect
5470 should be considered in the definition of this macro. Equivalent costs
5471 should probably only be given to addresses with different numbers of
5472 registers on machines with lots of registers.
5476 @section Adjusting the Instruction Scheduler
5478 The instruction scheduler may need a fair amount of machine-specific
5479 adjustment in order to produce good code. GCC provides several target
5480 hooks for this purpose. It is usually enough to define just a few of
5481 them: try the first ones in this list first.
5483 @deftypefn {Target Hook} int TARGET_SCHED_ISSUE_RATE (void)
5484 This hook returns the maximum number of instructions that can ever
5485 issue at the same time on the target machine. The default is one.
5486 Although the insn scheduler can define itself the possibility of issue
5487 an insn on the same cycle, the value can serve as an additional
5488 constraint to issue insns on the same simulated processor cycle (see
5489 hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}).
5490 This value must be constant over the entire compilation. If you need
5491 it to vary depending on what the instructions are, you must use
5492 @samp{TARGET_SCHED_VARIABLE_ISSUE}.
5494 For the automaton based pipeline interface, you could define this hook
5495 to return the value of the macro @code{MAX_DFA_ISSUE_RATE}.
5498 @deftypefn {Target Hook} int TARGET_SCHED_VARIABLE_ISSUE (FILE *@var{file}, int @var{verbose}, rtx @var{insn}, int @var{more})
5499 This hook is executed by the scheduler after it has scheduled an insn
5500 from the ready list. It should return the number of insns which can
5501 still be issued in the current cycle. The default is
5502 @samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and
5503 @code{USE}, which normally are not counted against the issue rate.
5504 You should define this hook if some insns take more machine resources
5505 than others, so that fewer insns can follow them in the same cycle.
5506 @var{file} is either a null pointer, or a stdio stream to write any
5507 debug output to. @var{verbose} is the verbose level provided by
5508 @option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that
5512 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_COST (rtx @var{insn}, rtx @var{link}, rtx @var{dep_insn}, int @var{cost})
5513 This function corrects the value of @var{cost} based on the
5514 relationship between @var{insn} and @var{dep_insn} through the
5515 dependence @var{link}. It should return the new value. The default
5516 is to make no adjustment to @var{cost}. This can be used for example
5517 to specify to the scheduler using the traditional pipeline description
5518 that an output- or anti-dependence does not incur the same cost as a
5519 data-dependence. If the scheduler using the automaton based pipeline
5520 description, the cost of anti-dependence is zero and the cost of
5521 output-dependence is maximum of one and the difference of latency
5522 times of the first and the second insns. If these values are not
5523 acceptable, you could use the hook to modify them too. See also
5524 @pxref{Automaton pipeline description}.
5527 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_PRIORITY (rtx @var{insn}, int @var{priority})
5528 This hook adjusts the integer scheduling priority @var{priority} of
5529 @var{insn}. It should return the new priority. Reduce the priority to
5530 execute @var{insn} earlier, increase the priority to execute @var{insn}
5531 later. Do not define this hook if you do not need to adjust the
5532 scheduling priorities of insns.
5535 @deftypefn {Target Hook} int TARGET_SCHED_REORDER (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_readyp}, int @var{clock})
5536 This hook is executed by the scheduler after it has scheduled the ready
5537 list, to allow the machine description to reorder it (for example to
5538 combine two small instructions together on @samp{VLIW} machines).
5539 @var{file} is either a null pointer, or a stdio stream to write any
5540 debug output to. @var{verbose} is the verbose level provided by
5541 @option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready
5542 list of instructions that are ready to be scheduled. @var{n_readyp} is
5543 a pointer to the number of elements in the ready list. The scheduler
5544 reads the ready list in reverse order, starting with
5545 @var{ready}[@var{*n_readyp}-1] and going to @var{ready}[0]. @var{clock}
5546 is the timer tick of the scheduler. You may modify the ready list and
5547 the number of ready insns. The return value is the number of insns that
5548 can issue this cycle; normally this is just @code{issue_rate}. See also
5549 @samp{TARGET_SCHED_REORDER2}.
5552 @deftypefn {Target Hook} int TARGET_SCHED_REORDER2 (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_ready}, @var{clock})
5553 Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That
5554 function is called whenever the scheduler starts a new cycle. This one
5555 is called once per iteration over a cycle, immediately after
5556 @samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and
5557 return the number of insns to be scheduled in the same cycle. Defining
5558 this hook can be useful if there are frequent situations where
5559 scheduling one insn causes other insns to become ready in the same
5560 cycle. These other insns can then be taken into account properly.
5563 @deftypefn {Target Hook} void TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK (rtx @var{head}, rtx @var{tail})
5564 This hook is called after evaluation forward dependencies of insns in
5565 chain given by two parameter values (@var{head} and @var{tail}
5566 correspondingly) but before insns scheduling of the insn chain. For
5567 example, it can be used for better insn classification if it requires
5568 analysis of dependencies. This hook can use backward and forward
5569 dependencies of the insn scheduler because they are already
5573 @deftypefn {Target Hook} void TARGET_SCHED_INIT (FILE *@var{file}, int @var{verbose}, int @var{max_ready})
5574 This hook is executed by the scheduler at the beginning of each block of
5575 instructions that are to be scheduled. @var{file} is either a null
5576 pointer, or a stdio stream to write any debug output to. @var{verbose}
5577 is the verbose level provided by @option{-fsched-verbose-@var{n}}.
5578 @var{max_ready} is the maximum number of insns in the current scheduling
5579 region that can be live at the same time. This can be used to allocate
5580 scratch space if it is needed, e.g. by @samp{TARGET_SCHED_REORDER}.
5583 @deftypefn {Target Hook} void TARGET_SCHED_FINISH (FILE *@var{file}, int @var{verbose})
5584 This hook is executed by the scheduler at the end of each block of
5585 instructions that are to be scheduled. It can be used to perform
5586 cleanup of any actions done by the other scheduling hooks. @var{file}
5587 is either a null pointer, or a stdio stream to write any debug output
5588 to. @var{verbose} is the verbose level provided by
5589 @option{-fsched-verbose-@var{n}}.
5592 @deftypefn {Target Hook} int TARGET_SCHED_USE_DFA_PIPELINE_INTERFACE (void)
5593 This hook is called many times during insn scheduling. If the hook
5594 returns nonzero, the automaton based pipeline description is used for
5595 insn scheduling. Otherwise the traditional pipeline description is
5596 used. The default is usage of the traditional pipeline description.
5598 You should also remember that to simplify the insn scheduler sources
5599 an empty traditional pipeline description interface is generated even
5600 if there is no a traditional pipeline description in the @file{.md}
5601 file. The same is true for the automaton based pipeline description.
5602 That means that you should be accurate in defining the hook.
5605 @deftypefn {Target Hook} int TARGET_SCHED_DFA_PRE_CYCLE_INSN (void)
5606 The hook returns an RTL insn. The automaton state used in the
5607 pipeline hazard recognizer is changed as if the insn were scheduled
5608 when the new simulated processor cycle starts. Usage of the hook may
5609 simplify the automaton pipeline description for some @acronym{VLIW}
5610 processors. If the hook is defined, it is used only for the automaton
5611 based pipeline description. The default is not to change the state
5612 when the new simulated processor cycle starts.
5615 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN (void)
5616 The hook can be used to initialize data used by the previous hook.
5619 @deftypefn {Target Hook} int TARGET_SCHED_DFA_POST_CYCLE_INSN (void)
5620 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
5621 to changed the state as if the insn were scheduled when the new
5622 simulated processor cycle finishes.
5625 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN (void)
5626 The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but
5627 used to initialize data used by the previous hook.
5630 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD (void)
5631 This hook controls better choosing an insn from the ready insn queue
5632 for the @acronym{DFA}-based insn scheduler. Usually the scheduler
5633 chooses the first insn from the queue. If the hook returns a positive
5634 value, an additional scheduler code tries all permutations of
5635 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()}
5636 subsequent ready insns to choose an insn whose issue will result in
5637 maximal number of issued insns on the same cycle. For the
5638 @acronym{VLIW} processor, the code could actually solve the problem of
5639 packing simple insns into the @acronym{VLIW} insn. Of course, if the
5640 rules of @acronym{VLIW} packing are described in the automaton.
5642 This code also could be used for superscalar @acronym{RISC}
5643 processors. Let us consider a superscalar @acronym{RISC} processor
5644 with 3 pipelines. Some insns can be executed in pipelines @var{A} or
5645 @var{B}, some insns can be executed only in pipelines @var{B} or
5646 @var{C}, and one insn can be executed in pipeline @var{B}. The
5647 processor may issue the 1st insn into @var{A} and the 2nd one into
5648 @var{B}. In this case, the 3rd insn will wait for freeing @var{B}
5649 until the next cycle. If the scheduler issues the 3rd insn the first,
5650 the processor could issue all 3 insns per cycle.
5652 Actually this code demonstrates advantages of the automaton based
5653 pipeline hazard recognizer. We try quickly and easy many insn
5654 schedules to choose the best one.
5656 The default is no multipass scheduling.
5659 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD (rtx)
5661 This hook controls what insns from the ready insn queue will be
5662 considered for the multipass insn scheduling. If the hook returns
5663 zero for insn passed as the parameter, the insn will be not chosen to
5666 The default is that any ready insns can be chosen to be issued.
5669 @deftypefn {Target Hook} int TARGET_SCHED_DFA_NEW_CYCLE (FILE *, int, rtx, int, int, int *)
5671 This hook is called by the insn scheduler before issuing insn passed
5672 as the third parameter on given cycle. If the hook returns nonzero,
5673 the insn is not issued on given processors cycle. Instead of that,
5674 the processor cycle is advanced. If the value passed through the last
5675 parameter is zero, the insn ready queue is not sorted on the new cycle
5676 start as usually. The first parameter passes file for debugging
5677 output. The second one passes the scheduler verbose level of the
5678 debugging output. The forth and the fifth parameter values are
5679 correspondingly processor cycle on which the previous insn has been
5680 issued and the current processor cycle.
5683 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_BUBBLES (void)
5684 The @acronym{DFA}-based scheduler could take the insertion of nop
5685 operations for better insn scheduling into account. It can be done
5686 only if the multi-pass insn scheduling works (see hook
5687 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD}).
5689 Let us consider a @acronym{VLIW} processor insn with 3 slots. Each
5690 insn can be placed only in one of the three slots. We have 3 ready
5691 insns @var{A}, @var{B}, and @var{C}. @var{A} and @var{C} can be
5692 placed only in the 1st slot, @var{B} can be placed only in the 3rd
5693 slot. We described the automaton which does not permit empty slot
5694 gaps between insns (usually such description is simpler). Without
5695 this code the scheduler would place each insn in 3 separate
5696 @acronym{VLIW} insns. If the scheduler places a nop insn into the 2nd
5697 slot, it could place the 3 insns into 2 @acronym{VLIW} insns. What is
5698 the nop insn is returned by hook @samp{TARGET_SCHED_DFA_BUBBLE}. Hook
5699 @samp{TARGET_SCHED_INIT_DFA_BUBBLES} can be used to initialize or
5700 create the nop insns.
5702 You should remember that the scheduler does not insert the nop insns.
5703 It is not wise because of the following optimizations. The scheduler
5704 only considers such possibility to improve the result schedule. The
5705 nop insns should be inserted lately, e.g. on the final phase.
5708 @deftypefn {Target Hook} rtx TARGET_SCHED_DFA_BUBBLE (int @var{index})
5709 This hook @samp{FIRST_CYCLE_MULTIPASS_SCHEDULING} is used to insert
5710 nop operations for better insn scheduling when @acronym{DFA}-based
5711 scheduler makes multipass insn scheduling (see also description of
5712 hook @samp{TARGET_SCHED_INIT_DFA_BUBBLES}). This hook
5713 returns a nop insn with given @var{index}. The indexes start with
5714 zero. The hook should return @code{NULL} if there are no more nop
5715 insns with indexes greater than given index.
5718 Macros in the following table are generated by the program
5719 @file{genattr} and can be useful for writing the hooks.
5721 @defmac TRADITIONAL_PIPELINE_INTERFACE
5722 The macro definition is generated if there is a traditional pipeline
5723 description in @file{.md} file. You should also remember that to
5724 simplify the insn scheduler sources an empty traditional pipeline
5725 description interface is generated even if there is no a traditional
5726 pipeline description in the @file{.md} file. The macro can be used to
5727 distinguish the two types of the traditional interface.
5730 @defmac DFA_PIPELINE_INTERFACE
5731 The macro definition is generated if there is an automaton pipeline
5732 description in @file{.md} file. You should also remember that to
5733 simplify the insn scheduler sources an empty automaton pipeline
5734 description interface is generated even if there is no an automaton
5735 pipeline description in the @file{.md} file. The macro can be used to
5736 distinguish the two types of the automaton interface.
5739 @defmac MAX_DFA_ISSUE_RATE
5740 The macro definition is generated in the automaton based pipeline
5741 description interface. Its value is calculated from the automaton
5742 based pipeline description and is equal to maximal number of all insns
5743 described in constructions @samp{define_insn_reservation} which can be
5744 issued on the same processor cycle.
5748 @section Dividing the Output into Sections (Texts, Data, @dots{})
5749 @c the above section title is WAY too long. maybe cut the part between
5750 @c the (...)? --mew 10feb93
5752 An object file is divided into sections containing different types of
5753 data. In the most common case, there are three sections: the @dfn{text
5754 section}, which holds instructions and read-only data; the @dfn{data
5755 section}, which holds initialized writable data; and the @dfn{bss
5756 section}, which holds uninitialized data. Some systems have other kinds
5759 The compiler must tell the assembler when to switch sections. These
5760 macros control what commands to output to tell the assembler this. You
5761 can also define additional sections.
5763 @defmac TEXT_SECTION_ASM_OP
5764 A C expression whose value is a string, including spacing, containing the
5765 assembler operation that should precede instructions and read-only data.
5766 Normally @code{"\t.text"} is right.
5769 @defmac TEXT_SECTION
5770 A C statement that switches to the default section containing instructions.
5771 Normally this is not needed, as simply defining @code{TEXT_SECTION_ASM_OP}
5772 is enough. The MIPS port uses this to sort all functions after all data
5776 @defmac HOT_TEXT_SECTION_NAME
5777 If defined, a C string constant for the name of the section containing most
5778 frequently executed functions of the program. If not defined, GCC will provide
5779 a default definition if the target supports named sections.
5782 @defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME
5783 If defined, a C string constant for the name of the section containing unlikely
5784 executed functions in the program.
5787 @defmac DATA_SECTION_ASM_OP
5788 A C expression whose value is a string, including spacing, containing the
5789 assembler operation to identify the following data as writable initialized
5790 data. Normally @code{"\t.data"} is right.
5793 @defmac READONLY_DATA_SECTION_ASM_OP
5794 A C expression whose value is a string, including spacing, containing the
5795 assembler operation to identify the following data as read-only initialized
5799 @defmac READONLY_DATA_SECTION
5800 A macro naming a function to call to switch to the proper section for
5801 read-only data. The default is to use @code{READONLY_DATA_SECTION_ASM_OP}
5802 if defined, else fall back to @code{text_section}.
5804 The most common definition will be @code{data_section}, if the target
5805 does not have a special read-only data section, and does not put data
5806 in the text section.
5809 @defmac SHARED_SECTION_ASM_OP
5810 If defined, a C expression whose value is a string, including spacing,
5811 containing the assembler operation to identify the following data as
5812 shared data. If not defined, @code{DATA_SECTION_ASM_OP} will be used.
5815 @defmac BSS_SECTION_ASM_OP
5816 If defined, a C expression whose value is a string, including spacing,
5817 containing the assembler operation to identify the following data as
5818 uninitialized global data. If not defined, and neither
5819 @code{ASM_OUTPUT_BSS} nor @code{ASM_OUTPUT_ALIGNED_BSS} are defined,
5820 uninitialized global data will be output in the data section if
5821 @option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be
5825 @defmac SHARED_BSS_SECTION_ASM_OP
5826 If defined, a C expression whose value is a string, including spacing,
5827 containing the assembler operation to identify the following data as
5828 uninitialized global shared data. If not defined, and
5829 @code{BSS_SECTION_ASM_OP} is, the latter will be used.
5832 @defmac INIT_SECTION_ASM_OP
5833 If defined, a C expression whose value is a string, including spacing,
5834 containing the assembler operation to identify the following data as
5835 initialization code. If not defined, GCC will assume such a section does
5839 @defmac FINI_SECTION_ASM_OP
5840 If defined, a C expression whose value is a string, including spacing,
5841 containing the assembler operation to identify the following data as
5842 finalization code. If not defined, GCC will assume such a section does
5846 @defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function})
5847 If defined, an ASM statement that switches to a different section
5848 via @var{section_op}, calls @var{function}, and switches back to
5849 the text section. This is used in @file{crtstuff.c} if
5850 @code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls
5851 to initialization and finalization functions from the init and fini
5852 sections. By default, this macro uses a simple function call. Some
5853 ports need hand-crafted assembly code to avoid dependencies on
5854 registers initialized in the function prologue or to ensure that
5855 constant pools don't end up too far way in the text section.
5858 @defmac FORCE_CODE_SECTION_ALIGN
5859 If defined, an ASM statement that aligns a code section to some
5860 arbitrary boundary. This is used to force all fragments of the
5861 @code{.init} and @code{.fini} sections to have to same alignment
5862 and thus prevent the linker from having to add any padding.
5867 @defmac EXTRA_SECTIONS
5868 A list of names for sections other than the standard two, which are
5869 @code{in_text} and @code{in_data}. You need not define this macro
5870 on a system with no other sections (that GCC needs to use).
5873 @findex text_section
5874 @findex data_section
5875 @defmac EXTRA_SECTION_FUNCTIONS
5876 One or more functions to be defined in @file{varasm.c}. These
5877 functions should do jobs analogous to those of @code{text_section} and
5878 @code{data_section}, for your additional sections. Do not define this
5879 macro if you do not define @code{EXTRA_SECTIONS}.
5882 @defmac JUMP_TABLES_IN_TEXT_SECTION
5883 Define this macro to be an expression with a nonzero value if jump
5884 tables (for @code{tablejump} insns) should be output in the text
5885 section, along with the assembler instructions. Otherwise, the
5886 readonly data section is used.
5888 This macro is irrelevant if there is no separate readonly data section.
5891 @deftypefn {Target Hook} void TARGET_ASM_SELECT_SECTION (tree @var{exp}, int @var{reloc}, unsigned HOST_WIDE_INT @var{align})
5892 Switches to the appropriate section for output of @var{exp}. You can
5893 assume that @var{exp} is either a @code{VAR_DECL} node or a constant of
5894 some sort. @var{reloc} indicates whether the initial value of @var{exp}
5895 requires link-time relocations. Bit 0 is set when variable contains
5896 local relocations only, while bit 1 is set for global relocations.
5897 Select the section by calling @code{data_section} or one of the
5898 alternatives for other sections. @var{align} is the constant alignment
5901 The default version of this function takes care of putting read-only
5902 variables in @code{readonly_data_section}.
5905 @deftypefn {Target Hook} void TARGET_ASM_UNIQUE_SECTION (tree @var{decl}, int @var{reloc})
5906 Build up a unique section name, expressed as a @code{STRING_CST} node,
5907 and assign it to @samp{DECL_SECTION_NAME (@var{decl})}.
5908 As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether
5909 the initial value of @var{exp} requires link-time relocations.
5911 The default version of this function appends the symbol name to the
5912 ELF section name that would normally be used for the symbol. For
5913 example, the function @code{foo} would be placed in @code{.text.foo}.
5914 Whatever the actual target object format, this is often good enough.
5917 @deftypefn {Target Hook} void TARGET_ASM_SELECT_RTX_SECTION (enum machine_mode @var{mode}, rtx @var{x}, unsigned HOST_WIDE_INT @var{align})
5918 Switches to the appropriate section for output of constant pool entry
5919 @var{x} in @var{mode}. You can assume that @var{x} is some kind of
5920 constant in RTL@. The argument @var{mode} is redundant except in the
5921 case of a @code{const_int} rtx. Select the section by calling
5922 @code{readonly_data_section} or one of the alternatives for other
5923 sections. @var{align} is the constant alignment in bits.
5925 The default version of this function takes care of putting symbolic
5926 constants in @code{flag_pic} mode in @code{data_section} and everything
5927 else in @code{readonly_data_section}.
5930 @deftypefn {Target Hook} void TARGET_ENCODE_SECTION_INFO (tree @var{decl}, rtx @var{rtl}, int @var{new_decl_p})
5931 Define this hook if references to a symbol or a constant must be
5932 treated differently depending on something about the variable or
5933 function named by the symbol (such as what section it is in).
5935 The hook is executed immediately after rtl has been created for
5936 @var{decl}, which may be a variable or function declaration or
5937 an entry in the constant pool. In either case, @var{rtl} is the
5938 rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})}
5939 in this hook; that field may not have been initialized yet.
5941 In the case of a constant, it is safe to assume that the rtl is
5942 a @code{mem} whose address is a @code{symbol_ref}. Most decls
5943 will also have this form, but that is not guaranteed. Global
5944 register variables, for instance, will have a @code{reg} for their
5945 rtl. (Normally the right thing to do with such unusual rtl is
5948 The @var{new_decl_p} argument will be true if this is the first time
5949 that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will
5950 be false for subsequent invocations, which will happen for duplicate
5951 declarations. Whether or not anything must be done for the duplicate
5952 declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}.
5953 @var{new_decl_p} is always true when the hook is called for a constant.
5955 @cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO}
5956 The usual thing for this hook to do is to record flags in the
5957 @code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}.
5958 Historically, the name string was modified if it was necessary to
5959 encode more than one bit of information, but this practice is now
5960 discouraged; use @code{SYMBOL_REF_FLAGS}.
5962 The default definition of this hook, @code{default_encode_section_info}
5963 in @file{varasm.c}, sets a number of commonly-useful bits in
5964 @code{SYMBOL_REF_FLAGS}. Check whether the default does what you need
5965 before overriding it.
5968 @deftypefn {Target Hook} const char *TARGET_STRIP_NAME_ENCODING (const char *name)
5969 Decode @var{name} and return the real name part, sans
5970 the characters that @code{TARGET_ENCODE_SECTION_INFO}
5974 @deftypefn {Target Hook} bool TARGET_IN_SMALL_DATA_P (tree @var{exp})
5975 Returns true if @var{exp} should be placed into a ``small data'' section.
5976 The default version of this hook always returns false.
5979 @deftypevar {Target Hook} bool TARGET_HAVE_SRODATA_SECTION
5980 Contains the value true if the target places read-only
5981 ``small data'' into a separate section. The default value is false.
5984 @deftypefn {Target Hook} bool TARGET_BINDS_LOCAL_P (tree @var{exp})
5985 Returns true if @var{exp} names an object for which name resolution
5986 rules must resolve to the current ``module'' (dynamic shared library
5987 or executable image).
5989 The default version of this hook implements the name resolution rules
5990 for ELF, which has a looser model of global name binding than other
5991 currently supported object file formats.
5994 @deftypevar {Target Hook} bool TARGET_HAVE_TLS
5995 Contains the value true if the target supports thread-local storage.
5996 The default value is false.
6001 @section Position Independent Code
6002 @cindex position independent code
6005 This section describes macros that help implement generation of position
6006 independent code. Simply defining these macros is not enough to
6007 generate valid PIC; you must also add support to the macros
6008 @code{GO_IF_LEGITIMATE_ADDRESS} and @code{PRINT_OPERAND_ADDRESS}, as
6009 well as @code{LEGITIMIZE_ADDRESS}. You must modify the definition of
6010 @samp{movsi} to do something appropriate when the source operand
6011 contains a symbolic address. You may also need to alter the handling of
6012 switch statements so that they use relative addresses.
6013 @c i rearranged the order of the macros above to try to force one of
6014 @c them to the next line, to eliminate an overfull hbox. --mew 10feb93
6016 @defmac PIC_OFFSET_TABLE_REGNUM
6017 The register number of the register used to address a table of static
6018 data addresses in memory. In some cases this register is defined by a
6019 processor's ``application binary interface'' (ABI)@. When this macro
6020 is defined, RTL is generated for this register once, as with the stack
6021 pointer and frame pointer registers. If this macro is not defined, it
6022 is up to the machine-dependent files to allocate such a register (if
6023 necessary). Note that this register must be fixed when in use (e.g.@:
6024 when @code{flag_pic} is true).
6027 @defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED
6028 Define this macro if the register defined by
6029 @code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. Do not define
6030 this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined.
6033 @defmac FINALIZE_PIC
6034 By generating position-independent code, when two different programs (A
6035 and B) share a common library (libC.a), the text of the library can be
6036 shared whether or not the library is linked at the same address for both
6037 programs. In some of these environments, position-independent code
6038 requires not only the use of different addressing modes, but also
6039 special code to enable the use of these addressing modes.
6041 The @code{FINALIZE_PIC} macro serves as a hook to emit these special
6042 codes once the function is being compiled into assembly code, but not
6043 before. (It is not done before, because in the case of compiling an
6044 inline function, it would lead to multiple PIC prologues being
6045 included in functions which used inline functions and were compiled to
6049 @defmac LEGITIMATE_PIC_OPERAND_P (@var{x})
6050 A C expression that is nonzero if @var{x} is a legitimate immediate
6051 operand on the target machine when generating position independent code.
6052 You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not
6053 check this. You can also assume @var{flag_pic} is true, so you need not
6054 check it either. You need not define this macro if all constants
6055 (including @code{SYMBOL_REF}) can be immediate operands when generating
6056 position independent code.
6059 @node Assembler Format
6060 @section Defining the Output Assembler Language
6062 This section describes macros whose principal purpose is to describe how
6063 to write instructions in assembler language---rather than what the
6067 * File Framework:: Structural information for the assembler file.
6068 * Data Output:: Output of constants (numbers, strings, addresses).
6069 * Uninitialized Data:: Output of uninitialized variables.
6070 * Label Output:: Output and generation of labels.
6071 * Initialization:: General principles of initialization
6072 and termination routines.
6073 * Macros for Initialization::
6074 Specific macros that control the handling of
6075 initialization and termination routines.
6076 * Instruction Output:: Output of actual instructions.
6077 * Dispatch Tables:: Output of jump tables.
6078 * Exception Region Output:: Output of exception region code.
6079 * Alignment Output:: Pseudo ops for alignment and skipping data.
6082 @node File Framework
6083 @subsection The Overall Framework of an Assembler File
6084 @cindex assembler format
6085 @cindex output of assembler code
6087 @c prevent bad page break with this line
6088 This describes the overall framework of an assembly file.
6090 @deftypefn {Target Hook} void TARGET_ASM_FILE_START ()
6091 @findex default_file_start
6092 Output to @code{asm_out_file} any text which the assembler expects to
6093 find at the beginning of a file. The default behavior is controlled
6094 by two flags, documented below. Unless your target's assembler is
6095 quite unusual, if you override the default, you should call
6096 @code{default_file_start} at some point in your target hook. This
6097 lets other target files rely on these variables.
6100 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_APP_OFF
6101 If this flag is true, the text of the macro @code{ASM_APP_OFF} will be
6102 printed as the very first line in the assembly file, unless
6103 @option{-fverbose-asm} is in effect. (If that macro has been defined
6104 to the empty string, this variable has no effect.) With the normal
6105 definition of @code{ASM_APP_OFF}, the effect is to notify the GNU
6106 assembler that it need not bother stripping comments or extra
6107 whitespace from its input. This allows it to work a bit faster.
6109 The default is false. You should not set it to true unless you have
6110 verified that your port does not generate any extra whitespace or
6111 comments that will cause GAS to issue errors in NO_APP mode.
6114 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_FILE_DIRECTIVE
6115 If this flag is true, @code{output_file_directive} will be called
6116 for the primary source file, immediately after printing
6117 @code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect
6118 this to be done. The default is false.
6121 @deftypefn {Target Hook} void TARGET_ASM_FILE_END ()
6122 Output to @code{asm_out_file} any text which the assembler expects
6123 to find at the end of a file. The default is to output nothing.
6126 @deftypefun void file_end_indicate_exec_stack ()
6127 Some systems use a common convention, the @samp{.note.GNU-stack}
6128 special section, to indicate whether or not an object file relies on
6129 the stack being executable. If your system uses this convention, you
6130 should define @code{TARGET_ASM_FILE_END} to this function. If you
6131 need to do other things in that hook, have your hook function call
6135 @defmac ASM_COMMENT_START
6136 A C string constant describing how to begin a comment in the target
6137 assembler language. The compiler assumes that the comment will end at
6138 the end of the line.
6142 A C string constant for text to be output before each @code{asm}
6143 statement or group of consecutive ones. Normally this is
6144 @code{"#APP"}, which is a comment that has no effect on most
6145 assemblers but tells the GNU assembler that it must check the lines
6146 that follow for all valid assembler constructs.
6150 A C string constant for text to be output after each @code{asm}
6151 statement or group of consecutive ones. Normally this is
6152 @code{"#NO_APP"}, which tells the GNU assembler to resume making the
6153 time-saving assumptions that are valid for ordinary compiler output.
6156 @defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
6157 A C statement to output COFF information or DWARF debugging information
6158 which indicates that filename @var{name} is the current source file to
6159 the stdio stream @var{stream}.
6161 This macro need not be defined if the standard form of output
6162 for the file format in use is appropriate.
6165 @defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string})
6166 A C statement to output the string @var{string} to the stdio stream
6167 @var{stream}. If you do not call the function @code{output_quoted_string}
6168 in your config files, GCC will only call it to output filenames to
6169 the assembler source. So you can use it to canonicalize the format
6170 of the filename using this macro.
6173 @defmac ASM_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter})
6174 A C statement to output DBX or SDB debugging information before code
6175 for line number @var{line} of the current source file to the
6176 stdio stream @var{stream}. @var{counter} is the number of time the
6177 macro was invoked, including the current invocation; it is intended
6178 to generate unique labels in the assembly output.
6180 This macro need not be defined if the standard form of debugging
6181 information for the debugger in use is appropriate.
6184 @defmac ASM_OUTPUT_IDENT (@var{stream}, @var{string})
6185 A C statement to output something to the assembler file to handle a
6186 @samp{#ident} directive containing the text @var{string}. If this
6187 macro is not defined, nothing is output for a @samp{#ident} directive.
6190 @deftypefn {Target Hook} void TARGET_ASM_NAMED_SECTION (const char *@var{name}, unsigned int @var{flags}, unsigned int @var{align})
6191 Output assembly directives to switch to section @var{name}. The section
6192 should have attributes as specified by @var{flags}, which is a bit mask
6193 of the @code{SECTION_*} flags defined in @file{output.h}. If @var{align}
6194 is nonzero, it contains an alignment in bytes to be used for the section,
6195 otherwise some target default should be used. Only targets that must
6196 specify an alignment within the section directive need pay attention to
6197 @var{align} -- we will still use @code{ASM_OUTPUT_ALIGN}.
6200 @deftypefn {Target Hook} bool TARGET_HAVE_NAMED_SECTIONS
6201 This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}.
6204 @deftypefn {Target Hook} {unsigned int} TARGET_SECTION_TYPE_FLAGS (tree @var{decl}, const char *@var{name}, int @var{reloc})
6205 Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION}
6206 based on a variable or function decl, a section name, and whether or not the
6207 declaration's initializer may contain runtime relocations. @var{decl} may be
6208 null, in which case read-write data should be assumed.
6210 The default version if this function handles choosing code vs data,
6211 read-only vs read-write data, and @code{flag_pic}. You should only
6212 need to override this if your target has special flags that might be
6213 set via @code{__attribute__}.
6218 @subsection Output of Data
6221 @deftypevr {Target Hook} {const char *} TARGET_ASM_BYTE_OP
6222 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP
6223 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP
6224 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP
6225 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP
6226 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP
6227 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP
6228 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP
6229 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP
6230 These hooks specify assembly directives for creating certain kinds
6231 of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a
6232 byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an
6233 aligned two-byte object, and so on. Any of the hooks may be
6234 @code{NULL}, indicating that no suitable directive is available.
6236 The compiler will print these strings at the start of a new line,
6237 followed immediately by the object's initial value. In most cases,
6238 the string should contain a tab, a pseudo-op, and then another tab.
6241 @deftypefn {Target Hook} bool TARGET_ASM_INTEGER (rtx @var{x}, unsigned int @var{size}, int @var{aligned_p})
6242 The @code{assemble_integer} function uses this hook to output an
6243 integer object. @var{x} is the object's value, @var{size} is its size
6244 in bytes and @var{aligned_p} indicates whether it is aligned. The
6245 function should return @code{true} if it was able to output the
6246 object. If it returns false, @code{assemble_integer} will try to
6247 split the object into smaller parts.
6249 The default implementation of this hook will use the
6250 @code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false}
6251 when the relevant string is @code{NULL}.
6254 @defmac OUTPUT_ADDR_CONST_EXTRA (@var{stream}, @var{x}, @var{fail})
6255 A C statement to recognize @var{rtx} patterns that
6256 @code{output_addr_const} can't deal with, and output assembly code to
6257 @var{stream} corresponding to the pattern @var{x}. This may be used to
6258 allow machine-dependent @code{UNSPEC}s to appear within constants.
6260 If @code{OUTPUT_ADDR_CONST_EXTRA} fails to recognize a pattern, it must
6261 @code{goto fail}, so that a standard error message is printed. If it
6262 prints an error message itself, by calling, for example,
6263 @code{output_operand_lossage}, it may just complete normally.
6266 @defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
6267 A C statement to output to the stdio stream @var{stream} an assembler
6268 instruction to assemble a string constant containing the @var{len}
6269 bytes at @var{ptr}. @var{ptr} will be a C expression of type
6270 @code{char *} and @var{len} a C expression of type @code{int}.
6272 If the assembler has a @code{.ascii} pseudo-op as found in the
6273 Berkeley Unix assembler, do not define the macro
6274 @code{ASM_OUTPUT_ASCII}.
6277 @defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n})
6278 A C statement to output word @var{n} of a function descriptor for
6279 @var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS}
6280 is defined, and is otherwise unused.
6283 @defmac CONSTANT_POOL_BEFORE_FUNCTION
6284 You may define this macro as a C expression. You should define the
6285 expression to have a nonzero value if GCC should output the constant
6286 pool for a function before the code for the function, or a zero value if
6287 GCC should output the constant pool after the function. If you do
6288 not define this macro, the usual case, GCC will output the constant
6289 pool before the function.
6292 @defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size})
6293 A C statement to output assembler commands to define the start of the
6294 constant pool for a function. @var{funname} is a string giving
6295 the name of the function. Should the return type of the function
6296 be required, it can be obtained via @var{fundecl}. @var{size}
6297 is the size, in bytes, of the constant pool that will be written
6298 immediately after this call.
6300 If no constant-pool prefix is required, the usual case, this macro need
6304 @defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto})
6305 A C statement (with or without semicolon) to output a constant in the
6306 constant pool, if it needs special treatment. (This macro need not do
6307 anything for RTL expressions that can be output normally.)
6309 The argument @var{file} is the standard I/O stream to output the
6310 assembler code on. @var{x} is the RTL expression for the constant to
6311 output, and @var{mode} is the machine mode (in case @var{x} is a
6312 @samp{const_int}). @var{align} is the required alignment for the value
6313 @var{x}; you should output an assembler directive to force this much
6316 The argument @var{labelno} is a number to use in an internal label for
6317 the address of this pool entry. The definition of this macro is
6318 responsible for outputting the label definition at the proper place.
6319 Here is how to do this:
6322 @code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno});
6325 When you output a pool entry specially, you should end with a
6326 @code{goto} to the label @var{jumpto}. This will prevent the same pool
6327 entry from being output a second time in the usual manner.
6329 You need not define this macro if it would do nothing.
6332 @defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size})
6333 A C statement to output assembler commands to at the end of the constant
6334 pool for a function. @var{funname} is a string giving the name of the
6335 function. Should the return type of the function be required, you can
6336 obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the
6337 constant pool that GCC wrote immediately before this call.
6339 If no constant-pool epilogue is required, the usual case, you need not
6343 @defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C})
6344 Define this macro as a C expression which is nonzero if @var{C} is
6345 used as a logical line separator by the assembler.
6347 If you do not define this macro, the default is that only
6348 the character @samp{;} is treated as a logical line separator.
6351 @deftypevr {Target Hook} {const char *} TARGET_ASM_OPEN_PAREN
6352 @deftypevrx {Target Hook} {const char *} TARGET_ASM_CLOSE_PAREN
6353 These target hooks are C string constants, describing the syntax in the
6354 assembler for grouping arithmetic expressions. If not overridden, they
6355 default to normal parentheses, which is correct for most assemblers.
6358 These macros are provided by @file{real.h} for writing the definitions
6359 of @code{ASM_OUTPUT_DOUBLE} and the like:
6361 @defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l})
6362 @defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l})
6363 @defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l})
6364 These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the target's
6365 floating point representation, and store its bit pattern in the variable
6366 @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE}, this variable should
6367 be a simple @code{long int}. For the others, it should be an array of
6368 @code{long int}. The number of elements in this array is determined by
6369 the size of the desired target floating point data type: 32 bits of it
6370 go in each @code{long int} array element. Each array element holds 32
6371 bits of the result, even if @code{long int} is wider than 32 bits on the
6374 The array element values are designed so that you can print them out
6375 using @code{fprintf} in the order they should appear in the target
6379 @node Uninitialized Data
6380 @subsection Output of Uninitialized Variables
6382 Each of the macros in this section is used to do the whole job of
6383 outputting a single uninitialized variable.
6385 @defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
6386 A C statement (sans semicolon) to output to the stdio stream
6387 @var{stream} the assembler definition of a common-label named
6388 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
6389 is the size rounded up to whatever alignment the caller wants.
6391 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
6392 output the name itself; before and after that, output the additional
6393 assembler syntax for defining the name, and a newline.
6395 This macro controls how the assembler definitions of uninitialized
6396 common global variables are output.
6399 @defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment})
6400 Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a
6401 separate, explicit argument. If you define this macro, it is used in
6402 place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in
6403 handling the required alignment of the variable. The alignment is specified
6404 as the number of bits.
6407 @defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
6408 Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the
6409 variable to be output, if there is one, or @code{NULL_TREE} if there
6410 is no corresponding variable. If you define this macro, GCC will use it
6411 in place of both @code{ASM_OUTPUT_COMMON} and
6412 @code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see
6413 the variable's decl in order to chose what to output.
6416 @defmac ASM_OUTPUT_SHARED_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
6417 If defined, it is similar to @code{ASM_OUTPUT_COMMON}, except that it
6418 is used when @var{name} is shared. If not defined, @code{ASM_OUTPUT_COMMON}
6422 @defmac ASM_OUTPUT_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded})
6423 A C statement (sans semicolon) to output to the stdio stream
6424 @var{stream} the assembler definition of uninitialized global @var{decl} named
6425 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
6426 is the size rounded up to whatever alignment the caller wants.
6428 Try to use function @code{asm_output_bss} defined in @file{varasm.c} when
6429 defining this macro. If unable, use the expression
6430 @code{assemble_name (@var{stream}, @var{name})} to output the name itself;
6431 before and after that, output the additional assembler syntax for defining
6432 the name, and a newline.
6434 This macro controls how the assembler definitions of uninitialized global
6435 variables are output. This macro exists to properly support languages like
6436 C++ which do not have @code{common} data. However, this macro currently
6437 is not defined for all targets. If this macro and
6438 @code{ASM_OUTPUT_ALIGNED_BSS} are not defined then @code{ASM_OUTPUT_COMMON}
6439 or @code{ASM_OUTPUT_ALIGNED_COMMON} or
6440 @code{ASM_OUTPUT_ALIGNED_DECL_COMMON} is used.
6443 @defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
6444 Like @code{ASM_OUTPUT_BSS} except takes the required alignment as a
6445 separate, explicit argument. If you define this macro, it is used in
6446 place of @code{ASM_OUTPUT_BSS}, and gives you more flexibility in
6447 handling the required alignment of the variable. The alignment is specified
6448 as the number of bits.
6450 Try to use function @code{asm_output_aligned_bss} defined in file
6451 @file{varasm.c} when defining this macro.
6454 @defmac ASM_OUTPUT_SHARED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded})
6455 If defined, it is similar to @code{ASM_OUTPUT_BSS}, except that it
6456 is used when @var{name} is shared. If not defined, @code{ASM_OUTPUT_BSS}
6460 @defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
6461 A C statement (sans semicolon) to output to the stdio stream
6462 @var{stream} the assembler definition of a local-common-label named
6463 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
6464 is the size rounded up to whatever alignment the caller wants.
6466 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
6467 output the name itself; before and after that, output the additional
6468 assembler syntax for defining the name, and a newline.
6470 This macro controls how the assembler definitions of uninitialized
6471 static variables are output.
6474 @defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment})
6475 Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a
6476 separate, explicit argument. If you define this macro, it is used in
6477 place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in
6478 handling the required alignment of the variable. The alignment is specified
6479 as the number of bits.
6482 @defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
6483 Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the
6484 variable to be output, if there is one, or @code{NULL_TREE} if there
6485 is no corresponding variable. If you define this macro, GCC will use it
6486 in place of both @code{ASM_OUTPUT_DECL} and
6487 @code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see
6488 the variable's decl in order to chose what to output.
6491 @defmac ASM_OUTPUT_SHARED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
6492 If defined, it is similar to @code{ASM_OUTPUT_LOCAL}, except that it
6493 is used when @var{name} is shared. If not defined, @code{ASM_OUTPUT_LOCAL}
6498 @subsection Output and Generation of Labels
6500 @c prevent bad page break with this line
6501 This is about outputting labels.
6503 @findex assemble_name
6504 @defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name})
6505 A C statement (sans semicolon) to output to the stdio stream
6506 @var{stream} the assembler definition of a label named @var{name}.
6507 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
6508 output the name itself; before and after that, output the additional
6509 assembler syntax for defining the name, and a newline. A default
6510 definition of this macro is provided which is correct for most systems.
6514 A C string containing the appropriate assembler directive to specify the
6515 size of a symbol, without any arguments. On systems that use ELF, the
6516 default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other
6517 systems, the default is not to define this macro.
6519 Define this macro only if it is correct to use the default definitions
6520 of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE}
6521 for your system. If you need your own custom definitions of those
6522 macros, or if you do not need explicit symbol sizes at all, do not
6526 @defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size})
6527 A C statement (sans semicolon) to output to the stdio stream
6528 @var{stream} a directive telling the assembler that the size of the
6529 symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}.
6530 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
6534 @defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name})
6535 A C statement (sans semicolon) to output to the stdio stream
6536 @var{stream} a directive telling the assembler to calculate the size of
6537 the symbol @var{name} by subtracting its address from the current
6540 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
6541 provided. The default assumes that the assembler recognizes a special
6542 @samp{.} symbol as referring to the current address, and can calculate
6543 the difference between this and another symbol. If your assembler does
6544 not recognize @samp{.} or cannot do calculations with it, you will need
6545 to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique.
6549 A C string containing the appropriate assembler directive to specify the
6550 type of a symbol, without any arguments. On systems that use ELF, the
6551 default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other
6552 systems, the default is not to define this macro.
6554 Define this macro only if it is correct to use the default definition of
6555 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
6556 custom definition of this macro, or if you do not need explicit symbol
6557 types at all, do not define this macro.
6560 @defmac TYPE_OPERAND_FMT
6561 A C string which specifies (using @code{printf} syntax) the format of
6562 the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the
6563 default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems,
6564 the default is not to define this macro.
6566 Define this macro only if it is correct to use the default definition of
6567 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
6568 custom definition of this macro, or if you do not need explicit symbol
6569 types at all, do not define this macro.
6572 @defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type})
6573 A C statement (sans semicolon) to output to the stdio stream
6574 @var{stream} a directive telling the assembler that the type of the
6575 symbol @var{name} is @var{type}. @var{type} is a C string; currently,
6576 that string is always either @samp{"function"} or @samp{"object"}, but
6577 you should not count on this.
6579 If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default
6580 definition of this macro is provided.
6583 @defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
6584 A C statement (sans semicolon) to output to the stdio stream
6585 @var{stream} any text necessary for declaring the name @var{name} of a
6586 function which is being defined. This macro is responsible for
6587 outputting the label definition (perhaps using
6588 @code{ASM_OUTPUT_LABEL}). The argument @var{decl} is the
6589 @code{FUNCTION_DECL} tree node representing the function.
6591 If this macro is not defined, then the function name is defined in the
6592 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
6594 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
6598 @defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl})
6599 A C statement (sans semicolon) to output to the stdio stream
6600 @var{stream} any text necessary for declaring the size of a function
6601 which is being defined. The argument @var{name} is the name of the
6602 function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node
6603 representing the function.
6605 If this macro is not defined, then the function size is not defined.
6607 You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition
6611 @defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl})
6612 A C statement (sans semicolon) to output to the stdio stream
6613 @var{stream} any text necessary for declaring the name @var{name} of an
6614 initialized variable which is being defined. This macro must output the
6615 label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument
6616 @var{decl} is the @code{VAR_DECL} tree node representing the variable.
6618 If this macro is not defined, then the variable name is defined in the
6619 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
6621 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or
6622 @code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro.
6625 @defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name})
6626 A C statement (sans semicolon) to output to the stdio stream
6627 @var{stream} any text necessary for claiming a register @var{regno}
6628 for a global variable @var{decl} with name @var{name}.
6630 If you don't define this macro, that is equivalent to defining it to do
6634 @defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend})
6635 A C statement (sans semicolon) to finish up declaring a variable name
6636 once the compiler has processed its initializer fully and thus has had a
6637 chance to determine the size of an array when controlled by an
6638 initializer. This is used on systems where it's necessary to declare
6639 something about the size of the object.
6641 If you don't define this macro, that is equivalent to defining it to do
6644 You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or
6645 @code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro.
6648 @deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_LABEL (FILE *@var{stream}, const char *@var{name})
6649 This target hook is a function to output to the stdio stream
6650 @var{stream} some commands that will make the label @var{name} global;
6651 that is, available for reference from other files.
6653 The default implementation relies on a proper definition of
6654 @code{GLOBAL_ASM_OP}.
6657 @defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name})
6658 A C statement (sans semicolon) to output to the stdio stream
6659 @var{stream} some commands that will make the label @var{name} weak;
6660 that is, available for reference from other files but only used if
6661 no other definition is available. Use the expression
6662 @code{assemble_name (@var{stream}, @var{name})} to output the name
6663 itself; before and after that, output the additional assembler syntax
6664 for making that name weak, and a newline.
6666 If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not
6667 support weak symbols and you should not define the @code{SUPPORTS_WEAK}
6671 @defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value})
6672 Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and
6673 @code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function
6674 or variable decl. If @var{value} is not @code{NULL}, this C statement
6675 should output to the stdio stream @var{stream} assembler code which
6676 defines (equates) the weak symbol @var{name} to have the value
6677 @var{value}. If @var{value} is @code{NULL}, it should output commands
6678 to make @var{name} weak.
6681 @defmac SUPPORTS_WEAK
6682 A C expression which evaluates to true if the target supports weak symbols.
6684 If you don't define this macro, @file{defaults.h} provides a default
6685 definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL}
6686 is defined, the default definition is @samp{1}; otherwise, it is
6687 @samp{0}. Define this macro if you want to control weak symbol support
6688 with a compiler flag such as @option{-melf}.
6691 @defmac MAKE_DECL_ONE_ONLY (@var{decl})
6692 A C statement (sans semicolon) to mark @var{decl} to be emitted as a
6693 public symbol such that extra copies in multiple translation units will
6694 be discarded by the linker. Define this macro if your object file
6695 format provides support for this concept, such as the @samp{COMDAT}
6696 section flags in the Microsoft Windows PE/COFF format, and this support
6697 requires changes to @var{decl}, such as putting it in a separate section.
6700 @defmac SUPPORTS_ONE_ONLY
6701 A C expression which evaluates to true if the target supports one-only
6704 If you don't define this macro, @file{varasm.c} provides a default
6705 definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default
6706 definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if
6707 you want to control one-only symbol support with a compiler flag, or if
6708 setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to
6709 be emitted as one-only.
6712 @deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_VISIBILITY (tree @var{decl}, const char *@var{visibility})
6713 This target hook is a function to output to @var{asm_out_file} some
6714 commands that will make the symbol(s) associated with @var{decl} have
6715 hidden, protected or internal visibility as specified by @var{visibility}.
6718 @defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
6719 A C statement (sans semicolon) to output to the stdio stream
6720 @var{stream} any text necessary for declaring the name of an external
6721 symbol named @var{name} which is referenced in this compilation but
6722 not defined. The value of @var{decl} is the tree node for the
6725 This macro need not be defined if it does not need to output anything.
6726 The GNU assembler and most Unix assemblers don't require anything.
6729 @defmac ASM_OUTPUT_EXTERNAL_LIBCALL (@var{stream}, @var{symref})
6730 A C statement (sans semicolon) to output on @var{stream} an assembler
6731 pseudo-op to declare a library function name external. The name of the
6732 library function is given by @var{symref}, which has type @code{rtx} and
6733 is a @code{symbol_ref}.
6735 This macro need not be defined if it does not need to output anything.
6736 The GNU assembler and most Unix assemblers don't require anything.
6739 @defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
6740 A C statement (sans semicolon) to output to the stdio stream
6741 @var{stream} a reference in assembler syntax to a label named
6742 @var{name}. This should add @samp{_} to the front of the name, if that
6743 is customary on your operating system, as it is in most Berkeley Unix
6744 systems. This macro is used in @code{assemble_name}.
6747 @defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym})
6748 A C statement (sans semicolon) to output a reference to
6749 @code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name}
6750 will be used to output the name of the symbol. This macro may be used
6751 to modify the way a symbol is referenced depending on information
6752 encoded by @code{TARGET_ENCODE_SECTION_INFO}.
6755 @defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf})
6756 A C statement (sans semicolon) to output a reference to @var{buf}, the
6757 result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined,
6758 @code{assemble_name} will be used to output the name of the symbol.
6759 This macro is not used by @code{output_asm_label}, or the @code{%l}
6760 specifier that calls it; the intention is that this macro should be set
6761 when it is necessary to output a label differently when its address is
6765 @deftypefn {Target Hook} void TARGET_ASM_INTERNAL_LABEL (FILE *@var{stream}, const char *@var{prefix}, unsigned long @var{labelno})
6766 A function to output to the stdio stream @var{stream} a label whose
6767 name is made from the string @var{prefix} and the number @var{labelno}.
6769 It is absolutely essential that these labels be distinct from the labels
6770 used for user-level functions and variables. Otherwise, certain programs
6771 will have name conflicts with internal labels.
6773 It is desirable to exclude internal labels from the symbol table of the
6774 object file. Most assemblers have a naming convention for labels that
6775 should be excluded; on many systems, the letter @samp{L} at the
6776 beginning of a label has this effect. You should find out what
6777 convention your system uses, and follow it.
6779 The default version of this function utilizes ASM_GENERATE_INTERNAL_LABEL.
6782 @defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num})
6783 A C statement to output to the stdio stream @var{stream} a debug info
6784 label whose name is made from the string @var{prefix} and the number
6785 @var{num}. This is useful for VLIW targets, where debug info labels
6786 may need to be treated differently than branch target labels. On some
6787 systems, branch target labels must be at the beginning of instruction
6788 bundles, but debug info labels can occur in the middle of instruction
6791 If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be
6795 @defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
6796 A C statement to store into the string @var{string} a label whose name
6797 is made from the string @var{prefix} and the number @var{num}.
6799 This string, when output subsequently by @code{assemble_name}, should
6800 produce the output that @code{(*targetm.asm_out.internal_label)} would produce
6801 with the same @var{prefix} and @var{num}.
6803 If the string begins with @samp{*}, then @code{assemble_name} will
6804 output the rest of the string unchanged. It is often convenient for
6805 @code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the
6806 string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets
6807 to output the string, and may change it. (Of course,
6808 @code{ASM_OUTPUT_LABELREF} is also part of your machine description, so
6809 you should know what it does on your machine.)
6812 @defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
6813 A C expression to assign to @var{outvar} (which is a variable of type
6814 @code{char *}) a newly allocated string made from the string
6815 @var{name} and the number @var{number}, with some suitable punctuation
6816 added. Use @code{alloca} to get space for the string.
6818 The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to
6819 produce an assembler label for an internal static variable whose name is
6820 @var{name}. Therefore, the string must be such as to result in valid
6821 assembler code. The argument @var{number} is different each time this
6822 macro is executed; it prevents conflicts between similarly-named
6823 internal static variables in different scopes.
6825 Ideally this string should not be a valid C identifier, to prevent any
6826 conflict with the user's own symbols. Most assemblers allow periods
6827 or percent signs in assembler symbols; putting at least one of these
6828 between the name and the number will suffice.
6830 If this macro is not defined, a default definition will be provided
6831 which is correct for most systems.
6834 @defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value})
6835 A C statement to output to the stdio stream @var{stream} assembler code
6836 which defines (equates) the symbol @var{name} to have the value @var{value}.
6839 If @code{SET_ASM_OP} is defined, a default definition is provided which is
6840 correct for most systems.
6843 @defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value})
6844 A C statement to output to the stdio stream @var{stream} assembler code
6845 which defines (equates) the symbol whose tree node is @var{decl_of_name}
6846 to have the value of the tree node @var{decl_of_value}. This macro will
6847 be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if
6848 the tree nodes are available.
6851 If @code{SET_ASM_OP} is defined, a default definition is provided which is
6852 correct for most systems.
6855 @defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value})
6856 A C statement to output to the stdio stream @var{stream} assembler code
6857 which defines (equates) the weak symbol @var{name} to have the value
6858 @var{value}. If @var{value} is @code{NULL}, it defines @var{name} as
6859 an undefined weak symbol.
6861 Define this macro if the target only supports weak aliases; define
6862 @code{ASM_OUTPUT_DEF} instead if possible.
6865 @defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name})
6866 Define this macro to override the default assembler names used for
6867 Objective-C methods.
6869 The default name is a unique method number followed by the name of the
6870 class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of
6871 the category is also included in the assembler name (e.g.@:
6874 These names are safe on most systems, but make debugging difficult since
6875 the method's selector is not present in the name. Therefore, particular
6876 systems define other ways of computing names.
6878 @var{buf} is an expression of type @code{char *} which gives you a
6879 buffer in which to store the name; its length is as long as
6880 @var{class_name}, @var{cat_name} and @var{sel_name} put together, plus
6881 50 characters extra.
6883 The argument @var{is_inst} specifies whether the method is an instance
6884 method or a class method; @var{class_name} is the name of the class;
6885 @var{cat_name} is the name of the category (or @code{NULL} if the method is not
6886 in a category); and @var{sel_name} is the name of the selector.
6888 On systems where the assembler can handle quoted names, you can use this
6889 macro to provide more human-readable names.
6892 @defmac ASM_DECLARE_CLASS_REFERENCE (@var{stream}, @var{name})
6893 A C statement (sans semicolon) to output to the stdio stream
6894 @var{stream} commands to declare that the label @var{name} is an
6895 Objective-C class reference. This is only needed for targets whose
6896 linkers have special support for NeXT-style runtimes.
6899 @defmac ASM_DECLARE_UNRESOLVED_REFERENCE (@var{stream}, @var{name})
6900 A C statement (sans semicolon) to output to the stdio stream
6901 @var{stream} commands to declare that the label @var{name} is an
6902 unresolved Objective-C class reference. This is only needed for targets
6903 whose linkers have special support for NeXT-style runtimes.
6906 @node Initialization
6907 @subsection How Initialization Functions Are Handled
6908 @cindex initialization routines
6909 @cindex termination routines
6910 @cindex constructors, output of
6911 @cindex destructors, output of
6913 The compiled code for certain languages includes @dfn{constructors}
6914 (also called @dfn{initialization routines})---functions to initialize
6915 data in the program when the program is started. These functions need
6916 to be called before the program is ``started''---that is to say, before
6917 @code{main} is called.
6919 Compiling some languages generates @dfn{destructors} (also called
6920 @dfn{termination routines}) that should be called when the program
6923 To make the initialization and termination functions work, the compiler
6924 must output something in the assembler code to cause those functions to
6925 be called at the appropriate time. When you port the compiler to a new
6926 system, you need to specify how to do this.
6928 There are two major ways that GCC currently supports the execution of
6929 initialization and termination functions. Each way has two variants.
6930 Much of the structure is common to all four variations.
6932 @findex __CTOR_LIST__
6933 @findex __DTOR_LIST__
6934 The linker must build two lists of these functions---a list of
6935 initialization functions, called @code{__CTOR_LIST__}, and a list of
6936 termination functions, called @code{__DTOR_LIST__}.
6938 Each list always begins with an ignored function pointer (which may hold
6939 0, @minus{}1, or a count of the function pointers after it, depending on
6940 the environment). This is followed by a series of zero or more function
6941 pointers to constructors (or destructors), followed by a function
6942 pointer containing zero.
6944 Depending on the operating system and its executable file format, either
6945 @file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup
6946 time and exit time. Constructors are called in reverse order of the
6947 list; destructors in forward order.
6949 The best way to handle static constructors works only for object file
6950 formats which provide arbitrarily-named sections. A section is set
6951 aside for a list of constructors, and another for a list of destructors.
6952 Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each
6953 object file that defines an initialization function also puts a word in
6954 the constructor section to point to that function. The linker
6955 accumulates all these words into one contiguous @samp{.ctors} section.
6956 Termination functions are handled similarly.
6958 This method will be chosen as the default by @file{target-def.h} if
6959 @code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not
6960 support arbitrary sections, but does support special designated
6961 constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP}
6962 and @code{DTORS_SECTION_ASM_OP} to achieve the same effect.
6964 When arbitrary sections are available, there are two variants, depending
6965 upon how the code in @file{crtstuff.c} is called. On systems that
6966 support a @dfn{.init} section which is executed at program startup,
6967 parts of @file{crtstuff.c} are compiled into that section. The
6968 program is linked by the @command{gcc} driver like this:
6971 ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o
6974 The prologue of a function (@code{__init}) appears in the @code{.init}
6975 section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise
6976 for the function @code{__fini} in the @dfn{.fini} section. Normally these
6977 files are provided by the operating system or by the GNU C library, but
6978 are provided by GCC for a few targets.
6980 The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets)
6981 compiled from @file{crtstuff.c}. They contain, among other things, code
6982 fragments within the @code{.init} and @code{.fini} sections that branch
6983 to routines in the @code{.text} section. The linker will pull all parts
6984 of a section together, which results in a complete @code{__init} function
6985 that invokes the routines we need at startup.
6987 To use this variant, you must define the @code{INIT_SECTION_ASM_OP}
6990 If no init section is available, when GCC compiles any function called
6991 @code{main} (or more accurately, any function designated as a program
6992 entry point by the language front end calling @code{expand_main_function}),
6993 it inserts a procedure call to @code{__main} as the first executable code
6994 after the function prologue. The @code{__main} function is defined
6995 in @file{libgcc2.c} and runs the global constructors.
6997 In file formats that don't support arbitrary sections, there are again
6998 two variants. In the simplest variant, the GNU linker (GNU @code{ld})
6999 and an `a.out' format must be used. In this case,
7000 @code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs}
7001 entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__},
7002 and with the address of the void function containing the initialization
7003 code as its value. The GNU linker recognizes this as a request to add
7004 the value to a @dfn{set}; the values are accumulated, and are eventually
7005 placed in the executable as a vector in the format described above, with
7006 a leading (ignored) count and a trailing zero element.
7007 @code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init
7008 section is available, the absence of @code{INIT_SECTION_ASM_OP} causes
7009 the compilation of @code{main} to call @code{__main} as above, starting
7010 the initialization process.
7012 The last variant uses neither arbitrary sections nor the GNU linker.
7013 This is preferable when you want to do dynamic linking and when using
7014 file formats which the GNU linker does not support, such as `ECOFF'@. In
7015 this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and
7016 termination functions are recognized simply by their names. This requires
7017 an extra program in the linkage step, called @command{collect2}. This program
7018 pretends to be the linker, for use with GCC; it does its job by running
7019 the ordinary linker, but also arranges to include the vectors of
7020 initialization and termination functions. These functions are called
7021 via @code{__main} as described above. In order to use this method,
7022 @code{use_collect2} must be defined in the target in @file{config.gcc}.
7025 The following section describes the specific macros that control and
7026 customize the handling of initialization and termination functions.
7029 @node Macros for Initialization
7030 @subsection Macros Controlling Initialization Routines
7032 Here are the macros that control how the compiler handles initialization
7033 and termination functions:
7035 @defmac INIT_SECTION_ASM_OP
7036 If defined, a C string constant, including spacing, for the assembler
7037 operation to identify the following data as initialization code. If not
7038 defined, GCC will assume such a section does not exist. When you are
7039 using special sections for initialization and termination functions, this
7040 macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to
7041 run the initialization functions.
7044 @defmac HAS_INIT_SECTION
7045 If defined, @code{main} will not call @code{__main} as described above.
7046 This macro should be defined for systems that control start-up code
7047 on a symbol-by-symbol basis, such as OSF/1, and should not
7048 be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}.
7051 @defmac LD_INIT_SWITCH
7052 If defined, a C string constant for a switch that tells the linker that
7053 the following symbol is an initialization routine.
7056 @defmac LD_FINI_SWITCH
7057 If defined, a C string constant for a switch that tells the linker that
7058 the following symbol is a finalization routine.
7061 @defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func})
7062 If defined, a C statement that will write a function that can be
7063 automatically called when a shared library is loaded. The function
7064 should call @var{func}, which takes no arguments. If not defined, and
7065 the object format requires an explicit initialization function, then a
7066 function called @code{_GLOBAL__DI} will be generated.
7068 This function and the following one are used by collect2 when linking a
7069 shared library that needs constructors or destructors, or has DWARF2
7070 exception tables embedded in the code.
7073 @defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func})
7074 If defined, a C statement that will write a function that can be
7075 automatically called when a shared library is unloaded. The function
7076 should call @var{func}, which takes no arguments. If not defined, and
7077 the object format requires an explicit finalization function, then a
7078 function called @code{_GLOBAL__DD} will be generated.
7081 @defmac INVOKE__main
7082 If defined, @code{main} will call @code{__main} despite the presence of
7083 @code{INIT_SECTION_ASM_OP}. This macro should be defined for systems
7084 where the init section is not actually run automatically, but is still
7085 useful for collecting the lists of constructors and destructors.
7088 @defmac SUPPORTS_INIT_PRIORITY
7089 If nonzero, the C++ @code{init_priority} attribute is supported and the
7090 compiler should emit instructions to control the order of initialization
7091 of objects. If zero, the compiler will issue an error message upon
7092 encountering an @code{init_priority} attribute.
7095 @deftypefn {Target Hook} bool TARGET_HAVE_CTORS_DTORS
7096 This value is true if the target supports some ``native'' method of
7097 collecting constructors and destructors to be run at startup and exit.
7098 It is false if we must use @command{collect2}.
7101 @deftypefn {Target Hook} void TARGET_ASM_CONSTRUCTOR (rtx @var{symbol}, int @var{priority})
7102 If defined, a function that outputs assembler code to arrange to call
7103 the function referenced by @var{symbol} at initialization time.
7105 Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking
7106 no arguments and with no return value. If the target supports initialization
7107 priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY};
7108 otherwise it must be @code{DEFAULT_INIT_PRIORITY}.
7110 If this macro is not defined by the target, a suitable default will
7111 be chosen if (1) the target supports arbitrary section names, (2) the
7112 target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2}
7116 @deftypefn {Target Hook} void TARGET_ASM_DESTRUCTOR (rtx @var{symbol}, int @var{priority})
7117 This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination
7118 functions rather than initialization functions.
7121 If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine
7122 generated for the generated object file will have static linkage.
7124 If your system uses @command{collect2} as the means of processing
7125 constructors, then that program normally uses @command{nm} to scan
7126 an object file for constructor functions to be called.
7128 On certain kinds of systems, you can define this macro to make
7129 @command{collect2} work faster (and, in some cases, make it work at all):
7131 @defmac OBJECT_FORMAT_COFF
7132 Define this macro if the system uses COFF (Common Object File Format)
7133 object files, so that @command{collect2} can assume this format and scan
7134 object files directly for dynamic constructor/destructor functions.
7136 This macro is effective only in a native compiler; @command{collect2} as
7137 part of a cross compiler always uses @command{nm} for the target machine.
7140 @defmac REAL_NM_FILE_NAME
7141 Define this macro as a C string constant containing the file name to use
7142 to execute @command{nm}. The default is to search the path normally for
7145 If your system supports shared libraries and has a program to list the
7146 dynamic dependencies of a given library or executable, you can define
7147 these macros to enable support for running initialization and
7148 termination functions in shared libraries:
7152 Define this macro to a C string constant containing the name of the program
7153 which lists dynamic dependencies, like @command{"ldd"} under SunOS 4.
7156 @defmac PARSE_LDD_OUTPUT (@var{ptr})
7157 Define this macro to be C code that extracts filenames from the output
7158 of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable
7159 of type @code{char *} that points to the beginning of a line of output
7160 from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the
7161 code must advance @var{ptr} to the beginning of the filename on that
7162 line. Otherwise, it must set @var{ptr} to @code{NULL}.
7165 @node Instruction Output
7166 @subsection Output of Assembler Instructions
7168 @c prevent bad page break with this line
7169 This describes assembler instruction output.
7171 @defmac REGISTER_NAMES
7172 A C initializer containing the assembler's names for the machine
7173 registers, each one as a C string constant. This is what translates
7174 register numbers in the compiler into assembler language.
7177 @defmac ADDITIONAL_REGISTER_NAMES
7178 If defined, a C initializer for an array of structures containing a name
7179 and a register number. This macro defines additional names for hard
7180 registers, thus allowing the @code{asm} option in declarations to refer
7181 to registers using alternate names.
7184 @defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
7185 Define this macro if you are using an unusual assembler that
7186 requires different names for the machine instructions.
7188 The definition is a C statement or statements which output an
7189 assembler instruction opcode to the stdio stream @var{stream}. The
7190 macro-operand @var{ptr} is a variable of type @code{char *} which
7191 points to the opcode name in its ``internal'' form---the form that is
7192 written in the machine description. The definition should output the
7193 opcode name to @var{stream}, performing any translation you desire, and
7194 increment the variable @var{ptr} to point at the end of the opcode
7195 so that it will not be output twice.
7197 In fact, your macro definition may process less than the entire opcode
7198 name, or more than the opcode name; but if you want to process text
7199 that includes @samp{%}-sequences to substitute operands, you must take
7200 care of the substitution yourself. Just be sure to increment
7201 @var{ptr} over whatever text should not be output normally.
7203 @findex recog_data.operand
7204 If you need to look at the operand values, they can be found as the
7205 elements of @code{recog_data.operand}.
7207 If the macro definition does nothing, the instruction is output
7211 @defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
7212 If defined, a C statement to be executed just prior to the output of
7213 assembler code for @var{insn}, to modify the extracted operands so
7214 they will be output differently.
7216 Here the argument @var{opvec} is the vector containing the operands
7217 extracted from @var{insn}, and @var{noperands} is the number of
7218 elements of the vector which contain meaningful data for this insn.
7219 The contents of this vector are what will be used to convert the insn
7220 template into assembler code, so you can change the assembler output
7221 by changing the contents of the vector.
7223 This macro is useful when various assembler syntaxes share a single
7224 file of instruction patterns; by defining this macro differently, you
7225 can cause a large class of instructions to be output differently (such
7226 as with rearranged operands). Naturally, variations in assembler
7227 syntax affecting individual insn patterns ought to be handled by
7228 writing conditional output routines in those patterns.
7230 If this macro is not defined, it is equivalent to a null statement.
7233 @defmac FINAL_PRESCAN_LABEL
7234 If defined, @code{FINAL_PRESCAN_INSN} will be called on each
7235 @code{CODE_LABEL}. In that case, @var{opvec} will be a null pointer and
7236 @var{noperands} will be zero.
7239 @defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
7240 A C compound statement to output to stdio stream @var{stream} the
7241 assembler syntax for an instruction operand @var{x}. @var{x} is an
7244 @var{code} is a value that can be used to specify one of several ways
7245 of printing the operand. It is used when identical operands must be
7246 printed differently depending on the context. @var{code} comes from
7247 the @samp{%} specification that was used to request printing of the
7248 operand. If the specification was just @samp{%@var{digit}} then
7249 @var{code} is 0; if the specification was @samp{%@var{ltr}
7250 @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
7253 If @var{x} is a register, this macro should print the register's name.
7254 The names can be found in an array @code{reg_names} whose type is
7255 @code{char *[]}. @code{reg_names} is initialized from
7256 @code{REGISTER_NAMES}.
7258 When the machine description has a specification @samp{%@var{punct}}
7259 (a @samp{%} followed by a punctuation character), this macro is called
7260 with a null pointer for @var{x} and the punctuation character for
7264 @defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code})
7265 A C expression which evaluates to true if @var{code} is a valid
7266 punctuation character for use in the @code{PRINT_OPERAND} macro. If
7267 @code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
7268 punctuation characters (except for the standard one, @samp{%}) are used
7272 @defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
7273 A C compound statement to output to stdio stream @var{stream} the
7274 assembler syntax for an instruction operand that is a memory reference
7275 whose address is @var{x}. @var{x} is an RTL expression.
7277 @cindex @code{TARGET_ENCODE_SECTION_INFO} usage
7278 On some machines, the syntax for a symbolic address depends on the
7279 section that the address refers to. On these machines, define the hook
7280 @code{TARGET_ENCODE_SECTION_INFO} to store the information into the
7281 @code{symbol_ref}, and then check for it here. @xref{Assembler
7285 @findex dbr_sequence_length
7286 @defmac DBR_OUTPUT_SEQEND (@var{file})
7287 A C statement, to be executed after all slot-filler instructions have
7288 been output. If necessary, call @code{dbr_sequence_length} to
7289 determine the number of slots filled in a sequence (zero if not
7290 currently outputting a sequence), to decide how many no-ops to output,
7293 Don't define this macro if it has nothing to do, but it is helpful in
7294 reading assembly output if the extent of the delay sequence is made
7295 explicit (e.g.@: with white space).
7298 @findex final_sequence
7299 Note that output routines for instructions with delay slots must be
7300 prepared to deal with not being output as part of a sequence
7301 (i.e.@: when the scheduling pass is not run, or when no slot fillers could be
7302 found.) The variable @code{final_sequence} is null when not
7303 processing a sequence, otherwise it contains the @code{sequence} rtx
7307 @defmac REGISTER_PREFIX
7308 @defmacx LOCAL_LABEL_PREFIX
7309 @defmacx USER_LABEL_PREFIX
7310 @defmacx IMMEDIATE_PREFIX
7311 If defined, C string expressions to be used for the @samp{%R}, @samp{%L},
7312 @samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see
7313 @file{final.c}). These are useful when a single @file{md} file must
7314 support multiple assembler formats. In that case, the various @file{tm.h}
7315 files can define these macros differently.
7318 @defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format})
7319 If defined this macro should expand to a series of @code{case}
7320 statements which will be parsed inside the @code{switch} statement of
7321 the @code{asm_fprintf} function. This allows targets to define extra
7322 printf formats which may useful when generating their assembler
7323 statements. Note that upper case letters are reserved for future
7324 generic extensions to asm_fprintf, and so are not available to target
7325 specific code. The output file is given by the parameter @var{file}.
7326 The varargs input pointer is @var{argptr} and the rest of the format
7327 string, starting the character after the one that is being switched
7328 upon, is pointed to by @var{format}.
7331 @defmac ASSEMBLER_DIALECT
7332 If your target supports multiple dialects of assembler language (such as
7333 different opcodes), define this macro as a C expression that gives the
7334 numeric index of the assembler language dialect to use, with zero as the
7337 If this macro is defined, you may use constructs of the form
7339 @samp{@{option0|option1|option2@dots{}@}}
7342 in the output templates of patterns (@pxref{Output Template}) or in the
7343 first argument of @code{asm_fprintf}. This construct outputs
7344 @samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of
7345 @code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters
7346 within these strings retain their usual meaning. If there are fewer
7347 alternatives within the braces than the value of
7348 @code{ASSEMBLER_DIALECT}, the construct outputs nothing.
7350 If you do not define this macro, the characters @samp{@{}, @samp{|} and
7351 @samp{@}} do not have any special meaning when used in templates or
7352 operands to @code{asm_fprintf}.
7354 Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX},
7355 @code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express
7356 the variations in assembler language syntax with that mechanism. Define
7357 @code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax
7358 if the syntax variant are larger and involve such things as different
7359 opcodes or operand order.
7362 @defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
7363 A C expression to output to @var{stream} some assembler code
7364 which will push hard register number @var{regno} onto the stack.
7365 The code need not be optimal, since this macro is used only when
7369 @defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
7370 A C expression to output to @var{stream} some assembler code
7371 which will pop hard register number @var{regno} off of the stack.
7372 The code need not be optimal, since this macro is used only when
7376 @node Dispatch Tables
7377 @subsection Output of Dispatch Tables
7379 @c prevent bad page break with this line
7380 This concerns dispatch tables.
7382 @cindex dispatch table
7383 @defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel})
7384 A C statement to output to the stdio stream @var{stream} an assembler
7385 pseudo-instruction to generate a difference between two labels.
7386 @var{value} and @var{rel} are the numbers of two internal labels. The
7387 definitions of these labels are output using
7388 @code{(*targetm.asm_out.internal_label)}, and they must be printed in the same
7389 way here. For example,
7392 fprintf (@var{stream}, "\t.word L%d-L%d\n",
7393 @var{value}, @var{rel})
7396 You must provide this macro on machines where the addresses in a
7397 dispatch table are relative to the table's own address. If defined, GCC
7398 will also use this macro on all machines when producing PIC@.
7399 @var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the
7400 mode and flags can be read.
7403 @defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
7404 This macro should be provided on machines where the addresses
7405 in a dispatch table are absolute.
7407 The definition should be a C statement to output to the stdio stream
7408 @var{stream} an assembler pseudo-instruction to generate a reference to
7409 a label. @var{value} is the number of an internal label whose
7410 definition is output using @code{(*targetm.asm_out.internal_label)}.
7414 fprintf (@var{stream}, "\t.word L%d\n", @var{value})
7418 @defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
7419 Define this if the label before a jump-table needs to be output
7420 specially. The first three arguments are the same as for
7421 @code{(*targetm.asm_out.internal_label)}; the fourth argument is the
7422 jump-table which follows (a @code{jump_insn} containing an
7423 @code{addr_vec} or @code{addr_diff_vec}).
7425 This feature is used on system V to output a @code{swbeg} statement
7428 If this macro is not defined, these labels are output with
7429 @code{(*targetm.asm_out.internal_label)}.
7432 @defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
7433 Define this if something special must be output at the end of a
7434 jump-table. The definition should be a C statement to be executed
7435 after the assembler code for the table is written. It should write
7436 the appropriate code to stdio stream @var{stream}. The argument
7437 @var{table} is the jump-table insn, and @var{num} is the label-number
7438 of the preceding label.
7440 If this macro is not defined, nothing special is output at the end of
7444 @node Exception Region Output
7445 @subsection Assembler Commands for Exception Regions
7447 @c prevent bad page break with this line
7449 This describes commands marking the start and the end of an exception
7452 @defmac EH_FRAME_SECTION_NAME
7453 If defined, a C string constant for the name of the section containing
7454 exception handling frame unwind information. If not defined, GCC will
7455 provide a default definition if the target supports named sections.
7456 @file{crtstuff.c} uses this macro to switch to the appropriate section.
7458 You should define this symbol if your target supports DWARF 2 frame
7459 unwind information and the default definition does not work.
7462 @defmac EH_FRAME_IN_DATA_SECTION
7463 If defined, DWARF 2 frame unwind information will be placed in the
7464 data section even though the target supports named sections. This
7465 might be necessary, for instance, if the system linker does garbage
7466 collection and sections cannot be marked as not to be collected.
7468 Do not define this macro unless @code{TARGET_ASM_NAMED_SECTION} is
7472 @defmac MASK_RETURN_ADDR
7473 An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so
7474 that it does not contain any extraneous set bits in it.
7477 @defmac DWARF2_UNWIND_INFO
7478 Define this macro to 0 if your target supports DWARF 2 frame unwind
7479 information, but it does not yet work with exception handling.
7480 Otherwise, if your target supports this information (if it defines
7481 @samp{INCOMING_RETURN_ADDR_RTX} and either @samp{UNALIGNED_INT_ASM_OP}
7482 or @samp{OBJECT_FORMAT_ELF}), GCC will provide a default definition of
7485 If this macro is defined to 1, the DWARF 2 unwinder will be the default
7486 exception handling mechanism; otherwise, @code{setjmp}/@code{longjmp} will be used by
7489 If this macro is defined to anything, the DWARF 2 unwinder will be used
7490 instead of inline unwinders and @code{__unwind_function} in the non-@code{setjmp} case.
7493 @defmac DWARF_CIE_DATA_ALIGNMENT
7494 This macro need only be defined if the target might save registers in the
7495 function prologue at an offset to the stack pointer that is not aligned to
7496 @code{UNITS_PER_WORD}. The definition should be the negative minimum
7497 alignment if @code{STACK_GROWS_DOWNWARD} is defined, and the positive
7498 minimum alignment otherwise. @xref{SDB and DWARF}. Only applicable if
7499 the target supports DWARF 2 frame unwind information.
7502 @deftypefn {Target Hook} void TARGET_ASM_EXCEPTION_SECTION ()
7503 If defined, a function that switches to the section in which the main
7504 exception table is to be placed (@pxref{Sections}). The default is a
7505 function that switches to a section named @code{.gcc_except_table} on
7506 machines that support named sections via
7507 @code{TARGET_ASM_NAMED_SECTION}, otherwise if @option{-fpic} or
7508 @option{-fPIC} is in effect, the @code{data_section}, otherwise the
7509 @code{readonly_data_section}.
7512 @deftypefn {Target Hook} void TARGET_ASM_EH_FRAME_SECTION ()
7513 If defined, a function that switches to the section in which the DWARF 2
7514 frame unwind information to be placed (@pxref{Sections}). The default
7515 is a function that outputs a standard GAS section directive, if
7516 @code{EH_FRAME_SECTION_NAME} is defined, or else a data section
7517 directive followed by a synthetic label.
7520 @deftypevar {Target Hook} bool TARGET_TERMINATE_DW2_EH_FRAME_INFO
7521 Contains the value true if the target should add a zero word onto the
7522 end of a Dwarf-2 frame info section when used for exception handling.
7523 Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and
7527 @deftypefn {Target Hook} rtx TARGET_DWARF_REGISTER_SPAN (rtx @var{reg})
7528 Given a register, this hook should return a parallel of registers to
7529 represent where to find the register pieces. Define this hook if the
7530 register and its mode are represented in Dwarf in non-contiguous
7531 locations, or if the register should be represented in more than one
7532 register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}.
7533 If not defined, the default is to return @code{NULL_RTX}.
7536 @node Alignment Output
7537 @subsection Assembler Commands for Alignment
7539 @c prevent bad page break with this line
7540 This describes commands for alignment.
7542 @defmac JUMP_ALIGN (@var{label})
7543 The alignment (log base 2) to put in front of @var{label}, which is
7544 a common destination of jumps and has no fallthru incoming edge.
7546 This macro need not be defined if you don't want any special alignment
7547 to be done at such a time. Most machine descriptions do not currently
7550 Unless it's necessary to inspect the @var{label} parameter, it is better
7551 to set the variable @var{align_jumps} in the target's
7552 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
7553 selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation.
7556 @defmac LABEL_ALIGN_AFTER_BARRIER (@var{label})
7557 The alignment (log base 2) to put in front of @var{label}, which follows
7560 This macro need not be defined if you don't want any special alignment
7561 to be done at such a time. Most machine descriptions do not currently
7565 @defmac LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP
7566 The maximum number of bytes to skip when applying
7567 @code{LABEL_ALIGN_AFTER_BARRIER}. This works only if
7568 @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
7571 @defmac LOOP_ALIGN (@var{label})
7572 The alignment (log base 2) to put in front of @var{label}, which follows
7573 a @code{NOTE_INSN_LOOP_BEG} note.
7575 This macro need not be defined if you don't want any special alignment
7576 to be done at such a time. Most machine descriptions do not currently
7579 Unless it's necessary to inspect the @var{label} parameter, it is better
7580 to set the variable @code{align_loops} in the target's
7581 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
7582 selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation.
7585 @defmac LOOP_ALIGN_MAX_SKIP
7586 The maximum number of bytes to skip when applying @code{LOOP_ALIGN}.
7587 This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
7590 @defmac LABEL_ALIGN (@var{label})
7591 The alignment (log base 2) to put in front of @var{label}.
7592 If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment,
7593 the maximum of the specified values is used.
7595 Unless it's necessary to inspect the @var{label} parameter, it is better
7596 to set the variable @code{align_labels} in the target's
7597 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
7598 selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation.
7601 @defmac LABEL_ALIGN_MAX_SKIP
7602 The maximum number of bytes to skip when applying @code{LABEL_ALIGN}.
7603 This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
7606 @defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
7607 A C statement to output to the stdio stream @var{stream} an assembler
7608 instruction to advance the location counter by @var{nbytes} bytes.
7609 Those bytes should be zero when loaded. @var{nbytes} will be a C
7610 expression of type @code{int}.
7613 @defmac ASM_NO_SKIP_IN_TEXT
7614 Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the
7615 text section because it fails to put zeros in the bytes that are skipped.
7616 This is true on many Unix systems, where the pseudo--op to skip bytes
7617 produces no-op instructions rather than zeros when used in the text
7621 @defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
7622 A C statement to output to the stdio stream @var{stream} an assembler
7623 command to advance the location counter to a multiple of 2 to the
7624 @var{power} bytes. @var{power} will be a C expression of type @code{int}.
7627 @defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power})
7628 Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used
7629 for padding, if necessary.
7632 @defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip})
7633 A C statement to output to the stdio stream @var{stream} an assembler
7634 command to advance the location counter to a multiple of 2 to the
7635 @var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to
7636 satisfy the alignment request. @var{power} and @var{max_skip} will be
7637 a C expression of type @code{int}.
7641 @node Debugging Info
7642 @section Controlling Debugging Information Format
7644 @c prevent bad page break with this line
7645 This describes how to specify debugging information.
7648 * All Debuggers:: Macros that affect all debugging formats uniformly.
7649 * DBX Options:: Macros enabling specific options in DBX format.
7650 * DBX Hooks:: Hook macros for varying DBX format.
7651 * File Names and DBX:: Macros controlling output of file names in DBX format.
7652 * SDB and DWARF:: Macros for SDB (COFF) and DWARF formats.
7653 * VMS Debug:: Macros for VMS debug format.
7657 @subsection Macros Affecting All Debugging Formats
7659 @c prevent bad page break with this line
7660 These macros affect all debugging formats.
7662 @defmac DBX_REGISTER_NUMBER (@var{regno})
7663 A C expression that returns the DBX register number for the compiler
7664 register number @var{regno}. In the default macro provided, the value
7665 of this expression will be @var{regno} itself. But sometimes there are
7666 some registers that the compiler knows about and DBX does not, or vice
7667 versa. In such cases, some register may need to have one number in the
7668 compiler and another for DBX@.
7670 If two registers have consecutive numbers inside GCC, and they can be
7671 used as a pair to hold a multiword value, then they @emph{must} have
7672 consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}.
7673 Otherwise, debuggers will be unable to access such a pair, because they
7674 expect register pairs to be consecutive in their own numbering scheme.
7676 If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that
7677 does not preserve register pairs, then what you must do instead is
7678 redefine the actual register numbering scheme.
7681 @defmac DEBUGGER_AUTO_OFFSET (@var{x})
7682 A C expression that returns the integer offset value for an automatic
7683 variable having address @var{x} (an RTL expression). The default
7684 computation assumes that @var{x} is based on the frame-pointer and
7685 gives the offset from the frame-pointer. This is required for targets
7686 that produce debugging output for DBX or COFF-style debugging output
7687 for SDB and allow the frame-pointer to be eliminated when the
7688 @option{-g} options is used.
7691 @defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x})
7692 A C expression that returns the integer offset value for an argument
7693 having address @var{x} (an RTL expression). The nominal offset is
7697 @defmac PREFERRED_DEBUGGING_TYPE
7698 A C expression that returns the type of debugging output GCC should
7699 produce when the user specifies just @option{-g}. Define
7700 this if you have arranged for GCC to support more than one format of
7701 debugging output. Currently, the allowable values are @code{DBX_DEBUG},
7702 @code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG},
7703 @code{XCOFF_DEBUG}, @code{VMS_DEBUG}, and @code{VMS_AND_DWARF2_DEBUG}.
7705 When the user specifies @option{-ggdb}, GCC normally also uses the
7706 value of this macro to select the debugging output format, but with two
7707 exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined and
7708 @code{LINKER_DOES_NOT_WORK_WITH_DWARF2} is not defined, GCC uses the
7709 value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is
7710 defined, GCC uses @code{DBX_DEBUG}.
7712 The value of this macro only affects the default debugging output; the
7713 user can always get a specific type of output by using @option{-gstabs},
7714 @option{-gcoff}, @option{-gdwarf-1}, @option{-gdwarf-2}, @option{-gxcoff},
7719 @subsection Specific Options for DBX Output
7721 @c prevent bad page break with this line
7722 These are specific options for DBX output.
7724 @defmac DBX_DEBUGGING_INFO
7725 Define this macro if GCC should produce debugging output for DBX
7726 in response to the @option{-g} option.
7729 @defmac XCOFF_DEBUGGING_INFO
7730 Define this macro if GCC should produce XCOFF format debugging output
7731 in response to the @option{-g} option. This is a variant of DBX format.
7734 @defmac DEFAULT_GDB_EXTENSIONS
7735 Define this macro to control whether GCC should by default generate
7736 GDB's extended version of DBX debugging information (assuming DBX-format
7737 debugging information is enabled at all). If you don't define the
7738 macro, the default is 1: always generate the extended information
7739 if there is any occasion to.
7742 @defmac DEBUG_SYMS_TEXT
7743 Define this macro if all @code{.stabs} commands should be output while
7744 in the text section.
7747 @defmac ASM_STABS_OP
7748 A C string constant, including spacing, naming the assembler pseudo op to
7749 use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol.
7750 If you don't define this macro, @code{"\t.stabs\t"} is used. This macro
7751 applies only to DBX debugging information format.
7754 @defmac ASM_STABD_OP
7755 A C string constant, including spacing, naming the assembler pseudo op to
7756 use instead of @code{"\t.stabd\t"} to define a debugging symbol whose
7757 value is the current location. If you don't define this macro,
7758 @code{"\t.stabd\t"} is used. This macro applies only to DBX debugging
7762 @defmac ASM_STABN_OP
7763 A C string constant, including spacing, naming the assembler pseudo op to
7764 use instead of @code{"\t.stabn\t"} to define a debugging symbol with no
7765 name. If you don't define this macro, @code{"\t.stabn\t"} is used. This
7766 macro applies only to DBX debugging information format.
7769 @defmac DBX_NO_XREFS
7770 Define this macro if DBX on your system does not support the construct
7771 @samp{xs@var{tagname}}. On some systems, this construct is used to
7772 describe a forward reference to a structure named @var{tagname}.
7773 On other systems, this construct is not supported at all.
7776 @defmac DBX_CONTIN_LENGTH
7777 A symbol name in DBX-format debugging information is normally
7778 continued (split into two separate @code{.stabs} directives) when it
7779 exceeds a certain length (by default, 80 characters). On some
7780 operating systems, DBX requires this splitting; on others, splitting
7781 must not be done. You can inhibit splitting by defining this macro
7782 with the value zero. You can override the default splitting-length by
7783 defining this macro as an expression for the length you desire.
7786 @defmac DBX_CONTIN_CHAR
7787 Normally continuation is indicated by adding a @samp{\} character to
7788 the end of a @code{.stabs} string when a continuation follows. To use
7789 a different character instead, define this macro as a character
7790 constant for the character you want to use. Do not define this macro
7791 if backslash is correct for your system.
7794 @defmac DBX_STATIC_STAB_DATA_SECTION
7795 Define this macro if it is necessary to go to the data section before
7796 outputting the @samp{.stabs} pseudo-op for a non-global static
7800 @defmac DBX_TYPE_DECL_STABS_CODE
7801 The value to use in the ``code'' field of the @code{.stabs} directive
7802 for a typedef. The default is @code{N_LSYM}.
7805 @defmac DBX_STATIC_CONST_VAR_CODE
7806 The value to use in the ``code'' field of the @code{.stabs} directive
7807 for a static variable located in the text section. DBX format does not
7808 provide any ``right'' way to do this. The default is @code{N_FUN}.
7811 @defmac DBX_REGPARM_STABS_CODE
7812 The value to use in the ``code'' field of the @code{.stabs} directive
7813 for a parameter passed in registers. DBX format does not provide any
7814 ``right'' way to do this. The default is @code{N_RSYM}.
7817 @defmac DBX_REGPARM_STABS_LETTER
7818 The letter to use in DBX symbol data to identify a symbol as a parameter
7819 passed in registers. DBX format does not customarily provide any way to
7820 do this. The default is @code{'P'}.
7823 @defmac DBX_MEMPARM_STABS_LETTER
7824 The letter to use in DBX symbol data to identify a symbol as a stack
7825 parameter. The default is @code{'p'}.
7828 @defmac DBX_FUNCTION_FIRST
7829 Define this macro if the DBX information for a function and its
7830 arguments should precede the assembler code for the function. Normally,
7831 in DBX format, the debugging information entirely follows the assembler
7835 @defmac DBX_BLOCKS_FUNCTION_RELATIVE
7836 Define this macro if the value of a symbol describing the scope of a
7837 block (@code{N_LBRAC} or @code{N_RBRAC}) should be relative to the start
7838 of the enclosing function. Normally, GCC uses an absolute address.
7841 @defmac DBX_USE_BINCL
7842 Define this macro if GCC should generate @code{N_BINCL} and
7843 @code{N_EINCL} stabs for included header files, as on Sun systems. This
7844 macro also directs GCC to output a type number as a pair of a file
7845 number and a type number within the file. Normally, GCC does not
7846 generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single
7847 number for a type number.
7851 @subsection Open-Ended Hooks for DBX Format
7853 @c prevent bad page break with this line
7854 These are hooks for DBX format.
7856 @defmac DBX_OUTPUT_LBRAC (@var{stream}, @var{name})
7857 Define this macro to say how to output to @var{stream} the debugging
7858 information for the start of a scope level for variable names. The
7859 argument @var{name} is the name of an assembler symbol (for use with
7860 @code{assemble_name}) whose value is the address where the scope begins.
7863 @defmac DBX_OUTPUT_RBRAC (@var{stream}, @var{name})
7864 Like @code{DBX_OUTPUT_LBRAC}, but for the end of a scope level.
7867 @defmac DBX_OUTPUT_NFUN (@var{stream}, @var{lscope_label}, @var{decl})
7868 Define this macro if the target machine requires special handling to
7869 output an @code{N_FUN} entry for the function @var{decl}.
7872 @defmac DBX_OUTPUT_FUNCTION_END (@var{stream}, @var{function})
7873 Define this macro if the target machine requires special output at the
7874 end of the debugging information for a function. The definition should
7875 be a C statement (sans semicolon) to output the appropriate information
7876 to @var{stream}. @var{function} is the @code{FUNCTION_DECL} node for
7880 @defmac DBX_OUTPUT_STANDARD_TYPES (@var{syms})
7881 Define this macro if you need to control the order of output of the
7882 standard data types at the beginning of compilation. The argument
7883 @var{syms} is a @code{tree} which is a chain of all the predefined
7884 global symbols, including names of data types.
7886 Normally, DBX output starts with definitions of the types for integers
7887 and characters, followed by all the other predefined types of the
7888 particular language in no particular order.
7890 On some machines, it is necessary to output different particular types
7891 first. To do this, define @code{DBX_OUTPUT_STANDARD_TYPES} to output
7892 those symbols in the necessary order. Any predefined types that you
7893 don't explicitly output will be output afterward in no particular order.
7895 Be careful not to define this macro so that it works only for C@. There
7896 are no global variables to access most of the built-in types, because
7897 another language may have another set of types. The way to output a
7898 particular type is to look through @var{syms} to see if you can find it.
7904 for (decl = syms; decl; decl = TREE_CHAIN (decl))
7905 if (!strcmp (IDENTIFIER_POINTER (DECL_NAME (decl)),
7907 dbxout_symbol (decl);
7913 This does nothing if the expected type does not exist.
7915 See the function @code{init_decl_processing} in @file{c-decl.c} to find
7916 the names to use for all the built-in C types.
7918 Here is another way of finding a particular type:
7920 @c this is still overfull. --mew 10feb93
7924 for (decl = syms; decl; decl = TREE_CHAIN (decl))
7925 if (TREE_CODE (decl) == TYPE_DECL
7926 && (TREE_CODE (TREE_TYPE (decl))
7928 && TYPE_PRECISION (TREE_TYPE (decl)) == 16
7929 && TYPE_UNSIGNED (TREE_TYPE (decl)))
7931 /* @r{This must be @code{unsigned short}.} */
7932 dbxout_symbol (decl);
7939 @defmac NO_DBX_FUNCTION_END
7940 Some stabs encapsulation formats (in particular ECOFF), cannot handle the
7941 @code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct.
7942 On those machines, define this macro to turn this feature off without
7943 disturbing the rest of the gdb extensions.
7946 @node File Names and DBX
7947 @subsection File Names in DBX Format
7949 @c prevent bad page break with this line
7950 This describes file names in DBX format.
7952 @defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name})
7953 A C statement to output DBX debugging information to the stdio stream
7954 @var{stream} which indicates that file @var{name} is the main source
7955 file---the file specified as the input file for compilation.
7956 This macro is called only once, at the beginning of compilation.
7958 This macro need not be defined if the standard form of output
7959 for DBX debugging information is appropriate.
7962 @defmac DBX_OUTPUT_MAIN_SOURCE_DIRECTORY (@var{stream}, @var{name})
7963 A C statement to output DBX debugging information to the stdio stream
7964 @var{stream} which indicates that the current directory during
7965 compilation is named @var{name}.
7967 This macro need not be defined if the standard form of output
7968 for DBX debugging information is appropriate.
7971 @defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name})
7972 A C statement to output DBX debugging information at the end of
7973 compilation of the main source file @var{name}.
7975 If you don't define this macro, nothing special is output at the end
7976 of compilation, which is correct for most machines.
7981 @subsection Macros for SDB and DWARF Output
7983 @c prevent bad page break with this line
7984 Here are macros for SDB and DWARF output.
7986 @defmac SDB_DEBUGGING_INFO
7987 Define this macro if GCC should produce COFF-style debugging output
7988 for SDB in response to the @option{-g} option.
7991 @defmac DWARF_DEBUGGING_INFO
7992 Define this macro if GCC should produce dwarf format debugging output
7993 in response to the @option{-g} option.
7996 @defmac DWARF2_DEBUGGING_INFO
7997 Define this macro if GCC should produce dwarf version 2 format
7998 debugging output in response to the @option{-g} option.
8000 To support optional call frame debugging information, you must also
8001 define @code{INCOMING_RETURN_ADDR_RTX} and either set
8002 @code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the
8003 prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save}
8004 as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't.
8007 @defmac DWARF2_FRAME_INFO
8008 Define this macro to a nonzero value if GCC should always output
8009 Dwarf 2 frame information. If @code{DWARF2_UNWIND_INFO}
8010 (@pxref{Exception Region Output} is nonzero, GCC will output this
8011 information not matter how you define @code{DWARF2_FRAME_INFO}.
8014 @defmac LINKER_DOES_NOT_WORK_WITH_DWARF2
8015 Define this macro if the linker does not work with Dwarf version 2.
8016 Normally, if the user specifies only @option{-ggdb} GCC will use Dwarf
8017 version 2 if available; this macro disables this. See the description
8018 of the @code{PREFERRED_DEBUGGING_TYPE} macro for more details.
8021 @defmac DWARF2_GENERATE_TEXT_SECTION_LABEL
8022 By default, the Dwarf 2 debugging information generator will generate a
8023 label to mark the beginning of the text section. If it is better simply
8024 to use the name of the text section itself, rather than an explicit label,
8025 to indicate the beginning of the text section, define this macro to zero.
8028 @defmac DWARF2_ASM_LINE_DEBUG_INFO
8029 Define this macro to be a nonzero value if the assembler can generate Dwarf 2
8030 line debug info sections. This will result in much more compact line number
8031 tables, and hence is desirable if it works.
8034 @defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
8035 A C statement to issue assembly directives that create a difference
8036 between the two given labels, using an integer of the given size.
8039 @defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label})
8040 A C statement to issue assembly directives that create a
8041 section-relative reference to the given label, using an integer of the
8045 @defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label})
8046 A C statement to issue assembly directives that create a self-relative
8047 reference to the given label, using an integer of the given size.
8050 @defmac PUT_SDB_@dots{}
8051 Define these macros to override the assembler syntax for the special
8052 SDB assembler directives. See @file{sdbout.c} for a list of these
8053 macros and their arguments. If the standard syntax is used, you need
8054 not define them yourself.
8058 Some assemblers do not support a semicolon as a delimiter, even between
8059 SDB assembler directives. In that case, define this macro to be the
8060 delimiter to use (usually @samp{\n}). It is not necessary to define
8061 a new set of @code{PUT_SDB_@var{op}} macros if this is the only change
8065 @defmac SDB_GENERATE_FAKE
8066 Define this macro to override the usual method of constructing a dummy
8067 name for anonymous structure and union types. See @file{sdbout.c} for
8071 @defmac SDB_ALLOW_UNKNOWN_REFERENCES
8072 Define this macro to allow references to unknown structure,
8073 union, or enumeration tags to be emitted. Standard COFF does not
8074 allow handling of unknown references, MIPS ECOFF has support for
8078 @defmac SDB_ALLOW_FORWARD_REFERENCES
8079 Define this macro to allow references to structure, union, or
8080 enumeration tags that have not yet been seen to be handled. Some
8081 assemblers choke if forward tags are used, while some require it.
8086 @subsection Macros for VMS Debug Format
8088 @c prevent bad page break with this line
8089 Here are macros for VMS debug format.
8091 @defmac VMS_DEBUGGING_INFO
8092 Define this macro if GCC should produce debugging output for VMS
8093 in response to the @option{-g} option. The default behavior for VMS
8094 is to generate minimal debug info for a traceback in the absence of
8095 @option{-g} unless explicitly overridden with @option{-g0}. This
8096 behavior is controlled by @code{OPTIMIZATION_OPTIONS} and
8097 @code{OVERRIDE_OPTIONS}.
8100 @node Floating Point
8101 @section Cross Compilation and Floating Point
8102 @cindex cross compilation and floating point
8103 @cindex floating point and cross compilation
8105 While all modern machines use twos-complement representation for integers,
8106 there are a variety of representations for floating point numbers. This
8107 means that in a cross-compiler the representation of floating point numbers
8108 in the compiled program may be different from that used in the machine
8109 doing the compilation.
8111 Because different representation systems may offer different amounts of
8112 range and precision, all floating point constants must be represented in
8113 the target machine's format. Therefore, the cross compiler cannot
8114 safely use the host machine's floating point arithmetic; it must emulate
8115 the target's arithmetic. To ensure consistency, GCC always uses
8116 emulation to work with floating point values, even when the host and
8117 target floating point formats are identical.
8119 The following macros are provided by @file{real.h} for the compiler to
8120 use. All parts of the compiler which generate or optimize
8121 floating-point calculations must use these macros. They may evaluate
8122 their operands more than once, so operands must not have side effects.
8124 @defmac REAL_VALUE_TYPE
8125 The C data type to be used to hold a floating point value in the target
8126 machine's format. Typically this is a @code{struct} containing an
8127 array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque
8131 @deftypefn Macro int REAL_VALUES_EQUAL (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
8132 Compares for equality the two values, @var{x} and @var{y}. If the target
8133 floating point format supports negative zeroes and/or NaNs,
8134 @samp{REAL_VALUES_EQUAL (-0.0, 0.0)} is true, and
8135 @samp{REAL_VALUES_EQUAL (NaN, NaN)} is false.
8138 @deftypefn Macro int REAL_VALUES_LESS (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
8139 Tests whether @var{x} is less than @var{y}.
8142 @deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x})
8143 Truncates @var{x} to a signed integer, rounding toward zero.
8146 @deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x})
8147 Truncates @var{x} to an unsigned integer, rounding toward zero. If
8148 @var{x} is negative, returns zero.
8151 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, enum machine_mode @var{mode})
8152 Converts @var{string} into a floating point number in the target machine's
8153 representation for mode @var{mode}. This routine can handle both
8154 decimal and hexadecimal floating point constants, using the syntax
8155 defined by the C language for both.
8158 @deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x})
8159 Returns 1 if @var{x} is negative (including negative zero), 0 otherwise.
8162 @deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x})
8163 Determines whether @var{x} represents infinity (positive or negative).
8166 @deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x})
8167 Determines whether @var{x} represents a ``NaN'' (not-a-number).
8170 @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})
8171 Calculates an arithmetic operation on the two floating point values
8172 @var{x} and @var{y}, storing the result in @var{output} (which must be a
8175 The operation to be performed is specified by @var{code}. Only the
8176 following codes are supported: @code{PLUS_EXPR}, @code{MINUS_EXPR},
8177 @code{MULT_EXPR}, @code{RDIV_EXPR}, @code{MAX_EXPR}, @code{MIN_EXPR}.
8179 If @code{REAL_ARITHMETIC} is asked to evaluate division by zero and the
8180 target's floating point format cannot represent infinity, it will call
8181 @code{abort}. Callers should check for this situation first, using
8182 @code{MODE_HAS_INFINITIES}. @xref{Storage Layout}.
8185 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x})
8186 Returns the negative of the floating point value @var{x}.
8189 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x})
8190 Returns the absolute value of @var{x}.
8193 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_TRUNCATE (REAL_VALUE_TYPE @var{mode}, enum machine_mode @var{x})
8194 Truncates the floating point value @var{x} to fit in @var{mode}. The
8195 return value is still a full-size @code{REAL_VALUE_TYPE}, but it has an
8196 appropriate bit pattern to be output asa floating constant whose
8197 precision accords with mode @var{mode}.
8200 @deftypefn Macro void REAL_VALUE_TO_INT (HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, REAL_VALUE_TYPE @var{x})
8201 Converts a floating point value @var{x} into a double-precision integer
8202 which is then stored into @var{low} and @var{high}. If the value is not
8203 integral, it is truncated.
8206 @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})
8207 Converts a double-precision integer found in @var{low} and @var{high},
8208 into a floating point value which is then stored into @var{x}. The
8209 value is truncated to fit in mode @var{mode}.
8212 @node Mode Switching
8213 @section Mode Switching Instructions
8214 @cindex mode switching
8215 The following macros control mode switching optimizations:
8217 @defmac OPTIMIZE_MODE_SWITCHING (@var{entity})
8218 Define this macro if the port needs extra instructions inserted for mode
8219 switching in an optimizing compilation.
8221 For an example, the SH4 can perform both single and double precision
8222 floating point operations, but to perform a single precision operation,
8223 the FPSCR PR bit has to be cleared, while for a double precision
8224 operation, this bit has to be set. Changing the PR bit requires a general
8225 purpose register as a scratch register, hence these FPSCR sets have to
8226 be inserted before reload, i.e.@: you can't put this into instruction emitting
8227 or @code{TARGET_MACHINE_DEPENDENT_REORG}.
8229 You can have multiple entities that are mode-switched, and select at run time
8230 which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should
8231 return nonzero for any @var{entity} that needs mode-switching.
8232 If you define this macro, you also have to define
8233 @code{NUM_MODES_FOR_MODE_SWITCHING}, @code{MODE_NEEDED},
8234 @code{MODE_PRIORITY_TO_MODE} and @code{EMIT_MODE_SET}.
8235 @code{NORMAL_MODE} is optional.
8238 @defmac NUM_MODES_FOR_MODE_SWITCHING
8239 If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as
8240 initializer for an array of integers. Each initializer element
8241 N refers to an entity that needs mode switching, and specifies the number
8242 of different modes that might need to be set for this entity.
8243 The position of the initializer in the initializer - starting counting at
8244 zero - determines the integer that is used to refer to the mode-switched
8246 In macros that take mode arguments / yield a mode result, modes are
8247 represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode
8248 switch is needed / supplied.
8251 @defmac MODE_NEEDED (@var{entity}, @var{insn})
8252 @var{entity} is an integer specifying a mode-switched entity. If
8253 @code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to
8254 return an integer value not larger than the corresponding element in
8255 @code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must
8256 be switched into prior to the execution of @var{insn}.
8259 @defmac NORMAL_MODE (@var{entity})
8260 If this macro is defined, it is evaluated for every @var{entity} that needs
8261 mode switching. It should evaluate to an integer, which is a mode that
8262 @var{entity} is assumed to be switched to at function entry and exit.
8265 @defmac MODE_PRIORITY_TO_MODE (@var{entity}, @var{n})
8266 This macro specifies the order in which modes for @var{entity} are processed.
8267 0 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the
8268 lowest. The value of the macro should be an integer designating a mode
8269 for @var{entity}. For any fixed @var{entity}, @code{mode_priority_to_mode}
8270 (@var{entity}, @var{n}) shall be a bijection in 0 @dots{}
8271 @code{num_modes_for_mode_switching[@var{entity}] - 1}.
8274 @defmac EMIT_MODE_SET (@var{entity}, @var{mode}, @var{hard_regs_live})
8275 Generate one or more insns to set @var{entity} to @var{mode}.
8276 @var{hard_reg_live} is the set of hard registers live at the point where
8277 the insn(s) are to be inserted.
8280 @node Target Attributes
8281 @section Defining target-specific uses of @code{__attribute__}
8282 @cindex target attributes
8283 @cindex machine attributes
8284 @cindex attributes, target-specific
8286 Target-specific attributes may be defined for functions, data and types.
8287 These are described using the following target hooks; they also need to
8288 be documented in @file{extend.texi}.
8290 @deftypevr {Target Hook} {const struct attribute_spec *} TARGET_ATTRIBUTE_TABLE
8291 If defined, this target hook points to an array of @samp{struct
8292 attribute_spec} (defined in @file{tree.h}) specifying the machine
8293 specific attributes for this target and some of the restrictions on the
8294 entities to which these attributes are applied and the arguments they
8298 @deftypefn {Target Hook} int TARGET_COMP_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2})
8299 If defined, this target hook is a function which returns zero if the attributes on
8300 @var{type1} and @var{type2} are incompatible, one if they are compatible,
8301 and two if they are nearly compatible (which causes a warning to be
8302 generated). If this is not defined, machine-specific attributes are
8303 supposed always to be compatible.
8306 @deftypefn {Target Hook} void TARGET_SET_DEFAULT_TYPE_ATTRIBUTES (tree @var{type})
8307 If defined, this target hook is a function which assigns default attributes to
8308 newly defined @var{type}.
8311 @deftypefn {Target Hook} tree TARGET_MERGE_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2})
8312 Define this target hook if the merging of type attributes needs special
8313 handling. If defined, the result is a list of the combined
8314 @code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed
8315 that @code{comptypes} has already been called and returned 1. This
8316 function may call @code{merge_attributes} to handle machine-independent
8320 @deftypefn {Target Hook} tree TARGET_MERGE_DECL_ATTRIBUTES (tree @var{olddecl}, tree @var{newdecl})
8321 Define this target hook if the merging of decl attributes needs special
8322 handling. If defined, the result is a list of the combined
8323 @code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}.
8324 @var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of
8325 when this is needed are when one attribute overrides another, or when an
8326 attribute is nullified by a subsequent definition. This function may
8327 call @code{merge_attributes} to handle machine-independent merging.
8329 @findex TARGET_DLLIMPORT_DECL_ATTRIBUTES
8330 If the only target-specific handling you require is @samp{dllimport} for
8331 Windows targets, you should define the macro
8332 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. This links in a function
8333 called @code{merge_dllimport_decl_attributes} which can then be defined
8334 as the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. This is done
8335 in @file{i386/cygwin.h} and @file{i386/i386.c}, for example.
8338 @deftypefn {Target Hook} void TARGET_INSERT_ATTRIBUTES (tree @var{node}, tree *@var{attr_ptr})
8339 Define this target hook if you want to be able to add attributes to a decl
8340 when it is being created. This is normally useful for back ends which
8341 wish to implement a pragma by using the attributes which correspond to
8342 the pragma's effect. The @var{node} argument is the decl which is being
8343 created. The @var{attr_ptr} argument is a pointer to the attribute list
8344 for this decl. The list itself should not be modified, since it may be
8345 shared with other decls, but attributes may be chained on the head of
8346 the list and @code{*@var{attr_ptr}} modified to point to the new
8347 attributes, or a copy of the list may be made if further changes are
8351 @deftypefn {Target Hook} bool TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P (tree @var{fndecl})
8353 This target hook returns @code{true} if it is ok to inline @var{fndecl}
8354 into the current function, despite its having target-specific
8355 attributes, @code{false} otherwise. By default, if a function has a
8356 target specific attribute attached to it, it will not be inlined.
8359 @node MIPS Coprocessors
8360 @section Defining coprocessor specifics for MIPS targets.
8361 @cindex MIPS coprocessor-definition macros
8363 The MIPS specification allows MIPS implementations to have as many as 4
8364 coprocessors, each with as many as 32 private registers. gcc supports
8365 accessing these registers and transferring values between the registers
8366 and memory using asm-ized variables. For example:
8369 register unsigned int cp0count asm ("c0r1");
8375 (``c0r1'' is the default name of register 1 in coprocessor 0; alternate
8376 names may be added as described below, or the default names may be
8377 overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.)
8379 Coprocessor registers are assumed to be epilogue-used; sets to them will
8380 be preserved even if it does not appear that the register is used again
8381 later in the function.
8383 Another note: according to the MIPS spec, coprocessor 1 (if present) is
8384 the FPU. One accesses COP1 registers through standard mips
8385 floating-point support; they are not included in this mechanism.
8387 There is one macro used in defining the MIPS coprocessor interface which
8388 you may want to override in subtargets; it is described below.
8390 @defmac ALL_COP_ADDITIONAL_REGISTER_NAMES
8391 A comma-separated list (with leading comma) of pairs describing the
8392 alternate names of coprocessor registers. The format of each entry should be
8394 @{ @var{alternatename}, @var{register_number}@}
8400 @section Miscellaneous Parameters
8401 @cindex parameters, miscellaneous
8403 @c prevent bad page break with this line
8404 Here are several miscellaneous parameters.
8406 @defmac PREDICATE_CODES
8407 Define this if you have defined special-purpose predicates in the file
8408 @file{@var{machine}.c}. This macro is called within an initializer of an
8409 array of structures. The first field in the structure is the name of a
8410 predicate and the second field is an array of rtl codes. For each
8411 predicate, list all rtl codes that can be in expressions matched by the
8412 predicate. The list should have a trailing comma. Here is an example
8413 of two entries in the list for a typical RISC machine:
8416 #define PREDICATE_CODES \
8417 @{"gen_reg_rtx_operand", @{SUBREG, REG@}@}, \
8418 @{"reg_or_short_cint_operand", @{SUBREG, REG, CONST_INT@}@},
8421 Defining this macro does not affect the generated code (however,
8422 incorrect definitions that omit an rtl code that may be matched by the
8423 predicate can cause the compiler to malfunction). Instead, it allows
8424 the table built by @file{genrecog} to be more compact and efficient,
8425 thus speeding up the compiler. The most important predicates to include
8426 in the list specified by this macro are those used in the most insn
8429 For each predicate function named in @code{PREDICATE_CODES}, a
8430 declaration will be generated in @file{insn-codes.h}.
8433 @defmac SPECIAL_MODE_PREDICATES
8434 Define this if you have special predicates that know special things
8435 about modes. Genrecog will warn about certain forms of
8436 @code{match_operand} without a mode; if the operand predicate is
8437 listed in @code{SPECIAL_MODE_PREDICATES}, the warning will be
8440 Here is an example from the IA-32 port (@code{ext_register_operand}
8441 specially checks for @code{HImode} or @code{SImode} in preparation
8442 for a byte extraction from @code{%ah} etc.).
8445 #define SPECIAL_MODE_PREDICATES \
8446 "ext_register_operand",
8450 @defmac CASE_VECTOR_MODE
8451 An alias for a machine mode name. This is the machine mode that
8452 elements of a jump-table should have.
8455 @defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body})
8456 Optional: return the preferred mode for an @code{addr_diff_vec}
8457 when the minimum and maximum offset are known. If you define this,
8458 it enables extra code in branch shortening to deal with @code{addr_diff_vec}.
8459 To make this work, you also have to define @code{INSN_ALIGN} and
8460 make the alignment for @code{addr_diff_vec} explicit.
8461 The @var{body} argument is provided so that the offset_unsigned and scale
8462 flags can be updated.
8465 @defmac CASE_VECTOR_PC_RELATIVE
8466 Define this macro to be a C expression to indicate when jump-tables
8467 should contain relative addresses. If jump-tables never contain
8468 relative addresses, then you need not define this macro.
8471 @defmac CASE_DROPS_THROUGH
8472 Define this if control falls through a @code{case} insn when the index
8473 value is out of range. This means the specified default-label is
8474 actually ignored by the @code{case} insn proper.
8477 @defmac CASE_VALUES_THRESHOLD
8478 Define this to be the smallest number of different values for which it
8479 is best to use a jump-table instead of a tree of conditional branches.
8480 The default is four for machines with a @code{casesi} instruction and
8481 five otherwise. This is best for most machines.
8484 @defmac CASE_USE_BIT_TESTS
8485 Define this macro to be a C expression to indicate whether C switch
8486 statements may be implemented by a sequence of bit tests. This is
8487 advantageous on processors that can efficiently implement left shift
8488 of 1 by the number of bits held in a register, but inappropriate on
8489 targets that would require a loop. By default, this macro returns
8490 @code{true} if the target defines an @code{ashlsi3} pattern, and
8491 @code{false} otherwise.
8494 @defmac WORD_REGISTER_OPERATIONS
8495 Define this macro if operations between registers with integral mode
8496 smaller than a word are always performed on the entire register.
8497 Most RISC machines have this property and most CISC machines do not.
8500 @defmac LOAD_EXTEND_OP (@var{mode})
8501 Define this macro to be a C expression indicating when insns that read
8502 memory in @var{mode}, an integral mode narrower than a word, set the
8503 bits outside of @var{mode} to be either the sign-extension or the
8504 zero-extension of the data read. Return @code{SIGN_EXTEND} for values
8505 of @var{mode} for which the
8506 insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and
8507 @code{NIL} for other modes.
8509 This macro is not called with @var{mode} non-integral or with a width
8510 greater than or equal to @code{BITS_PER_WORD}, so you may return any
8511 value in this case. Do not define this macro if it would always return
8512 @code{NIL}. On machines where this macro is defined, you will normally
8513 define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}.
8516 @defmac SHORT_IMMEDIATES_SIGN_EXTEND
8517 Define this macro if loading short immediate values into registers sign
8521 @defmac FIXUNS_TRUNC_LIKE_FIX_TRUNC
8522 Define this macro if the same instructions that convert a floating
8523 point number to a signed fixed point number also convert validly to an
8528 The maximum number of bytes that a single instruction can move quickly
8529 between memory and registers or between two memory locations.
8532 @defmac MAX_MOVE_MAX
8533 The maximum number of bytes that a single instruction can move quickly
8534 between memory and registers or between two memory locations. If this
8535 is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the
8536 constant value that is the largest value that @code{MOVE_MAX} can have
8540 @defmac SHIFT_COUNT_TRUNCATED
8541 A C expression that is nonzero if on this machine the number of bits
8542 actually used for the count of a shift operation is equal to the number
8543 of bits needed to represent the size of the object being shifted. When
8544 this macro is nonzero, the compiler will assume that it is safe to omit
8545 a sign-extend, zero-extend, and certain bitwise `and' instructions that
8546 truncates the count of a shift operation. On machines that have
8547 instructions that act on bit-fields at variable positions, which may
8548 include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED}
8549 also enables deletion of truncations of the values that serve as
8550 arguments to bit-field instructions.
8552 If both types of instructions truncate the count (for shifts) and
8553 position (for bit-field operations), or if no variable-position bit-field
8554 instructions exist, you should define this macro.
8556 However, on some machines, such as the 80386 and the 680x0, truncation
8557 only applies to shift operations and not the (real or pretended)
8558 bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on
8559 such machines. Instead, add patterns to the @file{md} file that include
8560 the implied truncation of the shift instructions.
8562 You need not define this macro if it would always have the value of zero.
8565 @defmac TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
8566 A C expression which is nonzero if on this machine it is safe to
8567 ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
8568 bits (where @var{outprec} is smaller than @var{inprec}) by merely
8569 operating on it as if it had only @var{outprec} bits.
8571 On many machines, this expression can be 1.
8573 @c rearranged this, removed the phrase "it is reported that". this was
8574 @c to fix an overfull hbox. --mew 10feb93
8575 When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for
8576 modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result.
8577 If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in
8578 such cases may improve things.
8581 @defmac STORE_FLAG_VALUE
8582 A C expression describing the value returned by a comparison operator
8583 with an integral mode and stored by a store-flag instruction
8584 (@samp{s@var{cond}}) when the condition is true. This description must
8585 apply to @emph{all} the @samp{s@var{cond}} patterns and all the
8586 comparison operators whose results have a @code{MODE_INT} mode.
8588 A value of 1 or @minus{}1 means that the instruction implementing the
8589 comparison operator returns exactly 1 or @minus{}1 when the comparison is true
8590 and 0 when the comparison is false. Otherwise, the value indicates
8591 which bits of the result are guaranteed to be 1 when the comparison is
8592 true. This value is interpreted in the mode of the comparison
8593 operation, which is given by the mode of the first operand in the
8594 @samp{s@var{cond}} pattern. Either the low bit or the sign bit of
8595 @code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by
8598 If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will
8599 generate code that depends only on the specified bits. It can also
8600 replace comparison operators with equivalent operations if they cause
8601 the required bits to be set, even if the remaining bits are undefined.
8602 For example, on a machine whose comparison operators return an
8603 @code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as
8604 @samp{0x80000000}, saying that just the sign bit is relevant, the
8608 (ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0))
8615 (ashift:SI @var{x} (const_int @var{n}))
8619 where @var{n} is the appropriate shift count to move the bit being
8620 tested into the sign bit.
8622 There is no way to describe a machine that always sets the low-order bit
8623 for a true value, but does not guarantee the value of any other bits,
8624 but we do not know of any machine that has such an instruction. If you
8625 are trying to port GCC to such a machine, include an instruction to
8626 perform a logical-and of the result with 1 in the pattern for the
8627 comparison operators and let us know at @email{gcc@@gcc.gnu.org}.
8629 Often, a machine will have multiple instructions that obtain a value
8630 from a comparison (or the condition codes). Here are rules to guide the
8631 choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions
8636 Use the shortest sequence that yields a valid definition for
8637 @code{STORE_FLAG_VALUE}. It is more efficient for the compiler to
8638 ``normalize'' the value (convert it to, e.g., 1 or 0) than for the
8639 comparison operators to do so because there may be opportunities to
8640 combine the normalization with other operations.
8643 For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being
8644 slightly preferred on machines with expensive jumps and 1 preferred on
8648 As a second choice, choose a value of @samp{0x80000001} if instructions
8649 exist that set both the sign and low-order bits but do not define the
8653 Otherwise, use a value of @samp{0x80000000}.
8656 Many machines can produce both the value chosen for
8657 @code{STORE_FLAG_VALUE} and its negation in the same number of
8658 instructions. On those machines, you should also define a pattern for
8659 those cases, e.g., one matching
8662 (set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C})))
8665 Some machines can also perform @code{and} or @code{plus} operations on
8666 condition code values with less instructions than the corresponding
8667 @samp{s@var{cond}} insn followed by @code{and} or @code{plus}. On those
8668 machines, define the appropriate patterns. Use the names @code{incscc}
8669 and @code{decscc}, respectively, for the patterns which perform
8670 @code{plus} or @code{minus} operations on condition code values. See
8671 @file{rs6000.md} for some examples. The GNU Superoptizer can be used to
8672 find such instruction sequences on other machines.
8674 If this macro is not defined, the default value, 1, is used. You need
8675 not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
8676 instructions, or if the value generated by these instructions is 1.
8679 @defmac FLOAT_STORE_FLAG_VALUE (@var{mode})
8680 A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is
8681 returned when comparison operators with floating-point results are true.
8682 Define this macro on machine that have comparison operations that return
8683 floating-point values. If there are no such operations, do not define
8687 @defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
8688 @defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
8689 A C expression that evaluates to true if the architecture defines a value
8690 for @code{clz} or @code{ctz} with a zero operand. If so, @var{value}
8691 should be set to this value. If this macro is not defined, the value of
8692 @code{clz} or @code{ctz} is assumed to be undefined.
8694 This macro must be defined if the target's expansion for @code{ffs}
8695 relies on a particular value to get correct results. Otherwise it
8696 is not necessary, though it may be used to optimize some corner cases.
8698 Note that regardless of this macro the ``definedness'' of @code{clz}
8699 and @code{ctz} at zero do @emph{not} extend to the builtin functions
8700 visible to the user. Thus one may be free to adjust the value at will
8701 to match the target expansion of these operations without fear of
8706 An alias for the machine mode for pointers. On most machines, define
8707 this to be the integer mode corresponding to the width of a hardware
8708 pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines.
8709 On some machines you must define this to be one of the partial integer
8710 modes, such as @code{PSImode}.
8712 The width of @code{Pmode} must be at least as large as the value of
8713 @code{POINTER_SIZE}. If it is not equal, you must define the macro
8714 @code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended
8718 @defmac FUNCTION_MODE
8719 An alias for the machine mode used for memory references to functions
8720 being called, in @code{call} RTL expressions. On most machines this
8721 should be @code{QImode}.
8724 @defmac INTEGRATE_THRESHOLD (@var{decl})
8725 A C expression for the maximum number of instructions above which the
8726 function @var{decl} should not be inlined. @var{decl} is a
8727 @code{FUNCTION_DECL} node.
8729 The default definition of this macro is 64 plus 8 times the number of
8730 arguments that the function accepts. Some people think a larger
8731 threshold should be used on RISC machines.
8734 @defmac STDC_0_IN_SYSTEM_HEADERS
8735 In normal operation, the preprocessor expands @code{__STDC__} to the
8736 constant 1, to signify that GCC conforms to ISO Standard C@. On some
8737 hosts, like Solaris, the system compiler uses a different convention,
8738 where @code{__STDC__} is normally 0, but is 1 if the user specifies
8739 strict conformance to the C Standard.
8741 Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host
8742 convention when processing system header files, but when processing user
8743 files @code{__STDC__} will always expand to 1.
8746 @defmac NO_IMPLICIT_EXTERN_C
8747 Define this macro if the system header files support C++ as well as C@.
8748 This macro inhibits the usual method of using system header files in
8749 C++, which is to pretend that the file's contents are enclosed in
8750 @samp{extern "C" @{@dots{}@}}.
8755 @defmac REGISTER_TARGET_PRAGMAS ()
8756 Define this macro if you want to implement any target-specific pragmas.
8757 If defined, it is a C expression which makes a series of calls to
8758 @code{c_register_pragma} for each pragma. The macro may also do any
8759 setup required for the pragmas.
8761 The primary reason to define this macro is to provide compatibility with
8762 other compilers for the same target. In general, we discourage
8763 definition of target-specific pragmas for GCC@.
8765 If the pragma can be implemented by attributes then you should consider
8766 defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well.
8768 Preprocessor macros that appear on pragma lines are not expanded. All
8769 @samp{#pragma} directives that do not match any registered pragma are
8770 silently ignored, unless the user specifies @option{-Wunknown-pragmas}.
8773 @deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
8775 Each call to @code{c_register_pragma} establishes one pragma. The
8776 @var{callback} routine will be called when the preprocessor encounters a
8780 #pragma [@var{space}] @var{name} @dots{}
8783 @var{space} is the case-sensitive namespace of the pragma, or
8784 @code{NULL} to put the pragma in the global namespace. The callback
8785 routine receives @var{pfile} as its first argument, which can be passed
8786 on to cpplib's functions if necessary. You can lex tokens after the
8787 @var{name} by calling @code{c_lex}. Tokens that are not read by the
8788 callback will be silently ignored. The end of the line is indicated by
8789 a token of type @code{CPP_EOF}
8791 For an example use of this routine, see @file{c4x.h} and the callback
8792 routines defined in @file{c4x-c.c}.
8794 Note that the use of @code{c_lex} is specific to the C and C++
8795 compilers. It will not work in the Java or Fortran compilers, or any
8796 other language compilers for that matter. Thus if @code{c_lex} is going
8797 to be called from target-specific code, it must only be done so when
8798 building the C and C++ compilers. This can be done by defining the
8799 variables @code{c_target_objs} and @code{cxx_target_objs} in the
8800 target entry in the @file{config.gcc} file. These variables should name
8801 the target-specific, language-specific object file which contains the
8802 code that uses @code{c_lex}. Note it will also be necessary to add a
8803 rule to the makefile fragment pointed to by @code{tmake_file} that shows
8804 how to build this object file.
8809 @defmac HANDLE_SYSV_PRAGMA
8810 Define this macro (to a value of 1) if you want the System V style
8811 pragmas @samp{#pragma pack(<n>)} and @samp{#pragma weak <name>
8812 [=<value>]} to be supported by gcc.
8814 The pack pragma specifies the maximum alignment (in bytes) of fields
8815 within a structure, in much the same way as the @samp{__aligned__} and
8816 @samp{__packed__} @code{__attribute__}s do. A pack value of zero resets
8817 the behavior to the default.
8819 A subtlety for Microsoft Visual C/C++ style bit-field packing
8820 (e.g. -mms-bitfields) for targets that support it:
8821 When a bit-field is inserted into a packed record, the whole size
8822 of the underlying type is used by one or more same-size adjacent
8823 bit-fields (that is, if its long:3, 32 bits is used in the record,
8824 and any additional adjacent long bit-fields are packed into the same
8825 chunk of 32 bits. However, if the size changes, a new field of that
8828 If both MS bit-fields and @samp{__attribute__((packed))} are used,
8829 the latter will take precedence. If @samp{__attribute__((packed))} is
8830 used on a single field when MS bit-fields are in use, it will take
8831 precedence for that field, but the alignment of the rest of the structure
8832 may affect its placement.
8834 The weak pragma only works if @code{SUPPORTS_WEAK} and
8835 @code{ASM_WEAKEN_LABEL} are defined. If enabled it allows the creation
8836 of specifically named weak labels, optionally with a value.
8841 @defmac HANDLE_PRAGMA_PACK_PUSH_POP
8842 Define this macro (to a value of 1) if you want to support the Win32
8843 style pragmas @samp{#pragma pack(push,@var{n})} and @samp{#pragma
8844 pack(pop)}. The @samp{pack(push,@var{n})} pragma specifies the maximum alignment
8845 (in bytes) of fields within a structure, in much the same way as the
8846 @samp{__aligned__} and @samp{__packed__} @code{__attribute__}s do. A
8847 pack value of zero resets the behavior to the default. Successive
8848 invocations of this pragma cause the previous values to be stacked, so
8849 that invocations of @samp{#pragma pack(pop)} will return to the previous
8853 @defmac DOLLARS_IN_IDENTIFIERS
8854 Define this macro to control use of the character @samp{$} in
8855 identifier names for the C family of languages. 0 means @samp{$} is
8856 not allowed by default; 1 means it is allowed. 1 is the default;
8857 there is no need to define this macro in that case.
8860 @defmac NO_DOLLAR_IN_LABEL
8861 Define this macro if the assembler does not accept the character
8862 @samp{$} in label names. By default constructors and destructors in
8863 G++ have @samp{$} in the identifiers. If this macro is defined,
8864 @samp{.} is used instead.
8867 @defmac NO_DOT_IN_LABEL
8868 Define this macro if the assembler does not accept the character
8869 @samp{.} in label names. By default constructors and destructors in G++
8870 have names that use @samp{.}. If this macro is defined, these names
8871 are rewritten to avoid @samp{.}.
8874 @defmac DEFAULT_MAIN_RETURN
8875 Define this macro if the target system expects every program's @code{main}
8876 function to return a standard ``success'' value by default (if no other
8877 value is explicitly returned).
8879 The definition should be a C statement (sans semicolon) to generate the
8880 appropriate rtl instructions. It is used only when compiling the end of
8884 @defmac INSN_SETS_ARE_DELAYED (@var{insn})
8885 Define this macro as a C expression that is nonzero if it is safe for the
8886 delay slot scheduler to place instructions in the delay slot of @var{insn},
8887 even if they appear to use a resource set or clobbered in @var{insn}.
8888 @var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that
8889 every @code{call_insn} has this behavior. On machines where some @code{insn}
8890 or @code{jump_insn} is really a function call and hence has this behavior,
8891 you should define this macro.
8893 You need not define this macro if it would always return zero.
8896 @defmac INSN_REFERENCES_ARE_DELAYED (@var{insn})
8897 Define this macro as a C expression that is nonzero if it is safe for the
8898 delay slot scheduler to place instructions in the delay slot of @var{insn},
8899 even if they appear to set or clobber a resource referenced in @var{insn}.
8900 @var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where
8901 some @code{insn} or @code{jump_insn} is really a function call and its operands
8902 are registers whose use is actually in the subroutine it calls, you should
8903 define this macro. Doing so allows the delay slot scheduler to move
8904 instructions which copy arguments into the argument registers into the delay
8907 You need not define this macro if it would always return zero.
8910 @defmac MULTIPLE_SYMBOL_SPACES
8911 Define this macro if in some cases global symbols from one translation
8912 unit may not be bound to undefined symbols in another translation unit
8913 without user intervention. For instance, under Microsoft Windows
8914 symbols must be explicitly imported from shared libraries (DLLs).
8917 @defmac MD_ASM_CLOBBERS (@var{clobbers})
8918 A C statement that adds to @var{clobbers} @code{STRING_CST} trees for
8919 any hard regs the port wishes to automatically clobber for all asms.
8922 @defmac MAX_INTEGER_COMPUTATION_MODE
8923 Define this to the largest integer machine mode which can be used for
8924 operations other than load, store and copy operations.
8926 You need only define this macro if the target holds values larger than
8927 @code{word_mode} in general purpose registers. Most targets should not define
8931 @defmac MATH_LIBRARY
8932 Define this macro as a C string constant for the linker argument to link
8933 in the system math library, or @samp{""} if the target does not have a
8934 separate math library.
8936 You need only define this macro if the default of @samp{"-lm"} is wrong.
8939 @defmac LIBRARY_PATH_ENV
8940 Define this macro as a C string constant for the environment variable that
8941 specifies where the linker should look for libraries.
8943 You need only define this macro if the default of @samp{"LIBRARY_PATH"}
8947 @defmac TARGET_HAS_F_SETLKW
8948 Define this macro if the target supports file locking with fcntl / F_SETLKW@.
8949 Note that this functionality is part of POSIX@.
8950 Defining @code{TARGET_HAS_F_SETLKW} will enable the test coverage code
8951 to use file locking when exiting a program, which avoids race conditions
8952 if the program has forked.
8955 @defmac MAX_CONDITIONAL_EXECUTE
8957 A C expression for the maximum number of instructions to execute via
8958 conditional execution instructions instead of a branch. A value of
8959 @code{BRANCH_COST}+1 is the default if the machine does not use cc0, and
8960 1 if it does use cc0.
8963 @defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr})
8964 Used if the target needs to perform machine-dependent modifications on the
8965 conditionals used for turning basic blocks into conditionally executed code.
8966 @var{ce_info} points to a data structure, @code{struct ce_if_block}, which
8967 contains information about the currently processed blocks. @var{true_expr}
8968 and @var{false_expr} are the tests that are used for converting the
8969 then-block and the else-block, respectively. Set either @var{true_expr} or
8970 @var{false_expr} to a null pointer if the tests cannot be converted.
8973 @defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr})
8974 Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated
8975 if-statements into conditions combined by @code{and} and @code{or} operations.
8976 @var{bb} contains the basic block that contains the test that is currently
8977 being processed and about to be turned into a condition.
8980 @defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn})
8981 A C expression to modify the @var{PATTERN} of an @var{INSN} that is to
8982 be converted to conditional execution format. @var{ce_info} points to
8983 a data structure, @code{struct ce_if_block}, which contains information
8984 about the currently processed blocks.
8987 @defmac IFCVT_MODIFY_FINAL (@var{ce_info})
8988 A C expression to perform any final machine dependent modifications in
8989 converting code to conditional execution. The involved basic blocks
8990 can be found in the @code{struct ce_if_block} structure that is pointed
8991 to by @var{ce_info}.
8994 @defmac IFCVT_MODIFY_CANCEL (@var{ce_info})
8995 A C expression to cancel any machine dependent modifications in
8996 converting code to conditional execution. The involved basic blocks
8997 can be found in the @code{struct ce_if_block} structure that is pointed
8998 to by @var{ce_info}.
9001 @defmac IFCVT_INIT_EXTRA_FIELDS (@var{ce_info})
9002 A C expression to initialize any extra fields in a @code{struct ce_if_block}
9003 structure, which are defined by the @code{IFCVT_EXTRA_FIELDS} macro.
9006 @defmac IFCVT_EXTRA_FIELDS
9007 If defined, it should expand to a set of field declarations that will be
9008 added to the @code{struct ce_if_block} structure. These should be initialized
9009 by the @code{IFCVT_INIT_EXTRA_FIELDS} macro.
9012 @deftypefn {Target Hook} void TARGET_MACHINE_DEPENDENT_REORG ()
9013 If non-null, this hook performs a target-specific pass over the
9014 instruction stream. The compiler will run it at all optimization levels,
9015 just before the point at which it normally does delayed-branch scheduling.
9017 The exact purpose of the hook varies from target to target. Some use
9018 it to do transformations that are necessary for correctness, such as
9019 laying out in-function constant pools or avoiding hardware hazards.
9020 Others use it as an opportunity to do some machine-dependent optimizations.
9022 You need not implement the hook if it has nothing to do. The default
9026 @deftypefn {Target Hook} void TARGET_INIT_BUILTINS ()
9027 Define this hook if you have any machine-specific built-in functions
9028 that need to be defined. It should be a function that performs the
9031 Machine specific built-in functions can be useful to expand special machine
9032 instructions that would otherwise not normally be generated because
9033 they have no equivalent in the source language (for example, SIMD vector
9034 instructions or prefetch instructions).
9036 To create a built-in function, call the function @code{builtin_function}
9037 which is defined by the language front end. You can use any type nodes set
9038 up by @code{build_common_tree_nodes} and @code{build_common_tree_nodes_2};
9039 only language front ends that use those two functions will call
9040 @samp{TARGET_INIT_BUILTINS}.
9043 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN (tree @var{exp}, rtx @var{target}, rtx @var{subtarget}, enum machine_mode @var{mode}, int @var{ignore})
9045 Expand a call to a machine specific built-in function that was set up by
9046 @samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the
9047 function call; the result should go to @var{target} if that is
9048 convenient, and have mode @var{mode} if that is convenient.
9049 @var{subtarget} may be used as the target for computing one of
9050 @var{exp}'s operands. @var{ignore} is nonzero if the value is to be
9051 ignored. This function should return the result of the call to the
9055 @defmac MD_CAN_REDIRECT_BRANCH (@var{branch1}, @var{branch2})
9057 Take a branch insn in @var{branch1} and another in @var{branch2}.
9058 Return true if redirecting @var{branch1} to the destination of
9059 @var{branch2} is possible.
9061 On some targets, branches may have a limited range. Optimizing the
9062 filling of delay slots can result in branches being redirected, and this
9063 may in turn cause a branch offset to overflow.
9066 @defmac ALLOCATE_INITIAL_VALUE (@var{hard_reg})
9068 When the initial value of a hard register has been copied in a pseudo
9069 register, it is often not necessary to actually allocate another register
9070 to this pseudo register, because the original hard register or a stack slot
9071 it has been saved into can be used. @code{ALLOCATE_INITIAL_VALUE}, if
9072 defined, is called at the start of register allocation once for each
9073 hard register that had its initial value copied by using
9074 @code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}.
9075 Possible values are @code{NULL_RTX}, if you don't want
9076 to do any special allocation, a @code{REG} rtx---that would typically be
9077 the hard register itself, if it is known not to be clobbered---or a
9079 If you are returning a @code{MEM}, this is only a hint for the allocator;
9080 it might decide to use another register anyways.
9081 You may use @code{current_function_leaf_function} in the definition of the
9082 macro, functions that use @code{REG_N_SETS}, to determine if the hard
9083 register in question will not be clobbered.
9086 @defmac TARGET_OBJECT_SUFFIX
9087 Define this macro to be a C string representing the suffix for object
9088 files on your target machine. If you do not define this macro, GCC will
9089 use @samp{.o} as the suffix for object files.
9092 @defmac TARGET_EXECUTABLE_SUFFIX
9093 Define this macro to be a C string representing the suffix to be
9094 automatically added to executable files on your target machine. If you
9095 do not define this macro, GCC will use the null string as the suffix for
9099 @defmac COLLECT_EXPORT_LIST
9100 If defined, @code{collect2} will scan the individual object files
9101 specified on its command line and create an export list for the linker.
9102 Define this macro for systems like AIX, where the linker discards
9103 object files that are not referenced from @code{main} and uses export
9107 @defmac MODIFY_JNI_METHOD_CALL (@var{mdecl})
9108 Define this macro to a C expression representing a variant of the
9109 method call @var{mdecl}, if Java Native Interface (JNI) methods
9110 must be invoked differently from other methods on your target.
9111 For example, on 32-bit Windows, JNI methods must be invoked using
9112 the @code{stdcall} calling convention and this macro is then
9113 defined as this expression:
9116 build_type_attribute_variant (@var{mdecl},
9118 (get_identifier ("stdcall"),
9123 @deftypefn {Target Hook} bool TARGET_CANNOT_MODIFY_JUMPS_P (void)
9124 This target hook returns @code{true} past the point in which new jump
9125 instructions could be created. On machines that require a register for
9126 every jump such as the SHmedia ISA of SH5, this point would typically be
9127 reload, so this target hook should be defined to a function such as:
9131 cannot_modify_jumps_past_reload_p ()
9133 return (reload_completed || reload_in_progress);
9138 @deftypefn {Target Hook} int TARGET_BRANCH_TARGET_REGISTER_CLASS (void)
9139 This target hook returns a register class for which branch target register
9140 optimizations should be applied. All registers in this class should be
9141 usable interchangeably. After reload, registers in this class will be
9142 re-allocated and loads will be hoisted out of loops and be subjected
9143 to inter-block scheduling.
9146 @deftypefn {Target Hook} bool TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED (bool @var{after_prologue_epilogue_gen})
9147 Branch target register optimization will by default exclude callee-saved
9149 that are not already live during the current function; if this target hook
9150 returns true, they will be included. The target code must than make sure
9151 that all target registers in the class returned by
9152 @samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are
9153 saved. @var{after_prologue_epilogue_gen} indicates if prologues and
9154 epilogues have already been generated. Note, even if you only return
9155 true when @var{after_prologue_epilogue_gen} is false, you still are likely
9156 to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET}
9157 to reserve space for caller-saved target registers.
9160 @defmac POWI_MAX_MULTS
9161 If defined, this macro is interpreted as a signed integer C expression
9162 that specifies the maximum number of floating point multiplications
9163 that should be emitted when expanding exponentiation by an integer
9164 constant inline. When this value is defined, exponentiation requiring
9165 more than this number of multiplications is implemented by calling the
9166 system library's @code{pow}, @code{powf} or @code{powl} routines.
9167 The default value places no upper bound on the multiplication count.