1 @c Copyright (C) 1988,1989,1992,1993,1994,1995,1996,1997,1998,1999,2000,2001,
2 @c 2002, 2003, 2004, 2005 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 * Registers:: Naming and describing the hardware registers.
35 * Register Classes:: Defining the classes of hardware registers.
36 * Stack and Calling:: Defining which way the stack grows and by how much.
37 * Varargs:: Defining the varargs macros.
38 * Trampolines:: Code set up at run time to enter a nested function.
39 * Library Calls:: Controlling how library routines are implicitly called.
40 * Addressing Modes:: Defining addressing modes valid for memory operands.
41 * Condition Code:: Defining how insns update the condition code.
42 * Costs:: Defining relative costs of different operations.
43 * Scheduling:: Adjusting the behavior of the instruction scheduler.
44 * Sections:: Dividing storage into text, data, and other sections.
45 * PIC:: Macros for position independent code.
46 * Assembler Format:: Defining how to write insns and pseudo-ops to output.
47 * Debugging Info:: Defining the format of debugging output.
48 * Floating Point:: Handling floating point for cross-compilers.
49 * Mode Switching:: Insertion of mode-switching instructions.
50 * Target Attributes:: Defining target-specific uses of @code{__attribute__}.
51 * MIPS Coprocessors:: MIPS coprocessor support and how to customize it.
52 * PCH Target:: Validity checking for precompiled headers.
53 * C++ ABI:: Controlling C++ ABI changes.
54 * Misc:: Everything else.
57 @node Target Structure
58 @section The Global @code{targetm} Variable
60 @cindex target functions
62 @deftypevar {struct gcc_target} targetm
63 The target @file{.c} file must define the global @code{targetm} variable
64 which contains pointers to functions and data relating to the target
65 machine. The variable is declared in @file{target.h};
66 @file{target-def.h} defines the macro @code{TARGET_INITIALIZER} which is
67 used to initialize the variable, and macros for the default initializers
68 for elements of the structure. The @file{.c} file should override those
69 macros for which the default definition is inappropriate. For example:
72 #include "target-def.h"
74 /* @r{Initialize the GCC target structure.} */
76 #undef TARGET_COMP_TYPE_ATTRIBUTES
77 #define TARGET_COMP_TYPE_ATTRIBUTES @var{machine}_comp_type_attributes
79 struct gcc_target targetm = TARGET_INITIALIZER;
83 Where a macro should be defined in the @file{.c} file in this manner to
84 form part of the @code{targetm} structure, it is documented below as a
85 ``Target Hook'' with a prototype. Many macros will change in future
86 from being defined in the @file{.h} file to being part of the
87 @code{targetm} structure.
90 @section Controlling the Compilation Driver, @file{gcc}
92 @cindex controlling the compilation driver
94 @c prevent bad page break with this line
95 You can control the compilation driver.
97 @defmac SWITCH_TAKES_ARG (@var{char})
98 A C expression which determines whether the option @option{-@var{char}}
99 takes arguments. The value should be the number of arguments that
100 option takes--zero, for many options.
102 By default, this macro is defined as
103 @code{DEFAULT_SWITCH_TAKES_ARG}, which handles the standard options
104 properly. You need not define @code{SWITCH_TAKES_ARG} unless you
105 wish to add additional options which take arguments. Any redefinition
106 should call @code{DEFAULT_SWITCH_TAKES_ARG} and then check for
110 @defmac WORD_SWITCH_TAKES_ARG (@var{name})
111 A C expression which determines whether the option @option{-@var{name}}
112 takes arguments. The value should be the number of arguments that
113 option takes--zero, for many options. This macro rather than
114 @code{SWITCH_TAKES_ARG} is used for multi-character option names.
116 By default, this macro is defined as
117 @code{DEFAULT_WORD_SWITCH_TAKES_ARG}, which handles the standard options
118 properly. You need not define @code{WORD_SWITCH_TAKES_ARG} unless you
119 wish to add additional options which take arguments. Any redefinition
120 should call @code{DEFAULT_WORD_SWITCH_TAKES_ARG} and then check for
124 @defmac SWITCH_CURTAILS_COMPILATION (@var{char})
125 A C expression which determines whether the option @option{-@var{char}}
126 stops compilation before the generation of an executable. The value is
127 boolean, nonzero if the option does stop an executable from being
128 generated, zero otherwise.
130 By default, this macro is defined as
131 @code{DEFAULT_SWITCH_CURTAILS_COMPILATION}, which handles the standard
132 options properly. You need not define
133 @code{SWITCH_CURTAILS_COMPILATION} unless you wish to add additional
134 options which affect the generation of an executable. Any redefinition
135 should call @code{DEFAULT_SWITCH_CURTAILS_COMPILATION} and then check
136 for additional options.
139 @defmac SWITCHES_NEED_SPACES
140 A string-valued C expression which enumerates the options for which
141 the linker needs a space between the option and its argument.
143 If this macro is not defined, the default value is @code{""}.
146 @defmac TARGET_OPTION_TRANSLATE_TABLE
147 If defined, a list of pairs of strings, the first of which is a
148 potential command line target to the @file{gcc} driver program, and the
149 second of which is a space-separated (tabs and other whitespace are not
150 supported) list of options with which to replace the first option. The
151 target defining this list is responsible for assuring that the results
152 are valid. Replacement options may not be the @code{--opt} style, they
153 must be the @code{-opt} style. It is the intention of this macro to
154 provide a mechanism for substitution that affects the multilibs chosen,
155 such as one option that enables many options, some of which select
156 multilibs. Example nonsensical definition, where @option{-malt-abi},
157 @option{-EB}, and @option{-mspoo} cause different multilibs to be chosen:
160 #define TARGET_OPTION_TRANSLATE_TABLE \
161 @{ "-fast", "-march=fast-foo -malt-abi -I/usr/fast-foo" @}, \
162 @{ "-compat", "-EB -malign=4 -mspoo" @}
166 @defmac DRIVER_SELF_SPECS
167 A list of specs for the driver itself. It should be a suitable
168 initializer for an array of strings, with no surrounding braces.
170 The driver applies these specs to its own command line between loading
171 default @file{specs} files (but not command-line specified ones) and
172 choosing the multilib directory or running any subcommands. It
173 applies them in the order given, so each spec can depend on the
174 options added by earlier ones. It is also possible to remove options
175 using @samp{%<@var{option}} in the usual way.
177 This macro can be useful when a port has several interdependent target
178 options. It provides a way of standardizing the command line so
179 that the other specs are easier to write.
181 Do not define this macro if it does not need to do anything.
184 @defmac OPTION_DEFAULT_SPECS
185 A list of specs used to support configure-time default options (i.e.@:
186 @option{--with} options) in the driver. It should be a suitable initializer
187 for an array of structures, each containing two strings, without the
188 outermost pair of surrounding braces.
190 The first item in the pair is the name of the default. This must match
191 the code in @file{config.gcc} for the target. The second item is a spec
192 to apply if a default with this name was specified. The string
193 @samp{%(VALUE)} in the spec will be replaced by the value of the default
194 everywhere it occurs.
196 The driver will apply these specs to its own command line between loading
197 default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using
198 the same mechanism as @code{DRIVER_SELF_SPECS}.
200 Do not define this macro if it does not need to do anything.
204 A C string constant that tells the GCC driver program options to
205 pass to CPP@. It can also specify how to translate options you
206 give to GCC into options for GCC to pass to the CPP@.
208 Do not define this macro if it does not need to do anything.
211 @defmac CPLUSPLUS_CPP_SPEC
212 This macro is just like @code{CPP_SPEC}, but is used for C++, rather
213 than C@. If you do not define this macro, then the value of
214 @code{CPP_SPEC} (if any) will be used instead.
218 A C string constant that tells the GCC driver program options to
219 pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language
221 It can also specify how to translate options you give to GCC into options
222 for GCC to pass to front ends.
224 Do not define this macro if it does not need to do anything.
228 A C string constant that tells the GCC driver program options to
229 pass to @code{cc1plus}. It can also specify how to translate options you
230 give to GCC into options for GCC to pass to the @code{cc1plus}.
232 Do not define this macro if it does not need to do anything.
233 Note that everything defined in CC1_SPEC is already passed to
234 @code{cc1plus} so there is no need to duplicate the contents of
235 CC1_SPEC in CC1PLUS_SPEC@.
239 A C string constant that tells the GCC driver program options to
240 pass to the assembler. It can also specify how to translate options
241 you give to GCC into options for GCC to pass to the assembler.
242 See the file @file{sun3.h} for an example of this.
244 Do not define this macro if it does not need to do anything.
247 @defmac ASM_FINAL_SPEC
248 A C string constant that tells the GCC driver program how to
249 run any programs which cleanup after the normal assembler.
250 Normally, this is not needed. See the file @file{mips.h} for
253 Do not define this macro if it does not need to do anything.
256 @defmac AS_NEEDS_DASH_FOR_PIPED_INPUT
257 Define this macro, with no value, if the driver should give the assembler
258 an argument consisting of a single dash, @option{-}, to instruct it to
259 read from its standard input (which will be a pipe connected to the
260 output of the compiler proper). This argument is given after any
261 @option{-o} option specifying the name of the output file.
263 If you do not define this macro, the assembler is assumed to read its
264 standard input if given no non-option arguments. If your assembler
265 cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct;
266 see @file{mips.h} for instance.
270 A C string constant that tells the GCC driver program options to
271 pass to the linker. It can also specify how to translate options you
272 give to GCC into options for GCC to pass to the linker.
274 Do not define this macro if it does not need to do anything.
278 Another C string constant used much like @code{LINK_SPEC}. The difference
279 between the two is that @code{LIB_SPEC} is used at the end of the
280 command given to the linker.
282 If this macro is not defined, a default is provided that
283 loads the standard C library from the usual place. See @file{gcc.c}.
287 Another C string constant that tells the GCC driver program
288 how and when to place a reference to @file{libgcc.a} into the
289 linker command line. This constant is placed both before and after
290 the value of @code{LIB_SPEC}.
292 If this macro is not defined, the GCC driver provides a default that
293 passes the string @option{-lgcc} to the linker.
296 @defmac REAL_LIBGCC_SPEC
297 By default, if @code{ENABLE_SHARED_LIBGCC} is defined, the
298 @code{LIBGCC_SPEC} is not directly used by the driver program but is
299 instead modified to refer to different versions of @file{libgcc.a}
300 depending on the values of the command line flags @option{-static},
301 @option{-shared}, @option{-static-libgcc}, and @option{-shared-libgcc}. On
302 targets where these modifications are inappropriate, define
303 @code{REAL_LIBGCC_SPEC} instead. @code{REAL_LIBGCC_SPEC} tells the
304 driver how to place a reference to @file{libgcc} on the link command
305 line, but, unlike @code{LIBGCC_SPEC}, it is used unmodified.
308 @defmac USE_LD_AS_NEEDED
309 A macro that controls the modifications to @code{LIBGCC_SPEC}
310 mentioned in @code{REAL_LIBGCC_SPEC}. If nonzero, a spec will be
311 generated that uses --as-needed and the shared libgcc in place of the
312 static exception handler library, when linking without any of
313 @code{-static}, @code{-static-libgcc}, or @code{-shared-libgcc}.
317 If defined, this C string constant is added to @code{LINK_SPEC}.
318 When @code{USE_LD_AS_NEEDED} is zero or undefined, it also affects
319 the modifications to @code{LIBGCC_SPEC} mentioned in
320 @code{REAL_LIBGCC_SPEC}.
323 @defmac STARTFILE_SPEC
324 Another C string constant used much like @code{LINK_SPEC}. The
325 difference between the two is that @code{STARTFILE_SPEC} is used at
326 the very beginning of the command given to the linker.
328 If this macro is not defined, a default is provided that loads the
329 standard C startup file from the usual place. See @file{gcc.c}.
333 Another C string constant used much like @code{LINK_SPEC}. The
334 difference between the two is that @code{ENDFILE_SPEC} is used at
335 the very end of the command given to the linker.
337 Do not define this macro if it does not need to do anything.
340 @defmac THREAD_MODEL_SPEC
341 GCC @code{-v} will print the thread model GCC was configured to use.
342 However, this doesn't work on platforms that are multilibbed on thread
343 models, such as AIX 4.3. On such platforms, define
344 @code{THREAD_MODEL_SPEC} such that it evaluates to a string without
345 blanks that names one of the recognized thread models. @code{%*}, the
346 default value of this macro, will expand to the value of
347 @code{thread_file} set in @file{config.gcc}.
350 @defmac SYSROOT_SUFFIX_SPEC
351 Define this macro to add a suffix to the target sysroot when GCC is
352 configured with a sysroot. This will cause GCC to search for usr/lib,
353 et al, within sysroot+suffix.
356 @defmac SYSROOT_HEADERS_SUFFIX_SPEC
357 Define this macro to add a headers_suffix to the target sysroot when
358 GCC is configured with a sysroot. This will cause GCC to pass the
359 updated sysroot+headers_suffix to CPP, causing it to search for
360 usr/include, et al, within sysroot+headers_suffix.
364 Define this macro to provide additional specifications to put in the
365 @file{specs} file that can be used in various specifications like
368 The definition should be an initializer for an array of structures,
369 containing a string constant, that defines the specification name, and a
370 string constant that provides the specification.
372 Do not define this macro if it does not need to do anything.
374 @code{EXTRA_SPECS} is useful when an architecture contains several
375 related targets, which have various @code{@dots{}_SPECS} which are similar
376 to each other, and the maintainer would like one central place to keep
379 For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to
380 define either @code{_CALL_SYSV} when the System V calling sequence is
381 used or @code{_CALL_AIX} when the older AIX-based calling sequence is
384 The @file{config/rs6000/rs6000.h} target file defines:
387 #define EXTRA_SPECS \
388 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @},
390 #define CPP_SYS_DEFAULT ""
393 The @file{config/rs6000/sysv.h} target file defines:
397 "%@{posix: -D_POSIX_SOURCE @} \
398 %@{mcall-sysv: -D_CALL_SYSV @} \
399 %@{!mcall-sysv: %(cpp_sysv_default) @} \
400 %@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}"
402 #undef CPP_SYSV_DEFAULT
403 #define CPP_SYSV_DEFAULT "-D_CALL_SYSV"
406 while the @file{config/rs6000/eabiaix.h} target file defines
407 @code{CPP_SYSV_DEFAULT} as:
410 #undef CPP_SYSV_DEFAULT
411 #define CPP_SYSV_DEFAULT "-D_CALL_AIX"
415 @defmac LINK_LIBGCC_SPECIAL_1
416 Define this macro if the driver program should find the library
417 @file{libgcc.a}. If you do not define this macro, the driver program will pass
418 the argument @option{-lgcc} to tell the linker to do the search.
421 @defmac LINK_GCC_C_SEQUENCE_SPEC
422 The sequence in which libgcc and libc are specified to the linker.
423 By default this is @code{%G %L %G}.
426 @defmac LINK_COMMAND_SPEC
427 A C string constant giving the complete command line need to execute the
428 linker. When you do this, you will need to update your port each time a
429 change is made to the link command line within @file{gcc.c}. Therefore,
430 define this macro only if you need to completely redefine the command
431 line for invoking the linker and there is no other way to accomplish
432 the effect you need. Overriding this macro may be avoidable by overriding
433 @code{LINK_GCC_C_SEQUENCE_SPEC} instead.
436 @defmac LINK_ELIMINATE_DUPLICATE_LDIRECTORIES
437 A nonzero value causes @command{collect2} to remove duplicate @option{-L@var{directory}} search
438 directories from linking commands. Do not give it a nonzero value if
439 removing duplicate search directories changes the linker's semantics.
442 @defmac MULTILIB_DEFAULTS
443 Define this macro as a C expression for the initializer of an array of
444 string to tell the driver program which options are defaults for this
445 target and thus do not need to be handled specially when using
446 @code{MULTILIB_OPTIONS}.
448 Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in
449 the target makefile fragment or if none of the options listed in
450 @code{MULTILIB_OPTIONS} are set by default.
451 @xref{Target Fragment}.
454 @defmac RELATIVE_PREFIX_NOT_LINKDIR
455 Define this macro to tell @command{gcc} that it should only translate
456 a @option{-B} prefix into a @option{-L} linker option if the prefix
457 indicates an absolute file name.
460 @defmac MD_EXEC_PREFIX
461 If defined, this macro is an additional prefix to try after
462 @code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched
463 when the @option{-b} option is used, or the compiler is built as a cross
464 compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it
465 to the list of directories used to find the assembler in @file{configure.in}.
468 @defmac STANDARD_STARTFILE_PREFIX
469 Define this macro as a C string constant if you wish to override the
470 standard choice of @code{libdir} as the default prefix to
471 try when searching for startup files such as @file{crt0.o}.
472 @code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler
473 is built as a cross compiler.
476 @defmac STANDARD_STARTFILE_PREFIX_1
477 Define this macro as a C string constant if you wish to override the
478 standard choice of @code{/lib} as a prefix to try after the default prefix
479 when searching for startup files such as @file{crt0.o}.
480 @code{STANDARD_STARTFILE_PREFIX_1} is not searched when the compiler
481 is built as a cross compiler.
484 @defmac STANDARD_STARTFILE_PREFIX_2
485 Define this macro as a C string constant if you wish to override the
486 standard choice of @code{/lib} as yet another prefix to try after the
487 default prefix when searching for startup files such as @file{crt0.o}.
488 @code{STANDARD_STARTFILE_PREFIX_2} is not searched when the compiler
489 is built as a cross compiler.
492 @defmac MD_STARTFILE_PREFIX
493 If defined, this macro supplies an additional prefix to try after the
494 standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the
495 @option{-b} option is used, or when the compiler is built as a cross
499 @defmac MD_STARTFILE_PREFIX_1
500 If defined, this macro supplies yet another prefix to try after the
501 standard prefixes. It is not searched when the @option{-b} option is
502 used, or when the compiler is built as a cross compiler.
505 @defmac INIT_ENVIRONMENT
506 Define this macro as a C string constant if you wish to set environment
507 variables for programs called by the driver, such as the assembler and
508 loader. The driver passes the value of this macro to @code{putenv} to
509 initialize the necessary environment variables.
512 @defmac LOCAL_INCLUDE_DIR
513 Define this macro as a C string constant if you wish to override the
514 standard choice of @file{/usr/local/include} as the default prefix to
515 try when searching for local header files. @code{LOCAL_INCLUDE_DIR}
516 comes before @code{SYSTEM_INCLUDE_DIR} in the search order.
518 Cross compilers do not search either @file{/usr/local/include} or its
522 @defmac MODIFY_TARGET_NAME
523 Define this macro if you wish to define command-line switches that
524 modify the default target name.
526 For each switch, you can include a string to be appended to the first
527 part of the configuration name or a string to be deleted from the
528 configuration name, if present. The definition should be an initializer
529 for an array of structures. Each array element should have three
530 elements: the switch name (a string constant, including the initial
531 dash), one of the enumeration codes @code{ADD} or @code{DELETE} to
532 indicate whether the string should be inserted or deleted, and the string
533 to be inserted or deleted (a string constant).
535 For example, on a machine where @samp{64} at the end of the
536 configuration name denotes a 64-bit target and you want the @option{-32}
537 and @option{-64} switches to select between 32- and 64-bit targets, you would
541 #define MODIFY_TARGET_NAME \
542 @{ @{ "-32", DELETE, "64"@}, \
543 @{"-64", ADD, "64"@}@}
547 @defmac SYSTEM_INCLUDE_DIR
548 Define this macro as a C string constant if you wish to specify a
549 system-specific directory to search for header files before the standard
550 directory. @code{SYSTEM_INCLUDE_DIR} comes before
551 @code{STANDARD_INCLUDE_DIR} in the search order.
553 Cross compilers do not use this macro and do not search the directory
557 @defmac STANDARD_INCLUDE_DIR
558 Define this macro as a C string constant if you wish to override the
559 standard choice of @file{/usr/include} as the default prefix to
560 try when searching for header files.
562 Cross compilers ignore this macro and do not search either
563 @file{/usr/include} or its replacement.
566 @defmac STANDARD_INCLUDE_COMPONENT
567 The ``component'' corresponding to @code{STANDARD_INCLUDE_DIR}.
568 See @code{INCLUDE_DEFAULTS}, below, for the description of components.
569 If you do not define this macro, no component is used.
572 @defmac INCLUDE_DEFAULTS
573 Define this macro if you wish to override the entire default search path
574 for include files. For a native compiler, the default search path
575 usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR},
576 @code{SYSTEM_INCLUDE_DIR}, @code{GPLUSPLUS_INCLUDE_DIR}, and
577 @code{STANDARD_INCLUDE_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR}
578 and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile},
579 and specify private search areas for GCC@. The directory
580 @code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs.
582 The definition should be an initializer for an array of structures.
583 Each array element should have four elements: the directory name (a
584 string constant), the component name (also a string constant), a flag
585 for C++-only directories,
586 and a flag showing that the includes in the directory don't need to be
587 wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of
588 the array with a null element.
590 The component name denotes what GNU package the include file is part of,
591 if any, in all uppercase letters. For example, it might be @samp{GCC}
592 or @samp{BINUTILS}. If the package is part of a vendor-supplied
593 operating system, code the component name as @samp{0}.
595 For example, here is the definition used for VAX/VMS:
598 #define INCLUDE_DEFAULTS \
600 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \
601 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \
602 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \
609 Here is the order of prefixes tried for exec files:
613 Any prefixes specified by the user with @option{-B}.
616 The environment variable @code{GCC_EXEC_PREFIX}, if any.
619 The directories specified by the environment variable @code{COMPILER_PATH}.
622 The macro @code{STANDARD_EXEC_PREFIX}.
625 @file{/usr/lib/gcc/}.
628 The macro @code{MD_EXEC_PREFIX}, if any.
631 Here is the order of prefixes tried for startfiles:
635 Any prefixes specified by the user with @option{-B}.
638 The environment variable @code{GCC_EXEC_PREFIX}, if any.
641 The directories specified by the environment variable @code{LIBRARY_PATH}
642 (or port-specific name; native only, cross compilers do not use this).
645 The macro @code{STANDARD_EXEC_PREFIX}.
648 @file{/usr/lib/gcc/}.
651 The macro @code{MD_EXEC_PREFIX}, if any.
654 The macro @code{MD_STARTFILE_PREFIX}, if any.
657 The macro @code{STANDARD_STARTFILE_PREFIX}.
666 @node Run-time Target
667 @section Run-time Target Specification
668 @cindex run-time target specification
669 @cindex predefined macros
670 @cindex target specifications
672 @c prevent bad page break with this line
673 Here are run-time target specifications.
675 @defmac TARGET_CPU_CPP_BUILTINS ()
676 This function-like macro expands to a block of code that defines
677 built-in preprocessor macros and assertions for the target cpu, using
678 the functions @code{builtin_define}, @code{builtin_define_std} and
679 @code{builtin_assert}. When the front end
680 calls this macro it provides a trailing semicolon, and since it has
681 finished command line option processing your code can use those
684 @code{builtin_assert} takes a string in the form you pass to the
685 command-line option @option{-A}, such as @code{cpu=mips}, and creates
686 the assertion. @code{builtin_define} takes a string in the form
687 accepted by option @option{-D} and unconditionally defines the macro.
689 @code{builtin_define_std} takes a string representing the name of an
690 object-like macro. If it doesn't lie in the user's namespace,
691 @code{builtin_define_std} defines it unconditionally. Otherwise, it
692 defines a version with two leading underscores, and another version
693 with two leading and trailing underscores, and defines the original
694 only if an ISO standard was not requested on the command line. For
695 example, passing @code{unix} defines @code{__unix}, @code{__unix__}
696 and possibly @code{unix}; passing @code{_mips} defines @code{__mips},
697 @code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64}
698 defines only @code{_ABI64}.
700 You can also test for the C dialect being compiled. The variable
701 @code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus}
702 or @code{clk_objective_c}. Note that if we are preprocessing
703 assembler, this variable will be @code{clk_c} but the function-like
704 macro @code{preprocessing_asm_p()} will return true, so you might want
705 to check for that first. If you need to check for strict ANSI, the
706 variable @code{flag_iso} can be used. The function-like macro
707 @code{preprocessing_trad_p()} can be used to check for traditional
711 @defmac TARGET_OS_CPP_BUILTINS ()
712 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
713 and is used for the target operating system instead.
716 @defmac TARGET_OBJFMT_CPP_BUILTINS ()
717 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
718 and is used for the target object format. @file{elfos.h} uses this
719 macro to define @code{__ELF__}, so you probably do not need to define
723 @deftypevar {extern int} target_flags
724 This variable is declared in @file{options.h}, which is included before
725 any target-specific headers.
728 @deftypevar {Target Hook} int TARGET_DEFAULT_TARGET_FLAGS
729 This variable specifies the initial value of @code{target_flags}.
730 Its default setting is 0.
733 @cindex optional hardware or system features
734 @cindex features, optional, in system conventions
736 @deftypefn {Target Hook} bool TARGET_HANDLE_OPTION (size_t @var{code}, const char *@var{arg}, int @var{value})
737 This hook is called whenever the user specifies one of the
738 target-specific options described by the @file{.opt} definition files
739 (@pxref{Options}). It has the opportunity to do some option-specific
740 processing and should return true if the option is valid. The default
741 definition does nothing but return true.
743 @var{code} specifies the @code{OPT_@var{name}} enumeration value
744 associated with the selected option; @var{name} is just a rendering of
745 the option name in which non-alphanumeric characters are replaced by
746 underscores. @var{arg} specifies the string argument and is null if
747 no argument was given. If the option is flagged as a @code{UInteger}
748 (@pxref{Option properties}), @var{value} is the numeric value of the
749 argument. Otherwise @var{value} is 1 if the positive form of the
750 option was used and 0 if the ``no-'' form was.
753 @defmac TARGET_VERSION
754 This macro is a C statement to print on @code{stderr} a string
755 describing the particular machine description choice. Every machine
756 description should define @code{TARGET_VERSION}. For example:
760 #define TARGET_VERSION \
761 fprintf (stderr, " (68k, Motorola syntax)");
763 #define TARGET_VERSION \
764 fprintf (stderr, " (68k, MIT syntax)");
769 @defmac OVERRIDE_OPTIONS
770 Sometimes certain combinations of command options do not make sense on
771 a particular target machine. You can define a macro
772 @code{OVERRIDE_OPTIONS} to take account of this. This macro, if
773 defined, is executed once just after all the command options have been
776 Don't use this macro to turn on various extra optimizations for
777 @option{-O}. That is what @code{OPTIMIZATION_OPTIONS} is for.
780 @defmac OPTIMIZATION_OPTIONS (@var{level}, @var{size})
781 Some machines may desire to change what optimizations are performed for
782 various optimization levels. This macro, if defined, is executed once
783 just after the optimization level is determined and before the remainder
784 of the command options have been parsed. Values set in this macro are
785 used as the default values for the other command line options.
787 @var{level} is the optimization level specified; 2 if @option{-O2} is
788 specified, 1 if @option{-O} is specified, and 0 if neither is specified.
790 @var{size} is nonzero if @option{-Os} is specified and zero otherwise.
792 You should not use this macro to change options that are not
793 machine-specific. These should uniformly selected by the same
794 optimization level on all supported machines. Use this macro to enable
795 machine-specific optimizations.
797 @strong{Do not examine @code{write_symbols} in
798 this macro!} The debugging options are not supposed to alter the
802 @defmac CAN_DEBUG_WITHOUT_FP
803 Define this macro if debugging can be performed even without a frame
804 pointer. If this macro is defined, GCC will turn on the
805 @option{-fomit-frame-pointer} option whenever @option{-O} is specified.
808 @node Per-Function Data
809 @section Defining data structures for per-function information.
810 @cindex per-function data
811 @cindex data structures
813 If the target needs to store information on a per-function basis, GCC
814 provides a macro and a couple of variables to allow this. Note, just
815 using statics to store the information is a bad idea, since GCC supports
816 nested functions, so you can be halfway through encoding one function
817 when another one comes along.
819 GCC defines a data structure called @code{struct function} which
820 contains all of the data specific to an individual function. This
821 structure contains a field called @code{machine} whose type is
822 @code{struct machine_function *}, which can be used by targets to point
823 to their own specific data.
825 If a target needs per-function specific data it should define the type
826 @code{struct machine_function} and also the macro @code{INIT_EXPANDERS}.
827 This macro should be used to initialize the function pointer
828 @code{init_machine_status}. This pointer is explained below.
830 One typical use of per-function, target specific data is to create an
831 RTX to hold the register containing the function's return address. This
832 RTX can then be used to implement the @code{__builtin_return_address}
833 function, for level 0.
835 Note---earlier implementations of GCC used a single data area to hold
836 all of the per-function information. Thus when processing of a nested
837 function began the old per-function data had to be pushed onto a
838 stack, and when the processing was finished, it had to be popped off the
839 stack. GCC used to provide function pointers called
840 @code{save_machine_status} and @code{restore_machine_status} to handle
841 the saving and restoring of the target specific information. Since the
842 single data area approach is no longer used, these pointers are no
845 @defmac INIT_EXPANDERS
846 Macro called to initialize any target specific information. This macro
847 is called once per function, before generation of any RTL has begun.
848 The intention of this macro is to allow the initialization of the
849 function pointer @code{init_machine_status}.
852 @deftypevar {void (*)(struct function *)} init_machine_status
853 If this function pointer is non-@code{NULL} it will be called once per
854 function, before function compilation starts, in order to allow the
855 target to perform any target specific initialization of the
856 @code{struct function} structure. It is intended that this would be
857 used to initialize the @code{machine} of that structure.
859 @code{struct machine_function} structures are expected to be freed by GC@.
860 Generally, any memory that they reference must be allocated by using
861 @code{ggc_alloc}, including the structure itself.
865 @section Storage Layout
866 @cindex storage layout
868 Note that the definitions of the macros in this table which are sizes or
869 alignments measured in bits do not need to be constant. They can be C
870 expressions that refer to static variables, such as the @code{target_flags}.
871 @xref{Run-time Target}.
873 @defmac BITS_BIG_ENDIAN
874 Define this macro to have the value 1 if the most significant bit in a
875 byte has the lowest number; otherwise define it to have the value zero.
876 This means that bit-field instructions count from the most significant
877 bit. If the machine has no bit-field instructions, then this must still
878 be defined, but it doesn't matter which value it is defined to. This
879 macro need not be a constant.
881 This macro does not affect the way structure fields are packed into
882 bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
885 @defmac BYTES_BIG_ENDIAN
886 Define this macro to have the value 1 if the most significant byte in a
887 word has the lowest number. This macro need not be a constant.
890 @defmac WORDS_BIG_ENDIAN
891 Define this macro to have the value 1 if, in a multiword object, the
892 most significant word has the lowest number. This applies to both
893 memory locations and registers; GCC fundamentally assumes that the
894 order of words in memory is the same as the order in registers. This
895 macro need not be a constant.
898 @defmac LIBGCC2_WORDS_BIG_ENDIAN
899 Define this macro if @code{WORDS_BIG_ENDIAN} is not constant. This must be a
900 constant value with the same meaning as @code{WORDS_BIG_ENDIAN}, which will be
901 used only when compiling @file{libgcc2.c}. Typically the value will be set
902 based on preprocessor defines.
905 @defmac FLOAT_WORDS_BIG_ENDIAN
906 Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or
907 @code{TFmode} floating point numbers are stored in memory with the word
908 containing the sign bit at the lowest address; otherwise define it to
909 have the value 0. This macro need not be a constant.
911 You need not define this macro if the ordering is the same as for
915 @defmac BITS_PER_UNIT
916 Define this macro to be the number of bits in an addressable storage
917 unit (byte). If you do not define this macro the default is 8.
920 @defmac BITS_PER_WORD
921 Number of bits in a word. If you do not define this macro, the default
922 is @code{BITS_PER_UNIT * UNITS_PER_WORD}.
925 @defmac MAX_BITS_PER_WORD
926 Maximum number of bits in a word. If this is undefined, the default is
927 @code{BITS_PER_WORD}. Otherwise, it is the constant value that is the
928 largest value that @code{BITS_PER_WORD} can have at run-time.
931 @defmac UNITS_PER_WORD
932 Number of storage units in a word; normally the size of a general-purpose
933 register, a power of two from 1 or 8.
936 @defmac MIN_UNITS_PER_WORD
937 Minimum number of units in a word. If this is undefined, the default is
938 @code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the
939 smallest value that @code{UNITS_PER_WORD} can have at run-time.
942 @defmac UNITS_PER_SIMD_WORD
943 Number of units in the vectors that the vectorizer can produce.
944 The default is equal to @code{UNITS_PER_WORD}, because the vectorizer
945 can do some transformations even in absence of specialized @acronym{SIMD}
950 Width of a pointer, in bits. You must specify a value no wider than the
951 width of @code{Pmode}. If it is not equal to the width of @code{Pmode},
952 you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify
953 a value the default is @code{BITS_PER_WORD}.
956 @defmac POINTERS_EXTEND_UNSIGNED
957 A C expression whose value is greater than zero if pointers that need to be
958 extended from being @code{POINTER_SIZE} bits wide to @code{Pmode} are to
959 be zero-extended and zero if they are to be sign-extended. If the value
960 is less then zero then there must be an "ptr_extend" instruction that
961 extends a pointer from @code{POINTER_SIZE} to @code{Pmode}.
963 You need not define this macro if the @code{POINTER_SIZE} is equal
964 to the width of @code{Pmode}.
967 @defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type})
968 A macro to update @var{m} and @var{unsignedp} when an object whose type
969 is @var{type} and which has the specified mode and signedness is to be
970 stored in a register. This macro is only called when @var{type} is a
973 On most RISC machines, which only have operations that operate on a full
974 register, define this macro to set @var{m} to @code{word_mode} if
975 @var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most
976 cases, only integer modes should be widened because wider-precision
977 floating-point operations are usually more expensive than their narrower
980 For most machines, the macro definition does not change @var{unsignedp}.
981 However, some machines, have instructions that preferentially handle
982 either signed or unsigned quantities of certain modes. For example, on
983 the DEC Alpha, 32-bit loads from memory and 32-bit add instructions
984 sign-extend the result to 64 bits. On such machines, set
985 @var{unsignedp} according to which kind of extension is more efficient.
987 Do not define this macro if it would never modify @var{m}.
990 @defmac PROMOTE_FUNCTION_MODE
991 Like @code{PROMOTE_MODE}, but is applied to outgoing function arguments or
992 function return values, as specified by @code{TARGET_PROMOTE_FUNCTION_ARGS}
993 and @code{TARGET_PROMOTE_FUNCTION_RETURN}, respectively.
995 The default is @code{PROMOTE_MODE}.
998 @deftypefn {Target Hook} bool TARGET_PROMOTE_FUNCTION_ARGS (tree @var{fntype})
999 This target hook should return @code{true} if the promotion described by
1000 @code{PROMOTE_FUNCTION_MODE} should be done for outgoing function
1004 @deftypefn {Target Hook} bool TARGET_PROMOTE_FUNCTION_RETURN (tree @var{fntype})
1005 This target hook should return @code{true} if the promotion described by
1006 @code{PROMOTE_FUNCTION_MODE} should be done for the return value of
1009 If this target hook returns @code{true}, @code{FUNCTION_VALUE} must
1010 perform the same promotions done by @code{PROMOTE_FUNCTION_MODE}.
1013 @defmac PARM_BOUNDARY
1014 Normal alignment required for function parameters on the stack, in
1015 bits. All stack parameters receive at least this much alignment
1016 regardless of data type. On most machines, this is the same as the
1020 @defmac STACK_BOUNDARY
1021 Define this macro to the minimum alignment enforced by hardware for the
1022 stack pointer on this machine. The definition is a C expression for the
1023 desired alignment (measured in bits). This value is used as a default
1024 if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines,
1025 this should be the same as @code{PARM_BOUNDARY}.
1028 @defmac PREFERRED_STACK_BOUNDARY
1029 Define this macro if you wish to preserve a certain alignment for the
1030 stack pointer, greater than what the hardware enforces. The definition
1031 is a C expression for the desired alignment (measured in bits). This
1032 macro must evaluate to a value equal to or larger than
1033 @code{STACK_BOUNDARY}.
1036 @defmac FUNCTION_BOUNDARY
1037 Alignment required for a function entry point, in bits.
1040 @defmac BIGGEST_ALIGNMENT
1041 Biggest alignment that any data type can require on this machine, in bits.
1044 @defmac MINIMUM_ATOMIC_ALIGNMENT
1045 If defined, the smallest alignment, in bits, that can be given to an
1046 object that can be referenced in one operation, without disturbing any
1047 nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger
1048 on machines that don't have byte or half-word store operations.
1051 @defmac BIGGEST_FIELD_ALIGNMENT
1052 Biggest alignment that any structure or union field can require on this
1053 machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for
1054 structure and union fields only, unless the field alignment has been set
1055 by the @code{__attribute__ ((aligned (@var{n})))} construct.
1058 @defmac ADJUST_FIELD_ALIGN (@var{field}, @var{computed})
1059 An expression for the alignment of a structure field @var{field} if the
1060 alignment computed in the usual way (including applying of
1061 @code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the
1062 alignment) is @var{computed}. It overrides alignment only if the
1063 field alignment has not been set by the
1064 @code{__attribute__ ((aligned (@var{n})))} construct.
1067 @defmac MAX_OFILE_ALIGNMENT
1068 Biggest alignment supported by the object file format of this machine.
1069 Use this macro to limit the alignment which can be specified using the
1070 @code{__attribute__ ((aligned (@var{n})))} construct. If not defined,
1071 the default value is @code{BIGGEST_ALIGNMENT}.
1074 @defmac DATA_ALIGNMENT (@var{type}, @var{basic-align})
1075 If defined, a C expression to compute the alignment for a variable in
1076 the static store. @var{type} is the data type, and @var{basic-align} is
1077 the alignment that the object would ordinarily have. The value of this
1078 macro is used instead of that alignment to align the object.
1080 If this macro is not defined, then @var{basic-align} is used.
1083 One use of this macro is to increase alignment of medium-size data to
1084 make it all fit in fewer cache lines. Another is to cause character
1085 arrays to be word-aligned so that @code{strcpy} calls that copy
1086 constants to character arrays can be done inline.
1089 @defmac CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align})
1090 If defined, a C expression to compute the alignment given to a constant
1091 that is being placed in memory. @var{constant} is the constant and
1092 @var{basic-align} is the alignment that the object would ordinarily
1093 have. The value of this macro is used instead of that alignment to
1096 If this macro is not defined, then @var{basic-align} is used.
1098 The typical use of this macro is to increase alignment for string
1099 constants to be word aligned so that @code{strcpy} calls that copy
1100 constants can be done inline.
1103 @defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align})
1104 If defined, a C expression to compute the alignment for a variable in
1105 the local store. @var{type} is the data type, and @var{basic-align} is
1106 the alignment that the object would ordinarily have. The value of this
1107 macro is used instead of that alignment to align the object.
1109 If this macro is not defined, then @var{basic-align} is used.
1111 One use of this macro is to increase alignment of medium-size data to
1112 make it all fit in fewer cache lines.
1115 @defmac EMPTY_FIELD_BOUNDARY
1116 Alignment in bits to be given to a structure bit-field that follows an
1117 empty field such as @code{int : 0;}.
1119 If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro.
1122 @defmac STRUCTURE_SIZE_BOUNDARY
1123 Number of bits which any structure or union's size must be a multiple of.
1124 Each structure or union's size is rounded up to a multiple of this.
1126 If you do not define this macro, the default is the same as
1127 @code{BITS_PER_UNIT}.
1130 @defmac STRICT_ALIGNMENT
1131 Define this macro to be the value 1 if instructions will fail to work
1132 if given data not on the nominal alignment. If instructions will merely
1133 go slower in that case, define this macro as 0.
1136 @defmac PCC_BITFIELD_TYPE_MATTERS
1137 Define this if you wish to imitate the way many other C compilers handle
1138 alignment of bit-fields and the structures that contain them.
1140 The behavior is that the type written for a named bit-field (@code{int},
1141 @code{short}, or other integer type) imposes an alignment for the entire
1142 structure, as if the structure really did contain an ordinary field of
1143 that type. In addition, the bit-field is placed within the structure so
1144 that it would fit within such a field, not crossing a boundary for it.
1146 Thus, on most machines, a named bit-field whose type is written as
1147 @code{int} would not cross a four-byte boundary, and would force
1148 four-byte alignment for the whole structure. (The alignment used may
1149 not be four bytes; it is controlled by the other alignment parameters.)
1151 An unnamed bit-field will not affect the alignment of the containing
1154 If the macro is defined, its definition should be a C expression;
1155 a nonzero value for the expression enables this behavior.
1157 Note that if this macro is not defined, or its value is zero, some
1158 bit-fields may cross more than one alignment boundary. The compiler can
1159 support such references if there are @samp{insv}, @samp{extv}, and
1160 @samp{extzv} insns that can directly reference memory.
1162 The other known way of making bit-fields work is to define
1163 @code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}.
1164 Then every structure can be accessed with fullwords.
1166 Unless the machine has bit-field instructions or you define
1167 @code{STRUCTURE_SIZE_BOUNDARY} that way, you must define
1168 @code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value.
1170 If your aim is to make GCC use the same conventions for laying out
1171 bit-fields as are used by another compiler, here is how to investigate
1172 what the other compiler does. Compile and run this program:
1191 printf ("Size of foo1 is %d\n",
1192 sizeof (struct foo1));
1193 printf ("Size of foo2 is %d\n",
1194 sizeof (struct foo2));
1199 If this prints 2 and 5, then the compiler's behavior is what you would
1200 get from @code{PCC_BITFIELD_TYPE_MATTERS}.
1203 @defmac BITFIELD_NBYTES_LIMITED
1204 Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited
1205 to aligning a bit-field within the structure.
1208 @deftypefn {Target Hook} bool TARGET_ALIGN_ANON_BITFIELDS (void)
1209 When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine
1210 whether unnamed bitfields affect the alignment of the containing
1211 structure. The hook should return true if the structure should inherit
1212 the alignment requirements of an unnamed bitfield's type.
1215 @defmac MEMBER_TYPE_FORCES_BLK (@var{field}, @var{mode})
1216 Return 1 if a structure or array containing @var{field} should be accessed using
1219 If @var{field} is the only field in the structure, @var{mode} is its
1220 mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the
1221 case where structures of one field would require the structure's mode to
1222 retain the field's mode.
1224 Normally, this is not needed. See the file @file{c4x.h} for an example
1225 of how to use this macro to prevent a structure having a floating point
1226 field from being accessed in an integer mode.
1229 @defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified})
1230 Define this macro as an expression for the alignment of a type (given
1231 by @var{type} as a tree node) if the alignment computed in the usual
1232 way is @var{computed} and the alignment explicitly specified was
1235 The default is to use @var{specified} if it is larger; otherwise, use
1236 the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT}
1239 @defmac MAX_FIXED_MODE_SIZE
1240 An integer expression for the size in bits of the largest integer
1241 machine mode that should actually be used. All integer machine modes of
1242 this size or smaller can be used for structures and unions with the
1243 appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE
1244 (DImode)} is assumed.
1247 @defmac STACK_SAVEAREA_MODE (@var{save_level})
1248 If defined, an expression of type @code{enum machine_mode} that
1249 specifies the mode of the save area operand of a
1250 @code{save_stack_@var{level}} named pattern (@pxref{Standard Names}).
1251 @var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or
1252 @code{SAVE_NONLOCAL} and selects which of the three named patterns is
1253 having its mode specified.
1255 You need not define this macro if it always returns @code{Pmode}. You
1256 would most commonly define this macro if the
1257 @code{save_stack_@var{level}} patterns need to support both a 32- and a
1261 @defmac STACK_SIZE_MODE
1262 If defined, an expression of type @code{enum machine_mode} that
1263 specifies the mode of the size increment operand of an
1264 @code{allocate_stack} named pattern (@pxref{Standard Names}).
1266 You need not define this macro if it always returns @code{word_mode}.
1267 You would most commonly define this macro if the @code{allocate_stack}
1268 pattern needs to support both a 32- and a 64-bit mode.
1271 @defmac TARGET_FLOAT_FORMAT
1272 A code distinguishing the floating point format of the target machine.
1273 There are four defined values:
1276 @item IEEE_FLOAT_FORMAT
1277 This code indicates IEEE floating point. It is the default; there is no
1278 need to define @code{TARGET_FLOAT_FORMAT} when the format is IEEE@.
1280 @item VAX_FLOAT_FORMAT
1281 This code indicates the ``F float'' (for @code{float}) and ``D float''
1282 or ``G float'' formats (for @code{double}) used on the VAX and PDP-11@.
1284 @item IBM_FLOAT_FORMAT
1285 This code indicates the format used on the IBM System/370.
1287 @item C4X_FLOAT_FORMAT
1288 This code indicates the format used on the TMS320C3x/C4x.
1291 If your target uses a floating point format other than these, you must
1292 define a new @var{name}_FLOAT_FORMAT code for it, and add support for
1293 it to @file{real.c}.
1295 The ordering of the component words of floating point values stored in
1296 memory is controlled by @code{FLOAT_WORDS_BIG_ENDIAN}.
1299 @defmac MODE_HAS_NANS (@var{mode})
1300 When defined, this macro should be true if @var{mode} has a NaN
1301 representation. The compiler assumes that NaNs are not equal to
1302 anything (including themselves) and that addition, subtraction,
1303 multiplication and division all return NaNs when one operand is
1306 By default, this macro is true if @var{mode} is a floating-point
1307 mode and the target floating-point format is IEEE@.
1310 @defmac MODE_HAS_INFINITIES (@var{mode})
1311 This macro should be true if @var{mode} can represent infinity. At
1312 present, the compiler uses this macro to decide whether @samp{x - x}
1313 is always defined. By default, the macro is true when @var{mode}
1314 is a floating-point mode and the target format is IEEE@.
1317 @defmac MODE_HAS_SIGNED_ZEROS (@var{mode})
1318 True if @var{mode} distinguishes between positive and negative zero.
1319 The rules are expected to follow the IEEE standard:
1323 @samp{x + x} has the same sign as @samp{x}.
1326 If the sum of two values with opposite sign is zero, the result is
1327 positive for all rounding modes expect towards @minus{}infinity, for
1328 which it is negative.
1331 The sign of a product or quotient is negative when exactly one
1332 of the operands is negative.
1335 The default definition is true if @var{mode} is a floating-point
1336 mode and the target format is IEEE@.
1339 @defmac MODE_HAS_SIGN_DEPENDENT_ROUNDING (@var{mode})
1340 If defined, this macro should be true for @var{mode} if it has at
1341 least one rounding mode in which @samp{x} and @samp{-x} can be
1342 rounded to numbers of different magnitude. Two such modes are
1343 towards @minus{}infinity and towards +infinity.
1345 The default definition of this macro is true if @var{mode} is
1346 a floating-point mode and the target format is IEEE@.
1349 @defmac ROUND_TOWARDS_ZERO
1350 If defined, this macro should be true if the prevailing rounding
1351 mode is towards zero. A true value has the following effects:
1355 @code{MODE_HAS_SIGN_DEPENDENT_ROUNDING} will be false for all modes.
1358 @file{libgcc.a}'s floating-point emulator will round towards zero
1359 rather than towards nearest.
1362 The compiler's floating-point emulator will round towards zero after
1363 doing arithmetic, and when converting from the internal float format to
1367 The macro does not affect the parsing of string literals. When the
1368 primary rounding mode is towards zero, library functions like
1369 @code{strtod} might still round towards nearest, and the compiler's
1370 parser should behave like the target's @code{strtod} where possible.
1372 Not defining this macro is equivalent to returning zero.
1375 @defmac LARGEST_EXPONENT_IS_NORMAL (@var{size})
1376 This macro should return true if floats with @var{size}
1377 bits do not have a NaN or infinity representation, but use the largest
1378 exponent for normal numbers instead.
1380 Defining this macro to true for @var{size} causes @code{MODE_HAS_NANS}
1381 and @code{MODE_HAS_INFINITIES} to be false for @var{size}-bit modes.
1382 It also affects the way @file{libgcc.a} and @file{real.c} emulate
1383 floating-point arithmetic.
1385 The default definition of this macro returns false for all sizes.
1388 @deftypefn {Target Hook} bool TARGET_VECTOR_OPAQUE_P (tree @var{type})
1389 This target hook should return @code{true} a vector is opaque. That
1390 is, if no cast is needed when copying a vector value of type
1391 @var{type} into another vector lvalue of the same size. Vector opaque
1392 types cannot be initialized. The default is that there are no such
1396 @deftypefn {Target Hook} bool TARGET_MS_BITFIELD_LAYOUT_P (tree @var{record_type})
1397 This target hook returns @code{true} if bit-fields in the given
1398 @var{record_type} are to be laid out following the rules of Microsoft
1399 Visual C/C++, namely: (i) a bit-field won't share the same storage
1400 unit with the previous bit-field if their underlying types have
1401 different sizes, and the bit-field will be aligned to the highest
1402 alignment of the underlying types of itself and of the previous
1403 bit-field; (ii) a zero-sized bit-field will affect the alignment of
1404 the whole enclosing structure, even if it is unnamed; except that
1405 (iii) a zero-sized bit-field will be disregarded unless it follows
1406 another bit-field of nonzero size. If this hook returns @code{true},
1407 other macros that control bit-field layout are ignored.
1409 When a bit-field is inserted into a packed record, the whole size
1410 of the underlying type is used by one or more same-size adjacent
1411 bit-fields (that is, if its long:3, 32 bits is used in the record,
1412 and any additional adjacent long bit-fields are packed into the same
1413 chunk of 32 bits. However, if the size changes, a new field of that
1414 size is allocated). In an unpacked record, this is the same as using
1415 alignment, but not equivalent when packing.
1417 If both MS bit-fields and @samp{__attribute__((packed))} are used,
1418 the latter will take precedence. If @samp{__attribute__((packed))} is
1419 used on a single field when MS bit-fields are in use, it will take
1420 precedence for that field, but the alignment of the rest of the structure
1421 may affect its placement.
1424 @deftypefn {Target Hook} {const char *} TARGET_MANGLE_FUNDAMENTAL_TYPE (tree @var{type})
1425 If your target defines any fundamental types, define this hook to
1426 return the appropriate encoding for these types as part of a C++
1427 mangled name. The @var{type} argument is the tree structure
1428 representing the type to be mangled. The hook may be applied to trees
1429 which are not target-specific fundamental types; it should return
1430 @code{NULL} for all such types, as well as arguments it does not
1431 recognize. If the return value is not @code{NULL}, it must point to
1432 a statically-allocated string constant.
1434 Target-specific fundamental types might be new fundamental types or
1435 qualified versions of ordinary fundamental types. Encode new
1436 fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name}
1437 is the name used for the type in source code, and @var{n} is the
1438 length of @var{name} in decimal. Encode qualified versions of
1439 ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where
1440 @var{name} is the name used for the type qualifier in source code,
1441 @var{n} is the length of @var{name} as above, and @var{code} is the
1442 code used to represent the unqualified version of this type. (See
1443 @code{write_builtin_type} in @file{cp/mangle.c} for the list of
1444 codes.) In both cases the spaces are for clarity; do not include any
1445 spaces in your string.
1447 The default version of this hook always returns @code{NULL}, which is
1448 appropriate for a target that does not define any new fundamental
1453 @section Layout of Source Language Data Types
1455 These macros define the sizes and other characteristics of the standard
1456 basic data types used in programs being compiled. Unlike the macros in
1457 the previous section, these apply to specific features of C and related
1458 languages, rather than to fundamental aspects of storage layout.
1460 @defmac INT_TYPE_SIZE
1461 A C expression for the size in bits of the type @code{int} on the
1462 target machine. If you don't define this, the default is one word.
1465 @defmac SHORT_TYPE_SIZE
1466 A C expression for the size in bits of the type @code{short} on the
1467 target machine. If you don't define this, the default is half a word.
1468 (If this would be less than one storage unit, it is rounded up to one
1472 @defmac LONG_TYPE_SIZE
1473 A C expression for the size in bits of the type @code{long} on the
1474 target machine. If you don't define this, the default is one word.
1477 @defmac ADA_LONG_TYPE_SIZE
1478 On some machines, the size used for the Ada equivalent of the type
1479 @code{long} by a native Ada compiler differs from that used by C@. In
1480 that situation, define this macro to be a C expression to be used for
1481 the size of that type. If you don't define this, the default is the
1482 value of @code{LONG_TYPE_SIZE}.
1485 @defmac LONG_LONG_TYPE_SIZE
1486 A C expression for the size in bits of the type @code{long long} on the
1487 target machine. If you don't define this, the default is two
1488 words. If you want to support GNU Ada on your machine, the value of this
1489 macro must be at least 64.
1492 @defmac CHAR_TYPE_SIZE
1493 A C expression for the size in bits of the type @code{char} on the
1494 target machine. If you don't define this, the default is
1495 @code{BITS_PER_UNIT}.
1498 @defmac BOOL_TYPE_SIZE
1499 A C expression for the size in bits of the C++ type @code{bool} and
1500 C99 type @code{_Bool} on the target machine. If you don't define
1501 this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}.
1504 @defmac FLOAT_TYPE_SIZE
1505 A C expression for the size in bits of the type @code{float} on the
1506 target machine. If you don't define this, the default is one word.
1509 @defmac DOUBLE_TYPE_SIZE
1510 A C expression for the size in bits of the type @code{double} on the
1511 target machine. If you don't define this, the default is two
1515 @defmac LONG_DOUBLE_TYPE_SIZE
1516 A C expression for the size in bits of the type @code{long double} on
1517 the target machine. If you don't define this, the default is two
1521 @defmac LIBGCC2_LONG_DOUBLE_TYPE_SIZE
1522 Define this macro if @code{LONG_DOUBLE_TYPE_SIZE} is not constant or
1523 if you want routines in @file{libgcc2.a} for a size other than
1524 @code{LONG_DOUBLE_TYPE_SIZE}. If you don't define this, the
1525 default is @code{LONG_DOUBLE_TYPE_SIZE}.
1528 @defmac LIBGCC2_HAS_DF_MODE
1529 Define this macro if neither @code{LIBGCC2_DOUBLE_TYPE_SIZE} nor
1530 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is
1531 @code{DFmode} but you want @code{DFmode} routines in @file{libgcc2.a}
1532 anyway. If you don't define this and either @code{LIBGCC2_DOUBLE_TYPE_SIZE}
1533 or @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64 then the default is 1,
1537 @defmac LIBGCC2_HAS_XF_MODE
1538 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1539 @code{XFmode} but you want @code{XFmode} routines in @file{libgcc2.a}
1540 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1541 is 80 then the default is 1, otherwise it is 0.
1544 @defmac LIBGCC2_HAS_TF_MODE
1545 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1546 @code{TFmode} but you want @code{TFmode} routines in @file{libgcc2.a}
1547 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1548 is 128 then the default is 1, otherwise it is 0.
1551 @defmac TARGET_FLT_EVAL_METHOD
1552 A C expression for the value for @code{FLT_EVAL_METHOD} in @file{float.h},
1553 assuming, if applicable, that the floating-point control word is in its
1554 default state. If you do not define this macro the value of
1555 @code{FLT_EVAL_METHOD} will be zero.
1558 @defmac WIDEST_HARDWARE_FP_SIZE
1559 A C expression for the size in bits of the widest floating-point format
1560 supported by the hardware. If you define this macro, you must specify a
1561 value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}.
1562 If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE}
1566 @defmac DEFAULT_SIGNED_CHAR
1567 An expression whose value is 1 or 0, according to whether the type
1568 @code{char} should be signed or unsigned by default. The user can
1569 always override this default with the options @option{-fsigned-char}
1570 and @option{-funsigned-char}.
1573 @deftypefn {Target Hook} bool TARGET_DEFAULT_SHORT_ENUMS (void)
1574 This target hook should return true if the compiler should give an
1575 @code{enum} type only as many bytes as it takes to represent the range
1576 of possible values of that type. It should return false if all
1577 @code{enum} types should be allocated like @code{int}.
1579 The default is to return false.
1583 A C expression for a string describing the name of the data type to use
1584 for size values. The typedef name @code{size_t} is defined using the
1585 contents of the string.
1587 The string can contain more than one keyword. If so, separate them with
1588 spaces, and write first any length keyword, then @code{unsigned} if
1589 appropriate, and finally @code{int}. The string must exactly match one
1590 of the data type names defined in the function
1591 @code{init_decl_processing} in the file @file{c-decl.c}. You may not
1592 omit @code{int} or change the order---that would cause the compiler to
1595 If you don't define this macro, the default is @code{"long unsigned
1599 @defmac PTRDIFF_TYPE
1600 A C expression for a string describing the name of the data type to use
1601 for the result of subtracting two pointers. The typedef name
1602 @code{ptrdiff_t} is defined using the contents of the string. See
1603 @code{SIZE_TYPE} above for more information.
1605 If you don't define this macro, the default is @code{"long int"}.
1609 A C expression for a string describing the name of the data type to use
1610 for wide characters. The typedef name @code{wchar_t} is defined using
1611 the contents of the string. See @code{SIZE_TYPE} above for more
1614 If you don't define this macro, the default is @code{"int"}.
1617 @defmac WCHAR_TYPE_SIZE
1618 A C expression for the size in bits of the data type for wide
1619 characters. This is used in @code{cpp}, which cannot make use of
1624 A C expression for a string describing the name of the data type to
1625 use for wide characters passed to @code{printf} and returned from
1626 @code{getwc}. The typedef name @code{wint_t} is defined using the
1627 contents of the string. See @code{SIZE_TYPE} above for more
1630 If you don't define this macro, the default is @code{"unsigned int"}.
1634 A C expression for a string describing the name of the data type that
1635 can represent any value of any standard or extended signed integer type.
1636 The typedef name @code{intmax_t} is defined using the contents of the
1637 string. See @code{SIZE_TYPE} above for more information.
1639 If you don't define this macro, the default is the first of
1640 @code{"int"}, @code{"long int"}, or @code{"long long int"} that has as
1641 much precision as @code{long long int}.
1644 @defmac UINTMAX_TYPE
1645 A C expression for a string describing the name of the data type that
1646 can represent any value of any standard or extended unsigned integer
1647 type. The typedef name @code{uintmax_t} is defined using the contents
1648 of the string. See @code{SIZE_TYPE} above for more information.
1650 If you don't define this macro, the default is the first of
1651 @code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long
1652 unsigned int"} that has as much precision as @code{long long unsigned
1656 @defmac TARGET_PTRMEMFUNC_VBIT_LOCATION
1657 The C++ compiler represents a pointer-to-member-function with a struct
1664 ptrdiff_t vtable_index;
1671 The C++ compiler must use one bit to indicate whether the function that
1672 will be called through a pointer-to-member-function is virtual.
1673 Normally, we assume that the low-order bit of a function pointer must
1674 always be zero. Then, by ensuring that the vtable_index is odd, we can
1675 distinguish which variant of the union is in use. But, on some
1676 platforms function pointers can be odd, and so this doesn't work. In
1677 that case, we use the low-order bit of the @code{delta} field, and shift
1678 the remainder of the @code{delta} field to the left.
1680 GCC will automatically make the right selection about where to store
1681 this bit using the @code{FUNCTION_BOUNDARY} setting for your platform.
1682 However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY}
1683 set such that functions always start at even addresses, but the lowest
1684 bit of pointers to functions indicate whether the function at that
1685 address is in ARM or Thumb mode. If this is the case of your
1686 architecture, you should define this macro to
1687 @code{ptrmemfunc_vbit_in_delta}.
1689 In general, you should not have to define this macro. On architectures
1690 in which function addresses are always even, according to
1691 @code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to
1692 @code{ptrmemfunc_vbit_in_pfn}.
1695 @defmac TARGET_VTABLE_USES_DESCRIPTORS
1696 Normally, the C++ compiler uses function pointers in vtables. This
1697 macro allows the target to change to use ``function descriptors''
1698 instead. Function descriptors are found on targets for whom a
1699 function pointer is actually a small data structure. Normally the
1700 data structure consists of the actual code address plus a data
1701 pointer to which the function's data is relative.
1703 If vtables are used, the value of this macro should be the number
1704 of words that the function descriptor occupies.
1707 @defmac TARGET_VTABLE_ENTRY_ALIGN
1708 By default, the vtable entries are void pointers, the so the alignment
1709 is the same as pointer alignment. The value of this macro specifies
1710 the alignment of the vtable entry in bits. It should be defined only
1711 when special alignment is necessary. */
1714 @defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE
1715 There are a few non-descriptor entries in the vtable at offsets below
1716 zero. If these entries must be padded (say, to preserve the alignment
1717 specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number
1718 of words in each data entry.
1722 @section Register Usage
1723 @cindex register usage
1725 This section explains how to describe what registers the target machine
1726 has, and how (in general) they can be used.
1728 The description of which registers a specific instruction can use is
1729 done with register classes; see @ref{Register Classes}. For information
1730 on using registers to access a stack frame, see @ref{Frame Registers}.
1731 For passing values in registers, see @ref{Register Arguments}.
1732 For returning values in registers, see @ref{Scalar Return}.
1735 * Register Basics:: Number and kinds of registers.
1736 * Allocation Order:: Order in which registers are allocated.
1737 * Values in Registers:: What kinds of values each reg can hold.
1738 * Leaf Functions:: Renumbering registers for leaf functions.
1739 * Stack Registers:: Handling a register stack such as 80387.
1742 @node Register Basics
1743 @subsection Basic Characteristics of Registers
1745 @c prevent bad page break with this line
1746 Registers have various characteristics.
1748 @defmac FIRST_PSEUDO_REGISTER
1749 Number of hardware registers known to the compiler. They receive
1750 numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1751 pseudo register's number really is assigned the number
1752 @code{FIRST_PSEUDO_REGISTER}.
1755 @defmac FIXED_REGISTERS
1756 @cindex fixed register
1757 An initializer that says which registers are used for fixed purposes
1758 all throughout the compiled code and are therefore not available for
1759 general allocation. These would include the stack pointer, the frame
1760 pointer (except on machines where that can be used as a general
1761 register when no frame pointer is needed), the program counter on
1762 machines where that is considered one of the addressable registers,
1763 and any other numbered register with a standard use.
1765 This information is expressed as a sequence of numbers, separated by
1766 commas and surrounded by braces. The @var{n}th number is 1 if
1767 register @var{n} is fixed, 0 otherwise.
1769 The table initialized from this macro, and the table initialized by
1770 the following one, may be overridden at run time either automatically,
1771 by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
1772 the user with the command options @option{-ffixed-@var{reg}},
1773 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}.
1776 @defmac CALL_USED_REGISTERS
1777 @cindex call-used register
1778 @cindex call-clobbered register
1779 @cindex call-saved register
1780 Like @code{FIXED_REGISTERS} but has 1 for each register that is
1781 clobbered (in general) by function calls as well as for fixed
1782 registers. This macro therefore identifies the registers that are not
1783 available for general allocation of values that must live across
1786 If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1787 automatically saves it on function entry and restores it on function
1788 exit, if the register is used within the function.
1791 @defmac CALL_REALLY_USED_REGISTERS
1792 @cindex call-used register
1793 @cindex call-clobbered register
1794 @cindex call-saved register
1795 Like @code{CALL_USED_REGISTERS} except this macro doesn't require
1796 that the entire set of @code{FIXED_REGISTERS} be included.
1797 (@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}).
1798 This macro is optional. If not specified, it defaults to the value
1799 of @code{CALL_USED_REGISTERS}.
1802 @defmac HARD_REGNO_CALL_PART_CLOBBERED (@var{regno}, @var{mode})
1803 @cindex call-used register
1804 @cindex call-clobbered register
1805 @cindex call-saved register
1806 A C expression that is nonzero if it is not permissible to store a
1807 value of mode @var{mode} in hard register number @var{regno} across a
1808 call without some part of it being clobbered. For most machines this
1809 macro need not be defined. It is only required for machines that do not
1810 preserve the entire contents of a register across a call.
1814 @findex call_used_regs
1817 @findex reg_class_contents
1818 @defmac CONDITIONAL_REGISTER_USAGE
1819 Zero or more C statements that may conditionally modify five variables
1820 @code{fixed_regs}, @code{call_used_regs}, @code{global_regs},
1821 @code{reg_names}, and @code{reg_class_contents}, to take into account
1822 any dependence of these register sets on target flags. The first three
1823 of these are of type @code{char []} (interpreted as Boolean vectors).
1824 @code{global_regs} is a @code{const char *[]}, and
1825 @code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is
1826 called, @code{fixed_regs}, @code{call_used_regs},
1827 @code{reg_class_contents}, and @code{reg_names} have been initialized
1828 from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS},
1829 @code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively.
1830 @code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}},
1831 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}
1832 command options have been applied.
1834 You need not define this macro if it has no work to do.
1836 @cindex disabling certain registers
1837 @cindex controlling register usage
1838 If the usage of an entire class of registers depends on the target
1839 flags, you may indicate this to GCC by using this macro to modify
1840 @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
1841 registers in the classes which should not be used by GCC@. Also define
1842 the macro @code{REG_CLASS_FROM_LETTER} / @code{REG_CLASS_FROM_CONSTRAINT}
1843 to return @code{NO_REGS} if it
1844 is called with a letter for a class that shouldn't be used.
1846 (However, if this class is not included in @code{GENERAL_REGS} and all
1847 of the insn patterns whose constraints permit this class are
1848 controlled by target switches, then GCC will automatically avoid using
1849 these registers when the target switches are opposed to them.)
1852 @defmac INCOMING_REGNO (@var{out})
1853 Define this macro if the target machine has register windows. This C
1854 expression returns the register number as seen by the called function
1855 corresponding to the register number @var{out} as seen by the calling
1856 function. Return @var{out} if register number @var{out} is not an
1860 @defmac OUTGOING_REGNO (@var{in})
1861 Define this macro if the target machine has register windows. This C
1862 expression returns the register number as seen by the calling function
1863 corresponding to the register number @var{in} as seen by the called
1864 function. Return @var{in} if register number @var{in} is not an inbound
1868 @defmac LOCAL_REGNO (@var{regno})
1869 Define this macro if the target machine has register windows. This C
1870 expression returns true if the register is call-saved but is in the
1871 register window. Unlike most call-saved registers, such registers
1872 need not be explicitly restored on function exit or during non-local
1877 If the program counter has a register number, define this as that
1878 register number. Otherwise, do not define it.
1881 @node Allocation Order
1882 @subsection Order of Allocation of Registers
1883 @cindex order of register allocation
1884 @cindex register allocation order
1886 @c prevent bad page break with this line
1887 Registers are allocated in order.
1889 @defmac REG_ALLOC_ORDER
1890 If defined, an initializer for a vector of integers, containing the
1891 numbers of hard registers in the order in which GCC should prefer
1892 to use them (from most preferred to least).
1894 If this macro is not defined, registers are used lowest numbered first
1895 (all else being equal).
1897 One use of this macro is on machines where the highest numbered
1898 registers must always be saved and the save-multiple-registers
1899 instruction supports only sequences of consecutive registers. On such
1900 machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists
1901 the highest numbered allocable register first.
1904 @defmac ORDER_REGS_FOR_LOCAL_ALLOC
1905 A C statement (sans semicolon) to choose the order in which to allocate
1906 hard registers for pseudo-registers local to a basic block.
1908 Store the desired register order in the array @code{reg_alloc_order}.
1909 Element 0 should be the register to allocate first; element 1, the next
1910 register; and so on.
1912 The macro body should not assume anything about the contents of
1913 @code{reg_alloc_order} before execution of the macro.
1915 On most machines, it is not necessary to define this macro.
1918 @node Values in Registers
1919 @subsection How Values Fit in Registers
1921 This section discusses the macros that describe which kinds of values
1922 (specifically, which machine modes) each register can hold, and how many
1923 consecutive registers are needed for a given mode.
1925 @defmac HARD_REGNO_NREGS (@var{regno}, @var{mode})
1926 A C expression for the number of consecutive hard registers, starting
1927 at register number @var{regno}, required to hold a value of mode
1930 On a machine where all registers are exactly one word, a suitable
1931 definition of this macro is
1934 #define HARD_REGNO_NREGS(REGNO, MODE) \
1935 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
1940 @defmac REGMODE_NATURAL_SIZE (@var{mode})
1941 Define this macro if the natural size of registers that hold values
1942 of mode @var{mode} is not the word size. It is a C expression that
1943 should give the natural size in bytes for the specified mode. It is
1944 used by the register allocator to try to optimize its results. This
1945 happens for example on SPARC 64-bit where the natural size of
1946 floating-point registers is still 32-bit.
1949 @defmac HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
1950 A C expression that is nonzero if it is permissible to store a value
1951 of mode @var{mode} in hard register number @var{regno} (or in several
1952 registers starting with that one). For a machine where all registers
1953 are equivalent, a suitable definition is
1956 #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
1959 You need not include code to check for the numbers of fixed registers,
1960 because the allocation mechanism considers them to be always occupied.
1962 @cindex register pairs
1963 On some machines, double-precision values must be kept in even/odd
1964 register pairs. You can implement that by defining this macro to reject
1965 odd register numbers for such modes.
1967 The minimum requirement for a mode to be OK in a register is that the
1968 @samp{mov@var{mode}} instruction pattern support moves between the
1969 register and other hard register in the same class and that moving a
1970 value into the register and back out not alter it.
1972 Since the same instruction used to move @code{word_mode} will work for
1973 all narrower integer modes, it is not necessary on any machine for
1974 @code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided
1975 you define patterns @samp{movhi}, etc., to take advantage of this. This
1976 is useful because of the interaction between @code{HARD_REGNO_MODE_OK}
1977 and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes
1980 Many machines have special registers for floating point arithmetic.
1981 Often people assume that floating point machine modes are allowed only
1982 in floating point registers. This is not true. Any registers that
1983 can hold integers can safely @emph{hold} a floating point machine
1984 mode, whether or not floating arithmetic can be done on it in those
1985 registers. Integer move instructions can be used to move the values.
1987 On some machines, though, the converse is true: fixed-point machine
1988 modes may not go in floating registers. This is true if the floating
1989 registers normalize any value stored in them, because storing a
1990 non-floating value there would garble it. In this case,
1991 @code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
1992 floating registers. But if the floating registers do not automatically
1993 normalize, if you can store any bit pattern in one and retrieve it
1994 unchanged without a trap, then any machine mode may go in a floating
1995 register, so you can define this macro to say so.
1997 The primary significance of special floating registers is rather that
1998 they are the registers acceptable in floating point arithmetic
1999 instructions. However, this is of no concern to
2000 @code{HARD_REGNO_MODE_OK}. You handle it by writing the proper
2001 constraints for those instructions.
2003 On some machines, the floating registers are especially slow to access,
2004 so that it is better to store a value in a stack frame than in such a
2005 register if floating point arithmetic is not being done. As long as the
2006 floating registers are not in class @code{GENERAL_REGS}, they will not
2007 be used unless some pattern's constraint asks for one.
2010 @defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to})
2011 A C expression that is nonzero if it is OK to rename a hard register
2012 @var{from} to another hard register @var{to}.
2014 One common use of this macro is to prevent renaming of a register to
2015 another register that is not saved by a prologue in an interrupt
2018 The default is always nonzero.
2021 @defmac MODES_TIEABLE_P (@var{mode1}, @var{mode2})
2022 A C expression that is nonzero if a value of mode
2023 @var{mode1} is accessible in mode @var{mode2} without copying.
2025 If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
2026 @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for
2027 any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})}
2028 should be nonzero. If they differ for any @var{r}, you should define
2029 this macro to return zero unless some other mechanism ensures the
2030 accessibility of the value in a narrower mode.
2032 You should define this macro to return nonzero in as many cases as
2033 possible since doing so will allow GCC to perform better register
2037 @defmac AVOID_CCMODE_COPIES
2038 Define this macro if the compiler should avoid copies to/from @code{CCmode}
2039 registers. You should only define this macro if support for copying to/from
2040 @code{CCmode} is incomplete.
2043 @node Leaf Functions
2044 @subsection Handling Leaf Functions
2046 @cindex leaf functions
2047 @cindex functions, leaf
2048 On some machines, a leaf function (i.e., one which makes no calls) can run
2049 more efficiently if it does not make its own register window. Often this
2050 means it is required to receive its arguments in the registers where they
2051 are passed by the caller, instead of the registers where they would
2054 The special treatment for leaf functions generally applies only when
2055 other conditions are met; for example, often they may use only those
2056 registers for its own variables and temporaries. We use the term ``leaf
2057 function'' to mean a function that is suitable for this special
2058 handling, so that functions with no calls are not necessarily ``leaf
2061 GCC assigns register numbers before it knows whether the function is
2062 suitable for leaf function treatment. So it needs to renumber the
2063 registers in order to output a leaf function. The following macros
2066 @defmac LEAF_REGISTERS
2067 Name of a char vector, indexed by hard register number, which
2068 contains 1 for a register that is allowable in a candidate for leaf
2071 If leaf function treatment involves renumbering the registers, then the
2072 registers marked here should be the ones before renumbering---those that
2073 GCC would ordinarily allocate. The registers which will actually be
2074 used in the assembler code, after renumbering, should not be marked with 1
2077 Define this macro only if the target machine offers a way to optimize
2078 the treatment of leaf functions.
2081 @defmac LEAF_REG_REMAP (@var{regno})
2082 A C expression whose value is the register number to which @var{regno}
2083 should be renumbered, when a function is treated as a leaf function.
2085 If @var{regno} is a register number which should not appear in a leaf
2086 function before renumbering, then the expression should yield @minus{}1, which
2087 will cause the compiler to abort.
2089 Define this macro only if the target machine offers a way to optimize the
2090 treatment of leaf functions, and registers need to be renumbered to do
2094 @findex current_function_is_leaf
2095 @findex current_function_uses_only_leaf_regs
2096 @code{TARGET_ASM_FUNCTION_PROLOGUE} and
2097 @code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions
2098 specially. They can test the C variable @code{current_function_is_leaf}
2099 which is nonzero for leaf functions. @code{current_function_is_leaf} is
2100 set prior to local register allocation and is valid for the remaining
2101 compiler passes. They can also test the C variable
2102 @code{current_function_uses_only_leaf_regs} which is nonzero for leaf
2103 functions which only use leaf registers.
2104 @code{current_function_uses_only_leaf_regs} is valid after all passes
2105 that modify the instructions have been run and is only useful if
2106 @code{LEAF_REGISTERS} is defined.
2107 @c changed this to fix overfull. ALSO: why the "it" at the beginning
2108 @c of the next paragraph?! --mew 2feb93
2110 @node Stack Registers
2111 @subsection Registers That Form a Stack
2113 There are special features to handle computers where some of the
2114 ``registers'' form a stack. Stack registers are normally written by
2115 pushing onto the stack, and are numbered relative to the top of the
2118 Currently, GCC can only handle one group of stack-like registers, and
2119 they must be consecutively numbered. Furthermore, the existing
2120 support for stack-like registers is specific to the 80387 floating
2121 point coprocessor. If you have a new architecture that uses
2122 stack-like registers, you will need to do substantial work on
2123 @file{reg-stack.c} and write your machine description to cooperate
2124 with it, as well as defining these macros.
2127 Define this if the machine has any stack-like registers.
2130 @defmac FIRST_STACK_REG
2131 The number of the first stack-like register. This one is the top
2135 @defmac LAST_STACK_REG
2136 The number of the last stack-like register. This one is the bottom of
2140 @node Register Classes
2141 @section Register Classes
2142 @cindex register class definitions
2143 @cindex class definitions, register
2145 On many machines, the numbered registers are not all equivalent.
2146 For example, certain registers may not be allowed for indexed addressing;
2147 certain registers may not be allowed in some instructions. These machine
2148 restrictions are described to the compiler using @dfn{register classes}.
2150 You define a number of register classes, giving each one a name and saying
2151 which of the registers belong to it. Then you can specify register classes
2152 that are allowed as operands to particular instruction patterns.
2156 In general, each register will belong to several classes. In fact, one
2157 class must be named @code{ALL_REGS} and contain all the registers. Another
2158 class must be named @code{NO_REGS} and contain no registers. Often the
2159 union of two classes will be another class; however, this is not required.
2161 @findex GENERAL_REGS
2162 One of the classes must be named @code{GENERAL_REGS}. There is nothing
2163 terribly special about the name, but the operand constraint letters
2164 @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
2165 the same as @code{ALL_REGS}, just define it as a macro which expands
2168 Order the classes so that if class @var{x} is contained in class @var{y}
2169 then @var{x} has a lower class number than @var{y}.
2171 The way classes other than @code{GENERAL_REGS} are specified in operand
2172 constraints is through machine-dependent operand constraint letters.
2173 You can define such letters to correspond to various classes, then use
2174 them in operand constraints.
2176 You should define a class for the union of two classes whenever some
2177 instruction allows both classes. For example, if an instruction allows
2178 either a floating point (coprocessor) register or a general register for a
2179 certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
2180 which includes both of them. Otherwise you will get suboptimal code.
2182 You must also specify certain redundant information about the register
2183 classes: for each class, which classes contain it and which ones are
2184 contained in it; for each pair of classes, the largest class contained
2187 When a value occupying several consecutive registers is expected in a
2188 certain class, all the registers used must belong to that class.
2189 Therefore, register classes cannot be used to enforce a requirement for
2190 a register pair to start with an even-numbered register. The way to
2191 specify this requirement is with @code{HARD_REGNO_MODE_OK}.
2193 Register classes used for input-operands of bitwise-and or shift
2194 instructions have a special requirement: each such class must have, for
2195 each fixed-point machine mode, a subclass whose registers can transfer that
2196 mode to or from memory. For example, on some machines, the operations for
2197 single-byte values (@code{QImode}) are limited to certain registers. When
2198 this is so, each register class that is used in a bitwise-and or shift
2199 instruction must have a subclass consisting of registers from which
2200 single-byte values can be loaded or stored. This is so that
2201 @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
2203 @deftp {Data type} {enum reg_class}
2204 An enumerated type that must be defined with all the register class names
2205 as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS}
2206 must be the last register class, followed by one more enumerated value,
2207 @code{LIM_REG_CLASSES}, which is not a register class but rather
2208 tells how many classes there are.
2210 Each register class has a number, which is the value of casting
2211 the class name to type @code{int}. The number serves as an index
2212 in many of the tables described below.
2215 @defmac N_REG_CLASSES
2216 The number of distinct register classes, defined as follows:
2219 #define N_REG_CLASSES (int) LIM_REG_CLASSES
2223 @defmac REG_CLASS_NAMES
2224 An initializer containing the names of the register classes as C string
2225 constants. These names are used in writing some of the debugging dumps.
2228 @defmac REG_CLASS_CONTENTS
2229 An initializer containing the contents of the register classes, as integers
2230 which are bit masks. The @var{n}th integer specifies the contents of class
2231 @var{n}. The way the integer @var{mask} is interpreted is that
2232 register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
2234 When the machine has more than 32 registers, an integer does not suffice.
2235 Then the integers are replaced by sub-initializers, braced groupings containing
2236 several integers. Each sub-initializer must be suitable as an initializer
2237 for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
2238 In this situation, the first integer in each sub-initializer corresponds to
2239 registers 0 through 31, the second integer to registers 32 through 63, and
2243 @defmac REGNO_REG_CLASS (@var{regno})
2244 A C expression whose value is a register class containing hard register
2245 @var{regno}. In general there is more than one such class; choose a class
2246 which is @dfn{minimal}, meaning that no smaller class also contains the
2250 @defmac BASE_REG_CLASS
2251 A macro whose definition is the name of the class to which a valid
2252 base register must belong. A base register is one used in an address
2253 which is the register value plus a displacement.
2256 @defmac MODE_BASE_REG_CLASS (@var{mode})
2257 This is a variation of the @code{BASE_REG_CLASS} macro which allows
2258 the selection of a base register in a mode dependent manner. If
2259 @var{mode} is VOIDmode then it should return the same value as
2260 @code{BASE_REG_CLASS}.
2263 @defmac MODE_BASE_REG_REG_CLASS (@var{mode})
2264 A C expression whose value is the register class to which a valid
2265 base register must belong in order to be used in a base plus index
2266 register address. You should define this macro if base plus index
2267 addresses have different requirements than other base register uses.
2270 @defmac INDEX_REG_CLASS
2271 A macro whose definition is the name of the class to which a valid
2272 index register must belong. An index register is one used in an
2273 address where its value is either multiplied by a scale factor or
2274 added to another register (as well as added to a displacement).
2277 @defmac CONSTRAINT_LEN (@var{char}, @var{str})
2278 For the constraint at the start of @var{str}, which starts with the letter
2279 @var{c}, return the length. This allows you to have register class /
2280 constant / extra constraints that are longer than a single letter;
2281 you don't need to define this macro if you can do with single-letter
2282 constraints only. The definition of this macro should use
2283 DEFAULT_CONSTRAINT_LEN for all the characters that you don't want
2284 to handle specially.
2285 There are some sanity checks in genoutput.c that check the constraint lengths
2286 for the md file, so you can also use this macro to help you while you are
2287 transitioning from a byzantine single-letter-constraint scheme: when you
2288 return a negative length for a constraint you want to re-use, genoutput
2289 will complain about every instance where it is used in the md file.
2292 @defmac REG_CLASS_FROM_LETTER (@var{char})
2293 A C expression which defines the machine-dependent operand constraint
2294 letters for register classes. If @var{char} is such a letter, the
2295 value should be the register class corresponding to it. Otherwise,
2296 the value should be @code{NO_REGS}. The register letter @samp{r},
2297 corresponding to class @code{GENERAL_REGS}, will not be passed
2298 to this macro; you do not need to handle it.
2301 @defmac REG_CLASS_FROM_CONSTRAINT (@var{char}, @var{str})
2302 Like @code{REG_CLASS_FROM_LETTER}, but you also get the constraint string
2303 passed in @var{str}, so that you can use suffixes to distinguish between
2307 @defmac REGNO_OK_FOR_BASE_P (@var{num})
2308 A C expression which is nonzero if register number @var{num} is
2309 suitable for use as a base register in operand addresses. It may be
2310 either a suitable hard register or a pseudo register that has been
2311 allocated such a hard register.
2314 @defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode})
2315 A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that
2316 that expression may examine the mode of the memory reference in
2317 @var{mode}. You should define this macro if the mode of the memory
2318 reference affects whether a register may be used as a base register. If
2319 you define this macro, the compiler will use it instead of
2320 @code{REGNO_OK_FOR_BASE_P}.
2323 @defmac REGNO_MODE_OK_FOR_REG_BASE_P (@var{num}, @var{mode})
2324 A C expression which is nonzero if register number @var{num} is suitable for
2325 use as a base register in base plus index operand addresses, accessing
2326 memory in mode @var{mode}. It may be either a suitable hard register or a
2327 pseudo register that has been allocated such a hard register. You should
2328 define this macro if base plus index addresses have different requirements
2329 than other base register uses.
2332 @defmac REGNO_OK_FOR_INDEX_P (@var{num})
2333 A C expression which is nonzero if register number @var{num} is
2334 suitable for use as an index register in operand addresses. It may be
2335 either a suitable hard register or a pseudo register that has been
2336 allocated such a hard register.
2338 The difference between an index register and a base register is that
2339 the index register may be scaled. If an address involves the sum of
2340 two registers, neither one of them scaled, then either one may be
2341 labeled the ``base'' and the other the ``index''; but whichever
2342 labeling is used must fit the machine's constraints of which registers
2343 may serve in each capacity. The compiler will try both labelings,
2344 looking for one that is valid, and will reload one or both registers
2345 only if neither labeling works.
2348 @defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
2349 A C expression that places additional restrictions on the register class
2350 to use when it is necessary to copy value @var{x} into a register in class
2351 @var{class}. The value is a register class; perhaps @var{class}, or perhaps
2352 another, smaller class. On many machines, the following definition is
2356 #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
2359 Sometimes returning a more restrictive class makes better code. For
2360 example, on the 68000, when @var{x} is an integer constant that is in range
2361 for a @samp{moveq} instruction, the value of this macro is always
2362 @code{DATA_REGS} as long as @var{class} includes the data registers.
2363 Requiring a data register guarantees that a @samp{moveq} will be used.
2365 One case where @code{PREFERRED_RELOAD_CLASS} must not return
2366 @var{class} is if @var{x} is a legitimate constant which cannot be
2367 loaded into some register class. By returning @code{NO_REGS} you can
2368 force @var{x} into a memory location. For example, rs6000 can load
2369 immediate values into general-purpose registers, but does not have an
2370 instruction for loading an immediate value into a floating-point
2371 register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2372 @var{x} is a floating-point constant. If the constant can't be loaded
2373 into any kind of register, code generation will be better if
2374 @code{LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2375 of using @code{PREFERRED_RELOAD_CLASS}.
2378 @defmac PREFERRED_OUTPUT_RELOAD_CLASS (@var{x}, @var{class})
2379 Like @code{PREFERRED_RELOAD_CLASS}, but for output reloads instead of
2380 input reloads. If you don't define this macro, the default is to use
2381 @var{class}, unchanged.
2384 @defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class})
2385 A C expression that places additional restrictions on the register class
2386 to use when it is necessary to be able to hold a value of mode
2387 @var{mode} in a reload register for which class @var{class} would
2390 Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when
2391 there are certain modes that simply can't go in certain reload classes.
2393 The value is a register class; perhaps @var{class}, or perhaps another,
2396 Don't define this macro unless the target machine has limitations which
2397 require the macro to do something nontrivial.
2400 @defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2401 @defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2402 @defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2403 Many machines have some registers that cannot be copied directly to or
2404 from memory or even from other types of registers. An example is the
2405 @samp{MQ} register, which on most machines, can only be copied to or
2406 from general registers, but not memory. Some machines allow copying all
2407 registers to and from memory, but require a scratch register for stores
2408 to some memory locations (e.g., those with symbolic address on the RT,
2409 and those with certain symbolic address on the SPARC when compiling
2410 PIC)@. In some cases, both an intermediate and a scratch register are
2413 You should define these macros to indicate to the reload phase that it may
2414 need to allocate at least one register for a reload in addition to the
2415 register to contain the data. Specifically, if copying @var{x} to a
2416 register @var{class} in @var{mode} requires an intermediate register,
2417 you should define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the
2418 largest register class all of whose registers can be used as
2419 intermediate registers or scratch registers.
2421 If copying a register @var{class} in @var{mode} to @var{x} requires an
2422 intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS}
2423 should be defined to return the largest register class required. If the
2424 requirements for input and output reloads are the same, the macro
2425 @code{SECONDARY_RELOAD_CLASS} should be used instead of defining both
2428 The values returned by these macros are often @code{GENERAL_REGS}.
2429 Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x}
2430 can be directly copied to or from a register of @var{class} in
2431 @var{mode} without requiring a scratch register. Do not define this
2432 macro if it would always return @code{NO_REGS}.
2434 If a scratch register is required (either with or without an
2435 intermediate register), you should define patterns for
2436 @samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required
2437 (@pxref{Standard Names}. These patterns, which will normally be
2438 implemented with a @code{define_expand}, should be similar to the
2439 @samp{mov@var{m}} patterns, except that operand 2 is the scratch
2442 Define constraints for the reload register and scratch register that
2443 contain a single register class. If the original reload register (whose
2444 class is @var{class}) can meet the constraint given in the pattern, the
2445 value returned by these macros is used for the class of the scratch
2446 register. Otherwise, two additional reload registers are required.
2447 Their classes are obtained from the constraints in the insn pattern.
2449 @var{x} might be a pseudo-register or a @code{subreg} of a
2450 pseudo-register, which could either be in a hard register or in memory.
2451 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2452 in memory and the hard register number if it is in a register.
2454 These macros should not be used in the case where a particular class of
2455 registers can only be copied to memory and not to another class of
2456 registers. In that case, secondary reload registers are not needed and
2457 would not be helpful. Instead, a stack location must be used to perform
2458 the copy and the @code{mov@var{m}} pattern should use memory as an
2459 intermediate storage. This case often occurs between floating-point and
2463 @defmac SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m})
2464 Certain machines have the property that some registers cannot be copied
2465 to some other registers without using memory. Define this macro on
2466 those machines to be a C expression that is nonzero if objects of mode
2467 @var{m} in registers of @var{class1} can only be copied to registers of
2468 class @var{class2} by storing a register of @var{class1} into memory
2469 and loading that memory location into a register of @var{class2}.
2471 Do not define this macro if its value would always be zero.
2474 @defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode})
2475 Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler
2476 allocates a stack slot for a memory location needed for register copies.
2477 If this macro is defined, the compiler instead uses the memory location
2478 defined by this macro.
2480 Do not define this macro if you do not define
2481 @code{SECONDARY_MEMORY_NEEDED}.
2484 @defmac SECONDARY_MEMORY_NEEDED_MODE (@var{mode})
2485 When the compiler needs a secondary memory location to copy between two
2486 registers of mode @var{mode}, it normally allocates sufficient memory to
2487 hold a quantity of @code{BITS_PER_WORD} bits and performs the store and
2488 load operations in a mode that many bits wide and whose class is the
2489 same as that of @var{mode}.
2491 This is right thing to do on most machines because it ensures that all
2492 bits of the register are copied and prevents accesses to the registers
2493 in a narrower mode, which some machines prohibit for floating-point
2496 However, this default behavior is not correct on some machines, such as
2497 the DEC Alpha, that store short integers in floating-point registers
2498 differently than in integer registers. On those machines, the default
2499 widening will not work correctly and you must define this macro to
2500 suppress that widening in some cases. See the file @file{alpha.h} for
2503 Do not define this macro if you do not define
2504 @code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that
2505 is @code{BITS_PER_WORD} bits wide is correct for your machine.
2508 @defmac SMALL_REGISTER_CLASSES
2509 On some machines, it is risky to let hard registers live across arbitrary
2510 insns. Typically, these machines have instructions that require values
2511 to be in specific registers (like an accumulator), and reload will fail
2512 if the required hard register is used for another purpose across such an
2515 Define @code{SMALL_REGISTER_CLASSES} to be an expression with a nonzero
2516 value on these machines. When this macro has a nonzero value, the
2517 compiler will try to minimize the lifetime of hard registers.
2519 It is always safe to define this macro with a nonzero value, but if you
2520 unnecessarily define it, you will reduce the amount of optimizations
2521 that can be performed in some cases. If you do not define this macro
2522 with a nonzero value when it is required, the compiler will run out of
2523 spill registers and print a fatal error message. For most machines, you
2524 should not define this macro at all.
2527 @defmac CLASS_LIKELY_SPILLED_P (@var{class})
2528 A C expression whose value is nonzero if pseudos that have been assigned
2529 to registers of class @var{class} would likely be spilled because
2530 registers of @var{class} are needed for spill registers.
2532 The default value of this macro returns 1 if @var{class} has exactly one
2533 register and zero otherwise. On most machines, this default should be
2534 used. Only define this macro to some other expression if pseudos
2535 allocated by @file{local-alloc.c} end up in memory because their hard
2536 registers were needed for spill registers. If this macro returns nonzero
2537 for those classes, those pseudos will only be allocated by
2538 @file{global.c}, which knows how to reallocate the pseudo to another
2539 register. If there would not be another register available for
2540 reallocation, you should not change the definition of this macro since
2541 the only effect of such a definition would be to slow down register
2545 @defmac CLASS_MAX_NREGS (@var{class}, @var{mode})
2546 A C expression for the maximum number of consecutive registers
2547 of class @var{class} needed to hold a value of mode @var{mode}.
2549 This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact,
2550 the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
2551 should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno},
2552 @var{mode})} for all @var{regno} values in the class @var{class}.
2554 This macro helps control the handling of multiple-word values
2558 @defmac CANNOT_CHANGE_MODE_CLASS (@var{from}, @var{to}, @var{class})
2559 If defined, a C expression that returns nonzero for a @var{class} for which
2560 a change from mode @var{from} to mode @var{to} is invalid.
2562 For the example, loading 32-bit integer or floating-point objects into
2563 floating-point registers on the Alpha extends them to 64 bits.
2564 Therefore loading a 64-bit object and then storing it as a 32-bit object
2565 does not store the low-order 32 bits, as would be the case for a normal
2566 register. Therefore, @file{alpha.h} defines @code{CANNOT_CHANGE_MODE_CLASS}
2570 #define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \
2571 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \
2572 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0)
2576 Three other special macros describe which operands fit which constraint
2579 @defmac CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
2580 A C expression that defines the machine-dependent operand constraint
2581 letters (@samp{I}, @samp{J}, @samp{K}, @dots{} @samp{P}) that specify
2582 particular ranges of integer values. If @var{c} is one of those
2583 letters, the expression should check that @var{value}, an integer, is in
2584 the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
2585 not one of those letters, the value should be 0 regardless of
2589 @defmac CONST_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2590 Like @code{CONST_OK_FOR_LETTER_P}, but you also get the constraint
2591 string passed in @var{str}, so that you can use suffixes to distinguish
2592 between different variants.
2595 @defmac CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
2596 A C expression that defines the machine-dependent operand constraint
2597 letters that specify particular ranges of @code{const_double} values
2598 (@samp{G} or @samp{H}).
2600 If @var{c} is one of those letters, the expression should check that
2601 @var{value}, an RTX of code @code{const_double}, is in the appropriate
2602 range and return 1 if so, 0 otherwise. If @var{c} is not one of those
2603 letters, the value should be 0 regardless of @var{value}.
2605 @code{const_double} is used for all floating-point constants and for
2606 @code{DImode} fixed-point constants. A given letter can accept either
2607 or both kinds of values. It can use @code{GET_MODE} to distinguish
2608 between these kinds.
2611 @defmac CONST_DOUBLE_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2612 Like @code{CONST_DOUBLE_OK_FOR_LETTER_P}, but you also get the constraint
2613 string passed in @var{str}, so that you can use suffixes to distinguish
2614 between different variants.
2617 @defmac EXTRA_CONSTRAINT (@var{value}, @var{c})
2618 A C expression that defines the optional machine-dependent constraint
2619 letters that can be used to segregate specific types of operands, usually
2620 memory references, for the target machine. Any letter that is not
2621 elsewhere defined and not matched by @code{REG_CLASS_FROM_LETTER} /
2622 @code{REG_CLASS_FROM_CONSTRAINT}
2623 may be used. Normally this macro will not be defined.
2625 If it is required for a particular target machine, it should return 1
2626 if @var{value} corresponds to the operand type represented by the
2627 constraint letter @var{c}. If @var{c} is not defined as an extra
2628 constraint, the value returned should be 0 regardless of @var{value}.
2630 For example, on the ROMP, load instructions cannot have their output
2631 in r0 if the memory reference contains a symbolic address. Constraint
2632 letter @samp{Q} is defined as representing a memory address that does
2633 @emph{not} contain a symbolic address. An alternative is specified with
2634 a @samp{Q} constraint on the input and @samp{r} on the output. The next
2635 alternative specifies @samp{m} on the input and a register class that
2636 does not include r0 on the output.
2639 @defmac EXTRA_CONSTRAINT_STR (@var{value}, @var{c}, @var{str})
2640 Like @code{EXTRA_CONSTRAINT}, but you also get the constraint string passed
2641 in @var{str}, so that you can use suffixes to distinguish between different
2645 @defmac EXTRA_MEMORY_CONSTRAINT (@var{c}, @var{str})
2646 A C expression that defines the optional machine-dependent constraint
2647 letters, amongst those accepted by @code{EXTRA_CONSTRAINT}, that should
2648 be treated like memory constraints by the reload pass.
2650 It should return 1 if the operand type represented by the constraint
2651 at the start of @var{str}, the first letter of which is the letter @var{c},
2652 comprises a subset of all memory references including
2653 all those whose address is simply a base register. This allows the reload
2654 pass to reload an operand, if it does not directly correspond to the operand
2655 type of @var{c}, by copying its address into a base register.
2657 For example, on the S/390, some instructions do not accept arbitrary
2658 memory references, but only those that do not make use of an index
2659 register. The constraint letter @samp{Q} is defined via
2660 @code{EXTRA_CONSTRAINT} as representing a memory address of this type.
2661 If the letter @samp{Q} is marked as @code{EXTRA_MEMORY_CONSTRAINT},
2662 a @samp{Q} constraint can handle any memory operand, because the
2663 reload pass knows it can be reloaded by copying the memory address
2664 into a base register if required. This is analogous to the way
2665 a @samp{o} constraint can handle any memory operand.
2668 @defmac EXTRA_ADDRESS_CONSTRAINT (@var{c}, @var{str})
2669 A C expression that defines the optional machine-dependent constraint
2670 letters, amongst those accepted by @code{EXTRA_CONSTRAINT} /
2671 @code{EXTRA_CONSTRAINT_STR}, that should
2672 be treated like address constraints by the reload pass.
2674 It should return 1 if the operand type represented by the constraint
2675 at the start of @var{str}, which starts with the letter @var{c}, comprises
2676 a subset of all memory addresses including
2677 all those that consist of just a base register. This allows the reload
2678 pass to reload an operand, if it does not directly correspond to the operand
2679 type of @var{str}, by copying it into a base register.
2681 Any constraint marked as @code{EXTRA_ADDRESS_CONSTRAINT} can only
2682 be used with the @code{address_operand} predicate. It is treated
2683 analogously to the @samp{p} constraint.
2686 @node Stack and Calling
2687 @section Stack Layout and Calling Conventions
2688 @cindex calling conventions
2690 @c prevent bad page break with this line
2691 This describes the stack layout and calling conventions.
2695 * Exception Handling::
2700 * Register Arguments::
2702 * Aggregate Return::
2707 * Stack Smashing Protection::
2711 @subsection Basic Stack Layout
2712 @cindex stack frame layout
2713 @cindex frame layout
2715 @c prevent bad page break with this line
2716 Here is the basic stack layout.
2718 @defmac STACK_GROWS_DOWNWARD
2719 Define this macro if pushing a word onto the stack moves the stack
2720 pointer to a smaller address.
2722 When we say, ``define this macro if @dots{}'', it means that the
2723 compiler checks this macro only with @code{#ifdef} so the precise
2724 definition used does not matter.
2727 @defmac STACK_PUSH_CODE
2728 This macro defines the operation used when something is pushed
2729 on the stack. In RTL, a push operation will be
2730 @code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})}
2732 The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC},
2733 and @code{POST_INC}. Which of these is correct depends on
2734 the stack direction and on whether the stack pointer points
2735 to the last item on the stack or whether it points to the
2736 space for the next item on the stack.
2738 The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is
2739 defined, which is almost always right, and @code{PRE_INC} otherwise,
2740 which is often wrong.
2743 @defmac FRAME_GROWS_DOWNWARD
2744 Define this macro to nonzero value if the addresses of local variable slots
2745 are at negative offsets from the frame pointer.
2748 @defmac ARGS_GROW_DOWNWARD
2749 Define this macro if successive arguments to a function occupy decreasing
2750 addresses on the stack.
2753 @defmac STARTING_FRAME_OFFSET
2754 Offset from the frame pointer to the first local variable slot to be allocated.
2756 If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by
2757 subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}.
2758 Otherwise, it is found by adding the length of the first slot to the
2759 value @code{STARTING_FRAME_OFFSET}.
2760 @c i'm not sure if the above is still correct.. had to change it to get
2761 @c rid of an overfull. --mew 2feb93
2764 @defmac STACK_ALIGNMENT_NEEDED
2765 Define to zero to disable final alignment of the stack during reload.
2766 The nonzero default for this macro is suitable for most ports.
2768 On ports where @code{STARTING_FRAME_OFFSET} is nonzero or where there
2769 is a register save block following the local block that doesn't require
2770 alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable
2771 stack alignment and do it in the backend.
2774 @defmac STACK_POINTER_OFFSET
2775 Offset from the stack pointer register to the first location at which
2776 outgoing arguments are placed. If not specified, the default value of
2777 zero is used. This is the proper value for most machines.
2779 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
2780 the first location at which outgoing arguments are placed.
2783 @defmac FIRST_PARM_OFFSET (@var{fundecl})
2784 Offset from the argument pointer register to the first argument's
2785 address. On some machines it may depend on the data type of the
2788 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
2789 the first argument's address.
2792 @defmac STACK_DYNAMIC_OFFSET (@var{fundecl})
2793 Offset from the stack pointer register to an item dynamically allocated
2794 on the stack, e.g., by @code{alloca}.
2796 The default value for this macro is @code{STACK_POINTER_OFFSET} plus the
2797 length of the outgoing arguments. The default is correct for most
2798 machines. See @file{function.c} for details.
2801 @defmac INITIAL_FRAME_ADDRESS_RTX
2802 A C expression whose value is RTL representing the address of the initial
2803 stack frame. This address is passed to @code{RETURN_ADDR_RTX} and
2804 @code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable
2805 default value will be used. Define this macro in order to make frame pointer
2806 elimination work in the presence of @code{__builtin_frame_address (count)} and
2807 @code{__builtin_return_address (count)} for @code{count} not equal to zero.
2810 @defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr})
2811 A C expression whose value is RTL representing the address in a stack
2812 frame where the pointer to the caller's frame is stored. Assume that
2813 @var{frameaddr} is an RTL expression for the address of the stack frame
2816 If you don't define this macro, the default is to return the value
2817 of @var{frameaddr}---that is, the stack frame address is also the
2818 address of the stack word that points to the previous frame.
2821 @defmac SETUP_FRAME_ADDRESSES
2822 If defined, a C expression that produces the machine-specific code to
2823 setup the stack so that arbitrary frames can be accessed. For example,
2824 on the SPARC, we must flush all of the register windows to the stack
2825 before we can access arbitrary stack frames. You will seldom need to
2829 @deftypefn {Target Hook} bool TARGET_BUILTIN_SETJMP_FRAME_VALUE ()
2830 This target hook should return an rtx that is used to store
2831 the address of the current frame into the built in @code{setjmp} buffer.
2832 The default value, @code{virtual_stack_vars_rtx}, is correct for most
2833 machines. One reason you may need to define this target hook is if
2834 @code{hard_frame_pointer_rtx} is the appropriate value on your machine.
2837 @defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr})
2838 A C expression whose value is RTL representing the value of the return
2839 address for the frame @var{count} steps up from the current frame, after
2840 the prologue. @var{frameaddr} is the frame pointer of the @var{count}
2841 frame, or the frame pointer of the @var{count} @minus{} 1 frame if
2842 @code{RETURN_ADDR_IN_PREVIOUS_FRAME} is defined.
2844 The value of the expression must always be the correct address when
2845 @var{count} is zero, but may be @code{NULL_RTX} if there is not way to
2846 determine the return address of other frames.
2849 @defmac RETURN_ADDR_IN_PREVIOUS_FRAME
2850 Define this if the return address of a particular stack frame is accessed
2851 from the frame pointer of the previous stack frame.
2854 @defmac INCOMING_RETURN_ADDR_RTX
2855 A C expression whose value is RTL representing the location of the
2856 incoming return address at the beginning of any function, before the
2857 prologue. This RTL is either a @code{REG}, indicating that the return
2858 value is saved in @samp{REG}, or a @code{MEM} representing a location in
2861 You only need to define this macro if you want to support call frame
2862 debugging information like that provided by DWARF 2.
2864 If this RTL is a @code{REG}, you should also define
2865 @code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}.
2868 @defmac DWARF_ALT_FRAME_RETURN_COLUMN
2869 A C expression whose value is an integer giving a DWARF 2 column
2870 number that may be used as an alternate return column. This should
2871 be defined only if @code{DWARF_FRAME_RETURN_COLUMN} is set to a
2872 general register, but an alternate column needs to be used for
2876 @defmac DWARF_ZERO_REG
2877 A C expression whose value is an integer giving a DWARF 2 register
2878 number that is considered to always have the value zero. This should
2879 only be defined if the target has an architected zero register, and
2880 someone decided it was a good idea to use that register number to
2881 terminate the stack backtrace. New ports should avoid this.
2884 @deftypefn {Target Hook} void TARGET_DWARF_HANDLE_FRAME_UNSPEC (const char *@var{label}, rtx @var{pattern}, int @var{index})
2885 This target hook allows the backend to emit frame-related insns that
2886 contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging
2887 info engine will invoke it on insns of the form
2889 (set (reg) (unspec [...] UNSPEC_INDEX))
2893 (set (reg) (unspec_volatile [...] UNSPECV_INDEX)).
2895 to let the backend emit the call frame instructions. @var{label} is
2896 the CFI label attached to the insn, @var{pattern} is the pattern of
2897 the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}.
2900 @defmac INCOMING_FRAME_SP_OFFSET
2901 A C expression whose value is an integer giving the offset, in bytes,
2902 from the value of the stack pointer register to the top of the stack
2903 frame at the beginning of any function, before the prologue. The top of
2904 the frame is defined to be the value of the stack pointer in the
2905 previous frame, just before the call instruction.
2907 You only need to define this macro if you want to support call frame
2908 debugging information like that provided by DWARF 2.
2911 @defmac ARG_POINTER_CFA_OFFSET (@var{fundecl})
2912 A C expression whose value is an integer giving the offset, in bytes,
2913 from the argument pointer to the canonical frame address (cfa). The
2914 final value should coincide with that calculated by
2915 @code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable
2916 during virtual register instantiation.
2918 The default value for this macro is @code{FIRST_PARM_OFFSET (fundecl)},
2919 which is correct for most machines; in general, the arguments are found
2920 immediately before the stack frame. Note that this is not the case on
2921 some targets that save registers into the caller's frame, such as SPARC
2922 and rs6000, and so such targets need to define this macro.
2924 You only need to define this macro if the default is incorrect, and you
2925 want to support call frame debugging information like that provided by
2929 @defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl})
2930 If defined, a C expression whose value is an integer giving the offset
2931 in bytes from the frame pointer to the canonical frame address (cfa).
2932 The final value should conincide with that calculated by
2933 @code{INCOMING_FRAME_SP_OFFSET}.
2935 Normally the CFA is calculated as an offset from the argument pointer,
2936 via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is
2937 variable due to the ABI, this may not be possible. If this macro is
2938 defined, it imples that the virtual register instantiation should be
2939 based on the frame pointer instead of the argument pointer. Only one
2940 of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET}
2944 @node Exception Handling
2945 @subsection Exception Handling Support
2946 @cindex exception handling
2948 @defmac EH_RETURN_DATA_REGNO (@var{N})
2949 A C expression whose value is the @var{N}th register number used for
2950 data by exception handlers, or @code{INVALID_REGNUM} if fewer than
2951 @var{N} registers are usable.
2953 The exception handling library routines communicate with the exception
2954 handlers via a set of agreed upon registers. Ideally these registers
2955 should be call-clobbered; it is possible to use call-saved registers,
2956 but may negatively impact code size. The target must support at least
2957 2 data registers, but should define 4 if there are enough free registers.
2959 You must define this macro if you want to support call frame exception
2960 handling like that provided by DWARF 2.
2963 @defmac EH_RETURN_STACKADJ_RTX
2964 A C expression whose value is RTL representing a location in which
2965 to store a stack adjustment to be applied before function return.
2966 This is used to unwind the stack to an exception handler's call frame.
2967 It will be assigned zero on code paths that return normally.
2969 Typically this is a call-clobbered hard register that is otherwise
2970 untouched by the epilogue, but could also be a stack slot.
2972 Do not define this macro if the stack pointer is saved and restored
2973 by the regular prolog and epilog code in the call frame itself; in
2974 this case, the exception handling library routines will update the
2975 stack location to be restored in place. Otherwise, you must define
2976 this macro if you want to support call frame exception handling like
2977 that provided by DWARF 2.
2980 @defmac EH_RETURN_HANDLER_RTX
2981 A C expression whose value is RTL representing a location in which
2982 to store the address of an exception handler to which we should
2983 return. It will not be assigned on code paths that return normally.
2985 Typically this is the location in the call frame at which the normal
2986 return address is stored. For targets that return by popping an
2987 address off the stack, this might be a memory address just below
2988 the @emph{target} call frame rather than inside the current call
2989 frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already
2990 been assigned, so it may be used to calculate the location of the
2993 Some targets have more complex requirements than storing to an
2994 address calculable during initial code generation. In that case
2995 the @code{eh_return} instruction pattern should be used instead.
2997 If you want to support call frame exception handling, you must
2998 define either this macro or the @code{eh_return} instruction pattern.
3001 @defmac RETURN_ADDR_OFFSET
3002 If defined, an integer-valued C expression for which rtl will be generated
3003 to add it to the exception handler address before it is searched in the
3004 exception handling tables, and to subtract it again from the address before
3005 using it to return to the exception handler.
3008 @defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global})
3009 This macro chooses the encoding of pointers embedded in the exception
3010 handling sections. If at all possible, this should be defined such
3011 that the exception handling section will not require dynamic relocations,
3012 and so may be read-only.
3014 @var{code} is 0 for data, 1 for code labels, 2 for function pointers.
3015 @var{global} is true if the symbol may be affected by dynamic relocations.
3016 The macro should return a combination of the @code{DW_EH_PE_*} defines
3017 as found in @file{dwarf2.h}.
3019 If this macro is not defined, pointers will not be encoded but
3020 represented directly.
3023 @defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done})
3024 This macro allows the target to emit whatever special magic is required
3025 to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}.
3026 Generic code takes care of pc-relative and indirect encodings; this must
3027 be defined if the target uses text-relative or data-relative encodings.
3029 This is a C statement that branches to @var{done} if the format was
3030 handled. @var{encoding} is the format chosen, @var{size} is the number
3031 of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF}
3035 @defmac MD_UNWIND_SUPPORT
3036 A string specifying a file to be #include'd in unwind-dw2.c. The file
3037 so included typically defines @code{MD_FALLBACK_FRAME_STATE_FOR}.
3040 @defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs})
3041 This macro allows the target to add cpu and operating system specific
3042 code to the call-frame unwinder for use when there is no unwind data
3043 available. The most common reason to implement this macro is to unwind
3044 through signal frames.
3046 This macro is called from @code{uw_frame_state_for} in @file{unwind-dw2.c}
3047 and @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context};
3048 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra}
3049 for the address of the code being executed and @code{context->cfa} for
3050 the stack pointer value. If the frame can be decoded, the register save
3051 addresses should be updated in @var{fs} and the macro should evaluate to
3052 @code{_URC_NO_REASON}. If the frame cannot be decoded, the macro should
3053 evaluate to @code{_URC_END_OF_STACK}.
3055 For proper signal handling in Java this macro is accompanied by
3056 @code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers.
3059 @defmac MD_HANDLE_UNWABI (@var{context}, @var{fs})
3060 This macro allows the target to add operating system specific code to the
3061 call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive,
3062 usually used for signal or interrupt frames.
3064 This macro is called from @code{uw_update_context} in @file{unwind-ia64.c}.
3065 @var{context} is an @code{_Unwind_Context};
3066 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi}
3067 for the abi and context in the @code{.unwabi} directive. If the
3068 @code{.unwabi} directive can be handled, the register save addresses should
3069 be updated in @var{fs}.
3072 @defmac TARGET_USES_WEAK_UNWIND_INFO
3073 A C expression that evaluates to true if the target requires unwind
3074 info to be given comdat linkage. Define it to be @code{1} if comdat
3075 linkage is necessary. The default is @code{0}.
3078 @node Stack Checking
3079 @subsection Specifying How Stack Checking is Done
3081 GCC will check that stack references are within the boundaries of
3082 the stack, if the @option{-fstack-check} is specified, in one of three ways:
3086 If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC
3087 will assume that you have arranged for stack checking to be done at
3088 appropriate places in the configuration files, e.g., in
3089 @code{TARGET_ASM_FUNCTION_PROLOGUE}. GCC will do not other special
3093 If @code{STACK_CHECK_BUILTIN} is zero and you defined a named pattern
3094 called @code{check_stack} in your @file{md} file, GCC will call that
3095 pattern with one argument which is the address to compare the stack
3096 value against. You must arrange for this pattern to report an error if
3097 the stack pointer is out of range.
3100 If neither of the above are true, GCC will generate code to periodically
3101 ``probe'' the stack pointer using the values of the macros defined below.
3104 Normally, you will use the default values of these macros, so GCC
3105 will use the third approach.
3107 @defmac STACK_CHECK_BUILTIN
3108 A nonzero value if stack checking is done by the configuration files in a
3109 machine-dependent manner. You should define this macro if stack checking
3110 is require by the ABI of your machine or if you would like to have to stack
3111 checking in some more efficient way than GCC's portable approach.
3112 The default value of this macro is zero.
3115 @defmac STACK_CHECK_PROBE_INTERVAL
3116 An integer representing the interval at which GCC must generate stack
3117 probe instructions. You will normally define this macro to be no larger
3118 than the size of the ``guard pages'' at the end of a stack area. The
3119 default value of 4096 is suitable for most systems.
3122 @defmac STACK_CHECK_PROBE_LOAD
3123 A integer which is nonzero if GCC should perform the stack probe
3124 as a load instruction and zero if GCC should use a store instruction.
3125 The default is zero, which is the most efficient choice on most systems.
3128 @defmac STACK_CHECK_PROTECT
3129 The number of bytes of stack needed to recover from a stack overflow,
3130 for languages where such a recovery is supported. The default value of
3131 75 words should be adequate for most machines.
3134 @defmac STACK_CHECK_MAX_FRAME_SIZE
3135 The maximum size of a stack frame, in bytes. GCC will generate probe
3136 instructions in non-leaf functions to ensure at least this many bytes of
3137 stack are available. If a stack frame is larger than this size, stack
3138 checking will not be reliable and GCC will issue a warning. The
3139 default is chosen so that GCC only generates one instruction on most
3140 systems. You should normally not change the default value of this macro.
3143 @defmac STACK_CHECK_FIXED_FRAME_SIZE
3144 GCC uses this value to generate the above warning message. It
3145 represents the amount of fixed frame used by a function, not including
3146 space for any callee-saved registers, temporaries and user variables.
3147 You need only specify an upper bound for this amount and will normally
3148 use the default of four words.
3151 @defmac STACK_CHECK_MAX_VAR_SIZE
3152 The maximum size, in bytes, of an object that GCC will place in the
3153 fixed area of the stack frame when the user specifies
3154 @option{-fstack-check}.
3155 GCC computed the default from the values of the above macros and you will
3156 normally not need to override that default.
3160 @node Frame Registers
3161 @subsection Registers That Address the Stack Frame
3163 @c prevent bad page break with this line
3164 This discusses registers that address the stack frame.
3166 @defmac STACK_POINTER_REGNUM
3167 The register number of the stack pointer register, which must also be a
3168 fixed register according to @code{FIXED_REGISTERS}. On most machines,
3169 the hardware determines which register this is.
3172 @defmac FRAME_POINTER_REGNUM
3173 The register number of the frame pointer register, which is used to
3174 access automatic variables in the stack frame. On some machines, the
3175 hardware determines which register this is. On other machines, you can
3176 choose any register you wish for this purpose.
3179 @defmac HARD_FRAME_POINTER_REGNUM
3180 On some machines the offset between the frame pointer and starting
3181 offset of the automatic variables is not known until after register
3182 allocation has been done (for example, because the saved registers are
3183 between these two locations). On those machines, define
3184 @code{FRAME_POINTER_REGNUM} the number of a special, fixed register to
3185 be used internally until the offset is known, and define
3186 @code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number
3187 used for the frame pointer.
3189 You should define this macro only in the very rare circumstances when it
3190 is not possible to calculate the offset between the frame pointer and
3191 the automatic variables until after register allocation has been
3192 completed. When this macro is defined, you must also indicate in your
3193 definition of @code{ELIMINABLE_REGS} how to eliminate
3194 @code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM}
3195 or @code{STACK_POINTER_REGNUM}.
3197 Do not define this macro if it would be the same as
3198 @code{FRAME_POINTER_REGNUM}.
3201 @defmac ARG_POINTER_REGNUM
3202 The register number of the arg pointer register, which is used to access
3203 the function's argument list. On some machines, this is the same as the
3204 frame pointer register. On some machines, the hardware determines which
3205 register this is. On other machines, you can choose any register you
3206 wish for this purpose. If this is not the same register as the frame
3207 pointer register, then you must mark it as a fixed register according to
3208 @code{FIXED_REGISTERS}, or arrange to be able to eliminate it
3209 (@pxref{Elimination}).
3212 @defmac RETURN_ADDRESS_POINTER_REGNUM
3213 The register number of the return address pointer register, which is used to
3214 access the current function's return address from the stack. On some
3215 machines, the return address is not at a fixed offset from the frame
3216 pointer or stack pointer or argument pointer. This register can be defined
3217 to point to the return address on the stack, and then be converted by
3218 @code{ELIMINABLE_REGS} into either the frame pointer or stack pointer.
3220 Do not define this macro unless there is no other way to get the return
3221 address from the stack.
3224 @defmac STATIC_CHAIN_REGNUM
3225 @defmacx STATIC_CHAIN_INCOMING_REGNUM
3226 Register numbers used for passing a function's static chain pointer. If
3227 register windows are used, the register number as seen by the called
3228 function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register
3229 number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If
3230 these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need
3233 The static chain register need not be a fixed register.
3235 If the static chain is passed in memory, these macros should not be
3236 defined; instead, the next two macros should be defined.
3239 @defmac STATIC_CHAIN
3240 @defmacx STATIC_CHAIN_INCOMING
3241 If the static chain is passed in memory, these macros provide rtx giving
3242 @code{mem} expressions that denote where they are stored.
3243 @code{STATIC_CHAIN} and @code{STATIC_CHAIN_INCOMING} give the locations
3244 as seen by the calling and called functions, respectively. Often the former
3245 will be at an offset from the stack pointer and the latter at an offset from
3248 @findex stack_pointer_rtx
3249 @findex frame_pointer_rtx
3250 @findex arg_pointer_rtx
3251 The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and
3252 @code{arg_pointer_rtx} will have been initialized prior to the use of these
3253 macros and should be used to refer to those items.
3255 If the static chain is passed in a register, the two previous macros should
3259 @defmac DWARF_FRAME_REGISTERS
3260 This macro specifies the maximum number of hard registers that can be
3261 saved in a call frame. This is used to size data structures used in
3262 DWARF2 exception handling.
3264 Prior to GCC 3.0, this macro was needed in order to establish a stable
3265 exception handling ABI in the face of adding new hard registers for ISA
3266 extensions. In GCC 3.0 and later, the EH ABI is insulated from changes
3267 in the number of hard registers. Nevertheless, this macro can still be
3268 used to reduce the runtime memory requirements of the exception handling
3269 routines, which can be substantial if the ISA contains a lot of
3270 registers that are not call-saved.
3272 If this macro is not defined, it defaults to
3273 @code{FIRST_PSEUDO_REGISTER}.
3276 @defmac PRE_GCC3_DWARF_FRAME_REGISTERS
3278 This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided
3279 for backward compatibility in pre GCC 3.0 compiled code.
3281 If this macro is not defined, it defaults to
3282 @code{DWARF_FRAME_REGISTERS}.
3285 @defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno})
3287 Define this macro if the target's representation for dwarf registers
3288 is different than the internal representation for unwind column.
3289 Given a dwarf register, this macro should return the internal unwind
3290 column number to use instead.
3292 See the PowerPC's SPE target for an example.
3295 @defmac DWARF_FRAME_REGNUM (@var{regno})
3297 Define this macro if the target's representation for dwarf registers
3298 used in .eh_frame or .debug_frame is different from that used in other
3299 debug info sections. Given a GCC hard register number, this macro
3300 should return the .eh_frame register number. The default is
3301 @code{DBX_REGISTER_NUMBER (@var{regno})}.
3305 @defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh})
3307 Define this macro to map register numbers held in the call frame info
3308 that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that
3309 should be output in .debug_frame (@code{@var{for_eh}} is zero) and
3310 .eh_frame (@code{@var{for_eh}} is nonzero). The default is to
3311 return @code{@var{regno}}.
3316 @subsection Eliminating Frame Pointer and Arg Pointer
3318 @c prevent bad page break with this line
3319 This is about eliminating the frame pointer and arg pointer.
3321 @defmac FRAME_POINTER_REQUIRED
3322 A C expression which is nonzero if a function must have and use a frame
3323 pointer. This expression is evaluated in the reload pass. If its value is
3324 nonzero the function will have a frame pointer.
3326 The expression can in principle examine the current function and decide
3327 according to the facts, but on most machines the constant 0 or the
3328 constant 1 suffices. Use 0 when the machine allows code to be generated
3329 with no frame pointer, and doing so saves some time or space. Use 1
3330 when there is no possible advantage to avoiding a frame pointer.
3332 In certain cases, the compiler does not know how to produce valid code
3333 without a frame pointer. The compiler recognizes those cases and
3334 automatically gives the function a frame pointer regardless of what
3335 @code{FRAME_POINTER_REQUIRED} says. You don't need to worry about
3338 In a function that does not require a frame pointer, the frame pointer
3339 register can be allocated for ordinary usage, unless you mark it as a
3340 fixed register. See @code{FIXED_REGISTERS} for more information.
3343 @findex get_frame_size
3344 @defmac INITIAL_FRAME_POINTER_OFFSET (@var{depth-var})
3345 A C statement to store in the variable @var{depth-var} the difference
3346 between the frame pointer and the stack pointer values immediately after
3347 the function prologue. The value would be computed from information
3348 such as the result of @code{get_frame_size ()} and the tables of
3349 registers @code{regs_ever_live} and @code{call_used_regs}.
3351 If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and
3352 need not be defined. Otherwise, it must be defined even if
3353 @code{FRAME_POINTER_REQUIRED} is defined to always be true; in that
3354 case, you may set @var{depth-var} to anything.
3357 @defmac ELIMINABLE_REGS
3358 If defined, this macro specifies a table of register pairs used to
3359 eliminate unneeded registers that point into the stack frame. If it is not
3360 defined, the only elimination attempted by the compiler is to replace
3361 references to the frame pointer with references to the stack pointer.
3363 The definition of this macro is a list of structure initializations, each
3364 of which specifies an original and replacement register.
3366 On some machines, the position of the argument pointer is not known until
3367 the compilation is completed. In such a case, a separate hard register
3368 must be used for the argument pointer. This register can be eliminated by
3369 replacing it with either the frame pointer or the argument pointer,
3370 depending on whether or not the frame pointer has been eliminated.
3372 In this case, you might specify:
3374 #define ELIMINABLE_REGS \
3375 @{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \
3376 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \
3377 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@}
3380 Note that the elimination of the argument pointer with the stack pointer is
3381 specified first since that is the preferred elimination.
3384 @defmac CAN_ELIMINATE (@var{from-reg}, @var{to-reg})
3385 A C expression that returns nonzero if the compiler is allowed to try
3386 to replace register number @var{from-reg} with register number
3387 @var{to-reg}. This macro need only be defined if @code{ELIMINABLE_REGS}
3388 is defined, and will usually be the constant 1, since most of the cases
3389 preventing register elimination are things that the compiler already
3393 @defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var})
3394 This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It
3395 specifies the initial difference between the specified pair of
3396 registers. This macro must be defined if @code{ELIMINABLE_REGS} is
3400 @node Stack Arguments
3401 @subsection Passing Function Arguments on the Stack
3402 @cindex arguments on stack
3403 @cindex stack arguments
3405 The macros in this section control how arguments are passed
3406 on the stack. See the following section for other macros that
3407 control passing certain arguments in registers.
3409 @deftypefn {Target Hook} bool TARGET_PROMOTE_PROTOTYPES (tree @var{fntype})
3410 This target hook returns @code{true} if an argument declared in a
3411 prototype as an integral type smaller than @code{int} should actually be
3412 passed as an @code{int}. In addition to avoiding errors in certain
3413 cases of mismatch, it also makes for better code on certain machines.
3414 The default is to not promote prototypes.
3418 A C expression. If nonzero, push insns will be used to pass
3420 If the target machine does not have a push instruction, set it to zero.
3421 That directs GCC to use an alternate strategy: to
3422 allocate the entire argument block and then store the arguments into
3423 it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too.
3426 @defmac PUSH_ARGS_REVERSED
3427 A C expression. If nonzero, function arguments will be evaluated from
3428 last to first, rather than from first to last. If this macro is not
3429 defined, it defaults to @code{PUSH_ARGS} on targets where the stack
3430 and args grow in opposite directions, and 0 otherwise.
3433 @defmac PUSH_ROUNDING (@var{npushed})
3434 A C expression that is the number of bytes actually pushed onto the
3435 stack when an instruction attempts to push @var{npushed} bytes.
3437 On some machines, the definition
3440 #define PUSH_ROUNDING(BYTES) (BYTES)
3444 will suffice. But on other machines, instructions that appear
3445 to push one byte actually push two bytes in an attempt to maintain
3446 alignment. Then the definition should be
3449 #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
3453 @findex current_function_outgoing_args_size
3454 @defmac ACCUMULATE_OUTGOING_ARGS
3455 A C expression. If nonzero, the maximum amount of space required for outgoing arguments
3456 will be computed and placed into the variable
3457 @code{current_function_outgoing_args_size}. No space will be pushed
3458 onto the stack for each call; instead, the function prologue should
3459 increase the stack frame size by this amount.
3461 Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS}
3465 @defmac REG_PARM_STACK_SPACE (@var{fndecl})
3466 Define this macro if functions should assume that stack space has been
3467 allocated for arguments even when their values are passed in
3470 The value of this macro is the size, in bytes, of the area reserved for
3471 arguments passed in registers for the function represented by @var{fndecl},
3472 which can be zero if GCC is calling a library function.
3474 This space can be allocated by the caller, or be a part of the
3475 machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says
3478 @c above is overfull. not sure what to do. --mew 5feb93 did
3479 @c something, not sure if it looks good. --mew 10feb93
3481 @defmac OUTGOING_REG_PARM_STACK_SPACE
3482 Define this if it is the responsibility of the caller to allocate the area
3483 reserved for arguments passed in registers.
3485 If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls
3486 whether the space for these arguments counts in the value of
3487 @code{current_function_outgoing_args_size}.
3490 @defmac STACK_PARMS_IN_REG_PARM_AREA
3491 Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the
3492 stack parameters don't skip the area specified by it.
3493 @c i changed this, makes more sens and it should have taken care of the
3494 @c overfull.. not as specific, tho. --mew 5feb93
3496 Normally, when a parameter is not passed in registers, it is placed on the
3497 stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro
3498 suppresses this behavior and causes the parameter to be passed on the
3499 stack in its natural location.
3502 @defmac RETURN_POPS_ARGS (@var{fundecl}, @var{funtype}, @var{stack-size})
3503 A C expression that should indicate the number of bytes of its own
3504 arguments that a function pops on returning, or 0 if the
3505 function pops no arguments and the caller must therefore pop them all
3506 after the function returns.
3508 @var{fundecl} is a C variable whose value is a tree node that describes
3509 the function in question. Normally it is a node of type
3510 @code{FUNCTION_DECL} that describes the declaration of the function.
3511 From this you can obtain the @code{DECL_ATTRIBUTES} of the function.
3513 @var{funtype} is a C variable whose value is a tree node that
3514 describes the function in question. Normally it is a node of type
3515 @code{FUNCTION_TYPE} that describes the data type of the function.
3516 From this it is possible to obtain the data types of the value and
3517 arguments (if known).
3519 When a call to a library function is being considered, @var{fundecl}
3520 will contain an identifier node for the library function. Thus, if
3521 you need to distinguish among various library functions, you can do so
3522 by their names. Note that ``library function'' in this context means
3523 a function used to perform arithmetic, whose name is known specially
3524 in the compiler and was not mentioned in the C code being compiled.
3526 @var{stack-size} is the number of bytes of arguments passed on the
3527 stack. If a variable number of bytes is passed, it is zero, and
3528 argument popping will always be the responsibility of the calling function.
3530 On the VAX, all functions always pop their arguments, so the definition
3531 of this macro is @var{stack-size}. On the 68000, using the standard
3532 calling convention, no functions pop their arguments, so the value of
3533 the macro is always 0 in this case. But an alternative calling
3534 convention is available in which functions that take a fixed number of
3535 arguments pop them but other functions (such as @code{printf}) pop
3536 nothing (the caller pops all). When this convention is in use,
3537 @var{funtype} is examined to determine whether a function takes a fixed
3538 number of arguments.
3541 @defmac CALL_POPS_ARGS (@var{cum})
3542 A C expression that should indicate the number of bytes a call sequence
3543 pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS}
3544 when compiling a function call.
3546 @var{cum} is the variable in which all arguments to the called function
3547 have been accumulated.
3549 On certain architectures, such as the SH5, a call trampoline is used
3550 that pops certain registers off the stack, depending on the arguments
3551 that have been passed to the function. Since this is a property of the
3552 call site, not of the called function, @code{RETURN_POPS_ARGS} is not
3556 @node Register Arguments
3557 @subsection Passing Arguments in Registers
3558 @cindex arguments in registers
3559 @cindex registers arguments
3561 This section describes the macros which let you control how various
3562 types of arguments are passed in registers or how they are arranged in
3565 @defmac FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
3566 A C expression that controls whether a function argument is passed
3567 in a register, and which register.
3569 The arguments are @var{cum}, which summarizes all the previous
3570 arguments; @var{mode}, the machine mode of the argument; @var{type},
3571 the data type of the argument as a tree node or 0 if that is not known
3572 (which happens for C support library functions); and @var{named},
3573 which is 1 for an ordinary argument and 0 for nameless arguments that
3574 correspond to @samp{@dots{}} in the called function's prototype.
3575 @var{type} can be an incomplete type if a syntax error has previously
3578 The value of the expression is usually either a @code{reg} RTX for the
3579 hard register in which to pass the argument, or zero to pass the
3580 argument on the stack.
3582 For machines like the VAX and 68000, where normally all arguments are
3583 pushed, zero suffices as a definition.
3585 The value of the expression can also be a @code{parallel} RTX@. This is
3586 used when an argument is passed in multiple locations. The mode of the
3587 @code{parallel} should be the mode of the entire argument. The
3588 @code{parallel} holds any number of @code{expr_list} pairs; each one
3589 describes where part of the argument is passed. In each
3590 @code{expr_list} the first operand must be a @code{reg} RTX for the hard
3591 register in which to pass this part of the argument, and the mode of the
3592 register RTX indicates how large this part of the argument is. The
3593 second operand of the @code{expr_list} is a @code{const_int} which gives
3594 the offset in bytes into the entire argument of where this part starts.
3595 As a special exception the first @code{expr_list} in the @code{parallel}
3596 RTX may have a first operand of zero. This indicates that the entire
3597 argument is also stored on the stack.
3599 The last time this macro is called, it is called with @code{MODE ==
3600 VOIDmode}, and its result is passed to the @code{call} or @code{call_value}
3601 pattern as operands 2 and 3 respectively.
3603 @cindex @file{stdarg.h} and register arguments
3604 The usual way to make the ISO library @file{stdarg.h} work on a machine
3605 where some arguments are usually passed in registers, is to cause
3606 nameless arguments to be passed on the stack instead. This is done
3607 by making @code{FUNCTION_ARG} return 0 whenever @var{named} is 0.
3609 @cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{FUNCTION_ARG}
3610 @cindex @code{REG_PARM_STACK_SPACE}, and @code{FUNCTION_ARG}
3611 You may use the hook @code{targetm.calls.must_pass_in_stack}
3612 in the definition of this macro to determine if this argument is of a
3613 type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE}
3614 is not defined and @code{FUNCTION_ARG} returns nonzero for such an
3615 argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is
3616 defined, the argument will be computed in the stack and then loaded into
3620 @deftypefn {Target Hook} bool TARGET_MUST_PASS_IN_STACK (enum machine_mode @var{mode}, tree @var{type})
3621 This target hook should return @code{true} if we should not pass @var{type}
3622 solely in registers. The file @file{expr.h} defines a
3623 definition that is usually appropriate, refer to @file{expr.h} for additional
3627 @defmac FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
3628 Define this macro if the target machine has ``register windows'', so
3629 that the register in which a function sees an arguments is not
3630 necessarily the same as the one in which the caller passed the
3633 For such machines, @code{FUNCTION_ARG} computes the register in which
3634 the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
3635 be defined in a similar fashion to tell the function being called
3636 where the arguments will arrive.
3638 If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
3639 serves both purposes.
3642 @deftypefn {Target Hook} int TARGET_ARG_PARTIAL_BYTES (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named})
3643 This target hook returns the number of bytes at the beginning of an
3644 argument that must be put in registers. The value must be zero for
3645 arguments that are passed entirely in registers or that are entirely
3646 pushed on the stack.
3648 On some machines, certain arguments must be passed partially in
3649 registers and partially in memory. On these machines, typically the
3650 first few words of arguments are passed in registers, and the rest
3651 on the stack. If a multi-word argument (a @code{double} or a
3652 structure) crosses that boundary, its first few words must be passed
3653 in registers and the rest must be pushed. This macro tells the
3654 compiler when this occurs, and how many bytes should go in registers.
3656 @code{FUNCTION_ARG} for these arguments should return the first
3657 register to be used by the caller for this argument; likewise
3658 @code{FUNCTION_INCOMING_ARG}, for the called function.
3661 @deftypefn {Target Hook} bool TARGET_PASS_BY_REFERENCE (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named})
3662 This target hook should return @code{true} if an argument at the
3663 position indicated by @var{cum} should be passed by reference. This
3664 predicate is queried after target independent reasons for being
3665 passed by reference, such as @code{TREE_ADDRESSABLE (type)}.
3667 If the hook returns true, a copy of that argument is made in memory and a
3668 pointer to the argument is passed instead of the argument itself.
3669 The pointer is passed in whatever way is appropriate for passing a pointer
3673 @deftypefn {Target Hook} bool TARGET_CALLEE_COPIES (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named})
3674 The function argument described by the parameters to this hook is
3675 known to be passed by reference. The hook should return true if the
3676 function argument should be copied by the callee instead of copied
3679 For any argument for which the hook returns true, if it can be
3680 determined that the argument is not modified, then a copy need
3683 The default version of this hook always returns false.
3686 @defmac CUMULATIVE_ARGS
3687 A C type for declaring a variable that is used as the first argument of
3688 @code{FUNCTION_ARG} and other related values. For some target machines,
3689 the type @code{int} suffices and can hold the number of bytes of
3692 There is no need to record in @code{CUMULATIVE_ARGS} anything about the
3693 arguments that have been passed on the stack. The compiler has other
3694 variables to keep track of that. For target machines on which all
3695 arguments are passed on the stack, there is no need to store anything in
3696 @code{CUMULATIVE_ARGS}; however, the data structure must exist and
3697 should not be empty, so use @code{int}.
3700 @defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args})
3701 A C statement (sans semicolon) for initializing the variable
3702 @var{cum} for the state at the beginning of the argument list. The
3703 variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype}
3704 is the tree node for the data type of the function which will receive
3705 the args, or 0 if the args are to a compiler support library function.
3706 For direct calls that are not libcalls, @var{fndecl} contain the
3707 declaration node of the function. @var{fndecl} is also set when
3708 @code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function
3709 being compiled. @var{n_named_args} is set to the number of named
3710 arguments, including a structure return address if it is passed as a
3711 parameter, when making a call. When processing incoming arguments,
3712 @var{n_named_args} is set to @minus{}1.
3714 When processing a call to a compiler support library function,
3715 @var{libname} identifies which one. It is a @code{symbol_ref} rtx which
3716 contains the name of the function, as a string. @var{libname} is 0 when
3717 an ordinary C function call is being processed. Thus, each time this
3718 macro is called, either @var{libname} or @var{fntype} is nonzero, but
3719 never both of them at once.
3722 @defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname})
3723 Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls,
3724 it gets a @code{MODE} argument instead of @var{fntype}, that would be
3725 @code{NULL}. @var{indirect} would always be zero, too. If this macro
3726 is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname,
3727 0)} is used instead.
3730 @defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname})
3731 Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of
3732 finding the arguments for the function being compiled. If this macro is
3733 undefined, @code{INIT_CUMULATIVE_ARGS} is used instead.
3735 The value passed for @var{libname} is always 0, since library routines
3736 with special calling conventions are never compiled with GCC@. The
3737 argument @var{libname} exists for symmetry with
3738 @code{INIT_CUMULATIVE_ARGS}.
3739 @c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe.
3740 @c --mew 5feb93 i switched the order of the sentences. --mew 10feb93
3743 @defmac FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
3744 A C statement (sans semicolon) to update the summarizer variable
3745 @var{cum} to advance past an argument in the argument list. The
3746 values @var{mode}, @var{type} and @var{named} describe that argument.
3747 Once this is done, the variable @var{cum} is suitable for analyzing
3748 the @emph{following} argument with @code{FUNCTION_ARG}, etc.
3750 This macro need not do anything if the argument in question was passed
3751 on the stack. The compiler knows how to track the amount of stack space
3752 used for arguments without any special help.
3755 @defmac FUNCTION_ARG_PADDING (@var{mode}, @var{type})
3756 If defined, a C expression which determines whether, and in which direction,
3757 to pad out an argument with extra space. The value should be of type
3758 @code{enum direction}: either @code{upward} to pad above the argument,
3759 @code{downward} to pad below, or @code{none} to inhibit padding.
3761 The @emph{amount} of padding is always just enough to reach the next
3762 multiple of @code{FUNCTION_ARG_BOUNDARY}; this macro does not control
3765 This macro has a default definition which is right for most systems.
3766 For little-endian machines, the default is to pad upward. For
3767 big-endian machines, the default is to pad downward for an argument of
3768 constant size shorter than an @code{int}, and upward otherwise.
3771 @defmac PAD_VARARGS_DOWN
3772 If defined, a C expression which determines whether the default
3773 implementation of va_arg will attempt to pad down before reading the
3774 next argument, if that argument is smaller than its aligned space as
3775 controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such
3776 arguments are padded down if @code{BYTES_BIG_ENDIAN} is true.
3779 @defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first})
3780 Specify padding for the last element of a block move between registers and
3781 memory. @var{first} is nonzero if this is the only element. Defining this
3782 macro allows better control of register function parameters on big-endian
3783 machines, without using @code{PARALLEL} rtl. In particular,
3784 @code{MUST_PASS_IN_STACK} need not test padding and mode of types in
3785 registers, as there is no longer a "wrong" part of a register; For example,
3786 a three byte aggregate may be passed in the high part of a register if so
3790 @defmac FUNCTION_ARG_BOUNDARY (@var{mode}, @var{type})
3791 If defined, a C expression that gives the alignment boundary, in bits,
3792 of an argument with the specified mode and type. If it is not defined,
3793 @code{PARM_BOUNDARY} is used for all arguments.
3796 @defmac FUNCTION_ARG_REGNO_P (@var{regno})
3797 A C expression that is nonzero if @var{regno} is the number of a hard
3798 register in which function arguments are sometimes passed. This does
3799 @emph{not} include implicit arguments such as the static chain and
3800 the structure-value address. On many machines, no registers can be
3801 used for this purpose since all function arguments are pushed on the
3805 @deftypefn {Target Hook} bool TARGET_SPLIT_COMPLEX_ARG (tree @var{type})
3806 This hook should return true if parameter of type @var{type} are passed
3807 as two scalar parameters. By default, GCC will attempt to pack complex
3808 arguments into the target's word size. Some ABIs require complex arguments
3809 to be split and treated as their individual components. For example, on
3810 AIX64, complex floats should be passed in a pair of floating point
3811 registers, even though a complex float would fit in one 64-bit floating
3814 The default value of this hook is @code{NULL}, which is treated as always
3818 @deftypefn {Target Hook} tree TARGET_BUILD_BUILTIN_VA_LIST (void)
3819 This hook returns a type node for @code{va_list} for the target.
3820 The default version of the hook returns @code{void*}.
3823 @deftypefn {Target Hook} tree TARGET_GIMPLIFY_VA_ARG_EXPR (tree @var{valist}, tree @var{type}, tree *@var{pre_p}, tree *@var{post_p})
3824 This hook performs target-specific gimplification of
3825 @code{VA_ARG_EXPR}. The first two parameters correspond to the
3826 arguments to @code{va_arg}; the latter two are as in
3827 @code{gimplify.c:gimplify_expr}.
3830 @deftypefn {Target Hook} bool TARGET_VALID_POINTER_MODE (enum machine_mode @var{mode})
3831 Define this to return nonzero if the port can handle pointers
3832 with machine mode @var{mode}. The default version of this
3833 hook returns true for both @code{ptr_mode} and @code{Pmode}.
3836 @deftypefn {Target Hook} bool TARGET_SCALAR_MODE_SUPPORTED_P (enum machine_mode @var{mode})
3837 Define this to return nonzero if the port is prepared to handle
3838 insns involving scalar mode @var{mode}. For a scalar mode to be
3839 considered supported, all the basic arithmetic and comparisons
3842 The default version of this hook returns true for any mode
3843 required to handle the basic C types (as defined by the port).
3844 Included here are the double-word arithmetic supported by the
3845 code in @file{optabs.c}.
3848 @deftypefn {Target Hook} bool TARGET_VECTOR_MODE_SUPPORTED_P (enum machine_mode @var{mode})
3849 Define this to return nonzero if the port is prepared to handle
3850 insns involving vector mode @var{mode}. At the very least, it
3851 must have move patterns for this mode.
3855 @subsection How Scalar Function Values Are Returned
3856 @cindex return values in registers
3857 @cindex values, returned by functions
3858 @cindex scalars, returned as values
3860 This section discusses the macros that control returning scalars as
3861 values---values that can fit in registers.
3863 @defmac FUNCTION_VALUE (@var{valtype}, @var{func})
3864 A C expression to create an RTX representing the place where a
3865 function returns a value of data type @var{valtype}. @var{valtype} is
3866 a tree node representing a data type. Write @code{TYPE_MODE
3867 (@var{valtype})} to get the machine mode used to represent that type.
3868 On many machines, only the mode is relevant. (Actually, on most
3869 machines, scalar values are returned in the same place regardless of
3872 The value of the expression is usually a @code{reg} RTX for the hard
3873 register where the return value is stored. The value can also be a
3874 @code{parallel} RTX, if the return value is in multiple places. See
3875 @code{FUNCTION_ARG} for an explanation of the @code{parallel} form.
3877 If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply the same
3878 promotion rules specified in @code{PROMOTE_MODE} if @var{valtype} is a
3881 If the precise function being called is known, @var{func} is a tree
3882 node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
3883 pointer. This makes it possible to use a different value-returning
3884 convention for specific functions when all their calls are
3887 @code{FUNCTION_VALUE} is not used for return vales with aggregate data
3888 types, because these are returned in another way. See
3889 @code{TARGET_STRUCT_VALUE_RTX} and related macros, below.
3892 @defmac FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
3893 Define this macro if the target machine has ``register windows''
3894 so that the register in which a function returns its value is not
3895 the same as the one in which the caller sees the value.
3897 For such machines, @code{FUNCTION_VALUE} computes the register in which
3898 the caller will see the value. @code{FUNCTION_OUTGOING_VALUE} should be
3899 defined in a similar fashion to tell the function where to put the
3902 If @code{FUNCTION_OUTGOING_VALUE} is not defined,
3903 @code{FUNCTION_VALUE} serves both purposes.
3905 @code{FUNCTION_OUTGOING_VALUE} is not used for return vales with
3906 aggregate data types, because these are returned in another way. See
3907 @code{TARGET_STRUCT_VALUE_RTX} and related macros, below.
3910 @defmac LIBCALL_VALUE (@var{mode})
3911 A C expression to create an RTX representing the place where a library
3912 function returns a value of mode @var{mode}. If the precise function
3913 being called is known, @var{func} is a tree node
3914 (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
3915 pointer. This makes it possible to use a different value-returning
3916 convention for specific functions when all their calls are
3919 Note that ``library function'' in this context means a compiler
3920 support routine, used to perform arithmetic, whose name is known
3921 specially by the compiler and was not mentioned in the C code being
3924 The definition of @code{LIBRARY_VALUE} need not be concerned aggregate
3925 data types, because none of the library functions returns such types.
3928 @defmac FUNCTION_VALUE_REGNO_P (@var{regno})
3929 A C expression that is nonzero if @var{regno} is the number of a hard
3930 register in which the values of called function may come back.
3932 A register whose use for returning values is limited to serving as the
3933 second of a pair (for a value of type @code{double}, say) need not be
3934 recognized by this macro. So for most machines, this definition
3938 #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
3941 If the machine has register windows, so that the caller and the called
3942 function use different registers for the return value, this macro
3943 should recognize only the caller's register numbers.
3946 @defmac APPLY_RESULT_SIZE
3947 Define this macro if @samp{untyped_call} and @samp{untyped_return}
3948 need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for
3949 saving and restoring an arbitrary return value.
3952 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MSB (tree @var{type})
3953 This hook should return true if values of type @var{type} are returned
3954 at the most significant end of a register (in other words, if they are
3955 padded at the least significant end). You can assume that @var{type}
3956 is returned in a register; the caller is required to check this.
3958 Note that the register provided by @code{FUNCTION_VALUE} must be able
3959 to hold the complete return value. For example, if a 1-, 2- or 3-byte
3960 structure is returned at the most significant end of a 4-byte register,
3961 @code{FUNCTION_VALUE} should provide an @code{SImode} rtx.
3964 @node Aggregate Return
3965 @subsection How Large Values Are Returned
3966 @cindex aggregates as return values
3967 @cindex large return values
3968 @cindex returning aggregate values
3969 @cindex structure value address
3971 When a function value's mode is @code{BLKmode} (and in some other
3972 cases), the value is not returned according to @code{FUNCTION_VALUE}
3973 (@pxref{Scalar Return}). Instead, the caller passes the address of a
3974 block of memory in which the value should be stored. This address
3975 is called the @dfn{structure value address}.
3977 This section describes how to control returning structure values in
3980 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MEMORY (tree @var{type}, tree @var{fntype})
3981 This target hook should return a nonzero value to say to return the
3982 function value in memory, just as large structures are always returned.
3983 Here @var{type} will be the data type of the value, and @var{fntype}
3984 will be the type of the function doing the returning, or @code{NULL} for
3987 Note that values of mode @code{BLKmode} must be explicitly handled
3988 by this function. Also, the option @option{-fpcc-struct-return}
3989 takes effect regardless of this macro. On most systems, it is
3990 possible to leave the hook undefined; this causes a default
3991 definition to be used, whose value is the constant 1 for @code{BLKmode}
3992 values, and 0 otherwise.
3994 Do not use this hook to indicate that structures and unions should always
3995 be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN}
3999 @defmac DEFAULT_PCC_STRUCT_RETURN
4000 Define this macro to be 1 if all structure and union return values must be
4001 in memory. Since this results in slower code, this should be defined
4002 only if needed for compatibility with other compilers or with an ABI@.
4003 If you define this macro to be 0, then the conventions used for structure
4004 and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY}
4007 If not defined, this defaults to the value 1.
4010 @deftypefn {Target Hook} rtx TARGET_STRUCT_VALUE_RTX (tree @var{fndecl}, int @var{incoming})
4011 This target hook should return the location of the structure value
4012 address (normally a @code{mem} or @code{reg}), or 0 if the address is
4013 passed as an ``invisible'' first argument. Note that @var{fndecl} may
4014 be @code{NULL}, for libcalls. You do not need to define this target
4015 hook if the address is always passed as an ``invisible'' first
4018 On some architectures the place where the structure value address
4019 is found by the called function is not the same place that the
4020 caller put it. This can be due to register windows, or it could
4021 be because the function prologue moves it to a different place.
4022 @var{incoming} is @code{true} when the location is needed in
4023 the context of the called function, and @code{false} in the context of
4026 If @var{incoming} is @code{true} and the address is to be found on the
4027 stack, return a @code{mem} which refers to the frame pointer.
4030 @defmac PCC_STATIC_STRUCT_RETURN
4031 Define this macro if the usual system convention on the target machine
4032 for returning structures and unions is for the called function to return
4033 the address of a static variable containing the value.
4035 Do not define this if the usual system convention is for the caller to
4036 pass an address to the subroutine.
4038 This macro has effect in @option{-fpcc-struct-return} mode, but it does
4039 nothing when you use @option{-freg-struct-return} mode.
4043 @subsection Caller-Saves Register Allocation
4045 If you enable it, GCC can save registers around function calls. This
4046 makes it possible to use call-clobbered registers to hold variables that
4047 must live across calls.
4049 @defmac CALLER_SAVE_PROFITABLE (@var{refs}, @var{calls})
4050 A C expression to determine whether it is worthwhile to consider placing
4051 a pseudo-register in a call-clobbered hard register and saving and
4052 restoring it around each function call. The expression should be 1 when
4053 this is worth doing, and 0 otherwise.
4055 If you don't define this macro, a default is used which is good on most
4056 machines: @code{4 * @var{calls} < @var{refs}}.
4059 @defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs})
4060 A C expression specifying which mode is required for saving @var{nregs}
4061 of a pseudo-register in call-clobbered hard register @var{regno}. If
4062 @var{regno} is unsuitable for caller save, @code{VOIDmode} should be
4063 returned. For most machines this macro need not be defined since GCC
4064 will select the smallest suitable mode.
4067 @node Function Entry
4068 @subsection Function Entry and Exit
4069 @cindex function entry and exit
4073 This section describes the macros that output function entry
4074 (@dfn{prologue}) and exit (@dfn{epilogue}) code.
4076 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_PROLOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4077 If defined, a function that outputs the assembler code for entry to a
4078 function. The prologue is responsible for setting up the stack frame,
4079 initializing the frame pointer register, saving registers that must be
4080 saved, and allocating @var{size} additional bytes of storage for the
4081 local variables. @var{size} is an integer. @var{file} is a stdio
4082 stream to which the assembler code should be output.
4084 The label for the beginning of the function need not be output by this
4085 macro. That has already been done when the macro is run.
4087 @findex regs_ever_live
4088 To determine which registers to save, the macro can refer to the array
4089 @code{regs_ever_live}: element @var{r} is nonzero if hard register
4090 @var{r} is used anywhere within the function. This implies the function
4091 prologue should save register @var{r}, provided it is not one of the
4092 call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use
4093 @code{regs_ever_live}.)
4095 On machines that have ``register windows'', the function entry code does
4096 not save on the stack the registers that are in the windows, even if
4097 they are supposed to be preserved by function calls; instead it takes
4098 appropriate steps to ``push'' the register stack, if any non-call-used
4099 registers are used in the function.
4101 @findex frame_pointer_needed
4102 On machines where functions may or may not have frame-pointers, the
4103 function entry code must vary accordingly; it must set up the frame
4104 pointer if one is wanted, and not otherwise. To determine whether a
4105 frame pointer is in wanted, the macro can refer to the variable
4106 @code{frame_pointer_needed}. The variable's value will be 1 at run
4107 time in a function that needs a frame pointer. @xref{Elimination}.
4109 The function entry code is responsible for allocating any stack space
4110 required for the function. This stack space consists of the regions
4111 listed below. In most cases, these regions are allocated in the
4112 order listed, with the last listed region closest to the top of the
4113 stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and
4114 the highest address if it is not defined). You can use a different order
4115 for a machine if doing so is more convenient or required for
4116 compatibility reasons. Except in cases where required by standard
4117 or by a debugger, there is no reason why the stack layout used by GCC
4118 need agree with that used by other compilers for a machine.
4121 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_END_PROLOGUE (FILE *@var{file})
4122 If defined, a function that outputs assembler code at the end of a
4123 prologue. This should be used when the function prologue is being
4124 emitted as RTL, and you have some extra assembler that needs to be
4125 emitted. @xref{prologue instruction pattern}.
4128 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_BEGIN_EPILOGUE (FILE *@var{file})
4129 If defined, a function that outputs assembler code at the start of an
4130 epilogue. This should be used when the function epilogue is being
4131 emitted as RTL, and you have some extra assembler that needs to be
4132 emitted. @xref{epilogue instruction pattern}.
4135 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_EPILOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4136 If defined, a function that outputs the assembler code for exit from a
4137 function. The epilogue is responsible for restoring the saved
4138 registers and stack pointer to their values when the function was
4139 called, and returning control to the caller. This macro takes the
4140 same arguments as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the
4141 registers to restore are determined from @code{regs_ever_live} and
4142 @code{CALL_USED_REGISTERS} in the same way.
4144 On some machines, there is a single instruction that does all the work
4145 of returning from the function. On these machines, give that
4146 instruction the name @samp{return} and do not define the macro
4147 @code{TARGET_ASM_FUNCTION_EPILOGUE} at all.
4149 Do not define a pattern named @samp{return} if you want the
4150 @code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target
4151 switches to control whether return instructions or epilogues are used,
4152 define a @samp{return} pattern with a validity condition that tests the
4153 target switches appropriately. If the @samp{return} pattern's validity
4154 condition is false, epilogues will be used.
4156 On machines where functions may or may not have frame-pointers, the
4157 function exit code must vary accordingly. Sometimes the code for these
4158 two cases is completely different. To determine whether a frame pointer
4159 is wanted, the macro can refer to the variable
4160 @code{frame_pointer_needed}. The variable's value will be 1 when compiling
4161 a function that needs a frame pointer.
4163 Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and
4164 @code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially.
4165 The C variable @code{current_function_is_leaf} is nonzero for such a
4166 function. @xref{Leaf Functions}.
4168 On some machines, some functions pop their arguments on exit while
4169 others leave that for the caller to do. For example, the 68020 when
4170 given @option{-mrtd} pops arguments in functions that take a fixed
4171 number of arguments.
4173 @findex current_function_pops_args
4174 Your definition of the macro @code{RETURN_POPS_ARGS} decides which
4175 functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE}
4176 needs to know what was decided. The variable that is called
4177 @code{current_function_pops_args} is the number of bytes of its
4178 arguments that a function should pop. @xref{Scalar Return}.
4179 @c what is the "its arguments" in the above sentence referring to, pray
4180 @c tell? --mew 5feb93
4185 @findex current_function_pretend_args_size
4186 A region of @code{current_function_pretend_args_size} bytes of
4187 uninitialized space just underneath the first argument arriving on the
4188 stack. (This may not be at the very start of the allocated stack region
4189 if the calling sequence has pushed anything else since pushing the stack
4190 arguments. But usually, on such machines, nothing else has been pushed
4191 yet, because the function prologue itself does all the pushing.) This
4192 region is used on machines where an argument may be passed partly in
4193 registers and partly in memory, and, in some cases to support the
4194 features in @code{<stdarg.h>}.
4197 An area of memory used to save certain registers used by the function.
4198 The size of this area, which may also include space for such things as
4199 the return address and pointers to previous stack frames, is
4200 machine-specific and usually depends on which registers have been used
4201 in the function. Machines with register windows often do not require
4205 A region of at least @var{size} bytes, possibly rounded up to an allocation
4206 boundary, to contain the local variables of the function. On some machines,
4207 this region and the save area may occur in the opposite order, with the
4208 save area closer to the top of the stack.
4211 @cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames
4212 Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of
4213 @code{current_function_outgoing_args_size} bytes to be used for outgoing
4214 argument lists of the function. @xref{Stack Arguments}.
4217 @defmac EXIT_IGNORE_STACK
4218 Define this macro as a C expression that is nonzero if the return
4219 instruction or the function epilogue ignores the value of the stack
4220 pointer; in other words, if it is safe to delete an instruction to
4221 adjust the stack pointer before a return from the function. The
4224 Note that this macro's value is relevant only for functions for which
4225 frame pointers are maintained. It is never safe to delete a final
4226 stack adjustment in a function that has no frame pointer, and the
4227 compiler knows this regardless of @code{EXIT_IGNORE_STACK}.
4230 @defmac EPILOGUE_USES (@var{regno})
4231 Define this macro as a C expression that is nonzero for registers that are
4232 used by the epilogue or the @samp{return} pattern. The stack and frame
4233 pointer registers are already be assumed to be used as needed.
4236 @defmac EH_USES (@var{regno})
4237 Define this macro as a C expression that is nonzero for registers that are
4238 used by the exception handling mechanism, and so should be considered live
4239 on entry to an exception edge.
4242 @defmac DELAY_SLOTS_FOR_EPILOGUE
4243 Define this macro if the function epilogue contains delay slots to which
4244 instructions from the rest of the function can be ``moved''. The
4245 definition should be a C expression whose value is an integer
4246 representing the number of delay slots there.
4249 @defmac ELIGIBLE_FOR_EPILOGUE_DELAY (@var{insn}, @var{n})
4250 A C expression that returns 1 if @var{insn} can be placed in delay
4251 slot number @var{n} of the epilogue.
4253 The argument @var{n} is an integer which identifies the delay slot now
4254 being considered (since different slots may have different rules of
4255 eligibility). It is never negative and is always less than the number
4256 of epilogue delay slots (what @code{DELAY_SLOTS_FOR_EPILOGUE} returns).
4257 If you reject a particular insn for a given delay slot, in principle, it
4258 may be reconsidered for a subsequent delay slot. Also, other insns may
4259 (at least in principle) be considered for the so far unfilled delay
4262 @findex current_function_epilogue_delay_list
4263 @findex final_scan_insn
4264 The insns accepted to fill the epilogue delay slots are put in an RTL
4265 list made with @code{insn_list} objects, stored in the variable
4266 @code{current_function_epilogue_delay_list}. The insn for the first
4267 delay slot comes first in the list. Your definition of the macro
4268 @code{TARGET_ASM_FUNCTION_EPILOGUE} should fill the delay slots by
4269 outputting the insns in this list, usually by calling
4270 @code{final_scan_insn}.
4272 You need not define this macro if you did not define
4273 @code{DELAY_SLOTS_FOR_EPILOGUE}.
4276 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_MI_THUNK (FILE *@var{file}, tree @var{thunk_fndecl}, HOST_WIDE_INT @var{delta}, HOST_WIDE_INT @var{vcall_offset}, tree @var{function})
4277 A function that outputs the assembler code for a thunk
4278 function, used to implement C++ virtual function calls with multiple
4279 inheritance. The thunk acts as a wrapper around a virtual function,
4280 adjusting the implicit object parameter before handing control off to
4283 First, emit code to add the integer @var{delta} to the location that
4284 contains the incoming first argument. Assume that this argument
4285 contains a pointer, and is the one used to pass the @code{this} pointer
4286 in C++. This is the incoming argument @emph{before} the function prologue,
4287 e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of
4288 all other incoming arguments.
4290 Then, if @var{vcall_offset} is nonzero, an additional adjustment should be
4291 made after adding @code{delta}. In particular, if @var{p} is the
4292 adjusted pointer, the following adjustment should be made:
4295 p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)]
4298 After the additions, emit code to jump to @var{function}, which is a
4299 @code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does
4300 not touch the return address. Hence returning from @var{FUNCTION} will
4301 return to whoever called the current @samp{thunk}.
4303 The effect must be as if @var{function} had been called directly with
4304 the adjusted first argument. This macro is responsible for emitting all
4305 of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE}
4306 and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked.
4308 The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function}
4309 have already been extracted from it.) It might possibly be useful on
4310 some targets, but probably not.
4312 If you do not define this macro, the target-independent code in the C++
4313 front end will generate a less efficient heavyweight thunk that calls
4314 @var{function} instead of jumping to it. The generic approach does
4315 not support varargs.
4318 @deftypefn {Target Hook} bool TARGET_ASM_CAN_OUTPUT_MI_THUNK (tree @var{thunk_fndecl}, HOST_WIDE_INT @var{delta}, HOST_WIDE_INT @var{vcall_offset}, tree @var{function})
4319 A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able
4320 to output the assembler code for the thunk function specified by the
4321 arguments it is passed, and false otherwise. In the latter case, the
4322 generic approach will be used by the C++ front end, with the limitations
4327 @subsection Generating Code for Profiling
4328 @cindex profiling, code generation
4330 These macros will help you generate code for profiling.
4332 @defmac FUNCTION_PROFILER (@var{file}, @var{labelno})
4333 A C statement or compound statement to output to @var{file} some
4334 assembler code to call the profiling subroutine @code{mcount}.
4337 The details of how @code{mcount} expects to be called are determined by
4338 your operating system environment, not by GCC@. To figure them out,
4339 compile a small program for profiling using the system's installed C
4340 compiler and look at the assembler code that results.
4342 Older implementations of @code{mcount} expect the address of a counter
4343 variable to be loaded into some register. The name of this variable is
4344 @samp{LP} followed by the number @var{labelno}, so you would generate
4345 the name using @samp{LP%d} in a @code{fprintf}.
4348 @defmac PROFILE_HOOK
4349 A C statement or compound statement to output to @var{file} some assembly
4350 code to call the profiling subroutine @code{mcount} even the target does
4351 not support profiling.
4354 @defmac NO_PROFILE_COUNTERS
4355 Define this macro if the @code{mcount} subroutine on your system does
4356 not need a counter variable allocated for each function. This is true
4357 for almost all modern implementations. If you define this macro, you
4358 must not use the @var{labelno} argument to @code{FUNCTION_PROFILER}.
4361 @defmac PROFILE_BEFORE_PROLOGUE
4362 Define this macro if the code for function profiling should come before
4363 the function prologue. Normally, the profiling code comes after.
4367 @subsection Permitting tail calls
4370 @deftypefn {Target Hook} bool TARGET_FUNCTION_OK_FOR_SIBCALL (tree @var{decl}, tree @var{exp})
4371 True if it is ok to do sibling call optimization for the specified
4372 call expression @var{exp}. @var{decl} will be the called function,
4373 or @code{NULL} if this is an indirect call.
4375 It is not uncommon for limitations of calling conventions to prevent
4376 tail calls to functions outside the current unit of translation, or
4377 during PIC compilation. The hook is used to enforce these restrictions,
4378 as the @code{sibcall} md pattern can not fail, or fall over to a
4379 ``normal'' call. The criteria for successful sibling call optimization
4380 may vary greatly between different architectures.
4383 @node Stack Smashing Protection
4384 @subsection Stack smashing protection
4385 @cindex stack smashing protection
4387 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_GUARD (void)
4388 This hook returns a @code{DECL} node for the external variable to use
4389 for the stack protection guard. This variable is initialized by the
4390 runtime to some random value and is used to initialize the guard value
4391 that is placed at the top of the local stack frame. The type of this
4392 variable must be @code{ptr_type_node}.
4394 The default version of this hook creates a variable called
4395 @samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}.
4398 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_FAIL (void)
4399 This hook returns a tree expression that alerts the runtime that the
4400 stack protect guard variable has been modified. This expression should
4401 involve a call to a @code{noreturn} function.
4403 The default version of this hook invokes a function called
4404 @samp{__stack_chk_fail}, taking no arguments. This function is
4405 normally defined in @file{libgcc2.c}.
4409 @section Implementing the Varargs Macros
4410 @cindex varargs implementation
4412 GCC comes with an implementation of @code{<varargs.h>} and
4413 @code{<stdarg.h>} that work without change on machines that pass arguments
4414 on the stack. Other machines require their own implementations of
4415 varargs, and the two machine independent header files must have
4416 conditionals to include it.
4418 ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in
4419 the calling convention for @code{va_start}. The traditional
4420 implementation takes just one argument, which is the variable in which
4421 to store the argument pointer. The ISO implementation of
4422 @code{va_start} takes an additional second argument. The user is
4423 supposed to write the last named argument of the function here.
4425 However, @code{va_start} should not use this argument. The way to find
4426 the end of the named arguments is with the built-in functions described
4429 @defmac __builtin_saveregs ()
4430 Use this built-in function to save the argument registers in memory so
4431 that the varargs mechanism can access them. Both ISO and traditional
4432 versions of @code{va_start} must use @code{__builtin_saveregs}, unless
4433 you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead.
4435 On some machines, @code{__builtin_saveregs} is open-coded under the
4436 control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On
4437 other machines, it calls a routine written in assembler language,
4438 found in @file{libgcc2.c}.
4440 Code generated for the call to @code{__builtin_saveregs} appears at the
4441 beginning of the function, as opposed to where the call to
4442 @code{__builtin_saveregs} is written, regardless of what the code is.
4443 This is because the registers must be saved before the function starts
4444 to use them for its own purposes.
4445 @c i rewrote the first sentence above to fix an overfull hbox. --mew
4449 @defmac __builtin_args_info (@var{category})
4450 Use this built-in function to find the first anonymous arguments in
4453 In general, a machine may have several categories of registers used for
4454 arguments, each for a particular category of data types. (For example,
4455 on some machines, floating-point registers are used for floating-point
4456 arguments while other arguments are passed in the general registers.)
4457 To make non-varargs functions use the proper calling convention, you
4458 have defined the @code{CUMULATIVE_ARGS} data type to record how many
4459 registers in each category have been used so far
4461 @code{__builtin_args_info} accesses the same data structure of type
4462 @code{CUMULATIVE_ARGS} after the ordinary argument layout is finished
4463 with it, with @var{category} specifying which word to access. Thus, the
4464 value indicates the first unused register in a given category.
4466 Normally, you would use @code{__builtin_args_info} in the implementation
4467 of @code{va_start}, accessing each category just once and storing the
4468 value in the @code{va_list} object. This is because @code{va_list} will
4469 have to update the values, and there is no way to alter the
4470 values accessed by @code{__builtin_args_info}.
4473 @defmac __builtin_next_arg (@var{lastarg})
4474 This is the equivalent of @code{__builtin_args_info}, for stack
4475 arguments. It returns the address of the first anonymous stack
4476 argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it
4477 returns the address of the location above the first anonymous stack
4478 argument. Use it in @code{va_start} to initialize the pointer for
4479 fetching arguments from the stack. Also use it in @code{va_start} to
4480 verify that the second parameter @var{lastarg} is the last named argument
4481 of the current function.
4484 @defmac __builtin_classify_type (@var{object})
4485 Since each machine has its own conventions for which data types are
4486 passed in which kind of register, your implementation of @code{va_arg}
4487 has to embody these conventions. The easiest way to categorize the
4488 specified data type is to use @code{__builtin_classify_type} together
4489 with @code{sizeof} and @code{__alignof__}.
4491 @code{__builtin_classify_type} ignores the value of @var{object},
4492 considering only its data type. It returns an integer describing what
4493 kind of type that is---integer, floating, pointer, structure, and so on.
4495 The file @file{typeclass.h} defines an enumeration that you can use to
4496 interpret the values of @code{__builtin_classify_type}.
4499 These machine description macros help implement varargs:
4501 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN_SAVEREGS (void)
4502 If defined, this hook produces the machine-specific code for a call to
4503 @code{__builtin_saveregs}. This code will be moved to the very
4504 beginning of the function, before any parameter access are made. The
4505 return value of this function should be an RTX that contains the value
4506 to use as the return of @code{__builtin_saveregs}.
4509 @deftypefn {Target Hook} void TARGET_SETUP_INCOMING_VARARGS (CUMULATIVE_ARGS *@var{args_so_far}, enum machine_mode @var{mode}, tree @var{type}, int *@var{pretend_args_size}, int @var{second_time})
4510 This target hook offers an alternative to using
4511 @code{__builtin_saveregs} and defining the hook
4512 @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous
4513 register arguments into the stack so that all the arguments appear to
4514 have been passed consecutively on the stack. Once this is done, you can
4515 use the standard implementation of varargs that works for machines that
4516 pass all their arguments on the stack.
4518 The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data
4519 structure, containing the values that are obtained after processing the
4520 named arguments. The arguments @var{mode} and @var{type} describe the
4521 last named argument---its machine mode and its data type as a tree node.
4523 The target hook should do two things: first, push onto the stack all the
4524 argument registers @emph{not} used for the named arguments, and second,
4525 store the size of the data thus pushed into the @code{int}-valued
4526 variable pointed to by @var{pretend_args_size}. The value that you
4527 store here will serve as additional offset for setting up the stack
4530 Because you must generate code to push the anonymous arguments at
4531 compile time without knowing their data types,
4532 @code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that
4533 have just a single category of argument register and use it uniformly
4536 If the argument @var{second_time} is nonzero, it means that the
4537 arguments of the function are being analyzed for the second time. This
4538 happens for an inline function, which is not actually compiled until the
4539 end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should
4540 not generate any instructions in this case.
4543 @deftypefn {Target Hook} bool TARGET_STRICT_ARGUMENT_NAMING (CUMULATIVE_ARGS *@var{ca})
4544 Define this hook to return @code{true} if the location where a function
4545 argument is passed depends on whether or not it is a named argument.
4547 This hook controls how the @var{named} argument to @code{FUNCTION_ARG}
4548 is set for varargs and stdarg functions. If this hook returns
4549 @code{true}, the @var{named} argument is always true for named
4550 arguments, and false for unnamed arguments. If it returns @code{false},
4551 but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true},
4552 then all arguments are treated as named. Otherwise, all named arguments
4553 except the last are treated as named.
4555 You need not define this hook if it always returns zero.
4558 @deftypefn {Target Hook} bool TARGET_PRETEND_OUTGOING_VARARGS_NAMED
4559 If you need to conditionally change ABIs so that one works with
4560 @code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither
4561 @code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was
4562 defined, then define this hook to return @code{true} if
4563 @code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise.
4564 Otherwise, you should not define this hook.
4568 @section Trampolines for Nested Functions
4569 @cindex trampolines for nested functions
4570 @cindex nested functions, trampolines for
4572 A @dfn{trampoline} is a small piece of code that is created at run time
4573 when the address of a nested function is taken. It normally resides on
4574 the stack, in the stack frame of the containing function. These macros
4575 tell GCC how to generate code to allocate and initialize a
4578 The instructions in the trampoline must do two things: load a constant
4579 address into the static chain register, and jump to the real address of
4580 the nested function. On CISC machines such as the m68k, this requires
4581 two instructions, a move immediate and a jump. Then the two addresses
4582 exist in the trampoline as word-long immediate operands. On RISC
4583 machines, it is often necessary to load each address into a register in
4584 two parts. Then pieces of each address form separate immediate
4587 The code generated to initialize the trampoline must store the variable
4588 parts---the static chain value and the function address---into the
4589 immediate operands of the instructions. On a CISC machine, this is
4590 simply a matter of copying each address to a memory reference at the
4591 proper offset from the start of the trampoline. On a RISC machine, it
4592 may be necessary to take out pieces of the address and store them
4595 @defmac TRAMPOLINE_TEMPLATE (@var{file})
4596 A C statement to output, on the stream @var{file}, assembler code for a
4597 block of data that contains the constant parts of a trampoline. This
4598 code should not include a label---the label is taken care of
4601 If you do not define this macro, it means no template is needed
4602 for the target. Do not define this macro on systems where the block move
4603 code to copy the trampoline into place would be larger than the code
4604 to generate it on the spot.
4607 @defmac TRAMPOLINE_SECTION
4608 The name of a subroutine to switch to the section in which the
4609 trampoline template is to be placed (@pxref{Sections}). The default is
4610 a value of @samp{readonly_data_section}, which places the trampoline in
4611 the section containing read-only data.
4614 @defmac TRAMPOLINE_SIZE
4615 A C expression for the size in bytes of the trampoline, as an integer.
4618 @defmac TRAMPOLINE_ALIGNMENT
4619 Alignment required for trampolines, in bits.
4621 If you don't define this macro, the value of @code{BIGGEST_ALIGNMENT}
4622 is used for aligning trampolines.
4625 @defmac INITIALIZE_TRAMPOLINE (@var{addr}, @var{fnaddr}, @var{static_chain})
4626 A C statement to initialize the variable parts of a trampoline.
4627 @var{addr} is an RTX for the address of the trampoline; @var{fnaddr} is
4628 an RTX for the address of the nested function; @var{static_chain} is an
4629 RTX for the static chain value that should be passed to the function
4633 @defmac TRAMPOLINE_ADJUST_ADDRESS (@var{addr})
4634 A C statement that should perform any machine-specific adjustment in
4635 the address of the trampoline. Its argument contains the address that
4636 was passed to @code{INITIALIZE_TRAMPOLINE}. In case the address to be
4637 used for a function call should be different from the address in which
4638 the template was stored, the different address should be assigned to
4639 @var{addr}. If this macro is not defined, @var{addr} will be used for
4642 @cindex @code{TARGET_ASM_FUNCTION_EPILOGUE} and trampolines
4643 @cindex @code{TARGET_ASM_FUNCTION_PROLOGUE} and trampolines
4644 If this macro is not defined, by default the trampoline is allocated as
4645 a stack slot. This default is right for most machines. The exceptions
4646 are machines where it is impossible to execute instructions in the stack
4647 area. On such machines, you may have to implement a separate stack,
4648 using this macro in conjunction with @code{TARGET_ASM_FUNCTION_PROLOGUE}
4649 and @code{TARGET_ASM_FUNCTION_EPILOGUE}.
4651 @var{fp} points to a data structure, a @code{struct function}, which
4652 describes the compilation status of the immediate containing function of
4653 the function which the trampoline is for. The stack slot for the
4654 trampoline is in the stack frame of this containing function. Other
4655 allocation strategies probably must do something analogous with this
4659 Implementing trampolines is difficult on many machines because they have
4660 separate instruction and data caches. Writing into a stack location
4661 fails to clear the memory in the instruction cache, so when the program
4662 jumps to that location, it executes the old contents.
4664 Here are two possible solutions. One is to clear the relevant parts of
4665 the instruction cache whenever a trampoline is set up. The other is to
4666 make all trampolines identical, by having them jump to a standard
4667 subroutine. The former technique makes trampoline execution faster; the
4668 latter makes initialization faster.
4670 To clear the instruction cache when a trampoline is initialized, define
4671 the following macro.
4673 @defmac CLEAR_INSN_CACHE (@var{beg}, @var{end})
4674 If defined, expands to a C expression clearing the @emph{instruction
4675 cache} in the specified interval. The definition of this macro would
4676 typically be a series of @code{asm} statements. Both @var{beg} and
4677 @var{end} are both pointer expressions.
4680 The operating system may also require the stack to be made executable
4681 before calling the trampoline. To implement this requirement, define
4682 the following macro.
4684 @defmac ENABLE_EXECUTE_STACK
4685 Define this macro if certain operations must be performed before executing
4686 code located on the stack. The macro should expand to a series of C
4687 file-scope constructs (e.g.@: functions) and provide a unique entry point
4688 named @code{__enable_execute_stack}. The target is responsible for
4689 emitting calls to the entry point in the code, for example from the
4690 @code{INITIALIZE_TRAMPOLINE} macro.
4693 To use a standard subroutine, define the following macro. In addition,
4694 you must make sure that the instructions in a trampoline fill an entire
4695 cache line with identical instructions, or else ensure that the
4696 beginning of the trampoline code is always aligned at the same point in
4697 its cache line. Look in @file{m68k.h} as a guide.
4699 @defmac TRANSFER_FROM_TRAMPOLINE
4700 Define this macro if trampolines need a special subroutine to do their
4701 work. The macro should expand to a series of @code{asm} statements
4702 which will be compiled with GCC@. They go in a library function named
4703 @code{__transfer_from_trampoline}.
4705 If you need to avoid executing the ordinary prologue code of a compiled
4706 C function when you jump to the subroutine, you can do so by placing a
4707 special label of your own in the assembler code. Use one @code{asm}
4708 statement to generate an assembler label, and another to make the label
4709 global. Then trampolines can use that label to jump directly to your
4710 special assembler code.
4714 @section Implicit Calls to Library Routines
4715 @cindex library subroutine names
4716 @cindex @file{libgcc.a}
4718 @c prevent bad page break with this line
4719 Here is an explanation of implicit calls to library routines.
4721 @defmac DECLARE_LIBRARY_RENAMES
4722 This macro, if defined, should expand to a piece of C code that will get
4723 expanded when compiling functions for libgcc.a. It can be used to
4724 provide alternate names for GCC's internal library functions if there
4725 are ABI-mandated names that the compiler should provide.
4728 @findex init_one_libfunc
4729 @findex set_optab_libfunc
4730 @deftypefn {Target Hook} void TARGET_INIT_LIBFUNCS (void)
4731 This hook should declare additional library routines or rename
4732 existing ones, using the functions @code{set_optab_libfunc} and
4733 @code{init_one_libfunc} defined in @file{optabs.c}.
4734 @code{init_optabs} calls this macro after initializing all the normal
4737 The default is to do nothing. Most ports don't need to define this hook.
4740 @defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison})
4741 This macro should return @code{true} if the library routine that
4742 implements the floating point comparison operator @var{comparison} in
4743 mode @var{mode} will return a boolean, and @var{false} if it will
4746 GCC's own floating point libraries return tristates from the
4747 comparison operators, so the default returns false always. Most ports
4748 don't need to define this macro.
4751 @defmac TARGET_LIB_INT_CMP_BIASED
4752 This macro should evaluate to @code{true} if the integer comparison
4753 functions (like @code{__cmpdi2}) return 0 to indicate that the first
4754 operand is smaller than the second, 1 to indicate that they are equal,
4755 and 2 to indicate that the first operand is greater than the second.
4756 If this macro evaluates to @code{false} the comparison functions return
4757 @minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines
4758 in @file{libgcc.a}, you do not need to define this macro.
4761 @cindex US Software GOFAST, floating point emulation library
4762 @cindex floating point emulation library, US Software GOFAST
4763 @cindex GOFAST, floating point emulation library
4764 @findex gofast_maybe_init_libfuncs
4765 @defmac US_SOFTWARE_GOFAST
4766 Define this macro if your system C library uses the US Software GOFAST
4767 library to provide floating point emulation.
4769 In addition to defining this macro, your architecture must set
4770 @code{TARGET_INIT_LIBFUNCS} to @code{gofast_maybe_init_libfuncs}, or
4771 else call that function from its version of that hook. It is defined
4772 in @file{config/gofast.h}, which must be included by your
4773 architecture's @file{@var{cpu}.c} file. See @file{sparc/sparc.c} for
4776 If this macro is defined, the
4777 @code{TARGET_FLOAT_LIB_COMPARE_RETURNS_BOOL} target hook must return
4778 false for @code{SFmode} and @code{DFmode} comparisons.
4781 @cindex @code{EDOM}, implicit usage
4784 The value of @code{EDOM} on the target machine, as a C integer constant
4785 expression. If you don't define this macro, GCC does not attempt to
4786 deposit the value of @code{EDOM} into @code{errno} directly. Look in
4787 @file{/usr/include/errno.h} to find the value of @code{EDOM} on your
4790 If you do not define @code{TARGET_EDOM}, then compiled code reports
4791 domain errors by calling the library function and letting it report the
4792 error. If mathematical functions on your system use @code{matherr} when
4793 there is an error, then you should leave @code{TARGET_EDOM} undefined so
4794 that @code{matherr} is used normally.
4797 @cindex @code{errno}, implicit usage
4798 @defmac GEN_ERRNO_RTX
4799 Define this macro as a C expression to create an rtl expression that
4800 refers to the global ``variable'' @code{errno}. (On certain systems,
4801 @code{errno} may not actually be a variable.) If you don't define this
4802 macro, a reasonable default is used.
4805 @cindex C99 math functions, implicit usage
4806 @defmac TARGET_C99_FUNCTIONS
4807 When this macro is nonzero, GCC will implicitly optimize @code{sin} calls into
4808 @code{sinf} and similarly for other functions defined by C99 standard. The
4809 default is nonzero that should be proper value for most modern systems, however
4810 number of existing systems lacks support for these functions in the runtime so
4811 they needs this macro to be redefined to 0.
4814 @defmac NEXT_OBJC_RUNTIME
4815 Define this macro to generate code for Objective-C message sending using
4816 the calling convention of the NeXT system. This calling convention
4817 involves passing the object, the selector and the method arguments all
4818 at once to the method-lookup library function.
4820 The default calling convention passes just the object and the selector
4821 to the lookup function, which returns a pointer to the method.
4824 @node Addressing Modes
4825 @section Addressing Modes
4826 @cindex addressing modes
4828 @c prevent bad page break with this line
4829 This is about addressing modes.
4831 @defmac HAVE_PRE_INCREMENT
4832 @defmacx HAVE_PRE_DECREMENT
4833 @defmacx HAVE_POST_INCREMENT
4834 @defmacx HAVE_POST_DECREMENT
4835 A C expression that is nonzero if the machine supports pre-increment,
4836 pre-decrement, post-increment, or post-decrement addressing respectively.
4839 @defmac HAVE_PRE_MODIFY_DISP
4840 @defmacx HAVE_POST_MODIFY_DISP
4841 A C expression that is nonzero if the machine supports pre- or
4842 post-address side-effect generation involving constants other than
4843 the size of the memory operand.
4846 @defmac HAVE_PRE_MODIFY_REG
4847 @defmacx HAVE_POST_MODIFY_REG
4848 A C expression that is nonzero if the machine supports pre- or
4849 post-address side-effect generation involving a register displacement.
4852 @defmac CONSTANT_ADDRESS_P (@var{x})
4853 A C expression that is 1 if the RTX @var{x} is a constant which
4854 is a valid address. On most machines, this can be defined as
4855 @code{CONSTANT_P (@var{x})}, but a few machines are more restrictive
4856 in which constant addresses are supported.
4859 @defmac CONSTANT_P (@var{x})
4860 @code{CONSTANT_P}, which is defined by target-independent code,
4861 accepts integer-values expressions whose values are not explicitly
4862 known, such as @code{symbol_ref}, @code{label_ref}, and @code{high}
4863 expressions and @code{const} arithmetic expressions, in addition to
4864 @code{const_int} and @code{const_double} expressions.
4867 @defmac MAX_REGS_PER_ADDRESS
4868 A number, the maximum number of registers that can appear in a valid
4869 memory address. Note that it is up to you to specify a value equal to
4870 the maximum number that @code{GO_IF_LEGITIMATE_ADDRESS} would ever
4874 @defmac GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
4875 A C compound statement with a conditional @code{goto @var{label};}
4876 executed if @var{x} (an RTX) is a legitimate memory address on the
4877 target machine for a memory operand of mode @var{mode}.
4879 It usually pays to define several simpler macros to serve as
4880 subroutines for this one. Otherwise it may be too complicated to
4883 This macro must exist in two variants: a strict variant and a
4884 non-strict one. The strict variant is used in the reload pass. It
4885 must be defined so that any pseudo-register that has not been
4886 allocated a hard register is considered a memory reference. In
4887 contexts where some kind of register is required, a pseudo-register
4888 with no hard register must be rejected.
4890 The non-strict variant is used in other passes. It must be defined to
4891 accept all pseudo-registers in every context where some kind of
4892 register is required.
4894 @findex REG_OK_STRICT
4895 Compiler source files that want to use the strict variant of this
4896 macro define the macro @code{REG_OK_STRICT}. You should use an
4897 @code{#ifdef REG_OK_STRICT} conditional to define the strict variant
4898 in that case and the non-strict variant otherwise.
4900 Subroutines to check for acceptable registers for various purposes (one
4901 for base registers, one for index registers, and so on) are typically
4902 among the subroutines used to define @code{GO_IF_LEGITIMATE_ADDRESS}.
4903 Then only these subroutine macros need have two variants; the higher
4904 levels of macros may be the same whether strict or not.
4906 Normally, constant addresses which are the sum of a @code{symbol_ref}
4907 and an integer are stored inside a @code{const} RTX to mark them as
4908 constant. Therefore, there is no need to recognize such sums
4909 specifically as legitimate addresses. Normally you would simply
4910 recognize any @code{const} as legitimate.
4912 Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
4913 sums that are not marked with @code{const}. It assumes that a naked
4914 @code{plus} indicates indexing. If so, then you @emph{must} reject such
4915 naked constant sums as illegitimate addresses, so that none of them will
4916 be given to @code{PRINT_OPERAND_ADDRESS}.
4918 @cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation
4919 On some machines, whether a symbolic address is legitimate depends on
4920 the section that the address refers to. On these machines, define the
4921 target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information
4922 into the @code{symbol_ref}, and then check for it here. When you see a
4923 @code{const}, you will have to look inside it to find the
4924 @code{symbol_ref} in order to determine the section. @xref{Assembler
4928 @defmac REG_OK_FOR_BASE_P (@var{x})
4929 A C expression that is nonzero if @var{x} (assumed to be a @code{reg}
4930 RTX) is valid for use as a base register. For hard registers, it
4931 should always accept those which the hardware permits and reject the
4932 others. Whether the macro accepts or rejects pseudo registers must be
4933 controlled by @code{REG_OK_STRICT} as described above. This usually
4934 requires two variant definitions, of which @code{REG_OK_STRICT}
4935 controls the one actually used.
4938 @defmac REG_MODE_OK_FOR_BASE_P (@var{x}, @var{mode})
4939 A C expression that is just like @code{REG_OK_FOR_BASE_P}, except that
4940 that expression may examine the mode of the memory reference in
4941 @var{mode}. You should define this macro if the mode of the memory
4942 reference affects whether a register may be used as a base register. If
4943 you define this macro, the compiler will use it instead of
4944 @code{REG_OK_FOR_BASE_P}.
4947 @defmac REG_MODE_OK_FOR_REG_BASE_P (@var{x}, @var{mode})
4948 A C expression which is nonzero if @var{x} (assumed to be a @code{reg} RTX)
4949 is suitable for use as a base register in base plus index operand addresses,
4950 accessing memory in mode @var{mode}. It may be either a suitable hard
4951 register or a pseudo register that has been allocated such a hard register.
4952 You should define this macro if base plus index addresses have different
4953 requirements than other base register uses.
4956 @defmac REG_OK_FOR_INDEX_P (@var{x})
4957 A C expression that is nonzero if @var{x} (assumed to be a @code{reg}
4958 RTX) is valid for use as an index register.
4960 The difference between an index register and a base register is that
4961 the index register may be scaled. If an address involves the sum of
4962 two registers, neither one of them scaled, then either one may be
4963 labeled the ``base'' and the other the ``index''; but whichever
4964 labeling is used must fit the machine's constraints of which registers
4965 may serve in each capacity. The compiler will try both labelings,
4966 looking for one that is valid, and will reload one or both registers
4967 only if neither labeling works.
4970 @defmac FIND_BASE_TERM (@var{x})
4971 A C expression to determine the base term of address @var{x}.
4972 This macro is used in only one place: `find_base_term' in alias.c.
4974 It is always safe for this macro to not be defined. It exists so
4975 that alias analysis can understand machine-dependent addresses.
4977 The typical use of this macro is to handle addresses containing
4978 a label_ref or symbol_ref within an UNSPEC@.
4981 @defmac LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win})
4982 A C compound statement that attempts to replace @var{x} with a valid
4983 memory address for an operand of mode @var{mode}. @var{win} will be a
4984 C statement label elsewhere in the code; the macro definition may use
4987 GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win});
4991 to avoid further processing if the address has become legitimate.
4993 @findex break_out_memory_refs
4994 @var{x} will always be the result of a call to @code{break_out_memory_refs},
4995 and @var{oldx} will be the operand that was given to that function to produce
4998 The code generated by this macro should not alter the substructure of
4999 @var{x}. If it transforms @var{x} into a more legitimate form, it
5000 should assign @var{x} (which will always be a C variable) a new value.
5002 It is not necessary for this macro to come up with a legitimate
5003 address. The compiler has standard ways of doing so in all cases. In
5004 fact, it is safe to omit this macro. But often a
5005 machine-dependent strategy can generate better code.
5008 @defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win})
5009 A C compound statement that attempts to replace @var{x}, which is an address
5010 that needs reloading, with a valid memory address for an operand of mode
5011 @var{mode}. @var{win} will be a C statement label elsewhere in the code.
5012 It is not necessary to define this macro, but it might be useful for
5013 performance reasons.
5015 For example, on the i386, it is sometimes possible to use a single
5016 reload register instead of two by reloading a sum of two pseudo
5017 registers into a register. On the other hand, for number of RISC
5018 processors offsets are limited so that often an intermediate address
5019 needs to be generated in order to address a stack slot. By defining
5020 @code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses
5021 generated for adjacent some stack slots can be made identical, and thus
5024 @emph{Note}: This macro should be used with caution. It is necessary
5025 to know something of how reload works in order to effectively use this,
5026 and it is quite easy to produce macros that build in too much knowledge
5027 of reload internals.
5029 @emph{Note}: This macro must be able to reload an address created by a
5030 previous invocation of this macro. If it fails to handle such addresses
5031 then the compiler may generate incorrect code or abort.
5034 The macro definition should use @code{push_reload} to indicate parts that
5035 need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually
5036 suitable to be passed unaltered to @code{push_reload}.
5038 The code generated by this macro must not alter the substructure of
5039 @var{x}. If it transforms @var{x} into a more legitimate form, it
5040 should assign @var{x} (which will always be a C variable) a new value.
5041 This also applies to parts that you change indirectly by calling
5044 @findex strict_memory_address_p
5045 The macro definition may use @code{strict_memory_address_p} to test if
5046 the address has become legitimate.
5049 If you want to change only a part of @var{x}, one standard way of doing
5050 this is to use @code{copy_rtx}. Note, however, that is unshares only a
5051 single level of rtl. Thus, if the part to be changed is not at the
5052 top level, you'll need to replace first the top level.
5053 It is not necessary for this macro to come up with a legitimate
5054 address; but often a machine-dependent strategy can generate better code.
5057 @defmac GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
5058 A C statement or compound statement with a conditional @code{goto
5059 @var{label};} executed if memory address @var{x} (an RTX) can have
5060 different meanings depending on the machine mode of the memory
5061 reference it is used for or if the address is valid for some modes
5064 Autoincrement and autodecrement addresses typically have mode-dependent
5065 effects because the amount of the increment or decrement is the size
5066 of the operand being addressed. Some machines have other mode-dependent
5067 addresses. Many RISC machines have no mode-dependent addresses.
5069 You may assume that @var{addr} is a valid address for the machine.
5072 @defmac LEGITIMATE_CONSTANT_P (@var{x})
5073 A C expression that is nonzero if @var{x} is a legitimate constant for
5074 an immediate operand on the target machine. You can assume that
5075 @var{x} satisfies @code{CONSTANT_P}, so you need not check this. In fact,
5076 @samp{1} is a suitable definition for this macro on machines where
5077 anything @code{CONSTANT_P} is valid.
5080 @deftypefn {Target Hook} rtx TARGET_DELEGITIMIZE_ADDRESS (rtx @var{x})
5081 This hook is used to undo the possibly obfuscating effects of the
5082 @code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target
5083 macros. Some backend implementations of these macros wrap symbol
5084 references inside an @code{UNSPEC} rtx to represent PIC or similar
5085 addressing modes. This target hook allows GCC's optimizers to understand
5086 the semantics of these opaque @code{UNSPEC}s by converting them back
5087 into their original form.
5090 @deftypefn {Target Hook} bool TARGET_CANNOT_FORCE_CONST_MEM (rtx @var{x})
5091 This hook should return true if @var{x} is of a form that cannot (or
5092 should not) be spilled to the constant pool. The default version of
5093 this hook returns false.
5095 The primary reason to define this hook is to prevent reload from
5096 deciding that a non-legitimate constant would be better reloaded
5097 from the constant pool instead of spilling and reloading a register
5098 holding the constant. This restriction is often true of addresses
5099 of TLS symbols for various targets.
5102 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD (void)
5103 This hook should return the DECL of a function @var{f} that given an
5104 address @var{addr} as an argument returns a mask @var{m} that can be
5105 used to extract from two vectors the relevant data that resides in
5106 @var{addr} in case @var{addr} is not properly aligned.
5108 The autovectrizer, when vectorizing a load operation from an address
5109 @var{addr} that may be unaligned, will generate two vector loads from
5110 the two aligned addresses around @var{addr}. It then generates a
5111 @code{REALIGN_LOAD} operation to extract the relevant data from the
5112 two loaded vectors. The first two arguments to @code{REALIGN_LOAD},
5113 @var{v1} and @var{v2}, are the two vectors, each of size @var{VS}, and
5114 the third argument, @var{OFF}, defines how the data will be extracted
5115 from these two vectors: if @var{OFF} is 0, then the returned vector is
5116 @var{v2}; otherwise, the returned vector is composed from the last
5117 @var{VS}-@var{OFF} elements of @var{v1} concatenated to the first
5118 @var{OFF} elements of @var{v2}.
5120 If this hook is defined, the autovectorizer will generate a call
5121 to @var{f} (using the DECL tree that this hook returns) and will
5122 use the return value of @var{f} as the argument @var{OFF} to
5123 @code{REALIGN_LOAD}. Therefore, the mask @var{m} returned by @var{f}
5124 should comply with the semantics expected by @code{REALIGN_LOAD}
5126 If this hook is not defined, then @var{addr} will be used as
5127 the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low
5128 log2(@var{VS})-1 bits of @var{addr} will be considered.
5131 @node Condition Code
5132 @section Condition Code Status
5133 @cindex condition code status
5135 @c prevent bad page break with this line
5136 This describes the condition code status.
5139 The file @file{conditions.h} defines a variable @code{cc_status} to
5140 describe how the condition code was computed (in case the interpretation of
5141 the condition code depends on the instruction that it was set by). This
5142 variable contains the RTL expressions on which the condition code is
5143 currently based, and several standard flags.
5145 Sometimes additional machine-specific flags must be defined in the machine
5146 description header file. It can also add additional machine-specific
5147 information by defining @code{CC_STATUS_MDEP}.
5149 @defmac CC_STATUS_MDEP
5150 C code for a data type which is used for declaring the @code{mdep}
5151 component of @code{cc_status}. It defaults to @code{int}.
5153 This macro is not used on machines that do not use @code{cc0}.
5156 @defmac CC_STATUS_MDEP_INIT
5157 A C expression to initialize the @code{mdep} field to ``empty''.
5158 The default definition does nothing, since most machines don't use
5159 the field anyway. If you want to use the field, you should probably
5160 define this macro to initialize it.
5162 This macro is not used on machines that do not use @code{cc0}.
5165 @defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn})
5166 A C compound statement to set the components of @code{cc_status}
5167 appropriately for an insn @var{insn} whose body is @var{exp}. It is
5168 this macro's responsibility to recognize insns that set the condition
5169 code as a byproduct of other activity as well as those that explicitly
5172 This macro is not used on machines that do not use @code{cc0}.
5174 If there are insns that do not set the condition code but do alter
5175 other machine registers, this macro must check to see whether they
5176 invalidate the expressions that the condition code is recorded as
5177 reflecting. For example, on the 68000, insns that store in address
5178 registers do not set the condition code, which means that usually
5179 @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
5180 insns. But suppose that the previous insn set the condition code
5181 based on location @samp{a4@@(102)} and the current insn stores a new
5182 value in @samp{a4}. Although the condition code is not changed by
5183 this, it will no longer be true that it reflects the contents of
5184 @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
5185 @code{cc_status} in this case to say that nothing is known about the
5186 condition code value.
5188 The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
5189 with the results of peephole optimization: insns whose patterns are
5190 @code{parallel} RTXs containing various @code{reg}, @code{mem} or
5191 constants which are just the operands. The RTL structure of these
5192 insns is not sufficient to indicate what the insns actually do. What
5193 @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
5194 @code{CC_STATUS_INIT}.
5196 A possible definition of @code{NOTICE_UPDATE_CC} is to call a function
5197 that looks at an attribute (@pxref{Insn Attributes}) named, for example,
5198 @samp{cc}. This avoids having detailed information about patterns in
5199 two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}.
5202 @defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y})
5203 Returns a mode from class @code{MODE_CC} to be used when comparison
5204 operation code @var{op} is applied to rtx @var{x} and @var{y}. For
5205 example, on the SPARC, @code{SELECT_CC_MODE} is defined as (see
5206 @pxref{Jump Patterns} for a description of the reason for this
5210 #define SELECT_CC_MODE(OP,X,Y) \
5211 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \
5212 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \
5213 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \
5214 || GET_CODE (X) == NEG) \
5215 ? CC_NOOVmode : CCmode))
5218 You should define this macro if and only if you define extra CC modes
5219 in @file{@var{machine}-modes.def}.
5222 @defmac CANONICALIZE_COMPARISON (@var{code}, @var{op0}, @var{op1})
5223 On some machines not all possible comparisons are defined, but you can
5224 convert an invalid comparison into a valid one. For example, the Alpha
5225 does not have a @code{GT} comparison, but you can use an @code{LT}
5226 comparison instead and swap the order of the operands.
5228 On such machines, define this macro to be a C statement to do any
5229 required conversions. @var{code} is the initial comparison code
5230 and @var{op0} and @var{op1} are the left and right operands of the
5231 comparison, respectively. You should modify @var{code}, @var{op0}, and
5232 @var{op1} as required.
5234 GCC will not assume that the comparison resulting from this macro is
5235 valid but will see if the resulting insn matches a pattern in the
5238 You need not define this macro if it would never change the comparison
5242 @defmac REVERSIBLE_CC_MODE (@var{mode})
5243 A C expression whose value is one if it is always safe to reverse a
5244 comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE}
5245 can ever return @var{mode} for a floating-point inequality comparison,
5246 then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero.
5248 You need not define this macro if it would always returns zero or if the
5249 floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}.
5250 For example, here is the definition used on the SPARC, where floating-point
5251 inequality comparisons are always given @code{CCFPEmode}:
5254 #define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode)
5258 @defmac REVERSE_CONDITION (@var{code}, @var{mode})
5259 A C expression whose value is reversed condition code of the @var{code} for
5260 comparison done in CC_MODE @var{mode}. The macro is used only in case
5261 @code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case
5262 machine has some non-standard way how to reverse certain conditionals. For
5263 instance in case all floating point conditions are non-trapping, compiler may
5264 freely convert unordered compares to ordered one. Then definition may look
5268 #define REVERSE_CONDITION(CODE, MODE) \
5269 ((MODE) != CCFPmode ? reverse_condition (CODE) \
5270 : reverse_condition_maybe_unordered (CODE))
5274 @defmac REVERSE_CONDEXEC_PREDICATES_P (@var{op1}, @var{op2})
5275 A C expression that returns true if the conditional execution predicate
5276 @var{op1}, a comparison operation, is the inverse of @var{op2} and vice
5277 versa. Define this to return 0 if the target has conditional execution
5278 predicates that cannot be reversed safely. There is no need to validate
5279 that the arguments of op1 and op2 are the same, this is done separately.
5280 If no expansion is specified, this macro is defined as follows:
5283 #define REVERSE_CONDEXEC_PREDICATES_P (x, y) \
5284 (GET_CODE ((x)) == reversed_comparison_code ((y), NULL))
5288 @deftypefn {Target Hook} bool TARGET_FIXED_CONDITION_CODE_REGS (unsigned int *, unsigned int *)
5289 On targets which do not use @code{(cc0)}, and which use a hard
5290 register rather than a pseudo-register to hold condition codes, the
5291 regular CSE passes are often not able to identify cases in which the
5292 hard register is set to a common value. Use this hook to enable a
5293 small pass which optimizes such cases. This hook should return true
5294 to enable this pass, and it should set the integers to which its
5295 arguments point to the hard register numbers used for condition codes.
5296 When there is only one such register, as is true on most systems, the
5297 integer pointed to by the second argument should be set to
5298 @code{INVALID_REGNUM}.
5300 The default version of this hook returns false.
5303 @deftypefn {Target Hook} enum machine_mode TARGET_CC_MODES_COMPATIBLE (enum machine_mode, enum machine_mode)
5304 On targets which use multiple condition code modes in class
5305 @code{MODE_CC}, it is sometimes the case that a comparison can be
5306 validly done in more than one mode. On such a system, define this
5307 target hook to take two mode arguments and to return a mode in which
5308 both comparisons may be validly done. If there is no such mode,
5309 return @code{VOIDmode}.
5311 The default version of this hook checks whether the modes are the
5312 same. If they are, it returns that mode. If they are different, it
5313 returns @code{VOIDmode}.
5317 @section Describing Relative Costs of Operations
5318 @cindex costs of instructions
5319 @cindex relative costs
5320 @cindex speed of instructions
5322 These macros let you describe the relative speed of various operations
5323 on the target machine.
5325 @defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to})
5326 A C expression for the cost of moving data of mode @var{mode} from a
5327 register in class @var{from} to one in class @var{to}. The classes are
5328 expressed using the enumeration values such as @code{GENERAL_REGS}. A
5329 value of 2 is the default; other values are interpreted relative to
5332 It is not required that the cost always equal 2 when @var{from} is the
5333 same as @var{to}; on some machines it is expensive to move between
5334 registers if they are not general registers.
5336 If reload sees an insn consisting of a single @code{set} between two
5337 hard registers, and if @code{REGISTER_MOVE_COST} applied to their
5338 classes returns a value of 2, reload does not check to ensure that the
5339 constraints of the insn are met. Setting a cost of other than 2 will
5340 allow reload to verify that the constraints are met. You should do this
5341 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
5344 @defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in})
5345 A C expression for the cost of moving data of mode @var{mode} between a
5346 register of class @var{class} and memory; @var{in} is zero if the value
5347 is to be written to memory, nonzero if it is to be read in. This cost
5348 is relative to those in @code{REGISTER_MOVE_COST}. If moving between
5349 registers and memory is more expensive than between two registers, you
5350 should define this macro to express the relative cost.
5352 If you do not define this macro, GCC uses a default cost of 4 plus
5353 the cost of copying via a secondary reload register, if one is
5354 needed. If your machine requires a secondary reload register to copy
5355 between memory and a register of @var{class} but the reload mechanism is
5356 more complex than copying via an intermediate, define this macro to
5357 reflect the actual cost of the move.
5359 GCC defines the function @code{memory_move_secondary_cost} if
5360 secondary reloads are needed. It computes the costs due to copying via
5361 a secondary register. If your machine copies from memory using a
5362 secondary register in the conventional way but the default base value of
5363 4 is not correct for your machine, define this macro to add some other
5364 value to the result of that function. The arguments to that function
5365 are the same as to this macro.
5369 A C expression for the cost of a branch instruction. A value of 1 is
5370 the default; other values are interpreted relative to that.
5373 Here are additional macros which do not specify precise relative costs,
5374 but only that certain actions are more expensive than GCC would
5377 @defmac SLOW_BYTE_ACCESS
5378 Define this macro as a C expression which is nonzero if accessing less
5379 than a word of memory (i.e.@: a @code{char} or a @code{short}) is no
5380 faster than accessing a word of memory, i.e., if such access
5381 require more than one instruction or if there is no difference in cost
5382 between byte and (aligned) word loads.
5384 When this macro is not defined, the compiler will access a field by
5385 finding the smallest containing object; when it is defined, a fullword
5386 load will be used if alignment permits. Unless bytes accesses are
5387 faster than word accesses, using word accesses is preferable since it
5388 may eliminate subsequent memory access if subsequent accesses occur to
5389 other fields in the same word of the structure, but to different bytes.
5392 @defmac SLOW_UNALIGNED_ACCESS (@var{mode}, @var{alignment})
5393 Define this macro to be the value 1 if memory accesses described by the
5394 @var{mode} and @var{alignment} parameters have a cost many times greater
5395 than aligned accesses, for example if they are emulated in a trap
5398 When this macro is nonzero, the compiler will act as if
5399 @code{STRICT_ALIGNMENT} were nonzero when generating code for block
5400 moves. This can cause significantly more instructions to be produced.
5401 Therefore, do not set this macro nonzero if unaligned accesses only add a
5402 cycle or two to the time for a memory access.
5404 If the value of this macro is always zero, it need not be defined. If
5405 this macro is defined, it should produce a nonzero value when
5406 @code{STRICT_ALIGNMENT} is nonzero.
5410 The threshold of number of scalar memory-to-memory move insns, @emph{below}
5411 which a sequence of insns should be generated instead of a
5412 string move insn or a library call. Increasing the value will always
5413 make code faster, but eventually incurs high cost in increased code size.
5415 Note that on machines where the corresponding move insn is a
5416 @code{define_expand} that emits a sequence of insns, this macro counts
5417 the number of such sequences.
5419 If you don't define this, a reasonable default is used.
5422 @defmac MOVE_BY_PIECES_P (@var{size}, @var{alignment})
5423 A C expression used to determine whether @code{move_by_pieces} will be used to
5424 copy a chunk of memory, or whether some other block move mechanism
5425 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
5426 than @code{MOVE_RATIO}.
5429 @defmac MOVE_MAX_PIECES
5430 A C expression used by @code{move_by_pieces} to determine the largest unit
5431 a load or store used to copy memory is. Defaults to @code{MOVE_MAX}.
5435 The threshold of number of scalar move insns, @emph{below} which a sequence
5436 of insns should be generated to clear memory instead of a string clear insn
5437 or a library call. Increasing the value will always make code faster, but
5438 eventually incurs high cost in increased code size.
5440 If you don't define this, a reasonable default is used.
5443 @defmac CLEAR_BY_PIECES_P (@var{size}, @var{alignment})
5444 A C expression used to determine whether @code{clear_by_pieces} will be used
5445 to clear a chunk of memory, or whether some other block clear mechanism
5446 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
5447 than @code{CLEAR_RATIO}.
5450 @defmac STORE_BY_PIECES_P (@var{size}, @var{alignment})
5451 A C expression used to determine whether @code{store_by_pieces} will be
5452 used to set a chunk of memory to a constant value, or whether some other
5453 mechanism will be used. Used by @code{__builtin_memset} when storing
5454 values other than constant zero and by @code{__builtin_strcpy} when
5455 when called with a constant source string.
5456 Defaults to 1 if @code{move_by_pieces_ninsns} returns less
5457 than @code{MOVE_RATIO}.
5460 @defmac USE_LOAD_POST_INCREMENT (@var{mode})
5461 A C expression used to determine whether a load postincrement is a good
5462 thing to use for a given mode. Defaults to the value of
5463 @code{HAVE_POST_INCREMENT}.
5466 @defmac USE_LOAD_POST_DECREMENT (@var{mode})
5467 A C expression used to determine whether a load postdecrement is a good
5468 thing to use for a given mode. Defaults to the value of
5469 @code{HAVE_POST_DECREMENT}.
5472 @defmac USE_LOAD_PRE_INCREMENT (@var{mode})
5473 A C expression used to determine whether a load preincrement is a good
5474 thing to use for a given mode. Defaults to the value of
5475 @code{HAVE_PRE_INCREMENT}.
5478 @defmac USE_LOAD_PRE_DECREMENT (@var{mode})
5479 A C expression used to determine whether a load predecrement is a good
5480 thing to use for a given mode. Defaults to the value of
5481 @code{HAVE_PRE_DECREMENT}.
5484 @defmac USE_STORE_POST_INCREMENT (@var{mode})
5485 A C expression used to determine whether a store postincrement is a good
5486 thing to use for a given mode. Defaults to the value of
5487 @code{HAVE_POST_INCREMENT}.
5490 @defmac USE_STORE_POST_DECREMENT (@var{mode})
5491 A C expression used to determine whether a store postdecrement is a good
5492 thing to use for a given mode. Defaults to the value of
5493 @code{HAVE_POST_DECREMENT}.
5496 @defmac USE_STORE_PRE_INCREMENT (@var{mode})
5497 This macro is used to determine whether a store preincrement is a good
5498 thing to use for a given mode. Defaults to the value of
5499 @code{HAVE_PRE_INCREMENT}.
5502 @defmac USE_STORE_PRE_DECREMENT (@var{mode})
5503 This macro is used to determine whether a store predecrement is a good
5504 thing to use for a given mode. Defaults to the value of
5505 @code{HAVE_PRE_DECREMENT}.
5508 @defmac NO_FUNCTION_CSE
5509 Define this macro if it is as good or better to call a constant
5510 function address than to call an address kept in a register.
5513 @defmac RANGE_TEST_NON_SHORT_CIRCUIT
5514 Define this macro if a non-short-circuit operation produced by
5515 @samp{fold_range_test ()} is optimal. This macro defaults to true if
5516 @code{BRANCH_COST} is greater than or equal to the value 2.
5519 @deftypefn {Target Hook} bool TARGET_RTX_COSTS (rtx @var{x}, int @var{code}, int @var{outer_code}, int *@var{total})
5520 This target hook describes the relative costs of RTL expressions.
5522 The cost may depend on the precise form of the expression, which is
5523 available for examination in @var{x}, and the rtx code of the expression
5524 in which it is contained, found in @var{outer_code}. @var{code} is the
5525 expression code---redundant, since it can be obtained with
5526 @code{GET_CODE (@var{x})}.
5528 In implementing this hook, you can use the construct
5529 @code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast
5532 On entry to the hook, @code{*@var{total}} contains a default estimate
5533 for the cost of the expression. The hook should modify this value as
5534 necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)}
5535 for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus
5536 operations, and @code{COSTS_N_INSNS (1)} for all other operations.
5538 When optimizing for code size, i.e.@: when @code{optimize_size} is
5539 nonzero, this target hook should be used to estimate the relative
5540 size cost of an expression, again relative to @code{COSTS_N_INSNS}.
5542 The hook returns true when all subexpressions of @var{x} have been
5543 processed, and false when @code{rtx_cost} should recurse.
5546 @deftypefn {Target Hook} int TARGET_ADDRESS_COST (rtx @var{address})
5547 This hook computes the cost of an addressing mode that contains
5548 @var{address}. If not defined, the cost is computed from
5549 the @var{address} expression and the @code{TARGET_RTX_COST} hook.
5551 For most CISC machines, the default cost is a good approximation of the
5552 true cost of the addressing mode. However, on RISC machines, all
5553 instructions normally have the same length and execution time. Hence
5554 all addresses will have equal costs.
5556 In cases where more than one form of an address is known, the form with
5557 the lowest cost will be used. If multiple forms have the same, lowest,
5558 cost, the one that is the most complex will be used.
5560 For example, suppose an address that is equal to the sum of a register
5561 and a constant is used twice in the same basic block. When this macro
5562 is not defined, the address will be computed in a register and memory
5563 references will be indirect through that register. On machines where
5564 the cost of the addressing mode containing the sum is no higher than
5565 that of a simple indirect reference, this will produce an additional
5566 instruction and possibly require an additional register. Proper
5567 specification of this macro eliminates this overhead for such machines.
5569 This hook is never called with an invalid address.
5571 On machines where an address involving more than one register is as
5572 cheap as an address computation involving only one register, defining
5573 @code{TARGET_ADDRESS_COST} to reflect this can cause two registers to
5574 be live over a region of code where only one would have been if
5575 @code{TARGET_ADDRESS_COST} were not defined in that manner. This effect
5576 should be considered in the definition of this macro. Equivalent costs
5577 should probably only be given to addresses with different numbers of
5578 registers on machines with lots of registers.
5582 @section Adjusting the Instruction Scheduler
5584 The instruction scheduler may need a fair amount of machine-specific
5585 adjustment in order to produce good code. GCC provides several target
5586 hooks for this purpose. It is usually enough to define just a few of
5587 them: try the first ones in this list first.
5589 @deftypefn {Target Hook} int TARGET_SCHED_ISSUE_RATE (void)
5590 This hook returns the maximum number of instructions that can ever
5591 issue at the same time on the target machine. The default is one.
5592 Although the insn scheduler can define itself the possibility of issue
5593 an insn on the same cycle, the value can serve as an additional
5594 constraint to issue insns on the same simulated processor cycle (see
5595 hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}).
5596 This value must be constant over the entire compilation. If you need
5597 it to vary depending on what the instructions are, you must use
5598 @samp{TARGET_SCHED_VARIABLE_ISSUE}.
5600 You could define this hook to return the value of the macro
5601 @code{MAX_DFA_ISSUE_RATE}.
5604 @deftypefn {Target Hook} int TARGET_SCHED_VARIABLE_ISSUE (FILE *@var{file}, int @var{verbose}, rtx @var{insn}, int @var{more})
5605 This hook is executed by the scheduler after it has scheduled an insn
5606 from the ready list. It should return the number of insns which can
5607 still be issued in the current cycle. The default is
5608 @samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and
5609 @code{USE}, which normally are not counted against the issue rate.
5610 You should define this hook if some insns take more machine resources
5611 than others, so that fewer insns can follow them in the same cycle.
5612 @var{file} is either a null pointer, or a stdio stream to write any
5613 debug output to. @var{verbose} is the verbose level provided by
5614 @option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that
5618 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_COST (rtx @var{insn}, rtx @var{link}, rtx @var{dep_insn}, int @var{cost})
5619 This function corrects the value of @var{cost} based on the
5620 relationship between @var{insn} and @var{dep_insn} through the
5621 dependence @var{link}. It should return the new value. The default
5622 is to make no adjustment to @var{cost}. This can be used for example
5623 to specify to the scheduler using the traditional pipeline description
5624 that an output- or anti-dependence does not incur the same cost as a
5625 data-dependence. If the scheduler using the automaton based pipeline
5626 description, the cost of anti-dependence is zero and the cost of
5627 output-dependence is maximum of one and the difference of latency
5628 times of the first and the second insns. If these values are not
5629 acceptable, you could use the hook to modify them too. See also
5630 @pxref{Processor pipeline description}.
5633 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_PRIORITY (rtx @var{insn}, int @var{priority})
5634 This hook adjusts the integer scheduling priority @var{priority} of
5635 @var{insn}. It should return the new priority. Reduce the priority to
5636 execute @var{insn} earlier, increase the priority to execute @var{insn}
5637 later. Do not define this hook if you do not need to adjust the
5638 scheduling priorities of insns.
5641 @deftypefn {Target Hook} int TARGET_SCHED_REORDER (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_readyp}, int @var{clock})
5642 This hook is executed by the scheduler after it has scheduled the ready
5643 list, to allow the machine description to reorder it (for example to
5644 combine two small instructions together on @samp{VLIW} machines).
5645 @var{file} is either a null pointer, or a stdio stream to write any
5646 debug output to. @var{verbose} is the verbose level provided by
5647 @option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready
5648 list of instructions that are ready to be scheduled. @var{n_readyp} is
5649 a pointer to the number of elements in the ready list. The scheduler
5650 reads the ready list in reverse order, starting with
5651 @var{ready}[@var{*n_readyp}-1] and going to @var{ready}[0]. @var{clock}
5652 is the timer tick of the scheduler. You may modify the ready list and
5653 the number of ready insns. The return value is the number of insns that
5654 can issue this cycle; normally this is just @code{issue_rate}. See also
5655 @samp{TARGET_SCHED_REORDER2}.
5658 @deftypefn {Target Hook} int TARGET_SCHED_REORDER2 (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_ready}, @var{clock})
5659 Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That
5660 function is called whenever the scheduler starts a new cycle. This one
5661 is called once per iteration over a cycle, immediately after
5662 @samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and
5663 return the number of insns to be scheduled in the same cycle. Defining
5664 this hook can be useful if there are frequent situations where
5665 scheduling one insn causes other insns to become ready in the same
5666 cycle. These other insns can then be taken into account properly.
5669 @deftypefn {Target Hook} void TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK (rtx @var{head}, rtx @var{tail})
5670 This hook is called after evaluation forward dependencies of insns in
5671 chain given by two parameter values (@var{head} and @var{tail}
5672 correspondingly) but before insns scheduling of the insn chain. For
5673 example, it can be used for better insn classification if it requires
5674 analysis of dependencies. This hook can use backward and forward
5675 dependencies of the insn scheduler because they are already
5679 @deftypefn {Target Hook} void TARGET_SCHED_INIT (FILE *@var{file}, int @var{verbose}, int @var{max_ready})
5680 This hook is executed by the scheduler at the beginning of each block of
5681 instructions that are to be scheduled. @var{file} is either a null
5682 pointer, or a stdio stream to write any debug output to. @var{verbose}
5683 is the verbose level provided by @option{-fsched-verbose-@var{n}}.
5684 @var{max_ready} is the maximum number of insns in the current scheduling
5685 region that can be live at the same time. This can be used to allocate
5686 scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}.
5689 @deftypefn {Target Hook} void TARGET_SCHED_FINISH (FILE *@var{file}, int @var{verbose})
5690 This hook is executed by the scheduler at the end of each block of
5691 instructions that are to be scheduled. It can be used to perform
5692 cleanup of any actions done by the other scheduling hooks. @var{file}
5693 is either a null pointer, or a stdio stream to write any debug output
5694 to. @var{verbose} is the verbose level provided by
5695 @option{-fsched-verbose-@var{n}}.
5698 @deftypefn {Target Hook} void TARGET_SCHED_INIT_GLOBAL (FILE *@var{file}, int @var{verbose}, int @var{old_max_uid})
5699 This hook is executed by the scheduler after function level initializations.
5700 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
5701 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
5702 @var{old_max_uid} is the maximum insn uid when scheduling begins.
5705 @deftypefn {Target Hook} void TARGET_SCHED_FINISH_GLOBAL (FILE *@var{file}, int @var{verbose})
5706 This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}.
5707 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
5708 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
5711 @deftypefn {Target Hook} int TARGET_SCHED_DFA_PRE_CYCLE_INSN (void)
5712 The hook returns an RTL insn. The automaton state used in the
5713 pipeline hazard recognizer is changed as if the insn were scheduled
5714 when the new simulated processor cycle starts. Usage of the hook may
5715 simplify the automaton pipeline description for some @acronym{VLIW}
5716 processors. If the hook is defined, it is used only for the automaton
5717 based pipeline description. The default is not to change the state
5718 when the new simulated processor cycle starts.
5721 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN (void)
5722 The hook can be used to initialize data used by the previous hook.
5725 @deftypefn {Target Hook} int TARGET_SCHED_DFA_POST_CYCLE_INSN (void)
5726 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
5727 to changed the state as if the insn were scheduled when the new
5728 simulated processor cycle finishes.
5731 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN (void)
5732 The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but
5733 used to initialize data used by the previous hook.
5736 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD (void)
5737 This hook controls better choosing an insn from the ready insn queue
5738 for the @acronym{DFA}-based insn scheduler. Usually the scheduler
5739 chooses the first insn from the queue. If the hook returns a positive
5740 value, an additional scheduler code tries all permutations of
5741 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()}
5742 subsequent ready insns to choose an insn whose issue will result in
5743 maximal number of issued insns on the same cycle. For the
5744 @acronym{VLIW} processor, the code could actually solve the problem of
5745 packing simple insns into the @acronym{VLIW} insn. Of course, if the
5746 rules of @acronym{VLIW} packing are described in the automaton.
5748 This code also could be used for superscalar @acronym{RISC}
5749 processors. Let us consider a superscalar @acronym{RISC} processor
5750 with 3 pipelines. Some insns can be executed in pipelines @var{A} or
5751 @var{B}, some insns can be executed only in pipelines @var{B} or
5752 @var{C}, and one insn can be executed in pipeline @var{B}. The
5753 processor may issue the 1st insn into @var{A} and the 2nd one into
5754 @var{B}. In this case, the 3rd insn will wait for freeing @var{B}
5755 until the next cycle. If the scheduler issues the 3rd insn the first,
5756 the processor could issue all 3 insns per cycle.
5758 Actually this code demonstrates advantages of the automaton based
5759 pipeline hazard recognizer. We try quickly and easy many insn
5760 schedules to choose the best one.
5762 The default is no multipass scheduling.
5765 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD (rtx)
5767 This hook controls what insns from the ready insn queue will be
5768 considered for the multipass insn scheduling. If the hook returns
5769 zero for insn passed as the parameter, the insn will be not chosen to
5772 The default is that any ready insns can be chosen to be issued.
5775 @deftypefn {Target Hook} int TARGET_SCHED_DFA_NEW_CYCLE (FILE *, int, rtx, int, int, int *)
5777 This hook is called by the insn scheduler before issuing insn passed
5778 as the third parameter on given cycle. If the hook returns nonzero,
5779 the insn is not issued on given processors cycle. Instead of that,
5780 the processor cycle is advanced. If the value passed through the last
5781 parameter is zero, the insn ready queue is not sorted on the new cycle
5782 start as usually. The first parameter passes file for debugging
5783 output. The second one passes the scheduler verbose level of the
5784 debugging output. The forth and the fifth parameter values are
5785 correspondingly processor cycle on which the previous insn has been
5786 issued and the current processor cycle.
5789 @deftypefn {Target Hook} bool TARGET_SCHED_IS_COSTLY_DEPENDENCE (rtx @var{insn1}, rtx @var{insn2}, rtx @var{dep_link}, int @var{dep_cost}, int @var{distance})
5790 This hook is used to define which dependences are considered costly by
5791 the target, so costly that it is not advisable to schedule the insns that
5792 are involved in the dependence too close to one another. The parameters
5793 to this hook are as follows: The second parameter @var{insn2} is dependent
5794 upon the first parameter @var{insn1}. The dependence between @var{insn1}
5795 and @var{insn2} is represented by the third parameter @var{dep_link}. The
5796 fourth parameter @var{cost} is the cost of the dependence, and the fifth
5797 parameter @var{distance} is the distance in cycles between the two insns.
5798 The hook returns @code{true} if considering the distance between the two
5799 insns the dependence between them is considered costly by the target,
5800 and @code{false} otherwise.
5802 Defining this hook can be useful in multiple-issue out-of-order machines,
5803 where (a) it's practically hopeless to predict the actual data/resource
5804 delays, however: (b) there's a better chance to predict the actual grouping
5805 that will be formed, and (c) correctly emulating the grouping can be very
5806 important. In such targets one may want to allow issuing dependent insns
5807 closer to one another---i.e., closer than the dependence distance; however,
5808 not in cases of "costly dependences", which this hooks allows to define.
5811 Macros in the following table are generated by the program
5812 @file{genattr} and can be useful for writing the hooks.
5814 @defmac MAX_DFA_ISSUE_RATE
5815 The macro definition is generated in the automaton based pipeline
5816 description interface. Its value is calculated from the automaton
5817 based pipeline description and is equal to maximal number of all insns
5818 described in constructions @samp{define_insn_reservation} which can be
5819 issued on the same processor cycle.
5823 @section Dividing the Output into Sections (Texts, Data, @dots{})
5824 @c the above section title is WAY too long. maybe cut the part between
5825 @c the (...)? --mew 10feb93
5827 An object file is divided into sections containing different types of
5828 data. In the most common case, there are three sections: the @dfn{text
5829 section}, which holds instructions and read-only data; the @dfn{data
5830 section}, which holds initialized writable data; and the @dfn{bss
5831 section}, which holds uninitialized data. Some systems have other kinds
5834 The compiler must tell the assembler when to switch sections. These
5835 macros control what commands to output to tell the assembler this. You
5836 can also define additional sections.
5838 @defmac TEXT_SECTION_ASM_OP
5839 A C expression whose value is a string, including spacing, containing the
5840 assembler operation that should precede instructions and read-only data.
5841 Normally @code{"\t.text"} is right.
5844 @defmac HOT_TEXT_SECTION_NAME
5845 If defined, a C string constant for the name of the section containing most
5846 frequently executed functions of the program. If not defined, GCC will provide
5847 a default definition if the target supports named sections.
5850 @defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME
5851 If defined, a C string constant for the name of the section containing unlikely
5852 executed functions in the program.
5855 @defmac DATA_SECTION_ASM_OP
5856 A C expression whose value is a string, including spacing, containing the
5857 assembler operation to identify the following data as writable initialized
5858 data. Normally @code{"\t.data"} is right.
5861 @defmac READONLY_DATA_SECTION_ASM_OP
5862 A C expression whose value is a string, including spacing, containing the
5863 assembler operation to identify the following data as read-only initialized
5867 @defmac READONLY_DATA_SECTION
5868 A macro naming a function to call to switch to the proper section for
5869 read-only data. The default is to use @code{READONLY_DATA_SECTION_ASM_OP}
5870 if defined, else fall back to @code{text_section}.
5872 The most common definition will be @code{data_section}, if the target
5873 does not have a special read-only data section, and does not put data
5874 in the text section.
5877 @defmac BSS_SECTION_ASM_OP
5878 If defined, a C expression whose value is a string, including spacing,
5879 containing the assembler operation to identify the following data as
5880 uninitialized global data. If not defined, and neither
5881 @code{ASM_OUTPUT_BSS} nor @code{ASM_OUTPUT_ALIGNED_BSS} are defined,
5882 uninitialized global data will be output in the data section if
5883 @option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be
5887 @defmac INIT_SECTION_ASM_OP
5888 If defined, a C expression whose value is a string, including spacing,
5889 containing the assembler operation to identify the following data as
5890 initialization code. If not defined, GCC will assume such a section does
5894 @defmac FINI_SECTION_ASM_OP
5895 If defined, a C expression whose value is a string, including spacing,
5896 containing the assembler operation to identify the following data as
5897 finalization code. If not defined, GCC will assume such a section does
5901 @defmac INIT_ARRAY_SECTION_ASM_OP
5902 If defined, a C expression whose value is a string, including spacing,
5903 containing the assembler operation to identify the following data as
5904 part of the @code{.init_array} (or equivalent) section. If not
5905 defined, GCC will assume such a section does not exist. Do not define
5906 both this macro and @code{INIT_SECTION_ASM_OP}.
5909 @defmac FINI_ARRAY_SECTION_ASM_OP
5910 If defined, a C expression whose value is a string, including spacing,
5911 containing the assembler operation to identify the following data as
5912 part of the @code{.fini_array} (or equivalent) section. If not
5913 defined, GCC will assume such a section does not exist. Do not define
5914 both this macro and @code{FINI_SECTION_ASM_OP}.
5917 @defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function})
5918 If defined, an ASM statement that switches to a different section
5919 via @var{section_op}, calls @var{function}, and switches back to
5920 the text section. This is used in @file{crtstuff.c} if
5921 @code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls
5922 to initialization and finalization functions from the init and fini
5923 sections. By default, this macro uses a simple function call. Some
5924 ports need hand-crafted assembly code to avoid dependencies on
5925 registers initialized in the function prologue or to ensure that
5926 constant pools don't end up too far way in the text section.
5929 @defmac FORCE_CODE_SECTION_ALIGN
5930 If defined, an ASM statement that aligns a code section to some
5931 arbitrary boundary. This is used to force all fragments of the
5932 @code{.init} and @code{.fini} sections to have to same alignment
5933 and thus prevent the linker from having to add any padding.
5938 @defmac EXTRA_SECTIONS
5939 A list of names for sections other than the standard two, which are
5940 @code{in_text} and @code{in_data}. You need not define this macro
5941 on a system with no other sections (that GCC needs to use).
5944 @findex text_section
5945 @findex data_section
5946 @defmac EXTRA_SECTION_FUNCTIONS
5947 One or more functions to be defined in @file{varasm.c}. These
5948 functions should do jobs analogous to those of @code{text_section} and
5949 @code{data_section}, for your additional sections. Do not define this
5950 macro if you do not define @code{EXTRA_SECTIONS}.
5953 @defmac JUMP_TABLES_IN_TEXT_SECTION
5954 Define this macro to be an expression with a nonzero value if jump
5955 tables (for @code{tablejump} insns) should be output in the text
5956 section, along with the assembler instructions. Otherwise, the
5957 readonly data section is used.
5959 This macro is irrelevant if there is no separate readonly data section.
5962 @deftypefn {Target Hook} void TARGET_ASM_SELECT_SECTION (tree @var{exp}, int @var{reloc}, unsigned HOST_WIDE_INT @var{align})
5963 Switches to the appropriate section for output of @var{exp}. You can
5964 assume that @var{exp} is either a @code{VAR_DECL} node or a constant of
5965 some sort. @var{reloc} indicates whether the initial value of @var{exp}
5966 requires link-time relocations. Bit 0 is set when variable contains
5967 local relocations only, while bit 1 is set for global relocations.
5968 Select the section by calling @code{data_section} or one of the
5969 alternatives for other sections. @var{align} is the constant alignment
5972 The default version of this function takes care of putting read-only
5973 variables in @code{readonly_data_section}.
5975 See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}.
5978 @defmac USE_SELECT_SECTION_FOR_FUNCTIONS
5979 Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called
5980 for @code{FUNCTION_DECL}s as well as for variables and constants.
5982 In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the
5983 function has been determined to be likely to be called, and nonzero if
5984 it is unlikely to be called.
5987 @deftypefn {Target Hook} void TARGET_ASM_UNIQUE_SECTION (tree @var{decl}, int @var{reloc})
5988 Build up a unique section name, expressed as a @code{STRING_CST} node,
5989 and assign it to @samp{DECL_SECTION_NAME (@var{decl})}.
5990 As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether
5991 the initial value of @var{exp} requires link-time relocations.
5993 The default version of this function appends the symbol name to the
5994 ELF section name that would normally be used for the symbol. For
5995 example, the function @code{foo} would be placed in @code{.text.foo}.
5996 Whatever the actual target object format, this is often good enough.
5999 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_RODATA_SECTION (tree @var{decl})
6000 Switches to a readonly data section associated with
6001 @samp{DECL_SECTION_NAME (@var{decl})}.
6002 The default version of this function switches to @code{.gnu.linkonce.r.name}
6003 section if function's section is @code{.gnu.linkonce.t.name}, to
6004 @code{.rodata.name} if function is in @code{.text.name} section
6005 and otherwise switches to the normal readonly data section.
6008 @deftypefn {Target Hook} void TARGET_ASM_SELECT_RTX_SECTION (enum machine_mode @var{mode}, rtx @var{x}, unsigned HOST_WIDE_INT @var{align})
6009 Switches to the appropriate section for output of constant pool entry
6010 @var{x} in @var{mode}. You can assume that @var{x} is some kind of
6011 constant in RTL@. The argument @var{mode} is redundant except in the
6012 case of a @code{const_int} rtx. Select the section by calling
6013 @code{readonly_data_section} or one of the alternatives for other
6014 sections. @var{align} is the constant alignment in bits.
6016 The default version of this function takes care of putting symbolic
6017 constants in @code{flag_pic} mode in @code{data_section} and everything
6018 else in @code{readonly_data_section}.
6021 @deftypefn {Target Hook} void TARGET_ENCODE_SECTION_INFO (tree @var{decl}, rtx @var{rtl}, int @var{new_decl_p})
6022 Define this hook if references to a symbol or a constant must be
6023 treated differently depending on something about the variable or
6024 function named by the symbol (such as what section it is in).
6026 The hook is executed immediately after rtl has been created for
6027 @var{decl}, which may be a variable or function declaration or
6028 an entry in the constant pool. In either case, @var{rtl} is the
6029 rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})}
6030 in this hook; that field may not have been initialized yet.
6032 In the case of a constant, it is safe to assume that the rtl is
6033 a @code{mem} whose address is a @code{symbol_ref}. Most decls
6034 will also have this form, but that is not guaranteed. Global
6035 register variables, for instance, will have a @code{reg} for their
6036 rtl. (Normally the right thing to do with such unusual rtl is
6039 The @var{new_decl_p} argument will be true if this is the first time
6040 that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will
6041 be false for subsequent invocations, which will happen for duplicate
6042 declarations. Whether or not anything must be done for the duplicate
6043 declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}.
6044 @var{new_decl_p} is always true when the hook is called for a constant.
6046 @cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO}
6047 The usual thing for this hook to do is to record flags in the
6048 @code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}.
6049 Historically, the name string was modified if it was necessary to
6050 encode more than one bit of information, but this practice is now
6051 discouraged; use @code{SYMBOL_REF_FLAGS}.
6053 The default definition of this hook, @code{default_encode_section_info}
6054 in @file{varasm.c}, sets a number of commonly-useful bits in
6055 @code{SYMBOL_REF_FLAGS}. Check whether the default does what you need
6056 before overriding it.
6059 @deftypefn {Target Hook} const char *TARGET_STRIP_NAME_ENCODING (const char *name)
6060 Decode @var{name} and return the real name part, sans
6061 the characters that @code{TARGET_ENCODE_SECTION_INFO}
6065 @deftypefn {Target Hook} bool TARGET_IN_SMALL_DATA_P (tree @var{exp})
6066 Returns true if @var{exp} should be placed into a ``small data'' section.
6067 The default version of this hook always returns false.
6070 @deftypevar {Target Hook} bool TARGET_HAVE_SRODATA_SECTION
6071 Contains the value true if the target places read-only
6072 ``small data'' into a separate section. The default value is false.
6075 @deftypefn {Target Hook} bool TARGET_BINDS_LOCAL_P (tree @var{exp})
6076 Returns true if @var{exp} names an object for which name resolution
6077 rules must resolve to the current ``module'' (dynamic shared library
6078 or executable image).
6080 The default version of this hook implements the name resolution rules
6081 for ELF, which has a looser model of global name binding than other
6082 currently supported object file formats.
6085 @deftypevar {Target Hook} bool TARGET_HAVE_TLS
6086 Contains the value true if the target supports thread-local storage.
6087 The default value is false.
6092 @section Position Independent Code
6093 @cindex position independent code
6096 This section describes macros that help implement generation of position
6097 independent code. Simply defining these macros is not enough to
6098 generate valid PIC; you must also add support to the macros
6099 @code{GO_IF_LEGITIMATE_ADDRESS} and @code{PRINT_OPERAND_ADDRESS}, as
6100 well as @code{LEGITIMIZE_ADDRESS}. You must modify the definition of
6101 @samp{movsi} to do something appropriate when the source operand
6102 contains a symbolic address. You may also need to alter the handling of
6103 switch statements so that they use relative addresses.
6104 @c i rearranged the order of the macros above to try to force one of
6105 @c them to the next line, to eliminate an overfull hbox. --mew 10feb93
6107 @defmac PIC_OFFSET_TABLE_REGNUM
6108 The register number of the register used to address a table of static
6109 data addresses in memory. In some cases this register is defined by a
6110 processor's ``application binary interface'' (ABI)@. When this macro
6111 is defined, RTL is generated for this register once, as with the stack
6112 pointer and frame pointer registers. If this macro is not defined, it
6113 is up to the machine-dependent files to allocate such a register (if
6114 necessary). Note that this register must be fixed when in use (e.g.@:
6115 when @code{flag_pic} is true).
6118 @defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED
6119 Define this macro if the register defined by
6120 @code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. Do not define
6121 this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined.
6124 @defmac LEGITIMATE_PIC_OPERAND_P (@var{x})
6125 A C expression that is nonzero if @var{x} is a legitimate immediate
6126 operand on the target machine when generating position independent code.
6127 You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not
6128 check this. You can also assume @var{flag_pic} is true, so you need not
6129 check it either. You need not define this macro if all constants
6130 (including @code{SYMBOL_REF}) can be immediate operands when generating
6131 position independent code.
6134 @node Assembler Format
6135 @section Defining the Output Assembler Language
6137 This section describes macros whose principal purpose is to describe how
6138 to write instructions in assembler language---rather than what the
6142 * File Framework:: Structural information for the assembler file.
6143 * Data Output:: Output of constants (numbers, strings, addresses).
6144 * Uninitialized Data:: Output of uninitialized variables.
6145 * Label Output:: Output and generation of labels.
6146 * Initialization:: General principles of initialization
6147 and termination routines.
6148 * Macros for Initialization::
6149 Specific macros that control the handling of
6150 initialization and termination routines.
6151 * Instruction Output:: Output of actual instructions.
6152 * Dispatch Tables:: Output of jump tables.
6153 * Exception Region Output:: Output of exception region code.
6154 * Alignment Output:: Pseudo ops for alignment and skipping data.
6157 @node File Framework
6158 @subsection The Overall Framework of an Assembler File
6159 @cindex assembler format
6160 @cindex output of assembler code
6162 @c prevent bad page break with this line
6163 This describes the overall framework of an assembly file.
6165 @deftypefn {Target Hook} void TARGET_ASM_FILE_START ()
6166 @findex default_file_start
6167 Output to @code{asm_out_file} any text which the assembler expects to
6168 find at the beginning of a file. The default behavior is controlled
6169 by two flags, documented below. Unless your target's assembler is
6170 quite unusual, if you override the default, you should call
6171 @code{default_file_start} at some point in your target hook. This
6172 lets other target files rely on these variables.
6175 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_APP_OFF
6176 If this flag is true, the text of the macro @code{ASM_APP_OFF} will be
6177 printed as the very first line in the assembly file, unless
6178 @option{-fverbose-asm} is in effect. (If that macro has been defined
6179 to the empty string, this variable has no effect.) With the normal
6180 definition of @code{ASM_APP_OFF}, the effect is to notify the GNU
6181 assembler that it need not bother stripping comments or extra
6182 whitespace from its input. This allows it to work a bit faster.
6184 The default is false. You should not set it to true unless you have
6185 verified that your port does not generate any extra whitespace or
6186 comments that will cause GAS to issue errors in NO_APP mode.
6189 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_FILE_DIRECTIVE
6190 If this flag is true, @code{output_file_directive} will be called
6191 for the primary source file, immediately after printing
6192 @code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect
6193 this to be done. The default is false.
6196 @deftypefn {Target Hook} void TARGET_ASM_FILE_END ()
6197 Output to @code{asm_out_file} any text which the assembler expects
6198 to find at the end of a file. The default is to output nothing.
6201 @deftypefun void file_end_indicate_exec_stack ()
6202 Some systems use a common convention, the @samp{.note.GNU-stack}
6203 special section, to indicate whether or not an object file relies on
6204 the stack being executable. If your system uses this convention, you
6205 should define @code{TARGET_ASM_FILE_END} to this function. If you
6206 need to do other things in that hook, have your hook function call
6210 @defmac ASM_COMMENT_START
6211 A C string constant describing how to begin a comment in the target
6212 assembler language. The compiler assumes that the comment will end at
6213 the end of the line.
6217 A C string constant for text to be output before each @code{asm}
6218 statement or group of consecutive ones. Normally this is
6219 @code{"#APP"}, which is a comment that has no effect on most
6220 assemblers but tells the GNU assembler that it must check the lines
6221 that follow for all valid assembler constructs.
6225 A C string constant for text to be output after each @code{asm}
6226 statement or group of consecutive ones. Normally this is
6227 @code{"#NO_APP"}, which tells the GNU assembler to resume making the
6228 time-saving assumptions that are valid for ordinary compiler output.
6231 @defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
6232 A C statement to output COFF information or DWARF debugging information
6233 which indicates that filename @var{name} is the current source file to
6234 the stdio stream @var{stream}.
6236 This macro need not be defined if the standard form of output
6237 for the file format in use is appropriate.
6240 @defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string})
6241 A C statement to output the string @var{string} to the stdio stream
6242 @var{stream}. If you do not call the function @code{output_quoted_string}
6243 in your config files, GCC will only call it to output filenames to
6244 the assembler source. So you can use it to canonicalize the format
6245 of the filename using this macro.
6248 @defmac ASM_OUTPUT_IDENT (@var{stream}, @var{string})
6249 A C statement to output something to the assembler file to handle a
6250 @samp{#ident} directive containing the text @var{string}. If this
6251 macro is not defined, nothing is output for a @samp{#ident} directive.
6254 @deftypefn {Target Hook} void TARGET_ASM_NAMED_SECTION (const char *@var{name}, unsigned int @var{flags}, unsigned int @var{align})
6255 Output assembly directives to switch to section @var{name}. The section
6256 should have attributes as specified by @var{flags}, which is a bit mask
6257 of the @code{SECTION_*} flags defined in @file{output.h}. If @var{align}
6258 is nonzero, it contains an alignment in bytes to be used for the section,
6259 otherwise some target default should be used. Only targets that must
6260 specify an alignment within the section directive need pay attention to
6261 @var{align} -- we will still use @code{ASM_OUTPUT_ALIGN}.
6264 @deftypefn {Target Hook} bool TARGET_HAVE_NAMED_SECTIONS
6265 This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}.
6268 @deftypefn {Target Hook} {unsigned int} TARGET_SECTION_TYPE_FLAGS (tree @var{decl}, const char *@var{name}, int @var{reloc})
6269 Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION}
6270 based on a variable or function decl, a section name, and whether or not the
6271 declaration's initializer may contain runtime relocations. @var{decl} may be
6272 null, in which case read-write data should be assumed.
6274 The default version if this function handles choosing code vs data,
6275 read-only vs read-write data, and @code{flag_pic}. You should only
6276 need to override this if your target has special flags that might be
6277 set via @code{__attribute__}.
6282 @subsection Output of Data
6285 @deftypevr {Target Hook} {const char *} TARGET_ASM_BYTE_OP
6286 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP
6287 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP
6288 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP
6289 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP
6290 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP
6291 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP
6292 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP
6293 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP
6294 These hooks specify assembly directives for creating certain kinds
6295 of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a
6296 byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an
6297 aligned two-byte object, and so on. Any of the hooks may be
6298 @code{NULL}, indicating that no suitable directive is available.
6300 The compiler will print these strings at the start of a new line,
6301 followed immediately by the object's initial value. In most cases,
6302 the string should contain a tab, a pseudo-op, and then another tab.
6305 @deftypefn {Target Hook} bool TARGET_ASM_INTEGER (rtx @var{x}, unsigned int @var{size}, int @var{aligned_p})
6306 The @code{assemble_integer} function uses this hook to output an
6307 integer object. @var{x} is the object's value, @var{size} is its size
6308 in bytes and @var{aligned_p} indicates whether it is aligned. The
6309 function should return @code{true} if it was able to output the
6310 object. If it returns false, @code{assemble_integer} will try to
6311 split the object into smaller parts.
6313 The default implementation of this hook will use the
6314 @code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false}
6315 when the relevant string is @code{NULL}.
6318 @defmac OUTPUT_ADDR_CONST_EXTRA (@var{stream}, @var{x}, @var{fail})
6319 A C statement to recognize @var{rtx} patterns that
6320 @code{output_addr_const} can't deal with, and output assembly code to
6321 @var{stream} corresponding to the pattern @var{x}. This may be used to
6322 allow machine-dependent @code{UNSPEC}s to appear within constants.
6324 If @code{OUTPUT_ADDR_CONST_EXTRA} fails to recognize a pattern, it must
6325 @code{goto fail}, so that a standard error message is printed. If it
6326 prints an error message itself, by calling, for example,
6327 @code{output_operand_lossage}, it may just complete normally.
6330 @defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
6331 A C statement to output to the stdio stream @var{stream} an assembler
6332 instruction to assemble a string constant containing the @var{len}
6333 bytes at @var{ptr}. @var{ptr} will be a C expression of type
6334 @code{char *} and @var{len} a C expression of type @code{int}.
6336 If the assembler has a @code{.ascii} pseudo-op as found in the
6337 Berkeley Unix assembler, do not define the macro
6338 @code{ASM_OUTPUT_ASCII}.
6341 @defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n})
6342 A C statement to output word @var{n} of a function descriptor for
6343 @var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS}
6344 is defined, and is otherwise unused.
6347 @defmac CONSTANT_POOL_BEFORE_FUNCTION
6348 You may define this macro as a C expression. You should define the
6349 expression to have a nonzero value if GCC should output the constant
6350 pool for a function before the code for the function, or a zero value if
6351 GCC should output the constant pool after the function. If you do
6352 not define this macro, the usual case, GCC will output the constant
6353 pool before the function.
6356 @defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size})
6357 A C statement to output assembler commands to define the start of the
6358 constant pool for a function. @var{funname} is a string giving
6359 the name of the function. Should the return type of the function
6360 be required, it can be obtained via @var{fundecl}. @var{size}
6361 is the size, in bytes, of the constant pool that will be written
6362 immediately after this call.
6364 If no constant-pool prefix is required, the usual case, this macro need
6368 @defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto})
6369 A C statement (with or without semicolon) to output a constant in the
6370 constant pool, if it needs special treatment. (This macro need not do
6371 anything for RTL expressions that can be output normally.)
6373 The argument @var{file} is the standard I/O stream to output the
6374 assembler code on. @var{x} is the RTL expression for the constant to
6375 output, and @var{mode} is the machine mode (in case @var{x} is a
6376 @samp{const_int}). @var{align} is the required alignment for the value
6377 @var{x}; you should output an assembler directive to force this much
6380 The argument @var{labelno} is a number to use in an internal label for
6381 the address of this pool entry. The definition of this macro is
6382 responsible for outputting the label definition at the proper place.
6383 Here is how to do this:
6386 @code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno});
6389 When you output a pool entry specially, you should end with a
6390 @code{goto} to the label @var{jumpto}. This will prevent the same pool
6391 entry from being output a second time in the usual manner.
6393 You need not define this macro if it would do nothing.
6396 @defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size})
6397 A C statement to output assembler commands to at the end of the constant
6398 pool for a function. @var{funname} is a string giving the name of the
6399 function. Should the return type of the function be required, you can
6400 obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the
6401 constant pool that GCC wrote immediately before this call.
6403 If no constant-pool epilogue is required, the usual case, you need not
6407 @defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C})
6408 Define this macro as a C expression which is nonzero if @var{C} is
6409 used as a logical line separator by the assembler.
6411 If you do not define this macro, the default is that only
6412 the character @samp{;} is treated as a logical line separator.
6415 @deftypevr {Target Hook} {const char *} TARGET_ASM_OPEN_PAREN
6416 @deftypevrx {Target Hook} {const char *} TARGET_ASM_CLOSE_PAREN
6417 These target hooks are C string constants, describing the syntax in the
6418 assembler for grouping arithmetic expressions. If not overridden, they
6419 default to normal parentheses, which is correct for most assemblers.
6422 These macros are provided by @file{real.h} for writing the definitions
6423 of @code{ASM_OUTPUT_DOUBLE} and the like:
6425 @defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l})
6426 @defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l})
6427 @defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l})
6428 These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the target's
6429 floating point representation, and store its bit pattern in the variable
6430 @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE}, this variable should
6431 be a simple @code{long int}. For the others, it should be an array of
6432 @code{long int}. The number of elements in this array is determined by
6433 the size of the desired target floating point data type: 32 bits of it
6434 go in each @code{long int} array element. Each array element holds 32
6435 bits of the result, even if @code{long int} is wider than 32 bits on the
6438 The array element values are designed so that you can print them out
6439 using @code{fprintf} in the order they should appear in the target
6443 @node Uninitialized Data
6444 @subsection Output of Uninitialized Variables
6446 Each of the macros in this section is used to do the whole job of
6447 outputting a single uninitialized variable.
6449 @defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
6450 A C statement (sans semicolon) to output to the stdio stream
6451 @var{stream} the assembler definition of a common-label named
6452 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
6453 is the size rounded up to whatever alignment the caller wants.
6455 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
6456 output the name itself; before and after that, output the additional
6457 assembler syntax for defining the name, and a newline.
6459 This macro controls how the assembler definitions of uninitialized
6460 common global variables are output.
6463 @defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment})
6464 Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a
6465 separate, explicit argument. If you define this macro, it is used in
6466 place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in
6467 handling the required alignment of the variable. The alignment is specified
6468 as the number of bits.
6471 @defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
6472 Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the
6473 variable to be output, if there is one, or @code{NULL_TREE} if there
6474 is no corresponding variable. If you define this macro, GCC will use it
6475 in place of both @code{ASM_OUTPUT_COMMON} and
6476 @code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see
6477 the variable's decl in order to chose what to output.
6480 @defmac ASM_OUTPUT_SHARED_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
6481 If defined, it is similar to @code{ASM_OUTPUT_COMMON}, except that it
6482 is used when @var{name} is shared. If not defined, @code{ASM_OUTPUT_COMMON}
6486 @defmac ASM_OUTPUT_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded})
6487 A C statement (sans semicolon) to output to the stdio stream
6488 @var{stream} the assembler definition of uninitialized global @var{decl} named
6489 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
6490 is the size rounded up to whatever alignment the caller wants.
6492 Try to use function @code{asm_output_bss} defined in @file{varasm.c} when
6493 defining this macro. If unable, use the expression
6494 @code{assemble_name (@var{stream}, @var{name})} to output the name itself;
6495 before and after that, output the additional assembler syntax for defining
6496 the name, and a newline.
6498 This macro controls how the assembler definitions of uninitialized global
6499 variables are output. This macro exists to properly support languages like
6500 C++ which do not have @code{common} data. However, this macro currently
6501 is not defined for all targets. If this macro and
6502 @code{ASM_OUTPUT_ALIGNED_BSS} are not defined then @code{ASM_OUTPUT_COMMON}
6503 or @code{ASM_OUTPUT_ALIGNED_COMMON} or
6504 @code{ASM_OUTPUT_ALIGNED_DECL_COMMON} is used.
6507 @defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
6508 Like @code{ASM_OUTPUT_BSS} except takes the required alignment as a
6509 separate, explicit argument. If you define this macro, it is used in
6510 place of @code{ASM_OUTPUT_BSS}, and gives you more flexibility in
6511 handling the required alignment of the variable. The alignment is specified
6512 as the number of bits.
6514 Try to use function @code{asm_output_aligned_bss} defined in file
6515 @file{varasm.c} when defining this macro.
6518 @defmac ASM_OUTPUT_SHARED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded})
6519 If defined, it is similar to @code{ASM_OUTPUT_BSS}, except that it
6520 is used when @var{name} is shared. If not defined, @code{ASM_OUTPUT_BSS}
6524 @defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
6525 A C statement (sans semicolon) to output to the stdio stream
6526 @var{stream} the assembler definition of a local-common-label named
6527 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
6528 is the size rounded up to whatever alignment the caller wants.
6530 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
6531 output the name itself; before and after that, output the additional
6532 assembler syntax for defining the name, and a newline.
6534 This macro controls how the assembler definitions of uninitialized
6535 static variables are output.
6538 @defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment})
6539 Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a
6540 separate, explicit argument. If you define this macro, it is used in
6541 place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in
6542 handling the required alignment of the variable. The alignment is specified
6543 as the number of bits.
6546 @defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
6547 Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the
6548 variable to be output, if there is one, or @code{NULL_TREE} if there
6549 is no corresponding variable. If you define this macro, GCC will use it
6550 in place of both @code{ASM_OUTPUT_DECL} and
6551 @code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see
6552 the variable's decl in order to chose what to output.
6555 @defmac ASM_OUTPUT_SHARED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
6556 If defined, it is similar to @code{ASM_OUTPUT_LOCAL}, except that it
6557 is used when @var{name} is shared. If not defined, @code{ASM_OUTPUT_LOCAL}
6562 @subsection Output and Generation of Labels
6564 @c prevent bad page break with this line
6565 This is about outputting labels.
6567 @findex assemble_name
6568 @defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name})
6569 A C statement (sans semicolon) to output to the stdio stream
6570 @var{stream} the assembler definition of a label named @var{name}.
6571 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
6572 output the name itself; before and after that, output the additional
6573 assembler syntax for defining the name, and a newline. A default
6574 definition of this macro is provided which is correct for most systems.
6577 @findex assemble_name_raw
6578 @defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name})
6579 Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known
6580 to refer to a compiler-generated label. The default definition uses
6581 @code{assemble_name_raw}, which is like @code{assemble_name} except
6582 that it is more efficient.
6586 A C string containing the appropriate assembler directive to specify the
6587 size of a symbol, without any arguments. On systems that use ELF, the
6588 default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other
6589 systems, the default is not to define this macro.
6591 Define this macro only if it is correct to use the default definitions
6592 of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE}
6593 for your system. If you need your own custom definitions of those
6594 macros, or if you do not need explicit symbol sizes at all, do not
6598 @defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size})
6599 A C statement (sans semicolon) to output to the stdio stream
6600 @var{stream} a directive telling the assembler that the size of the
6601 symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}.
6602 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
6606 @defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name})
6607 A C statement (sans semicolon) to output to the stdio stream
6608 @var{stream} a directive telling the assembler to calculate the size of
6609 the symbol @var{name} by subtracting its address from the current
6612 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
6613 provided. The default assumes that the assembler recognizes a special
6614 @samp{.} symbol as referring to the current address, and can calculate
6615 the difference between this and another symbol. If your assembler does
6616 not recognize @samp{.} or cannot do calculations with it, you will need
6617 to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique.
6621 A C string containing the appropriate assembler directive to specify the
6622 type of a symbol, without any arguments. On systems that use ELF, the
6623 default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other
6624 systems, the default is not to define this macro.
6626 Define this macro only if it is correct to use the default definition of
6627 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
6628 custom definition of this macro, or if you do not need explicit symbol
6629 types at all, do not define this macro.
6632 @defmac TYPE_OPERAND_FMT
6633 A C string which specifies (using @code{printf} syntax) the format of
6634 the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the
6635 default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems,
6636 the default is not to define this macro.
6638 Define this macro only if it is correct to use the default definition of
6639 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
6640 custom definition of this macro, or if you do not need explicit symbol
6641 types at all, do not define this macro.
6644 @defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type})
6645 A C statement (sans semicolon) to output to the stdio stream
6646 @var{stream} a directive telling the assembler that the type of the
6647 symbol @var{name} is @var{type}. @var{type} is a C string; currently,
6648 that string is always either @samp{"function"} or @samp{"object"}, but
6649 you should not count on this.
6651 If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default
6652 definition of this macro is provided.
6655 @defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
6656 A C statement (sans semicolon) to output to the stdio stream
6657 @var{stream} any text necessary for declaring the name @var{name} of a
6658 function which is being defined. This macro is responsible for
6659 outputting the label definition (perhaps using
6660 @code{ASM_OUTPUT_LABEL}). The argument @var{decl} is the
6661 @code{FUNCTION_DECL} tree node representing the function.
6663 If this macro is not defined, then the function name is defined in the
6664 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
6666 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
6670 @defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl})
6671 A C statement (sans semicolon) to output to the stdio stream
6672 @var{stream} any text necessary for declaring the size of a function
6673 which is being defined. The argument @var{name} is the name of the
6674 function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node
6675 representing the function.
6677 If this macro is not defined, then the function size is not defined.
6679 You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition
6683 @defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl})
6684 A C statement (sans semicolon) to output to the stdio stream
6685 @var{stream} any text necessary for declaring the name @var{name} of an
6686 initialized variable which is being defined. This macro must output the
6687 label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument
6688 @var{decl} is the @code{VAR_DECL} tree node representing the variable.
6690 If this macro is not defined, then the variable name is defined in the
6691 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
6693 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or
6694 @code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro.
6697 @defmac ASM_DECLARE_CONSTANT_NAME (@var{stream}, @var{name}, @var{exp}, @var{size})
6698 A C statement (sans semicolon) to output to the stdio stream
6699 @var{stream} any text necessary for declaring the name @var{name} of a
6700 constant which is being defined. This macro is responsible for
6701 outputting the label definition (perhaps using
6702 @code{ASM_OUTPUT_LABEL}). The argument @var{exp} is the
6703 value of the constant, and @var{size} is the size of the constant
6704 in bytes. @var{name} will be an internal label.
6706 If this macro is not defined, then the @var{name} is defined in the
6707 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
6709 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
6713 @defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name})
6714 A C statement (sans semicolon) to output to the stdio stream
6715 @var{stream} any text necessary for claiming a register @var{regno}
6716 for a global variable @var{decl} with name @var{name}.
6718 If you don't define this macro, that is equivalent to defining it to do
6722 @defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend})
6723 A C statement (sans semicolon) to finish up declaring a variable name
6724 once the compiler has processed its initializer fully and thus has had a
6725 chance to determine the size of an array when controlled by an
6726 initializer. This is used on systems where it's necessary to declare
6727 something about the size of the object.
6729 If you don't define this macro, that is equivalent to defining it to do
6732 You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or
6733 @code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro.
6736 @deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_LABEL (FILE *@var{stream}, const char *@var{name})
6737 This target hook is a function to output to the stdio stream
6738 @var{stream} some commands that will make the label @var{name} global;
6739 that is, available for reference from other files.
6741 The default implementation relies on a proper definition of
6742 @code{GLOBAL_ASM_OP}.
6745 @defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name})
6746 A C statement (sans semicolon) to output to the stdio stream
6747 @var{stream} some commands that will make the label @var{name} weak;
6748 that is, available for reference from other files but only used if
6749 no other definition is available. Use the expression
6750 @code{assemble_name (@var{stream}, @var{name})} to output the name
6751 itself; before and after that, output the additional assembler syntax
6752 for making that name weak, and a newline.
6754 If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not
6755 support weak symbols and you should not define the @code{SUPPORTS_WEAK}
6759 @defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value})
6760 Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and
6761 @code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function
6762 or variable decl. If @var{value} is not @code{NULL}, this C statement
6763 should output to the stdio stream @var{stream} assembler code which
6764 defines (equates) the weak symbol @var{name} to have the value
6765 @var{value}. If @var{value} is @code{NULL}, it should output commands
6766 to make @var{name} weak.
6769 @defmac SUPPORTS_WEAK
6770 A C expression which evaluates to true if the target supports weak symbols.
6772 If you don't define this macro, @file{defaults.h} provides a default
6773 definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL}
6774 is defined, the default definition is @samp{1}; otherwise, it is
6775 @samp{0}. Define this macro if you want to control weak symbol support
6776 with a compiler flag such as @option{-melf}.
6779 @defmac MAKE_DECL_ONE_ONLY (@var{decl})
6780 A C statement (sans semicolon) to mark @var{decl} to be emitted as a
6781 public symbol such that extra copies in multiple translation units will
6782 be discarded by the linker. Define this macro if your object file
6783 format provides support for this concept, such as the @samp{COMDAT}
6784 section flags in the Microsoft Windows PE/COFF format, and this support
6785 requires changes to @var{decl}, such as putting it in a separate section.
6788 @defmac SUPPORTS_ONE_ONLY
6789 A C expression which evaluates to true if the target supports one-only
6792 If you don't define this macro, @file{varasm.c} provides a default
6793 definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default
6794 definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if
6795 you want to control one-only symbol support with a compiler flag, or if
6796 setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to
6797 be emitted as one-only.
6800 @deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_VISIBILITY (tree @var{decl}, const char *@var{visibility})
6801 This target hook is a function to output to @var{asm_out_file} some
6802 commands that will make the symbol(s) associated with @var{decl} have
6803 hidden, protected or internal visibility as specified by @var{visibility}.
6806 @defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC
6807 A C expression that evaluates to true if the target's linker expects
6808 that weak symbols do not appear in a static archive's table of contents.
6809 The default is @code{0}.
6811 Leaving weak symbols out of an archive's table of contents means that,
6812 if a symbol will only have a definition in one translation unit and
6813 will have undefined references from other translation units, that
6814 symbol should not be weak. Defining this macro to be nonzero will
6815 thus have the effect that certain symbols that would normally be weak
6816 (explicit template instantiations, and vtables for polymorphic classes
6817 with noninline key methods) will instead be nonweak.
6819 The C++ ABI requires this macro to be zero. Define this macro for
6820 targets where full C++ ABI compliance is impossible and where linker
6821 restrictions require weak symbols to be left out of a static archive's
6825 @defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
6826 A C statement (sans semicolon) to output to the stdio stream
6827 @var{stream} any text necessary for declaring the name of an external
6828 symbol named @var{name} which is referenced in this compilation but
6829 not defined. The value of @var{decl} is the tree node for the
6832 This macro need not be defined if it does not need to output anything.
6833 The GNU assembler and most Unix assemblers don't require anything.
6836 @deftypefn {Target Hook} void TARGET_ASM_EXTERNAL_LIBCALL (rtx @var{symref})
6837 This target hook is a function to output to @var{asm_out_file} an assembler
6838 pseudo-op to declare a library function name external. The name of the
6839 library function is given by @var{symref}, which is a @code{symbol_ref}.
6842 @deftypefn {Target Hook} void TARGET_ASM_MARK_DECL_PRESERVED (tree @var{decl})
6843 This target hook is a function to output to @var{asm_out_file} an assembler
6844 directive to annotate used symbol. Darwin target use .no_dead_code_strip
6848 @defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
6849 A C statement (sans semicolon) to output to the stdio stream
6850 @var{stream} a reference in assembler syntax to a label named
6851 @var{name}. This should add @samp{_} to the front of the name, if that
6852 is customary on your operating system, as it is in most Berkeley Unix
6853 systems. This macro is used in @code{assemble_name}.
6856 @defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym})
6857 A C statement (sans semicolon) to output a reference to
6858 @code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name}
6859 will be used to output the name of the symbol. This macro may be used
6860 to modify the way a symbol is referenced depending on information
6861 encoded by @code{TARGET_ENCODE_SECTION_INFO}.
6864 @defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf})
6865 A C statement (sans semicolon) to output a reference to @var{buf}, the
6866 result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined,
6867 @code{assemble_name} will be used to output the name of the symbol.
6868 This macro is not used by @code{output_asm_label}, or the @code{%l}
6869 specifier that calls it; the intention is that this macro should be set
6870 when it is necessary to output a label differently when its address is
6874 @deftypefn {Target Hook} void TARGET_ASM_INTERNAL_LABEL (FILE *@var{stream}, const char *@var{prefix}, unsigned long @var{labelno})
6875 A function to output to the stdio stream @var{stream} a label whose
6876 name is made from the string @var{prefix} and the number @var{labelno}.
6878 It is absolutely essential that these labels be distinct from the labels
6879 used for user-level functions and variables. Otherwise, certain programs
6880 will have name conflicts with internal labels.
6882 It is desirable to exclude internal labels from the symbol table of the
6883 object file. Most assemblers have a naming convention for labels that
6884 should be excluded; on many systems, the letter @samp{L} at the
6885 beginning of a label has this effect. You should find out what
6886 convention your system uses, and follow it.
6888 The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}.
6891 @defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num})
6892 A C statement to output to the stdio stream @var{stream} a debug info
6893 label whose name is made from the string @var{prefix} and the number
6894 @var{num}. This is useful for VLIW targets, where debug info labels
6895 may need to be treated differently than branch target labels. On some
6896 systems, branch target labels must be at the beginning of instruction
6897 bundles, but debug info labels can occur in the middle of instruction
6900 If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be
6904 @defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
6905 A C statement to store into the string @var{string} a label whose name
6906 is made from the string @var{prefix} and the number @var{num}.
6908 This string, when output subsequently by @code{assemble_name}, should
6909 produce the output that @code{(*targetm.asm_out.internal_label)} would produce
6910 with the same @var{prefix} and @var{num}.
6912 If the string begins with @samp{*}, then @code{assemble_name} will
6913 output the rest of the string unchanged. It is often convenient for
6914 @code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the
6915 string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets
6916 to output the string, and may change it. (Of course,
6917 @code{ASM_OUTPUT_LABELREF} is also part of your machine description, so
6918 you should know what it does on your machine.)
6921 @defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
6922 A C expression to assign to @var{outvar} (which is a variable of type
6923 @code{char *}) a newly allocated string made from the string
6924 @var{name} and the number @var{number}, with some suitable punctuation
6925 added. Use @code{alloca} to get space for the string.
6927 The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to
6928 produce an assembler label for an internal static variable whose name is
6929 @var{name}. Therefore, the string must be such as to result in valid
6930 assembler code. The argument @var{number} is different each time this
6931 macro is executed; it prevents conflicts between similarly-named
6932 internal static variables in different scopes.
6934 Ideally this string should not be a valid C identifier, to prevent any
6935 conflict with the user's own symbols. Most assemblers allow periods
6936 or percent signs in assembler symbols; putting at least one of these
6937 between the name and the number will suffice.
6939 If this macro is not defined, a default definition will be provided
6940 which is correct for most systems.
6943 @defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value})
6944 A C statement to output to the stdio stream @var{stream} assembler code
6945 which defines (equates) the symbol @var{name} to have the value @var{value}.
6948 If @code{SET_ASM_OP} is defined, a default definition is provided which is
6949 correct for most systems.
6952 @defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value})
6953 A C statement to output to the stdio stream @var{stream} assembler code
6954 which defines (equates) the symbol whose tree node is @var{decl_of_name}
6955 to have the value of the tree node @var{decl_of_value}. This macro will
6956 be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if
6957 the tree nodes are available.
6960 If @code{SET_ASM_OP} is defined, a default definition is provided which is
6961 correct for most systems.
6964 @defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value})
6965 A C statement that evaluates to true if the assembler code which defines
6966 (equates) the symbol whose tree node is @var{decl_of_name} to have the value
6967 of the tree node @var{decl_of_value} should be emitted near the end of the
6968 current compilation unit. The default is to not defer output of defines.
6969 This macro affects defines output by @samp{ASM_OUTPUT_DEF} and
6970 @samp{ASM_OUTPUT_DEF_FROM_DECLS}.
6973 @defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value})
6974 A C statement to output to the stdio stream @var{stream} assembler code
6975 which defines (equates) the weak symbol @var{name} to have the value
6976 @var{value}. If @var{value} is @code{NULL}, it defines @var{name} as
6977 an undefined weak symbol.
6979 Define this macro if the target only supports weak aliases; define
6980 @code{ASM_OUTPUT_DEF} instead if possible.
6983 @defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name})
6984 Define this macro to override the default assembler names used for
6985 Objective-C methods.
6987 The default name is a unique method number followed by the name of the
6988 class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of
6989 the category is also included in the assembler name (e.g.@:
6992 These names are safe on most systems, but make debugging difficult since
6993 the method's selector is not present in the name. Therefore, particular
6994 systems define other ways of computing names.
6996 @var{buf} is an expression of type @code{char *} which gives you a
6997 buffer in which to store the name; its length is as long as
6998 @var{class_name}, @var{cat_name} and @var{sel_name} put together, plus
6999 50 characters extra.
7001 The argument @var{is_inst} specifies whether the method is an instance
7002 method or a class method; @var{class_name} is the name of the class;
7003 @var{cat_name} is the name of the category (or @code{NULL} if the method is not
7004 in a category); and @var{sel_name} is the name of the selector.
7006 On systems where the assembler can handle quoted names, you can use this
7007 macro to provide more human-readable names.
7010 @defmac ASM_DECLARE_CLASS_REFERENCE (@var{stream}, @var{name})
7011 A C statement (sans semicolon) to output to the stdio stream
7012 @var{stream} commands to declare that the label @var{name} is an
7013 Objective-C class reference. This is only needed for targets whose
7014 linkers have special support for NeXT-style runtimes.
7017 @defmac ASM_DECLARE_UNRESOLVED_REFERENCE (@var{stream}, @var{name})
7018 A C statement (sans semicolon) to output to the stdio stream
7019 @var{stream} commands to declare that the label @var{name} is an
7020 unresolved Objective-C class reference. This is only needed for targets
7021 whose linkers have special support for NeXT-style runtimes.
7024 @node Initialization
7025 @subsection How Initialization Functions Are Handled
7026 @cindex initialization routines
7027 @cindex termination routines
7028 @cindex constructors, output of
7029 @cindex destructors, output of
7031 The compiled code for certain languages includes @dfn{constructors}
7032 (also called @dfn{initialization routines})---functions to initialize
7033 data in the program when the program is started. These functions need
7034 to be called before the program is ``started''---that is to say, before
7035 @code{main} is called.
7037 Compiling some languages generates @dfn{destructors} (also called
7038 @dfn{termination routines}) that should be called when the program
7041 To make the initialization and termination functions work, the compiler
7042 must output something in the assembler code to cause those functions to
7043 be called at the appropriate time. When you port the compiler to a new
7044 system, you need to specify how to do this.
7046 There are two major ways that GCC currently supports the execution of
7047 initialization and termination functions. Each way has two variants.
7048 Much of the structure is common to all four variations.
7050 @findex __CTOR_LIST__
7051 @findex __DTOR_LIST__
7052 The linker must build two lists of these functions---a list of
7053 initialization functions, called @code{__CTOR_LIST__}, and a list of
7054 termination functions, called @code{__DTOR_LIST__}.
7056 Each list always begins with an ignored function pointer (which may hold
7057 0, @minus{}1, or a count of the function pointers after it, depending on
7058 the environment). This is followed by a series of zero or more function
7059 pointers to constructors (or destructors), followed by a function
7060 pointer containing zero.
7062 Depending on the operating system and its executable file format, either
7063 @file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup
7064 time and exit time. Constructors are called in reverse order of the
7065 list; destructors in forward order.
7067 The best way to handle static constructors works only for object file
7068 formats which provide arbitrarily-named sections. A section is set
7069 aside for a list of constructors, and another for a list of destructors.
7070 Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each
7071 object file that defines an initialization function also puts a word in
7072 the constructor section to point to that function. The linker
7073 accumulates all these words into one contiguous @samp{.ctors} section.
7074 Termination functions are handled similarly.
7076 This method will be chosen as the default by @file{target-def.h} if
7077 @code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not
7078 support arbitrary sections, but does support special designated
7079 constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP}
7080 and @code{DTORS_SECTION_ASM_OP} to achieve the same effect.
7082 When arbitrary sections are available, there are two variants, depending
7083 upon how the code in @file{crtstuff.c} is called. On systems that
7084 support a @dfn{.init} section which is executed at program startup,
7085 parts of @file{crtstuff.c} are compiled into that section. The
7086 program is linked by the @command{gcc} driver like this:
7089 ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o
7092 The prologue of a function (@code{__init}) appears in the @code{.init}
7093 section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise
7094 for the function @code{__fini} in the @dfn{.fini} section. Normally these
7095 files are provided by the operating system or by the GNU C library, but
7096 are provided by GCC for a few targets.
7098 The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets)
7099 compiled from @file{crtstuff.c}. They contain, among other things, code
7100 fragments within the @code{.init} and @code{.fini} sections that branch
7101 to routines in the @code{.text} section. The linker will pull all parts
7102 of a section together, which results in a complete @code{__init} function
7103 that invokes the routines we need at startup.
7105 To use this variant, you must define the @code{INIT_SECTION_ASM_OP}
7108 If no init section is available, when GCC compiles any function called
7109 @code{main} (or more accurately, any function designated as a program
7110 entry point by the language front end calling @code{expand_main_function}),
7111 it inserts a procedure call to @code{__main} as the first executable code
7112 after the function prologue. The @code{__main} function is defined
7113 in @file{libgcc2.c} and runs the global constructors.
7115 In file formats that don't support arbitrary sections, there are again
7116 two variants. In the simplest variant, the GNU linker (GNU @code{ld})
7117 and an `a.out' format must be used. In this case,
7118 @code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs}
7119 entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__},
7120 and with the address of the void function containing the initialization
7121 code as its value. The GNU linker recognizes this as a request to add
7122 the value to a @dfn{set}; the values are accumulated, and are eventually
7123 placed in the executable as a vector in the format described above, with
7124 a leading (ignored) count and a trailing zero element.
7125 @code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init
7126 section is available, the absence of @code{INIT_SECTION_ASM_OP} causes
7127 the compilation of @code{main} to call @code{__main} as above, starting
7128 the initialization process.
7130 The last variant uses neither arbitrary sections nor the GNU linker.
7131 This is preferable when you want to do dynamic linking and when using
7132 file formats which the GNU linker does not support, such as `ECOFF'@. In
7133 this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and
7134 termination functions are recognized simply by their names. This requires
7135 an extra program in the linkage step, called @command{collect2}. This program
7136 pretends to be the linker, for use with GCC; it does its job by running
7137 the ordinary linker, but also arranges to include the vectors of
7138 initialization and termination functions. These functions are called
7139 via @code{__main} as described above. In order to use this method,
7140 @code{use_collect2} must be defined in the target in @file{config.gcc}.
7143 The following section describes the specific macros that control and
7144 customize the handling of initialization and termination functions.
7147 @node Macros for Initialization
7148 @subsection Macros Controlling Initialization Routines
7150 Here are the macros that control how the compiler handles initialization
7151 and termination functions:
7153 @defmac INIT_SECTION_ASM_OP
7154 If defined, a C string constant, including spacing, for the assembler
7155 operation to identify the following data as initialization code. If not
7156 defined, GCC will assume such a section does not exist. When you are
7157 using special sections for initialization and termination functions, this
7158 macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to
7159 run the initialization functions.
7162 @defmac HAS_INIT_SECTION
7163 If defined, @code{main} will not call @code{__main} as described above.
7164 This macro should be defined for systems that control start-up code
7165 on a symbol-by-symbol basis, such as OSF/1, and should not
7166 be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}.
7169 @defmac LD_INIT_SWITCH
7170 If defined, a C string constant for a switch that tells the linker that
7171 the following symbol is an initialization routine.
7174 @defmac LD_FINI_SWITCH
7175 If defined, a C string constant for a switch that tells the linker that
7176 the following symbol is a finalization routine.
7179 @defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func})
7180 If defined, a C statement that will write a function that can be
7181 automatically called when a shared library is loaded. The function
7182 should call @var{func}, which takes no arguments. If not defined, and
7183 the object format requires an explicit initialization function, then a
7184 function called @code{_GLOBAL__DI} will be generated.
7186 This function and the following one are used by collect2 when linking a
7187 shared library that needs constructors or destructors, or has DWARF2
7188 exception tables embedded in the code.
7191 @defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func})
7192 If defined, a C statement that will write a function that can be
7193 automatically called when a shared library is unloaded. The function
7194 should call @var{func}, which takes no arguments. If not defined, and
7195 the object format requires an explicit finalization function, then a
7196 function called @code{_GLOBAL__DD} will be generated.
7199 @defmac INVOKE__main
7200 If defined, @code{main} will call @code{__main} despite the presence of
7201 @code{INIT_SECTION_ASM_OP}. This macro should be defined for systems
7202 where the init section is not actually run automatically, but is still
7203 useful for collecting the lists of constructors and destructors.
7206 @defmac SUPPORTS_INIT_PRIORITY
7207 If nonzero, the C++ @code{init_priority} attribute is supported and the
7208 compiler should emit instructions to control the order of initialization
7209 of objects. If zero, the compiler will issue an error message upon
7210 encountering an @code{init_priority} attribute.
7213 @deftypefn {Target Hook} bool TARGET_HAVE_CTORS_DTORS
7214 This value is true if the target supports some ``native'' method of
7215 collecting constructors and destructors to be run at startup and exit.
7216 It is false if we must use @command{collect2}.
7219 @deftypefn {Target Hook} void TARGET_ASM_CONSTRUCTOR (rtx @var{symbol}, int @var{priority})
7220 If defined, a function that outputs assembler code to arrange to call
7221 the function referenced by @var{symbol} at initialization time.
7223 Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking
7224 no arguments and with no return value. If the target supports initialization
7225 priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY};
7226 otherwise it must be @code{DEFAULT_INIT_PRIORITY}.
7228 If this macro is not defined by the target, a suitable default will
7229 be chosen if (1) the target supports arbitrary section names, (2) the
7230 target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2}
7234 @deftypefn {Target Hook} void TARGET_ASM_DESTRUCTOR (rtx @var{symbol}, int @var{priority})
7235 This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination
7236 functions rather than initialization functions.
7239 If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine
7240 generated for the generated object file will have static linkage.
7242 If your system uses @command{collect2} as the means of processing
7243 constructors, then that program normally uses @command{nm} to scan
7244 an object file for constructor functions to be called.
7246 On certain kinds of systems, you can define this macro to make
7247 @command{collect2} work faster (and, in some cases, make it work at all):
7249 @defmac OBJECT_FORMAT_COFF
7250 Define this macro if the system uses COFF (Common Object File Format)
7251 object files, so that @command{collect2} can assume this format and scan
7252 object files directly for dynamic constructor/destructor functions.
7254 This macro is effective only in a native compiler; @command{collect2} as
7255 part of a cross compiler always uses @command{nm} for the target machine.
7258 @defmac REAL_NM_FILE_NAME
7259 Define this macro as a C string constant containing the file name to use
7260 to execute @command{nm}. The default is to search the path normally for
7263 If your system supports shared libraries and has a program to list the
7264 dynamic dependencies of a given library or executable, you can define
7265 these macros to enable support for running initialization and
7266 termination functions in shared libraries:
7270 Define this macro to a C string constant containing the name of the program
7271 which lists dynamic dependencies, like @command{"ldd"} under SunOS 4.
7274 @defmac PARSE_LDD_OUTPUT (@var{ptr})
7275 Define this macro to be C code that extracts filenames from the output
7276 of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable
7277 of type @code{char *} that points to the beginning of a line of output
7278 from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the
7279 code must advance @var{ptr} to the beginning of the filename on that
7280 line. Otherwise, it must set @var{ptr} to @code{NULL}.
7283 @node Instruction Output
7284 @subsection Output of Assembler Instructions
7286 @c prevent bad page break with this line
7287 This describes assembler instruction output.
7289 @defmac REGISTER_NAMES
7290 A C initializer containing the assembler's names for the machine
7291 registers, each one as a C string constant. This is what translates
7292 register numbers in the compiler into assembler language.
7295 @defmac ADDITIONAL_REGISTER_NAMES
7296 If defined, a C initializer for an array of structures containing a name
7297 and a register number. This macro defines additional names for hard
7298 registers, thus allowing the @code{asm} option in declarations to refer
7299 to registers using alternate names.
7302 @defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
7303 Define this macro if you are using an unusual assembler that
7304 requires different names for the machine instructions.
7306 The definition is a C statement or statements which output an
7307 assembler instruction opcode to the stdio stream @var{stream}. The
7308 macro-operand @var{ptr} is a variable of type @code{char *} which
7309 points to the opcode name in its ``internal'' form---the form that is
7310 written in the machine description. The definition should output the
7311 opcode name to @var{stream}, performing any translation you desire, and
7312 increment the variable @var{ptr} to point at the end of the opcode
7313 so that it will not be output twice.
7315 In fact, your macro definition may process less than the entire opcode
7316 name, or more than the opcode name; but if you want to process text
7317 that includes @samp{%}-sequences to substitute operands, you must take
7318 care of the substitution yourself. Just be sure to increment
7319 @var{ptr} over whatever text should not be output normally.
7321 @findex recog_data.operand
7322 If you need to look at the operand values, they can be found as the
7323 elements of @code{recog_data.operand}.
7325 If the macro definition does nothing, the instruction is output
7329 @defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
7330 If defined, a C statement to be executed just prior to the output of
7331 assembler code for @var{insn}, to modify the extracted operands so
7332 they will be output differently.
7334 Here the argument @var{opvec} is the vector containing the operands
7335 extracted from @var{insn}, and @var{noperands} is the number of
7336 elements of the vector which contain meaningful data for this insn.
7337 The contents of this vector are what will be used to convert the insn
7338 template into assembler code, so you can change the assembler output
7339 by changing the contents of the vector.
7341 This macro is useful when various assembler syntaxes share a single
7342 file of instruction patterns; by defining this macro differently, you
7343 can cause a large class of instructions to be output differently (such
7344 as with rearranged operands). Naturally, variations in assembler
7345 syntax affecting individual insn patterns ought to be handled by
7346 writing conditional output routines in those patterns.
7348 If this macro is not defined, it is equivalent to a null statement.
7351 @defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
7352 A C compound statement to output to stdio stream @var{stream} the
7353 assembler syntax for an instruction operand @var{x}. @var{x} is an
7356 @var{code} is a value that can be used to specify one of several ways
7357 of printing the operand. It is used when identical operands must be
7358 printed differently depending on the context. @var{code} comes from
7359 the @samp{%} specification that was used to request printing of the
7360 operand. If the specification was just @samp{%@var{digit}} then
7361 @var{code} is 0; if the specification was @samp{%@var{ltr}
7362 @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
7365 If @var{x} is a register, this macro should print the register's name.
7366 The names can be found in an array @code{reg_names} whose type is
7367 @code{char *[]}. @code{reg_names} is initialized from
7368 @code{REGISTER_NAMES}.
7370 When the machine description has a specification @samp{%@var{punct}}
7371 (a @samp{%} followed by a punctuation character), this macro is called
7372 with a null pointer for @var{x} and the punctuation character for
7376 @defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code})
7377 A C expression which evaluates to true if @var{code} is a valid
7378 punctuation character for use in the @code{PRINT_OPERAND} macro. If
7379 @code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
7380 punctuation characters (except for the standard one, @samp{%}) are used
7384 @defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
7385 A C compound statement to output to stdio stream @var{stream} the
7386 assembler syntax for an instruction operand that is a memory reference
7387 whose address is @var{x}. @var{x} is an RTL expression.
7389 @cindex @code{TARGET_ENCODE_SECTION_INFO} usage
7390 On some machines, the syntax for a symbolic address depends on the
7391 section that the address refers to. On these machines, define the hook
7392 @code{TARGET_ENCODE_SECTION_INFO} to store the information into the
7393 @code{symbol_ref}, and then check for it here. @xref{Assembler
7397 @findex dbr_sequence_length
7398 @defmac DBR_OUTPUT_SEQEND (@var{file})
7399 A C statement, to be executed after all slot-filler instructions have
7400 been output. If necessary, call @code{dbr_sequence_length} to
7401 determine the number of slots filled in a sequence (zero if not
7402 currently outputting a sequence), to decide how many no-ops to output,
7405 Don't define this macro if it has nothing to do, but it is helpful in
7406 reading assembly output if the extent of the delay sequence is made
7407 explicit (e.g.@: with white space).
7410 @findex final_sequence
7411 Note that output routines for instructions with delay slots must be
7412 prepared to deal with not being output as part of a sequence
7413 (i.e.@: when the scheduling pass is not run, or when no slot fillers could be
7414 found.) The variable @code{final_sequence} is null when not
7415 processing a sequence, otherwise it contains the @code{sequence} rtx
7419 @defmac REGISTER_PREFIX
7420 @defmacx LOCAL_LABEL_PREFIX
7421 @defmacx USER_LABEL_PREFIX
7422 @defmacx IMMEDIATE_PREFIX
7423 If defined, C string expressions to be used for the @samp{%R}, @samp{%L},
7424 @samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see
7425 @file{final.c}). These are useful when a single @file{md} file must
7426 support multiple assembler formats. In that case, the various @file{tm.h}
7427 files can define these macros differently.
7430 @defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format})
7431 If defined this macro should expand to a series of @code{case}
7432 statements which will be parsed inside the @code{switch} statement of
7433 the @code{asm_fprintf} function. This allows targets to define extra
7434 printf formats which may useful when generating their assembler
7435 statements. Note that uppercase letters are reserved for future
7436 generic extensions to asm_fprintf, and so are not available to target
7437 specific code. The output file is given by the parameter @var{file}.
7438 The varargs input pointer is @var{argptr} and the rest of the format
7439 string, starting the character after the one that is being switched
7440 upon, is pointed to by @var{format}.
7443 @defmac ASSEMBLER_DIALECT
7444 If your target supports multiple dialects of assembler language (such as
7445 different opcodes), define this macro as a C expression that gives the
7446 numeric index of the assembler language dialect to use, with zero as the
7449 If this macro is defined, you may use constructs of the form
7451 @samp{@{option0|option1|option2@dots{}@}}
7454 in the output templates of patterns (@pxref{Output Template}) or in the
7455 first argument of @code{asm_fprintf}. This construct outputs
7456 @samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of
7457 @code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters
7458 within these strings retain their usual meaning. If there are fewer
7459 alternatives within the braces than the value of
7460 @code{ASSEMBLER_DIALECT}, the construct outputs nothing.
7462 If you do not define this macro, the characters @samp{@{}, @samp{|} and
7463 @samp{@}} do not have any special meaning when used in templates or
7464 operands to @code{asm_fprintf}.
7466 Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX},
7467 @code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express
7468 the variations in assembler language syntax with that mechanism. Define
7469 @code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax
7470 if the syntax variant are larger and involve such things as different
7471 opcodes or operand order.
7474 @defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
7475 A C expression to output to @var{stream} some assembler code
7476 which will push hard register number @var{regno} onto the stack.
7477 The code need not be optimal, since this macro is used only when
7481 @defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
7482 A C expression to output to @var{stream} some assembler code
7483 which will pop hard register number @var{regno} off of the stack.
7484 The code need not be optimal, since this macro is used only when
7488 @node Dispatch Tables
7489 @subsection Output of Dispatch Tables
7491 @c prevent bad page break with this line
7492 This concerns dispatch tables.
7494 @cindex dispatch table
7495 @defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel})
7496 A C statement to output to the stdio stream @var{stream} an assembler
7497 pseudo-instruction to generate a difference between two labels.
7498 @var{value} and @var{rel} are the numbers of two internal labels. The
7499 definitions of these labels are output using
7500 @code{(*targetm.asm_out.internal_label)}, and they must be printed in the same
7501 way here. For example,
7504 fprintf (@var{stream}, "\t.word L%d-L%d\n",
7505 @var{value}, @var{rel})
7508 You must provide this macro on machines where the addresses in a
7509 dispatch table are relative to the table's own address. If defined, GCC
7510 will also use this macro on all machines when producing PIC@.
7511 @var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the
7512 mode and flags can be read.
7515 @defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
7516 This macro should be provided on machines where the addresses
7517 in a dispatch table are absolute.
7519 The definition should be a C statement to output to the stdio stream
7520 @var{stream} an assembler pseudo-instruction to generate a reference to
7521 a label. @var{value} is the number of an internal label whose
7522 definition is output using @code{(*targetm.asm_out.internal_label)}.
7526 fprintf (@var{stream}, "\t.word L%d\n", @var{value})
7530 @defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
7531 Define this if the label before a jump-table needs to be output
7532 specially. The first three arguments are the same as for
7533 @code{(*targetm.asm_out.internal_label)}; the fourth argument is the
7534 jump-table which follows (a @code{jump_insn} containing an
7535 @code{addr_vec} or @code{addr_diff_vec}).
7537 This feature is used on system V to output a @code{swbeg} statement
7540 If this macro is not defined, these labels are output with
7541 @code{(*targetm.asm_out.internal_label)}.
7544 @defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
7545 Define this if something special must be output at the end of a
7546 jump-table. The definition should be a C statement to be executed
7547 after the assembler code for the table is written. It should write
7548 the appropriate code to stdio stream @var{stream}. The argument
7549 @var{table} is the jump-table insn, and @var{num} is the label-number
7550 of the preceding label.
7552 If this macro is not defined, nothing special is output at the end of
7556 @deftypefn {Target Hook} void TARGET_ASM_EMIT_UNWIND_LABEL (@var{stream}, @var{decl}, @var{for_eh}, @var{empty})
7557 This target hook emits a label at the beginning of each FDE@. It
7558 should be defined on targets where FDEs need special labels, and it
7559 should write the appropriate label, for the FDE associated with the
7560 function declaration @var{decl}, to the stdio stream @var{stream}.
7561 The third argument, @var{for_eh}, is a boolean: true if this is for an
7562 exception table. The fourth argument, @var{empty}, is a boolean:
7563 true if this is a placeholder label for an omitted FDE@.
7565 The default is that FDEs are not given nonlocal labels.
7568 @deftypefn {Taget Hook} void TARGET_UNWIND_EMIT (FILE * @var{stream}, rtx @var{insn})
7569 This target hook emits and assembly directives required to unwind the
7570 given instruction. This is only used when TARGET_UNWIND_INFO is set.
7573 @node Exception Region Output
7574 @subsection Assembler Commands for Exception Regions
7576 @c prevent bad page break with this line
7578 This describes commands marking the start and the end of an exception
7581 @defmac EH_FRAME_SECTION_NAME
7582 If defined, a C string constant for the name of the section containing
7583 exception handling frame unwind information. If not defined, GCC will
7584 provide a default definition if the target supports named sections.
7585 @file{crtstuff.c} uses this macro to switch to the appropriate section.
7587 You should define this symbol if your target supports DWARF 2 frame
7588 unwind information and the default definition does not work.
7591 @defmac EH_FRAME_IN_DATA_SECTION
7592 If defined, DWARF 2 frame unwind information will be placed in the
7593 data section even though the target supports named sections. This
7594 might be necessary, for instance, if the system linker does garbage
7595 collection and sections cannot be marked as not to be collected.
7597 Do not define this macro unless @code{TARGET_ASM_NAMED_SECTION} is
7601 @defmac EH_TABLES_CAN_BE_READ_ONLY
7602 Define this macro to 1 if your target is such that no frame unwind
7603 information encoding used with non-PIC code will ever require a
7604 runtime relocation, but the linker may not support merging read-only
7605 and read-write sections into a single read-write section.
7608 @defmac MASK_RETURN_ADDR
7609 An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so
7610 that it does not contain any extraneous set bits in it.
7613 @defmac DWARF2_UNWIND_INFO
7614 Define this macro to 0 if your target supports DWARF 2 frame unwind
7615 information, but it does not yet work with exception handling.
7616 Otherwise, if your target supports this information (if it defines
7617 @samp{INCOMING_RETURN_ADDR_RTX} and either @samp{UNALIGNED_INT_ASM_OP}
7618 or @samp{OBJECT_FORMAT_ELF}), GCC will provide a default definition of
7621 If @code{TARGET_UNWIND_INFO} is defined, the target specific unwinder
7622 will be used in all cases. Defining this macro will enable the generation
7623 of DWARF 2 frame debugging information.
7625 If @code{TARGET_UNWIND_INFO} is not defined, and this macro is defined to 1,
7626 the DWARF 2 unwinder will be the default exception handling mechanism;
7627 otherwise, @code{setjmp}/@code{longjmp} will be used by default.
7630 @defmac TARGET_UNWIND_INFO
7631 Define this macro if your target has ABI specified unwind tables. Usually
7632 these will be output by @code{TARGET_UNWIND_EMIT}.
7635 @deftypevar {Target Hook} bool TARGET_UNWID_TABLES_DEFAULT
7636 This variable should be set to @code{true} if the target ABI requires unwinding
7637 tables even when exceptions are not used.
7640 @defmac MUST_USE_SJLJ_EXCEPTIONS
7641 This macro need only be defined if @code{DWARF2_UNWIND_INFO} is
7642 runtime-variable. In that case, @file{except.h} cannot correctly
7643 determine the corresponding definition of
7644 @code{MUST_USE_SJLJ_EXCEPTIONS}, so the target must provide it directly.
7647 @defmac DWARF_CIE_DATA_ALIGNMENT
7648 This macro need only be defined if the target might save registers in the
7649 function prologue at an offset to the stack pointer that is not aligned to
7650 @code{UNITS_PER_WORD}. The definition should be the negative minimum
7651 alignment if @code{STACK_GROWS_DOWNWARD} is defined, and the positive
7652 minimum alignment otherwise. @xref{SDB and DWARF}. Only applicable if
7653 the target supports DWARF 2 frame unwind information.
7656 @deftypefn {Target Hook} void TARGET_ASM_EXCEPTION_SECTION ()
7657 If defined, a function that switches to the section in which the main
7658 exception table is to be placed (@pxref{Sections}). The default is a
7659 function that switches to a section named @code{.gcc_except_table} on
7660 machines that support named sections via
7661 @code{TARGET_ASM_NAMED_SECTION}, otherwise if @option{-fpic} or
7662 @option{-fPIC} is in effect, the @code{data_section}, otherwise the
7663 @code{readonly_data_section}.
7666 @deftypefn {Target Hook} void TARGET_ASM_EH_FRAME_SECTION ()
7667 If defined, a function that switches to the section in which the DWARF 2
7668 frame unwind information to be placed (@pxref{Sections}). The default
7669 is a function that outputs a standard GAS section directive, if
7670 @code{EH_FRAME_SECTION_NAME} is defined, or else a data section
7671 directive followed by a synthetic label.
7674 @deftypevar {Target Hook} bool TARGET_TERMINATE_DW2_EH_FRAME_INFO
7675 Contains the value true if the target should add a zero word onto the
7676 end of a Dwarf-2 frame info section when used for exception handling.
7677 Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and
7681 @deftypefn {Target Hook} rtx TARGET_DWARF_REGISTER_SPAN (rtx @var{reg})
7682 Given a register, this hook should return a parallel of registers to
7683 represent where to find the register pieces. Define this hook if the
7684 register and its mode are represented in Dwarf in non-contiguous
7685 locations, or if the register should be represented in more than one
7686 register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}.
7687 If not defined, the default is to return @code{NULL_RTX}.
7690 @deftypefn {Target Hook} bool TARGET_ASM_TTYPE (rtx @var{sym})
7691 This hook is used to output a reference from a frame unwinding table to
7692 the type_info object identified by @var{sym}. It should return @code{true}
7693 if the reference was output. Returning @code{false} will cause the
7694 reference to be output using the normal Dwarf2 routines.
7697 @deftypefn {Target Hook} bool TARGET_ARM_EABI_UNWINDER
7698 This hook should be set to @code{true} on targets that use an ARM EABI
7699 based unwinding library, and @code{false} on other targets. This effects
7700 the format of unwinding tables, and how the unwinder in entered after
7701 running a cleanup. The default is @code{false}.
7704 @node Alignment Output
7705 @subsection Assembler Commands for Alignment
7707 @c prevent bad page break with this line
7708 This describes commands for alignment.
7710 @defmac JUMP_ALIGN (@var{label})
7711 The alignment (log base 2) to put in front of @var{label}, which is
7712 a common destination of jumps and has no fallthru incoming edge.
7714 This macro need not be defined if you don't want any special alignment
7715 to be done at such a time. Most machine descriptions do not currently
7718 Unless it's necessary to inspect the @var{label} parameter, it is better
7719 to set the variable @var{align_jumps} in the target's
7720 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
7721 selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation.
7724 @defmac LABEL_ALIGN_AFTER_BARRIER (@var{label})
7725 The alignment (log base 2) to put in front of @var{label}, which follows
7728 This macro need not be defined if you don't want any special alignment
7729 to be done at such a time. Most machine descriptions do not currently
7733 @defmac LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP
7734 The maximum number of bytes to skip when applying
7735 @code{LABEL_ALIGN_AFTER_BARRIER}. This works only if
7736 @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
7739 @defmac LOOP_ALIGN (@var{label})
7740 The alignment (log base 2) to put in front of @var{label}, which follows
7741 a @code{NOTE_INSN_LOOP_BEG} note.
7743 This macro need not be defined if you don't want any special alignment
7744 to be done at such a time. Most machine descriptions do not currently
7747 Unless it's necessary to inspect the @var{label} parameter, it is better
7748 to set the variable @code{align_loops} in the target's
7749 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
7750 selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation.
7753 @defmac LOOP_ALIGN_MAX_SKIP
7754 The maximum number of bytes to skip when applying @code{LOOP_ALIGN}.
7755 This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
7758 @defmac LABEL_ALIGN (@var{label})
7759 The alignment (log base 2) to put in front of @var{label}.
7760 If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment,
7761 the maximum of the specified values is used.
7763 Unless it's necessary to inspect the @var{label} parameter, it is better
7764 to set the variable @code{align_labels} in the target's
7765 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
7766 selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation.
7769 @defmac LABEL_ALIGN_MAX_SKIP
7770 The maximum number of bytes to skip when applying @code{LABEL_ALIGN}.
7771 This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
7774 @defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
7775 A C statement to output to the stdio stream @var{stream} an assembler
7776 instruction to advance the location counter by @var{nbytes} bytes.
7777 Those bytes should be zero when loaded. @var{nbytes} will be a C
7778 expression of type @code{int}.
7781 @defmac ASM_NO_SKIP_IN_TEXT
7782 Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the
7783 text section because it fails to put zeros in the bytes that are skipped.
7784 This is true on many Unix systems, where the pseudo--op to skip bytes
7785 produces no-op instructions rather than zeros when used in the text
7789 @defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
7790 A C statement to output to the stdio stream @var{stream} an assembler
7791 command to advance the location counter to a multiple of 2 to the
7792 @var{power} bytes. @var{power} will be a C expression of type @code{int}.
7795 @defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power})
7796 Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used
7797 for padding, if necessary.
7800 @defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip})
7801 A C statement to output to the stdio stream @var{stream} an assembler
7802 command to advance the location counter to a multiple of 2 to the
7803 @var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to
7804 satisfy the alignment request. @var{power} and @var{max_skip} will be
7805 a C expression of type @code{int}.
7809 @node Debugging Info
7810 @section Controlling Debugging Information Format
7812 @c prevent bad page break with this line
7813 This describes how to specify debugging information.
7816 * All Debuggers:: Macros that affect all debugging formats uniformly.
7817 * DBX Options:: Macros enabling specific options in DBX format.
7818 * DBX Hooks:: Hook macros for varying DBX format.
7819 * File Names and DBX:: Macros controlling output of file names in DBX format.
7820 * SDB and DWARF:: Macros for SDB (COFF) and DWARF formats.
7821 * VMS Debug:: Macros for VMS debug format.
7825 @subsection Macros Affecting All Debugging Formats
7827 @c prevent bad page break with this line
7828 These macros affect all debugging formats.
7830 @defmac DBX_REGISTER_NUMBER (@var{regno})
7831 A C expression that returns the DBX register number for the compiler
7832 register number @var{regno}. In the default macro provided, the value
7833 of this expression will be @var{regno} itself. But sometimes there are
7834 some registers that the compiler knows about and DBX does not, or vice
7835 versa. In such cases, some register may need to have one number in the
7836 compiler and another for DBX@.
7838 If two registers have consecutive numbers inside GCC, and they can be
7839 used as a pair to hold a multiword value, then they @emph{must} have
7840 consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}.
7841 Otherwise, debuggers will be unable to access such a pair, because they
7842 expect register pairs to be consecutive in their own numbering scheme.
7844 If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that
7845 does not preserve register pairs, then what you must do instead is
7846 redefine the actual register numbering scheme.
7849 @defmac DEBUGGER_AUTO_OFFSET (@var{x})
7850 A C expression that returns the integer offset value for an automatic
7851 variable having address @var{x} (an RTL expression). The default
7852 computation assumes that @var{x} is based on the frame-pointer and
7853 gives the offset from the frame-pointer. This is required for targets
7854 that produce debugging output for DBX or COFF-style debugging output
7855 for SDB and allow the frame-pointer to be eliminated when the
7856 @option{-g} options is used.
7859 @defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x})
7860 A C expression that returns the integer offset value for an argument
7861 having address @var{x} (an RTL expression). The nominal offset is
7865 @defmac PREFERRED_DEBUGGING_TYPE
7866 A C expression that returns the type of debugging output GCC should
7867 produce when the user specifies just @option{-g}. Define
7868 this if you have arranged for GCC to support more than one format of
7869 debugging output. Currently, the allowable values are @code{DBX_DEBUG},
7870 @code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG},
7871 @code{XCOFF_DEBUG}, @code{VMS_DEBUG}, and @code{VMS_AND_DWARF2_DEBUG}.
7873 When the user specifies @option{-ggdb}, GCC normally also uses the
7874 value of this macro to select the debugging output format, but with two
7875 exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the
7876 value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is
7877 defined, GCC uses @code{DBX_DEBUG}.
7879 The value of this macro only affects the default debugging output; the
7880 user can always get a specific type of output by using @option{-gstabs},
7881 @option{-gcoff}, @option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}.
7885 @subsection Specific Options for DBX Output
7887 @c prevent bad page break with this line
7888 These are specific options for DBX output.
7890 @defmac DBX_DEBUGGING_INFO
7891 Define this macro if GCC should produce debugging output for DBX
7892 in response to the @option{-g} option.
7895 @defmac XCOFF_DEBUGGING_INFO
7896 Define this macro if GCC should produce XCOFF format debugging output
7897 in response to the @option{-g} option. This is a variant of DBX format.
7900 @defmac DEFAULT_GDB_EXTENSIONS
7901 Define this macro to control whether GCC should by default generate
7902 GDB's extended version of DBX debugging information (assuming DBX-format
7903 debugging information is enabled at all). If you don't define the
7904 macro, the default is 1: always generate the extended information
7905 if there is any occasion to.
7908 @defmac DEBUG_SYMS_TEXT
7909 Define this macro if all @code{.stabs} commands should be output while
7910 in the text section.
7913 @defmac ASM_STABS_OP
7914 A C string constant, including spacing, naming the assembler pseudo op to
7915 use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol.
7916 If you don't define this macro, @code{"\t.stabs\t"} is used. This macro
7917 applies only to DBX debugging information format.
7920 @defmac ASM_STABD_OP
7921 A C string constant, including spacing, naming the assembler pseudo op to
7922 use instead of @code{"\t.stabd\t"} to define a debugging symbol whose
7923 value is the current location. If you don't define this macro,
7924 @code{"\t.stabd\t"} is used. This macro applies only to DBX debugging
7928 @defmac ASM_STABN_OP
7929 A C string constant, including spacing, naming the assembler pseudo op to
7930 use instead of @code{"\t.stabn\t"} to define a debugging symbol with no
7931 name. If you don't define this macro, @code{"\t.stabn\t"} is used. This
7932 macro applies only to DBX debugging information format.
7935 @defmac DBX_NO_XREFS
7936 Define this macro if DBX on your system does not support the construct
7937 @samp{xs@var{tagname}}. On some systems, this construct is used to
7938 describe a forward reference to a structure named @var{tagname}.
7939 On other systems, this construct is not supported at all.
7942 @defmac DBX_CONTIN_LENGTH
7943 A symbol name in DBX-format debugging information is normally
7944 continued (split into two separate @code{.stabs} directives) when it
7945 exceeds a certain length (by default, 80 characters). On some
7946 operating systems, DBX requires this splitting; on others, splitting
7947 must not be done. You can inhibit splitting by defining this macro
7948 with the value zero. You can override the default splitting-length by
7949 defining this macro as an expression for the length you desire.
7952 @defmac DBX_CONTIN_CHAR
7953 Normally continuation is indicated by adding a @samp{\} character to
7954 the end of a @code{.stabs} string when a continuation follows. To use
7955 a different character instead, define this macro as a character
7956 constant for the character you want to use. Do not define this macro
7957 if backslash is correct for your system.
7960 @defmac DBX_STATIC_STAB_DATA_SECTION
7961 Define this macro if it is necessary to go to the data section before
7962 outputting the @samp{.stabs} pseudo-op for a non-global static
7966 @defmac DBX_TYPE_DECL_STABS_CODE
7967 The value to use in the ``code'' field of the @code{.stabs} directive
7968 for a typedef. The default is @code{N_LSYM}.
7971 @defmac DBX_STATIC_CONST_VAR_CODE
7972 The value to use in the ``code'' field of the @code{.stabs} directive
7973 for a static variable located in the text section. DBX format does not
7974 provide any ``right'' way to do this. The default is @code{N_FUN}.
7977 @defmac DBX_REGPARM_STABS_CODE
7978 The value to use in the ``code'' field of the @code{.stabs} directive
7979 for a parameter passed in registers. DBX format does not provide any
7980 ``right'' way to do this. The default is @code{N_RSYM}.
7983 @defmac DBX_REGPARM_STABS_LETTER
7984 The letter to use in DBX symbol data to identify a symbol as a parameter
7985 passed in registers. DBX format does not customarily provide any way to
7986 do this. The default is @code{'P'}.
7989 @defmac DBX_FUNCTION_FIRST
7990 Define this macro if the DBX information for a function and its
7991 arguments should precede the assembler code for the function. Normally,
7992 in DBX format, the debugging information entirely follows the assembler
7996 @defmac DBX_BLOCKS_FUNCTION_RELATIVE
7997 Define this macro, with value 1, if the value of a symbol describing
7998 the scope of a block (@code{N_LBRAC} or @code{N_RBRAC}) should be
7999 relative to the start of the enclosing function. Normally, GCC uses
8000 an absolute address.
8003 @defmac DBX_LINES_FUNCTION_RELATIVE
8004 Define this macro, with value 1, if the value of a symbol indicating
8005 the current line number (@code{N_SLINE}) should be relative to the
8006 start of the enclosing function. Normally, GCC uses an absolute address.
8009 @defmac DBX_USE_BINCL
8010 Define this macro if GCC should generate @code{N_BINCL} and
8011 @code{N_EINCL} stabs for included header files, as on Sun systems. This
8012 macro also directs GCC to output a type number as a pair of a file
8013 number and a type number within the file. Normally, GCC does not
8014 generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single
8015 number for a type number.
8019 @subsection Open-Ended Hooks for DBX Format
8021 @c prevent bad page break with this line
8022 These are hooks for DBX format.
8024 @defmac DBX_OUTPUT_LBRAC (@var{stream}, @var{name})
8025 Define this macro to say how to output to @var{stream} the debugging
8026 information for the start of a scope level for variable names. The
8027 argument @var{name} is the name of an assembler symbol (for use with
8028 @code{assemble_name}) whose value is the address where the scope begins.
8031 @defmac DBX_OUTPUT_RBRAC (@var{stream}, @var{name})
8032 Like @code{DBX_OUTPUT_LBRAC}, but for the end of a scope level.
8035 @defmac DBX_OUTPUT_NFUN (@var{stream}, @var{lscope_label}, @var{decl})
8036 Define this macro if the target machine requires special handling to
8037 output an @code{N_FUN} entry for the function @var{decl}.
8040 @defmac DBX_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter})
8041 A C statement to output DBX debugging information before code for line
8042 number @var{line} of the current source file to the stdio stream
8043 @var{stream}. @var{counter} is the number of time the macro was
8044 invoked, including the current invocation; it is intended to generate
8045 unique labels in the assembly output.
8047 This macro should not be defined if the default output is correct, or
8048 if it can be made correct by defining @code{DBX_LINES_FUNCTION_RELATIVE}.
8051 @defmac NO_DBX_FUNCTION_END
8052 Some stabs encapsulation formats (in particular ECOFF), cannot handle the
8053 @code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct.
8054 On those machines, define this macro to turn this feature off without
8055 disturbing the rest of the gdb extensions.
8058 @defmac NO_DBX_BNSYM_ENSYM
8059 Some assemblers cannot handle the @code{.stabd BNSYM/ENSYM,0,0} gdb dbx
8060 extension construct. On those machines, define this macro to turn this
8061 feature off without disturbing the rest of the gdb extensions.
8064 @node File Names and DBX
8065 @subsection File Names in DBX Format
8067 @c prevent bad page break with this line
8068 This describes file names in DBX format.
8070 @defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name})
8071 A C statement to output DBX debugging information to the stdio stream
8072 @var{stream}, which indicates that file @var{name} is the main source
8073 file---the file specified as the input file for compilation.
8074 This macro is called only once, at the beginning of compilation.
8076 This macro need not be defined if the standard form of output
8077 for DBX debugging information is appropriate.
8079 It may be necessary to refer to a label equal to the beginning of the
8080 text section. You can use @samp{assemble_name (stream, ltext_label_name)}
8081 to do so. If you do this, you must also set the variable
8082 @var{used_ltext_label_name} to @code{true}.
8085 @defmac NO_DBX_MAIN_SOURCE_DIRECTORY
8086 Define this macro, with value 1, if GCC should not emit an indication
8087 of the current directory for compilation and current source language at
8088 the beginning of the file.
8091 @defmac NO_DBX_GCC_MARKER
8092 Define this macro, with value 1, if GCC should not emit an indication
8093 that this object file was compiled by GCC@. The default is to emit
8094 an @code{N_OPT} stab at the beginning of every source file, with
8095 @samp{gcc2_compiled.} for the string and value 0.
8098 @defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name})
8099 A C statement to output DBX debugging information at the end of
8100 compilation of the main source file @var{name}. Output should be
8101 written to the stdio stream @var{stream}.
8103 If you don't define this macro, nothing special is output at the end
8104 of compilation, which is correct for most machines.
8107 @defmac DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END
8108 Define this macro @emph{instead of} defining
8109 @code{DBX_OUTPUT_MAIN_SOURCE_FILE_END}, if what needs to be output at
8110 the end of compilation is a @code{N_SO} stab with an empty string,
8111 whose value is the highest absolute text address in the file.
8116 @subsection Macros for SDB and DWARF Output
8118 @c prevent bad page break with this line
8119 Here are macros for SDB and DWARF output.
8121 @defmac SDB_DEBUGGING_INFO
8122 Define this macro if GCC should produce COFF-style debugging output
8123 for SDB in response to the @option{-g} option.
8126 @defmac DWARF2_DEBUGGING_INFO
8127 Define this macro if GCC should produce dwarf version 2 format
8128 debugging output in response to the @option{-g} option.
8130 @deftypefn {Target Hook} int TARGET_DWARF_CALLING_CONVENTION (tree @var{function})
8131 Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to
8132 be emitted for each function. Instead of an integer return the enum
8133 value for the @code{DW_CC_} tag.
8136 To support optional call frame debugging information, you must also
8137 define @code{INCOMING_RETURN_ADDR_RTX} and either set
8138 @code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the
8139 prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save}
8140 as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't.
8143 @defmac DWARF2_FRAME_INFO
8144 Define this macro to a nonzero value if GCC should always output
8145 Dwarf 2 frame information. If @code{DWARF2_UNWIND_INFO}
8146 (@pxref{Exception Region Output} is nonzero, GCC will output this
8147 information not matter how you define @code{DWARF2_FRAME_INFO}.
8150 @defmac DWARF2_ASM_LINE_DEBUG_INFO
8151 Define this macro to be a nonzero value if the assembler can generate Dwarf 2
8152 line debug info sections. This will result in much more compact line number
8153 tables, and hence is desirable if it works.
8156 @defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
8157 A C statement to issue assembly directives that create a difference
8158 between the two given labels, using an integer of the given size.
8161 @defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label})
8162 A C statement to issue assembly directives that create a
8163 section-relative reference to the given label, using an integer of the
8167 @defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label})
8168 A C statement to issue assembly directives that create a self-relative
8169 reference to the given label, using an integer of the given size.
8172 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_DWARF_DTPREL (FILE *@var{FILE}, int @var{size}, rtx @var{x})
8173 If defined, this target hook is a function which outputs a DTP-relative
8174 reference to the given TLS symbol of the specified size.
8177 @defmac PUT_SDB_@dots{}
8178 Define these macros to override the assembler syntax for the special
8179 SDB assembler directives. See @file{sdbout.c} for a list of these
8180 macros and their arguments. If the standard syntax is used, you need
8181 not define them yourself.
8185 Some assemblers do not support a semicolon as a delimiter, even between
8186 SDB assembler directives. In that case, define this macro to be the
8187 delimiter to use (usually @samp{\n}). It is not necessary to define
8188 a new set of @code{PUT_SDB_@var{op}} macros if this is the only change
8192 @defmac SDB_ALLOW_UNKNOWN_REFERENCES
8193 Define this macro to allow references to unknown structure,
8194 union, or enumeration tags to be emitted. Standard COFF does not
8195 allow handling of unknown references, MIPS ECOFF has support for
8199 @defmac SDB_ALLOW_FORWARD_REFERENCES
8200 Define this macro to allow references to structure, union, or
8201 enumeration tags that have not yet been seen to be handled. Some
8202 assemblers choke if forward tags are used, while some require it.
8205 @defmac SDB_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
8206 A C statement to output SDB debugging information before code for line
8207 number @var{line} of the current source file to the stdio stream
8208 @var{stream}. The default is to emit an @code{.ln} directive.
8213 @subsection Macros for VMS Debug Format
8215 @c prevent bad page break with this line
8216 Here are macros for VMS debug format.
8218 @defmac VMS_DEBUGGING_INFO
8219 Define this macro if GCC should produce debugging output for VMS
8220 in response to the @option{-g} option. The default behavior for VMS
8221 is to generate minimal debug info for a traceback in the absence of
8222 @option{-g} unless explicitly overridden with @option{-g0}. This
8223 behavior is controlled by @code{OPTIMIZATION_OPTIONS} and
8224 @code{OVERRIDE_OPTIONS}.
8227 @node Floating Point
8228 @section Cross Compilation and Floating Point
8229 @cindex cross compilation and floating point
8230 @cindex floating point and cross compilation
8232 While all modern machines use twos-complement representation for integers,
8233 there are a variety of representations for floating point numbers. This
8234 means that in a cross-compiler the representation of floating point numbers
8235 in the compiled program may be different from that used in the machine
8236 doing the compilation.
8238 Because different representation systems may offer different amounts of
8239 range and precision, all floating point constants must be represented in
8240 the target machine's format. Therefore, the cross compiler cannot
8241 safely use the host machine's floating point arithmetic; it must emulate
8242 the target's arithmetic. To ensure consistency, GCC always uses
8243 emulation to work with floating point values, even when the host and
8244 target floating point formats are identical.
8246 The following macros are provided by @file{real.h} for the compiler to
8247 use. All parts of the compiler which generate or optimize
8248 floating-point calculations must use these macros. They may evaluate
8249 their operands more than once, so operands must not have side effects.
8251 @defmac REAL_VALUE_TYPE
8252 The C data type to be used to hold a floating point value in the target
8253 machine's format. Typically this is a @code{struct} containing an
8254 array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque
8258 @deftypefn Macro int REAL_VALUES_EQUAL (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
8259 Compares for equality the two values, @var{x} and @var{y}. If the target
8260 floating point format supports negative zeroes and/or NaNs,
8261 @samp{REAL_VALUES_EQUAL (-0.0, 0.0)} is true, and
8262 @samp{REAL_VALUES_EQUAL (NaN, NaN)} is false.
8265 @deftypefn Macro int REAL_VALUES_LESS (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
8266 Tests whether @var{x} is less than @var{y}.
8269 @deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x})
8270 Truncates @var{x} to a signed integer, rounding toward zero.
8273 @deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x})
8274 Truncates @var{x} to an unsigned integer, rounding toward zero. If
8275 @var{x} is negative, returns zero.
8278 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, enum machine_mode @var{mode})
8279 Converts @var{string} into a floating point number in the target machine's
8280 representation for mode @var{mode}. This routine can handle both
8281 decimal and hexadecimal floating point constants, using the syntax
8282 defined by the C language for both.
8285 @deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x})
8286 Returns 1 if @var{x} is negative (including negative zero), 0 otherwise.
8289 @deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x})
8290 Determines whether @var{x} represents infinity (positive or negative).
8293 @deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x})
8294 Determines whether @var{x} represents a ``NaN'' (not-a-number).
8297 @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})
8298 Calculates an arithmetic operation on the two floating point values
8299 @var{x} and @var{y}, storing the result in @var{output} (which must be a
8302 The operation to be performed is specified by @var{code}. Only the
8303 following codes are supported: @code{PLUS_EXPR}, @code{MINUS_EXPR},
8304 @code{MULT_EXPR}, @code{RDIV_EXPR}, @code{MAX_EXPR}, @code{MIN_EXPR}.
8306 If @code{REAL_ARITHMETIC} is asked to evaluate division by zero and the
8307 target's floating point format cannot represent infinity, it will call
8308 @code{abort}. Callers should check for this situation first, using
8309 @code{MODE_HAS_INFINITIES}. @xref{Storage Layout}.
8312 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x})
8313 Returns the negative of the floating point value @var{x}.
8316 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x})
8317 Returns the absolute value of @var{x}.
8320 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_TRUNCATE (REAL_VALUE_TYPE @var{mode}, enum machine_mode @var{x})
8321 Truncates the floating point value @var{x} to fit in @var{mode}. The
8322 return value is still a full-size @code{REAL_VALUE_TYPE}, but it has an
8323 appropriate bit pattern to be output asa floating constant whose
8324 precision accords with mode @var{mode}.
8327 @deftypefn Macro void REAL_VALUE_TO_INT (HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, REAL_VALUE_TYPE @var{x})
8328 Converts a floating point value @var{x} into a double-precision integer
8329 which is then stored into @var{low} and @var{high}. If the value is not
8330 integral, it is truncated.
8333 @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})
8334 Converts a double-precision integer found in @var{low} and @var{high},
8335 into a floating point value which is then stored into @var{x}. The
8336 value is truncated to fit in mode @var{mode}.
8339 @node Mode Switching
8340 @section Mode Switching Instructions
8341 @cindex mode switching
8342 The following macros control mode switching optimizations:
8344 @defmac OPTIMIZE_MODE_SWITCHING (@var{entity})
8345 Define this macro if the port needs extra instructions inserted for mode
8346 switching in an optimizing compilation.
8348 For an example, the SH4 can perform both single and double precision
8349 floating point operations, but to perform a single precision operation,
8350 the FPSCR PR bit has to be cleared, while for a double precision
8351 operation, this bit has to be set. Changing the PR bit requires a general
8352 purpose register as a scratch register, hence these FPSCR sets have to
8353 be inserted before reload, i.e.@: you can't put this into instruction emitting
8354 or @code{TARGET_MACHINE_DEPENDENT_REORG}.
8356 You can have multiple entities that are mode-switched, and select at run time
8357 which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should
8358 return nonzero for any @var{entity} that needs mode-switching.
8359 If you define this macro, you also have to define
8360 @code{NUM_MODES_FOR_MODE_SWITCHING}, @code{MODE_NEEDED},
8361 @code{MODE_PRIORITY_TO_MODE} and @code{EMIT_MODE_SET}.
8362 @code{MODE_AFTER}, @code{MODE_ENTRY}, and @code{MODE_EXIT}
8366 @defmac NUM_MODES_FOR_MODE_SWITCHING
8367 If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as
8368 initializer for an array of integers. Each initializer element
8369 N refers to an entity that needs mode switching, and specifies the number
8370 of different modes that might need to be set for this entity.
8371 The position of the initializer in the initializer---starting counting at
8372 zero---determines the integer that is used to refer to the mode-switched
8374 In macros that take mode arguments / yield a mode result, modes are
8375 represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode
8376 switch is needed / supplied.
8379 @defmac MODE_NEEDED (@var{entity}, @var{insn})
8380 @var{entity} is an integer specifying a mode-switched entity. If
8381 @code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to
8382 return an integer value not larger than the corresponding element in
8383 @code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must
8384 be switched into prior to the execution of @var{insn}.
8387 @defmac MODE_AFTER (@var{mode}, @var{insn})
8388 If this macro is defined, it is evaluated for every @var{insn} during
8389 mode switching. It determines the mode that an insn results in (if
8390 different from the incoming mode).
8393 @defmac MODE_ENTRY (@var{entity})
8394 If this macro is defined, it is evaluated for every @var{entity} that needs
8395 mode switching. It should evaluate to an integer, which is a mode that
8396 @var{entity} is assumed to be switched to at function entry. If @code{MODE_ENTRY}
8397 is defined then @code{MODE_EXIT} must be defined.
8400 @defmac MODE_EXIT (@var{entity})
8401 If this macro is defined, it is evaluated for every @var{entity} that needs
8402 mode switching. It should evaluate to an integer, which is a mode that
8403 @var{entity} is assumed to be switched to at function exit. If @code{MODE_EXIT}
8404 is defined then @code{MODE_ENTRY} must be defined.
8407 @defmac MODE_PRIORITY_TO_MODE (@var{entity}, @var{n})
8408 This macro specifies the order in which modes for @var{entity} are processed.
8409 0 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the
8410 lowest. The value of the macro should be an integer designating a mode
8411 for @var{entity}. For any fixed @var{entity}, @code{mode_priority_to_mode}
8412 (@var{entity}, @var{n}) shall be a bijection in 0 @dots{}
8413 @code{num_modes_for_mode_switching[@var{entity}] - 1}.
8416 @defmac EMIT_MODE_SET (@var{entity}, @var{mode}, @var{hard_regs_live})
8417 Generate one or more insns to set @var{entity} to @var{mode}.
8418 @var{hard_reg_live} is the set of hard registers live at the point where
8419 the insn(s) are to be inserted.
8422 @node Target Attributes
8423 @section Defining target-specific uses of @code{__attribute__}
8424 @cindex target attributes
8425 @cindex machine attributes
8426 @cindex attributes, target-specific
8428 Target-specific attributes may be defined for functions, data and types.
8429 These are described using the following target hooks; they also need to
8430 be documented in @file{extend.texi}.
8432 @deftypevr {Target Hook} {const struct attribute_spec *} TARGET_ATTRIBUTE_TABLE
8433 If defined, this target hook points to an array of @samp{struct
8434 attribute_spec} (defined in @file{tree.h}) specifying the machine
8435 specific attributes for this target and some of the restrictions on the
8436 entities to which these attributes are applied and the arguments they
8440 @deftypefn {Target Hook} int TARGET_COMP_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2})
8441 If defined, this target hook is a function which returns zero if the attributes on
8442 @var{type1} and @var{type2} are incompatible, one if they are compatible,
8443 and two if they are nearly compatible (which causes a warning to be
8444 generated). If this is not defined, machine-specific attributes are
8445 supposed always to be compatible.
8448 @deftypefn {Target Hook} void TARGET_SET_DEFAULT_TYPE_ATTRIBUTES (tree @var{type})
8449 If defined, this target hook is a function which assigns default attributes to
8450 newly defined @var{type}.
8453 @deftypefn {Target Hook} tree TARGET_MERGE_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2})
8454 Define this target hook if the merging of type attributes needs special
8455 handling. If defined, the result is a list of the combined
8456 @code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed
8457 that @code{comptypes} has already been called and returned 1. This
8458 function may call @code{merge_attributes} to handle machine-independent
8462 @deftypefn {Target Hook} tree TARGET_MERGE_DECL_ATTRIBUTES (tree @var{olddecl}, tree @var{newdecl})
8463 Define this target hook if the merging of decl attributes needs special
8464 handling. If defined, the result is a list of the combined
8465 @code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}.
8466 @var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of
8467 when this is needed are when one attribute overrides another, or when an
8468 attribute is nullified by a subsequent definition. This function may
8469 call @code{merge_attributes} to handle machine-independent merging.
8471 @findex TARGET_DLLIMPORT_DECL_ATTRIBUTES
8472 If the only target-specific handling you require is @samp{dllimport}
8473 for Microsoft Windows targets, you should define the macro
8474 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler
8475 will then define a function called
8476 @code{merge_dllimport_decl_attributes} which can then be defined as
8477 the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also
8478 add @code{handle_dll_attribute} in the attribute table for your port
8479 to perform initial processing of the @samp{dllimport} and
8480 @samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and
8481 @file{i386/i386.c}, for example.
8484 @deftypefn {Target Hook} bool TARGET_VALID_DLLIMPORT_ATTRIBUTE_P (tree @var{decl})
8485 @var{decl} is a variable or function with @code{__attribute__((dllimport))}
8486 specified. Use this hook if the target needs to add extra validation
8487 checks to @code{handle_dll_attribute}.
8490 @defmac TARGET_DECLSPEC
8491 Define this macro to a nonzero value if you want to treat
8492 @code{__declspec(X)} as equivalent to @code{__attribute((X))}. By
8493 default, this behavior is enabled only for targets that define
8494 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation
8495 of @code{__declspec} is via a built-in macro, but you should not rely
8496 on this implementation detail.
8499 @deftypefn {Target Hook} void TARGET_INSERT_ATTRIBUTES (tree @var{node}, tree *@var{attr_ptr})
8500 Define this target hook if you want to be able to add attributes to a decl
8501 when it is being created. This is normally useful for back ends which
8502 wish to implement a pragma by using the attributes which correspond to
8503 the pragma's effect. The @var{node} argument is the decl which is being
8504 created. The @var{attr_ptr} argument is a pointer to the attribute list
8505 for this decl. The list itself should not be modified, since it may be
8506 shared with other decls, but attributes may be chained on the head of
8507 the list and @code{*@var{attr_ptr}} modified to point to the new
8508 attributes, or a copy of the list may be made if further changes are
8512 @deftypefn {Target Hook} bool TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P (tree @var{fndecl})
8514 This target hook returns @code{true} if it is ok to inline @var{fndecl}
8515 into the current function, despite its having target-specific
8516 attributes, @code{false} otherwise. By default, if a function has a
8517 target specific attribute attached to it, it will not be inlined.
8520 @node MIPS Coprocessors
8521 @section Defining coprocessor specifics for MIPS targets.
8522 @cindex MIPS coprocessor-definition macros
8524 The MIPS specification allows MIPS implementations to have as many as 4
8525 coprocessors, each with as many as 32 private registers. GCC supports
8526 accessing these registers and transferring values between the registers
8527 and memory using asm-ized variables. For example:
8530 register unsigned int cp0count asm ("c0r1");
8536 (``c0r1'' is the default name of register 1 in coprocessor 0; alternate
8537 names may be added as described below, or the default names may be
8538 overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.)
8540 Coprocessor registers are assumed to be epilogue-used; sets to them will
8541 be preserved even if it does not appear that the register is used again
8542 later in the function.
8544 Another note: according to the MIPS spec, coprocessor 1 (if present) is
8545 the FPU@. One accesses COP1 registers through standard mips
8546 floating-point support; they are not included in this mechanism.
8548 There is one macro used in defining the MIPS coprocessor interface which
8549 you may want to override in subtargets; it is described below.
8551 @defmac ALL_COP_ADDITIONAL_REGISTER_NAMES
8552 A comma-separated list (with leading comma) of pairs describing the
8553 alternate names of coprocessor registers. The format of each entry should be
8555 @{ @var{alternatename}, @var{register_number}@}
8561 @section Parameters for Precompiled Header Validity Checking
8562 @cindex parameters, precompiled headers
8564 @deftypefn {Target Hook} void *TARGET_GET_PCH_VALIDITY (size_t *@var{sz})
8565 This hook returns the data needed by @code{TARGET_PCH_VALID_P} and sets
8566 @samp{*@var{sz}} to the size of the data in bytes.
8569 @deftypefn {Target Hook} const char *TARGET_PCH_VALID_P (const void *@var{data}, size_t @var{sz})
8570 This hook checks whether the options used to create a PCH file are
8571 compatible with the current settings. It returns @code{NULL}
8572 if so and a suitable error message if not. Error messages will
8573 be presented to the user and must be localized using @samp{_(@var{msg})}.
8575 @var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY}
8576 when the PCH file was created and @var{sz} is the size of that data in bytes.
8577 It's safe to assume that the data was created by the same version of the
8578 compiler, so no format checking is needed.
8580 The default definition of @code{default_pch_valid_p} should be
8581 suitable for most targets.
8584 @deftypefn {Target Hook} const char *TARGET_CHECK_PCH_TARGET_FLAGS (int @var{pch_flags})
8585 If this hook is nonnull, the default implementation of
8586 @code{TARGET_PCH_VALID_P} will use it to check for compatible values
8587 of @code{target_flags}. @var{pch_flags} specifies the value that
8588 @code{target_flags} had when the PCH file was created. The return
8589 value is the same as for @code{TARGET_PCH_VALID_P}.
8593 @section C++ ABI parameters
8594 @cindex parameters, c++ abi
8596 @deftypefn {Target Hook} tree TARGET_CXX_GUARD_TYPE (void)
8597 Define this hook to override the integer type used for guard variables.
8598 These are used to implement one-time construction of static objects. The
8599 default is long_long_integer_type_node.
8602 @deftypefn {Target Hook} bool TARGET_CXX_GUARD_MASK_BIT (void)
8603 This hook determines how guard variables are used. It should return
8604 @code{false} (the default) if first byte should be used. A return value of
8605 @code{true} indicates the least significant bit should be used.
8608 @deftypefn {Target Hook} tree TARGET_CXX_GET_COOKIE_SIZE (tree @var{type})
8609 This hook returns the size of the cookie to use when allocating an array
8610 whose elements have the indicated @var{type}. Assumes that it is already
8611 known that a cookie is needed. The default is
8612 @code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the
8613 IA64/Generic C++ ABI@.
8616 @deftypefn {Target Hook} bool TARGET_CXX_COOKIE_HAS_SIZE (void)
8617 This hook should return @code{true} if the element size should be stored in
8618 array cookies. The default is to return @code{false}.
8621 @deftypefn {Target Hook} int TARGET_CXX_IMPORT_EXPORT_CLASS (tree @var{type}, int @var{import_export})
8622 If defined by a backend this hook allows the decision made to export
8623 class @var{type} to be overruled. Upon entry @var{import_export}
8624 will contain 1 if the class is going to be exported, @minus{}1 if it is going
8625 to be imported and 0 otherwise. This function should return the
8626 modified value and perform any other actions necessary to support the
8627 backend's targeted operating system.
8630 @deftypefn {Target Hook} bool TARGET_CXX_CDTOR_RETURNS_THIS (void)
8631 This hook should return @code{true} if constructors and destructors return
8632 the address of the object created/destroyed. The default is to return
8636 @deftypefn {Target Hook} bool TARGET_CXX_KEY_METHOD_MAY_BE_INLINE (void)
8637 This hook returns true if the key method for a class (i.e., the method
8638 which, if defined in the current translation unit, causes the virtual
8639 table to be emitted) may be an inline function. Under the standard
8640 Itanium C++ ABI the key method may be an inline function so long as
8641 the function is not declared inline in the class definition. Under
8642 some variants of the ABI, an inline function can never be the key
8643 method. The default is to return @code{true}.
8646 @deftypefn {Target Hook} void TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY (tree @var{decl})
8647 @var{decl} is a virtual table, virtual table table, typeinfo object,
8648 or other similar implicit class data object that will be emitted with
8649 external linkage in this translation unit. No ELF visibility has been
8650 explicitly specified. If the target needs to specify a visibility
8651 other than that of the containing class, use this hook to set
8652 @code{DECL_VISIBILITY} and @code{DECL_VISIBILITY_SPECIFIED}.
8655 @deftypefn {Target Hook} bool TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT (void)
8656 This hook returns true (the default) if virtual tables and other
8657 similar implicit class data objects are always COMDAT if they have
8658 external linkage. If this hook returns false, then class data for
8659 classes whose virtual table will be emitted in only one translation
8660 unit will not be COMDAT.
8663 @deftypefn {Target Hook} bool TARGET_CXX_USE_AEABI_ATEXIT (void)
8664 This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI)
8665 should be used to register static destructors when @option{-fuse-cxa-atexit}
8666 is in effect. The default is to return false to use @code{__cxa_atexit}.
8669 @deftypefn {Target Hook} void TARGET_CXX_ADJUST_CLASS_AT_DEFINITION (tree @var{type})
8670 @var{type} is a C++ class (i.e., RECORD_TYPE or UNION_TYPE) that has just been
8671 defined. Use this hook to make adjustments to the class (eg, tweak
8672 visibility or perform any other required target modifications).
8676 @section Miscellaneous Parameters
8677 @cindex parameters, miscellaneous
8679 @c prevent bad page break with this line
8680 Here are several miscellaneous parameters.
8682 @defmac HAS_LONG_COND_BRANCH
8683 Define this boolean macro to indicate whether or not your architecture
8684 has conditional branches that can span all of memory. It is used in
8685 conjunction with an optimization that partitions hot and cold basic
8686 blocks into separate sections of the executable. If this macro is
8687 set to false, gcc will convert any conditional branches that attempt
8688 to cross between sections into unconditional branches or indirect jumps.
8691 @defmac HAS_LONG_UNCOND_BRANCH
8692 Define this boolean macro to indicate whether or not your architecture
8693 has unconditional branches that can span all of memory. It is used in
8694 conjunction with an optimization that partitions hot and cold basic
8695 blocks into separate sections of the executable. If this macro is
8696 set to false, gcc will convert any unconditional branches that attempt
8697 to cross between sections into indirect jumps.
8700 @defmac CASE_VECTOR_MODE
8701 An alias for a machine mode name. This is the machine mode that
8702 elements of a jump-table should have.
8705 @defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body})
8706 Optional: return the preferred mode for an @code{addr_diff_vec}
8707 when the minimum and maximum offset are known. If you define this,
8708 it enables extra code in branch shortening to deal with @code{addr_diff_vec}.
8709 To make this work, you also have to define @code{INSN_ALIGN} and
8710 make the alignment for @code{addr_diff_vec} explicit.
8711 The @var{body} argument is provided so that the offset_unsigned and scale
8712 flags can be updated.
8715 @defmac CASE_VECTOR_PC_RELATIVE
8716 Define this macro to be a C expression to indicate when jump-tables
8717 should contain relative addresses. You need not define this macro if
8718 jump-tables never contain relative addresses, or jump-tables should
8719 contain relative addresses only when @option{-fPIC} or @option{-fPIC}
8723 @defmac CASE_VALUES_THRESHOLD
8724 Define this to be the smallest number of different values for which it
8725 is best to use a jump-table instead of a tree of conditional branches.
8726 The default is four for machines with a @code{casesi} instruction and
8727 five otherwise. This is best for most machines.
8730 @defmac CASE_USE_BIT_TESTS
8731 Define this macro to be a C expression to indicate whether C switch
8732 statements may be implemented by a sequence of bit tests. This is
8733 advantageous on processors that can efficiently implement left shift
8734 of 1 by the number of bits held in a register, but inappropriate on
8735 targets that would require a loop. By default, this macro returns
8736 @code{true} if the target defines an @code{ashlsi3} pattern, and
8737 @code{false} otherwise.
8740 @defmac WORD_REGISTER_OPERATIONS
8741 Define this macro if operations between registers with integral mode
8742 smaller than a word are always performed on the entire register.
8743 Most RISC machines have this property and most CISC machines do not.
8746 @defmac LOAD_EXTEND_OP (@var{mem_mode})
8747 Define this macro to be a C expression indicating when insns that read
8748 memory in @var{mem_mode}, an integral mode narrower than a word, set the
8749 bits outside of @var{mem_mode} to be either the sign-extension or the
8750 zero-extension of the data read. Return @code{SIGN_EXTEND} for values
8751 of @var{mem_mode} for which the
8752 insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and
8753 @code{UNKNOWN} for other modes.
8755 This macro is not called with @var{mem_mode} non-integral or with a width
8756 greater than or equal to @code{BITS_PER_WORD}, so you may return any
8757 value in this case. Do not define this macro if it would always return
8758 @code{UNKNOWN}. On machines where this macro is defined, you will normally
8759 define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}.
8761 You may return a non-@code{UNKNOWN} value even if for some hard registers
8762 the sign extension is not performed, if for the @code{REGNO_REG_CLASS}
8763 of these hard registers @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero
8764 when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any
8765 integral mode larger than this but not larger than @code{word_mode}.
8767 You must return @code{UNKNOWN} if for some hard registers that allow this
8768 mode, @code{CANNOT_CHANGE_MODE_CLASS} says that they cannot change to
8769 @code{word_mode}, but that they can change to another integral mode that
8770 is larger then @var{mem_mode} but still smaller than @code{word_mode}.
8773 @defmac SHORT_IMMEDIATES_SIGN_EXTEND
8774 Define this macro if loading short immediate values into registers sign
8778 @defmac FIXUNS_TRUNC_LIKE_FIX_TRUNC
8779 Define this macro if the same instructions that convert a floating
8780 point number to a signed fixed point number also convert validly to an
8785 The maximum number of bytes that a single instruction can move quickly
8786 between memory and registers or between two memory locations.
8789 @defmac MAX_MOVE_MAX
8790 The maximum number of bytes that a single instruction can move quickly
8791 between memory and registers or between two memory locations. If this
8792 is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the
8793 constant value that is the largest value that @code{MOVE_MAX} can have
8797 @defmac SHIFT_COUNT_TRUNCATED
8798 A C expression that is nonzero if on this machine the number of bits
8799 actually used for the count of a shift operation is equal to the number
8800 of bits needed to represent the size of the object being shifted. When
8801 this macro is nonzero, the compiler will assume that it is safe to omit
8802 a sign-extend, zero-extend, and certain bitwise `and' instructions that
8803 truncates the count of a shift operation. On machines that have
8804 instructions that act on bit-fields at variable positions, which may
8805 include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED}
8806 also enables deletion of truncations of the values that serve as
8807 arguments to bit-field instructions.
8809 If both types of instructions truncate the count (for shifts) and
8810 position (for bit-field operations), or if no variable-position bit-field
8811 instructions exist, you should define this macro.
8813 However, on some machines, such as the 80386 and the 680x0, truncation
8814 only applies to shift operations and not the (real or pretended)
8815 bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on
8816 such machines. Instead, add patterns to the @file{md} file that include
8817 the implied truncation of the shift instructions.
8819 You need not define this macro if it would always have the value of zero.
8822 @anchor{TARGET_SHIFT_TRUNCATION_MASK}
8823 @deftypefn {Target Hook} int TARGET_SHIFT_TRUNCATION_MASK (enum machine_mode @var{mode})
8824 This function describes how the standard shift patterns for @var{mode}
8825 deal with shifts by negative amounts or by more than the width of the mode.
8826 @xref{shift patterns}.
8828 On many machines, the shift patterns will apply a mask @var{m} to the
8829 shift count, meaning that a fixed-width shift of @var{x} by @var{y} is
8830 equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If
8831 this is true for mode @var{mode}, the function should return @var{m},
8832 otherwise it should return 0. A return value of 0 indicates that no
8833 particular behavior is guaranteed.
8835 Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does
8836 @emph{not} apply to general shift rtxes; it applies only to instructions
8837 that are generated by the named shift patterns.
8839 The default implementation of this function returns
8840 @code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED}
8841 and 0 otherwise. This definition is always safe, but if
8842 @code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns
8843 nevertheless truncate the shift count, you may get better code
8847 @defmac TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
8848 A C expression which is nonzero if on this machine it is safe to
8849 ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
8850 bits (where @var{outprec} is smaller than @var{inprec}) by merely
8851 operating on it as if it had only @var{outprec} bits.
8853 On many machines, this expression can be 1.
8855 @c rearranged this, removed the phrase "it is reported that". this was
8856 @c to fix an overfull hbox. --mew 10feb93
8857 When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for
8858 modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result.
8859 If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in
8860 such cases may improve things.
8863 @defmac STORE_FLAG_VALUE
8864 A C expression describing the value returned by a comparison operator
8865 with an integral mode and stored by a store-flag instruction
8866 (@samp{s@var{cond}}) when the condition is true. This description must
8867 apply to @emph{all} the @samp{s@var{cond}} patterns and all the
8868 comparison operators whose results have a @code{MODE_INT} mode.
8870 A value of 1 or @minus{}1 means that the instruction implementing the
8871 comparison operator returns exactly 1 or @minus{}1 when the comparison is true
8872 and 0 when the comparison is false. Otherwise, the value indicates
8873 which bits of the result are guaranteed to be 1 when the comparison is
8874 true. This value is interpreted in the mode of the comparison
8875 operation, which is given by the mode of the first operand in the
8876 @samp{s@var{cond}} pattern. Either the low bit or the sign bit of
8877 @code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by
8880 If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will
8881 generate code that depends only on the specified bits. It can also
8882 replace comparison operators with equivalent operations if they cause
8883 the required bits to be set, even if the remaining bits are undefined.
8884 For example, on a machine whose comparison operators return an
8885 @code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as
8886 @samp{0x80000000}, saying that just the sign bit is relevant, the
8890 (ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0))
8897 (ashift:SI @var{x} (const_int @var{n}))
8901 where @var{n} is the appropriate shift count to move the bit being
8902 tested into the sign bit.
8904 There is no way to describe a machine that always sets the low-order bit
8905 for a true value, but does not guarantee the value of any other bits,
8906 but we do not know of any machine that has such an instruction. If you
8907 are trying to port GCC to such a machine, include an instruction to
8908 perform a logical-and of the result with 1 in the pattern for the
8909 comparison operators and let us know at @email{gcc@@gcc.gnu.org}.
8911 Often, a machine will have multiple instructions that obtain a value
8912 from a comparison (or the condition codes). Here are rules to guide the
8913 choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions
8918 Use the shortest sequence that yields a valid definition for
8919 @code{STORE_FLAG_VALUE}. It is more efficient for the compiler to
8920 ``normalize'' the value (convert it to, e.g., 1 or 0) than for the
8921 comparison operators to do so because there may be opportunities to
8922 combine the normalization with other operations.
8925 For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being
8926 slightly preferred on machines with expensive jumps and 1 preferred on
8930 As a second choice, choose a value of @samp{0x80000001} if instructions
8931 exist that set both the sign and low-order bits but do not define the
8935 Otherwise, use a value of @samp{0x80000000}.
8938 Many machines can produce both the value chosen for
8939 @code{STORE_FLAG_VALUE} and its negation in the same number of
8940 instructions. On those machines, you should also define a pattern for
8941 those cases, e.g., one matching
8944 (set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C})))
8947 Some machines can also perform @code{and} or @code{plus} operations on
8948 condition code values with less instructions than the corresponding
8949 @samp{s@var{cond}} insn followed by @code{and} or @code{plus}. On those
8950 machines, define the appropriate patterns. Use the names @code{incscc}
8951 and @code{decscc}, respectively, for the patterns which perform
8952 @code{plus} or @code{minus} operations on condition code values. See
8953 @file{rs6000.md} for some examples. The GNU Superoptizer can be used to
8954 find such instruction sequences on other machines.
8956 If this macro is not defined, the default value, 1, is used. You need
8957 not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
8958 instructions, or if the value generated by these instructions is 1.
8961 @defmac FLOAT_STORE_FLAG_VALUE (@var{mode})
8962 A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is
8963 returned when comparison operators with floating-point results are true.
8964 Define this macro on machines that have comparison operations that return
8965 floating-point values. If there are no such operations, do not define
8969 @defmac VECTOR_STORE_FLAG_VALUE (@var{mode})
8970 A C expression that gives a rtx representing the nonzero true element
8971 for vector comparisons. The returned rtx should be valid for the inner
8972 mode of @var{mode} which is guaranteed to be a vector mode. Define
8973 this macro on machines that have vector comparison operations that
8974 return a vector result. If there are no such operations, do not define
8975 this macro. Typically, this macro is defined as @code{const1_rtx} or
8976 @code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent
8977 the compiler optimizing such vector comparison operations for the
8981 @defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
8982 @defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
8983 A C expression that evaluates to true if the architecture defines a value
8984 for @code{clz} or @code{ctz} with a zero operand. If so, @var{value}
8985 should be set to this value. If this macro is not defined, the value of
8986 @code{clz} or @code{ctz} is assumed to be undefined.
8988 This macro must be defined if the target's expansion for @code{ffs}
8989 relies on a particular value to get correct results. Otherwise it
8990 is not necessary, though it may be used to optimize some corner cases.
8992 Note that regardless of this macro the ``definedness'' of @code{clz}
8993 and @code{ctz} at zero do @emph{not} extend to the builtin functions
8994 visible to the user. Thus one may be free to adjust the value at will
8995 to match the target expansion of these operations without fear of
9000 An alias for the machine mode for pointers. On most machines, define
9001 this to be the integer mode corresponding to the width of a hardware
9002 pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines.
9003 On some machines you must define this to be one of the partial integer
9004 modes, such as @code{PSImode}.
9006 The width of @code{Pmode} must be at least as large as the value of
9007 @code{POINTER_SIZE}. If it is not equal, you must define the macro
9008 @code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended
9012 @defmac FUNCTION_MODE
9013 An alias for the machine mode used for memory references to functions
9014 being called, in @code{call} RTL expressions. On most machines this
9015 should be @code{QImode}.
9018 @defmac STDC_0_IN_SYSTEM_HEADERS
9019 In normal operation, the preprocessor expands @code{__STDC__} to the
9020 constant 1, to signify that GCC conforms to ISO Standard C@. On some
9021 hosts, like Solaris, the system compiler uses a different convention,
9022 where @code{__STDC__} is normally 0, but is 1 if the user specifies
9023 strict conformance to the C Standard.
9025 Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host
9026 convention when processing system header files, but when processing user
9027 files @code{__STDC__} will always expand to 1.
9030 @defmac NO_IMPLICIT_EXTERN_C
9031 Define this macro if the system header files support C++ as well as C@.
9032 This macro inhibits the usual method of using system header files in
9033 C++, which is to pretend that the file's contents are enclosed in
9034 @samp{extern "C" @{@dots{}@}}.
9039 @defmac REGISTER_TARGET_PRAGMAS ()
9040 Define this macro if you want to implement any target-specific pragmas.
9041 If defined, it is a C expression which makes a series of calls to
9042 @code{c_register_pragma} or @code{c_register_pragma_with_expansion}
9043 for each pragma. The macro may also do any
9044 setup required for the pragmas.
9046 The primary reason to define this macro is to provide compatibility with
9047 other compilers for the same target. In general, we discourage
9048 definition of target-specific pragmas for GCC@.
9050 If the pragma can be implemented by attributes then you should consider
9051 defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well.
9053 Preprocessor macros that appear on pragma lines are not expanded. All
9054 @samp{#pragma} directives that do not match any registered pragma are
9055 silently ignored, unless the user specifies @option{-Wunknown-pragmas}.
9058 @deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
9059 @deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
9061 Each call to @code{c_register_pragma} or
9062 @code{c_register_pragma_with_expansion} establishes one pragma. The
9063 @var{callback} routine will be called when the preprocessor encounters a
9067 #pragma [@var{space}] @var{name} @dots{}
9070 @var{space} is the case-sensitive namespace of the pragma, or
9071 @code{NULL} to put the pragma in the global namespace. The callback
9072 routine receives @var{pfile} as its first argument, which can be passed
9073 on to cpplib's functions if necessary. You can lex tokens after the
9074 @var{name} by calling @code{c_lex}. Tokens that are not read by the
9075 callback will be silently ignored. The end of the line is indicated by
9076 a token of type @code{CPP_EOF}. Macro expansion occurs on the
9077 arguments of pragmas registered with
9078 @code{c_register_pragma_with_expansion} but not on the arguments of
9079 pragmas registered with @code{c_register_pragma}.
9081 For an example use of this routine, see @file{c4x.h} and the callback
9082 routines defined in @file{c4x-c.c}.
9084 Note that the use of @code{c_lex} is specific to the C and C++
9085 compilers. It will not work in the Java or Fortran compilers, or any
9086 other language compilers for that matter. Thus if @code{c_lex} is going
9087 to be called from target-specific code, it must only be done so when
9088 building the C and C++ compilers. This can be done by defining the
9089 variables @code{c_target_objs} and @code{cxx_target_objs} in the
9090 target entry in the @file{config.gcc} file. These variables should name
9091 the target-specific, language-specific object file which contains the
9092 code that uses @code{c_lex}. Note it will also be necessary to add a
9093 rule to the makefile fragment pointed to by @code{tmake_file} that shows
9094 how to build this object file.
9099 @defmac HANDLE_SYSV_PRAGMA
9100 Define this macro (to a value of 1) if you want the System V style
9101 pragmas @samp{#pragma pack(<n>)} and @samp{#pragma weak <name>
9102 [=<value>]} to be supported by gcc.
9104 The pack pragma specifies the maximum alignment (in bytes) of fields
9105 within a structure, in much the same way as the @samp{__aligned__} and
9106 @samp{__packed__} @code{__attribute__}s do. A pack value of zero resets
9107 the behavior to the default.
9109 A subtlety for Microsoft Visual C/C++ style bit-field packing
9110 (e.g.@: -mms-bitfields) for targets that support it:
9111 When a bit-field is inserted into a packed record, the whole size
9112 of the underlying type is used by one or more same-size adjacent
9113 bit-fields (that is, if its long:3, 32 bits is used in the record,
9114 and any additional adjacent long bit-fields are packed into the same
9115 chunk of 32 bits. However, if the size changes, a new field of that
9118 If both MS bit-fields and @samp{__attribute__((packed))} are used,
9119 the latter will take precedence. If @samp{__attribute__((packed))} is
9120 used on a single field when MS bit-fields are in use, it will take
9121 precedence for that field, but the alignment of the rest of the structure
9122 may affect its placement.
9124 The weak pragma only works if @code{SUPPORTS_WEAK} and
9125 @code{ASM_WEAKEN_LABEL} are defined. If enabled it allows the creation
9126 of specifically named weak labels, optionally with a value.
9131 @defmac HANDLE_PRAGMA_PACK_PUSH_POP
9132 Define this macro (to a value of 1) if you want to support the Win32
9133 style pragmas @samp{#pragma pack(push[,@var{n}])} and @samp{#pragma
9134 pack(pop)}. The @samp{pack(push,[@var{n}])} pragma specifies the maximum
9135 alignment (in bytes) of fields within a structure, in much the same way as
9136 the @samp{__aligned__} and @samp{__packed__} @code{__attribute__}s do. A
9137 pack value of zero resets the behavior to the default. Successive
9138 invocations of this pragma cause the previous values to be stacked, so
9139 that invocations of @samp{#pragma pack(pop)} will return to the previous
9143 @defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION
9144 Define this macro, as well as
9145 @code{HANDLE_SYSV_PRAGMA}, if macros should be expanded in the
9146 arguments of @samp{#pragma pack}.
9149 @defmac TARGET_DEFAULT_PACK_STRUCT
9150 If your target requires a structure packing default other than 0 (meaning
9151 the machine default), define this macro to the necessary value (in bytes).
9152 This must be a value that would also valid to be used with
9153 @samp{#pragma pack()} (that is, a small power of two).
9156 @defmac DOLLARS_IN_IDENTIFIERS
9157 Define this macro to control use of the character @samp{$} in
9158 identifier names for the C family of languages. 0 means @samp{$} is
9159 not allowed by default; 1 means it is allowed. 1 is the default;
9160 there is no need to define this macro in that case.
9163 @defmac NO_DOLLAR_IN_LABEL
9164 Define this macro if the assembler does not accept the character
9165 @samp{$} in label names. By default constructors and destructors in
9166 G++ have @samp{$} in the identifiers. If this macro is defined,
9167 @samp{.} is used instead.
9170 @defmac NO_DOT_IN_LABEL
9171 Define this macro if the assembler does not accept the character
9172 @samp{.} in label names. By default constructors and destructors in G++
9173 have names that use @samp{.}. If this macro is defined, these names
9174 are rewritten to avoid @samp{.}.
9177 @defmac INSN_SETS_ARE_DELAYED (@var{insn})
9178 Define this macro as a C expression that is nonzero if it is safe for the
9179 delay slot scheduler to place instructions in the delay slot of @var{insn},
9180 even if they appear to use a resource set or clobbered in @var{insn}.
9181 @var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that
9182 every @code{call_insn} has this behavior. On machines where some @code{insn}
9183 or @code{jump_insn} is really a function call and hence has this behavior,
9184 you should define this macro.
9186 You need not define this macro if it would always return zero.
9189 @defmac INSN_REFERENCES_ARE_DELAYED (@var{insn})
9190 Define this macro as a C expression that is nonzero if it is safe for the
9191 delay slot scheduler to place instructions in the delay slot of @var{insn},
9192 even if they appear to set or clobber a resource referenced in @var{insn}.
9193 @var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where
9194 some @code{insn} or @code{jump_insn} is really a function call and its operands
9195 are registers whose use is actually in the subroutine it calls, you should
9196 define this macro. Doing so allows the delay slot scheduler to move
9197 instructions which copy arguments into the argument registers into the delay
9200 You need not define this macro if it would always return zero.
9203 @defmac MULTIPLE_SYMBOL_SPACES
9204 Define this macro as a C expression that is nonzero if, in some cases,
9205 global symbols from one translation unit may not be bound to undefined
9206 symbols in another translation unit without user intervention. For
9207 instance, under Microsoft Windows symbols must be explicitly imported
9208 from shared libraries (DLLs).
9210 You need not define this macro if it would always evaluate to zero.
9213 @deftypefn {Target Hook} tree TARGET_MD_ASM_CLOBBERS (tree @var{outputs}, tree @var{inputs}, tree @var{clobbers})
9214 This target hook should add to @var{clobbers} @code{STRING_CST} trees for
9215 any hard regs the port wishes to automatically clobber for an asm.
9216 It should return the result of the last @code{tree_cons} used to add a
9217 clobber. The @var{outputs}, @var{inputs} and @var{clobber} lists are the
9218 corresponding parameters to the asm and may be inspected to avoid
9219 clobbering a register that is an input or output of the asm. You can use
9220 @code{decl_overlaps_hard_reg_set_p}, declared in @file{tree.h}, to test
9221 for overlap with regards to asm-declared registers.
9224 @defmac MATH_LIBRARY
9225 Define this macro as a C string constant for the linker argument to link
9226 in the system math library, or @samp{""} if the target does not have a
9227 separate math library.
9229 You need only define this macro if the default of @samp{"-lm"} is wrong.
9232 @defmac LIBRARY_PATH_ENV
9233 Define this macro as a C string constant for the environment variable that
9234 specifies where the linker should look for libraries.
9236 You need only define this macro if the default of @samp{"LIBRARY_PATH"}
9240 @defmac TARGET_POSIX_IO
9241 Define this macro if the target supports the following POSIX@ file
9242 functions, access, mkdir and file locking with fcntl / F_SETLKW@.
9243 Defining @code{TARGET_POSIX_IO} will enable the test coverage code
9244 to use file locking when exiting a program, which avoids race conditions
9245 if the program has forked. It will also create directories at run-time
9246 for cross-profiling.
9249 @defmac MAX_CONDITIONAL_EXECUTE
9251 A C expression for the maximum number of instructions to execute via
9252 conditional execution instructions instead of a branch. A value of
9253 @code{BRANCH_COST}+1 is the default if the machine does not use cc0, and
9254 1 if it does use cc0.
9257 @defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr})
9258 Used if the target needs to perform machine-dependent modifications on the
9259 conditionals used for turning basic blocks into conditionally executed code.
9260 @var{ce_info} points to a data structure, @code{struct ce_if_block}, which
9261 contains information about the currently processed blocks. @var{true_expr}
9262 and @var{false_expr} are the tests that are used for converting the
9263 then-block and the else-block, respectively. Set either @var{true_expr} or
9264 @var{false_expr} to a null pointer if the tests cannot be converted.
9267 @defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr})
9268 Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated
9269 if-statements into conditions combined by @code{and} and @code{or} operations.
9270 @var{bb} contains the basic block that contains the test that is currently
9271 being processed and about to be turned into a condition.
9274 @defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn})
9275 A C expression to modify the @var{PATTERN} of an @var{INSN} that is to
9276 be converted to conditional execution format. @var{ce_info} points to
9277 a data structure, @code{struct ce_if_block}, which contains information
9278 about the currently processed blocks.
9281 @defmac IFCVT_MODIFY_FINAL (@var{ce_info})
9282 A C expression to perform any final machine dependent modifications in
9283 converting code to conditional execution. The involved basic blocks
9284 can be found in the @code{struct ce_if_block} structure that is pointed
9285 to by @var{ce_info}.
9288 @defmac IFCVT_MODIFY_CANCEL (@var{ce_info})
9289 A C expression to cancel any machine dependent modifications in
9290 converting code to conditional execution. The involved basic blocks
9291 can be found in the @code{struct ce_if_block} structure that is pointed
9292 to by @var{ce_info}.
9295 @defmac IFCVT_INIT_EXTRA_FIELDS (@var{ce_info})
9296 A C expression to initialize any extra fields in a @code{struct ce_if_block}
9297 structure, which are defined by the @code{IFCVT_EXTRA_FIELDS} macro.
9300 @defmac IFCVT_EXTRA_FIELDS
9301 If defined, it should expand to a set of field declarations that will be
9302 added to the @code{struct ce_if_block} structure. These should be initialized
9303 by the @code{IFCVT_INIT_EXTRA_FIELDS} macro.
9306 @deftypefn {Target Hook} void TARGET_MACHINE_DEPENDENT_REORG ()
9307 If non-null, this hook performs a target-specific pass over the
9308 instruction stream. The compiler will run it at all optimization levels,
9309 just before the point at which it normally does delayed-branch scheduling.
9311 The exact purpose of the hook varies from target to target. Some use
9312 it to do transformations that are necessary for correctness, such as
9313 laying out in-function constant pools or avoiding hardware hazards.
9314 Others use it as an opportunity to do some machine-dependent optimizations.
9316 You need not implement the hook if it has nothing to do. The default
9320 @deftypefn {Target Hook} void TARGET_INIT_BUILTINS ()
9321 Define this hook if you have any machine-specific built-in functions
9322 that need to be defined. It should be a function that performs the
9325 Machine specific built-in functions can be useful to expand special machine
9326 instructions that would otherwise not normally be generated because
9327 they have no equivalent in the source language (for example, SIMD vector
9328 instructions or prefetch instructions).
9330 To create a built-in function, call the function
9331 @code{lang_hooks.builtin_function}
9332 which is defined by the language front end. You can use any type nodes set
9333 up by @code{build_common_tree_nodes} and @code{build_common_tree_nodes_2};
9334 only language front ends that use those two functions will call
9335 @samp{TARGET_INIT_BUILTINS}.
9338 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN (tree @var{exp}, rtx @var{target}, rtx @var{subtarget}, enum machine_mode @var{mode}, int @var{ignore})
9340 Expand a call to a machine specific built-in function that was set up by
9341 @samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the
9342 function call; the result should go to @var{target} if that is
9343 convenient, and have mode @var{mode} if that is convenient.
9344 @var{subtarget} may be used as the target for computing one of
9345 @var{exp}'s operands. @var{ignore} is nonzero if the value is to be
9346 ignored. This function should return the result of the call to the
9350 @deftypefn {Target Hook} tree TARGET_RESOLVE_OVERLOADED_BUILTIN (tree @var{fndecl}, tree @var{arglist})
9352 Select a replacement for a machine specific built-in function that
9353 was set up by @samp{TARGET_INIT_BUILTINS}. This is done
9354 @emph{before} regular type checking, and so allows the target to
9355 implement a crude form of function overloading. @var{fndecl} is the
9356 declaration of the built-in function. @var{arglist} is the list of
9357 arguments passed to the built-in function. The result is a
9358 complete expression that implements the operation, usually
9359 another @code{CALL_EXPR}.
9362 @deftypefn {Target Hook} tree TARGET_FOLD_BUILTIN (tree @var{fndecl}, tree @var{arglist}, bool @var{ignore})
9364 Fold a call to a machine specific built-in function that was set up by
9365 @samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the
9366 built-in function. @var{arglist} is the list of arguments passed to
9367 the built-in function. The result is another tree containing a
9368 simplified expression for the call's result. If @var{ignore} is true
9369 the value will be ignored.
9372 @deftypefn {Target Hook} const char * TARGET_INVALID_WITHIN_DOLOOP (rtx @var{insn})
9374 Take an instruction in @var{insn} and return NULL if it is valid within a
9375 low-overhead loop, otherwise return a string why doloop could not be applied.
9377 Many targets use special registers for low-overhead looping. For any
9378 instruction that clobbers these this function should return a string indicating
9379 the reason why the doloop could not be applied.
9380 By default, the RTL loop optimizer does not use a present doloop pattern for
9381 loops containing function calls or branch on table instructions.
9384 @defmac MD_CAN_REDIRECT_BRANCH (@var{branch1}, @var{branch2})
9386 Take a branch insn in @var{branch1} and another in @var{branch2}.
9387 Return true if redirecting @var{branch1} to the destination of
9388 @var{branch2} is possible.
9390 On some targets, branches may have a limited range. Optimizing the
9391 filling of delay slots can result in branches being redirected, and this
9392 may in turn cause a branch offset to overflow.
9395 @deftypefn {Target Hook} bool TARGET_COMMUTATIVE_P (rtx @var{x}, @var{outer_code})
9396 This target hook returns @code{true} if @var{x} is considered to be commutative.
9397 Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider
9398 PLUS to be commutative inside a MEM. @var{outer_code} is the rtx code
9399 of the enclosing rtl, if known, otherwise it is UNKNOWN.
9402 @deftypefn {Target Hook} rtx TARGET_ALLOCATE_INITIAL_VALUE (rtx @var{hard_reg})
9404 When the initial value of a hard register has been copied in a pseudo
9405 register, it is often not necessary to actually allocate another register
9406 to this pseudo register, because the original hard register or a stack slot
9407 it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE}
9408 is called at the start of register allocation once for each hard register
9409 that had its initial value copied by using
9410 @code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}.
9411 Possible values are @code{NULL_RTX}, if you don't want
9412 to do any special allocation, a @code{REG} rtx---that would typically be
9413 the hard register itself, if it is known not to be clobbered---or a
9415 If you are returning a @code{MEM}, this is only a hint for the allocator;
9416 it might decide to use another register anyways.
9417 You may use @code{current_function_leaf_function} in the hook, functions
9418 that use @code{REG_N_SETS}, to determine if the hard
9419 register in question will not be clobbered.
9420 The default value of this hook is @code{NULL}, which disables any special
9424 @defmac TARGET_OBJECT_SUFFIX
9425 Define this macro to be a C string representing the suffix for object
9426 files on your target machine. If you do not define this macro, GCC will
9427 use @samp{.o} as the suffix for object files.
9430 @defmac TARGET_EXECUTABLE_SUFFIX
9431 Define this macro to be a C string representing the suffix to be
9432 automatically added to executable files on your target machine. If you
9433 do not define this macro, GCC will use the null string as the suffix for
9437 @defmac COLLECT_EXPORT_LIST
9438 If defined, @code{collect2} will scan the individual object files
9439 specified on its command line and create an export list for the linker.
9440 Define this macro for systems like AIX, where the linker discards
9441 object files that are not referenced from @code{main} and uses export
9445 @defmac MODIFY_JNI_METHOD_CALL (@var{mdecl})
9446 Define this macro to a C expression representing a variant of the
9447 method call @var{mdecl}, if Java Native Interface (JNI) methods
9448 must be invoked differently from other methods on your target.
9449 For example, on 32-bit Microsoft Windows, JNI methods must be invoked using
9450 the @code{stdcall} calling convention and this macro is then
9451 defined as this expression:
9454 build_type_attribute_variant (@var{mdecl},
9456 (get_identifier ("stdcall"),
9461 @deftypefn {Target Hook} bool TARGET_CANNOT_MODIFY_JUMPS_P (void)
9462 This target hook returns @code{true} past the point in which new jump
9463 instructions could be created. On machines that require a register for
9464 every jump such as the SHmedia ISA of SH5, this point would typically be
9465 reload, so this target hook should be defined to a function such as:
9469 cannot_modify_jumps_past_reload_p ()
9471 return (reload_completed || reload_in_progress);
9476 @deftypefn {Target Hook} int TARGET_BRANCH_TARGET_REGISTER_CLASS (void)
9477 This target hook returns a register class for which branch target register
9478 optimizations should be applied. All registers in this class should be
9479 usable interchangeably. After reload, registers in this class will be
9480 re-allocated and loads will be hoisted out of loops and be subjected
9481 to inter-block scheduling.
9484 @deftypefn {Target Hook} bool TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED (bool @var{after_prologue_epilogue_gen})
9485 Branch target register optimization will by default exclude callee-saved
9487 that are not already live during the current function; if this target hook
9488 returns true, they will be included. The target code must than make sure
9489 that all target registers in the class returned by
9490 @samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are
9491 saved. @var{after_prologue_epilogue_gen} indicates if prologues and
9492 epilogues have already been generated. Note, even if you only return
9493 true when @var{after_prologue_epilogue_gen} is false, you still are likely
9494 to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET}
9495 to reserve space for caller-saved target registers.
9498 @defmac POWI_MAX_MULTS
9499 If defined, this macro is interpreted as a signed integer C expression
9500 that specifies the maximum number of floating point multiplications
9501 that should be emitted when expanding exponentiation by an integer
9502 constant inline. When this value is defined, exponentiation requiring
9503 more than this number of multiplications is implemented by calling the
9504 system library's @code{pow}, @code{powf} or @code{powl} routines.
9505 The default value places no upper bound on the multiplication count.
9508 @deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
9509 This target hook should register any extra include files for the
9510 target. The parameter @var{stdinc} indicates if normal include files
9511 are present. The parameter @var{sysroot} is the system root directory.
9512 The parameter @var{iprefix} is the prefix for the gcc directory.
9515 @deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
9516 This target hook should register any extra include files for the
9517 target before any standard headers. The parameter @var{stdinc}
9518 indicates if normal include files are present. The parameter
9519 @var{sysroot} is the system root directory. The parameter
9520 @var{iprefix} is the prefix for the gcc directory.
9523 @deftypefn Macro void TARGET_OPTF (char *@var{path})
9524 This target hook should register special include paths for the target.
9525 The parameter @var{path} is the include to register. On Darwin
9526 systems, this is used for Framework includes, which have semantics
9527 that are different from @option{-I}.
9530 @deftypefn {Target Hook} bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl})
9531 This target hook returns @code{true} if it is safe to use a local alias
9532 for a virtual function @var{fndecl} when constructing thunks,
9533 @code{false} otherwise. By default, the hook returns @code{true} for all
9534 functions, if a target supports aliases (i.e.@: defines
9535 @code{ASM_OUTPUT_DEF}), @code{false} otherwise,
9538 @defmac TARGET_FORMAT_TYPES
9539 If defined, this macro is the name of a global variable containing
9540 target-specific format checking information for the @option{-Wformat}
9541 option. The default is to have no target-specific format checks.
9544 @defmac TARGET_N_FORMAT_TYPES
9545 If defined, this macro is the number of entries in
9546 @code{TARGET_FORMAT_TYPES}.
9549 @deftypefn {Target Hook} bool TARGET_RELAXED_ORDERING
9550 If set to @code{true}, means that the target's memory model does not
9551 guarantee that loads which do not depend on one another will access
9552 main memory in the order of the instruction stream; if ordering is
9553 important, an explicit memory barrier must be used. This is true of
9554 many recent processors which implement a policy of ``relaxed,''
9555 ``weak,'' or ``release'' memory consistency, such as Alpha, PowerPC,
9556 and ia64. The default is @code{false}.
9559 @deftypefn {Target Hook} const char *TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN (tree @var{typelist}, tree @var{funcdecl}, tree @var{val})
9560 If defined, this macro returns the diagnostic message when it is
9561 illegal to pass argument @var{val} to function @var{funcdecl}
9562 with prototype @var{typelist}.
9565 @deftypefn {Target Hook} {const char *} TARGET_INVALID_CONVERSION (tree @var{fromtype}, tree @var{totype})
9566 If defined, this macro returns the diagnostic message when it is
9567 invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL}
9568 if validity should be determined by the front end.
9571 @deftypefn {Target Hook} {const char *} TARGET_INVALID_UNARY_OP (int @var{op}, tree @var{type})
9572 If defined, this macro returns the diagnostic message when it is
9573 invalid to apply operation @var{op} (where unary plus is denoted by
9574 @code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL}
9575 if validity should be determined by the front end.
9578 @deftypefn {Target Hook} {const char *} TARGET_INVALID_BINARY_OP (int @var{op}, tree @var{type1}, tree @var{type2})
9579 If defined, this macro returns the diagnostic message when it is
9580 invalid to apply operation @var{op} to operands of types @var{type1}
9581 and @var{type2}, or @code{NULL} if validity should be determined by
9585 @defmac TARGET_USE_JCR_SECTION
9586 This macro determines whether to use the JCR section to register Java
9587 classes. By default, TARGET_USE_JCR_SECTION is defined to 1 if both
9588 SUPPORTS_WEAK and TARGET_HAVE_NAMED_SECTIONS are true, else 0.