1 @c Copyright (C) 1988,1989,1992,1993,1994,1995,1996,1997,1998,1999,2000,2001,
2 @c 2002, 2003, 2004, 2005, 2006, 2007, 2008 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 * Old Constraints:: The old way to define machine-specific constraints.
37 * Stack and Calling:: Defining which way the stack grows and by how much.
38 * Varargs:: Defining the varargs macros.
39 * Trampolines:: Code set up at run time to enter a nested function.
40 * Library Calls:: Controlling how library routines are implicitly called.
41 * Addressing Modes:: Defining addressing modes valid for memory operands.
42 * Anchored Addresses:: Defining how @option{-fsection-anchors} should work.
43 * Condition Code:: Defining how insns update the condition code.
44 * Costs:: Defining relative costs of different operations.
45 * Scheduling:: Adjusting the behavior of the instruction scheduler.
46 * Sections:: Dividing storage into text, data, and other sections.
47 * PIC:: Macros for position independent code.
48 * Assembler Format:: Defining how to write insns and pseudo-ops to output.
49 * Debugging Info:: Defining the format of debugging output.
50 * Floating Point:: Handling floating point for cross-compilers.
51 * Mode Switching:: Insertion of mode-switching instructions.
52 * Target Attributes:: Defining target-specific uses of @code{__attribute__}.
53 * Emulated TLS:: Emulated TLS support.
54 * MIPS Coprocessors:: MIPS coprocessor support and how to customize it.
55 * PCH Target:: Validity checking for precompiled headers.
56 * C++ ABI:: Controlling C++ ABI changes.
57 * Misc:: Everything else.
60 @node Target Structure
61 @section The Global @code{targetm} Variable
63 @cindex target functions
65 @deftypevar {struct gcc_target} targetm
66 The target @file{.c} file must define the global @code{targetm} variable
67 which contains pointers to functions and data relating to the target
68 machine. The variable is declared in @file{target.h};
69 @file{target-def.h} defines the macro @code{TARGET_INITIALIZER} which is
70 used to initialize the variable, and macros for the default initializers
71 for elements of the structure. The @file{.c} file should override those
72 macros for which the default definition is inappropriate. For example:
75 #include "target-def.h"
77 /* @r{Initialize the GCC target structure.} */
79 #undef TARGET_COMP_TYPE_ATTRIBUTES
80 #define TARGET_COMP_TYPE_ATTRIBUTES @var{machine}_comp_type_attributes
82 struct gcc_target targetm = TARGET_INITIALIZER;
86 Where a macro should be defined in the @file{.c} file in this manner to
87 form part of the @code{targetm} structure, it is documented below as a
88 ``Target Hook'' with a prototype. Many macros will change in future
89 from being defined in the @file{.h} file to being part of the
90 @code{targetm} structure.
93 @section Controlling the Compilation Driver, @file{gcc}
95 @cindex controlling the compilation driver
97 @c prevent bad page break with this line
98 You can control the compilation driver.
100 @defmac SWITCH_TAKES_ARG (@var{char})
101 A C expression which determines whether the option @option{-@var{char}}
102 takes arguments. The value should be the number of arguments that
103 option takes--zero, for many options.
105 By default, this macro is defined as
106 @code{DEFAULT_SWITCH_TAKES_ARG}, which handles the standard options
107 properly. You need not define @code{SWITCH_TAKES_ARG} unless you
108 wish to add additional options which take arguments. Any redefinition
109 should call @code{DEFAULT_SWITCH_TAKES_ARG} and then check for
113 @defmac WORD_SWITCH_TAKES_ARG (@var{name})
114 A C expression which determines whether the option @option{-@var{name}}
115 takes arguments. The value should be the number of arguments that
116 option takes--zero, for many options. This macro rather than
117 @code{SWITCH_TAKES_ARG} is used for multi-character option names.
119 By default, this macro is defined as
120 @code{DEFAULT_WORD_SWITCH_TAKES_ARG}, which handles the standard options
121 properly. You need not define @code{WORD_SWITCH_TAKES_ARG} unless you
122 wish to add additional options which take arguments. Any redefinition
123 should call @code{DEFAULT_WORD_SWITCH_TAKES_ARG} and then check for
127 @defmac SWITCH_CURTAILS_COMPILATION (@var{char})
128 A C expression which determines whether the option @option{-@var{char}}
129 stops compilation before the generation of an executable. The value is
130 boolean, nonzero if the option does stop an executable from being
131 generated, zero otherwise.
133 By default, this macro is defined as
134 @code{DEFAULT_SWITCH_CURTAILS_COMPILATION}, which handles the standard
135 options properly. You need not define
136 @code{SWITCH_CURTAILS_COMPILATION} unless you wish to add additional
137 options which affect the generation of an executable. Any redefinition
138 should call @code{DEFAULT_SWITCH_CURTAILS_COMPILATION} and then check
139 for additional options.
142 @defmac SWITCHES_NEED_SPACES
143 A string-valued C expression which enumerates the options for which
144 the linker needs a space between the option and its argument.
146 If this macro is not defined, the default value is @code{""}.
149 @defmac TARGET_OPTION_TRANSLATE_TABLE
150 If defined, a list of pairs of strings, the first of which is a
151 potential command line target to the @file{gcc} driver program, and the
152 second of which is a space-separated (tabs and other whitespace are not
153 supported) list of options with which to replace the first option. The
154 target defining this list is responsible for assuring that the results
155 are valid. Replacement options may not be the @code{--opt} style, they
156 must be the @code{-opt} style. It is the intention of this macro to
157 provide a mechanism for substitution that affects the multilibs chosen,
158 such as one option that enables many options, some of which select
159 multilibs. Example nonsensical definition, where @option{-malt-abi},
160 @option{-EB}, and @option{-mspoo} cause different multilibs to be chosen:
163 #define TARGET_OPTION_TRANSLATE_TABLE \
164 @{ "-fast", "-march=fast-foo -malt-abi -I/usr/fast-foo" @}, \
165 @{ "-compat", "-EB -malign=4 -mspoo" @}
169 @defmac DRIVER_SELF_SPECS
170 A list of specs for the driver itself. It should be a suitable
171 initializer for an array of strings, with no surrounding braces.
173 The driver applies these specs to its own command line between loading
174 default @file{specs} files (but not command-line specified ones) and
175 choosing the multilib directory or running any subcommands. It
176 applies them in the order given, so each spec can depend on the
177 options added by earlier ones. It is also possible to remove options
178 using @samp{%<@var{option}} in the usual way.
180 This macro can be useful when a port has several interdependent target
181 options. It provides a way of standardizing the command line so
182 that the other specs are easier to write.
184 Do not define this macro if it does not need to do anything.
187 @defmac OPTION_DEFAULT_SPECS
188 A list of specs used to support configure-time default options (i.e.@:
189 @option{--with} options) in the driver. It should be a suitable initializer
190 for an array of structures, each containing two strings, without the
191 outermost pair of surrounding braces.
193 The first item in the pair is the name of the default. This must match
194 the code in @file{config.gcc} for the target. The second item is a spec
195 to apply if a default with this name was specified. The string
196 @samp{%(VALUE)} in the spec will be replaced by the value of the default
197 everywhere it occurs.
199 The driver will apply these specs to its own command line between loading
200 default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using
201 the same mechanism as @code{DRIVER_SELF_SPECS}.
203 Do not define this macro if it does not need to do anything.
207 A C string constant that tells the GCC driver program options to
208 pass to CPP@. It can also specify how to translate options you
209 give to GCC into options for GCC to pass to the CPP@.
211 Do not define this macro if it does not need to do anything.
214 @defmac CPLUSPLUS_CPP_SPEC
215 This macro is just like @code{CPP_SPEC}, but is used for C++, rather
216 than C@. If you do not define this macro, then the value of
217 @code{CPP_SPEC} (if any) will be used instead.
221 A C string constant that tells the GCC driver program options to
222 pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language
224 It can also specify how to translate options you give to GCC into options
225 for GCC to pass to front ends.
227 Do not define this macro if it does not need to do anything.
231 A C string constant that tells the GCC driver program options to
232 pass to @code{cc1plus}. It can also specify how to translate options you
233 give to GCC into options for GCC to pass to the @code{cc1plus}.
235 Do not define this macro if it does not need to do anything.
236 Note that everything defined in CC1_SPEC is already passed to
237 @code{cc1plus} so there is no need to duplicate the contents of
238 CC1_SPEC in CC1PLUS_SPEC@.
242 A C string constant that tells the GCC driver program options to
243 pass to the assembler. It can also specify how to translate options
244 you give to GCC into options for GCC to pass to the assembler.
245 See the file @file{sun3.h} for an example of this.
247 Do not define this macro if it does not need to do anything.
250 @defmac ASM_FINAL_SPEC
251 A C string constant that tells the GCC driver program how to
252 run any programs which cleanup after the normal assembler.
253 Normally, this is not needed. See the file @file{mips.h} for
256 Do not define this macro if it does not need to do anything.
259 @defmac AS_NEEDS_DASH_FOR_PIPED_INPUT
260 Define this macro, with no value, if the driver should give the assembler
261 an argument consisting of a single dash, @option{-}, to instruct it to
262 read from its standard input (which will be a pipe connected to the
263 output of the compiler proper). This argument is given after any
264 @option{-o} option specifying the name of the output file.
266 If you do not define this macro, the assembler is assumed to read its
267 standard input if given no non-option arguments. If your assembler
268 cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct;
269 see @file{mips.h} for instance.
273 A C string constant that tells the GCC driver program options to
274 pass to the linker. It can also specify how to translate options you
275 give to GCC into options for GCC to pass to the linker.
277 Do not define this macro if it does not need to do anything.
281 Another C string constant used much like @code{LINK_SPEC}. The difference
282 between the two is that @code{LIB_SPEC} is used at the end of the
283 command given to the linker.
285 If this macro is not defined, a default is provided that
286 loads the standard C library from the usual place. See @file{gcc.c}.
290 Another C string constant that tells the GCC driver program
291 how and when to place a reference to @file{libgcc.a} into the
292 linker command line. This constant is placed both before and after
293 the value of @code{LIB_SPEC}.
295 If this macro is not defined, the GCC driver provides a default that
296 passes the string @option{-lgcc} to the linker.
299 @defmac REAL_LIBGCC_SPEC
300 By default, if @code{ENABLE_SHARED_LIBGCC} is defined, the
301 @code{LIBGCC_SPEC} is not directly used by the driver program but is
302 instead modified to refer to different versions of @file{libgcc.a}
303 depending on the values of the command line flags @option{-static},
304 @option{-shared}, @option{-static-libgcc}, and @option{-shared-libgcc}. On
305 targets where these modifications are inappropriate, define
306 @code{REAL_LIBGCC_SPEC} instead. @code{REAL_LIBGCC_SPEC} tells the
307 driver how to place a reference to @file{libgcc} on the link command
308 line, but, unlike @code{LIBGCC_SPEC}, it is used unmodified.
311 @defmac USE_LD_AS_NEEDED
312 A macro that controls the modifications to @code{LIBGCC_SPEC}
313 mentioned in @code{REAL_LIBGCC_SPEC}. If nonzero, a spec will be
314 generated that uses --as-needed and the shared libgcc in place of the
315 static exception handler library, when linking without any of
316 @code{-static}, @code{-static-libgcc}, or @code{-shared-libgcc}.
320 If defined, this C string constant is added to @code{LINK_SPEC}.
321 When @code{USE_LD_AS_NEEDED} is zero or undefined, it also affects
322 the modifications to @code{LIBGCC_SPEC} mentioned in
323 @code{REAL_LIBGCC_SPEC}.
326 @defmac STARTFILE_SPEC
327 Another C string constant used much like @code{LINK_SPEC}. The
328 difference between the two is that @code{STARTFILE_SPEC} is used at
329 the very beginning of the command given to the linker.
331 If this macro is not defined, a default is provided that loads the
332 standard C startup file from the usual place. See @file{gcc.c}.
336 Another C string constant used much like @code{LINK_SPEC}. The
337 difference between the two is that @code{ENDFILE_SPEC} is used at
338 the very end of the command given to the linker.
340 Do not define this macro if it does not need to do anything.
343 @defmac THREAD_MODEL_SPEC
344 GCC @code{-v} will print the thread model GCC was configured to use.
345 However, this doesn't work on platforms that are multilibbed on thread
346 models, such as AIX 4.3. On such platforms, define
347 @code{THREAD_MODEL_SPEC} such that it evaluates to a string without
348 blanks that names one of the recognized thread models. @code{%*}, the
349 default value of this macro, will expand to the value of
350 @code{thread_file} set in @file{config.gcc}.
353 @defmac SYSROOT_SUFFIX_SPEC
354 Define this macro to add a suffix to the target sysroot when GCC is
355 configured with a sysroot. This will cause GCC to search for usr/lib,
356 et al, within sysroot+suffix.
359 @defmac SYSROOT_HEADERS_SUFFIX_SPEC
360 Define this macro to add a headers_suffix to the target sysroot when
361 GCC is configured with a sysroot. This will cause GCC to pass the
362 updated sysroot+headers_suffix to CPP, causing it to search for
363 usr/include, et al, within sysroot+headers_suffix.
367 Define this macro to provide additional specifications to put in the
368 @file{specs} file that can be used in various specifications like
371 The definition should be an initializer for an array of structures,
372 containing a string constant, that defines the specification name, and a
373 string constant that provides the specification.
375 Do not define this macro if it does not need to do anything.
377 @code{EXTRA_SPECS} is useful when an architecture contains several
378 related targets, which have various @code{@dots{}_SPECS} which are similar
379 to each other, and the maintainer would like one central place to keep
382 For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to
383 define either @code{_CALL_SYSV} when the System V calling sequence is
384 used or @code{_CALL_AIX} when the older AIX-based calling sequence is
387 The @file{config/rs6000/rs6000.h} target file defines:
390 #define EXTRA_SPECS \
391 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @},
393 #define CPP_SYS_DEFAULT ""
396 The @file{config/rs6000/sysv.h} target file defines:
400 "%@{posix: -D_POSIX_SOURCE @} \
401 %@{mcall-sysv: -D_CALL_SYSV @} \
402 %@{!mcall-sysv: %(cpp_sysv_default) @} \
403 %@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}"
405 #undef CPP_SYSV_DEFAULT
406 #define CPP_SYSV_DEFAULT "-D_CALL_SYSV"
409 while the @file{config/rs6000/eabiaix.h} target file defines
410 @code{CPP_SYSV_DEFAULT} as:
413 #undef CPP_SYSV_DEFAULT
414 #define CPP_SYSV_DEFAULT "-D_CALL_AIX"
418 @defmac LINK_LIBGCC_SPECIAL_1
419 Define this macro if the driver program should find the library
420 @file{libgcc.a}. If you do not define this macro, the driver program will pass
421 the argument @option{-lgcc} to tell the linker to do the search.
424 @defmac LINK_GCC_C_SEQUENCE_SPEC
425 The sequence in which libgcc and libc are specified to the linker.
426 By default this is @code{%G %L %G}.
429 @defmac LINK_COMMAND_SPEC
430 A C string constant giving the complete command line need to execute the
431 linker. When you do this, you will need to update your port each time a
432 change is made to the link command line within @file{gcc.c}. Therefore,
433 define this macro only if you need to completely redefine the command
434 line for invoking the linker and there is no other way to accomplish
435 the effect you need. Overriding this macro may be avoidable by overriding
436 @code{LINK_GCC_C_SEQUENCE_SPEC} instead.
439 @defmac LINK_ELIMINATE_DUPLICATE_LDIRECTORIES
440 A nonzero value causes @command{collect2} to remove duplicate @option{-L@var{directory}} search
441 directories from linking commands. Do not give it a nonzero value if
442 removing duplicate search directories changes the linker's semantics.
445 @defmac MULTILIB_DEFAULTS
446 Define this macro as a C expression for the initializer of an array of
447 string to tell the driver program which options are defaults for this
448 target and thus do not need to be handled specially when using
449 @code{MULTILIB_OPTIONS}.
451 Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in
452 the target makefile fragment or if none of the options listed in
453 @code{MULTILIB_OPTIONS} are set by default.
454 @xref{Target Fragment}.
457 @defmac RELATIVE_PREFIX_NOT_LINKDIR
458 Define this macro to tell @command{gcc} that it should only translate
459 a @option{-B} prefix into a @option{-L} linker option if the prefix
460 indicates an absolute file name.
463 @defmac MD_EXEC_PREFIX
464 If defined, this macro is an additional prefix to try after
465 @code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched
466 when the @option{-b} option is used, or the compiler is built as a cross
467 compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it
468 to the list of directories used to find the assembler in @file{configure.in}.
471 @defmac STANDARD_STARTFILE_PREFIX
472 Define this macro as a C string constant if you wish to override the
473 standard choice of @code{libdir} as the default prefix to
474 try when searching for startup files such as @file{crt0.o}.
475 @code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler
476 is built as a cross compiler.
479 @defmac STANDARD_STARTFILE_PREFIX_1
480 Define this macro as a C string constant if you wish to override the
481 standard choice of @code{/lib} as a prefix to try after the default prefix
482 when searching for startup files such as @file{crt0.o}.
483 @code{STANDARD_STARTFILE_PREFIX_1} is not searched when the compiler
484 is built as a cross compiler.
487 @defmac STANDARD_STARTFILE_PREFIX_2
488 Define this macro as a C string constant if you wish to override the
489 standard choice of @code{/lib} as yet another prefix to try after the
490 default prefix when searching for startup files such as @file{crt0.o}.
491 @code{STANDARD_STARTFILE_PREFIX_2} is not searched when the compiler
492 is built as a cross compiler.
495 @defmac MD_STARTFILE_PREFIX
496 If defined, this macro supplies an additional prefix to try after the
497 standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the
498 @option{-b} option is used, or when the compiler is built as a cross
502 @defmac MD_STARTFILE_PREFIX_1
503 If defined, this macro supplies yet another prefix to try after the
504 standard prefixes. It is not searched when the @option{-b} option is
505 used, or when the compiler is built as a cross compiler.
508 @defmac INIT_ENVIRONMENT
509 Define this macro as a C string constant if you wish to set environment
510 variables for programs called by the driver, such as the assembler and
511 loader. The driver passes the value of this macro to @code{putenv} to
512 initialize the necessary environment variables.
515 @defmac LOCAL_INCLUDE_DIR
516 Define this macro as a C string constant if you wish to override the
517 standard choice of @file{/usr/local/include} as the default prefix to
518 try when searching for local header files. @code{LOCAL_INCLUDE_DIR}
519 comes before @code{SYSTEM_INCLUDE_DIR} in the search order.
521 Cross compilers do not search either @file{/usr/local/include} or its
525 @defmac MODIFY_TARGET_NAME
526 Define this macro if you wish to define command-line switches that
527 modify the default target name.
529 For each switch, you can include a string to be appended to the first
530 part of the configuration name or a string to be deleted from the
531 configuration name, if present. The definition should be an initializer
532 for an array of structures. Each array element should have three
533 elements: the switch name (a string constant, including the initial
534 dash), one of the enumeration codes @code{ADD} or @code{DELETE} to
535 indicate whether the string should be inserted or deleted, and the string
536 to be inserted or deleted (a string constant).
538 For example, on a machine where @samp{64} at the end of the
539 configuration name denotes a 64-bit target and you want the @option{-32}
540 and @option{-64} switches to select between 32- and 64-bit targets, you would
544 #define MODIFY_TARGET_NAME \
545 @{ @{ "-32", DELETE, "64"@}, \
546 @{"-64", ADD, "64"@}@}
550 @defmac SYSTEM_INCLUDE_DIR
551 Define this macro as a C string constant if you wish to specify a
552 system-specific directory to search for header files before the standard
553 directory. @code{SYSTEM_INCLUDE_DIR} comes before
554 @code{STANDARD_INCLUDE_DIR} in the search order.
556 Cross compilers do not use this macro and do not search the directory
560 @defmac STANDARD_INCLUDE_DIR
561 Define this macro as a C string constant if you wish to override the
562 standard choice of @file{/usr/include} as the default prefix to
563 try when searching for header files.
565 Cross compilers ignore this macro and do not search either
566 @file{/usr/include} or its replacement.
569 @defmac STANDARD_INCLUDE_COMPONENT
570 The ``component'' corresponding to @code{STANDARD_INCLUDE_DIR}.
571 See @code{INCLUDE_DEFAULTS}, below, for the description of components.
572 If you do not define this macro, no component is used.
575 @defmac INCLUDE_DEFAULTS
576 Define this macro if you wish to override the entire default search path
577 for include files. For a native compiler, the default search path
578 usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR},
579 @code{SYSTEM_INCLUDE_DIR}, @code{GPLUSPLUS_INCLUDE_DIR}, and
580 @code{STANDARD_INCLUDE_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR}
581 and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile},
582 and specify private search areas for GCC@. The directory
583 @code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs.
585 The definition should be an initializer for an array of structures.
586 Each array element should have four elements: the directory name (a
587 string constant), the component name (also a string constant), a flag
588 for C++-only directories,
589 and a flag showing that the includes in the directory don't need to be
590 wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of
591 the array with a null element.
593 The component name denotes what GNU package the include file is part of,
594 if any, in all uppercase letters. For example, it might be @samp{GCC}
595 or @samp{BINUTILS}. If the package is part of a vendor-supplied
596 operating system, code the component name as @samp{0}.
598 For example, here is the definition used for VAX/VMS:
601 #define INCLUDE_DEFAULTS \
603 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \
604 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \
605 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \
612 Here is the order of prefixes tried for exec files:
616 Any prefixes specified by the user with @option{-B}.
619 The environment variable @code{GCC_EXEC_PREFIX} or, if @code{GCC_EXEC_PREFIX}
620 is not set and the compiler has not been installed in the configure-time
621 @var{prefix}, the location in which the compiler has actually been installed.
624 The directories specified by the environment variable @code{COMPILER_PATH}.
627 The macro @code{STANDARD_EXEC_PREFIX}, if the compiler has been installed
628 in the configured-time @var{prefix}.
631 The location @file{/usr/libexec/gcc/}, but only if this is a native compiler.
634 The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
637 The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
641 Here is the order of prefixes tried for startfiles:
645 Any prefixes specified by the user with @option{-B}.
648 The environment variable @code{GCC_EXEC_PREFIX} or its automatically determined
649 value based on the installed toolchain location.
652 The directories specified by the environment variable @code{LIBRARY_PATH}
653 (or port-specific name; native only, cross compilers do not use this).
656 The macro @code{STANDARD_EXEC_PREFIX}, but only if the toolchain is installed
657 in the configured @var{prefix} or this is a native compiler.
660 The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
663 The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
667 The macro @code{MD_STARTFILE_PREFIX}, if defined, but only if this is a
668 native compiler, or we have a target system root.
671 The macro @code{MD_STARTFILE_PREFIX_1}, if defined, but only if this is a
672 native compiler, or we have a target system root.
675 The macro @code{STANDARD_STARTFILE_PREFIX}, with any sysroot modifications.
676 If this path is relative it will be prefixed by @code{GCC_EXEC_PREFIX} and
677 the machine suffix or @code{STANDARD_EXEC_PREFIX} and the machine suffix.
680 The macro @code{STANDARD_STARTFILE_PREFIX_1}, but only if this is a native
681 compiler, or we have a target system root. The default for this macro is
685 The macro @code{STANDARD_STARTFILE_PREFIX_2}, but only if this is a native
686 compiler, or we have a target system root. The default for this macro is
690 @node Run-time Target
691 @section Run-time Target Specification
692 @cindex run-time target specification
693 @cindex predefined macros
694 @cindex target specifications
696 @c prevent bad page break with this line
697 Here are run-time target specifications.
699 @defmac TARGET_CPU_CPP_BUILTINS ()
700 This function-like macro expands to a block of code that defines
701 built-in preprocessor macros and assertions for the target CPU, using
702 the functions @code{builtin_define}, @code{builtin_define_std} and
703 @code{builtin_assert}. When the front end
704 calls this macro it provides a trailing semicolon, and since it has
705 finished command line option processing your code can use those
708 @code{builtin_assert} takes a string in the form you pass to the
709 command-line option @option{-A}, such as @code{cpu=mips}, and creates
710 the assertion. @code{builtin_define} takes a string in the form
711 accepted by option @option{-D} and unconditionally defines the macro.
713 @code{builtin_define_std} takes a string representing the name of an
714 object-like macro. If it doesn't lie in the user's namespace,
715 @code{builtin_define_std} defines it unconditionally. Otherwise, it
716 defines a version with two leading underscores, and another version
717 with two leading and trailing underscores, and defines the original
718 only if an ISO standard was not requested on the command line. For
719 example, passing @code{unix} defines @code{__unix}, @code{__unix__}
720 and possibly @code{unix}; passing @code{_mips} defines @code{__mips},
721 @code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64}
722 defines only @code{_ABI64}.
724 You can also test for the C dialect being compiled. The variable
725 @code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus}
726 or @code{clk_objective_c}. Note that if we are preprocessing
727 assembler, this variable will be @code{clk_c} but the function-like
728 macro @code{preprocessing_asm_p()} will return true, so you might want
729 to check for that first. If you need to check for strict ANSI, the
730 variable @code{flag_iso} can be used. The function-like macro
731 @code{preprocessing_trad_p()} can be used to check for traditional
735 @defmac TARGET_OS_CPP_BUILTINS ()
736 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
737 and is used for the target operating system instead.
740 @defmac TARGET_OBJFMT_CPP_BUILTINS ()
741 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
742 and is used for the target object format. @file{elfos.h} uses this
743 macro to define @code{__ELF__}, so you probably do not need to define
747 @deftypevar {extern int} target_flags
748 This variable is declared in @file{options.h}, which is included before
749 any target-specific headers.
752 @deftypevar {Target Hook} int TARGET_DEFAULT_TARGET_FLAGS
753 This variable specifies the initial value of @code{target_flags}.
754 Its default setting is 0.
757 @cindex optional hardware or system features
758 @cindex features, optional, in system conventions
760 @deftypefn {Target Hook} bool TARGET_HANDLE_OPTION (size_t @var{code}, const char *@var{arg}, int @var{value})
761 This hook is called whenever the user specifies one of the
762 target-specific options described by the @file{.opt} definition files
763 (@pxref{Options}). It has the opportunity to do some option-specific
764 processing and should return true if the option is valid. The default
765 definition does nothing but return true.
767 @var{code} specifies the @code{OPT_@var{name}} enumeration value
768 associated with the selected option; @var{name} is just a rendering of
769 the option name in which non-alphanumeric characters are replaced by
770 underscores. @var{arg} specifies the string argument and is null if
771 no argument was given. If the option is flagged as a @code{UInteger}
772 (@pxref{Option properties}), @var{value} is the numeric value of the
773 argument. Otherwise @var{value} is 1 if the positive form of the
774 option was used and 0 if the ``no-'' form was.
777 @deftypefn {Target Hook} bool TARGET_HANDLE_C_OPTION (size_t @var{code}, const char *@var{arg}, int @var{value})
778 This target hook is called whenever the user specifies one of the
779 target-specific C language family options described by the @file{.opt}
780 definition files(@pxref{Options}). It has the opportunity to do some
781 option-specific processing and should return true if the option is
782 valid. The default definition does nothing but return false.
784 In general, you should use @code{TARGET_HANDLE_OPTION} to handle
785 options. However, if processing an option requires routines that are
786 only available in the C (and related language) front ends, then you
787 should use @code{TARGET_HANDLE_C_OPTION} instead.
790 @defmac TARGET_VERSION
791 This macro is a C statement to print on @code{stderr} a string
792 describing the particular machine description choice. Every machine
793 description should define @code{TARGET_VERSION}. For example:
797 #define TARGET_VERSION \
798 fprintf (stderr, " (68k, Motorola syntax)");
800 #define TARGET_VERSION \
801 fprintf (stderr, " (68k, MIT syntax)");
806 @defmac OVERRIDE_OPTIONS
807 Sometimes certain combinations of command options do not make sense on
808 a particular target machine. You can define a macro
809 @code{OVERRIDE_OPTIONS} to take account of this. This macro, if
810 defined, is executed once just after all the command options have been
813 Don't use this macro to turn on various extra optimizations for
814 @option{-O}. That is what @code{OPTIMIZATION_OPTIONS} is for.
817 @defmac C_COMMON_OVERRIDE_OPTIONS
818 This is similar to @code{OVERRIDE_OPTIONS} but is only used in the C
819 language frontends (C, Objective-C, C++, Objective-C++) and so can be
820 used to alter option flag variables which only exist in those
824 @defmac OPTIMIZATION_OPTIONS (@var{level}, @var{size})
825 Some machines may desire to change what optimizations are performed for
826 various optimization levels. This macro, if defined, is executed once
827 just after the optimization level is determined and before the remainder
828 of the command options have been parsed. Values set in this macro are
829 used as the default values for the other command line options.
831 @var{level} is the optimization level specified; 2 if @option{-O2} is
832 specified, 1 if @option{-O} is specified, and 0 if neither is specified.
834 @var{size} is nonzero if @option{-Os} is specified and zero otherwise.
836 You should not use this macro to change options that are not
837 machine-specific. These should uniformly selected by the same
838 optimization level on all supported machines. Use this macro to enable
839 machine-specific optimizations.
841 @strong{Do not examine @code{write_symbols} in
842 this macro!} The debugging options are not supposed to alter the
846 @deftypefn {Target Hook} bool TARGET_HELP (void)
847 This hook is called in response to the user invoking
848 @option{--target-help} on the command line. It gives the target a
849 chance to display extra information on the target specific command
850 line options found in its @file{.opt} file.
853 @defmac CAN_DEBUG_WITHOUT_FP
854 Define this macro if debugging can be performed even without a frame
855 pointer. If this macro is defined, GCC will turn on the
856 @option{-fomit-frame-pointer} option whenever @option{-O} is specified.
859 @node Per-Function Data
860 @section Defining data structures for per-function information.
861 @cindex per-function data
862 @cindex data structures
864 If the target needs to store information on a per-function basis, GCC
865 provides a macro and a couple of variables to allow this. Note, just
866 using statics to store the information is a bad idea, since GCC supports
867 nested functions, so you can be halfway through encoding one function
868 when another one comes along.
870 GCC defines a data structure called @code{struct function} which
871 contains all of the data specific to an individual function. This
872 structure contains a field called @code{machine} whose type is
873 @code{struct machine_function *}, which can be used by targets to point
874 to their own specific data.
876 If a target needs per-function specific data it should define the type
877 @code{struct machine_function} and also the macro @code{INIT_EXPANDERS}.
878 This macro should be used to initialize the function pointer
879 @code{init_machine_status}. This pointer is explained below.
881 One typical use of per-function, target specific data is to create an
882 RTX to hold the register containing the function's return address. This
883 RTX can then be used to implement the @code{__builtin_return_address}
884 function, for level 0.
886 Note---earlier implementations of GCC used a single data area to hold
887 all of the per-function information. Thus when processing of a nested
888 function began the old per-function data had to be pushed onto a
889 stack, and when the processing was finished, it had to be popped off the
890 stack. GCC used to provide function pointers called
891 @code{save_machine_status} and @code{restore_machine_status} to handle
892 the saving and restoring of the target specific information. Since the
893 single data area approach is no longer used, these pointers are no
896 @defmac INIT_EXPANDERS
897 Macro called to initialize any target specific information. This macro
898 is called once per function, before generation of any RTL has begun.
899 The intention of this macro is to allow the initialization of the
900 function pointer @code{init_machine_status}.
903 @deftypevar {void (*)(struct function *)} init_machine_status
904 If this function pointer is non-@code{NULL} it will be called once per
905 function, before function compilation starts, in order to allow the
906 target to perform any target specific initialization of the
907 @code{struct function} structure. It is intended that this would be
908 used to initialize the @code{machine} of that structure.
910 @code{struct machine_function} structures are expected to be freed by GC@.
911 Generally, any memory that they reference must be allocated by using
912 @code{ggc_alloc}, including the structure itself.
916 @section Storage Layout
917 @cindex storage layout
919 Note that the definitions of the macros in this table which are sizes or
920 alignments measured in bits do not need to be constant. They can be C
921 expressions that refer to static variables, such as the @code{target_flags}.
922 @xref{Run-time Target}.
924 @defmac BITS_BIG_ENDIAN
925 Define this macro to have the value 1 if the most significant bit in a
926 byte has the lowest number; otherwise define it to have the value zero.
927 This means that bit-field instructions count from the most significant
928 bit. If the machine has no bit-field instructions, then this must still
929 be defined, but it doesn't matter which value it is defined to. This
930 macro need not be a constant.
932 This macro does not affect the way structure fields are packed into
933 bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
936 @defmac BYTES_BIG_ENDIAN
937 Define this macro to have the value 1 if the most significant byte in a
938 word has the lowest number. This macro need not be a constant.
941 @defmac WORDS_BIG_ENDIAN
942 Define this macro to have the value 1 if, in a multiword object, the
943 most significant word has the lowest number. This applies to both
944 memory locations and registers; GCC fundamentally assumes that the
945 order of words in memory is the same as the order in registers. This
946 macro need not be a constant.
949 @defmac LIBGCC2_WORDS_BIG_ENDIAN
950 Define this macro if @code{WORDS_BIG_ENDIAN} is not constant. This must be a
951 constant value with the same meaning as @code{WORDS_BIG_ENDIAN}, which will be
952 used only when compiling @file{libgcc2.c}. Typically the value will be set
953 based on preprocessor defines.
956 @defmac FLOAT_WORDS_BIG_ENDIAN
957 Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or
958 @code{TFmode} floating point numbers are stored in memory with the word
959 containing the sign bit at the lowest address; otherwise define it to
960 have the value 0. This macro need not be a constant.
962 You need not define this macro if the ordering is the same as for
966 @defmac BITS_PER_UNIT
967 Define this macro to be the number of bits in an addressable storage
968 unit (byte). If you do not define this macro the default is 8.
971 @defmac BITS_PER_WORD
972 Number of bits in a word. If you do not define this macro, the default
973 is @code{BITS_PER_UNIT * UNITS_PER_WORD}.
976 @defmac MAX_BITS_PER_WORD
977 Maximum number of bits in a word. If this is undefined, the default is
978 @code{BITS_PER_WORD}. Otherwise, it is the constant value that is the
979 largest value that @code{BITS_PER_WORD} can have at run-time.
982 @defmac UNITS_PER_WORD
983 Number of storage units in a word; normally the size of a general-purpose
984 register, a power of two from 1 or 8.
987 @defmac MIN_UNITS_PER_WORD
988 Minimum number of units in a word. If this is undefined, the default is
989 @code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the
990 smallest value that @code{UNITS_PER_WORD} can have at run-time.
993 @defmac UNITS_PER_SIMD_WORD
994 Number of units in the vectors that the vectorizer can produce.
995 The default is equal to @code{UNITS_PER_WORD}, because the vectorizer
996 can do some transformations even in absence of specialized @acronym{SIMD}
1000 @defmac POINTER_SIZE
1001 Width of a pointer, in bits. You must specify a value no wider than the
1002 width of @code{Pmode}. If it is not equal to the width of @code{Pmode},
1003 you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify
1004 a value the default is @code{BITS_PER_WORD}.
1007 @defmac POINTERS_EXTEND_UNSIGNED
1008 A C expression that determines how pointers should be extended from
1009 @code{ptr_mode} to either @code{Pmode} or @code{word_mode}. It is
1010 greater than zero if pointers should be zero-extended, zero if they
1011 should be sign-extended, and negative if some other sort of conversion
1012 is needed. In the last case, the extension is done by the target's
1013 @code{ptr_extend} instruction.
1015 You need not define this macro if the @code{ptr_mode}, @code{Pmode}
1016 and @code{word_mode} are all the same width.
1019 @defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type})
1020 A macro to update @var{m} and @var{unsignedp} when an object whose type
1021 is @var{type} and which has the specified mode and signedness is to be
1022 stored in a register. This macro is only called when @var{type} is a
1025 On most RISC machines, which only have operations that operate on a full
1026 register, define this macro to set @var{m} to @code{word_mode} if
1027 @var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most
1028 cases, only integer modes should be widened because wider-precision
1029 floating-point operations are usually more expensive than their narrower
1032 For most machines, the macro definition does not change @var{unsignedp}.
1033 However, some machines, have instructions that preferentially handle
1034 either signed or unsigned quantities of certain modes. For example, on
1035 the DEC Alpha, 32-bit loads from memory and 32-bit add instructions
1036 sign-extend the result to 64 bits. On such machines, set
1037 @var{unsignedp} according to which kind of extension is more efficient.
1039 Do not define this macro if it would never modify @var{m}.
1042 @defmac PROMOTE_FUNCTION_MODE
1043 Like @code{PROMOTE_MODE}, but is applied to outgoing function arguments or
1044 function return values, as specified by @code{TARGET_PROMOTE_FUNCTION_ARGS}
1045 and @code{TARGET_PROMOTE_FUNCTION_RETURN}, respectively.
1047 The default is @code{PROMOTE_MODE}.
1050 @deftypefn {Target Hook} bool TARGET_PROMOTE_FUNCTION_ARGS (tree @var{fntype})
1051 This target hook should return @code{true} if the promotion described by
1052 @code{PROMOTE_FUNCTION_MODE} should be done for outgoing function
1056 @deftypefn {Target Hook} bool TARGET_PROMOTE_FUNCTION_RETURN (tree @var{fntype})
1057 This target hook should return @code{true} if the promotion described by
1058 @code{PROMOTE_FUNCTION_MODE} should be done for the return value of
1061 If this target hook returns @code{true}, @code{TARGET_FUNCTION_VALUE}
1062 must perform the same promotions done by @code{PROMOTE_FUNCTION_MODE}.
1065 @defmac PARM_BOUNDARY
1066 Normal alignment required for function parameters on the stack, in
1067 bits. All stack parameters receive at least this much alignment
1068 regardless of data type. On most machines, this is the same as the
1072 @defmac STACK_BOUNDARY
1073 Define this macro to the minimum alignment enforced by hardware for the
1074 stack pointer on this machine. The definition is a C expression for the
1075 desired alignment (measured in bits). This value is used as a default
1076 if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines,
1077 this should be the same as @code{PARM_BOUNDARY}.
1080 @defmac PREFERRED_STACK_BOUNDARY
1081 Define this macro if you wish to preserve a certain alignment for the
1082 stack pointer, greater than what the hardware enforces. The definition
1083 is a C expression for the desired alignment (measured in bits). This
1084 macro must evaluate to a value equal to or larger than
1085 @code{STACK_BOUNDARY}.
1088 @defmac FUNCTION_BOUNDARY
1089 Alignment required for a function entry point, in bits.
1092 @defmac BIGGEST_ALIGNMENT
1093 Biggest alignment that any data type can require on this machine, in
1094 bits. Note that this is not the biggest alignment that is supported,
1095 just the biggest alignment that, when violated, may cause a fault.
1098 @defmac MINIMUM_ATOMIC_ALIGNMENT
1099 If defined, the smallest alignment, in bits, that can be given to an
1100 object that can be referenced in one operation, without disturbing any
1101 nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger
1102 on machines that don't have byte or half-word store operations.
1105 @defmac BIGGEST_FIELD_ALIGNMENT
1106 Biggest alignment that any structure or union field can require on this
1107 machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for
1108 structure and union fields only, unless the field alignment has been set
1109 by the @code{__attribute__ ((aligned (@var{n})))} construct.
1112 @defmac ADJUST_FIELD_ALIGN (@var{field}, @var{computed})
1113 An expression for the alignment of a structure field @var{field} if the
1114 alignment computed in the usual way (including applying of
1115 @code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the
1116 alignment) is @var{computed}. It overrides alignment only if the
1117 field alignment has not been set by the
1118 @code{__attribute__ ((aligned (@var{n})))} construct.
1121 @defmac MAX_OFILE_ALIGNMENT
1122 Biggest alignment supported by the object file format of this machine.
1123 Use this macro to limit the alignment which can be specified using the
1124 @code{__attribute__ ((aligned (@var{n})))} construct. If not defined,
1125 the default value is @code{BIGGEST_ALIGNMENT}.
1127 On systems that use ELF, the default (in @file{config/elfos.h}) is
1128 the largest supported 32-bit ELF section alignment representable on
1129 a 32-bit host e.g. @samp{(((unsigned HOST_WIDEST_INT) 1 << 28) * 8)}.
1130 On 32-bit ELF the largest supported section alignment in bits is
1131 @samp{(0x80000000 * 8)}, but this is not representable on 32-bit hosts.
1134 @defmac DATA_ALIGNMENT (@var{type}, @var{basic-align})
1135 If defined, a C expression to compute the alignment for a variable in
1136 the static store. @var{type} is the data type, and @var{basic-align} is
1137 the alignment that the object would ordinarily have. The value of this
1138 macro is used instead of that alignment to align the object.
1140 If this macro is not defined, then @var{basic-align} is used.
1143 One use of this macro is to increase alignment of medium-size data to
1144 make it all fit in fewer cache lines. Another is to cause character
1145 arrays to be word-aligned so that @code{strcpy} calls that copy
1146 constants to character arrays can be done inline.
1149 @defmac CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align})
1150 If defined, a C expression to compute the alignment given to a constant
1151 that is being placed in memory. @var{constant} is the constant and
1152 @var{basic-align} is the alignment that the object would ordinarily
1153 have. The value of this macro is used instead of that alignment to
1156 If this macro is not defined, then @var{basic-align} is used.
1158 The typical use of this macro is to increase alignment for string
1159 constants to be word aligned so that @code{strcpy} calls that copy
1160 constants can be done inline.
1163 @defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align})
1164 If defined, a C expression to compute the alignment for a variable in
1165 the local store. @var{type} is the data type, and @var{basic-align} is
1166 the alignment that the object would ordinarily have. The value of this
1167 macro is used instead of that alignment to align the object.
1169 If this macro is not defined, then @var{basic-align} is used.
1171 One use of this macro is to increase alignment of medium-size data to
1172 make it all fit in fewer cache lines.
1175 @defmac EMPTY_FIELD_BOUNDARY
1176 Alignment in bits to be given to a structure bit-field that follows an
1177 empty field such as @code{int : 0;}.
1179 If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro.
1182 @defmac STRUCTURE_SIZE_BOUNDARY
1183 Number of bits which any structure or union's size must be a multiple of.
1184 Each structure or union's size is rounded up to a multiple of this.
1186 If you do not define this macro, the default is the same as
1187 @code{BITS_PER_UNIT}.
1190 @defmac STRICT_ALIGNMENT
1191 Define this macro to be the value 1 if instructions will fail to work
1192 if given data not on the nominal alignment. If instructions will merely
1193 go slower in that case, define this macro as 0.
1196 @defmac PCC_BITFIELD_TYPE_MATTERS
1197 Define this if you wish to imitate the way many other C compilers handle
1198 alignment of bit-fields and the structures that contain them.
1200 The behavior is that the type written for a named bit-field (@code{int},
1201 @code{short}, or other integer type) imposes an alignment for the entire
1202 structure, as if the structure really did contain an ordinary field of
1203 that type. In addition, the bit-field is placed within the structure so
1204 that it would fit within such a field, not crossing a boundary for it.
1206 Thus, on most machines, a named bit-field whose type is written as
1207 @code{int} would not cross a four-byte boundary, and would force
1208 four-byte alignment for the whole structure. (The alignment used may
1209 not be four bytes; it is controlled by the other alignment parameters.)
1211 An unnamed bit-field will not affect the alignment of the containing
1214 If the macro is defined, its definition should be a C expression;
1215 a nonzero value for the expression enables this behavior.
1217 Note that if this macro is not defined, or its value is zero, some
1218 bit-fields may cross more than one alignment boundary. The compiler can
1219 support such references if there are @samp{insv}, @samp{extv}, and
1220 @samp{extzv} insns that can directly reference memory.
1222 The other known way of making bit-fields work is to define
1223 @code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}.
1224 Then every structure can be accessed with fullwords.
1226 Unless the machine has bit-field instructions or you define
1227 @code{STRUCTURE_SIZE_BOUNDARY} that way, you must define
1228 @code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value.
1230 If your aim is to make GCC use the same conventions for laying out
1231 bit-fields as are used by another compiler, here is how to investigate
1232 what the other compiler does. Compile and run this program:
1251 printf ("Size of foo1 is %d\n",
1252 sizeof (struct foo1));
1253 printf ("Size of foo2 is %d\n",
1254 sizeof (struct foo2));
1259 If this prints 2 and 5, then the compiler's behavior is what you would
1260 get from @code{PCC_BITFIELD_TYPE_MATTERS}.
1263 @defmac BITFIELD_NBYTES_LIMITED
1264 Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited
1265 to aligning a bit-field within the structure.
1268 @deftypefn {Target Hook} bool TARGET_ALIGN_ANON_BITFIELDS (void)
1269 When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine
1270 whether unnamed bitfields affect the alignment of the containing
1271 structure. The hook should return true if the structure should inherit
1272 the alignment requirements of an unnamed bitfield's type.
1275 @deftypefn {Target Hook} bool TARGET_NARROW_VOLATILE_BITFIELDS (void)
1276 This target hook should return @code{true} if accesses to volatile bitfields
1277 should use the narrowest mode possible. It should return @code{false} if
1278 these accesses should use the bitfield container type.
1280 The default is @code{!TARGET_STRICT_ALIGN}.
1283 @defmac MEMBER_TYPE_FORCES_BLK (@var{field}, @var{mode})
1284 Return 1 if a structure or array containing @var{field} should be accessed using
1287 If @var{field} is the only field in the structure, @var{mode} is its
1288 mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the
1289 case where structures of one field would require the structure's mode to
1290 retain the field's mode.
1292 Normally, this is not needed.
1295 @defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified})
1296 Define this macro as an expression for the alignment of a type (given
1297 by @var{type} as a tree node) if the alignment computed in the usual
1298 way is @var{computed} and the alignment explicitly specified was
1301 The default is to use @var{specified} if it is larger; otherwise, use
1302 the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT}
1305 @defmac MAX_FIXED_MODE_SIZE
1306 An integer expression for the size in bits of the largest integer
1307 machine mode that should actually be used. All integer machine modes of
1308 this size or smaller can be used for structures and unions with the
1309 appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE
1310 (DImode)} is assumed.
1313 @defmac STACK_SAVEAREA_MODE (@var{save_level})
1314 If defined, an expression of type @code{enum machine_mode} that
1315 specifies the mode of the save area operand of a
1316 @code{save_stack_@var{level}} named pattern (@pxref{Standard Names}).
1317 @var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or
1318 @code{SAVE_NONLOCAL} and selects which of the three named patterns is
1319 having its mode specified.
1321 You need not define this macro if it always returns @code{Pmode}. You
1322 would most commonly define this macro if the
1323 @code{save_stack_@var{level}} patterns need to support both a 32- and a
1327 @defmac STACK_SIZE_MODE
1328 If defined, an expression of type @code{enum machine_mode} that
1329 specifies the mode of the size increment operand of an
1330 @code{allocate_stack} named pattern (@pxref{Standard Names}).
1332 You need not define this macro if it always returns @code{word_mode}.
1333 You would most commonly define this macro if the @code{allocate_stack}
1334 pattern needs to support both a 32- and a 64-bit mode.
1337 @deftypefn {Target Hook} {enum machine_mode} TARGET_LIBGCC_CMP_RETURN_MODE ()
1338 This target hook should return the mode to be used for the return value
1339 of compare instructions expanded to libgcc calls. If not defined
1340 @code{word_mode} is returned which is the right choice for a majority of
1344 @deftypefn {Target Hook} {enum machine_mode} TARGET_LIBGCC_SHIFT_COUNT_MODE ()
1345 This target hook should return the mode to be used for the shift count operand
1346 of shift instructions expanded to libgcc calls. If not defined
1347 @code{word_mode} is returned which is the right choice for a majority of
1351 @defmac TARGET_FLOAT_FORMAT
1352 A code distinguishing the floating point format of the target machine.
1353 There are two defined values:
1356 @item IEEE_FLOAT_FORMAT
1357 This code indicates IEEE floating point. It is the default; there is no
1358 need to define @code{TARGET_FLOAT_FORMAT} when the format is IEEE@.
1360 @item VAX_FLOAT_FORMAT
1361 This code indicates the ``F float'' (for @code{float}) and ``D float''
1362 or ``G float'' formats (for @code{double}) used on the VAX and PDP-11@.
1365 If your target uses a floating point format other than these, you must
1366 define a new @var{name}_FLOAT_FORMAT code for it, and add support for
1367 it to @file{real.c}.
1369 The ordering of the component words of floating point values stored in
1370 memory is controlled by @code{FLOAT_WORDS_BIG_ENDIAN}.
1373 @defmac MODE_HAS_NANS (@var{mode})
1374 When defined, this macro should be true if @var{mode} has a NaN
1375 representation. The compiler assumes that NaNs are not equal to
1376 anything (including themselves) and that addition, subtraction,
1377 multiplication and division all return NaNs when one operand is
1380 By default, this macro is true if @var{mode} is a floating-point
1381 mode and the target floating-point format is IEEE@.
1384 @defmac MODE_HAS_INFINITIES (@var{mode})
1385 This macro should be true if @var{mode} can represent infinity. At
1386 present, the compiler uses this macro to decide whether @samp{x - x}
1387 is always defined. By default, the macro is true when @var{mode}
1388 is a floating-point mode and the target format is IEEE@.
1391 @defmac MODE_HAS_SIGNED_ZEROS (@var{mode})
1392 True if @var{mode} distinguishes between positive and negative zero.
1393 The rules are expected to follow the IEEE standard:
1397 @samp{x + x} has the same sign as @samp{x}.
1400 If the sum of two values with opposite sign is zero, the result is
1401 positive for all rounding modes expect towards @minus{}infinity, for
1402 which it is negative.
1405 The sign of a product or quotient is negative when exactly one
1406 of the operands is negative.
1409 The default definition is true if @var{mode} is a floating-point
1410 mode and the target format is IEEE@.
1413 @defmac MODE_HAS_SIGN_DEPENDENT_ROUNDING (@var{mode})
1414 If defined, this macro should be true for @var{mode} if it has at
1415 least one rounding mode in which @samp{x} and @samp{-x} can be
1416 rounded to numbers of different magnitude. Two such modes are
1417 towards @minus{}infinity and towards +infinity.
1419 The default definition of this macro is true if @var{mode} is
1420 a floating-point mode and the target format is IEEE@.
1423 @defmac ROUND_TOWARDS_ZERO
1424 If defined, this macro should be true if the prevailing rounding
1425 mode is towards zero. A true value has the following effects:
1429 @code{MODE_HAS_SIGN_DEPENDENT_ROUNDING} will be false for all modes.
1432 @file{libgcc.a}'s floating-point emulator will round towards zero
1433 rather than towards nearest.
1436 The compiler's floating-point emulator will round towards zero after
1437 doing arithmetic, and when converting from the internal float format to
1441 The macro does not affect the parsing of string literals. When the
1442 primary rounding mode is towards zero, library functions like
1443 @code{strtod} might still round towards nearest, and the compiler's
1444 parser should behave like the target's @code{strtod} where possible.
1446 Not defining this macro is equivalent to returning zero.
1449 @defmac LARGEST_EXPONENT_IS_NORMAL (@var{size})
1450 This macro should return true if floats with @var{size}
1451 bits do not have a NaN or infinity representation, but use the largest
1452 exponent for normal numbers instead.
1454 Defining this macro to true for @var{size} causes @code{MODE_HAS_NANS}
1455 and @code{MODE_HAS_INFINITIES} to be false for @var{size}-bit modes.
1456 It also affects the way @file{libgcc.a} and @file{real.c} emulate
1457 floating-point arithmetic.
1459 The default definition of this macro returns false for all sizes.
1462 @deftypefn {Target Hook} bool TARGET_VECTOR_OPAQUE_P (tree @var{type})
1463 This target hook should return @code{true} a vector is opaque. That
1464 is, if no cast is needed when copying a vector value of type
1465 @var{type} into another vector lvalue of the same size. Vector opaque
1466 types cannot be initialized. The default is that there are no such
1470 @deftypefn {Target Hook} bool TARGET_MS_BITFIELD_LAYOUT_P (tree @var{record_type})
1471 This target hook returns @code{true} if bit-fields in the given
1472 @var{record_type} are to be laid out following the rules of Microsoft
1473 Visual C/C++, namely: (i) a bit-field won't share the same storage
1474 unit with the previous bit-field if their underlying types have
1475 different sizes, and the bit-field will be aligned to the highest
1476 alignment of the underlying types of itself and of the previous
1477 bit-field; (ii) a zero-sized bit-field will affect the alignment of
1478 the whole enclosing structure, even if it is unnamed; except that
1479 (iii) a zero-sized bit-field will be disregarded unless it follows
1480 another bit-field of nonzero size. If this hook returns @code{true},
1481 other macros that control bit-field layout are ignored.
1483 When a bit-field is inserted into a packed record, the whole size
1484 of the underlying type is used by one or more same-size adjacent
1485 bit-fields (that is, if its long:3, 32 bits is used in the record,
1486 and any additional adjacent long bit-fields are packed into the same
1487 chunk of 32 bits. However, if the size changes, a new field of that
1488 size is allocated). In an unpacked record, this is the same as using
1489 alignment, but not equivalent when packing.
1491 If both MS bit-fields and @samp{__attribute__((packed))} are used,
1492 the latter will take precedence. If @samp{__attribute__((packed))} is
1493 used on a single field when MS bit-fields are in use, it will take
1494 precedence for that field, but the alignment of the rest of the structure
1495 may affect its placement.
1498 @deftypefn {Target Hook} {bool} TARGET_DECIMAL_FLOAT_SUPPORTED_P (void)
1499 Returns true if the target supports decimal floating point.
1502 @deftypefn {Target Hook} {bool} TARGET_FIXED_POINT_SUPPORTED_P (void)
1503 Returns true if the target supports fixed-point arithmetic.
1506 @deftypefn {Target Hook} void TARGET_EXPAND_TO_RTL_HOOK (void)
1507 This hook is called just before expansion into rtl, allowing the target
1508 to perform additional initializations or analysis before the expansion.
1509 For example, the rs6000 port uses it to allocate a scratch stack slot
1510 for use in copying SDmode values between memory and floating point
1511 registers whenever the function being expanded has any SDmode
1515 @deftypefn {Target Hook} void TARGET_INSTANTIATE_DECLS (void)
1516 This hook allows the backend to perform additional instantiations on rtl
1517 that are not actually in any insns yet, but will be later.
1520 @deftypefn {Target Hook} {const char *} TARGET_MANGLE_TYPE (tree @var{type})
1521 If your target defines any fundamental types, or any types your target
1522 uses should be mangled differently from the default, define this hook
1523 to return the appropriate encoding for these types as part of a C++
1524 mangled name. The @var{type} argument is the tree structure representing
1525 the type to be mangled. The hook may be applied to trees which are
1526 not target-specific fundamental types; it should return @code{NULL}
1527 for all such types, as well as arguments it does not recognize. If the
1528 return value is not @code{NULL}, it must point to a statically-allocated
1531 Target-specific fundamental types might be new fundamental types or
1532 qualified versions of ordinary fundamental types. Encode new
1533 fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name}
1534 is the name used for the type in source code, and @var{n} is the
1535 length of @var{name} in decimal. Encode qualified versions of
1536 ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where
1537 @var{name} is the name used for the type qualifier in source code,
1538 @var{n} is the length of @var{name} as above, and @var{code} is the
1539 code used to represent the unqualified version of this type. (See
1540 @code{write_builtin_type} in @file{cp/mangle.c} for the list of
1541 codes.) In both cases the spaces are for clarity; do not include any
1542 spaces in your string.
1544 This hook is applied to types prior to typedef resolution. If the mangled
1545 name for a particular type depends only on that type's main variant, you
1546 can perform typedef resolution yourself using @code{TYPE_MAIN_VARIANT}
1549 The default version of this hook always returns @code{NULL}, which is
1550 appropriate for a target that does not define any new fundamental
1555 @section Layout of Source Language Data Types
1557 These macros define the sizes and other characteristics of the standard
1558 basic data types used in programs being compiled. Unlike the macros in
1559 the previous section, these apply to specific features of C and related
1560 languages, rather than to fundamental aspects of storage layout.
1562 @defmac INT_TYPE_SIZE
1563 A C expression for the size in bits of the type @code{int} on the
1564 target machine. If you don't define this, the default is one word.
1567 @defmac SHORT_TYPE_SIZE
1568 A C expression for the size in bits of the type @code{short} on the
1569 target machine. If you don't define this, the default is half a word.
1570 (If this would be less than one storage unit, it is rounded up to one
1574 @defmac LONG_TYPE_SIZE
1575 A C expression for the size in bits of the type @code{long} on the
1576 target machine. If you don't define this, the default is one word.
1579 @defmac ADA_LONG_TYPE_SIZE
1580 On some machines, the size used for the Ada equivalent of the type
1581 @code{long} by a native Ada compiler differs from that used by C@. In
1582 that situation, define this macro to be a C expression to be used for
1583 the size of that type. If you don't define this, the default is the
1584 value of @code{LONG_TYPE_SIZE}.
1587 @defmac LONG_LONG_TYPE_SIZE
1588 A C expression for the size in bits of the type @code{long long} on the
1589 target machine. If you don't define this, the default is two
1590 words. If you want to support GNU Ada on your machine, the value of this
1591 macro must be at least 64.
1594 @defmac CHAR_TYPE_SIZE
1595 A C expression for the size in bits of the type @code{char} on the
1596 target machine. If you don't define this, the default is
1597 @code{BITS_PER_UNIT}.
1600 @defmac BOOL_TYPE_SIZE
1601 A C expression for the size in bits of the C++ type @code{bool} and
1602 C99 type @code{_Bool} on the target machine. If you don't define
1603 this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}.
1606 @defmac FLOAT_TYPE_SIZE
1607 A C expression for the size in bits of the type @code{float} on the
1608 target machine. If you don't define this, the default is one word.
1611 @defmac DOUBLE_TYPE_SIZE
1612 A C expression for the size in bits of the type @code{double} on the
1613 target machine. If you don't define this, the default is two
1617 @defmac LONG_DOUBLE_TYPE_SIZE
1618 A C expression for the size in bits of the type @code{long double} on
1619 the target machine. If you don't define this, the default is two
1623 @defmac SHORT_FRACT_TYPE_SIZE
1624 A C expression for the size in bits of the type @code{short _Fract} on
1625 the target machine. If you don't define this, the default is
1626 @code{BITS_PER_UNIT}.
1629 @defmac FRACT_TYPE_SIZE
1630 A C expression for the size in bits of the type @code{_Fract} on
1631 the target machine. If you don't define this, the default is
1632 @code{BITS_PER_UNIT * 2}.
1635 @defmac LONG_FRACT_TYPE_SIZE
1636 A C expression for the size in bits of the type @code{long _Fract} on
1637 the target machine. If you don't define this, the default is
1638 @code{BITS_PER_UNIT * 4}.
1641 @defmac LONG_LONG_FRACT_TYPE_SIZE
1642 A C expression for the size in bits of the type @code{long long _Fract} on
1643 the target machine. If you don't define this, the default is
1644 @code{BITS_PER_UNIT * 8}.
1647 @defmac SHORT_ACCUM_TYPE_SIZE
1648 A C expression for the size in bits of the type @code{short _Accum} on
1649 the target machine. If you don't define this, the default is
1650 @code{BITS_PER_UNIT * 2}.
1653 @defmac ACCUM_TYPE_SIZE
1654 A C expression for the size in bits of the type @code{_Accum} on
1655 the target machine. If you don't define this, the default is
1656 @code{BITS_PER_UNIT * 4}.
1659 @defmac LONG_ACCUM_TYPE_SIZE
1660 A C expression for the size in bits of the type @code{long _Accum} on
1661 the target machine. If you don't define this, the default is
1662 @code{BITS_PER_UNIT * 8}.
1665 @defmac LONG_LONG_ACCUM_TYPE_SIZE
1666 A C expression for the size in bits of the type @code{long long _Accum} on
1667 the target machine. If you don't define this, the default is
1668 @code{BITS_PER_UNIT * 16}.
1671 @defmac LIBGCC2_LONG_DOUBLE_TYPE_SIZE
1672 Define this macro if @code{LONG_DOUBLE_TYPE_SIZE} is not constant or
1673 if you want routines in @file{libgcc2.a} for a size other than
1674 @code{LONG_DOUBLE_TYPE_SIZE}. If you don't define this, the
1675 default is @code{LONG_DOUBLE_TYPE_SIZE}.
1678 @defmac LIBGCC2_HAS_DF_MODE
1679 Define this macro if neither @code{LIBGCC2_DOUBLE_TYPE_SIZE} nor
1680 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is
1681 @code{DFmode} but you want @code{DFmode} routines in @file{libgcc2.a}
1682 anyway. If you don't define this and either @code{LIBGCC2_DOUBLE_TYPE_SIZE}
1683 or @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64 then the default is 1,
1687 @defmac LIBGCC2_HAS_XF_MODE
1688 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1689 @code{XFmode} but you want @code{XFmode} routines in @file{libgcc2.a}
1690 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1691 is 80 then the default is 1, otherwise it is 0.
1694 @defmac LIBGCC2_HAS_TF_MODE
1695 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1696 @code{TFmode} but you want @code{TFmode} routines in @file{libgcc2.a}
1697 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1698 is 128 then the default is 1, otherwise it is 0.
1705 Define these macros to be the size in bits of the mantissa of
1706 @code{SFmode}, @code{DFmode}, @code{XFmode} and @code{TFmode} values,
1707 if the defaults in @file{libgcc2.h} are inappropriate. By default,
1708 @code{FLT_MANT_DIG} is used for @code{SF_SIZE}, @code{LDBL_MANT_DIG}
1709 for @code{XF_SIZE} and @code{TF_SIZE}, and @code{DBL_MANT_DIG} or
1710 @code{LDBL_MANT_DIG} for @code{DF_SIZE} according to whether
1711 @code{LIBGCC2_DOUBLE_TYPE_SIZE} or
1712 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64.
1715 @defmac TARGET_FLT_EVAL_METHOD
1716 A C expression for the value for @code{FLT_EVAL_METHOD} in @file{float.h},
1717 assuming, if applicable, that the floating-point control word is in its
1718 default state. If you do not define this macro the value of
1719 @code{FLT_EVAL_METHOD} will be zero.
1722 @defmac WIDEST_HARDWARE_FP_SIZE
1723 A C expression for the size in bits of the widest floating-point format
1724 supported by the hardware. If you define this macro, you must specify a
1725 value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}.
1726 If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE}
1730 @defmac DEFAULT_SIGNED_CHAR
1731 An expression whose value is 1 or 0, according to whether the type
1732 @code{char} should be signed or unsigned by default. The user can
1733 always override this default with the options @option{-fsigned-char}
1734 and @option{-funsigned-char}.
1737 @deftypefn {Target Hook} bool TARGET_DEFAULT_SHORT_ENUMS (void)
1738 This target hook should return true if the compiler should give an
1739 @code{enum} type only as many bytes as it takes to represent the range
1740 of possible values of that type. It should return false if all
1741 @code{enum} types should be allocated like @code{int}.
1743 The default is to return false.
1747 A C expression for a string describing the name of the data type to use
1748 for size values. The typedef name @code{size_t} is defined using the
1749 contents of the string.
1751 The string can contain more than one keyword. If so, separate them with
1752 spaces, and write first any length keyword, then @code{unsigned} if
1753 appropriate, and finally @code{int}. The string must exactly match one
1754 of the data type names defined in the function
1755 @code{init_decl_processing} in the file @file{c-decl.c}. You may not
1756 omit @code{int} or change the order---that would cause the compiler to
1759 If you don't define this macro, the default is @code{"long unsigned
1763 @defmac PTRDIFF_TYPE
1764 A C expression for a string describing the name of the data type to use
1765 for the result of subtracting two pointers. The typedef name
1766 @code{ptrdiff_t} is defined using the contents of the string. See
1767 @code{SIZE_TYPE} above for more information.
1769 If you don't define this macro, the default is @code{"long int"}.
1773 A C expression for a string describing the name of the data type to use
1774 for wide characters. The typedef name @code{wchar_t} is defined using
1775 the contents of the string. See @code{SIZE_TYPE} above for more
1778 If you don't define this macro, the default is @code{"int"}.
1781 @defmac WCHAR_TYPE_SIZE
1782 A C expression for the size in bits of the data type for wide
1783 characters. This is used in @code{cpp}, which cannot make use of
1788 A C expression for a string describing the name of the data type to
1789 use for wide characters passed to @code{printf} and returned from
1790 @code{getwc}. The typedef name @code{wint_t} is defined using the
1791 contents of the string. See @code{SIZE_TYPE} above for more
1794 If you don't define this macro, the default is @code{"unsigned int"}.
1798 A C expression for a string describing the name of the data type that
1799 can represent any value of any standard or extended signed integer type.
1800 The typedef name @code{intmax_t} is defined using the contents of the
1801 string. See @code{SIZE_TYPE} above for more information.
1803 If you don't define this macro, the default is the first of
1804 @code{"int"}, @code{"long int"}, or @code{"long long int"} that has as
1805 much precision as @code{long long int}.
1808 @defmac UINTMAX_TYPE
1809 A C expression for a string describing the name of the data type that
1810 can represent any value of any standard or extended unsigned integer
1811 type. The typedef name @code{uintmax_t} is defined using the contents
1812 of the string. See @code{SIZE_TYPE} above for more information.
1814 If you don't define this macro, the default is the first of
1815 @code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long
1816 unsigned int"} that has as much precision as @code{long long unsigned
1820 @defmac TARGET_PTRMEMFUNC_VBIT_LOCATION
1821 The C++ compiler represents a pointer-to-member-function with a struct
1828 ptrdiff_t vtable_index;
1835 The C++ compiler must use one bit to indicate whether the function that
1836 will be called through a pointer-to-member-function is virtual.
1837 Normally, we assume that the low-order bit of a function pointer must
1838 always be zero. Then, by ensuring that the vtable_index is odd, we can
1839 distinguish which variant of the union is in use. But, on some
1840 platforms function pointers can be odd, and so this doesn't work. In
1841 that case, we use the low-order bit of the @code{delta} field, and shift
1842 the remainder of the @code{delta} field to the left.
1844 GCC will automatically make the right selection about where to store
1845 this bit using the @code{FUNCTION_BOUNDARY} setting for your platform.
1846 However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY}
1847 set such that functions always start at even addresses, but the lowest
1848 bit of pointers to functions indicate whether the function at that
1849 address is in ARM or Thumb mode. If this is the case of your
1850 architecture, you should define this macro to
1851 @code{ptrmemfunc_vbit_in_delta}.
1853 In general, you should not have to define this macro. On architectures
1854 in which function addresses are always even, according to
1855 @code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to
1856 @code{ptrmemfunc_vbit_in_pfn}.
1859 @defmac TARGET_VTABLE_USES_DESCRIPTORS
1860 Normally, the C++ compiler uses function pointers in vtables. This
1861 macro allows the target to change to use ``function descriptors''
1862 instead. Function descriptors are found on targets for whom a
1863 function pointer is actually a small data structure. Normally the
1864 data structure consists of the actual code address plus a data
1865 pointer to which the function's data is relative.
1867 If vtables are used, the value of this macro should be the number
1868 of words that the function descriptor occupies.
1871 @defmac TARGET_VTABLE_ENTRY_ALIGN
1872 By default, the vtable entries are void pointers, the so the alignment
1873 is the same as pointer alignment. The value of this macro specifies
1874 the alignment of the vtable entry in bits. It should be defined only
1875 when special alignment is necessary. */
1878 @defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE
1879 There are a few non-descriptor entries in the vtable at offsets below
1880 zero. If these entries must be padded (say, to preserve the alignment
1881 specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number
1882 of words in each data entry.
1886 @section Register Usage
1887 @cindex register usage
1889 This section explains how to describe what registers the target machine
1890 has, and how (in general) they can be used.
1892 The description of which registers a specific instruction can use is
1893 done with register classes; see @ref{Register Classes}. For information
1894 on using registers to access a stack frame, see @ref{Frame Registers}.
1895 For passing values in registers, see @ref{Register Arguments}.
1896 For returning values in registers, see @ref{Scalar Return}.
1899 * Register Basics:: Number and kinds of registers.
1900 * Allocation Order:: Order in which registers are allocated.
1901 * Values in Registers:: What kinds of values each reg can hold.
1902 * Leaf Functions:: Renumbering registers for leaf functions.
1903 * Stack Registers:: Handling a register stack such as 80387.
1906 @node Register Basics
1907 @subsection Basic Characteristics of Registers
1909 @c prevent bad page break with this line
1910 Registers have various characteristics.
1912 @defmac FIRST_PSEUDO_REGISTER
1913 Number of hardware registers known to the compiler. They receive
1914 numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1915 pseudo register's number really is assigned the number
1916 @code{FIRST_PSEUDO_REGISTER}.
1919 @defmac FIXED_REGISTERS
1920 @cindex fixed register
1921 An initializer that says which registers are used for fixed purposes
1922 all throughout the compiled code and are therefore not available for
1923 general allocation. These would include the stack pointer, the frame
1924 pointer (except on machines where that can be used as a general
1925 register when no frame pointer is needed), the program counter on
1926 machines where that is considered one of the addressable registers,
1927 and any other numbered register with a standard use.
1929 This information is expressed as a sequence of numbers, separated by
1930 commas and surrounded by braces. The @var{n}th number is 1 if
1931 register @var{n} is fixed, 0 otherwise.
1933 The table initialized from this macro, and the table initialized by
1934 the following one, may be overridden at run time either automatically,
1935 by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
1936 the user with the command options @option{-ffixed-@var{reg}},
1937 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}.
1940 @defmac CALL_USED_REGISTERS
1941 @cindex call-used register
1942 @cindex call-clobbered register
1943 @cindex call-saved register
1944 Like @code{FIXED_REGISTERS} but has 1 for each register that is
1945 clobbered (in general) by function calls as well as for fixed
1946 registers. This macro therefore identifies the registers that are not
1947 available for general allocation of values that must live across
1950 If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1951 automatically saves it on function entry and restores it on function
1952 exit, if the register is used within the function.
1955 @defmac CALL_REALLY_USED_REGISTERS
1956 @cindex call-used register
1957 @cindex call-clobbered register
1958 @cindex call-saved register
1959 Like @code{CALL_USED_REGISTERS} except this macro doesn't require
1960 that the entire set of @code{FIXED_REGISTERS} be included.
1961 (@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}).
1962 This macro is optional. If not specified, it defaults to the value
1963 of @code{CALL_USED_REGISTERS}.
1966 @defmac HARD_REGNO_CALL_PART_CLOBBERED (@var{regno}, @var{mode})
1967 @cindex call-used register
1968 @cindex call-clobbered register
1969 @cindex call-saved register
1970 A C expression that is nonzero if it is not permissible to store a
1971 value of mode @var{mode} in hard register number @var{regno} across a
1972 call without some part of it being clobbered. For most machines this
1973 macro need not be defined. It is only required for machines that do not
1974 preserve the entire contents of a register across a call.
1978 @findex call_used_regs
1981 @findex reg_class_contents
1982 @defmac CONDITIONAL_REGISTER_USAGE
1983 Zero or more C statements that may conditionally modify five variables
1984 @code{fixed_regs}, @code{call_used_regs}, @code{global_regs},
1985 @code{reg_names}, and @code{reg_class_contents}, to take into account
1986 any dependence of these register sets on target flags. The first three
1987 of these are of type @code{char []} (interpreted as Boolean vectors).
1988 @code{global_regs} is a @code{const char *[]}, and
1989 @code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is
1990 called, @code{fixed_regs}, @code{call_used_regs},
1991 @code{reg_class_contents}, and @code{reg_names} have been initialized
1992 from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS},
1993 @code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively.
1994 @code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}},
1995 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}
1996 command options have been applied.
1998 You need not define this macro if it has no work to do.
2000 @cindex disabling certain registers
2001 @cindex controlling register usage
2002 If the usage of an entire class of registers depends on the target
2003 flags, you may indicate this to GCC by using this macro to modify
2004 @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
2005 registers in the classes which should not be used by GCC@. Also define
2006 the macro @code{REG_CLASS_FROM_LETTER} / @code{REG_CLASS_FROM_CONSTRAINT}
2007 to return @code{NO_REGS} if it
2008 is called with a letter for a class that shouldn't be used.
2010 (However, if this class is not included in @code{GENERAL_REGS} and all
2011 of the insn patterns whose constraints permit this class are
2012 controlled by target switches, then GCC will automatically avoid using
2013 these registers when the target switches are opposed to them.)
2016 @defmac INCOMING_REGNO (@var{out})
2017 Define this macro if the target machine has register windows. This C
2018 expression returns the register number as seen by the called function
2019 corresponding to the register number @var{out} as seen by the calling
2020 function. Return @var{out} if register number @var{out} is not an
2024 @defmac OUTGOING_REGNO (@var{in})
2025 Define this macro if the target machine has register windows. This C
2026 expression returns the register number as seen by the calling function
2027 corresponding to the register number @var{in} as seen by the called
2028 function. Return @var{in} if register number @var{in} is not an inbound
2032 @defmac LOCAL_REGNO (@var{regno})
2033 Define this macro if the target machine has register windows. This C
2034 expression returns true if the register is call-saved but is in the
2035 register window. Unlike most call-saved registers, such registers
2036 need not be explicitly restored on function exit or during non-local
2041 If the program counter has a register number, define this as that
2042 register number. Otherwise, do not define it.
2045 @node Allocation Order
2046 @subsection Order of Allocation of Registers
2047 @cindex order of register allocation
2048 @cindex register allocation order
2050 @c prevent bad page break with this line
2051 Registers are allocated in order.
2053 @defmac REG_ALLOC_ORDER
2054 If defined, an initializer for a vector of integers, containing the
2055 numbers of hard registers in the order in which GCC should prefer
2056 to use them (from most preferred to least).
2058 If this macro is not defined, registers are used lowest numbered first
2059 (all else being equal).
2061 One use of this macro is on machines where the highest numbered
2062 registers must always be saved and the save-multiple-registers
2063 instruction supports only sequences of consecutive registers. On such
2064 machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists
2065 the highest numbered allocable register first.
2068 @defmac ORDER_REGS_FOR_LOCAL_ALLOC
2069 A C statement (sans semicolon) to choose the order in which to allocate
2070 hard registers for pseudo-registers local to a basic block.
2072 Store the desired register order in the array @code{reg_alloc_order}.
2073 Element 0 should be the register to allocate first; element 1, the next
2074 register; and so on.
2076 The macro body should not assume anything about the contents of
2077 @code{reg_alloc_order} before execution of the macro.
2079 On most machines, it is not necessary to define this macro.
2082 @node Values in Registers
2083 @subsection How Values Fit in Registers
2085 This section discusses the macros that describe which kinds of values
2086 (specifically, which machine modes) each register can hold, and how many
2087 consecutive registers are needed for a given mode.
2089 @defmac HARD_REGNO_NREGS (@var{regno}, @var{mode})
2090 A C expression for the number of consecutive hard registers, starting
2091 at register number @var{regno}, required to hold a value of mode
2092 @var{mode}. This macro must never return zero, even if a register
2093 cannot hold the requested mode - indicate that with HARD_REGNO_MODE_OK
2094 and/or CANNOT_CHANGE_MODE_CLASS instead.
2096 On a machine where all registers are exactly one word, a suitable
2097 definition of this macro is
2100 #define HARD_REGNO_NREGS(REGNO, MODE) \
2101 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
2106 @defmac HARD_REGNO_NREGS_HAS_PADDING (@var{regno}, @var{mode})
2107 A C expression that is nonzero if a value of mode @var{mode}, stored
2108 in memory, ends with padding that causes it to take up more space than
2109 in registers starting at register number @var{regno} (as determined by
2110 multiplying GCC's notion of the size of the register when containing
2111 this mode by the number of registers returned by
2112 @code{HARD_REGNO_NREGS}). By default this is zero.
2114 For example, if a floating-point value is stored in three 32-bit
2115 registers but takes up 128 bits in memory, then this would be
2118 This macros only needs to be defined if there are cases where
2119 @code{subreg_get_info}
2120 would otherwise wrongly determine that a @code{subreg} can be
2121 represented by an offset to the register number, when in fact such a
2122 @code{subreg} would contain some of the padding not stored in
2123 registers and so not be representable.
2126 @defmac HARD_REGNO_NREGS_WITH_PADDING (@var{regno}, @var{mode})
2127 For values of @var{regno} and @var{mode} for which
2128 @code{HARD_REGNO_NREGS_HAS_PADDING} returns nonzero, a C expression
2129 returning the greater number of registers required to hold the value
2130 including any padding. In the example above, the value would be four.
2133 @defmac REGMODE_NATURAL_SIZE (@var{mode})
2134 Define this macro if the natural size of registers that hold values
2135 of mode @var{mode} is not the word size. It is a C expression that
2136 should give the natural size in bytes for the specified mode. It is
2137 used by the register allocator to try to optimize its results. This
2138 happens for example on SPARC 64-bit where the natural size of
2139 floating-point registers is still 32-bit.
2142 @defmac HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
2143 A C expression that is nonzero if it is permissible to store a value
2144 of mode @var{mode} in hard register number @var{regno} (or in several
2145 registers starting with that one). For a machine where all registers
2146 are equivalent, a suitable definition is
2149 #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
2152 You need not include code to check for the numbers of fixed registers,
2153 because the allocation mechanism considers them to be always occupied.
2155 @cindex register pairs
2156 On some machines, double-precision values must be kept in even/odd
2157 register pairs. You can implement that by defining this macro to reject
2158 odd register numbers for such modes.
2160 The minimum requirement for a mode to be OK in a register is that the
2161 @samp{mov@var{mode}} instruction pattern support moves between the
2162 register and other hard register in the same class and that moving a
2163 value into the register and back out not alter it.
2165 Since the same instruction used to move @code{word_mode} will work for
2166 all narrower integer modes, it is not necessary on any machine for
2167 @code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided
2168 you define patterns @samp{movhi}, etc., to take advantage of this. This
2169 is useful because of the interaction between @code{HARD_REGNO_MODE_OK}
2170 and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes
2173 Many machines have special registers for floating point arithmetic.
2174 Often people assume that floating point machine modes are allowed only
2175 in floating point registers. This is not true. Any registers that
2176 can hold integers can safely @emph{hold} a floating point machine
2177 mode, whether or not floating arithmetic can be done on it in those
2178 registers. Integer move instructions can be used to move the values.
2180 On some machines, though, the converse is true: fixed-point machine
2181 modes may not go in floating registers. This is true if the floating
2182 registers normalize any value stored in them, because storing a
2183 non-floating value there would garble it. In this case,
2184 @code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
2185 floating registers. But if the floating registers do not automatically
2186 normalize, if you can store any bit pattern in one and retrieve it
2187 unchanged without a trap, then any machine mode may go in a floating
2188 register, so you can define this macro to say so.
2190 The primary significance of special floating registers is rather that
2191 they are the registers acceptable in floating point arithmetic
2192 instructions. However, this is of no concern to
2193 @code{HARD_REGNO_MODE_OK}. You handle it by writing the proper
2194 constraints for those instructions.
2196 On some machines, the floating registers are especially slow to access,
2197 so that it is better to store a value in a stack frame than in such a
2198 register if floating point arithmetic is not being done. As long as the
2199 floating registers are not in class @code{GENERAL_REGS}, they will not
2200 be used unless some pattern's constraint asks for one.
2203 @defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to})
2204 A C expression that is nonzero if it is OK to rename a hard register
2205 @var{from} to another hard register @var{to}.
2207 One common use of this macro is to prevent renaming of a register to
2208 another register that is not saved by a prologue in an interrupt
2211 The default is always nonzero.
2214 @defmac MODES_TIEABLE_P (@var{mode1}, @var{mode2})
2215 A C expression that is nonzero if a value of mode
2216 @var{mode1} is accessible in mode @var{mode2} without copying.
2218 If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
2219 @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for
2220 any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})}
2221 should be nonzero. If they differ for any @var{r}, you should define
2222 this macro to return zero unless some other mechanism ensures the
2223 accessibility of the value in a narrower mode.
2225 You should define this macro to return nonzero in as many cases as
2226 possible since doing so will allow GCC to perform better register
2230 @defmac AVOID_CCMODE_COPIES
2231 Define this macro if the compiler should avoid copies to/from @code{CCmode}
2232 registers. You should only define this macro if support for copying to/from
2233 @code{CCmode} is incomplete.
2236 @node Leaf Functions
2237 @subsection Handling Leaf Functions
2239 @cindex leaf functions
2240 @cindex functions, leaf
2241 On some machines, a leaf function (i.e., one which makes no calls) can run
2242 more efficiently if it does not make its own register window. Often this
2243 means it is required to receive its arguments in the registers where they
2244 are passed by the caller, instead of the registers where they would
2247 The special treatment for leaf functions generally applies only when
2248 other conditions are met; for example, often they may use only those
2249 registers for its own variables and temporaries. We use the term ``leaf
2250 function'' to mean a function that is suitable for this special
2251 handling, so that functions with no calls are not necessarily ``leaf
2254 GCC assigns register numbers before it knows whether the function is
2255 suitable for leaf function treatment. So it needs to renumber the
2256 registers in order to output a leaf function. The following macros
2259 @defmac LEAF_REGISTERS
2260 Name of a char vector, indexed by hard register number, which
2261 contains 1 for a register that is allowable in a candidate for leaf
2264 If leaf function treatment involves renumbering the registers, then the
2265 registers marked here should be the ones before renumbering---those that
2266 GCC would ordinarily allocate. The registers which will actually be
2267 used in the assembler code, after renumbering, should not be marked with 1
2270 Define this macro only if the target machine offers a way to optimize
2271 the treatment of leaf functions.
2274 @defmac LEAF_REG_REMAP (@var{regno})
2275 A C expression whose value is the register number to which @var{regno}
2276 should be renumbered, when a function is treated as a leaf function.
2278 If @var{regno} is a register number which should not appear in a leaf
2279 function before renumbering, then the expression should yield @minus{}1, which
2280 will cause the compiler to abort.
2282 Define this macro only if the target machine offers a way to optimize the
2283 treatment of leaf functions, and registers need to be renumbered to do
2287 @findex current_function_is_leaf
2288 @findex current_function_uses_only_leaf_regs
2289 @code{TARGET_ASM_FUNCTION_PROLOGUE} and
2290 @code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions
2291 specially. They can test the C variable @code{current_function_is_leaf}
2292 which is nonzero for leaf functions. @code{current_function_is_leaf} is
2293 set prior to local register allocation and is valid for the remaining
2294 compiler passes. They can also test the C variable
2295 @code{current_function_uses_only_leaf_regs} which is nonzero for leaf
2296 functions which only use leaf registers.
2297 @code{current_function_uses_only_leaf_regs} is valid after all passes
2298 that modify the instructions have been run and is only useful if
2299 @code{LEAF_REGISTERS} is defined.
2300 @c changed this to fix overfull. ALSO: why the "it" at the beginning
2301 @c of the next paragraph?! --mew 2feb93
2303 @node Stack Registers
2304 @subsection Registers That Form a Stack
2306 There are special features to handle computers where some of the
2307 ``registers'' form a stack. Stack registers are normally written by
2308 pushing onto the stack, and are numbered relative to the top of the
2311 Currently, GCC can only handle one group of stack-like registers, and
2312 they must be consecutively numbered. Furthermore, the existing
2313 support for stack-like registers is specific to the 80387 floating
2314 point coprocessor. If you have a new architecture that uses
2315 stack-like registers, you will need to do substantial work on
2316 @file{reg-stack.c} and write your machine description to cooperate
2317 with it, as well as defining these macros.
2320 Define this if the machine has any stack-like registers.
2323 @defmac FIRST_STACK_REG
2324 The number of the first stack-like register. This one is the top
2328 @defmac LAST_STACK_REG
2329 The number of the last stack-like register. This one is the bottom of
2333 @node Register Classes
2334 @section Register Classes
2335 @cindex register class definitions
2336 @cindex class definitions, register
2338 On many machines, the numbered registers are not all equivalent.
2339 For example, certain registers may not be allowed for indexed addressing;
2340 certain registers may not be allowed in some instructions. These machine
2341 restrictions are described to the compiler using @dfn{register classes}.
2343 You define a number of register classes, giving each one a name and saying
2344 which of the registers belong to it. Then you can specify register classes
2345 that are allowed as operands to particular instruction patterns.
2349 In general, each register will belong to several classes. In fact, one
2350 class must be named @code{ALL_REGS} and contain all the registers. Another
2351 class must be named @code{NO_REGS} and contain no registers. Often the
2352 union of two classes will be another class; however, this is not required.
2354 @findex GENERAL_REGS
2355 One of the classes must be named @code{GENERAL_REGS}. There is nothing
2356 terribly special about the name, but the operand constraint letters
2357 @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
2358 the same as @code{ALL_REGS}, just define it as a macro which expands
2361 Order the classes so that if class @var{x} is contained in class @var{y}
2362 then @var{x} has a lower class number than @var{y}.
2364 The way classes other than @code{GENERAL_REGS} are specified in operand
2365 constraints is through machine-dependent operand constraint letters.
2366 You can define such letters to correspond to various classes, then use
2367 them in operand constraints.
2369 You should define a class for the union of two classes whenever some
2370 instruction allows both classes. For example, if an instruction allows
2371 either a floating point (coprocessor) register or a general register for a
2372 certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
2373 which includes both of them. Otherwise you will get suboptimal code.
2375 You must also specify certain redundant information about the register
2376 classes: for each class, which classes contain it and which ones are
2377 contained in it; for each pair of classes, the largest class contained
2380 When a value occupying several consecutive registers is expected in a
2381 certain class, all the registers used must belong to that class.
2382 Therefore, register classes cannot be used to enforce a requirement for
2383 a register pair to start with an even-numbered register. The way to
2384 specify this requirement is with @code{HARD_REGNO_MODE_OK}.
2386 Register classes used for input-operands of bitwise-and or shift
2387 instructions have a special requirement: each such class must have, for
2388 each fixed-point machine mode, a subclass whose registers can transfer that
2389 mode to or from memory. For example, on some machines, the operations for
2390 single-byte values (@code{QImode}) are limited to certain registers. When
2391 this is so, each register class that is used in a bitwise-and or shift
2392 instruction must have a subclass consisting of registers from which
2393 single-byte values can be loaded or stored. This is so that
2394 @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
2396 @deftp {Data type} {enum reg_class}
2397 An enumerated type that must be defined with all the register class names
2398 as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS}
2399 must be the last register class, followed by one more enumerated value,
2400 @code{LIM_REG_CLASSES}, which is not a register class but rather
2401 tells how many classes there are.
2403 Each register class has a number, which is the value of casting
2404 the class name to type @code{int}. The number serves as an index
2405 in many of the tables described below.
2408 @defmac N_REG_CLASSES
2409 The number of distinct register classes, defined as follows:
2412 #define N_REG_CLASSES (int) LIM_REG_CLASSES
2416 @defmac REG_CLASS_NAMES
2417 An initializer containing the names of the register classes as C string
2418 constants. These names are used in writing some of the debugging dumps.
2421 @defmac REG_CLASS_CONTENTS
2422 An initializer containing the contents of the register classes, as integers
2423 which are bit masks. The @var{n}th integer specifies the contents of class
2424 @var{n}. The way the integer @var{mask} is interpreted is that
2425 register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
2427 When the machine has more than 32 registers, an integer does not suffice.
2428 Then the integers are replaced by sub-initializers, braced groupings containing
2429 several integers. Each sub-initializer must be suitable as an initializer
2430 for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
2431 In this situation, the first integer in each sub-initializer corresponds to
2432 registers 0 through 31, the second integer to registers 32 through 63, and
2436 @defmac REGNO_REG_CLASS (@var{regno})
2437 A C expression whose value is a register class containing hard register
2438 @var{regno}. In general there is more than one such class; choose a class
2439 which is @dfn{minimal}, meaning that no smaller class also contains the
2443 @defmac BASE_REG_CLASS
2444 A macro whose definition is the name of the class to which a valid
2445 base register must belong. A base register is one used in an address
2446 which is the register value plus a displacement.
2449 @defmac MODE_BASE_REG_CLASS (@var{mode})
2450 This is a variation of the @code{BASE_REG_CLASS} macro which allows
2451 the selection of a base register in a mode dependent manner. If
2452 @var{mode} is VOIDmode then it should return the same value as
2453 @code{BASE_REG_CLASS}.
2456 @defmac MODE_BASE_REG_REG_CLASS (@var{mode})
2457 A C expression whose value is the register class to which a valid
2458 base register must belong in order to be used in a base plus index
2459 register address. You should define this macro if base plus index
2460 addresses have different requirements than other base register uses.
2463 @defmac MODE_CODE_BASE_REG_CLASS (@var{mode}, @var{outer_code}, @var{index_code})
2464 A C expression whose value is the register class to which a valid
2465 base register must belong. @var{outer_code} and @var{index_code} define the
2466 context in which the base register occurs. @var{outer_code} is the code of
2467 the immediately enclosing expression (@code{MEM} for the top level of an
2468 address, @code{ADDRESS} for something that occurs in an
2469 @code{address_operand}). @var{index_code} is the code of the corresponding
2470 index expression if @var{outer_code} is @code{PLUS}; @code{SCRATCH} otherwise.
2473 @defmac INDEX_REG_CLASS
2474 A macro whose definition is the name of the class to which a valid
2475 index register must belong. An index register is one used in an
2476 address where its value is either multiplied by a scale factor or
2477 added to another register (as well as added to a displacement).
2480 @defmac REGNO_OK_FOR_BASE_P (@var{num})
2481 A C expression which is nonzero if register number @var{num} is
2482 suitable for use as a base register in operand addresses. It may be
2483 either a suitable hard register or a pseudo register that has been
2484 allocated such a hard register.
2487 @defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode})
2488 A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that
2489 that expression may examine the mode of the memory reference in
2490 @var{mode}. You should define this macro if the mode of the memory
2491 reference affects whether a register may be used as a base register. If
2492 you define this macro, the compiler will use it instead of
2493 @code{REGNO_OK_FOR_BASE_P}. The mode may be @code{VOIDmode} for
2494 addresses that appear outside a @code{MEM}, i.e., as an
2495 @code{address_operand}.
2499 @defmac REGNO_MODE_OK_FOR_REG_BASE_P (@var{num}, @var{mode})
2500 A C expression which is nonzero if register number @var{num} is suitable for
2501 use as a base register in base plus index operand addresses, accessing
2502 memory in mode @var{mode}. It may be either a suitable hard register or a
2503 pseudo register that has been allocated such a hard register. You should
2504 define this macro if base plus index addresses have different requirements
2505 than other base register uses.
2507 Use of this macro is deprecated; please use the more general
2508 @code{REGNO_MODE_CODE_OK_FOR_BASE_P}.
2511 @defmac REGNO_MODE_CODE_OK_FOR_BASE_P (@var{num}, @var{mode}, @var{outer_code}, @var{index_code})
2512 A C expression that is just like @code{REGNO_MODE_OK_FOR_BASE_P}, except
2513 that that expression may examine the context in which the register
2514 appears in the memory reference. @var{outer_code} is the code of the
2515 immediately enclosing expression (@code{MEM} if at the top level of the
2516 address, @code{ADDRESS} for something that occurs in an
2517 @code{address_operand}). @var{index_code} is the code of the
2518 corresponding index expression if @var{outer_code} is @code{PLUS};
2519 @code{SCRATCH} otherwise. The mode may be @code{VOIDmode} for addresses
2520 that appear outside a @code{MEM}, i.e., as an @code{address_operand}.
2523 @defmac REGNO_OK_FOR_INDEX_P (@var{num})
2524 A C expression which is nonzero if register number @var{num} is
2525 suitable for use as an index register in operand addresses. It may be
2526 either a suitable hard register or a pseudo register that has been
2527 allocated such a hard register.
2529 The difference between an index register and a base register is that
2530 the index register may be scaled. If an address involves the sum of
2531 two registers, neither one of them scaled, then either one may be
2532 labeled the ``base'' and the other the ``index''; but whichever
2533 labeling is used must fit the machine's constraints of which registers
2534 may serve in each capacity. The compiler will try both labelings,
2535 looking for one that is valid, and will reload one or both registers
2536 only if neither labeling works.
2539 @defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
2540 A C expression that places additional restrictions on the register class
2541 to use when it is necessary to copy value @var{x} into a register in class
2542 @var{class}. The value is a register class; perhaps @var{class}, or perhaps
2543 another, smaller class. On many machines, the following definition is
2547 #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
2550 Sometimes returning a more restrictive class makes better code. For
2551 example, on the 68000, when @var{x} is an integer constant that is in range
2552 for a @samp{moveq} instruction, the value of this macro is always
2553 @code{DATA_REGS} as long as @var{class} includes the data registers.
2554 Requiring a data register guarantees that a @samp{moveq} will be used.
2556 One case where @code{PREFERRED_RELOAD_CLASS} must not return
2557 @var{class} is if @var{x} is a legitimate constant which cannot be
2558 loaded into some register class. By returning @code{NO_REGS} you can
2559 force @var{x} into a memory location. For example, rs6000 can load
2560 immediate values into general-purpose registers, but does not have an
2561 instruction for loading an immediate value into a floating-point
2562 register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2563 @var{x} is a floating-point constant. If the constant can't be loaded
2564 into any kind of register, code generation will be better if
2565 @code{LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2566 of using @code{PREFERRED_RELOAD_CLASS}.
2568 If an insn has pseudos in it after register allocation, reload will go
2569 through the alternatives and call repeatedly @code{PREFERRED_RELOAD_CLASS}
2570 to find the best one. Returning @code{NO_REGS}, in this case, makes
2571 reload add a @code{!} in front of the constraint: the x86 back-end uses
2572 this feature to discourage usage of 387 registers when math is done in
2573 the SSE registers (and vice versa).
2576 @defmac PREFERRED_OUTPUT_RELOAD_CLASS (@var{x}, @var{class})
2577 Like @code{PREFERRED_RELOAD_CLASS}, but for output reloads instead of
2578 input reloads. If you don't define this macro, the default is to use
2579 @var{class}, unchanged.
2581 You can also use @code{PREFERRED_OUTPUT_RELOAD_CLASS} to discourage
2582 reload from using some alternatives, like @code{PREFERRED_RELOAD_CLASS}.
2585 @defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class})
2586 A C expression that places additional restrictions on the register class
2587 to use when it is necessary to be able to hold a value of mode
2588 @var{mode} in a reload register for which class @var{class} would
2591 Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when
2592 there are certain modes that simply can't go in certain reload classes.
2594 The value is a register class; perhaps @var{class}, or perhaps another,
2597 Don't define this macro unless the target machine has limitations which
2598 require the macro to do something nontrivial.
2601 @deftypefn {Target Hook} {enum reg_class} TARGET_SECONDARY_RELOAD (bool @var{in_p}, rtx @var{x}, enum reg_class @var{reload_class}, enum machine_mode @var{reload_mode}, secondary_reload_info *@var{sri})
2602 Many machines have some registers that cannot be copied directly to or
2603 from memory or even from other types of registers. An example is the
2604 @samp{MQ} register, which on most machines, can only be copied to or
2605 from general registers, but not memory. Below, we shall be using the
2606 term 'intermediate register' when a move operation cannot be performed
2607 directly, but has to be done by copying the source into the intermediate
2608 register first, and then copying the intermediate register to the
2609 destination. An intermediate register always has the same mode as
2610 source and destination. Since it holds the actual value being copied,
2611 reload might apply optimizations to re-use an intermediate register
2612 and eliding the copy from the source when it can determine that the
2613 intermediate register still holds the required value.
2615 Another kind of secondary reload is required on some machines which
2616 allow copying all registers to and from memory, but require a scratch
2617 register for stores to some memory locations (e.g., those with symbolic
2618 address on the RT, and those with certain symbolic address on the SPARC
2619 when compiling PIC)@. Scratch registers need not have the same mode
2620 as the value being copied, and usually hold a different value that
2621 that being copied. Special patterns in the md file are needed to
2622 describe how the copy is performed with the help of the scratch register;
2623 these patterns also describe the number, register class(es) and mode(s)
2624 of the scratch register(s).
2626 In some cases, both an intermediate and a scratch register are required.
2628 For input reloads, this target hook is called with nonzero @var{in_p},
2629 and @var{x} is an rtx that needs to be copied to a register of class
2630 @var{reload_class} in @var{reload_mode}. For output reloads, this target
2631 hook is called with zero @var{in_p}, and a register of class @var{reload_class}
2632 needs to be copied to rtx @var{x} in @var{reload_mode}.
2634 If copying a register of @var{reload_class} from/to @var{x} requires
2635 an intermediate register, the hook @code{secondary_reload} should
2636 return the register class required for this intermediate register.
2637 If no intermediate register is required, it should return NO_REGS.
2638 If more than one intermediate register is required, describe the one
2639 that is closest in the copy chain to the reload register.
2641 If scratch registers are needed, you also have to describe how to
2642 perform the copy from/to the reload register to/from this
2643 closest intermediate register. Or if no intermediate register is
2644 required, but still a scratch register is needed, describe the
2645 copy from/to the reload register to/from the reload operand @var{x}.
2647 You do this by setting @code{sri->icode} to the instruction code of a pattern
2648 in the md file which performs the move. Operands 0 and 1 are the output
2649 and input of this copy, respectively. Operands from operand 2 onward are
2650 for scratch operands. These scratch operands must have a mode, and a
2651 single-register-class
2652 @c [later: or memory]
2655 When an intermediate register is used, the @code{secondary_reload}
2656 hook will be called again to determine how to copy the intermediate
2657 register to/from the reload operand @var{x}, so your hook must also
2658 have code to handle the register class of the intermediate operand.
2660 @c [For later: maybe we'll allow multi-alternative reload patterns -
2661 @c the port maintainer could name a mov<mode> pattern that has clobbers -
2662 @c and match the constraints of input and output to determine the required
2663 @c alternative. A restriction would be that constraints used to match
2664 @c against reloads registers would have to be written as register class
2665 @c constraints, or we need a new target macro / hook that tells us if an
2666 @c arbitrary constraint can match an unknown register of a given class.
2667 @c Such a macro / hook would also be useful in other places.]
2670 @var{x} might be a pseudo-register or a @code{subreg} of a
2671 pseudo-register, which could either be in a hard register or in memory.
2672 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2673 in memory and the hard register number if it is in a register.
2675 Scratch operands in memory (constraint @code{"=m"} / @code{"=&m"}) are
2676 currently not supported. For the time being, you will have to continue
2677 to use @code{SECONDARY_MEMORY_NEEDED} for that purpose.
2679 @code{copy_cost} also uses this target hook to find out how values are
2680 copied. If you want it to include some extra cost for the need to allocate
2681 (a) scratch register(s), set @code{sri->extra_cost} to the additional cost.
2682 Or if two dependent moves are supposed to have a lower cost than the sum
2683 of the individual moves due to expected fortuitous scheduling and/or special
2684 forwarding logic, you can set @code{sri->extra_cost} to a negative amount.
2687 @defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2688 @defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2689 @defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2690 These macros are obsolete, new ports should use the target hook
2691 @code{TARGET_SECONDARY_RELOAD} instead.
2693 These are obsolete macros, replaced by the @code{TARGET_SECONDARY_RELOAD}
2694 target hook. Older ports still define these macros to indicate to the
2695 reload phase that it may
2696 need to allocate at least one register for a reload in addition to the
2697 register to contain the data. Specifically, if copying @var{x} to a
2698 register @var{class} in @var{mode} requires an intermediate register,
2699 you were supposed to define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the
2700 largest register class all of whose registers can be used as
2701 intermediate registers or scratch registers.
2703 If copying a register @var{class} in @var{mode} to @var{x} requires an
2704 intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS}
2705 was supposed to be defined be defined to return the largest register
2706 class required. If the
2707 requirements for input and output reloads were the same, the macro
2708 @code{SECONDARY_RELOAD_CLASS} should have been used instead of defining both
2711 The values returned by these macros are often @code{GENERAL_REGS}.
2712 Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x}
2713 can be directly copied to or from a register of @var{class} in
2714 @var{mode} without requiring a scratch register. Do not define this
2715 macro if it would always return @code{NO_REGS}.
2717 If a scratch register is required (either with or without an
2718 intermediate register), you were supposed to define patterns for
2719 @samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required
2720 (@pxref{Standard Names}. These patterns, which were normally
2721 implemented with a @code{define_expand}, should be similar to the
2722 @samp{mov@var{m}} patterns, except that operand 2 is the scratch
2725 These patterns need constraints for the reload register and scratch
2727 contain a single register class. If the original reload register (whose
2728 class is @var{class}) can meet the constraint given in the pattern, the
2729 value returned by these macros is used for the class of the scratch
2730 register. Otherwise, two additional reload registers are required.
2731 Their classes are obtained from the constraints in the insn pattern.
2733 @var{x} might be a pseudo-register or a @code{subreg} of a
2734 pseudo-register, which could either be in a hard register or in memory.
2735 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2736 in memory and the hard register number if it is in a register.
2738 These macros should not be used in the case where a particular class of
2739 registers can only be copied to memory and not to another class of
2740 registers. In that case, secondary reload registers are not needed and
2741 would not be helpful. Instead, a stack location must be used to perform
2742 the copy and the @code{mov@var{m}} pattern should use memory as an
2743 intermediate storage. This case often occurs between floating-point and
2747 @defmac SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m})
2748 Certain machines have the property that some registers cannot be copied
2749 to some other registers without using memory. Define this macro on
2750 those machines to be a C expression that is nonzero if objects of mode
2751 @var{m} in registers of @var{class1} can only be copied to registers of
2752 class @var{class2} by storing a register of @var{class1} into memory
2753 and loading that memory location into a register of @var{class2}.
2755 Do not define this macro if its value would always be zero.
2758 @defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode})
2759 Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler
2760 allocates a stack slot for a memory location needed for register copies.
2761 If this macro is defined, the compiler instead uses the memory location
2762 defined by this macro.
2764 Do not define this macro if you do not define
2765 @code{SECONDARY_MEMORY_NEEDED}.
2768 @defmac SECONDARY_MEMORY_NEEDED_MODE (@var{mode})
2769 When the compiler needs a secondary memory location to copy between two
2770 registers of mode @var{mode}, it normally allocates sufficient memory to
2771 hold a quantity of @code{BITS_PER_WORD} bits and performs the store and
2772 load operations in a mode that many bits wide and whose class is the
2773 same as that of @var{mode}.
2775 This is right thing to do on most machines because it ensures that all
2776 bits of the register are copied and prevents accesses to the registers
2777 in a narrower mode, which some machines prohibit for floating-point
2780 However, this default behavior is not correct on some machines, such as
2781 the DEC Alpha, that store short integers in floating-point registers
2782 differently than in integer registers. On those machines, the default
2783 widening will not work correctly and you must define this macro to
2784 suppress that widening in some cases. See the file @file{alpha.h} for
2787 Do not define this macro if you do not define
2788 @code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that
2789 is @code{BITS_PER_WORD} bits wide is correct for your machine.
2792 @defmac SMALL_REGISTER_CLASSES
2793 On some machines, it is risky to let hard registers live across arbitrary
2794 insns. Typically, these machines have instructions that require values
2795 to be in specific registers (like an accumulator), and reload will fail
2796 if the required hard register is used for another purpose across such an
2799 Define @code{SMALL_REGISTER_CLASSES} to be an expression with a nonzero
2800 value on these machines. When this macro has a nonzero value, the
2801 compiler will try to minimize the lifetime of hard registers.
2803 It is always safe to define this macro with a nonzero value, but if you
2804 unnecessarily define it, you will reduce the amount of optimizations
2805 that can be performed in some cases. If you do not define this macro
2806 with a nonzero value when it is required, the compiler will run out of
2807 spill registers and print a fatal error message. For most machines, you
2808 should not define this macro at all.
2811 @defmac CLASS_LIKELY_SPILLED_P (@var{class})
2812 A C expression whose value is nonzero if pseudos that have been assigned
2813 to registers of class @var{class} would likely be spilled because
2814 registers of @var{class} are needed for spill registers.
2816 The default value of this macro returns 1 if @var{class} has exactly one
2817 register and zero otherwise. On most machines, this default should be
2818 used. Only define this macro to some other expression if pseudos
2819 allocated by @file{local-alloc.c} end up in memory because their hard
2820 registers were needed for spill registers. If this macro returns nonzero
2821 for those classes, those pseudos will only be allocated by
2822 @file{global.c}, which knows how to reallocate the pseudo to another
2823 register. If there would not be another register available for
2824 reallocation, you should not change the definition of this macro since
2825 the only effect of such a definition would be to slow down register
2829 @defmac CLASS_MAX_NREGS (@var{class}, @var{mode})
2830 A C expression for the maximum number of consecutive registers
2831 of class @var{class} needed to hold a value of mode @var{mode}.
2833 This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact,
2834 the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
2835 should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno},
2836 @var{mode})} for all @var{regno} values in the class @var{class}.
2838 This macro helps control the handling of multiple-word values
2842 @defmac CANNOT_CHANGE_MODE_CLASS (@var{from}, @var{to}, @var{class})
2843 If defined, a C expression that returns nonzero for a @var{class} for which
2844 a change from mode @var{from} to mode @var{to} is invalid.
2846 For the example, loading 32-bit integer or floating-point objects into
2847 floating-point registers on the Alpha extends them to 64 bits.
2848 Therefore loading a 64-bit object and then storing it as a 32-bit object
2849 does not store the low-order 32 bits, as would be the case for a normal
2850 register. Therefore, @file{alpha.h} defines @code{CANNOT_CHANGE_MODE_CLASS}
2854 #define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \
2855 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \
2856 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0)
2860 @node Old Constraints
2861 @section Obsolete Macros for Defining Constraints
2862 @cindex defining constraints, obsolete method
2863 @cindex constraints, defining, obsolete method
2865 Machine-specific constraints can be defined with these macros instead
2866 of the machine description constructs described in @ref{Define
2867 Constraints}. This mechanism is obsolete. New ports should not use
2868 it; old ports should convert to the new mechanism.
2870 @defmac CONSTRAINT_LEN (@var{char}, @var{str})
2871 For the constraint at the start of @var{str}, which starts with the letter
2872 @var{c}, return the length. This allows you to have register class /
2873 constant / extra constraints that are longer than a single letter;
2874 you don't need to define this macro if you can do with single-letter
2875 constraints only. The definition of this macro should use
2876 DEFAULT_CONSTRAINT_LEN for all the characters that you don't want
2877 to handle specially.
2878 There are some sanity checks in genoutput.c that check the constraint lengths
2879 for the md file, so you can also use this macro to help you while you are
2880 transitioning from a byzantine single-letter-constraint scheme: when you
2881 return a negative length for a constraint you want to re-use, genoutput
2882 will complain about every instance where it is used in the md file.
2885 @defmac REG_CLASS_FROM_LETTER (@var{char})
2886 A C expression which defines the machine-dependent operand constraint
2887 letters for register classes. If @var{char} is such a letter, the
2888 value should be the register class corresponding to it. Otherwise,
2889 the value should be @code{NO_REGS}. The register letter @samp{r},
2890 corresponding to class @code{GENERAL_REGS}, will not be passed
2891 to this macro; you do not need to handle it.
2894 @defmac REG_CLASS_FROM_CONSTRAINT (@var{char}, @var{str})
2895 Like @code{REG_CLASS_FROM_LETTER}, but you also get the constraint string
2896 passed in @var{str}, so that you can use suffixes to distinguish between
2900 @defmac CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
2901 A C expression that defines the machine-dependent operand constraint
2902 letters (@samp{I}, @samp{J}, @samp{K}, @dots{} @samp{P}) that specify
2903 particular ranges of integer values. If @var{c} is one of those
2904 letters, the expression should check that @var{value}, an integer, is in
2905 the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
2906 not one of those letters, the value should be 0 regardless of
2910 @defmac CONST_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2911 Like @code{CONST_OK_FOR_LETTER_P}, but you also get the constraint
2912 string passed in @var{str}, so that you can use suffixes to distinguish
2913 between different variants.
2916 @defmac CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
2917 A C expression that defines the machine-dependent operand constraint
2918 letters that specify particular ranges of @code{const_double} values
2919 (@samp{G} or @samp{H}).
2921 If @var{c} is one of those letters, the expression should check that
2922 @var{value}, an RTX of code @code{const_double}, is in the appropriate
2923 range and return 1 if so, 0 otherwise. If @var{c} is not one of those
2924 letters, the value should be 0 regardless of @var{value}.
2926 @code{const_double} is used for all floating-point constants and for
2927 @code{DImode} fixed-point constants. A given letter can accept either
2928 or both kinds of values. It can use @code{GET_MODE} to distinguish
2929 between these kinds.
2932 @defmac CONST_DOUBLE_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2933 Like @code{CONST_DOUBLE_OK_FOR_LETTER_P}, but you also get the constraint
2934 string passed in @var{str}, so that you can use suffixes to distinguish
2935 between different variants.
2938 @defmac EXTRA_CONSTRAINT (@var{value}, @var{c})
2939 A C expression that defines the optional machine-dependent constraint
2940 letters that can be used to segregate specific types of operands, usually
2941 memory references, for the target machine. Any letter that is not
2942 elsewhere defined and not matched by @code{REG_CLASS_FROM_LETTER} /
2943 @code{REG_CLASS_FROM_CONSTRAINT}
2944 may be used. Normally this macro will not be defined.
2946 If it is required for a particular target machine, it should return 1
2947 if @var{value} corresponds to the operand type represented by the
2948 constraint letter @var{c}. If @var{c} is not defined as an extra
2949 constraint, the value returned should be 0 regardless of @var{value}.
2951 For example, on the ROMP, load instructions cannot have their output
2952 in r0 if the memory reference contains a symbolic address. Constraint
2953 letter @samp{Q} is defined as representing a memory address that does
2954 @emph{not} contain a symbolic address. An alternative is specified with
2955 a @samp{Q} constraint on the input and @samp{r} on the output. The next
2956 alternative specifies @samp{m} on the input and a register class that
2957 does not include r0 on the output.
2960 @defmac EXTRA_CONSTRAINT_STR (@var{value}, @var{c}, @var{str})
2961 Like @code{EXTRA_CONSTRAINT}, but you also get the constraint string passed
2962 in @var{str}, so that you can use suffixes to distinguish between different
2966 @defmac EXTRA_MEMORY_CONSTRAINT (@var{c}, @var{str})
2967 A C expression that defines the optional machine-dependent constraint
2968 letters, amongst those accepted by @code{EXTRA_CONSTRAINT}, that should
2969 be treated like memory constraints by the reload pass.
2971 It should return 1 if the operand type represented by the constraint
2972 at the start of @var{str}, the first letter of which is the letter @var{c},
2973 comprises a subset of all memory references including
2974 all those whose address is simply a base register. This allows the reload
2975 pass to reload an operand, if it does not directly correspond to the operand
2976 type of @var{c}, by copying its address into a base register.
2978 For example, on the S/390, some instructions do not accept arbitrary
2979 memory references, but only those that do not make use of an index
2980 register. The constraint letter @samp{Q} is defined via
2981 @code{EXTRA_CONSTRAINT} as representing a memory address of this type.
2982 If the letter @samp{Q} is marked as @code{EXTRA_MEMORY_CONSTRAINT},
2983 a @samp{Q} constraint can handle any memory operand, because the
2984 reload pass knows it can be reloaded by copying the memory address
2985 into a base register if required. This is analogous to the way
2986 a @samp{o} constraint can handle any memory operand.
2989 @defmac EXTRA_ADDRESS_CONSTRAINT (@var{c}, @var{str})
2990 A C expression that defines the optional machine-dependent constraint
2991 letters, amongst those accepted by @code{EXTRA_CONSTRAINT} /
2992 @code{EXTRA_CONSTRAINT_STR}, that should
2993 be treated like address constraints by the reload pass.
2995 It should return 1 if the operand type represented by the constraint
2996 at the start of @var{str}, which starts with the letter @var{c}, comprises
2997 a subset of all memory addresses including
2998 all those that consist of just a base register. This allows the reload
2999 pass to reload an operand, if it does not directly correspond to the operand
3000 type of @var{str}, by copying it into a base register.
3002 Any constraint marked as @code{EXTRA_ADDRESS_CONSTRAINT} can only
3003 be used with the @code{address_operand} predicate. It is treated
3004 analogously to the @samp{p} constraint.
3007 @node Stack and Calling
3008 @section Stack Layout and Calling Conventions
3009 @cindex calling conventions
3011 @c prevent bad page break with this line
3012 This describes the stack layout and calling conventions.
3016 * Exception Handling::
3021 * Register Arguments::
3023 * Aggregate Return::
3028 * Stack Smashing Protection::
3032 @subsection Basic Stack Layout
3033 @cindex stack frame layout
3034 @cindex frame layout
3036 @c prevent bad page break with this line
3037 Here is the basic stack layout.
3039 @defmac STACK_GROWS_DOWNWARD
3040 Define this macro if pushing a word onto the stack moves the stack
3041 pointer to a smaller address.
3043 When we say, ``define this macro if @dots{}'', it means that the
3044 compiler checks this macro only with @code{#ifdef} so the precise
3045 definition used does not matter.
3048 @defmac STACK_PUSH_CODE
3049 This macro defines the operation used when something is pushed
3050 on the stack. In RTL, a push operation will be
3051 @code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})}
3053 The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC},
3054 and @code{POST_INC}. Which of these is correct depends on
3055 the stack direction and on whether the stack pointer points
3056 to the last item on the stack or whether it points to the
3057 space for the next item on the stack.
3059 The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is
3060 defined, which is almost always right, and @code{PRE_INC} otherwise,
3061 which is often wrong.
3064 @defmac FRAME_GROWS_DOWNWARD
3065 Define this macro to nonzero value if the addresses of local variable slots
3066 are at negative offsets from the frame pointer.
3069 @defmac ARGS_GROW_DOWNWARD
3070 Define this macro if successive arguments to a function occupy decreasing
3071 addresses on the stack.
3074 @defmac STARTING_FRAME_OFFSET
3075 Offset from the frame pointer to the first local variable slot to be allocated.
3077 If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by
3078 subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}.
3079 Otherwise, it is found by adding the length of the first slot to the
3080 value @code{STARTING_FRAME_OFFSET}.
3081 @c i'm not sure if the above is still correct.. had to change it to get
3082 @c rid of an overfull. --mew 2feb93
3085 @defmac STACK_ALIGNMENT_NEEDED
3086 Define to zero to disable final alignment of the stack during reload.
3087 The nonzero default for this macro is suitable for most ports.
3089 On ports where @code{STARTING_FRAME_OFFSET} is nonzero or where there
3090 is a register save block following the local block that doesn't require
3091 alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable
3092 stack alignment and do it in the backend.
3095 @defmac STACK_POINTER_OFFSET
3096 Offset from the stack pointer register to the first location at which
3097 outgoing arguments are placed. If not specified, the default value of
3098 zero is used. This is the proper value for most machines.
3100 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
3101 the first location at which outgoing arguments are placed.
3104 @defmac FIRST_PARM_OFFSET (@var{fundecl})
3105 Offset from the argument pointer register to the first argument's
3106 address. On some machines it may depend on the data type of the
3109 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
3110 the first argument's address.
3113 @defmac STACK_DYNAMIC_OFFSET (@var{fundecl})
3114 Offset from the stack pointer register to an item dynamically allocated
3115 on the stack, e.g., by @code{alloca}.
3117 The default value for this macro is @code{STACK_POINTER_OFFSET} plus the
3118 length of the outgoing arguments. The default is correct for most
3119 machines. See @file{function.c} for details.
3122 @defmac INITIAL_FRAME_ADDRESS_RTX
3123 A C expression whose value is RTL representing the address of the initial
3124 stack frame. This address is passed to @code{RETURN_ADDR_RTX} and
3125 @code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable
3126 default value will be used. Define this macro in order to make frame pointer
3127 elimination work in the presence of @code{__builtin_frame_address (count)} and
3128 @code{__builtin_return_address (count)} for @code{count} not equal to zero.
3131 @defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr})
3132 A C expression whose value is RTL representing the address in a stack
3133 frame where the pointer to the caller's frame is stored. Assume that
3134 @var{frameaddr} is an RTL expression for the address of the stack frame
3137 If you don't define this macro, the default is to return the value
3138 of @var{frameaddr}---that is, the stack frame address is also the
3139 address of the stack word that points to the previous frame.
3142 @defmac SETUP_FRAME_ADDRESSES
3143 If defined, a C expression that produces the machine-specific code to
3144 setup the stack so that arbitrary frames can be accessed. For example,
3145 on the SPARC, we must flush all of the register windows to the stack
3146 before we can access arbitrary stack frames. You will seldom need to
3150 @deftypefn {Target Hook} bool TARGET_BUILTIN_SETJMP_FRAME_VALUE ()
3151 This target hook should return an rtx that is used to store
3152 the address of the current frame into the built in @code{setjmp} buffer.
3153 The default value, @code{virtual_stack_vars_rtx}, is correct for most
3154 machines. One reason you may need to define this target hook is if
3155 @code{hard_frame_pointer_rtx} is the appropriate value on your machine.
3158 @defmac FRAME_ADDR_RTX (@var{frameaddr})
3159 A C expression whose value is RTL representing the value of the frame
3160 address for the current frame. @var{frameaddr} is the frame pointer
3161 of the current frame. This is used for __builtin_frame_address.
3162 You need only define this macro if the frame address is not the same
3163 as the frame pointer. Most machines do not need to define it.
3166 @defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr})
3167 A C expression whose value is RTL representing the value of the return
3168 address for the frame @var{count} steps up from the current frame, after
3169 the prologue. @var{frameaddr} is the frame pointer of the @var{count}
3170 frame, or the frame pointer of the @var{count} @minus{} 1 frame if
3171 @code{RETURN_ADDR_IN_PREVIOUS_FRAME} is defined.
3173 The value of the expression must always be the correct address when
3174 @var{count} is zero, but may be @code{NULL_RTX} if there is no way to
3175 determine the return address of other frames.
3178 @defmac RETURN_ADDR_IN_PREVIOUS_FRAME
3179 Define this if the return address of a particular stack frame is accessed
3180 from the frame pointer of the previous stack frame.
3183 @defmac INCOMING_RETURN_ADDR_RTX
3184 A C expression whose value is RTL representing the location of the
3185 incoming return address at the beginning of any function, before the
3186 prologue. This RTL is either a @code{REG}, indicating that the return
3187 value is saved in @samp{REG}, or a @code{MEM} representing a location in
3190 You only need to define this macro if you want to support call frame
3191 debugging information like that provided by DWARF 2.
3193 If this RTL is a @code{REG}, you should also define
3194 @code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}.
3197 @defmac DWARF_ALT_FRAME_RETURN_COLUMN
3198 A C expression whose value is an integer giving a DWARF 2 column
3199 number that may be used as an alternative return column. The column
3200 must not correspond to any gcc hard register (that is, it must not
3201 be in the range of @code{DWARF_FRAME_REGNUM}).
3203 This macro can be useful if @code{DWARF_FRAME_RETURN_COLUMN} is set to a
3204 general register, but an alternative column needs to be used for signal
3205 frames. Some targets have also used different frame return columns
3209 @defmac DWARF_ZERO_REG
3210 A C expression whose value is an integer giving a DWARF 2 register
3211 number that is considered to always have the value zero. This should
3212 only be defined if the target has an architected zero register, and
3213 someone decided it was a good idea to use that register number to
3214 terminate the stack backtrace. New ports should avoid this.
3217 @deftypefn {Target Hook} void TARGET_DWARF_HANDLE_FRAME_UNSPEC (const char *@var{label}, rtx @var{pattern}, int @var{index})
3218 This target hook allows the backend to emit frame-related insns that
3219 contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging
3220 info engine will invoke it on insns of the form
3222 (set (reg) (unspec [@dots{}] UNSPEC_INDEX))
3226 (set (reg) (unspec_volatile [@dots{}] UNSPECV_INDEX)).
3228 to let the backend emit the call frame instructions. @var{label} is
3229 the CFI label attached to the insn, @var{pattern} is the pattern of
3230 the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}.
3233 @defmac INCOMING_FRAME_SP_OFFSET
3234 A C expression whose value is an integer giving the offset, in bytes,
3235 from the value of the stack pointer register to the top of the stack
3236 frame at the beginning of any function, before the prologue. The top of
3237 the frame is defined to be the value of the stack pointer in the
3238 previous frame, just before the call instruction.
3240 You only need to define this macro if you want to support call frame
3241 debugging information like that provided by DWARF 2.
3244 @defmac ARG_POINTER_CFA_OFFSET (@var{fundecl})
3245 A C expression whose value is an integer giving the offset, in bytes,
3246 from the argument pointer to the canonical frame address (cfa). The
3247 final value should coincide with that calculated by
3248 @code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable
3249 during virtual register instantiation.
3251 The default value for this macro is @code{FIRST_PARM_OFFSET (fundecl)},
3252 which is correct for most machines; in general, the arguments are found
3253 immediately before the stack frame. Note that this is not the case on
3254 some targets that save registers into the caller's frame, such as SPARC
3255 and rs6000, and so such targets need to define this macro.
3257 You only need to define this macro if the default is incorrect, and you
3258 want to support call frame debugging information like that provided by
3262 @defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl})
3263 If defined, a C expression whose value is an integer giving the offset
3264 in bytes from the frame pointer to the canonical frame address (cfa).
3265 The final value should coincide with that calculated by
3266 @code{INCOMING_FRAME_SP_OFFSET}.
3268 Normally the CFA is calculated as an offset from the argument pointer,
3269 via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is
3270 variable due to the ABI, this may not be possible. If this macro is
3271 defined, it implies that the virtual register instantiation should be
3272 based on the frame pointer instead of the argument pointer. Only one
3273 of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET}
3277 @defmac CFA_FRAME_BASE_OFFSET (@var{fundecl})
3278 If defined, a C expression whose value is an integer giving the offset
3279 in bytes from the canonical frame address (cfa) to the frame base used
3280 in DWARF 2 debug information. The default is zero. A different value
3281 may reduce the size of debug information on some ports.
3284 @node Exception Handling
3285 @subsection Exception Handling Support
3286 @cindex exception handling
3288 @defmac EH_RETURN_DATA_REGNO (@var{N})
3289 A C expression whose value is the @var{N}th register number used for
3290 data by exception handlers, or @code{INVALID_REGNUM} if fewer than
3291 @var{N} registers are usable.
3293 The exception handling library routines communicate with the exception
3294 handlers via a set of agreed upon registers. Ideally these registers
3295 should be call-clobbered; it is possible to use call-saved registers,
3296 but may negatively impact code size. The target must support at least
3297 2 data registers, but should define 4 if there are enough free registers.
3299 You must define this macro if you want to support call frame exception
3300 handling like that provided by DWARF 2.
3303 @defmac EH_RETURN_STACKADJ_RTX
3304 A C expression whose value is RTL representing a location in which
3305 to store a stack adjustment to be applied before function return.
3306 This is used to unwind the stack to an exception handler's call frame.
3307 It will be assigned zero on code paths that return normally.
3309 Typically this is a call-clobbered hard register that is otherwise
3310 untouched by the epilogue, but could also be a stack slot.
3312 Do not define this macro if the stack pointer is saved and restored
3313 by the regular prolog and epilog code in the call frame itself; in
3314 this case, the exception handling library routines will update the
3315 stack location to be restored in place. Otherwise, you must define
3316 this macro if you want to support call frame exception handling like
3317 that provided by DWARF 2.
3320 @defmac EH_RETURN_HANDLER_RTX
3321 A C expression whose value is RTL representing a location in which
3322 to store the address of an exception handler to which we should
3323 return. It will not be assigned on code paths that return normally.
3325 Typically this is the location in the call frame at which the normal
3326 return address is stored. For targets that return by popping an
3327 address off the stack, this might be a memory address just below
3328 the @emph{target} call frame rather than inside the current call
3329 frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already
3330 been assigned, so it may be used to calculate the location of the
3333 Some targets have more complex requirements than storing to an
3334 address calculable during initial code generation. In that case
3335 the @code{eh_return} instruction pattern should be used instead.
3337 If you want to support call frame exception handling, you must
3338 define either this macro or the @code{eh_return} instruction pattern.
3341 @defmac RETURN_ADDR_OFFSET
3342 If defined, an integer-valued C expression for which rtl will be generated
3343 to add it to the exception handler address before it is searched in the
3344 exception handling tables, and to subtract it again from the address before
3345 using it to return to the exception handler.
3348 @defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global})
3349 This macro chooses the encoding of pointers embedded in the exception
3350 handling sections. If at all possible, this should be defined such
3351 that the exception handling section will not require dynamic relocations,
3352 and so may be read-only.
3354 @var{code} is 0 for data, 1 for code labels, 2 for function pointers.
3355 @var{global} is true if the symbol may be affected by dynamic relocations.
3356 The macro should return a combination of the @code{DW_EH_PE_*} defines
3357 as found in @file{dwarf2.h}.
3359 If this macro is not defined, pointers will not be encoded but
3360 represented directly.
3363 @defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done})
3364 This macro allows the target to emit whatever special magic is required
3365 to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}.
3366 Generic code takes care of pc-relative and indirect encodings; this must
3367 be defined if the target uses text-relative or data-relative encodings.
3369 This is a C statement that branches to @var{done} if the format was
3370 handled. @var{encoding} is the format chosen, @var{size} is the number
3371 of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF}
3375 @defmac MD_UNWIND_SUPPORT
3376 A string specifying a file to be #include'd in unwind-dw2.c. The file
3377 so included typically defines @code{MD_FALLBACK_FRAME_STATE_FOR}.
3380 @defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs})
3381 This macro allows the target to add CPU and operating system specific
3382 code to the call-frame unwinder for use when there is no unwind data
3383 available. The most common reason to implement this macro is to unwind
3384 through signal frames.
3386 This macro is called from @code{uw_frame_state_for} in
3387 @file{unwind-dw2.c}, @file{unwind-dw2-xtensa.c} and
3388 @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context};
3389 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra}
3390 for the address of the code being executed and @code{context->cfa} for
3391 the stack pointer value. If the frame can be decoded, the register
3392 save addresses should be updated in @var{fs} and the macro should
3393 evaluate to @code{_URC_NO_REASON}. If the frame cannot be decoded,
3394 the macro should evaluate to @code{_URC_END_OF_STACK}.
3396 For proper signal handling in Java this macro is accompanied by
3397 @code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers.
3400 @defmac MD_HANDLE_UNWABI (@var{context}, @var{fs})
3401 This macro allows the target to add operating system specific code to the
3402 call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive,
3403 usually used for signal or interrupt frames.
3405 This macro is called from @code{uw_update_context} in @file{unwind-ia64.c}.
3406 @var{context} is an @code{_Unwind_Context};
3407 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi}
3408 for the abi and context in the @code{.unwabi} directive. If the
3409 @code{.unwabi} directive can be handled, the register save addresses should
3410 be updated in @var{fs}.
3413 @defmac TARGET_USES_WEAK_UNWIND_INFO
3414 A C expression that evaluates to true if the target requires unwind
3415 info to be given comdat linkage. Define it to be @code{1} if comdat
3416 linkage is necessary. The default is @code{0}.
3419 @node Stack Checking
3420 @subsection Specifying How Stack Checking is Done
3422 GCC will check that stack references are within the boundaries of
3423 the stack, if the @option{-fstack-check} is specified, in one of three ways:
3427 If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC
3428 will assume that you have arranged for stack checking to be done at
3429 appropriate places in the configuration files, e.g., in
3430 @code{TARGET_ASM_FUNCTION_PROLOGUE}. GCC will do not other special
3434 If @code{STACK_CHECK_BUILTIN} is zero and you defined a named pattern
3435 called @code{check_stack} in your @file{md} file, GCC will call that
3436 pattern with one argument which is the address to compare the stack
3437 value against. You must arrange for this pattern to report an error if
3438 the stack pointer is out of range.
3441 If neither of the above are true, GCC will generate code to periodically
3442 ``probe'' the stack pointer using the values of the macros defined below.
3445 Normally, you will use the default values of these macros, so GCC
3446 will use the third approach.
3448 @defmac STACK_CHECK_BUILTIN
3449 A nonzero value if stack checking is done by the configuration files in a
3450 machine-dependent manner. You should define this macro if stack checking
3451 is require by the ABI of your machine or if you would like to have to stack
3452 checking in some more efficient way than GCC's portable approach.
3453 The default value of this macro is zero.
3456 @defmac STACK_CHECK_PROBE_INTERVAL
3457 An integer representing the interval at which GCC must generate stack
3458 probe instructions. You will normally define this macro to be no larger
3459 than the size of the ``guard pages'' at the end of a stack area. The
3460 default value of 4096 is suitable for most systems.
3463 @defmac STACK_CHECK_PROBE_LOAD
3464 A integer which is nonzero if GCC should perform the stack probe
3465 as a load instruction and zero if GCC should use a store instruction.
3466 The default is zero, which is the most efficient choice on most systems.
3469 @defmac STACK_CHECK_PROTECT
3470 The number of bytes of stack needed to recover from a stack overflow,
3471 for languages where such a recovery is supported. The default value of
3472 75 words should be adequate for most machines.
3475 @defmac STACK_CHECK_MAX_FRAME_SIZE
3476 The maximum size of a stack frame, in bytes. GCC will generate probe
3477 instructions in non-leaf functions to ensure at least this many bytes of
3478 stack are available. If a stack frame is larger than this size, stack
3479 checking will not be reliable and GCC will issue a warning. The
3480 default is chosen so that GCC only generates one instruction on most
3481 systems. You should normally not change the default value of this macro.
3484 @defmac STACK_CHECK_FIXED_FRAME_SIZE
3485 GCC uses this value to generate the above warning message. It
3486 represents the amount of fixed frame used by a function, not including
3487 space for any callee-saved registers, temporaries and user variables.
3488 You need only specify an upper bound for this amount and will normally
3489 use the default of four words.
3492 @defmac STACK_CHECK_MAX_VAR_SIZE
3493 The maximum size, in bytes, of an object that GCC will place in the
3494 fixed area of the stack frame when the user specifies
3495 @option{-fstack-check}.
3496 GCC computed the default from the values of the above macros and you will
3497 normally not need to override that default.
3501 @node Frame Registers
3502 @subsection Registers That Address the Stack Frame
3504 @c prevent bad page break with this line
3505 This discusses registers that address the stack frame.
3507 @defmac STACK_POINTER_REGNUM
3508 The register number of the stack pointer register, which must also be a
3509 fixed register according to @code{FIXED_REGISTERS}. On most machines,
3510 the hardware determines which register this is.
3513 @defmac FRAME_POINTER_REGNUM
3514 The register number of the frame pointer register, which is used to
3515 access automatic variables in the stack frame. On some machines, the
3516 hardware determines which register this is. On other machines, you can
3517 choose any register you wish for this purpose.
3520 @defmac HARD_FRAME_POINTER_REGNUM
3521 On some machines the offset between the frame pointer and starting
3522 offset of the automatic variables is not known until after register
3523 allocation has been done (for example, because the saved registers are
3524 between these two locations). On those machines, define
3525 @code{FRAME_POINTER_REGNUM} the number of a special, fixed register to
3526 be used internally until the offset is known, and define
3527 @code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number
3528 used for the frame pointer.
3530 You should define this macro only in the very rare circumstances when it
3531 is not possible to calculate the offset between the frame pointer and
3532 the automatic variables until after register allocation has been
3533 completed. When this macro is defined, you must also indicate in your
3534 definition of @code{ELIMINABLE_REGS} how to eliminate
3535 @code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM}
3536 or @code{STACK_POINTER_REGNUM}.
3538 Do not define this macro if it would be the same as
3539 @code{FRAME_POINTER_REGNUM}.
3542 @defmac ARG_POINTER_REGNUM
3543 The register number of the arg pointer register, which is used to access
3544 the function's argument list. On some machines, this is the same as the
3545 frame pointer register. On some machines, the hardware determines which
3546 register this is. On other machines, you can choose any register you
3547 wish for this purpose. If this is not the same register as the frame
3548 pointer register, then you must mark it as a fixed register according to
3549 @code{FIXED_REGISTERS}, or arrange to be able to eliminate it
3550 (@pxref{Elimination}).
3553 @defmac RETURN_ADDRESS_POINTER_REGNUM
3554 The register number of the return address pointer register, which is used to
3555 access the current function's return address from the stack. On some
3556 machines, the return address is not at a fixed offset from the frame
3557 pointer or stack pointer or argument pointer. This register can be defined
3558 to point to the return address on the stack, and then be converted by
3559 @code{ELIMINABLE_REGS} into either the frame pointer or stack pointer.
3561 Do not define this macro unless there is no other way to get the return
3562 address from the stack.
3565 @defmac STATIC_CHAIN_REGNUM
3566 @defmacx STATIC_CHAIN_INCOMING_REGNUM
3567 Register numbers used for passing a function's static chain pointer. If
3568 register windows are used, the register number as seen by the called
3569 function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register
3570 number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If
3571 these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need
3574 The static chain register need not be a fixed register.
3576 If the static chain is passed in memory, these macros should not be
3577 defined; instead, the next two macros should be defined.
3580 @defmac STATIC_CHAIN
3581 @defmacx STATIC_CHAIN_INCOMING
3582 If the static chain is passed in memory, these macros provide rtx giving
3583 @code{mem} expressions that denote where they are stored.
3584 @code{STATIC_CHAIN} and @code{STATIC_CHAIN_INCOMING} give the locations
3585 as seen by the calling and called functions, respectively. Often the former
3586 will be at an offset from the stack pointer and the latter at an offset from
3589 @findex stack_pointer_rtx
3590 @findex frame_pointer_rtx
3591 @findex arg_pointer_rtx
3592 The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and
3593 @code{arg_pointer_rtx} will have been initialized prior to the use of these
3594 macros and should be used to refer to those items.
3596 If the static chain is passed in a register, the two previous macros should
3600 @defmac DWARF_FRAME_REGISTERS
3601 This macro specifies the maximum number of hard registers that can be
3602 saved in a call frame. This is used to size data structures used in
3603 DWARF2 exception handling.
3605 Prior to GCC 3.0, this macro was needed in order to establish a stable
3606 exception handling ABI in the face of adding new hard registers for ISA
3607 extensions. In GCC 3.0 and later, the EH ABI is insulated from changes
3608 in the number of hard registers. Nevertheless, this macro can still be
3609 used to reduce the runtime memory requirements of the exception handling
3610 routines, which can be substantial if the ISA contains a lot of
3611 registers that are not call-saved.
3613 If this macro is not defined, it defaults to
3614 @code{FIRST_PSEUDO_REGISTER}.
3617 @defmac PRE_GCC3_DWARF_FRAME_REGISTERS
3619 This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided
3620 for backward compatibility in pre GCC 3.0 compiled code.
3622 If this macro is not defined, it defaults to
3623 @code{DWARF_FRAME_REGISTERS}.
3626 @defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno})
3628 Define this macro if the target's representation for dwarf registers
3629 is different than the internal representation for unwind column.
3630 Given a dwarf register, this macro should return the internal unwind
3631 column number to use instead.
3633 See the PowerPC's SPE target for an example.
3636 @defmac DWARF_FRAME_REGNUM (@var{regno})
3638 Define this macro if the target's representation for dwarf registers
3639 used in .eh_frame or .debug_frame is different from that used in other
3640 debug info sections. Given a GCC hard register number, this macro
3641 should return the .eh_frame register number. The default is
3642 @code{DBX_REGISTER_NUMBER (@var{regno})}.
3646 @defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh})
3648 Define this macro to map register numbers held in the call frame info
3649 that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that
3650 should be output in .debug_frame (@code{@var{for_eh}} is zero) and
3651 .eh_frame (@code{@var{for_eh}} is nonzero). The default is to
3652 return @code{@var{regno}}.
3657 @subsection Eliminating Frame Pointer and Arg Pointer
3659 @c prevent bad page break with this line
3660 This is about eliminating the frame pointer and arg pointer.
3662 @defmac FRAME_POINTER_REQUIRED
3663 A C expression which is nonzero if a function must have and use a frame
3664 pointer. This expression is evaluated in the reload pass. If its value is
3665 nonzero the function will have a frame pointer.
3667 The expression can in principle examine the current function and decide
3668 according to the facts, but on most machines the constant 0 or the
3669 constant 1 suffices. Use 0 when the machine allows code to be generated
3670 with no frame pointer, and doing so saves some time or space. Use 1
3671 when there is no possible advantage to avoiding a frame pointer.
3673 In certain cases, the compiler does not know how to produce valid code
3674 without a frame pointer. The compiler recognizes those cases and
3675 automatically gives the function a frame pointer regardless of what
3676 @code{FRAME_POINTER_REQUIRED} says. You don't need to worry about
3679 In a function that does not require a frame pointer, the frame pointer
3680 register can be allocated for ordinary usage, unless you mark it as a
3681 fixed register. See @code{FIXED_REGISTERS} for more information.
3684 @findex get_frame_size
3685 @defmac INITIAL_FRAME_POINTER_OFFSET (@var{depth-var})
3686 A C statement to store in the variable @var{depth-var} the difference
3687 between the frame pointer and the stack pointer values immediately after
3688 the function prologue. The value would be computed from information
3689 such as the result of @code{get_frame_size ()} and the tables of
3690 registers @code{regs_ever_live} and @code{call_used_regs}.
3692 If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and
3693 need not be defined. Otherwise, it must be defined even if
3694 @code{FRAME_POINTER_REQUIRED} is defined to always be true; in that
3695 case, you may set @var{depth-var} to anything.
3698 @defmac ELIMINABLE_REGS
3699 If defined, this macro specifies a table of register pairs used to
3700 eliminate unneeded registers that point into the stack frame. If it is not
3701 defined, the only elimination attempted by the compiler is to replace
3702 references to the frame pointer with references to the stack pointer.
3704 The definition of this macro is a list of structure initializations, each
3705 of which specifies an original and replacement register.
3707 On some machines, the position of the argument pointer is not known until
3708 the compilation is completed. In such a case, a separate hard register
3709 must be used for the argument pointer. This register can be eliminated by
3710 replacing it with either the frame pointer or the argument pointer,
3711 depending on whether or not the frame pointer has been eliminated.
3713 In this case, you might specify:
3715 #define ELIMINABLE_REGS \
3716 @{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \
3717 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \
3718 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@}
3721 Note that the elimination of the argument pointer with the stack pointer is
3722 specified first since that is the preferred elimination.
3725 @defmac CAN_ELIMINATE (@var{from-reg}, @var{to-reg})
3726 A C expression that returns nonzero if the compiler is allowed to try
3727 to replace register number @var{from-reg} with register number
3728 @var{to-reg}. This macro need only be defined if @code{ELIMINABLE_REGS}
3729 is defined, and will usually be the constant 1, since most of the cases
3730 preventing register elimination are things that the compiler already
3734 @defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var})
3735 This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It
3736 specifies the initial difference between the specified pair of
3737 registers. This macro must be defined if @code{ELIMINABLE_REGS} is
3741 @node Stack Arguments
3742 @subsection Passing Function Arguments on the Stack
3743 @cindex arguments on stack
3744 @cindex stack arguments
3746 The macros in this section control how arguments are passed
3747 on the stack. See the following section for other macros that
3748 control passing certain arguments in registers.
3750 @deftypefn {Target Hook} bool TARGET_PROMOTE_PROTOTYPES (tree @var{fntype})
3751 This target hook returns @code{true} if an argument declared in a
3752 prototype as an integral type smaller than @code{int} should actually be
3753 passed as an @code{int}. In addition to avoiding errors in certain
3754 cases of mismatch, it also makes for better code on certain machines.
3755 The default is to not promote prototypes.
3759 A C expression. If nonzero, push insns will be used to pass
3761 If the target machine does not have a push instruction, set it to zero.
3762 That directs GCC to use an alternate strategy: to
3763 allocate the entire argument block and then store the arguments into
3764 it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too.
3767 @defmac PUSH_ARGS_REVERSED
3768 A C expression. If nonzero, function arguments will be evaluated from
3769 last to first, rather than from first to last. If this macro is not
3770 defined, it defaults to @code{PUSH_ARGS} on targets where the stack
3771 and args grow in opposite directions, and 0 otherwise.
3774 @defmac PUSH_ROUNDING (@var{npushed})
3775 A C expression that is the number of bytes actually pushed onto the
3776 stack when an instruction attempts to push @var{npushed} bytes.
3778 On some machines, the definition
3781 #define PUSH_ROUNDING(BYTES) (BYTES)
3785 will suffice. But on other machines, instructions that appear
3786 to push one byte actually push two bytes in an attempt to maintain
3787 alignment. Then the definition should be
3790 #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
3794 @findex current_function_outgoing_args_size
3795 @defmac ACCUMULATE_OUTGOING_ARGS
3796 A C expression. If nonzero, the maximum amount of space required for outgoing arguments
3797 will be computed and placed into the variable
3798 @code{current_function_outgoing_args_size}. No space will be pushed
3799 onto the stack for each call; instead, the function prologue should
3800 increase the stack frame size by this amount.
3802 Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS}
3806 @defmac REG_PARM_STACK_SPACE (@var{fndecl})
3807 Define this macro if functions should assume that stack space has been
3808 allocated for arguments even when their values are passed in
3811 The value of this macro is the size, in bytes, of the area reserved for
3812 arguments passed in registers for the function represented by @var{fndecl},
3813 which can be zero if GCC is calling a library function.
3815 This space can be allocated by the caller, or be a part of the
3816 machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says
3819 @c above is overfull. not sure what to do. --mew 5feb93 did
3820 @c something, not sure if it looks good. --mew 10feb93
3822 @defmac OUTGOING_REG_PARM_STACK_SPACE (@var{fntype})
3823 Define this to a nonzero value if it is the responsibility of the
3824 caller to allocate the area reserved for arguments passed in registers
3825 when calling a function of @var{fntype}. @var{fntype} may be NULL
3826 if the function called is a library function.
3828 If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls
3829 whether the space for these arguments counts in the value of
3830 @code{current_function_outgoing_args_size}.
3833 @defmac STACK_PARMS_IN_REG_PARM_AREA
3834 Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the
3835 stack parameters don't skip the area specified by it.
3836 @c i changed this, makes more sens and it should have taken care of the
3837 @c overfull.. not as specific, tho. --mew 5feb93
3839 Normally, when a parameter is not passed in registers, it is placed on the
3840 stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro
3841 suppresses this behavior and causes the parameter to be passed on the
3842 stack in its natural location.
3845 @defmac RETURN_POPS_ARGS (@var{fundecl}, @var{funtype}, @var{stack-size})
3846 A C expression that should indicate the number of bytes of its own
3847 arguments that a function pops on returning, or 0 if the
3848 function pops no arguments and the caller must therefore pop them all
3849 after the function returns.
3851 @var{fundecl} is a C variable whose value is a tree node that describes
3852 the function in question. Normally it is a node of type
3853 @code{FUNCTION_DECL} that describes the declaration of the function.
3854 From this you can obtain the @code{DECL_ATTRIBUTES} of the function.
3856 @var{funtype} is a C variable whose value is a tree node that
3857 describes the function in question. Normally it is a node of type
3858 @code{FUNCTION_TYPE} that describes the data type of the function.
3859 From this it is possible to obtain the data types of the value and
3860 arguments (if known).
3862 When a call to a library function is being considered, @var{fundecl}
3863 will contain an identifier node for the library function. Thus, if
3864 you need to distinguish among various library functions, you can do so
3865 by their names. Note that ``library function'' in this context means
3866 a function used to perform arithmetic, whose name is known specially
3867 in the compiler and was not mentioned in the C code being compiled.
3869 @var{stack-size} is the number of bytes of arguments passed on the
3870 stack. If a variable number of bytes is passed, it is zero, and
3871 argument popping will always be the responsibility of the calling function.
3873 On the VAX, all functions always pop their arguments, so the definition
3874 of this macro is @var{stack-size}. On the 68000, using the standard
3875 calling convention, no functions pop their arguments, so the value of
3876 the macro is always 0 in this case. But an alternative calling
3877 convention is available in which functions that take a fixed number of
3878 arguments pop them but other functions (such as @code{printf}) pop
3879 nothing (the caller pops all). When this convention is in use,
3880 @var{funtype} is examined to determine whether a function takes a fixed
3881 number of arguments.
3884 @defmac CALL_POPS_ARGS (@var{cum})
3885 A C expression that should indicate the number of bytes a call sequence
3886 pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS}
3887 when compiling a function call.
3889 @var{cum} is the variable in which all arguments to the called function
3890 have been accumulated.
3892 On certain architectures, such as the SH5, a call trampoline is used
3893 that pops certain registers off the stack, depending on the arguments
3894 that have been passed to the function. Since this is a property of the
3895 call site, not of the called function, @code{RETURN_POPS_ARGS} is not
3899 @node Register Arguments
3900 @subsection Passing Arguments in Registers
3901 @cindex arguments in registers
3902 @cindex registers arguments
3904 This section describes the macros which let you control how various
3905 types of arguments are passed in registers or how they are arranged in
3908 @defmac FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
3909 A C expression that controls whether a function argument is passed
3910 in a register, and which register.
3912 The arguments are @var{cum}, which summarizes all the previous
3913 arguments; @var{mode}, the machine mode of the argument; @var{type},
3914 the data type of the argument as a tree node or 0 if that is not known
3915 (which happens for C support library functions); and @var{named},
3916 which is 1 for an ordinary argument and 0 for nameless arguments that
3917 correspond to @samp{@dots{}} in the called function's prototype.
3918 @var{type} can be an incomplete type if a syntax error has previously
3921 The value of the expression is usually either a @code{reg} RTX for the
3922 hard register in which to pass the argument, or zero to pass the
3923 argument on the stack.
3925 For machines like the VAX and 68000, where normally all arguments are
3926 pushed, zero suffices as a definition.
3928 The value of the expression can also be a @code{parallel} RTX@. This is
3929 used when an argument is passed in multiple locations. The mode of the
3930 @code{parallel} should be the mode of the entire argument. The
3931 @code{parallel} holds any number of @code{expr_list} pairs; each one
3932 describes where part of the argument is passed. In each
3933 @code{expr_list} the first operand must be a @code{reg} RTX for the hard
3934 register in which to pass this part of the argument, and the mode of the
3935 register RTX indicates how large this part of the argument is. The
3936 second operand of the @code{expr_list} is a @code{const_int} which gives
3937 the offset in bytes into the entire argument of where this part starts.
3938 As a special exception the first @code{expr_list} in the @code{parallel}
3939 RTX may have a first operand of zero. This indicates that the entire
3940 argument is also stored on the stack.
3942 The last time this macro is called, it is called with @code{MODE ==
3943 VOIDmode}, and its result is passed to the @code{call} or @code{call_value}
3944 pattern as operands 2 and 3 respectively.
3946 @cindex @file{stdarg.h} and register arguments
3947 The usual way to make the ISO library @file{stdarg.h} work on a machine
3948 where some arguments are usually passed in registers, is to cause
3949 nameless arguments to be passed on the stack instead. This is done
3950 by making @code{FUNCTION_ARG} return 0 whenever @var{named} is 0.
3952 @cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{FUNCTION_ARG}
3953 @cindex @code{REG_PARM_STACK_SPACE}, and @code{FUNCTION_ARG}
3954 You may use the hook @code{targetm.calls.must_pass_in_stack}
3955 in the definition of this macro to determine if this argument is of a
3956 type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE}
3957 is not defined and @code{FUNCTION_ARG} returns nonzero for such an
3958 argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is
3959 defined, the argument will be computed in the stack and then loaded into
3963 @deftypefn {Target Hook} bool TARGET_MUST_PASS_IN_STACK (enum machine_mode @var{mode}, tree @var{type})
3964 This target hook should return @code{true} if we should not pass @var{type}
3965 solely in registers. The file @file{expr.h} defines a
3966 definition that is usually appropriate, refer to @file{expr.h} for additional
3970 @defmac FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
3971 Define this macro if the target machine has ``register windows'', so
3972 that the register in which a function sees an arguments is not
3973 necessarily the same as the one in which the caller passed the
3976 For such machines, @code{FUNCTION_ARG} computes the register in which
3977 the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
3978 be defined in a similar fashion to tell the function being called
3979 where the arguments will arrive.
3981 If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
3982 serves both purposes.
3985 @deftypefn {Target Hook} int TARGET_ARG_PARTIAL_BYTES (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named})
3986 This target hook returns the number of bytes at the beginning of an
3987 argument that must be put in registers. The value must be zero for
3988 arguments that are passed entirely in registers or that are entirely
3989 pushed on the stack.
3991 On some machines, certain arguments must be passed partially in
3992 registers and partially in memory. On these machines, typically the
3993 first few words of arguments are passed in registers, and the rest
3994 on the stack. If a multi-word argument (a @code{double} or a
3995 structure) crosses that boundary, its first few words must be passed
3996 in registers and the rest must be pushed. This macro tells the
3997 compiler when this occurs, and how many bytes should go in registers.
3999 @code{FUNCTION_ARG} for these arguments should return the first
4000 register to be used by the caller for this argument; likewise
4001 @code{FUNCTION_INCOMING_ARG}, for the called function.
4004 @deftypefn {Target Hook} bool TARGET_PASS_BY_REFERENCE (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named})
4005 This target hook should return @code{true} if an argument at the
4006 position indicated by @var{cum} should be passed by reference. This
4007 predicate is queried after target independent reasons for being
4008 passed by reference, such as @code{TREE_ADDRESSABLE (type)}.
4010 If the hook returns true, a copy of that argument is made in memory and a
4011 pointer to the argument is passed instead of the argument itself.
4012 The pointer is passed in whatever way is appropriate for passing a pointer
4016 @deftypefn {Target Hook} bool TARGET_CALLEE_COPIES (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named})
4017 The function argument described by the parameters to this hook is
4018 known to be passed by reference. The hook should return true if the
4019 function argument should be copied by the callee instead of copied
4022 For any argument for which the hook returns true, if it can be
4023 determined that the argument is not modified, then a copy need
4026 The default version of this hook always returns false.
4029 @defmac CUMULATIVE_ARGS
4030 A C type for declaring a variable that is used as the first argument of
4031 @code{FUNCTION_ARG} and other related values. For some target machines,
4032 the type @code{int} suffices and can hold the number of bytes of
4035 There is no need to record in @code{CUMULATIVE_ARGS} anything about the
4036 arguments that have been passed on the stack. The compiler has other
4037 variables to keep track of that. For target machines on which all
4038 arguments are passed on the stack, there is no need to store anything in
4039 @code{CUMULATIVE_ARGS}; however, the data structure must exist and
4040 should not be empty, so use @code{int}.
4043 @defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args})
4044 A C statement (sans semicolon) for initializing the variable
4045 @var{cum} for the state at the beginning of the argument list. The
4046 variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype}
4047 is the tree node for the data type of the function which will receive
4048 the args, or 0 if the args are to a compiler support library function.
4049 For direct calls that are not libcalls, @var{fndecl} contain the
4050 declaration node of the function. @var{fndecl} is also set when
4051 @code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function
4052 being compiled. @var{n_named_args} is set to the number of named
4053 arguments, including a structure return address if it is passed as a
4054 parameter, when making a call. When processing incoming arguments,
4055 @var{n_named_args} is set to @minus{}1.
4057 When processing a call to a compiler support library function,
4058 @var{libname} identifies which one. It is a @code{symbol_ref} rtx which
4059 contains the name of the function, as a string. @var{libname} is 0 when
4060 an ordinary C function call is being processed. Thus, each time this
4061 macro is called, either @var{libname} or @var{fntype} is nonzero, but
4062 never both of them at once.
4065 @defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname})
4066 Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls,
4067 it gets a @code{MODE} argument instead of @var{fntype}, that would be
4068 @code{NULL}. @var{indirect} would always be zero, too. If this macro
4069 is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname,
4070 0)} is used instead.
4073 @defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname})
4074 Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of
4075 finding the arguments for the function being compiled. If this macro is
4076 undefined, @code{INIT_CUMULATIVE_ARGS} is used instead.
4078 The value passed for @var{libname} is always 0, since library routines
4079 with special calling conventions are never compiled with GCC@. The
4080 argument @var{libname} exists for symmetry with
4081 @code{INIT_CUMULATIVE_ARGS}.
4082 @c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe.
4083 @c --mew 5feb93 i switched the order of the sentences. --mew 10feb93
4086 @defmac FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
4087 A C statement (sans semicolon) to update the summarizer variable
4088 @var{cum} to advance past an argument in the argument list. The
4089 values @var{mode}, @var{type} and @var{named} describe that argument.
4090 Once this is done, the variable @var{cum} is suitable for analyzing
4091 the @emph{following} argument with @code{FUNCTION_ARG}, etc.
4093 This macro need not do anything if the argument in question was passed
4094 on the stack. The compiler knows how to track the amount of stack space
4095 used for arguments without any special help.
4098 @defmac FUNCTION_ARG_PADDING (@var{mode}, @var{type})
4099 If defined, a C expression which determines whether, and in which direction,
4100 to pad out an argument with extra space. The value should be of type
4101 @code{enum direction}: either @code{upward} to pad above the argument,
4102 @code{downward} to pad below, or @code{none} to inhibit padding.
4104 The @emph{amount} of padding is always just enough to reach the next
4105 multiple of @code{FUNCTION_ARG_BOUNDARY}; this macro does not control
4108 This macro has a default definition which is right for most systems.
4109 For little-endian machines, the default is to pad upward. For
4110 big-endian machines, the default is to pad downward for an argument of
4111 constant size shorter than an @code{int}, and upward otherwise.
4114 @defmac PAD_VARARGS_DOWN
4115 If defined, a C expression which determines whether the default
4116 implementation of va_arg will attempt to pad down before reading the
4117 next argument, if that argument is smaller than its aligned space as
4118 controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such
4119 arguments are padded down if @code{BYTES_BIG_ENDIAN} is true.
4122 @defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first})
4123 Specify padding for the last element of a block move between registers and
4124 memory. @var{first} is nonzero if this is the only element. Defining this
4125 macro allows better control of register function parameters on big-endian
4126 machines, without using @code{PARALLEL} rtl. In particular,
4127 @code{MUST_PASS_IN_STACK} need not test padding and mode of types in
4128 registers, as there is no longer a "wrong" part of a register; For example,
4129 a three byte aggregate may be passed in the high part of a register if so
4133 @defmac FUNCTION_ARG_BOUNDARY (@var{mode}, @var{type})
4134 If defined, a C expression that gives the alignment boundary, in bits,
4135 of an argument with the specified mode and type. If it is not defined,
4136 @code{PARM_BOUNDARY} is used for all arguments.
4139 @defmac FUNCTION_ARG_REGNO_P (@var{regno})
4140 A C expression that is nonzero if @var{regno} is the number of a hard
4141 register in which function arguments are sometimes passed. This does
4142 @emph{not} include implicit arguments such as the static chain and
4143 the structure-value address. On many machines, no registers can be
4144 used for this purpose since all function arguments are pushed on the
4148 @deftypefn {Target Hook} bool TARGET_SPLIT_COMPLEX_ARG (tree @var{type})
4149 This hook should return true if parameter of type @var{type} are passed
4150 as two scalar parameters. By default, GCC will attempt to pack complex
4151 arguments into the target's word size. Some ABIs require complex arguments
4152 to be split and treated as their individual components. For example, on
4153 AIX64, complex floats should be passed in a pair of floating point
4154 registers, even though a complex float would fit in one 64-bit floating
4157 The default value of this hook is @code{NULL}, which is treated as always
4161 @deftypefn {Target Hook} tree TARGET_BUILD_BUILTIN_VA_LIST (void)
4162 This hook returns a type node for @code{va_list} for the target.
4163 The default version of the hook returns @code{void*}.
4166 @deftypefn {Target Hook} tree TARGET_GIMPLIFY_VA_ARG_EXPR (tree @var{valist}, tree @var{type}, tree *@var{pre_p}, tree *@var{post_p})
4167 This hook performs target-specific gimplification of
4168 @code{VA_ARG_EXPR}. The first two parameters correspond to the
4169 arguments to @code{va_arg}; the latter two are as in
4170 @code{gimplify.c:gimplify_expr}.
4173 @deftypefn {Target Hook} bool TARGET_VALID_POINTER_MODE (enum machine_mode @var{mode})
4174 Define this to return nonzero if the port can handle pointers
4175 with machine mode @var{mode}. The default version of this
4176 hook returns true for both @code{ptr_mode} and @code{Pmode}.
4179 @deftypefn {Target Hook} bool TARGET_SCALAR_MODE_SUPPORTED_P (enum machine_mode @var{mode})
4180 Define this to return nonzero if the port is prepared to handle
4181 insns involving scalar mode @var{mode}. For a scalar mode to be
4182 considered supported, all the basic arithmetic and comparisons
4185 The default version of this hook returns true for any mode
4186 required to handle the basic C types (as defined by the port).
4187 Included here are the double-word arithmetic supported by the
4188 code in @file{optabs.c}.
4191 @deftypefn {Target Hook} bool TARGET_VECTOR_MODE_SUPPORTED_P (enum machine_mode @var{mode})
4192 Define this to return nonzero if the port is prepared to handle
4193 insns involving vector mode @var{mode}. At the very least, it
4194 must have move patterns for this mode.
4198 @subsection How Scalar Function Values Are Returned
4199 @cindex return values in registers
4200 @cindex values, returned by functions
4201 @cindex scalars, returned as values
4203 This section discusses the macros that control returning scalars as
4204 values---values that can fit in registers.
4206 @deftypefn {Target Hook} rtx TARGET_FUNCTION_VALUE (tree @var{ret_type}, tree @var{fn_decl_or_type}, bool @var{outgoing})
4208 Define this to return an RTX representing the place where a function
4209 returns or receives a value of data type @var{ret_type}, a tree node
4210 node representing a data type. @var{fn_decl_or_type} is a tree node
4211 representing @code{FUNCTION_DECL} or @code{FUNCTION_TYPE} of a
4212 function being called. If @var{outgoing} is false, the hook should
4213 compute the register in which the caller will see the return value.
4214 Otherwise, the hook should return an RTX representing the place where
4215 a function returns a value.
4217 On many machines, only @code{TYPE_MODE (@var{ret_type})} is relevant.
4218 (Actually, on most machines, scalar values are returned in the same
4219 place regardless of mode.) The value of the expression is usually a
4220 @code{reg} RTX for the hard register where the return value is stored.
4221 The value can also be a @code{parallel} RTX, if the return value is in
4222 multiple places. See @code{FUNCTION_ARG} for an explanation of the
4223 @code{parallel} form. Note that the callee will populate every
4224 location specified in the @code{parallel}, but if the first element of
4225 the @code{parallel} contains the whole return value, callers will use
4226 that element as the canonical location and ignore the others. The m68k
4227 port uses this type of @code{parallel} to return pointers in both
4228 @samp{%a0} (the canonical location) and @samp{%d0}.
4230 If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply
4231 the same promotion rules specified in @code{PROMOTE_MODE} if
4232 @var{valtype} is a scalar type.
4234 If the precise function being called is known, @var{func} is a tree
4235 node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
4236 pointer. This makes it possible to use a different value-returning
4237 convention for specific functions when all their calls are
4240 Some target machines have ``register windows'' so that the register in
4241 which a function returns its value is not the same as the one in which
4242 the caller sees the value. For such machines, you should return
4243 different RTX depending on @var{outgoing}.
4245 @code{TARGET_FUNCTION_VALUE} is not used for return values with
4246 aggregate data types, because these are returned in another way. See
4247 @code{TARGET_STRUCT_VALUE_RTX} and related macros, below.
4250 @defmac FUNCTION_VALUE (@var{valtype}, @var{func})
4251 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for
4252 a new target instead.
4255 @defmac FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
4256 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for
4257 a new target instead.
4260 @defmac LIBCALL_VALUE (@var{mode})
4261 A C expression to create an RTX representing the place where a library
4262 function returns a value of mode @var{mode}. If the precise function
4263 being called is known, @var{func} is a tree node
4264 (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
4265 pointer. This makes it possible to use a different value-returning
4266 convention for specific functions when all their calls are
4269 Note that ``library function'' in this context means a compiler
4270 support routine, used to perform arithmetic, whose name is known
4271 specially by the compiler and was not mentioned in the C code being
4274 The definition of @code{LIBRARY_VALUE} need not be concerned aggregate
4275 data types, because none of the library functions returns such types.
4278 @defmac FUNCTION_VALUE_REGNO_P (@var{regno})
4279 A C expression that is nonzero if @var{regno} is the number of a hard
4280 register in which the values of called function may come back.
4282 A register whose use for returning values is limited to serving as the
4283 second of a pair (for a value of type @code{double}, say) need not be
4284 recognized by this macro. So for most machines, this definition
4288 #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
4291 If the machine has register windows, so that the caller and the called
4292 function use different registers for the return value, this macro
4293 should recognize only the caller's register numbers.
4296 @defmac APPLY_RESULT_SIZE
4297 Define this macro if @samp{untyped_call} and @samp{untyped_return}
4298 need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for
4299 saving and restoring an arbitrary return value.
4302 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MSB (tree @var{type})
4303 This hook should return true if values of type @var{type} are returned
4304 at the most significant end of a register (in other words, if they are
4305 padded at the least significant end). You can assume that @var{type}
4306 is returned in a register; the caller is required to check this.
4308 Note that the register provided by @code{TARGET_FUNCTION_VALUE} must
4309 be able to hold the complete return value. For example, if a 1-, 2-
4310 or 3-byte structure is returned at the most significant end of a
4311 4-byte register, @code{TARGET_FUNCTION_VALUE} should provide an
4315 @node Aggregate Return
4316 @subsection How Large Values Are Returned
4317 @cindex aggregates as return values
4318 @cindex large return values
4319 @cindex returning aggregate values
4320 @cindex structure value address
4322 When a function value's mode is @code{BLKmode} (and in some other
4323 cases), the value is not returned according to
4324 @code{TARGET_FUNCTION_VALUE} (@pxref{Scalar Return}). Instead, the
4325 caller passes the address of a block of memory in which the value
4326 should be stored. This address is called the @dfn{structure value
4329 This section describes how to control returning structure values in
4332 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MEMORY (tree @var{type}, tree @var{fntype})
4333 This target hook should return a nonzero value to say to return the
4334 function value in memory, just as large structures are always returned.
4335 Here @var{type} will be the data type of the value, and @var{fntype}
4336 will be the type of the function doing the returning, or @code{NULL} for
4339 Note that values of mode @code{BLKmode} must be explicitly handled
4340 by this function. Also, the option @option{-fpcc-struct-return}
4341 takes effect regardless of this macro. On most systems, it is
4342 possible to leave the hook undefined; this causes a default
4343 definition to be used, whose value is the constant 1 for @code{BLKmode}
4344 values, and 0 otherwise.
4346 Do not use this hook to indicate that structures and unions should always
4347 be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN}
4351 @defmac DEFAULT_PCC_STRUCT_RETURN
4352 Define this macro to be 1 if all structure and union return values must be
4353 in memory. Since this results in slower code, this should be defined
4354 only if needed for compatibility with other compilers or with an ABI@.
4355 If you define this macro to be 0, then the conventions used for structure
4356 and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY}
4359 If not defined, this defaults to the value 1.
4362 @deftypefn {Target Hook} rtx TARGET_STRUCT_VALUE_RTX (tree @var{fndecl}, int @var{incoming})
4363 This target hook should return the location of the structure value
4364 address (normally a @code{mem} or @code{reg}), or 0 if the address is
4365 passed as an ``invisible'' first argument. Note that @var{fndecl} may
4366 be @code{NULL}, for libcalls. You do not need to define this target
4367 hook if the address is always passed as an ``invisible'' first
4370 On some architectures the place where the structure value address
4371 is found by the called function is not the same place that the
4372 caller put it. This can be due to register windows, or it could
4373 be because the function prologue moves it to a different place.
4374 @var{incoming} is @code{1} or @code{2} when the location is needed in
4375 the context of the called function, and @code{0} in the context of
4378 If @var{incoming} is nonzero and the address is to be found on the
4379 stack, return a @code{mem} which refers to the frame pointer. If
4380 @var{incoming} is @code{2}, the result is being used to fetch the
4381 structure value address at the beginning of a function. If you need
4382 to emit adjusting code, you should do it at this point.
4385 @defmac PCC_STATIC_STRUCT_RETURN
4386 Define this macro if the usual system convention on the target machine
4387 for returning structures and unions is for the called function to return
4388 the address of a static variable containing the value.
4390 Do not define this if the usual system convention is for the caller to
4391 pass an address to the subroutine.
4393 This macro has effect in @option{-fpcc-struct-return} mode, but it does
4394 nothing when you use @option{-freg-struct-return} mode.
4398 @subsection Caller-Saves Register Allocation
4400 If you enable it, GCC can save registers around function calls. This
4401 makes it possible to use call-clobbered registers to hold variables that
4402 must live across calls.
4404 @defmac CALLER_SAVE_PROFITABLE (@var{refs}, @var{calls})
4405 A C expression to determine whether it is worthwhile to consider placing
4406 a pseudo-register in a call-clobbered hard register and saving and
4407 restoring it around each function call. The expression should be 1 when
4408 this is worth doing, and 0 otherwise.
4410 If you don't define this macro, a default is used which is good on most
4411 machines: @code{4 * @var{calls} < @var{refs}}.
4414 @defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs})
4415 A C expression specifying which mode is required for saving @var{nregs}
4416 of a pseudo-register in call-clobbered hard register @var{regno}. If
4417 @var{regno} is unsuitable for caller save, @code{VOIDmode} should be
4418 returned. For most machines this macro need not be defined since GCC
4419 will select the smallest suitable mode.
4422 @node Function Entry
4423 @subsection Function Entry and Exit
4424 @cindex function entry and exit
4428 This section describes the macros that output function entry
4429 (@dfn{prologue}) and exit (@dfn{epilogue}) code.
4431 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_PROLOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4432 If defined, a function that outputs the assembler code for entry to a
4433 function. The prologue is responsible for setting up the stack frame,
4434 initializing the frame pointer register, saving registers that must be
4435 saved, and allocating @var{size} additional bytes of storage for the
4436 local variables. @var{size} is an integer. @var{file} is a stdio
4437 stream to which the assembler code should be output.
4439 The label for the beginning of the function need not be output by this
4440 macro. That has already been done when the macro is run.
4442 @findex regs_ever_live
4443 To determine which registers to save, the macro can refer to the array
4444 @code{regs_ever_live}: element @var{r} is nonzero if hard register
4445 @var{r} is used anywhere within the function. This implies the function
4446 prologue should save register @var{r}, provided it is not one of the
4447 call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use
4448 @code{regs_ever_live}.)
4450 On machines that have ``register windows'', the function entry code does
4451 not save on the stack the registers that are in the windows, even if
4452 they are supposed to be preserved by function calls; instead it takes
4453 appropriate steps to ``push'' the register stack, if any non-call-used
4454 registers are used in the function.
4456 @findex frame_pointer_needed
4457 On machines where functions may or may not have frame-pointers, the
4458 function entry code must vary accordingly; it must set up the frame
4459 pointer if one is wanted, and not otherwise. To determine whether a
4460 frame pointer is in wanted, the macro can refer to the variable
4461 @code{frame_pointer_needed}. The variable's value will be 1 at run
4462 time in a function that needs a frame pointer. @xref{Elimination}.
4464 The function entry code is responsible for allocating any stack space
4465 required for the function. This stack space consists of the regions
4466 listed below. In most cases, these regions are allocated in the
4467 order listed, with the last listed region closest to the top of the
4468 stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and
4469 the highest address if it is not defined). You can use a different order
4470 for a machine if doing so is more convenient or required for
4471 compatibility reasons. Except in cases where required by standard
4472 or by a debugger, there is no reason why the stack layout used by GCC
4473 need agree with that used by other compilers for a machine.
4476 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_END_PROLOGUE (FILE *@var{file})
4477 If defined, a function that outputs assembler code at the end of a
4478 prologue. This should be used when the function prologue is being
4479 emitted as RTL, and you have some extra assembler that needs to be
4480 emitted. @xref{prologue instruction pattern}.
4483 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_BEGIN_EPILOGUE (FILE *@var{file})
4484 If defined, a function that outputs assembler code at the start of an
4485 epilogue. This should be used when the function epilogue is being
4486 emitted as RTL, and you have some extra assembler that needs to be
4487 emitted. @xref{epilogue instruction pattern}.
4490 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_EPILOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4491 If defined, a function that outputs the assembler code for exit from a
4492 function. The epilogue is responsible for restoring the saved
4493 registers and stack pointer to their values when the function was
4494 called, and returning control to the caller. This macro takes the
4495 same arguments as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the
4496 registers to restore are determined from @code{regs_ever_live} and
4497 @code{CALL_USED_REGISTERS} in the same way.
4499 On some machines, there is a single instruction that does all the work
4500 of returning from the function. On these machines, give that
4501 instruction the name @samp{return} and do not define the macro
4502 @code{TARGET_ASM_FUNCTION_EPILOGUE} at all.
4504 Do not define a pattern named @samp{return} if you want the
4505 @code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target
4506 switches to control whether return instructions or epilogues are used,
4507 define a @samp{return} pattern with a validity condition that tests the
4508 target switches appropriately. If the @samp{return} pattern's validity
4509 condition is false, epilogues will be used.
4511 On machines where functions may or may not have frame-pointers, the
4512 function exit code must vary accordingly. Sometimes the code for these
4513 two cases is completely different. To determine whether a frame pointer
4514 is wanted, the macro can refer to the variable
4515 @code{frame_pointer_needed}. The variable's value will be 1 when compiling
4516 a function that needs a frame pointer.
4518 Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and
4519 @code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially.
4520 The C variable @code{current_function_is_leaf} is nonzero for such a
4521 function. @xref{Leaf Functions}.
4523 On some machines, some functions pop their arguments on exit while
4524 others leave that for the caller to do. For example, the 68020 when
4525 given @option{-mrtd} pops arguments in functions that take a fixed
4526 number of arguments.
4528 @findex current_function_pops_args
4529 Your definition of the macro @code{RETURN_POPS_ARGS} decides which
4530 functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE}
4531 needs to know what was decided. The variable that is called
4532 @code{current_function_pops_args} is the number of bytes of its
4533 arguments that a function should pop. @xref{Scalar Return}.
4534 @c what is the "its arguments" in the above sentence referring to, pray
4535 @c tell? --mew 5feb93
4540 @findex current_function_pretend_args_size
4541 A region of @code{current_function_pretend_args_size} bytes of
4542 uninitialized space just underneath the first argument arriving on the
4543 stack. (This may not be at the very start of the allocated stack region
4544 if the calling sequence has pushed anything else since pushing the stack
4545 arguments. But usually, on such machines, nothing else has been pushed
4546 yet, because the function prologue itself does all the pushing.) This
4547 region is used on machines where an argument may be passed partly in
4548 registers and partly in memory, and, in some cases to support the
4549 features in @code{<stdarg.h>}.
4552 An area of memory used to save certain registers used by the function.
4553 The size of this area, which may also include space for such things as
4554 the return address and pointers to previous stack frames, is
4555 machine-specific and usually depends on which registers have been used
4556 in the function. Machines with register windows often do not require
4560 A region of at least @var{size} bytes, possibly rounded up to an allocation
4561 boundary, to contain the local variables of the function. On some machines,
4562 this region and the save area may occur in the opposite order, with the
4563 save area closer to the top of the stack.
4566 @cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames
4567 Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of
4568 @code{current_function_outgoing_args_size} bytes to be used for outgoing
4569 argument lists of the function. @xref{Stack Arguments}.
4572 @defmac EXIT_IGNORE_STACK
4573 Define this macro as a C expression that is nonzero if the return
4574 instruction or the function epilogue ignores the value of the stack
4575 pointer; in other words, if it is safe to delete an instruction to
4576 adjust the stack pointer before a return from the function. The
4579 Note that this macro's value is relevant only for functions for which
4580 frame pointers are maintained. It is never safe to delete a final
4581 stack adjustment in a function that has no frame pointer, and the
4582 compiler knows this regardless of @code{EXIT_IGNORE_STACK}.
4585 @defmac EPILOGUE_USES (@var{regno})
4586 Define this macro as a C expression that is nonzero for registers that are
4587 used by the epilogue or the @samp{return} pattern. The stack and frame
4588 pointer registers are already assumed to be used as needed.
4591 @defmac EH_USES (@var{regno})
4592 Define this macro as a C expression that is nonzero for registers that are
4593 used by the exception handling mechanism, and so should be considered live
4594 on entry to an exception edge.
4597 @defmac DELAY_SLOTS_FOR_EPILOGUE
4598 Define this macro if the function epilogue contains delay slots to which
4599 instructions from the rest of the function can be ``moved''. The
4600 definition should be a C expression whose value is an integer
4601 representing the number of delay slots there.
4604 @defmac ELIGIBLE_FOR_EPILOGUE_DELAY (@var{insn}, @var{n})
4605 A C expression that returns 1 if @var{insn} can be placed in delay
4606 slot number @var{n} of the epilogue.
4608 The argument @var{n} is an integer which identifies the delay slot now
4609 being considered (since different slots may have different rules of
4610 eligibility). It is never negative and is always less than the number
4611 of epilogue delay slots (what @code{DELAY_SLOTS_FOR_EPILOGUE} returns).
4612 If you reject a particular insn for a given delay slot, in principle, it
4613 may be reconsidered for a subsequent delay slot. Also, other insns may
4614 (at least in principle) be considered for the so far unfilled delay
4617 @findex current_function_epilogue_delay_list
4618 @findex final_scan_insn
4619 The insns accepted to fill the epilogue delay slots are put in an RTL
4620 list made with @code{insn_list} objects, stored in the variable
4621 @code{current_function_epilogue_delay_list}. The insn for the first
4622 delay slot comes first in the list. Your definition of the macro
4623 @code{TARGET_ASM_FUNCTION_EPILOGUE} should fill the delay slots by
4624 outputting the insns in this list, usually by calling
4625 @code{final_scan_insn}.
4627 You need not define this macro if you did not define
4628 @code{DELAY_SLOTS_FOR_EPILOGUE}.
4631 @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})
4632 A function that outputs the assembler code for a thunk
4633 function, used to implement C++ virtual function calls with multiple
4634 inheritance. The thunk acts as a wrapper around a virtual function,
4635 adjusting the implicit object parameter before handing control off to
4638 First, emit code to add the integer @var{delta} to the location that
4639 contains the incoming first argument. Assume that this argument
4640 contains a pointer, and is the one used to pass the @code{this} pointer
4641 in C++. This is the incoming argument @emph{before} the function prologue,
4642 e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of
4643 all other incoming arguments.
4645 Then, if @var{vcall_offset} is nonzero, an additional adjustment should be
4646 made after adding @code{delta}. In particular, if @var{p} is the
4647 adjusted pointer, the following adjustment should be made:
4650 p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)]
4653 After the additions, emit code to jump to @var{function}, which is a
4654 @code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does
4655 not touch the return address. Hence returning from @var{FUNCTION} will
4656 return to whoever called the current @samp{thunk}.
4658 The effect must be as if @var{function} had been called directly with
4659 the adjusted first argument. This macro is responsible for emitting all
4660 of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE}
4661 and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked.
4663 The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function}
4664 have already been extracted from it.) It might possibly be useful on
4665 some targets, but probably not.
4667 If you do not define this macro, the target-independent code in the C++
4668 front end will generate a less efficient heavyweight thunk that calls
4669 @var{function} instead of jumping to it. The generic approach does
4670 not support varargs.
4673 @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})
4674 A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able
4675 to output the assembler code for the thunk function specified by the
4676 arguments it is passed, and false otherwise. In the latter case, the
4677 generic approach will be used by the C++ front end, with the limitations
4682 @subsection Generating Code for Profiling
4683 @cindex profiling, code generation
4685 These macros will help you generate code for profiling.
4687 @defmac FUNCTION_PROFILER (@var{file}, @var{labelno})
4688 A C statement or compound statement to output to @var{file} some
4689 assembler code to call the profiling subroutine @code{mcount}.
4692 The details of how @code{mcount} expects to be called are determined by
4693 your operating system environment, not by GCC@. To figure them out,
4694 compile a small program for profiling using the system's installed C
4695 compiler and look at the assembler code that results.
4697 Older implementations of @code{mcount} expect the address of a counter
4698 variable to be loaded into some register. The name of this variable is
4699 @samp{LP} followed by the number @var{labelno}, so you would generate
4700 the name using @samp{LP%d} in a @code{fprintf}.
4703 @defmac PROFILE_HOOK
4704 A C statement or compound statement to output to @var{file} some assembly
4705 code to call the profiling subroutine @code{mcount} even the target does
4706 not support profiling.
4709 @defmac NO_PROFILE_COUNTERS
4710 Define this macro to be an expression with a nonzero value if the
4711 @code{mcount} subroutine on your system does not need a counter variable
4712 allocated for each function. This is true for almost all modern
4713 implementations. If you define this macro, you must not use the
4714 @var{labelno} argument to @code{FUNCTION_PROFILER}.
4717 @defmac PROFILE_BEFORE_PROLOGUE
4718 Define this macro if the code for function profiling should come before
4719 the function prologue. Normally, the profiling code comes after.
4723 @subsection Permitting tail calls
4726 @deftypefn {Target Hook} bool TARGET_FUNCTION_OK_FOR_SIBCALL (tree @var{decl}, tree @var{exp})
4727 True if it is ok to do sibling call optimization for the specified
4728 call expression @var{exp}. @var{decl} will be the called function,
4729 or @code{NULL} if this is an indirect call.
4731 It is not uncommon for limitations of calling conventions to prevent
4732 tail calls to functions outside the current unit of translation, or
4733 during PIC compilation. The hook is used to enforce these restrictions,
4734 as the @code{sibcall} md pattern can not fail, or fall over to a
4735 ``normal'' call. The criteria for successful sibling call optimization
4736 may vary greatly between different architectures.
4739 @deftypefn {Target Hook} void TARGET_EXTRA_LIVE_ON_ENTRY (bitmap *@var{regs})
4740 Add any hard registers to @var{regs} that are live on entry to the
4741 function. This hook only needs to be defined to provide registers that
4742 cannot be found by examination of FUNCTION_ARG_REGNO_P, the callee saved
4743 registers, STATIC_CHAIN_INCOMING_REGNUM, STATIC_CHAIN_REGNUM,
4744 TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM, EH_USES,
4745 FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM.
4748 @node Stack Smashing Protection
4749 @subsection Stack smashing protection
4750 @cindex stack smashing protection
4752 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_GUARD (void)
4753 This hook returns a @code{DECL} node for the external variable to use
4754 for the stack protection guard. This variable is initialized by the
4755 runtime to some random value and is used to initialize the guard value
4756 that is placed at the top of the local stack frame. The type of this
4757 variable must be @code{ptr_type_node}.
4759 The default version of this hook creates a variable called
4760 @samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}.
4763 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_FAIL (void)
4764 This hook returns a tree expression that alerts the runtime that the
4765 stack protect guard variable has been modified. This expression should
4766 involve a call to a @code{noreturn} function.
4768 The default version of this hook invokes a function called
4769 @samp{__stack_chk_fail}, taking no arguments. This function is
4770 normally defined in @file{libgcc2.c}.
4774 @section Implementing the Varargs Macros
4775 @cindex varargs implementation
4777 GCC comes with an implementation of @code{<varargs.h>} and
4778 @code{<stdarg.h>} that work without change on machines that pass arguments
4779 on the stack. Other machines require their own implementations of
4780 varargs, and the two machine independent header files must have
4781 conditionals to include it.
4783 ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in
4784 the calling convention for @code{va_start}. The traditional
4785 implementation takes just one argument, which is the variable in which
4786 to store the argument pointer. The ISO implementation of
4787 @code{va_start} takes an additional second argument. The user is
4788 supposed to write the last named argument of the function here.
4790 However, @code{va_start} should not use this argument. The way to find
4791 the end of the named arguments is with the built-in functions described
4794 @defmac __builtin_saveregs ()
4795 Use this built-in function to save the argument registers in memory so
4796 that the varargs mechanism can access them. Both ISO and traditional
4797 versions of @code{va_start} must use @code{__builtin_saveregs}, unless
4798 you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead.
4800 On some machines, @code{__builtin_saveregs} is open-coded under the
4801 control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On
4802 other machines, it calls a routine written in assembler language,
4803 found in @file{libgcc2.c}.
4805 Code generated for the call to @code{__builtin_saveregs} appears at the
4806 beginning of the function, as opposed to where the call to
4807 @code{__builtin_saveregs} is written, regardless of what the code is.
4808 This is because the registers must be saved before the function starts
4809 to use them for its own purposes.
4810 @c i rewrote the first sentence above to fix an overfull hbox. --mew
4814 @defmac __builtin_args_info (@var{category})
4815 Use this built-in function to find the first anonymous arguments in
4818 In general, a machine may have several categories of registers used for
4819 arguments, each for a particular category of data types. (For example,
4820 on some machines, floating-point registers are used for floating-point
4821 arguments while other arguments are passed in the general registers.)
4822 To make non-varargs functions use the proper calling convention, you
4823 have defined the @code{CUMULATIVE_ARGS} data type to record how many
4824 registers in each category have been used so far
4826 @code{__builtin_args_info} accesses the same data structure of type
4827 @code{CUMULATIVE_ARGS} after the ordinary argument layout is finished
4828 with it, with @var{category} specifying which word to access. Thus, the
4829 value indicates the first unused register in a given category.
4831 Normally, you would use @code{__builtin_args_info} in the implementation
4832 of @code{va_start}, accessing each category just once and storing the
4833 value in the @code{va_list} object. This is because @code{va_list} will
4834 have to update the values, and there is no way to alter the
4835 values accessed by @code{__builtin_args_info}.
4838 @defmac __builtin_next_arg (@var{lastarg})
4839 This is the equivalent of @code{__builtin_args_info}, for stack
4840 arguments. It returns the address of the first anonymous stack
4841 argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it
4842 returns the address of the location above the first anonymous stack
4843 argument. Use it in @code{va_start} to initialize the pointer for
4844 fetching arguments from the stack. Also use it in @code{va_start} to
4845 verify that the second parameter @var{lastarg} is the last named argument
4846 of the current function.
4849 @defmac __builtin_classify_type (@var{object})
4850 Since each machine has its own conventions for which data types are
4851 passed in which kind of register, your implementation of @code{va_arg}
4852 has to embody these conventions. The easiest way to categorize the
4853 specified data type is to use @code{__builtin_classify_type} together
4854 with @code{sizeof} and @code{__alignof__}.
4856 @code{__builtin_classify_type} ignores the value of @var{object},
4857 considering only its data type. It returns an integer describing what
4858 kind of type that is---integer, floating, pointer, structure, and so on.
4860 The file @file{typeclass.h} defines an enumeration that you can use to
4861 interpret the values of @code{__builtin_classify_type}.
4864 These machine description macros help implement varargs:
4866 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN_SAVEREGS (void)
4867 If defined, this hook produces the machine-specific code for a call to
4868 @code{__builtin_saveregs}. This code will be moved to the very
4869 beginning of the function, before any parameter access are made. The
4870 return value of this function should be an RTX that contains the value
4871 to use as the return of @code{__builtin_saveregs}.
4874 @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})
4875 This target hook offers an alternative to using
4876 @code{__builtin_saveregs} and defining the hook
4877 @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous
4878 register arguments into the stack so that all the arguments appear to
4879 have been passed consecutively on the stack. Once this is done, you can
4880 use the standard implementation of varargs that works for machines that
4881 pass all their arguments on the stack.
4883 The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data
4884 structure, containing the values that are obtained after processing the
4885 named arguments. The arguments @var{mode} and @var{type} describe the
4886 last named argument---its machine mode and its data type as a tree node.
4888 The target hook should do two things: first, push onto the stack all the
4889 argument registers @emph{not} used for the named arguments, and second,
4890 store the size of the data thus pushed into the @code{int}-valued
4891 variable pointed to by @var{pretend_args_size}. The value that you
4892 store here will serve as additional offset for setting up the stack
4895 Because you must generate code to push the anonymous arguments at
4896 compile time without knowing their data types,
4897 @code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that
4898 have just a single category of argument register and use it uniformly
4901 If the argument @var{second_time} is nonzero, it means that the
4902 arguments of the function are being analyzed for the second time. This
4903 happens for an inline function, which is not actually compiled until the
4904 end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should
4905 not generate any instructions in this case.
4908 @deftypefn {Target Hook} bool TARGET_STRICT_ARGUMENT_NAMING (CUMULATIVE_ARGS *@var{ca})
4909 Define this hook to return @code{true} if the location where a function
4910 argument is passed depends on whether or not it is a named argument.
4912 This hook controls how the @var{named} argument to @code{FUNCTION_ARG}
4913 is set for varargs and stdarg functions. If this hook returns
4914 @code{true}, the @var{named} argument is always true for named
4915 arguments, and false for unnamed arguments. If it returns @code{false},
4916 but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true},
4917 then all arguments are treated as named. Otherwise, all named arguments
4918 except the last are treated as named.
4920 You need not define this hook if it always returns zero.
4923 @deftypefn {Target Hook} bool TARGET_PRETEND_OUTGOING_VARARGS_NAMED
4924 If you need to conditionally change ABIs so that one works with
4925 @code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither
4926 @code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was
4927 defined, then define this hook to return @code{true} if
4928 @code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise.
4929 Otherwise, you should not define this hook.
4933 @section Trampolines for Nested Functions
4934 @cindex trampolines for nested functions
4935 @cindex nested functions, trampolines for
4937 A @dfn{trampoline} is a small piece of code that is created at run time
4938 when the address of a nested function is taken. It normally resides on
4939 the stack, in the stack frame of the containing function. These macros
4940 tell GCC how to generate code to allocate and initialize a
4943 The instructions in the trampoline must do two things: load a constant
4944 address into the static chain register, and jump to the real address of
4945 the nested function. On CISC machines such as the m68k, this requires
4946 two instructions, a move immediate and a jump. Then the two addresses
4947 exist in the trampoline as word-long immediate operands. On RISC
4948 machines, it is often necessary to load each address into a register in
4949 two parts. Then pieces of each address form separate immediate
4952 The code generated to initialize the trampoline must store the variable
4953 parts---the static chain value and the function address---into the
4954 immediate operands of the instructions. On a CISC machine, this is
4955 simply a matter of copying each address to a memory reference at the
4956 proper offset from the start of the trampoline. On a RISC machine, it
4957 may be necessary to take out pieces of the address and store them
4960 @defmac TRAMPOLINE_TEMPLATE (@var{file})
4961 A C statement to output, on the stream @var{file}, assembler code for a
4962 block of data that contains the constant parts of a trampoline. This
4963 code should not include a label---the label is taken care of
4966 If you do not define this macro, it means no template is needed
4967 for the target. Do not define this macro on systems where the block move
4968 code to copy the trampoline into place would be larger than the code
4969 to generate it on the spot.
4972 @defmac TRAMPOLINE_SECTION
4973 Return the section into which the trampoline template is to be placed
4974 (@pxref{Sections}). The default value is @code{readonly_data_section}.
4977 @defmac TRAMPOLINE_SIZE
4978 A C expression for the size in bytes of the trampoline, as an integer.
4981 @defmac TRAMPOLINE_ALIGNMENT
4982 Alignment required for trampolines, in bits.
4984 If you don't define this macro, the value of @code{BIGGEST_ALIGNMENT}
4985 is used for aligning trampolines.
4988 @defmac INITIALIZE_TRAMPOLINE (@var{addr}, @var{fnaddr}, @var{static_chain})
4989 A C statement to initialize the variable parts of a trampoline.
4990 @var{addr} is an RTX for the address of the trampoline; @var{fnaddr} is
4991 an RTX for the address of the nested function; @var{static_chain} is an
4992 RTX for the static chain value that should be passed to the function
4996 @defmac TRAMPOLINE_ADJUST_ADDRESS (@var{addr})
4997 A C statement that should perform any machine-specific adjustment in
4998 the address of the trampoline. Its argument contains the address that
4999 was passed to @code{INITIALIZE_TRAMPOLINE}. In case the address to be
5000 used for a function call should be different from the address in which
5001 the template was stored, the different address should be assigned to
5002 @var{addr}. If this macro is not defined, @var{addr} will be used for
5005 @cindex @code{TARGET_ASM_FUNCTION_EPILOGUE} and trampolines
5006 @cindex @code{TARGET_ASM_FUNCTION_PROLOGUE} and trampolines
5007 If this macro is not defined, by default the trampoline is allocated as
5008 a stack slot. This default is right for most machines. The exceptions
5009 are machines where it is impossible to execute instructions in the stack
5010 area. On such machines, you may have to implement a separate stack,
5011 using this macro in conjunction with @code{TARGET_ASM_FUNCTION_PROLOGUE}
5012 and @code{TARGET_ASM_FUNCTION_EPILOGUE}.
5014 @var{fp} points to a data structure, a @code{struct function}, which
5015 describes the compilation status of the immediate containing function of
5016 the function which the trampoline is for. The stack slot for the
5017 trampoline is in the stack frame of this containing function. Other
5018 allocation strategies probably must do something analogous with this
5022 Implementing trampolines is difficult on many machines because they have
5023 separate instruction and data caches. Writing into a stack location
5024 fails to clear the memory in the instruction cache, so when the program
5025 jumps to that location, it executes the old contents.
5027 Here are two possible solutions. One is to clear the relevant parts of
5028 the instruction cache whenever a trampoline is set up. The other is to
5029 make all trampolines identical, by having them jump to a standard
5030 subroutine. The former technique makes trampoline execution faster; the
5031 latter makes initialization faster.
5033 To clear the instruction cache when a trampoline is initialized, define
5034 the following macro.
5036 @defmac CLEAR_INSN_CACHE (@var{beg}, @var{end})
5037 If defined, expands to a C expression clearing the @emph{instruction
5038 cache} in the specified interval. The definition of this macro would
5039 typically be a series of @code{asm} statements. Both @var{beg} and
5040 @var{end} are both pointer expressions.
5043 The operating system may also require the stack to be made executable
5044 before calling the trampoline. To implement this requirement, define
5045 the following macro.
5047 @defmac ENABLE_EXECUTE_STACK
5048 Define this macro if certain operations must be performed before executing
5049 code located on the stack. The macro should expand to a series of C
5050 file-scope constructs (e.g.@: functions) and provide a unique entry point
5051 named @code{__enable_execute_stack}. The target is responsible for
5052 emitting calls to the entry point in the code, for example from the
5053 @code{INITIALIZE_TRAMPOLINE} macro.
5056 To use a standard subroutine, define the following macro. In addition,
5057 you must make sure that the instructions in a trampoline fill an entire
5058 cache line with identical instructions, or else ensure that the
5059 beginning of the trampoline code is always aligned at the same point in
5060 its cache line. Look in @file{m68k.h} as a guide.
5062 @defmac TRANSFER_FROM_TRAMPOLINE
5063 Define this macro if trampolines need a special subroutine to do their
5064 work. The macro should expand to a series of @code{asm} statements
5065 which will be compiled with GCC@. They go in a library function named
5066 @code{__transfer_from_trampoline}.
5068 If you need to avoid executing the ordinary prologue code of a compiled
5069 C function when you jump to the subroutine, you can do so by placing a
5070 special label of your own in the assembler code. Use one @code{asm}
5071 statement to generate an assembler label, and another to make the label
5072 global. Then trampolines can use that label to jump directly to your
5073 special assembler code.
5077 @section Implicit Calls to Library Routines
5078 @cindex library subroutine names
5079 @cindex @file{libgcc.a}
5081 @c prevent bad page break with this line
5082 Here is an explanation of implicit calls to library routines.
5084 @defmac DECLARE_LIBRARY_RENAMES
5085 This macro, if defined, should expand to a piece of C code that will get
5086 expanded when compiling functions for libgcc.a. It can be used to
5087 provide alternate names for GCC's internal library functions if there
5088 are ABI-mandated names that the compiler should provide.
5091 @findex init_one_libfunc
5092 @findex set_optab_libfunc
5093 @deftypefn {Target Hook} void TARGET_INIT_LIBFUNCS (void)
5094 This hook should declare additional library routines or rename
5095 existing ones, using the functions @code{set_optab_libfunc} and
5096 @code{init_one_libfunc} defined in @file{optabs.c}.
5097 @code{init_optabs} calls this macro after initializing all the normal
5100 The default is to do nothing. Most ports don't need to define this hook.
5103 @defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison})
5104 This macro should return @code{true} if the library routine that
5105 implements the floating point comparison operator @var{comparison} in
5106 mode @var{mode} will return a boolean, and @var{false} if it will
5109 GCC's own floating point libraries return tristates from the
5110 comparison operators, so the default returns false always. Most ports
5111 don't need to define this macro.
5114 @defmac TARGET_LIB_INT_CMP_BIASED
5115 This macro should evaluate to @code{true} if the integer comparison
5116 functions (like @code{__cmpdi2}) return 0 to indicate that the first
5117 operand is smaller than the second, 1 to indicate that they are equal,
5118 and 2 to indicate that the first operand is greater than the second.
5119 If this macro evaluates to @code{false} the comparison functions return
5120 @minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines
5121 in @file{libgcc.a}, you do not need to define this macro.
5124 @cindex US Software GOFAST, floating point emulation library
5125 @cindex floating point emulation library, US Software GOFAST
5126 @cindex GOFAST, floating point emulation library
5127 @findex gofast_maybe_init_libfuncs
5128 @defmac US_SOFTWARE_GOFAST
5129 Define this macro if your system C library uses the US Software GOFAST
5130 library to provide floating point emulation.
5132 In addition to defining this macro, your architecture must set
5133 @code{TARGET_INIT_LIBFUNCS} to @code{gofast_maybe_init_libfuncs}, or
5134 else call that function from its version of that hook. It is defined
5135 in @file{config/gofast.h}, which must be included by your
5136 architecture's @file{@var{cpu}.c} file. See @file{sparc/sparc.c} for
5139 If this macro is defined, the
5140 @code{TARGET_FLOAT_LIB_COMPARE_RETURNS_BOOL} target hook must return
5141 false for @code{SFmode} and @code{DFmode} comparisons.
5144 @cindex @code{EDOM}, implicit usage
5147 The value of @code{EDOM} on the target machine, as a C integer constant
5148 expression. If you don't define this macro, GCC does not attempt to
5149 deposit the value of @code{EDOM} into @code{errno} directly. Look in
5150 @file{/usr/include/errno.h} to find the value of @code{EDOM} on your
5153 If you do not define @code{TARGET_EDOM}, then compiled code reports
5154 domain errors by calling the library function and letting it report the
5155 error. If mathematical functions on your system use @code{matherr} when
5156 there is an error, then you should leave @code{TARGET_EDOM} undefined so
5157 that @code{matherr} is used normally.
5160 @cindex @code{errno}, implicit usage
5161 @defmac GEN_ERRNO_RTX
5162 Define this macro as a C expression to create an rtl expression that
5163 refers to the global ``variable'' @code{errno}. (On certain systems,
5164 @code{errno} may not actually be a variable.) If you don't define this
5165 macro, a reasonable default is used.
5168 @cindex C99 math functions, implicit usage
5169 @defmac TARGET_C99_FUNCTIONS
5170 When this macro is nonzero, GCC will implicitly optimize @code{sin} calls into
5171 @code{sinf} and similarly for other functions defined by C99 standard. The
5172 default is nonzero that should be proper value for most modern systems, however
5173 number of existing systems lacks support for these functions in the runtime so
5174 they needs this macro to be redefined to 0.
5177 @cindex sincos math function, implicit usage
5178 @defmac TARGET_HAS_SINCOS
5179 When this macro is nonzero, GCC will implicitly optimize calls to @code{sin}
5180 and @code{cos} with the same argument to a call to @code{sincos}. The
5181 default is zero. The target has to provide the following functions:
5183 void sincos(double x, double *sin, double *cos);
5184 void sincosf(float x, float *sin, float *cos);
5185 void sincosl(long double x, long double *sin, long double *cos);
5189 @defmac NEXT_OBJC_RUNTIME
5190 Define this macro to generate code for Objective-C message sending using
5191 the calling convention of the NeXT system. This calling convention
5192 involves passing the object, the selector and the method arguments all
5193 at once to the method-lookup library function.
5195 The default calling convention passes just the object and the selector
5196 to the lookup function, which returns a pointer to the method.
5199 @node Addressing Modes
5200 @section Addressing Modes
5201 @cindex addressing modes
5203 @c prevent bad page break with this line
5204 This is about addressing modes.
5206 @defmac HAVE_PRE_INCREMENT
5207 @defmacx HAVE_PRE_DECREMENT
5208 @defmacx HAVE_POST_INCREMENT
5209 @defmacx HAVE_POST_DECREMENT
5210 A C expression that is nonzero if the machine supports pre-increment,
5211 pre-decrement, post-increment, or post-decrement addressing respectively.
5214 @defmac HAVE_PRE_MODIFY_DISP
5215 @defmacx HAVE_POST_MODIFY_DISP
5216 A C expression that is nonzero if the machine supports pre- or
5217 post-address side-effect generation involving constants other than
5218 the size of the memory operand.
5221 @defmac HAVE_PRE_MODIFY_REG
5222 @defmacx HAVE_POST_MODIFY_REG
5223 A C expression that is nonzero if the machine supports pre- or
5224 post-address side-effect generation involving a register displacement.
5227 @defmac CONSTANT_ADDRESS_P (@var{x})
5228 A C expression that is 1 if the RTX @var{x} is a constant which
5229 is a valid address. On most machines, this can be defined as
5230 @code{CONSTANT_P (@var{x})}, but a few machines are more restrictive
5231 in which constant addresses are supported.
5234 @defmac CONSTANT_P (@var{x})
5235 @code{CONSTANT_P}, which is defined by target-independent code,
5236 accepts integer-values expressions whose values are not explicitly
5237 known, such as @code{symbol_ref}, @code{label_ref}, and @code{high}
5238 expressions and @code{const} arithmetic expressions, in addition to
5239 @code{const_int} and @code{const_double} expressions.
5242 @defmac MAX_REGS_PER_ADDRESS
5243 A number, the maximum number of registers that can appear in a valid
5244 memory address. Note that it is up to you to specify a value equal to
5245 the maximum number that @code{GO_IF_LEGITIMATE_ADDRESS} would ever
5249 @defmac GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
5250 A C compound statement with a conditional @code{goto @var{label};}
5251 executed if @var{x} (an RTX) is a legitimate memory address on the
5252 target machine for a memory operand of mode @var{mode}.
5254 It usually pays to define several simpler macros to serve as
5255 subroutines for this one. Otherwise it may be too complicated to
5258 This macro must exist in two variants: a strict variant and a
5259 non-strict one. The strict variant is used in the reload pass. It
5260 must be defined so that any pseudo-register that has not been
5261 allocated a hard register is considered a memory reference. In
5262 contexts where some kind of register is required, a pseudo-register
5263 with no hard register must be rejected.
5265 The non-strict variant is used in other passes. It must be defined to
5266 accept all pseudo-registers in every context where some kind of
5267 register is required.
5269 @findex REG_OK_STRICT
5270 Compiler source files that want to use the strict variant of this
5271 macro define the macro @code{REG_OK_STRICT}. You should use an
5272 @code{#ifdef REG_OK_STRICT} conditional to define the strict variant
5273 in that case and the non-strict variant otherwise.
5275 Subroutines to check for acceptable registers for various purposes (one
5276 for base registers, one for index registers, and so on) are typically
5277 among the subroutines used to define @code{GO_IF_LEGITIMATE_ADDRESS}.
5278 Then only these subroutine macros need have two variants; the higher
5279 levels of macros may be the same whether strict or not.
5281 Normally, constant addresses which are the sum of a @code{symbol_ref}
5282 and an integer are stored inside a @code{const} RTX to mark them as
5283 constant. Therefore, there is no need to recognize such sums
5284 specifically as legitimate addresses. Normally you would simply
5285 recognize any @code{const} as legitimate.
5287 Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
5288 sums that are not marked with @code{const}. It assumes that a naked
5289 @code{plus} indicates indexing. If so, then you @emph{must} reject such
5290 naked constant sums as illegitimate addresses, so that none of them will
5291 be given to @code{PRINT_OPERAND_ADDRESS}.
5293 @cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation
5294 On some machines, whether a symbolic address is legitimate depends on
5295 the section that the address refers to. On these machines, define the
5296 target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information
5297 into the @code{symbol_ref}, and then check for it here. When you see a
5298 @code{const}, you will have to look inside it to find the
5299 @code{symbol_ref} in order to determine the section. @xref{Assembler
5303 @defmac FIND_BASE_TERM (@var{x})
5304 A C expression to determine the base term of address @var{x}.
5305 This macro is used in only one place: `find_base_term' in alias.c.
5307 It is always safe for this macro to not be defined. It exists so
5308 that alias analysis can understand machine-dependent addresses.
5310 The typical use of this macro is to handle addresses containing
5311 a label_ref or symbol_ref within an UNSPEC@.
5314 @defmac LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win})
5315 A C compound statement that attempts to replace @var{x} with a valid
5316 memory address for an operand of mode @var{mode}. @var{win} will be a
5317 C statement label elsewhere in the code; the macro definition may use
5320 GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win});
5324 to avoid further processing if the address has become legitimate.
5326 @findex break_out_memory_refs
5327 @var{x} will always be the result of a call to @code{break_out_memory_refs},
5328 and @var{oldx} will be the operand that was given to that function to produce
5331 The code generated by this macro should not alter the substructure of
5332 @var{x}. If it transforms @var{x} into a more legitimate form, it
5333 should assign @var{x} (which will always be a C variable) a new value.
5335 It is not necessary for this macro to come up with a legitimate
5336 address. The compiler has standard ways of doing so in all cases. In
5337 fact, it is safe to omit this macro. But often a
5338 machine-dependent strategy can generate better code.
5341 @defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win})
5342 A C compound statement that attempts to replace @var{x}, which is an address
5343 that needs reloading, with a valid memory address for an operand of mode
5344 @var{mode}. @var{win} will be a C statement label elsewhere in the code.
5345 It is not necessary to define this macro, but it might be useful for
5346 performance reasons.
5348 For example, on the i386, it is sometimes possible to use a single
5349 reload register instead of two by reloading a sum of two pseudo
5350 registers into a register. On the other hand, for number of RISC
5351 processors offsets are limited so that often an intermediate address
5352 needs to be generated in order to address a stack slot. By defining
5353 @code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses
5354 generated for adjacent some stack slots can be made identical, and thus
5357 @emph{Note}: This macro should be used with caution. It is necessary
5358 to know something of how reload works in order to effectively use this,
5359 and it is quite easy to produce macros that build in too much knowledge
5360 of reload internals.
5362 @emph{Note}: This macro must be able to reload an address created by a
5363 previous invocation of this macro. If it fails to handle such addresses
5364 then the compiler may generate incorrect code or abort.
5367 The macro definition should use @code{push_reload} to indicate parts that
5368 need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually
5369 suitable to be passed unaltered to @code{push_reload}.
5371 The code generated by this macro must not alter the substructure of
5372 @var{x}. If it transforms @var{x} into a more legitimate form, it
5373 should assign @var{x} (which will always be a C variable) a new value.
5374 This also applies to parts that you change indirectly by calling
5377 @findex strict_memory_address_p
5378 The macro definition may use @code{strict_memory_address_p} to test if
5379 the address has become legitimate.
5382 If you want to change only a part of @var{x}, one standard way of doing
5383 this is to use @code{copy_rtx}. Note, however, that it unshares only a
5384 single level of rtl. Thus, if the part to be changed is not at the
5385 top level, you'll need to replace first the top level.
5386 It is not necessary for this macro to come up with a legitimate
5387 address; but often a machine-dependent strategy can generate better code.
5390 @defmac GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
5391 A C statement or compound statement with a conditional @code{goto
5392 @var{label};} executed if memory address @var{x} (an RTX) can have
5393 different meanings depending on the machine mode of the memory
5394 reference it is used for or if the address is valid for some modes
5397 Autoincrement and autodecrement addresses typically have mode-dependent
5398 effects because the amount of the increment or decrement is the size
5399 of the operand being addressed. Some machines have other mode-dependent
5400 addresses. Many RISC machines have no mode-dependent addresses.
5402 You may assume that @var{addr} is a valid address for the machine.
5405 @defmac LEGITIMATE_CONSTANT_P (@var{x})
5406 A C expression that is nonzero if @var{x} is a legitimate constant for
5407 an immediate operand on the target machine. You can assume that
5408 @var{x} satisfies @code{CONSTANT_P}, so you need not check this. In fact,
5409 @samp{1} is a suitable definition for this macro on machines where
5410 anything @code{CONSTANT_P} is valid.
5413 @deftypefn {Target Hook} rtx TARGET_DELEGITIMIZE_ADDRESS (rtx @var{x})
5414 This hook is used to undo the possibly obfuscating effects of the
5415 @code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target
5416 macros. Some backend implementations of these macros wrap symbol
5417 references inside an @code{UNSPEC} rtx to represent PIC or similar
5418 addressing modes. This target hook allows GCC's optimizers to understand
5419 the semantics of these opaque @code{UNSPEC}s by converting them back
5420 into their original form.
5423 @deftypefn {Target Hook} bool TARGET_CANNOT_FORCE_CONST_MEM (rtx @var{x})
5424 This hook should return true if @var{x} is of a form that cannot (or
5425 should not) be spilled to the constant pool. The default version of
5426 this hook returns false.
5428 The primary reason to define this hook is to prevent reload from
5429 deciding that a non-legitimate constant would be better reloaded
5430 from the constant pool instead of spilling and reloading a register
5431 holding the constant. This restriction is often true of addresses
5432 of TLS symbols for various targets.
5435 @deftypefn {Target Hook} bool TARGET_USE_BLOCKS_FOR_CONSTANT_P (enum machine_mode @var{mode}, rtx @var{x})
5436 This hook should return true if pool entries for constant @var{x} can
5437 be placed in an @code{object_block} structure. @var{mode} is the mode
5440 The default version returns false for all constants.
5443 @deftypefn {Target Hook} tree TARGET_BUILTIN_RECIPROCAL (enum tree_code @var{fn}, bool @var{tm_fn}, bool @var{sqrt})
5444 This hook should return the DECL of a function that implements reciprocal of
5445 the builtin function with builtin function code @var{fn}, or
5446 @code{NULL_TREE} if such a function is not available. @var{tm_fn} is true
5447 when @var{fn} is a code of a machine-dependent builtin function. When
5448 @var{sqrt} is true, additional optimizations that apply only to the reciprocal
5449 of a square root function are performed, and only reciprocals of @code{sqrt}
5453 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD (void)
5454 This hook should return the DECL of a function @var{f} that given an
5455 address @var{addr} as an argument returns a mask @var{m} that can be
5456 used to extract from two vectors the relevant data that resides in
5457 @var{addr} in case @var{addr} is not properly aligned.
5459 The autovectorizer, when vectorizing a load operation from an address
5460 @var{addr} that may be unaligned, will generate two vector loads from
5461 the two aligned addresses around @var{addr}. It then generates a
5462 @code{REALIGN_LOAD} operation to extract the relevant data from the
5463 two loaded vectors. The first two arguments to @code{REALIGN_LOAD},
5464 @var{v1} and @var{v2}, are the two vectors, each of size @var{VS}, and
5465 the third argument, @var{OFF}, defines how the data will be extracted
5466 from these two vectors: if @var{OFF} is 0, then the returned vector is
5467 @var{v2}; otherwise, the returned vector is composed from the last
5468 @var{VS}-@var{OFF} elements of @var{v1} concatenated to the first
5469 @var{OFF} elements of @var{v2}.
5471 If this hook is defined, the autovectorizer will generate a call
5472 to @var{f} (using the DECL tree that this hook returns) and will
5473 use the return value of @var{f} as the argument @var{OFF} to
5474 @code{REALIGN_LOAD}. Therefore, the mask @var{m} returned by @var{f}
5475 should comply with the semantics expected by @code{REALIGN_LOAD}
5477 If this hook is not defined, then @var{addr} will be used as
5478 the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low
5479 log2(@var{VS})-1 bits of @var{addr} will be considered.
5482 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN (tree @var{x})
5483 This hook should return the DECL of a function @var{f} that implements
5484 widening multiplication of the even elements of two input vectors of type @var{x}.
5486 If this hook is defined, the autovectorizer will use it along with the
5487 @code{TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD} target hook when vectorizing
5488 widening multiplication in cases that the order of the results does not have to be
5489 preserved (e.g.@: used only by a reduction computation). Otherwise, the
5490 @code{widen_mult_hi/lo} idioms will be used.
5493 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD (tree @var{x})
5494 This hook should return the DECL of a function @var{f} that implements
5495 widening multiplication of the odd elements of two input vectors of type @var{x}.
5497 If this hook is defined, the autovectorizer will use it along with the
5498 @code{TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN} target hook when vectorizing
5499 widening multiplication in cases that the order of the results does not have to be
5500 preserved (e.g.@: used only by a reduction computation). Otherwise, the
5501 @code{widen_mult_hi/lo} idioms will be used.
5504 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_CONVERSION (enum tree_code @var{code}, tree @var{type})
5505 This hook should return the DECL of a function that implements conversion of the
5506 input vector of type @var{type}.
5507 If @var{type} is an integral type, the result of the conversion is a vector of
5508 floating-point type of the same size.
5509 If @var{type} is a floating-point type, the result of the conversion is a vector
5510 of integral type of the same size.
5511 @var{code} specifies how the conversion is to be applied
5512 (truncation, rounding, etc.).
5514 If this hook is defined, the autovectorizer will use the
5515 @code{TARGET_VECTORIZE_BUILTIN_CONVERSION} target hook when vectorizing
5516 conversion. Otherwise, it will return @code{NULL_TREE}.
5519 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION (enum built_in_function @var{code}, tree @var{vec_type_out}, tree @var{vec_type_in})
5520 This hook should return the decl of a function that implements the vectorized
5521 variant of the builtin function with builtin function code @var{code} or
5522 @code{NULL_TREE} if such a function is not available. The return type of
5523 the vectorized function shall be of vector type @var{vec_type_out} and the
5524 argument types should be @var{vec_type_in}.
5527 @node Anchored Addresses
5528 @section Anchored Addresses
5529 @cindex anchored addresses
5530 @cindex @option{-fsection-anchors}
5532 GCC usually addresses every static object as a separate entity.
5533 For example, if we have:
5537 int foo (void) @{ return a + b + c; @}
5540 the code for @code{foo} will usually calculate three separate symbolic
5541 addresses: those of @code{a}, @code{b} and @code{c}. On some targets,
5542 it would be better to calculate just one symbolic address and access
5543 the three variables relative to it. The equivalent pseudocode would
5549 register int *xr = &x;
5550 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x];
5554 (which isn't valid C). We refer to shared addresses like @code{x} as
5555 ``section anchors''. Their use is controlled by @option{-fsection-anchors}.
5557 The hooks below describe the target properties that GCC needs to know
5558 in order to make effective use of section anchors. It won't use
5559 section anchors at all unless either @code{TARGET_MIN_ANCHOR_OFFSET}
5560 or @code{TARGET_MAX_ANCHOR_OFFSET} is set to a nonzero value.
5562 @deftypevar {Target Hook} HOST_WIDE_INT TARGET_MIN_ANCHOR_OFFSET
5563 The minimum offset that should be applied to a section anchor.
5564 On most targets, it should be the smallest offset that can be
5565 applied to a base register while still giving a legitimate address
5566 for every mode. The default value is 0.
5569 @deftypevar {Target Hook} HOST_WIDE_INT TARGET_MAX_ANCHOR_OFFSET
5570 Like @code{TARGET_MIN_ANCHOR_OFFSET}, but the maximum (inclusive)
5571 offset that should be applied to section anchors. The default
5575 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_ANCHOR (rtx @var{x})
5576 Write the assembly code to define section anchor @var{x}, which is a
5577 @code{SYMBOL_REF} for which @samp{SYMBOL_REF_ANCHOR_P (@var{x})} is true.
5578 The hook is called with the assembly output position set to the beginning
5579 of @code{SYMBOL_REF_BLOCK (@var{x})}.
5581 If @code{ASM_OUTPUT_DEF} is available, the hook's default definition uses
5582 it to define the symbol as @samp{. + SYMBOL_REF_BLOCK_OFFSET (@var{x})}.
5583 If @code{ASM_OUTPUT_DEF} is not available, the hook's default definition
5584 is @code{NULL}, which disables the use of section anchors altogether.
5587 @deftypefn {Target Hook} bool TARGET_USE_ANCHORS_FOR_SYMBOL_P (rtx @var{x})
5588 Return true if GCC should attempt to use anchors to access @code{SYMBOL_REF}
5589 @var{x}. You can assume @samp{SYMBOL_REF_HAS_BLOCK_INFO_P (@var{x})} and
5590 @samp{!SYMBOL_REF_ANCHOR_P (@var{x})}.
5592 The default version is correct for most targets, but you might need to
5593 intercept this hook to handle things like target-specific attributes
5594 or target-specific sections.
5597 @node Condition Code
5598 @section Condition Code Status
5599 @cindex condition code status
5601 @c prevent bad page break with this line
5602 This describes the condition code status.
5605 The file @file{conditions.h} defines a variable @code{cc_status} to
5606 describe how the condition code was computed (in case the interpretation of
5607 the condition code depends on the instruction that it was set by). This
5608 variable contains the RTL expressions on which the condition code is
5609 currently based, and several standard flags.
5611 Sometimes additional machine-specific flags must be defined in the machine
5612 description header file. It can also add additional machine-specific
5613 information by defining @code{CC_STATUS_MDEP}.
5615 @defmac CC_STATUS_MDEP
5616 C code for a data type which is used for declaring the @code{mdep}
5617 component of @code{cc_status}. It defaults to @code{int}.
5619 This macro is not used on machines that do not use @code{cc0}.
5622 @defmac CC_STATUS_MDEP_INIT
5623 A C expression to initialize the @code{mdep} field to ``empty''.
5624 The default definition does nothing, since most machines don't use
5625 the field anyway. If you want to use the field, you should probably
5626 define this macro to initialize it.
5628 This macro is not used on machines that do not use @code{cc0}.
5631 @defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn})
5632 A C compound statement to set the components of @code{cc_status}
5633 appropriately for an insn @var{insn} whose body is @var{exp}. It is
5634 this macro's responsibility to recognize insns that set the condition
5635 code as a byproduct of other activity as well as those that explicitly
5638 This macro is not used on machines that do not use @code{cc0}.
5640 If there are insns that do not set the condition code but do alter
5641 other machine registers, this macro must check to see whether they
5642 invalidate the expressions that the condition code is recorded as
5643 reflecting. For example, on the 68000, insns that store in address
5644 registers do not set the condition code, which means that usually
5645 @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
5646 insns. But suppose that the previous insn set the condition code
5647 based on location @samp{a4@@(102)} and the current insn stores a new
5648 value in @samp{a4}. Although the condition code is not changed by
5649 this, it will no longer be true that it reflects the contents of
5650 @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
5651 @code{cc_status} in this case to say that nothing is known about the
5652 condition code value.
5654 The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
5655 with the results of peephole optimization: insns whose patterns are
5656 @code{parallel} RTXs containing various @code{reg}, @code{mem} or
5657 constants which are just the operands. The RTL structure of these
5658 insns is not sufficient to indicate what the insns actually do. What
5659 @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
5660 @code{CC_STATUS_INIT}.
5662 A possible definition of @code{NOTICE_UPDATE_CC} is to call a function
5663 that looks at an attribute (@pxref{Insn Attributes}) named, for example,
5664 @samp{cc}. This avoids having detailed information about patterns in
5665 two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}.
5668 @defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y})
5669 Returns a mode from class @code{MODE_CC} to be used when comparison
5670 operation code @var{op} is applied to rtx @var{x} and @var{y}. For
5671 example, on the SPARC, @code{SELECT_CC_MODE} is defined as (see
5672 @pxref{Jump Patterns} for a description of the reason for this
5676 #define SELECT_CC_MODE(OP,X,Y) \
5677 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \
5678 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \
5679 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \
5680 || GET_CODE (X) == NEG) \
5681 ? CC_NOOVmode : CCmode))
5684 You should define this macro if and only if you define extra CC modes
5685 in @file{@var{machine}-modes.def}.
5688 @defmac CANONICALIZE_COMPARISON (@var{code}, @var{op0}, @var{op1})
5689 On some machines not all possible comparisons are defined, but you can
5690 convert an invalid comparison into a valid one. For example, the Alpha
5691 does not have a @code{GT} comparison, but you can use an @code{LT}
5692 comparison instead and swap the order of the operands.
5694 On such machines, define this macro to be a C statement to do any
5695 required conversions. @var{code} is the initial comparison code
5696 and @var{op0} and @var{op1} are the left and right operands of the
5697 comparison, respectively. You should modify @var{code}, @var{op0}, and
5698 @var{op1} as required.
5700 GCC will not assume that the comparison resulting from this macro is
5701 valid but will see if the resulting insn matches a pattern in the
5704 You need not define this macro if it would never change the comparison
5708 @defmac REVERSIBLE_CC_MODE (@var{mode})
5709 A C expression whose value is one if it is always safe to reverse a
5710 comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE}
5711 can ever return @var{mode} for a floating-point inequality comparison,
5712 then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero.
5714 You need not define this macro if it would always returns zero or if the
5715 floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}.
5716 For example, here is the definition used on the SPARC, where floating-point
5717 inequality comparisons are always given @code{CCFPEmode}:
5720 #define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode)
5724 @defmac REVERSE_CONDITION (@var{code}, @var{mode})
5725 A C expression whose value is reversed condition code of the @var{code} for
5726 comparison done in CC_MODE @var{mode}. The macro is used only in case
5727 @code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case
5728 machine has some non-standard way how to reverse certain conditionals. For
5729 instance in case all floating point conditions are non-trapping, compiler may
5730 freely convert unordered compares to ordered one. Then definition may look
5734 #define REVERSE_CONDITION(CODE, MODE) \
5735 ((MODE) != CCFPmode ? reverse_condition (CODE) \
5736 : reverse_condition_maybe_unordered (CODE))
5740 @defmac REVERSE_CONDEXEC_PREDICATES_P (@var{op1}, @var{op2})
5741 A C expression that returns true if the conditional execution predicate
5742 @var{op1}, a comparison operation, is the inverse of @var{op2} and vice
5743 versa. Define this to return 0 if the target has conditional execution
5744 predicates that cannot be reversed safely. There is no need to validate
5745 that the arguments of op1 and op2 are the same, this is done separately.
5746 If no expansion is specified, this macro is defined as follows:
5749 #define REVERSE_CONDEXEC_PREDICATES_P (x, y) \
5750 (GET_CODE ((x)) == reversed_comparison_code ((y), NULL))
5754 @deftypefn {Target Hook} bool TARGET_FIXED_CONDITION_CODE_REGS (unsigned int *, unsigned int *)
5755 On targets which do not use @code{(cc0)}, and which use a hard
5756 register rather than a pseudo-register to hold condition codes, the
5757 regular CSE passes are often not able to identify cases in which the
5758 hard register is set to a common value. Use this hook to enable a
5759 small pass which optimizes such cases. This hook should return true
5760 to enable this pass, and it should set the integers to which its
5761 arguments point to the hard register numbers used for condition codes.
5762 When there is only one such register, as is true on most systems, the
5763 integer pointed to by the second argument should be set to
5764 @code{INVALID_REGNUM}.
5766 The default version of this hook returns false.
5769 @deftypefn {Target Hook} enum machine_mode TARGET_CC_MODES_COMPATIBLE (enum machine_mode, enum machine_mode)
5770 On targets which use multiple condition code modes in class
5771 @code{MODE_CC}, it is sometimes the case that a comparison can be
5772 validly done in more than one mode. On such a system, define this
5773 target hook to take two mode arguments and to return a mode in which
5774 both comparisons may be validly done. If there is no such mode,
5775 return @code{VOIDmode}.
5777 The default version of this hook checks whether the modes are the
5778 same. If they are, it returns that mode. If they are different, it
5779 returns @code{VOIDmode}.
5783 @section Describing Relative Costs of Operations
5784 @cindex costs of instructions
5785 @cindex relative costs
5786 @cindex speed of instructions
5788 These macros let you describe the relative speed of various operations
5789 on the target machine.
5791 @defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to})
5792 A C expression for the cost of moving data of mode @var{mode} from a
5793 register in class @var{from} to one in class @var{to}. The classes are
5794 expressed using the enumeration values such as @code{GENERAL_REGS}. A
5795 value of 2 is the default; other values are interpreted relative to
5798 It is not required that the cost always equal 2 when @var{from} is the
5799 same as @var{to}; on some machines it is expensive to move between
5800 registers if they are not general registers.
5802 If reload sees an insn consisting of a single @code{set} between two
5803 hard registers, and if @code{REGISTER_MOVE_COST} applied to their
5804 classes returns a value of 2, reload does not check to ensure that the
5805 constraints of the insn are met. Setting a cost of other than 2 will
5806 allow reload to verify that the constraints are met. You should do this
5807 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
5810 @defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in})
5811 A C expression for the cost of moving data of mode @var{mode} between a
5812 register of class @var{class} and memory; @var{in} is zero if the value
5813 is to be written to memory, nonzero if it is to be read in. This cost
5814 is relative to those in @code{REGISTER_MOVE_COST}. If moving between
5815 registers and memory is more expensive than between two registers, you
5816 should define this macro to express the relative cost.
5818 If you do not define this macro, GCC uses a default cost of 4 plus
5819 the cost of copying via a secondary reload register, if one is
5820 needed. If your machine requires a secondary reload register to copy
5821 between memory and a register of @var{class} but the reload mechanism is
5822 more complex than copying via an intermediate, define this macro to
5823 reflect the actual cost of the move.
5825 GCC defines the function @code{memory_move_secondary_cost} if
5826 secondary reloads are needed. It computes the costs due to copying via
5827 a secondary register. If your machine copies from memory using a
5828 secondary register in the conventional way but the default base value of
5829 4 is not correct for your machine, define this macro to add some other
5830 value to the result of that function. The arguments to that function
5831 are the same as to this macro.
5835 A C expression for the cost of a branch instruction. A value of 1 is
5836 the default; other values are interpreted relative to that.
5839 Here are additional macros which do not specify precise relative costs,
5840 but only that certain actions are more expensive than GCC would
5843 @defmac SLOW_BYTE_ACCESS
5844 Define this macro as a C expression which is nonzero if accessing less
5845 than a word of memory (i.e.@: a @code{char} or a @code{short}) is no
5846 faster than accessing a word of memory, i.e., if such access
5847 require more than one instruction or if there is no difference in cost
5848 between byte and (aligned) word loads.
5850 When this macro is not defined, the compiler will access a field by
5851 finding the smallest containing object; when it is defined, a fullword
5852 load will be used if alignment permits. Unless bytes accesses are
5853 faster than word accesses, using word accesses is preferable since it
5854 may eliminate subsequent memory access if subsequent accesses occur to
5855 other fields in the same word of the structure, but to different bytes.
5858 @defmac SLOW_UNALIGNED_ACCESS (@var{mode}, @var{alignment})
5859 Define this macro to be the value 1 if memory accesses described by the
5860 @var{mode} and @var{alignment} parameters have a cost many times greater
5861 than aligned accesses, for example if they are emulated in a trap
5864 When this macro is nonzero, the compiler will act as if
5865 @code{STRICT_ALIGNMENT} were nonzero when generating code for block
5866 moves. This can cause significantly more instructions to be produced.
5867 Therefore, do not set this macro nonzero if unaligned accesses only add a
5868 cycle or two to the time for a memory access.
5870 If the value of this macro is always zero, it need not be defined. If
5871 this macro is defined, it should produce a nonzero value when
5872 @code{STRICT_ALIGNMENT} is nonzero.
5876 The threshold of number of scalar memory-to-memory move insns, @emph{below}
5877 which a sequence of insns should be generated instead of a
5878 string move insn or a library call. Increasing the value will always
5879 make code faster, but eventually incurs high cost in increased code size.
5881 Note that on machines where the corresponding move insn is a
5882 @code{define_expand} that emits a sequence of insns, this macro counts
5883 the number of such sequences.
5885 If you don't define this, a reasonable default is used.
5888 @defmac MOVE_BY_PIECES_P (@var{size}, @var{alignment})
5889 A C expression used to determine whether @code{move_by_pieces} will be used to
5890 copy a chunk of memory, or whether some other block move mechanism
5891 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
5892 than @code{MOVE_RATIO}.
5895 @defmac MOVE_MAX_PIECES
5896 A C expression used by @code{move_by_pieces} to determine the largest unit
5897 a load or store used to copy memory is. Defaults to @code{MOVE_MAX}.
5901 The threshold of number of scalar move insns, @emph{below} which a sequence
5902 of insns should be generated to clear memory instead of a string clear insn
5903 or a library call. Increasing the value will always make code faster, but
5904 eventually incurs high cost in increased code size.
5906 If you don't define this, a reasonable default is used.
5909 @defmac CLEAR_BY_PIECES_P (@var{size}, @var{alignment})
5910 A C expression used to determine whether @code{clear_by_pieces} will be used
5911 to clear a chunk of memory, or whether some other block clear mechanism
5912 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
5913 than @code{CLEAR_RATIO}.
5917 The threshold of number of scalar move insns, @emph{below} which a sequence
5918 of insns should be generated to set memory to a constant value, instead of
5919 a block set insn or a library call.
5920 Increasing the value will always make code faster, but
5921 eventually incurs high cost in increased code size.
5923 If you don't define this, it defaults to the value of @code{MOVE_RATIO}.
5926 @defmac SET_BY_PIECES_P (@var{size}, @var{alignment})
5927 A C expression used to determine whether @code{store_by_pieces} will be
5928 used to set a chunk of memory to a constant value, or whether some
5929 other mechanism will be used. Used by @code{__builtin_memset} when
5930 storing values other than constant zero.
5931 Defaults to 1 if @code{move_by_pieces_ninsns} returns less
5932 than @code{SET_RATIO}.
5935 @defmac STORE_BY_PIECES_P (@var{size}, @var{alignment})
5936 A C expression used to determine whether @code{store_by_pieces} will be
5937 used to set a chunk of memory to a constant string value, or whether some
5938 other mechanism will be used. Used by @code{__builtin_strcpy} when
5939 called with a constant source string.
5940 Defaults to 1 if @code{move_by_pieces_ninsns} returns less
5941 than @code{MOVE_RATIO}.
5944 @defmac USE_LOAD_POST_INCREMENT (@var{mode})
5945 A C expression used to determine whether a load postincrement is a good
5946 thing to use for a given mode. Defaults to the value of
5947 @code{HAVE_POST_INCREMENT}.
5950 @defmac USE_LOAD_POST_DECREMENT (@var{mode})
5951 A C expression used to determine whether a load postdecrement is a good
5952 thing to use for a given mode. Defaults to the value of
5953 @code{HAVE_POST_DECREMENT}.
5956 @defmac USE_LOAD_PRE_INCREMENT (@var{mode})
5957 A C expression used to determine whether a load preincrement is a good
5958 thing to use for a given mode. Defaults to the value of
5959 @code{HAVE_PRE_INCREMENT}.
5962 @defmac USE_LOAD_PRE_DECREMENT (@var{mode})
5963 A C expression used to determine whether a load predecrement is a good
5964 thing to use for a given mode. Defaults to the value of
5965 @code{HAVE_PRE_DECREMENT}.
5968 @defmac USE_STORE_POST_INCREMENT (@var{mode})
5969 A C expression used to determine whether a store postincrement is a good
5970 thing to use for a given mode. Defaults to the value of
5971 @code{HAVE_POST_INCREMENT}.
5974 @defmac USE_STORE_POST_DECREMENT (@var{mode})
5975 A C expression used to determine whether a store postdecrement is a good
5976 thing to use for a given mode. Defaults to the value of
5977 @code{HAVE_POST_DECREMENT}.
5980 @defmac USE_STORE_PRE_INCREMENT (@var{mode})
5981 This macro is used to determine whether a store preincrement is a good
5982 thing to use for a given mode. Defaults to the value of
5983 @code{HAVE_PRE_INCREMENT}.
5986 @defmac USE_STORE_PRE_DECREMENT (@var{mode})
5987 This macro is used to determine whether a store predecrement is a good
5988 thing to use for a given mode. Defaults to the value of
5989 @code{HAVE_PRE_DECREMENT}.
5992 @defmac NO_FUNCTION_CSE
5993 Define this macro if it is as good or better to call a constant
5994 function address than to call an address kept in a register.
5997 @defmac RANGE_TEST_NON_SHORT_CIRCUIT
5998 Define this macro if a non-short-circuit operation produced by
5999 @samp{fold_range_test ()} is optimal. This macro defaults to true if
6000 @code{BRANCH_COST} is greater than or equal to the value 2.
6003 @deftypefn {Target Hook} bool TARGET_RTX_COSTS (rtx @var{x}, int @var{code}, int @var{outer_code}, int *@var{total})
6004 This target hook describes the relative costs of RTL expressions.
6006 The cost may depend on the precise form of the expression, which is
6007 available for examination in @var{x}, and the rtx code of the expression
6008 in which it is contained, found in @var{outer_code}. @var{code} is the
6009 expression code---redundant, since it can be obtained with
6010 @code{GET_CODE (@var{x})}.
6012 In implementing this hook, you can use the construct
6013 @code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast
6016 On entry to the hook, @code{*@var{total}} contains a default estimate
6017 for the cost of the expression. The hook should modify this value as
6018 necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)}
6019 for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus
6020 operations, and @code{COSTS_N_INSNS (1)} for all other operations.
6022 When optimizing for code size, i.e.@: when @code{optimize_size} is
6023 nonzero, this target hook should be used to estimate the relative
6024 size cost of an expression, again relative to @code{COSTS_N_INSNS}.
6026 The hook returns true when all subexpressions of @var{x} have been
6027 processed, and false when @code{rtx_cost} should recurse.
6030 @deftypefn {Target Hook} int TARGET_ADDRESS_COST (rtx @var{address})
6031 This hook computes the cost of an addressing mode that contains
6032 @var{address}. If not defined, the cost is computed from
6033 the @var{address} expression and the @code{TARGET_RTX_COST} hook.
6035 For most CISC machines, the default cost is a good approximation of the
6036 true cost of the addressing mode. However, on RISC machines, all
6037 instructions normally have the same length and execution time. Hence
6038 all addresses will have equal costs.
6040 In cases where more than one form of an address is known, the form with
6041 the lowest cost will be used. If multiple forms have the same, lowest,
6042 cost, the one that is the most complex will be used.
6044 For example, suppose an address that is equal to the sum of a register
6045 and a constant is used twice in the same basic block. When this macro
6046 is not defined, the address will be computed in a register and memory
6047 references will be indirect through that register. On machines where
6048 the cost of the addressing mode containing the sum is no higher than
6049 that of a simple indirect reference, this will produce an additional
6050 instruction and possibly require an additional register. Proper
6051 specification of this macro eliminates this overhead for such machines.
6053 This hook is never called with an invalid address.
6055 On machines where an address involving more than one register is as
6056 cheap as an address computation involving only one register, defining
6057 @code{TARGET_ADDRESS_COST} to reflect this can cause two registers to
6058 be live over a region of code where only one would have been if
6059 @code{TARGET_ADDRESS_COST} were not defined in that manner. This effect
6060 should be considered in the definition of this macro. Equivalent costs
6061 should probably only be given to addresses with different numbers of
6062 registers on machines with lots of registers.
6066 @section Adjusting the Instruction Scheduler
6068 The instruction scheduler may need a fair amount of machine-specific
6069 adjustment in order to produce good code. GCC provides several target
6070 hooks for this purpose. It is usually enough to define just a few of
6071 them: try the first ones in this list first.
6073 @deftypefn {Target Hook} int TARGET_SCHED_ISSUE_RATE (void)
6074 This hook returns the maximum number of instructions that can ever
6075 issue at the same time on the target machine. The default is one.
6076 Although the insn scheduler can define itself the possibility of issue
6077 an insn on the same cycle, the value can serve as an additional
6078 constraint to issue insns on the same simulated processor cycle (see
6079 hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}).
6080 This value must be constant over the entire compilation. If you need
6081 it to vary depending on what the instructions are, you must use
6082 @samp{TARGET_SCHED_VARIABLE_ISSUE}.
6085 @deftypefn {Target Hook} int TARGET_SCHED_VARIABLE_ISSUE (FILE *@var{file}, int @var{verbose}, rtx @var{insn}, int @var{more})
6086 This hook is executed by the scheduler after it has scheduled an insn
6087 from the ready list. It should return the number of insns which can
6088 still be issued in the current cycle. The default is
6089 @samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and
6090 @code{USE}, which normally are not counted against the issue rate.
6091 You should define this hook if some insns take more machine resources
6092 than others, so that fewer insns can follow them in the same cycle.
6093 @var{file} is either a null pointer, or a stdio stream to write any
6094 debug output to. @var{verbose} is the verbose level provided by
6095 @option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that
6099 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_COST (rtx @var{insn}, rtx @var{link}, rtx @var{dep_insn}, int @var{cost})
6100 This function corrects the value of @var{cost} based on the
6101 relationship between @var{insn} and @var{dep_insn} through the
6102 dependence @var{link}. It should return the new value. The default
6103 is to make no adjustment to @var{cost}. This can be used for example
6104 to specify to the scheduler using the traditional pipeline description
6105 that an output- or anti-dependence does not incur the same cost as a
6106 data-dependence. If the scheduler using the automaton based pipeline
6107 description, the cost of anti-dependence is zero and the cost of
6108 output-dependence is maximum of one and the difference of latency
6109 times of the first and the second insns. If these values are not
6110 acceptable, you could use the hook to modify them too. See also
6111 @pxref{Processor pipeline description}.
6114 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_PRIORITY (rtx @var{insn}, int @var{priority})
6115 This hook adjusts the integer scheduling priority @var{priority} of
6116 @var{insn}. It should return the new priority. Increase the priority to
6117 execute @var{insn} earlier, reduce the priority to execute @var{insn}
6118 later. Do not define this hook if you do not need to adjust the
6119 scheduling priorities of insns.
6122 @deftypefn {Target Hook} int TARGET_SCHED_REORDER (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_readyp}, int @var{clock})
6123 This hook is executed by the scheduler after it has scheduled the ready
6124 list, to allow the machine description to reorder it (for example to
6125 combine two small instructions together on @samp{VLIW} machines).
6126 @var{file} is either a null pointer, or a stdio stream to write any
6127 debug output to. @var{verbose} is the verbose level provided by
6128 @option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready
6129 list of instructions that are ready to be scheduled. @var{n_readyp} is
6130 a pointer to the number of elements in the ready list. The scheduler
6131 reads the ready list in reverse order, starting with
6132 @var{ready}[@var{*n_readyp}-1] and going to @var{ready}[0]. @var{clock}
6133 is the timer tick of the scheduler. You may modify the ready list and
6134 the number of ready insns. The return value is the number of insns that
6135 can issue this cycle; normally this is just @code{issue_rate}. See also
6136 @samp{TARGET_SCHED_REORDER2}.
6139 @deftypefn {Target Hook} int TARGET_SCHED_REORDER2 (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_ready}, @var{clock})
6140 Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That
6141 function is called whenever the scheduler starts a new cycle. This one
6142 is called once per iteration over a cycle, immediately after
6143 @samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and
6144 return the number of insns to be scheduled in the same cycle. Defining
6145 this hook can be useful if there are frequent situations where
6146 scheduling one insn causes other insns to become ready in the same
6147 cycle. These other insns can then be taken into account properly.
6150 @deftypefn {Target Hook} void TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK (rtx @var{head}, rtx @var{tail})
6151 This hook is called after evaluation forward dependencies of insns in
6152 chain given by two parameter values (@var{head} and @var{tail}
6153 correspondingly) but before insns scheduling of the insn chain. For
6154 example, it can be used for better insn classification if it requires
6155 analysis of dependencies. This hook can use backward and forward
6156 dependencies of the insn scheduler because they are already
6160 @deftypefn {Target Hook} void TARGET_SCHED_INIT (FILE *@var{file}, int @var{verbose}, int @var{max_ready})
6161 This hook is executed by the scheduler at the beginning of each block of
6162 instructions that are to be scheduled. @var{file} is either a null
6163 pointer, or a stdio stream to write any debug output to. @var{verbose}
6164 is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6165 @var{max_ready} is the maximum number of insns in the current scheduling
6166 region that can be live at the same time. This can be used to allocate
6167 scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}.
6170 @deftypefn {Target Hook} void TARGET_SCHED_FINISH (FILE *@var{file}, int @var{verbose})
6171 This hook is executed by the scheduler at the end of each block of
6172 instructions that are to be scheduled. It can be used to perform
6173 cleanup of any actions done by the other scheduling hooks. @var{file}
6174 is either a null pointer, or a stdio stream to write any debug output
6175 to. @var{verbose} is the verbose level provided by
6176 @option{-fsched-verbose-@var{n}}.
6179 @deftypefn {Target Hook} void TARGET_SCHED_INIT_GLOBAL (FILE *@var{file}, int @var{verbose}, int @var{old_max_uid})
6180 This hook is executed by the scheduler after function level initializations.
6181 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
6182 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6183 @var{old_max_uid} is the maximum insn uid when scheduling begins.
6186 @deftypefn {Target Hook} void TARGET_SCHED_FINISH_GLOBAL (FILE *@var{file}, int @var{verbose})
6187 This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}.
6188 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
6189 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6192 @deftypefn {Target Hook} int TARGET_SCHED_DFA_PRE_CYCLE_INSN (void)
6193 The hook returns an RTL insn. The automaton state used in the
6194 pipeline hazard recognizer is changed as if the insn were scheduled
6195 when the new simulated processor cycle starts. Usage of the hook may
6196 simplify the automaton pipeline description for some @acronym{VLIW}
6197 processors. If the hook is defined, it is used only for the automaton
6198 based pipeline description. The default is not to change the state
6199 when the new simulated processor cycle starts.
6202 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN (void)
6203 The hook can be used to initialize data used by the previous hook.
6206 @deftypefn {Target Hook} int TARGET_SCHED_DFA_POST_CYCLE_INSN (void)
6207 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
6208 to changed the state as if the insn were scheduled when the new
6209 simulated processor cycle finishes.
6212 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN (void)
6213 The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but
6214 used to initialize data used by the previous hook.
6217 @deftypefn {Target Hook} void TARGET_SCHED_DFA_PRE_CYCLE_ADVANCE (void)
6218 The hook to notify target that the current simulated cycle is about to finish.
6219 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
6220 to change the state in more complicated situations - e.g., when advancing
6221 state on a single insn is not enough.
6224 @deftypefn {Target Hook} void TARGET_SCHED_DFA_POST_CYCLE_ADVANCE (void)
6225 The hook to notify target that new simulated cycle has just started.
6226 The hook is analogous to @samp{TARGET_SCHED_DFA_POST_CYCLE_INSN} but used
6227 to change the state in more complicated situations - e.g., when advancing
6228 state on a single insn is not enough.
6231 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD (void)
6232 This hook controls better choosing an insn from the ready insn queue
6233 for the @acronym{DFA}-based insn scheduler. Usually the scheduler
6234 chooses the first insn from the queue. If the hook returns a positive
6235 value, an additional scheduler code tries all permutations of
6236 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()}
6237 subsequent ready insns to choose an insn whose issue will result in
6238 maximal number of issued insns on the same cycle. For the
6239 @acronym{VLIW} processor, the code could actually solve the problem of
6240 packing simple insns into the @acronym{VLIW} insn. Of course, if the
6241 rules of @acronym{VLIW} packing are described in the automaton.
6243 This code also could be used for superscalar @acronym{RISC}
6244 processors. Let us consider a superscalar @acronym{RISC} processor
6245 with 3 pipelines. Some insns can be executed in pipelines @var{A} or
6246 @var{B}, some insns can be executed only in pipelines @var{B} or
6247 @var{C}, and one insn can be executed in pipeline @var{B}. The
6248 processor may issue the 1st insn into @var{A} and the 2nd one into
6249 @var{B}. In this case, the 3rd insn will wait for freeing @var{B}
6250 until the next cycle. If the scheduler issues the 3rd insn the first,
6251 the processor could issue all 3 insns per cycle.
6253 Actually this code demonstrates advantages of the automaton based
6254 pipeline hazard recognizer. We try quickly and easy many insn
6255 schedules to choose the best one.
6257 The default is no multipass scheduling.
6260 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD (rtx)
6262 This hook controls what insns from the ready insn queue will be
6263 considered for the multipass insn scheduling. If the hook returns
6264 zero for insn passed as the parameter, the insn will be not chosen to
6267 The default is that any ready insns can be chosen to be issued.
6270 @deftypefn {Target Hook} int TARGET_SCHED_DFA_NEW_CYCLE (FILE *, int, rtx, int, int, int *)
6272 This hook is called by the insn scheduler before issuing insn passed
6273 as the third parameter on given cycle. If the hook returns nonzero,
6274 the insn is not issued on given processors cycle. Instead of that,
6275 the processor cycle is advanced. If the value passed through the last
6276 parameter is zero, the insn ready queue is not sorted on the new cycle
6277 start as usually. The first parameter passes file for debugging
6278 output. The second one passes the scheduler verbose level of the
6279 debugging output. The forth and the fifth parameter values are
6280 correspondingly processor cycle on which the previous insn has been
6281 issued and the current processor cycle.
6284 @deftypefn {Target Hook} bool TARGET_SCHED_IS_COSTLY_DEPENDENCE (struct dep_def *@var{_dep}, int @var{cost}, int @var{distance})
6285 This hook is used to define which dependences are considered costly by
6286 the target, so costly that it is not advisable to schedule the insns that
6287 are involved in the dependence too close to one another. The parameters
6288 to this hook are as follows: The first parameter @var{_dep} is the dependence
6289 being evaluated. The second parameter @var{cost} is the cost of the
6290 dependence, and the third
6291 parameter @var{distance} is the distance in cycles between the two insns.
6292 The hook returns @code{true} if considering the distance between the two
6293 insns the dependence between them is considered costly by the target,
6294 and @code{false} otherwise.
6296 Defining this hook can be useful in multiple-issue out-of-order machines,
6297 where (a) it's practically hopeless to predict the actual data/resource
6298 delays, however: (b) there's a better chance to predict the actual grouping
6299 that will be formed, and (c) correctly emulating the grouping can be very
6300 important. In such targets one may want to allow issuing dependent insns
6301 closer to one another---i.e., closer than the dependence distance; however,
6302 not in cases of "costly dependences", which this hooks allows to define.
6305 @deftypefn {Target Hook} void TARGET_SCHED_H_I_D_EXTENDED (void)
6306 This hook is called by the insn scheduler after emitting a new instruction to
6307 the instruction stream. The hook notifies a target backend to extend its
6308 per instruction data structures.
6311 @deftypefn {Target Hook} int TARGET_SCHED_SPECULATE_INSN (rtx @var{insn}, int @var{request}, rtx *@var{new_pat})
6312 This hook is called by the insn scheduler when @var{insn} has only
6313 speculative dependencies and therefore can be scheduled speculatively.
6314 The hook is used to check if the pattern of @var{insn} has a speculative
6315 version and, in case of successful check, to generate that speculative
6316 pattern. The hook should return 1, if the instruction has a speculative form,
6317 or -1, if it doesn't. @var{request} describes the type of requested
6318 speculation. If the return value equals 1 then @var{new_pat} is assigned
6319 the generated speculative pattern.
6322 @deftypefn {Target Hook} int TARGET_SCHED_NEEDS_BLOCK_P (rtx @var{insn})
6323 This hook is called by the insn scheduler during generation of recovery code
6324 for @var{insn}. It should return nonzero, if the corresponding check
6325 instruction should branch to recovery code, or zero otherwise.
6328 @deftypefn {Target Hook} rtx TARGET_SCHED_GEN_CHECK (rtx @var{insn}, rtx @var{label}, int @var{mutate_p})
6329 This hook is called by the insn scheduler to generate a pattern for recovery
6330 check instruction. If @var{mutate_p} is zero, then @var{insn} is a
6331 speculative instruction for which the check should be generated.
6332 @var{label} is either a label of a basic block, where recovery code should
6333 be emitted, or a null pointer, when requested check doesn't branch to
6334 recovery code (a simple check). If @var{mutate_p} is nonzero, then
6335 a pattern for a branchy check corresponding to a simple check denoted by
6336 @var{insn} should be generated. In this case @var{label} can't be null.
6339 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD_SPEC (rtx @var{insn})
6340 This hook is used as a workaround for
6341 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD} not being
6342 called on the first instruction of the ready list. The hook is used to
6343 discard speculative instruction that stand first in the ready list from
6344 being scheduled on the current cycle. For non-speculative instructions,
6345 the hook should always return nonzero. For example, in the ia64 backend
6346 the hook is used to cancel data speculative insns when the ALAT table
6350 @deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_FLAGS (unsigned int *@var{flags}, spec_info_t @var{spec_info})
6351 This hook is used by the insn scheduler to find out what features should be
6352 enabled/used. @var{flags} initially may have either the SCHED_RGN or SCHED_EBB
6353 bit set. This denotes the scheduler pass for which the data should be
6354 provided. The target backend should modify @var{flags} by modifying
6355 the bits corresponding to the following features: USE_DEPS_LIST, USE_GLAT,
6356 DETACH_LIFE_INFO, and DO_SPECULATION@. For the DO_SPECULATION feature
6357 an additional structure @var{spec_info} should be filled by the target.
6358 The structure describes speculation types that can be used in the scheduler.
6361 @deftypefn {Target Hook} int TARGET_SCHED_SMS_RES_MII (struct ddg *@var{g})
6362 This hook is called by the swing modulo scheduler to calculate a
6363 resource-based lower bound which is based on the resources available in
6364 the machine and the resources required by each instruction. The target
6365 backend can use @var{g} to calculate such bound. A very simple lower
6366 bound will be used in case this hook is not implemented: the total number
6367 of instructions divided by the issue rate.
6371 @section Dividing the Output into Sections (Texts, Data, @dots{})
6372 @c the above section title is WAY too long. maybe cut the part between
6373 @c the (...)? --mew 10feb93
6375 An object file is divided into sections containing different types of
6376 data. In the most common case, there are three sections: the @dfn{text
6377 section}, which holds instructions and read-only data; the @dfn{data
6378 section}, which holds initialized writable data; and the @dfn{bss
6379 section}, which holds uninitialized data. Some systems have other kinds
6382 @file{varasm.c} provides several well-known sections, such as
6383 @code{text_section}, @code{data_section} and @code{bss_section}.
6384 The normal way of controlling a @code{@var{foo}_section} variable
6385 is to define the associated @code{@var{FOO}_SECTION_ASM_OP} macro,
6386 as described below. The macros are only read once, when @file{varasm.c}
6387 initializes itself, so their values must be run-time constants.
6388 They may however depend on command-line flags.
6390 @emph{Note:} Some run-time files, such @file{crtstuff.c}, also make
6391 use of the @code{@var{FOO}_SECTION_ASM_OP} macros, and expect them
6392 to be string literals.
6394 Some assemblers require a different string to be written every time a
6395 section is selected. If your assembler falls into this category, you
6396 should define the @code{TARGET_ASM_INIT_SECTIONS} hook and use
6397 @code{get_unnamed_section} to set up the sections.
6399 You must always create a @code{text_section}, either by defining
6400 @code{TEXT_SECTION_ASM_OP} or by initializing @code{text_section}
6401 in @code{TARGET_ASM_INIT_SECTIONS}. The same is true of
6402 @code{data_section} and @code{DATA_SECTION_ASM_OP}. If you do not
6403 create a distinct @code{readonly_data_section}, the default is to
6404 reuse @code{text_section}.
6406 All the other @file{varasm.c} sections are optional, and are null
6407 if the target does not provide them.
6409 @defmac TEXT_SECTION_ASM_OP
6410 A C expression whose value is a string, including spacing, containing the
6411 assembler operation that should precede instructions and read-only data.
6412 Normally @code{"\t.text"} is right.
6415 @defmac HOT_TEXT_SECTION_NAME
6416 If defined, a C string constant for the name of the section containing most
6417 frequently executed functions of the program. If not defined, GCC will provide
6418 a default definition if the target supports named sections.
6421 @defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME
6422 If defined, a C string constant for the name of the section containing unlikely
6423 executed functions in the program.
6426 @defmac DATA_SECTION_ASM_OP
6427 A C expression whose value is a string, including spacing, containing the
6428 assembler operation to identify the following data as writable initialized
6429 data. Normally @code{"\t.data"} is right.
6432 @defmac SDATA_SECTION_ASM_OP
6433 If defined, a C expression whose value is a string, including spacing,
6434 containing the assembler operation to identify the following data as
6435 initialized, writable small data.
6438 @defmac READONLY_DATA_SECTION_ASM_OP
6439 A C expression whose value is a string, including spacing, containing the
6440 assembler operation to identify the following data as read-only initialized
6444 @defmac BSS_SECTION_ASM_OP
6445 If defined, a C expression whose value is a string, including spacing,
6446 containing the assembler operation to identify the following data as
6447 uninitialized global data. If not defined, and neither
6448 @code{ASM_OUTPUT_BSS} nor @code{ASM_OUTPUT_ALIGNED_BSS} are defined,
6449 uninitialized global data will be output in the data section if
6450 @option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be
6454 @defmac SBSS_SECTION_ASM_OP
6455 If defined, a C expression whose value is a string, including spacing,
6456 containing the assembler operation to identify the following data as
6457 uninitialized, writable small data.
6460 @defmac INIT_SECTION_ASM_OP
6461 If defined, a C expression whose value is a string, including spacing,
6462 containing the assembler operation to identify the following data as
6463 initialization code. If not defined, GCC will assume such a section does
6464 not exist. This section has no corresponding @code{init_section}
6465 variable; it is used entirely in runtime code.
6468 @defmac FINI_SECTION_ASM_OP
6469 If defined, a C expression whose value is a string, including spacing,
6470 containing the assembler operation to identify the following data as
6471 finalization code. If not defined, GCC will assume such a section does
6472 not exist. This section has no corresponding @code{fini_section}
6473 variable; it is used entirely in runtime code.
6476 @defmac INIT_ARRAY_SECTION_ASM_OP
6477 If defined, a C expression whose value is a string, including spacing,
6478 containing the assembler operation to identify the following data as
6479 part of the @code{.init_array} (or equivalent) section. If not
6480 defined, GCC will assume such a section does not exist. Do not define
6481 both this macro and @code{INIT_SECTION_ASM_OP}.
6484 @defmac FINI_ARRAY_SECTION_ASM_OP
6485 If defined, a C expression whose value is a string, including spacing,
6486 containing the assembler operation to identify the following data as
6487 part of the @code{.fini_array} (or equivalent) section. If not
6488 defined, GCC will assume such a section does not exist. Do not define
6489 both this macro and @code{FINI_SECTION_ASM_OP}.
6492 @defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function})
6493 If defined, an ASM statement that switches to a different section
6494 via @var{section_op}, calls @var{function}, and switches back to
6495 the text section. This is used in @file{crtstuff.c} if
6496 @code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls
6497 to initialization and finalization functions from the init and fini
6498 sections. By default, this macro uses a simple function call. Some
6499 ports need hand-crafted assembly code to avoid dependencies on
6500 registers initialized in the function prologue or to ensure that
6501 constant pools don't end up too far way in the text section.
6504 @defmac TARGET_LIBGCC_SDATA_SECTION
6505 If defined, a string which names the section into which small
6506 variables defined in crtstuff and libgcc should go. This is useful
6507 when the target has options for optimizing access to small data, and
6508 you want the crtstuff and libgcc routines to be conservative in what
6509 they expect of your application yet liberal in what your application
6510 expects. For example, for targets with a @code{.sdata} section (like
6511 MIPS), you could compile crtstuff with @code{-G 0} so that it doesn't
6512 require small data support from your application, but use this macro
6513 to put small data into @code{.sdata} so that your application can
6514 access these variables whether it uses small data or not.
6517 @defmac FORCE_CODE_SECTION_ALIGN
6518 If defined, an ASM statement that aligns a code section to some
6519 arbitrary boundary. This is used to force all fragments of the
6520 @code{.init} and @code{.fini} sections to have to same alignment
6521 and thus prevent the linker from having to add any padding.
6524 @defmac JUMP_TABLES_IN_TEXT_SECTION
6525 Define this macro to be an expression with a nonzero value if jump
6526 tables (for @code{tablejump} insns) should be output in the text
6527 section, along with the assembler instructions. Otherwise, the
6528 readonly data section is used.
6530 This macro is irrelevant if there is no separate readonly data section.
6533 @deftypefn {Target Hook} void TARGET_ASM_INIT_SECTIONS (void)
6534 Define this hook if you need to do something special to set up the
6535 @file{varasm.c} sections, or if your target has some special sections
6536 of its own that you need to create.
6538 GCC calls this hook after processing the command line, but before writing
6539 any assembly code, and before calling any of the section-returning hooks
6543 @deftypefn {Target Hook} TARGET_ASM_RELOC_RW_MASK (void)
6544 Return a mask describing how relocations should be treated when
6545 selecting sections. Bit 1 should be set if global relocations
6546 should be placed in a read-write section; bit 0 should be set if
6547 local relocations should be placed in a read-write section.
6549 The default version of this function returns 3 when @option{-fpic}
6550 is in effect, and 0 otherwise. The hook is typically redefined
6551 when the target cannot support (some kinds of) dynamic relocations
6552 in read-only sections even in executables.
6555 @deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_SECTION (tree @var{exp}, int @var{reloc}, unsigned HOST_WIDE_INT @var{align})
6556 Return the section into which @var{exp} should be placed. You can
6557 assume that @var{exp} is either a @code{VAR_DECL} node or a constant of
6558 some sort. @var{reloc} indicates whether the initial value of @var{exp}
6559 requires link-time relocations. Bit 0 is set when variable contains
6560 local relocations only, while bit 1 is set for global relocations.
6561 @var{align} is the constant alignment in bits.
6563 The default version of this function takes care of putting read-only
6564 variables in @code{readonly_data_section}.
6566 See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}.
6569 @defmac USE_SELECT_SECTION_FOR_FUNCTIONS
6570 Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called
6571 for @code{FUNCTION_DECL}s as well as for variables and constants.
6573 In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the
6574 function has been determined to be likely to be called, and nonzero if
6575 it is unlikely to be called.
6578 @deftypefn {Target Hook} void TARGET_ASM_UNIQUE_SECTION (tree @var{decl}, int @var{reloc})
6579 Build up a unique section name, expressed as a @code{STRING_CST} node,
6580 and assign it to @samp{DECL_SECTION_NAME (@var{decl})}.
6581 As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether
6582 the initial value of @var{exp} requires link-time relocations.
6584 The default version of this function appends the symbol name to the
6585 ELF section name that would normally be used for the symbol. For
6586 example, the function @code{foo} would be placed in @code{.text.foo}.
6587 Whatever the actual target object format, this is often good enough.
6590 @deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_RODATA_SECTION (tree @var{decl})
6591 Return the readonly data section associated with
6592 @samp{DECL_SECTION_NAME (@var{decl})}.
6593 The default version of this function selects @code{.gnu.linkonce.r.name} if
6594 the function's section is @code{.gnu.linkonce.t.name}, @code{.rodata.name}
6595 if function is in @code{.text.name}, and the normal readonly-data section
6599 @deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_RTX_SECTION (enum machine_mode @var{mode}, rtx @var{x}, unsigned HOST_WIDE_INT @var{align})
6600 Return the section into which a constant @var{x}, of mode @var{mode},
6601 should be placed. You can assume that @var{x} is some kind of
6602 constant in RTL@. The argument @var{mode} is redundant except in the
6603 case of a @code{const_int} rtx. @var{align} is the constant alignment
6606 The default version of this function takes care of putting symbolic
6607 constants in @code{flag_pic} mode in @code{data_section} and everything
6608 else in @code{readonly_data_section}.
6611 @deftypefn {Target Hook} void TARGET_MANGLE_DECL_ASSEMBLER_NAME (tree @var{decl}, tree @var{id})
6612 Define this hook if you need to postprocess the assembler name generated
6613 by target-independent code. The @var{id} provided to this hook will be
6614 the computed name (e.g., the macro @code{DECL_NAME} of the @var{decl} in C,
6615 or the mangled name of the @var{decl} in C++). The return value of the
6616 hook is an @code{IDENTIFIER_NODE} for the appropriate mangled name on
6617 your target system. The default implementation of this hook just
6618 returns the @var{id} provided.
6621 @deftypefn {Target Hook} void TARGET_ENCODE_SECTION_INFO (tree @var{decl}, rtx @var{rtl}, int @var{new_decl_p})
6622 Define this hook if references to a symbol or a constant must be
6623 treated differently depending on something about the variable or
6624 function named by the symbol (such as what section it is in).
6626 The hook is executed immediately after rtl has been created for
6627 @var{decl}, which may be a variable or function declaration or
6628 an entry in the constant pool. In either case, @var{rtl} is the
6629 rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})}
6630 in this hook; that field may not have been initialized yet.
6632 In the case of a constant, it is safe to assume that the rtl is
6633 a @code{mem} whose address is a @code{symbol_ref}. Most decls
6634 will also have this form, but that is not guaranteed. Global
6635 register variables, for instance, will have a @code{reg} for their
6636 rtl. (Normally the right thing to do with such unusual rtl is
6639 The @var{new_decl_p} argument will be true if this is the first time
6640 that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will
6641 be false for subsequent invocations, which will happen for duplicate
6642 declarations. Whether or not anything must be done for the duplicate
6643 declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}.
6644 @var{new_decl_p} is always true when the hook is called for a constant.
6646 @cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO}
6647 The usual thing for this hook to do is to record flags in the
6648 @code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}.
6649 Historically, the name string was modified if it was necessary to
6650 encode more than one bit of information, but this practice is now
6651 discouraged; use @code{SYMBOL_REF_FLAGS}.
6653 The default definition of this hook, @code{default_encode_section_info}
6654 in @file{varasm.c}, sets a number of commonly-useful bits in
6655 @code{SYMBOL_REF_FLAGS}. Check whether the default does what you need
6656 before overriding it.
6659 @deftypefn {Target Hook} const char *TARGET_STRIP_NAME_ENCODING (const char *name)
6660 Decode @var{name} and return the real name part, sans
6661 the characters that @code{TARGET_ENCODE_SECTION_INFO}
6665 @deftypefn {Target Hook} bool TARGET_IN_SMALL_DATA_P (tree @var{exp})
6666 Returns true if @var{exp} should be placed into a ``small data'' section.
6667 The default version of this hook always returns false.
6670 @deftypevar {Target Hook} bool TARGET_HAVE_SRODATA_SECTION
6671 Contains the value true if the target places read-only
6672 ``small data'' into a separate section. The default value is false.
6675 @deftypefn {Target Hook} bool TARGET_BINDS_LOCAL_P (tree @var{exp})
6676 Returns true if @var{exp} names an object for which name resolution
6677 rules must resolve to the current ``module'' (dynamic shared library
6678 or executable image).
6680 The default version of this hook implements the name resolution rules
6681 for ELF, which has a looser model of global name binding than other
6682 currently supported object file formats.
6685 @deftypevar {Target Hook} bool TARGET_HAVE_TLS
6686 Contains the value true if the target supports thread-local storage.
6687 The default value is false.
6692 @section Position Independent Code
6693 @cindex position independent code
6696 This section describes macros that help implement generation of position
6697 independent code. Simply defining these macros is not enough to
6698 generate valid PIC; you must also add support to the macros
6699 @code{GO_IF_LEGITIMATE_ADDRESS} and @code{PRINT_OPERAND_ADDRESS}, as
6700 well as @code{LEGITIMIZE_ADDRESS}. You must modify the definition of
6701 @samp{movsi} to do something appropriate when the source operand
6702 contains a symbolic address. You may also need to alter the handling of
6703 switch statements so that they use relative addresses.
6704 @c i rearranged the order of the macros above to try to force one of
6705 @c them to the next line, to eliminate an overfull hbox. --mew 10feb93
6707 @defmac PIC_OFFSET_TABLE_REGNUM
6708 The register number of the register used to address a table of static
6709 data addresses in memory. In some cases this register is defined by a
6710 processor's ``application binary interface'' (ABI)@. When this macro
6711 is defined, RTL is generated for this register once, as with the stack
6712 pointer and frame pointer registers. If this macro is not defined, it
6713 is up to the machine-dependent files to allocate such a register (if
6714 necessary). Note that this register must be fixed when in use (e.g.@:
6715 when @code{flag_pic} is true).
6718 @defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED
6719 Define this macro if the register defined by
6720 @code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. Do not define
6721 this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined.
6724 @defmac LEGITIMATE_PIC_OPERAND_P (@var{x})
6725 A C expression that is nonzero if @var{x} is a legitimate immediate
6726 operand on the target machine when generating position independent code.
6727 You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not
6728 check this. You can also assume @var{flag_pic} is true, so you need not
6729 check it either. You need not define this macro if all constants
6730 (including @code{SYMBOL_REF}) can be immediate operands when generating
6731 position independent code.
6734 @node Assembler Format
6735 @section Defining the Output Assembler Language
6737 This section describes macros whose principal purpose is to describe how
6738 to write instructions in assembler language---rather than what the
6742 * File Framework:: Structural information for the assembler file.
6743 * Data Output:: Output of constants (numbers, strings, addresses).
6744 * Uninitialized Data:: Output of uninitialized variables.
6745 * Label Output:: Output and generation of labels.
6746 * Initialization:: General principles of initialization
6747 and termination routines.
6748 * Macros for Initialization::
6749 Specific macros that control the handling of
6750 initialization and termination routines.
6751 * Instruction Output:: Output of actual instructions.
6752 * Dispatch Tables:: Output of jump tables.
6753 * Exception Region Output:: Output of exception region code.
6754 * Alignment Output:: Pseudo ops for alignment and skipping data.
6757 @node File Framework
6758 @subsection The Overall Framework of an Assembler File
6759 @cindex assembler format
6760 @cindex output of assembler code
6762 @c prevent bad page break with this line
6763 This describes the overall framework of an assembly file.
6765 @deftypefn {Target Hook} void TARGET_ASM_FILE_START ()
6766 @findex default_file_start
6767 Output to @code{asm_out_file} any text which the assembler expects to
6768 find at the beginning of a file. The default behavior is controlled
6769 by two flags, documented below. Unless your target's assembler is
6770 quite unusual, if you override the default, you should call
6771 @code{default_file_start} at some point in your target hook. This
6772 lets other target files rely on these variables.
6775 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_APP_OFF
6776 If this flag is true, the text of the macro @code{ASM_APP_OFF} will be
6777 printed as the very first line in the assembly file, unless
6778 @option{-fverbose-asm} is in effect. (If that macro has been defined
6779 to the empty string, this variable has no effect.) With the normal
6780 definition of @code{ASM_APP_OFF}, the effect is to notify the GNU
6781 assembler that it need not bother stripping comments or extra
6782 whitespace from its input. This allows it to work a bit faster.
6784 The default is false. You should not set it to true unless you have
6785 verified that your port does not generate any extra whitespace or
6786 comments that will cause GAS to issue errors in NO_APP mode.
6789 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_FILE_DIRECTIVE
6790 If this flag is true, @code{output_file_directive} will be called
6791 for the primary source file, immediately after printing
6792 @code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect
6793 this to be done. The default is false.
6796 @deftypefn {Target Hook} void TARGET_ASM_FILE_END ()
6797 Output to @code{asm_out_file} any text which the assembler expects
6798 to find at the end of a file. The default is to output nothing.
6801 @deftypefun void file_end_indicate_exec_stack ()
6802 Some systems use a common convention, the @samp{.note.GNU-stack}
6803 special section, to indicate whether or not an object file relies on
6804 the stack being executable. If your system uses this convention, you
6805 should define @code{TARGET_ASM_FILE_END} to this function. If you
6806 need to do other things in that hook, have your hook function call
6810 @defmac ASM_COMMENT_START
6811 A C string constant describing how to begin a comment in the target
6812 assembler language. The compiler assumes that the comment will end at
6813 the end of the line.
6817 A C string constant for text to be output before each @code{asm}
6818 statement or group of consecutive ones. Normally this is
6819 @code{"#APP"}, which is a comment that has no effect on most
6820 assemblers but tells the GNU assembler that it must check the lines
6821 that follow for all valid assembler constructs.
6825 A C string constant for text to be output after each @code{asm}
6826 statement or group of consecutive ones. Normally this is
6827 @code{"#NO_APP"}, which tells the GNU assembler to resume making the
6828 time-saving assumptions that are valid for ordinary compiler output.
6831 @defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
6832 A C statement to output COFF information or DWARF debugging information
6833 which indicates that filename @var{name} is the current source file to
6834 the stdio stream @var{stream}.
6836 This macro need not be defined if the standard form of output
6837 for the file format in use is appropriate.
6840 @defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string})
6841 A C statement to output the string @var{string} to the stdio stream
6842 @var{stream}. If you do not call the function @code{output_quoted_string}
6843 in your config files, GCC will only call it to output filenames to
6844 the assembler source. So you can use it to canonicalize the format
6845 of the filename using this macro.
6848 @defmac ASM_OUTPUT_IDENT (@var{stream}, @var{string})
6849 A C statement to output something to the assembler file to handle a
6850 @samp{#ident} directive containing the text @var{string}. If this
6851 macro is not defined, nothing is output for a @samp{#ident} directive.
6854 @deftypefn {Target Hook} void TARGET_ASM_NAMED_SECTION (const char *@var{name}, unsigned int @var{flags}, unsigned int @var{align})
6855 Output assembly directives to switch to section @var{name}. The section
6856 should have attributes as specified by @var{flags}, which is a bit mask
6857 of the @code{SECTION_*} flags defined in @file{output.h}. If @var{align}
6858 is nonzero, it contains an alignment in bytes to be used for the section,
6859 otherwise some target default should be used. Only targets that must
6860 specify an alignment within the section directive need pay attention to
6861 @var{align} -- we will still use @code{ASM_OUTPUT_ALIGN}.
6864 @deftypefn {Target Hook} bool TARGET_HAVE_NAMED_SECTIONS
6865 This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}.
6868 @anchor{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}
6869 @deftypefn {Target Hook} bool TARGET_HAVE_SWITCHABLE_BSS_SECTIONS
6870 This flag is true if we can create zeroed data by switching to a BSS
6871 section and then using @code{ASM_OUTPUT_SKIP} to allocate the space.
6872 This is true on most ELF targets.
6875 @deftypefn {Target Hook} {unsigned int} TARGET_SECTION_TYPE_FLAGS (tree @var{decl}, const char *@var{name}, int @var{reloc})
6876 Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION}
6877 based on a variable or function decl, a section name, and whether or not the
6878 declaration's initializer may contain runtime relocations. @var{decl} may be
6879 null, in which case read-write data should be assumed.
6881 The default version of this function handles choosing code vs data,
6882 read-only vs read-write data, and @code{flag_pic}. You should only
6883 need to override this if your target has special flags that might be
6884 set via @code{__attribute__}.
6887 @deftypefn {Target Hook} {int} TARGET_ASM_RECORD_GCC_SWITCHES (print_switch_type @var{type}, const char * @var{text})
6888 Provides the target with the ability to record the gcc command line
6889 switches that have been passed to the compiler, and options that are
6890 enabled. The @var{type} argument specifies what is being recorded.
6891 It can take the following values:
6894 @item SWITCH_TYPE_PASSED
6895 @var{text} is a command line switch that has been set by the user.
6897 @item SWITCH_TYPE_ENABLED
6898 @var{text} is an option which has been enabled. This might be as a
6899 direct result of a command line switch, or because it is enabled by
6900 default or because it has been enabled as a side effect of a different
6901 command line switch. For example, the @option{-O2} switch enables
6902 various different individual optimization passes.
6904 @item SWITCH_TYPE_DESCRIPTIVE
6905 @var{text} is either NULL or some descriptive text which should be
6906 ignored. If @var{text} is NULL then it is being used to warn the
6907 target hook that either recording is starting or ending. The first
6908 time @var{type} is SWITCH_TYPE_DESCRIPTIVE and @var{text} is NULL, the
6909 warning is for start up and the second time the warning is for
6910 wind down. This feature is to allow the target hook to make any
6911 necessary preparations before it starts to record switches and to
6912 perform any necessary tidying up after it has finished recording
6915 @item SWITCH_TYPE_LINE_START
6916 This option can be ignored by this target hook.
6918 @item SWITCH_TYPE_LINE_END
6919 This option can be ignored by this target hook.
6922 The hook's return value must be zero. Other return values may be
6923 supported in the future.
6925 By default this hook is set to NULL, but an example implementation is
6926 provided for ELF based targets. Called @var{elf_record_gcc_switches},
6927 it records the switches as ASCII text inside a new, string mergeable
6928 section in the assembler output file. The name of the new section is
6929 provided by the @code{TARGET_ASM_RECORD_GCC_SWITCHES_SECTION} target
6933 @deftypefn {Target Hook} {const char *} TARGET_ASM_RECORD_GCC_SWITCHES_SECTION
6934 This is the name of the section that will be created by the example
6935 ELF implementation of the @code{TARGET_ASM_RECORD_GCC_SWITCHES} target
6941 @subsection Output of Data
6944 @deftypevr {Target Hook} {const char *} TARGET_ASM_BYTE_OP
6945 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP
6946 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP
6947 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP
6948 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP
6949 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP
6950 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP
6951 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP
6952 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP
6953 These hooks specify assembly directives for creating certain kinds
6954 of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a
6955 byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an
6956 aligned two-byte object, and so on. Any of the hooks may be
6957 @code{NULL}, indicating that no suitable directive is available.
6959 The compiler will print these strings at the start of a new line,
6960 followed immediately by the object's initial value. In most cases,
6961 the string should contain a tab, a pseudo-op, and then another tab.
6964 @deftypefn {Target Hook} bool TARGET_ASM_INTEGER (rtx @var{x}, unsigned int @var{size}, int @var{aligned_p})
6965 The @code{assemble_integer} function uses this hook to output an
6966 integer object. @var{x} is the object's value, @var{size} is its size
6967 in bytes and @var{aligned_p} indicates whether it is aligned. The
6968 function should return @code{true} if it was able to output the
6969 object. If it returns false, @code{assemble_integer} will try to
6970 split the object into smaller parts.
6972 The default implementation of this hook will use the
6973 @code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false}
6974 when the relevant string is @code{NULL}.
6977 @defmac OUTPUT_ADDR_CONST_EXTRA (@var{stream}, @var{x}, @var{fail})
6978 A C statement to recognize @var{rtx} patterns that
6979 @code{output_addr_const} can't deal with, and output assembly code to
6980 @var{stream} corresponding to the pattern @var{x}. This may be used to
6981 allow machine-dependent @code{UNSPEC}s to appear within constants.
6983 If @code{OUTPUT_ADDR_CONST_EXTRA} fails to recognize a pattern, it must
6984 @code{goto fail}, so that a standard error message is printed. If it
6985 prints an error message itself, by calling, for example,
6986 @code{output_operand_lossage}, it may just complete normally.
6989 @defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
6990 A C statement to output to the stdio stream @var{stream} an assembler
6991 instruction to assemble a string constant containing the @var{len}
6992 bytes at @var{ptr}. @var{ptr} will be a C expression of type
6993 @code{char *} and @var{len} a C expression of type @code{int}.
6995 If the assembler has a @code{.ascii} pseudo-op as found in the
6996 Berkeley Unix assembler, do not define the macro
6997 @code{ASM_OUTPUT_ASCII}.
7000 @defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n})
7001 A C statement to output word @var{n} of a function descriptor for
7002 @var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS}
7003 is defined, and is otherwise unused.
7006 @defmac CONSTANT_POOL_BEFORE_FUNCTION
7007 You may define this macro as a C expression. You should define the
7008 expression to have a nonzero value if GCC should output the constant
7009 pool for a function before the code for the function, or a zero value if
7010 GCC should output the constant pool after the function. If you do
7011 not define this macro, the usual case, GCC will output the constant
7012 pool before the function.
7015 @defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size})
7016 A C statement to output assembler commands to define the start of the
7017 constant pool for a function. @var{funname} is a string giving
7018 the name of the function. Should the return type of the function
7019 be required, it can be obtained via @var{fundecl}. @var{size}
7020 is the size, in bytes, of the constant pool that will be written
7021 immediately after this call.
7023 If no constant-pool prefix is required, the usual case, this macro need
7027 @defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto})
7028 A C statement (with or without semicolon) to output a constant in the
7029 constant pool, if it needs special treatment. (This macro need not do
7030 anything for RTL expressions that can be output normally.)
7032 The argument @var{file} is the standard I/O stream to output the
7033 assembler code on. @var{x} is the RTL expression for the constant to
7034 output, and @var{mode} is the machine mode (in case @var{x} is a
7035 @samp{const_int}). @var{align} is the required alignment for the value
7036 @var{x}; you should output an assembler directive to force this much
7039 The argument @var{labelno} is a number to use in an internal label for
7040 the address of this pool entry. The definition of this macro is
7041 responsible for outputting the label definition at the proper place.
7042 Here is how to do this:
7045 @code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno});
7048 When you output a pool entry specially, you should end with a
7049 @code{goto} to the label @var{jumpto}. This will prevent the same pool
7050 entry from being output a second time in the usual manner.
7052 You need not define this macro if it would do nothing.
7055 @defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size})
7056 A C statement to output assembler commands to at the end of the constant
7057 pool for a function. @var{funname} is a string giving the name of the
7058 function. Should the return type of the function be required, you can
7059 obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the
7060 constant pool that GCC wrote immediately before this call.
7062 If no constant-pool epilogue is required, the usual case, you need not
7066 @defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C}, @var{STR})
7067 Define this macro as a C expression which is nonzero if @var{C} is
7068 used as a logical line separator by the assembler. @var{STR} points
7069 to the position in the string where @var{C} was found; this can be used if
7070 a line separator uses multiple characters.
7072 If you do not define this macro, the default is that only
7073 the character @samp{;} is treated as a logical line separator.
7076 @deftypevr {Target Hook} {const char *} TARGET_ASM_OPEN_PAREN
7077 @deftypevrx {Target Hook} {const char *} TARGET_ASM_CLOSE_PAREN
7078 These target hooks are C string constants, describing the syntax in the
7079 assembler for grouping arithmetic expressions. If not overridden, they
7080 default to normal parentheses, which is correct for most assemblers.
7083 These macros are provided by @file{real.h} for writing the definitions
7084 of @code{ASM_OUTPUT_DOUBLE} and the like:
7086 @defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l})
7087 @defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l})
7088 @defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l})
7089 @defmacx REAL_VALUE_TO_TARGET_DECIMAL32 (@var{x}, @var{l})
7090 @defmacx REAL_VALUE_TO_TARGET_DECIMAL64 (@var{x}, @var{l})
7091 @defmacx REAL_VALUE_TO_TARGET_DECIMAL128 (@var{x}, @var{l})
7092 These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the
7093 target's floating point representation, and store its bit pattern in
7094 the variable @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE} and
7095 @code{REAL_VALUE_TO_TARGET_DECIMAL32}, this variable should be a
7096 simple @code{long int}. For the others, it should be an array of
7097 @code{long int}. The number of elements in this array is determined
7098 by the size of the desired target floating point data type: 32 bits of
7099 it go in each @code{long int} array element. Each array element holds
7100 32 bits of the result, even if @code{long int} is wider than 32 bits
7101 on the host machine.
7103 The array element values are designed so that you can print them out
7104 using @code{fprintf} in the order they should appear in the target
7108 @node Uninitialized Data
7109 @subsection Output of Uninitialized Variables
7111 Each of the macros in this section is used to do the whole job of
7112 outputting a single uninitialized variable.
7114 @defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
7115 A C statement (sans semicolon) to output to the stdio stream
7116 @var{stream} the assembler definition of a common-label named
7117 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7118 is the size rounded up to whatever alignment the caller wants.
7120 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7121 output the name itself; before and after that, output the additional
7122 assembler syntax for defining the name, and a newline.
7124 This macro controls how the assembler definitions of uninitialized
7125 common global variables are output.
7128 @defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment})
7129 Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a
7130 separate, explicit argument. If you define this macro, it is used in
7131 place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in
7132 handling the required alignment of the variable. The alignment is specified
7133 as the number of bits.
7136 @defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7137 Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the
7138 variable to be output, if there is one, or @code{NULL_TREE} if there
7139 is no corresponding variable. If you define this macro, GCC will use it
7140 in place of both @code{ASM_OUTPUT_COMMON} and
7141 @code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see
7142 the variable's decl in order to chose what to output.
7145 @defmac ASM_OUTPUT_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded})
7146 A C statement (sans semicolon) to output to the stdio stream
7147 @var{stream} the assembler definition of uninitialized global @var{decl} named
7148 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7149 is the size rounded up to whatever alignment the caller wants.
7151 Try to use function @code{asm_output_bss} defined in @file{varasm.c} when
7152 defining this macro. If unable, use the expression
7153 @code{assemble_name (@var{stream}, @var{name})} to output the name itself;
7154 before and after that, output the additional assembler syntax for defining
7155 the name, and a newline.
7157 There are two ways of handling global BSS@. One is to define either
7158 this macro or its aligned counterpart, @code{ASM_OUTPUT_ALIGNED_BSS}.
7159 The other is to have @code{TARGET_ASM_SELECT_SECTION} return a
7160 switchable BSS section (@pxref{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}).
7161 You do not need to do both.
7163 Some languages do not have @code{common} data, and require a
7164 non-common form of global BSS in order to handle uninitialized globals
7165 efficiently. C++ is one example of this. However, if the target does
7166 not support global BSS, the front end may choose to make globals
7167 common in order to save space in the object file.
7170 @defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7171 Like @code{ASM_OUTPUT_BSS} except takes the required alignment as a
7172 separate, explicit argument. If you define this macro, it is used in
7173 place of @code{ASM_OUTPUT_BSS}, and gives you more flexibility in
7174 handling the required alignment of the variable. The alignment is specified
7175 as the number of bits.
7177 Try to use function @code{asm_output_aligned_bss} defined in file
7178 @file{varasm.c} when defining this macro.
7181 @defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
7182 A C statement (sans semicolon) to output to the stdio stream
7183 @var{stream} the assembler definition of a local-common-label named
7184 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7185 is the size rounded up to whatever alignment the caller wants.
7187 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7188 output the name itself; before and after that, output the additional
7189 assembler syntax for defining the name, and a newline.
7191 This macro controls how the assembler definitions of uninitialized
7192 static variables are output.
7195 @defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment})
7196 Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a
7197 separate, explicit argument. If you define this macro, it is used in
7198 place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in
7199 handling the required alignment of the variable. The alignment is specified
7200 as the number of bits.
7203 @defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7204 Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the
7205 variable to be output, if there is one, or @code{NULL_TREE} if there
7206 is no corresponding variable. If you define this macro, GCC will use it
7207 in place of both @code{ASM_OUTPUT_DECL} and
7208 @code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see
7209 the variable's decl in order to chose what to output.
7213 @subsection Output and Generation of Labels
7215 @c prevent bad page break with this line
7216 This is about outputting labels.
7218 @findex assemble_name
7219 @defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name})
7220 A C statement (sans semicolon) to output to the stdio stream
7221 @var{stream} the assembler definition of a label named @var{name}.
7222 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7223 output the name itself; before and after that, output the additional
7224 assembler syntax for defining the name, and a newline. A default
7225 definition of this macro is provided which is correct for most systems.
7228 @findex assemble_name_raw
7229 @defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name})
7230 Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known
7231 to refer to a compiler-generated label. The default definition uses
7232 @code{assemble_name_raw}, which is like @code{assemble_name} except
7233 that it is more efficient.
7237 A C string containing the appropriate assembler directive to specify the
7238 size of a symbol, without any arguments. On systems that use ELF, the
7239 default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other
7240 systems, the default is not to define this macro.
7242 Define this macro only if it is correct to use the default definitions
7243 of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE}
7244 for your system. If you need your own custom definitions of those
7245 macros, or if you do not need explicit symbol sizes at all, do not
7249 @defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size})
7250 A C statement (sans semicolon) to output to the stdio stream
7251 @var{stream} a directive telling the assembler that the size of the
7252 symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}.
7253 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
7257 @defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name})
7258 A C statement (sans semicolon) to output to the stdio stream
7259 @var{stream} a directive telling the assembler to calculate the size of
7260 the symbol @var{name} by subtracting its address from the current
7263 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
7264 provided. The default assumes that the assembler recognizes a special
7265 @samp{.} symbol as referring to the current address, and can calculate
7266 the difference between this and another symbol. If your assembler does
7267 not recognize @samp{.} or cannot do calculations with it, you will need
7268 to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique.
7272 A C string containing the appropriate assembler directive to specify the
7273 type of a symbol, without any arguments. On systems that use ELF, the
7274 default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other
7275 systems, the default is not to define this macro.
7277 Define this macro only if it is correct to use the default definition of
7278 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
7279 custom definition of this macro, or if you do not need explicit symbol
7280 types at all, do not define this macro.
7283 @defmac TYPE_OPERAND_FMT
7284 A C string which specifies (using @code{printf} syntax) the format of
7285 the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the
7286 default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems,
7287 the default is not to define this macro.
7289 Define this macro only if it is correct to use the default definition of
7290 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
7291 custom definition of this macro, or if you do not need explicit symbol
7292 types at all, do not define this macro.
7295 @defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type})
7296 A C statement (sans semicolon) to output to the stdio stream
7297 @var{stream} a directive telling the assembler that the type of the
7298 symbol @var{name} is @var{type}. @var{type} is a C string; currently,
7299 that string is always either @samp{"function"} or @samp{"object"}, but
7300 you should not count on this.
7302 If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default
7303 definition of this macro is provided.
7306 @defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
7307 A C statement (sans semicolon) to output to the stdio stream
7308 @var{stream} any text necessary for declaring the name @var{name} of a
7309 function which is being defined. This macro is responsible for
7310 outputting the label definition (perhaps using
7311 @code{ASM_OUTPUT_LABEL}). The argument @var{decl} is the
7312 @code{FUNCTION_DECL} tree node representing the function.
7314 If this macro is not defined, then the function name is defined in the
7315 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7317 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
7321 @defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl})
7322 A C statement (sans semicolon) to output to the stdio stream
7323 @var{stream} any text necessary for declaring the size of a function
7324 which is being defined. The argument @var{name} is the name of the
7325 function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node
7326 representing the function.
7328 If this macro is not defined, then the function size is not defined.
7330 You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition
7334 @defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl})
7335 A C statement (sans semicolon) to output to the stdio stream
7336 @var{stream} any text necessary for declaring the name @var{name} of an
7337 initialized variable which is being defined. This macro must output the
7338 label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument
7339 @var{decl} is the @code{VAR_DECL} tree node representing the variable.
7341 If this macro is not defined, then the variable name is defined in the
7342 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7344 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or
7345 @code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro.
7348 @defmac ASM_DECLARE_CONSTANT_NAME (@var{stream}, @var{name}, @var{exp}, @var{size})
7349 A C statement (sans semicolon) to output to the stdio stream
7350 @var{stream} any text necessary for declaring the name @var{name} of a
7351 constant which is being defined. This macro is responsible for
7352 outputting the label definition (perhaps using
7353 @code{ASM_OUTPUT_LABEL}). The argument @var{exp} is the
7354 value of the constant, and @var{size} is the size of the constant
7355 in bytes. @var{name} will be an internal label.
7357 If this macro is not defined, then the @var{name} is defined in the
7358 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7360 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
7364 @defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name})
7365 A C statement (sans semicolon) to output to the stdio stream
7366 @var{stream} any text necessary for claiming a register @var{regno}
7367 for a global variable @var{decl} with name @var{name}.
7369 If you don't define this macro, that is equivalent to defining it to do
7373 @defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend})
7374 A C statement (sans semicolon) to finish up declaring a variable name
7375 once the compiler has processed its initializer fully and thus has had a
7376 chance to determine the size of an array when controlled by an
7377 initializer. This is used on systems where it's necessary to declare
7378 something about the size of the object.
7380 If you don't define this macro, that is equivalent to defining it to do
7383 You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or
7384 @code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro.
7387 @deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_LABEL (FILE *@var{stream}, const char *@var{name})
7388 This target hook is a function to output to the stdio stream
7389 @var{stream} some commands that will make the label @var{name} global;
7390 that is, available for reference from other files.
7392 The default implementation relies on a proper definition of
7393 @code{GLOBAL_ASM_OP}.
7396 @deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_DECL_NAME (FILE *@var{stream}, tree @var{decl})
7397 This target hook is a function to output to the stdio stream
7398 @var{stream} some commands that will make the name associated with @var{decl}
7399 global; that is, available for reference from other files.
7401 The default implementation uses the TARGET_ASM_GLOBALIZE_LABEL target hook.
7404 @defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name})
7405 A C statement (sans semicolon) to output to the stdio stream
7406 @var{stream} some commands that will make the label @var{name} weak;
7407 that is, available for reference from other files but only used if
7408 no other definition is available. Use the expression
7409 @code{assemble_name (@var{stream}, @var{name})} to output the name
7410 itself; before and after that, output the additional assembler syntax
7411 for making that name weak, and a newline.
7413 If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not
7414 support weak symbols and you should not define the @code{SUPPORTS_WEAK}
7418 @defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value})
7419 Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and
7420 @code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function
7421 or variable decl. If @var{value} is not @code{NULL}, this C statement
7422 should output to the stdio stream @var{stream} assembler code which
7423 defines (equates) the weak symbol @var{name} to have the value
7424 @var{value}. If @var{value} is @code{NULL}, it should output commands
7425 to make @var{name} weak.
7428 @defmac ASM_OUTPUT_WEAKREF (@var{stream}, @var{decl}, @var{name}, @var{value})
7429 Outputs a directive that enables @var{name} to be used to refer to
7430 symbol @var{value} with weak-symbol semantics. @code{decl} is the
7431 declaration of @code{name}.
7434 @defmac SUPPORTS_WEAK
7435 A C expression which evaluates to true if the target supports weak symbols.
7437 If you don't define this macro, @file{defaults.h} provides a default
7438 definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL}
7439 is defined, the default definition is @samp{1}; otherwise, it is
7440 @samp{0}. Define this macro if you want to control weak symbol support
7441 with a compiler flag such as @option{-melf}.
7444 @defmac MAKE_DECL_ONE_ONLY (@var{decl})
7445 A C statement (sans semicolon) to mark @var{decl} to be emitted as a
7446 public symbol such that extra copies in multiple translation units will
7447 be discarded by the linker. Define this macro if your object file
7448 format provides support for this concept, such as the @samp{COMDAT}
7449 section flags in the Microsoft Windows PE/COFF format, and this support
7450 requires changes to @var{decl}, such as putting it in a separate section.
7453 @defmac SUPPORTS_ONE_ONLY
7454 A C expression which evaluates to true if the target supports one-only
7457 If you don't define this macro, @file{varasm.c} provides a default
7458 definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default
7459 definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if
7460 you want to control one-only symbol support with a compiler flag, or if
7461 setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to
7462 be emitted as one-only.
7465 @deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_VISIBILITY (tree @var{decl}, const char *@var{visibility})
7466 This target hook is a function to output to @var{asm_out_file} some
7467 commands that will make the symbol(s) associated with @var{decl} have
7468 hidden, protected or internal visibility as specified by @var{visibility}.
7471 @defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC
7472 A C expression that evaluates to true if the target's linker expects
7473 that weak symbols do not appear in a static archive's table of contents.
7474 The default is @code{0}.
7476 Leaving weak symbols out of an archive's table of contents means that,
7477 if a symbol will only have a definition in one translation unit and
7478 will have undefined references from other translation units, that
7479 symbol should not be weak. Defining this macro to be nonzero will
7480 thus have the effect that certain symbols that would normally be weak
7481 (explicit template instantiations, and vtables for polymorphic classes
7482 with noninline key methods) will instead be nonweak.
7484 The C++ ABI requires this macro to be zero. Define this macro for
7485 targets where full C++ ABI compliance is impossible and where linker
7486 restrictions require weak symbols to be left out of a static archive's
7490 @defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
7491 A C statement (sans semicolon) to output to the stdio stream
7492 @var{stream} any text necessary for declaring the name of an external
7493 symbol named @var{name} which is referenced in this compilation but
7494 not defined. The value of @var{decl} is the tree node for the
7497 This macro need not be defined if it does not need to output anything.
7498 The GNU assembler and most Unix assemblers don't require anything.
7501 @deftypefn {Target Hook} void TARGET_ASM_EXTERNAL_LIBCALL (rtx @var{symref})
7502 This target hook is a function to output to @var{asm_out_file} an assembler
7503 pseudo-op to declare a library function name external. The name of the
7504 library function is given by @var{symref}, which is a @code{symbol_ref}.
7507 @deftypefn {Target Hook} void TARGET_ASM_MARK_DECL_PRESERVED (tree @var{decl})
7508 This target hook is a function to output to @var{asm_out_file} an assembler
7509 directive to annotate used symbol. Darwin target use .no_dead_code_strip
7513 @defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
7514 A C statement (sans semicolon) to output to the stdio stream
7515 @var{stream} a reference in assembler syntax to a label named
7516 @var{name}. This should add @samp{_} to the front of the name, if that
7517 is customary on your operating system, as it is in most Berkeley Unix
7518 systems. This macro is used in @code{assemble_name}.
7521 @defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym})
7522 A C statement (sans semicolon) to output a reference to
7523 @code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name}
7524 will be used to output the name of the symbol. This macro may be used
7525 to modify the way a symbol is referenced depending on information
7526 encoded by @code{TARGET_ENCODE_SECTION_INFO}.
7529 @defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf})
7530 A C statement (sans semicolon) to output a reference to @var{buf}, the
7531 result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined,
7532 @code{assemble_name} will be used to output the name of the symbol.
7533 This macro is not used by @code{output_asm_label}, or the @code{%l}
7534 specifier that calls it; the intention is that this macro should be set
7535 when it is necessary to output a label differently when its address is
7539 @deftypefn {Target Hook} void TARGET_ASM_INTERNAL_LABEL (FILE *@var{stream}, const char *@var{prefix}, unsigned long @var{labelno})
7540 A function to output to the stdio stream @var{stream} a label whose
7541 name is made from the string @var{prefix} and the number @var{labelno}.
7543 It is absolutely essential that these labels be distinct from the labels
7544 used for user-level functions and variables. Otherwise, certain programs
7545 will have name conflicts with internal labels.
7547 It is desirable to exclude internal labels from the symbol table of the
7548 object file. Most assemblers have a naming convention for labels that
7549 should be excluded; on many systems, the letter @samp{L} at the
7550 beginning of a label has this effect. You should find out what
7551 convention your system uses, and follow it.
7553 The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}.
7556 @defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num})
7557 A C statement to output to the stdio stream @var{stream} a debug info
7558 label whose name is made from the string @var{prefix} and the number
7559 @var{num}. This is useful for VLIW targets, where debug info labels
7560 may need to be treated differently than branch target labels. On some
7561 systems, branch target labels must be at the beginning of instruction
7562 bundles, but debug info labels can occur in the middle of instruction
7565 If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be
7569 @defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
7570 A C statement to store into the string @var{string} a label whose name
7571 is made from the string @var{prefix} and the number @var{num}.
7573 This string, when output subsequently by @code{assemble_name}, should
7574 produce the output that @code{(*targetm.asm_out.internal_label)} would produce
7575 with the same @var{prefix} and @var{num}.
7577 If the string begins with @samp{*}, then @code{assemble_name} will
7578 output the rest of the string unchanged. It is often convenient for
7579 @code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the
7580 string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets
7581 to output the string, and may change it. (Of course,
7582 @code{ASM_OUTPUT_LABELREF} is also part of your machine description, so
7583 you should know what it does on your machine.)
7586 @defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
7587 A C expression to assign to @var{outvar} (which is a variable of type
7588 @code{char *}) a newly allocated string made from the string
7589 @var{name} and the number @var{number}, with some suitable punctuation
7590 added. Use @code{alloca} to get space for the string.
7592 The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to
7593 produce an assembler label for an internal static variable whose name is
7594 @var{name}. Therefore, the string must be such as to result in valid
7595 assembler code. The argument @var{number} is different each time this
7596 macro is executed; it prevents conflicts between similarly-named
7597 internal static variables in different scopes.
7599 Ideally this string should not be a valid C identifier, to prevent any
7600 conflict with the user's own symbols. Most assemblers allow periods
7601 or percent signs in assembler symbols; putting at least one of these
7602 between the name and the number will suffice.
7604 If this macro is not defined, a default definition will be provided
7605 which is correct for most systems.
7608 @defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value})
7609 A C statement to output to the stdio stream @var{stream} assembler code
7610 which defines (equates) the symbol @var{name} to have the value @var{value}.
7613 If @code{SET_ASM_OP} is defined, a default definition is provided which is
7614 correct for most systems.
7617 @defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value})
7618 A C statement to output to the stdio stream @var{stream} assembler code
7619 which defines (equates) the symbol whose tree node is @var{decl_of_name}
7620 to have the value of the tree node @var{decl_of_value}. This macro will
7621 be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if
7622 the tree nodes are available.
7625 If @code{SET_ASM_OP} is defined, a default definition is provided which is
7626 correct for most systems.
7629 @defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value})
7630 A C statement that evaluates to true if the assembler code which defines
7631 (equates) the symbol whose tree node is @var{decl_of_name} to have the value
7632 of the tree node @var{decl_of_value} should be emitted near the end of the
7633 current compilation unit. The default is to not defer output of defines.
7634 This macro affects defines output by @samp{ASM_OUTPUT_DEF} and
7635 @samp{ASM_OUTPUT_DEF_FROM_DECLS}.
7638 @defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value})
7639 A C statement to output to the stdio stream @var{stream} assembler code
7640 which defines (equates) the weak symbol @var{name} to have the value
7641 @var{value}. If @var{value} is @code{NULL}, it defines @var{name} as
7642 an undefined weak symbol.
7644 Define this macro if the target only supports weak aliases; define
7645 @code{ASM_OUTPUT_DEF} instead if possible.
7648 @defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name})
7649 Define this macro to override the default assembler names used for
7650 Objective-C methods.
7652 The default name is a unique method number followed by the name of the
7653 class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of
7654 the category is also included in the assembler name (e.g.@:
7657 These names are safe on most systems, but make debugging difficult since
7658 the method's selector is not present in the name. Therefore, particular
7659 systems define other ways of computing names.
7661 @var{buf} is an expression of type @code{char *} which gives you a
7662 buffer in which to store the name; its length is as long as
7663 @var{class_name}, @var{cat_name} and @var{sel_name} put together, plus
7664 50 characters extra.
7666 The argument @var{is_inst} specifies whether the method is an instance
7667 method or a class method; @var{class_name} is the name of the class;
7668 @var{cat_name} is the name of the category (or @code{NULL} if the method is not
7669 in a category); and @var{sel_name} is the name of the selector.
7671 On systems where the assembler can handle quoted names, you can use this
7672 macro to provide more human-readable names.
7675 @defmac ASM_DECLARE_CLASS_REFERENCE (@var{stream}, @var{name})
7676 A C statement (sans semicolon) to output to the stdio stream
7677 @var{stream} commands to declare that the label @var{name} is an
7678 Objective-C class reference. This is only needed for targets whose
7679 linkers have special support for NeXT-style runtimes.
7682 @defmac ASM_DECLARE_UNRESOLVED_REFERENCE (@var{stream}, @var{name})
7683 A C statement (sans semicolon) to output to the stdio stream
7684 @var{stream} commands to declare that the label @var{name} is an
7685 unresolved Objective-C class reference. This is only needed for targets
7686 whose linkers have special support for NeXT-style runtimes.
7689 @node Initialization
7690 @subsection How Initialization Functions Are Handled
7691 @cindex initialization routines
7692 @cindex termination routines
7693 @cindex constructors, output of
7694 @cindex destructors, output of
7696 The compiled code for certain languages includes @dfn{constructors}
7697 (also called @dfn{initialization routines})---functions to initialize
7698 data in the program when the program is started. These functions need
7699 to be called before the program is ``started''---that is to say, before
7700 @code{main} is called.
7702 Compiling some languages generates @dfn{destructors} (also called
7703 @dfn{termination routines}) that should be called when the program
7706 To make the initialization and termination functions work, the compiler
7707 must output something in the assembler code to cause those functions to
7708 be called at the appropriate time. When you port the compiler to a new
7709 system, you need to specify how to do this.
7711 There are two major ways that GCC currently supports the execution of
7712 initialization and termination functions. Each way has two variants.
7713 Much of the structure is common to all four variations.
7715 @findex __CTOR_LIST__
7716 @findex __DTOR_LIST__
7717 The linker must build two lists of these functions---a list of
7718 initialization functions, called @code{__CTOR_LIST__}, and a list of
7719 termination functions, called @code{__DTOR_LIST__}.
7721 Each list always begins with an ignored function pointer (which may hold
7722 0, @minus{}1, or a count of the function pointers after it, depending on
7723 the environment). This is followed by a series of zero or more function
7724 pointers to constructors (or destructors), followed by a function
7725 pointer containing zero.
7727 Depending on the operating system and its executable file format, either
7728 @file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup
7729 time and exit time. Constructors are called in reverse order of the
7730 list; destructors in forward order.
7732 The best way to handle static constructors works only for object file
7733 formats which provide arbitrarily-named sections. A section is set
7734 aside for a list of constructors, and another for a list of destructors.
7735 Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each
7736 object file that defines an initialization function also puts a word in
7737 the constructor section to point to that function. The linker
7738 accumulates all these words into one contiguous @samp{.ctors} section.
7739 Termination functions are handled similarly.
7741 This method will be chosen as the default by @file{target-def.h} if
7742 @code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not
7743 support arbitrary sections, but does support special designated
7744 constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP}
7745 and @code{DTORS_SECTION_ASM_OP} to achieve the same effect.
7747 When arbitrary sections are available, there are two variants, depending
7748 upon how the code in @file{crtstuff.c} is called. On systems that
7749 support a @dfn{.init} section which is executed at program startup,
7750 parts of @file{crtstuff.c} are compiled into that section. The
7751 program is linked by the @command{gcc} driver like this:
7754 ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o
7757 The prologue of a function (@code{__init}) appears in the @code{.init}
7758 section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise
7759 for the function @code{__fini} in the @dfn{.fini} section. Normally these
7760 files are provided by the operating system or by the GNU C library, but
7761 are provided by GCC for a few targets.
7763 The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets)
7764 compiled from @file{crtstuff.c}. They contain, among other things, code
7765 fragments within the @code{.init} and @code{.fini} sections that branch
7766 to routines in the @code{.text} section. The linker will pull all parts
7767 of a section together, which results in a complete @code{__init} function
7768 that invokes the routines we need at startup.
7770 To use this variant, you must define the @code{INIT_SECTION_ASM_OP}
7773 If no init section is available, when GCC compiles any function called
7774 @code{main} (or more accurately, any function designated as a program
7775 entry point by the language front end calling @code{expand_main_function}),
7776 it inserts a procedure call to @code{__main} as the first executable code
7777 after the function prologue. The @code{__main} function is defined
7778 in @file{libgcc2.c} and runs the global constructors.
7780 In file formats that don't support arbitrary sections, there are again
7781 two variants. In the simplest variant, the GNU linker (GNU @code{ld})
7782 and an `a.out' format must be used. In this case,
7783 @code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs}
7784 entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__},
7785 and with the address of the void function containing the initialization
7786 code as its value. The GNU linker recognizes this as a request to add
7787 the value to a @dfn{set}; the values are accumulated, and are eventually
7788 placed in the executable as a vector in the format described above, with
7789 a leading (ignored) count and a trailing zero element.
7790 @code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init
7791 section is available, the absence of @code{INIT_SECTION_ASM_OP} causes
7792 the compilation of @code{main} to call @code{__main} as above, starting
7793 the initialization process.
7795 The last variant uses neither arbitrary sections nor the GNU linker.
7796 This is preferable when you want to do dynamic linking and when using
7797 file formats which the GNU linker does not support, such as `ECOFF'@. In
7798 this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and
7799 termination functions are recognized simply by their names. This requires
7800 an extra program in the linkage step, called @command{collect2}. This program
7801 pretends to be the linker, for use with GCC; it does its job by running
7802 the ordinary linker, but also arranges to include the vectors of
7803 initialization and termination functions. These functions are called
7804 via @code{__main} as described above. In order to use this method,
7805 @code{use_collect2} must be defined in the target in @file{config.gcc}.
7808 The following section describes the specific macros that control and
7809 customize the handling of initialization and termination functions.
7812 @node Macros for Initialization
7813 @subsection Macros Controlling Initialization Routines
7815 Here are the macros that control how the compiler handles initialization
7816 and termination functions:
7818 @defmac INIT_SECTION_ASM_OP
7819 If defined, a C string constant, including spacing, for the assembler
7820 operation to identify the following data as initialization code. If not
7821 defined, GCC will assume such a section does not exist. When you are
7822 using special sections for initialization and termination functions, this
7823 macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to
7824 run the initialization functions.
7827 @defmac HAS_INIT_SECTION
7828 If defined, @code{main} will not call @code{__main} as described above.
7829 This macro should be defined for systems that control start-up code
7830 on a symbol-by-symbol basis, such as OSF/1, and should not
7831 be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}.
7834 @defmac LD_INIT_SWITCH
7835 If defined, a C string constant for a switch that tells the linker that
7836 the following symbol is an initialization routine.
7839 @defmac LD_FINI_SWITCH
7840 If defined, a C string constant for a switch that tells the linker that
7841 the following symbol is a finalization routine.
7844 @defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func})
7845 If defined, a C statement that will write a function that can be
7846 automatically called when a shared library is loaded. The function
7847 should call @var{func}, which takes no arguments. If not defined, and
7848 the object format requires an explicit initialization function, then a
7849 function called @code{_GLOBAL__DI} will be generated.
7851 This function and the following one are used by collect2 when linking a
7852 shared library that needs constructors or destructors, or has DWARF2
7853 exception tables embedded in the code.
7856 @defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func})
7857 If defined, a C statement that will write a function that can be
7858 automatically called when a shared library is unloaded. The function
7859 should call @var{func}, which takes no arguments. If not defined, and
7860 the object format requires an explicit finalization function, then a
7861 function called @code{_GLOBAL__DD} will be generated.
7864 @defmac INVOKE__main
7865 If defined, @code{main} will call @code{__main} despite the presence of
7866 @code{INIT_SECTION_ASM_OP}. This macro should be defined for systems
7867 where the init section is not actually run automatically, but is still
7868 useful for collecting the lists of constructors and destructors.
7871 @defmac SUPPORTS_INIT_PRIORITY
7872 If nonzero, the C++ @code{init_priority} attribute is supported and the
7873 compiler should emit instructions to control the order of initialization
7874 of objects. If zero, the compiler will issue an error message upon
7875 encountering an @code{init_priority} attribute.
7878 @deftypefn {Target Hook} bool TARGET_HAVE_CTORS_DTORS
7879 This value is true if the target supports some ``native'' method of
7880 collecting constructors and destructors to be run at startup and exit.
7881 It is false if we must use @command{collect2}.
7884 @deftypefn {Target Hook} void TARGET_ASM_CONSTRUCTOR (rtx @var{symbol}, int @var{priority})
7885 If defined, a function that outputs assembler code to arrange to call
7886 the function referenced by @var{symbol} at initialization time.
7888 Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking
7889 no arguments and with no return value. If the target supports initialization
7890 priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY};
7891 otherwise it must be @code{DEFAULT_INIT_PRIORITY}.
7893 If this macro is not defined by the target, a suitable default will
7894 be chosen if (1) the target supports arbitrary section names, (2) the
7895 target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2}
7899 @deftypefn {Target Hook} void TARGET_ASM_DESTRUCTOR (rtx @var{symbol}, int @var{priority})
7900 This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination
7901 functions rather than initialization functions.
7904 If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine
7905 generated for the generated object file will have static linkage.
7907 If your system uses @command{collect2} as the means of processing
7908 constructors, then that program normally uses @command{nm} to scan
7909 an object file for constructor functions to be called.
7911 On certain kinds of systems, you can define this macro to make
7912 @command{collect2} work faster (and, in some cases, make it work at all):
7914 @defmac OBJECT_FORMAT_COFF
7915 Define this macro if the system uses COFF (Common Object File Format)
7916 object files, so that @command{collect2} can assume this format and scan
7917 object files directly for dynamic constructor/destructor functions.
7919 This macro is effective only in a native compiler; @command{collect2} as
7920 part of a cross compiler always uses @command{nm} for the target machine.
7923 @defmac REAL_NM_FILE_NAME
7924 Define this macro as a C string constant containing the file name to use
7925 to execute @command{nm}. The default is to search the path normally for
7928 If your system supports shared libraries and has a program to list the
7929 dynamic dependencies of a given library or executable, you can define
7930 these macros to enable support for running initialization and
7931 termination functions in shared libraries:
7935 Define this macro to a C string constant containing the name of the program
7936 which lists dynamic dependencies, like @command{"ldd"} under SunOS 4.
7939 @defmac PARSE_LDD_OUTPUT (@var{ptr})
7940 Define this macro to be C code that extracts filenames from the output
7941 of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable
7942 of type @code{char *} that points to the beginning of a line of output
7943 from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the
7944 code must advance @var{ptr} to the beginning of the filename on that
7945 line. Otherwise, it must set @var{ptr} to @code{NULL}.
7948 @defmac SHLIB_SUFFIX
7949 Define this macro to a C string constant containing the default shared
7950 library extension of the target (e.g., @samp{".so"}). @command{collect2}
7951 strips version information after this suffix when generating global
7952 constructor and destructor names. This define is only needed on targets
7953 that use @command{collect2} to process constructors and destructors.
7956 @node Instruction Output
7957 @subsection Output of Assembler Instructions
7959 @c prevent bad page break with this line
7960 This describes assembler instruction output.
7962 @defmac REGISTER_NAMES
7963 A C initializer containing the assembler's names for the machine
7964 registers, each one as a C string constant. This is what translates
7965 register numbers in the compiler into assembler language.
7968 @defmac ADDITIONAL_REGISTER_NAMES
7969 If defined, a C initializer for an array of structures containing a name
7970 and a register number. This macro defines additional names for hard
7971 registers, thus allowing the @code{asm} option in declarations to refer
7972 to registers using alternate names.
7975 @defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
7976 Define this macro if you are using an unusual assembler that
7977 requires different names for the machine instructions.
7979 The definition is a C statement or statements which output an
7980 assembler instruction opcode to the stdio stream @var{stream}. The
7981 macro-operand @var{ptr} is a variable of type @code{char *} which
7982 points to the opcode name in its ``internal'' form---the form that is
7983 written in the machine description. The definition should output the
7984 opcode name to @var{stream}, performing any translation you desire, and
7985 increment the variable @var{ptr} to point at the end of the opcode
7986 so that it will not be output twice.
7988 In fact, your macro definition may process less than the entire opcode
7989 name, or more than the opcode name; but if you want to process text
7990 that includes @samp{%}-sequences to substitute operands, you must take
7991 care of the substitution yourself. Just be sure to increment
7992 @var{ptr} over whatever text should not be output normally.
7994 @findex recog_data.operand
7995 If you need to look at the operand values, they can be found as the
7996 elements of @code{recog_data.operand}.
7998 If the macro definition does nothing, the instruction is output
8002 @defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
8003 If defined, a C statement to be executed just prior to the output of
8004 assembler code for @var{insn}, to modify the extracted operands so
8005 they will be output differently.
8007 Here the argument @var{opvec} is the vector containing the operands
8008 extracted from @var{insn}, and @var{noperands} is the number of
8009 elements of the vector which contain meaningful data for this insn.
8010 The contents of this vector are what will be used to convert the insn
8011 template into assembler code, so you can change the assembler output
8012 by changing the contents of the vector.
8014 This macro is useful when various assembler syntaxes share a single
8015 file of instruction patterns; by defining this macro differently, you
8016 can cause a large class of instructions to be output differently (such
8017 as with rearranged operands). Naturally, variations in assembler
8018 syntax affecting individual insn patterns ought to be handled by
8019 writing conditional output routines in those patterns.
8021 If this macro is not defined, it is equivalent to a null statement.
8024 @defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
8025 A C compound statement to output to stdio stream @var{stream} the
8026 assembler syntax for an instruction operand @var{x}. @var{x} is an
8029 @var{code} is a value that can be used to specify one of several ways
8030 of printing the operand. It is used when identical operands must be
8031 printed differently depending on the context. @var{code} comes from
8032 the @samp{%} specification that was used to request printing of the
8033 operand. If the specification was just @samp{%@var{digit}} then
8034 @var{code} is 0; if the specification was @samp{%@var{ltr}
8035 @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
8038 If @var{x} is a register, this macro should print the register's name.
8039 The names can be found in an array @code{reg_names} whose type is
8040 @code{char *[]}. @code{reg_names} is initialized from
8041 @code{REGISTER_NAMES}.
8043 When the machine description has a specification @samp{%@var{punct}}
8044 (a @samp{%} followed by a punctuation character), this macro is called
8045 with a null pointer for @var{x} and the punctuation character for
8049 @defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code})
8050 A C expression which evaluates to true if @var{code} is a valid
8051 punctuation character for use in the @code{PRINT_OPERAND} macro. If
8052 @code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
8053 punctuation characters (except for the standard one, @samp{%}) are used
8057 @defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
8058 A C compound statement to output to stdio stream @var{stream} the
8059 assembler syntax for an instruction operand that is a memory reference
8060 whose address is @var{x}. @var{x} is an RTL expression.
8062 @cindex @code{TARGET_ENCODE_SECTION_INFO} usage
8063 On some machines, the syntax for a symbolic address depends on the
8064 section that the address refers to. On these machines, define the hook
8065 @code{TARGET_ENCODE_SECTION_INFO} to store the information into the
8066 @code{symbol_ref}, and then check for it here. @xref{Assembler
8070 @findex dbr_sequence_length
8071 @defmac DBR_OUTPUT_SEQEND (@var{file})
8072 A C statement, to be executed after all slot-filler instructions have
8073 been output. If necessary, call @code{dbr_sequence_length} to
8074 determine the number of slots filled in a sequence (zero if not
8075 currently outputting a sequence), to decide how many no-ops to output,
8078 Don't define this macro if it has nothing to do, but it is helpful in
8079 reading assembly output if the extent of the delay sequence is made
8080 explicit (e.g.@: with white space).
8083 @findex final_sequence
8084 Note that output routines for instructions with delay slots must be
8085 prepared to deal with not being output as part of a sequence
8086 (i.e.@: when the scheduling pass is not run, or when no slot fillers could be
8087 found.) The variable @code{final_sequence} is null when not
8088 processing a sequence, otherwise it contains the @code{sequence} rtx
8092 @defmac REGISTER_PREFIX
8093 @defmacx LOCAL_LABEL_PREFIX
8094 @defmacx USER_LABEL_PREFIX
8095 @defmacx IMMEDIATE_PREFIX
8096 If defined, C string expressions to be used for the @samp{%R}, @samp{%L},
8097 @samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see
8098 @file{final.c}). These are useful when a single @file{md} file must
8099 support multiple assembler formats. In that case, the various @file{tm.h}
8100 files can define these macros differently.
8103 @defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format})
8104 If defined this macro should expand to a series of @code{case}
8105 statements which will be parsed inside the @code{switch} statement of
8106 the @code{asm_fprintf} function. This allows targets to define extra
8107 printf formats which may useful when generating their assembler
8108 statements. Note that uppercase letters are reserved for future
8109 generic extensions to asm_fprintf, and so are not available to target
8110 specific code. The output file is given by the parameter @var{file}.
8111 The varargs input pointer is @var{argptr} and the rest of the format
8112 string, starting the character after the one that is being switched
8113 upon, is pointed to by @var{format}.
8116 @defmac ASSEMBLER_DIALECT
8117 If your target supports multiple dialects of assembler language (such as
8118 different opcodes), define this macro as a C expression that gives the
8119 numeric index of the assembler language dialect to use, with zero as the
8122 If this macro is defined, you may use constructs of the form
8124 @samp{@{option0|option1|option2@dots{}@}}
8127 in the output templates of patterns (@pxref{Output Template}) or in the
8128 first argument of @code{asm_fprintf}. This construct outputs
8129 @samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of
8130 @code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters
8131 within these strings retain their usual meaning. If there are fewer
8132 alternatives within the braces than the value of
8133 @code{ASSEMBLER_DIALECT}, the construct outputs nothing.
8135 If you do not define this macro, the characters @samp{@{}, @samp{|} and
8136 @samp{@}} do not have any special meaning when used in templates or
8137 operands to @code{asm_fprintf}.
8139 Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX},
8140 @code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express
8141 the variations in assembler language syntax with that mechanism. Define
8142 @code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax
8143 if the syntax variant are larger and involve such things as different
8144 opcodes or operand order.
8147 @defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
8148 A C expression to output to @var{stream} some assembler code
8149 which will push hard register number @var{regno} onto the stack.
8150 The code need not be optimal, since this macro is used only when
8154 @defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
8155 A C expression to output to @var{stream} some assembler code
8156 which will pop hard register number @var{regno} off of the stack.
8157 The code need not be optimal, since this macro is used only when
8161 @node Dispatch Tables
8162 @subsection Output of Dispatch Tables
8164 @c prevent bad page break with this line
8165 This concerns dispatch tables.
8167 @cindex dispatch table
8168 @defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel})
8169 A C statement to output to the stdio stream @var{stream} an assembler
8170 pseudo-instruction to generate a difference between two labels.
8171 @var{value} and @var{rel} are the numbers of two internal labels. The
8172 definitions of these labels are output using
8173 @code{(*targetm.asm_out.internal_label)}, and they must be printed in the same
8174 way here. For example,
8177 fprintf (@var{stream}, "\t.word L%d-L%d\n",
8178 @var{value}, @var{rel})
8181 You must provide this macro on machines where the addresses in a
8182 dispatch table are relative to the table's own address. If defined, GCC
8183 will also use this macro on all machines when producing PIC@.
8184 @var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the
8185 mode and flags can be read.
8188 @defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
8189 This macro should be provided on machines where the addresses
8190 in a dispatch table are absolute.
8192 The definition should be a C statement to output to the stdio stream
8193 @var{stream} an assembler pseudo-instruction to generate a reference to
8194 a label. @var{value} is the number of an internal label whose
8195 definition is output using @code{(*targetm.asm_out.internal_label)}.
8199 fprintf (@var{stream}, "\t.word L%d\n", @var{value})
8203 @defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
8204 Define this if the label before a jump-table needs to be output
8205 specially. The first three arguments are the same as for
8206 @code{(*targetm.asm_out.internal_label)}; the fourth argument is the
8207 jump-table which follows (a @code{jump_insn} containing an
8208 @code{addr_vec} or @code{addr_diff_vec}).
8210 This feature is used on system V to output a @code{swbeg} statement
8213 If this macro is not defined, these labels are output with
8214 @code{(*targetm.asm_out.internal_label)}.
8217 @defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
8218 Define this if something special must be output at the end of a
8219 jump-table. The definition should be a C statement to be executed
8220 after the assembler code for the table is written. It should write
8221 the appropriate code to stdio stream @var{stream}. The argument
8222 @var{table} is the jump-table insn, and @var{num} is the label-number
8223 of the preceding label.
8225 If this macro is not defined, nothing special is output at the end of
8229 @deftypefn {Target Hook} void TARGET_ASM_EMIT_UNWIND_LABEL (@var{stream}, @var{decl}, @var{for_eh}, @var{empty})
8230 This target hook emits a label at the beginning of each FDE@. It
8231 should be defined on targets where FDEs need special labels, and it
8232 should write the appropriate label, for the FDE associated with the
8233 function declaration @var{decl}, to the stdio stream @var{stream}.
8234 The third argument, @var{for_eh}, is a boolean: true if this is for an
8235 exception table. The fourth argument, @var{empty}, is a boolean:
8236 true if this is a placeholder label for an omitted FDE@.
8238 The default is that FDEs are not given nonlocal labels.
8241 @deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL (@var{stream})
8242 This target hook emits a label at the beginning of the exception table.
8243 It should be defined on targets where it is desirable for the table
8244 to be broken up according to function.
8246 The default is that no label is emitted.
8249 @deftypefn {Target Hook} void TARGET_UNWIND_EMIT (FILE * @var{stream}, rtx @var{insn})
8250 This target hook emits and assembly directives required to unwind the
8251 given instruction. This is only used when TARGET_UNWIND_INFO is set.
8254 @node Exception Region Output
8255 @subsection Assembler Commands for Exception Regions
8257 @c prevent bad page break with this line
8259 This describes commands marking the start and the end of an exception
8262 @defmac EH_FRAME_SECTION_NAME
8263 If defined, a C string constant for the name of the section containing
8264 exception handling frame unwind information. If not defined, GCC will
8265 provide a default definition if the target supports named sections.
8266 @file{crtstuff.c} uses this macro to switch to the appropriate section.
8268 You should define this symbol if your target supports DWARF 2 frame
8269 unwind information and the default definition does not work.
8272 @defmac EH_FRAME_IN_DATA_SECTION
8273 If defined, DWARF 2 frame unwind information will be placed in the
8274 data section even though the target supports named sections. This
8275 might be necessary, for instance, if the system linker does garbage
8276 collection and sections cannot be marked as not to be collected.
8278 Do not define this macro unless @code{TARGET_ASM_NAMED_SECTION} is
8282 @defmac EH_TABLES_CAN_BE_READ_ONLY
8283 Define this macro to 1 if your target is such that no frame unwind
8284 information encoding used with non-PIC code will ever require a
8285 runtime relocation, but the linker may not support merging read-only
8286 and read-write sections into a single read-write section.
8289 @defmac MASK_RETURN_ADDR
8290 An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so
8291 that it does not contain any extraneous set bits in it.
8294 @defmac DWARF2_UNWIND_INFO
8295 Define this macro to 0 if your target supports DWARF 2 frame unwind
8296 information, but it does not yet work with exception handling.
8297 Otherwise, if your target supports this information (if it defines
8298 @samp{INCOMING_RETURN_ADDR_RTX} and either @samp{UNALIGNED_INT_ASM_OP}
8299 or @samp{OBJECT_FORMAT_ELF}), GCC will provide a default definition of 1.
8301 If @code{TARGET_UNWIND_INFO} is defined, the target specific unwinder
8302 will be used in all cases. Defining this macro will enable the generation
8303 of DWARF 2 frame debugging information.
8305 If @code{TARGET_UNWIND_INFO} is not defined, and this macro is defined to 1,
8306 the DWARF 2 unwinder will be the default exception handling mechanism;
8307 otherwise, the @code{setjmp}/@code{longjmp}-based scheme will be used by
8311 @defmac TARGET_UNWIND_INFO
8312 Define this macro if your target has ABI specified unwind tables. Usually
8313 these will be output by @code{TARGET_UNWIND_EMIT}.
8316 @deftypevar {Target Hook} bool TARGET_UNWIND_TABLES_DEFAULT
8317 This variable should be set to @code{true} if the target ABI requires unwinding
8318 tables even when exceptions are not used.
8321 @defmac MUST_USE_SJLJ_EXCEPTIONS
8322 This macro need only be defined if @code{DWARF2_UNWIND_INFO} is
8323 runtime-variable. In that case, @file{except.h} cannot correctly
8324 determine the corresponding definition of @code{MUST_USE_SJLJ_EXCEPTIONS},
8325 so the target must provide it directly.
8328 @defmac DONT_USE_BUILTIN_SETJMP
8329 Define this macro to 1 if the @code{setjmp}/@code{longjmp}-based scheme
8330 should use the @code{setjmp}/@code{longjmp} functions from the C library
8331 instead of the @code{__builtin_setjmp}/@code{__builtin_longjmp} machinery.
8334 @defmac DWARF_CIE_DATA_ALIGNMENT
8335 This macro need only be defined if the target might save registers in the
8336 function prologue at an offset to the stack pointer that is not aligned to
8337 @code{UNITS_PER_WORD}. The definition should be the negative minimum
8338 alignment if @code{STACK_GROWS_DOWNWARD} is defined, and the positive
8339 minimum alignment otherwise. @xref{SDB and DWARF}. Only applicable if
8340 the target supports DWARF 2 frame unwind information.
8343 @deftypevar {Target Hook} bool TARGET_TERMINATE_DW2_EH_FRAME_INFO
8344 Contains the value true if the target should add a zero word onto the
8345 end of a Dwarf-2 frame info section when used for exception handling.
8346 Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and
8350 @deftypefn {Target Hook} rtx TARGET_DWARF_REGISTER_SPAN (rtx @var{reg})
8351 Given a register, this hook should return a parallel of registers to
8352 represent where to find the register pieces. Define this hook if the
8353 register and its mode are represented in Dwarf in non-contiguous
8354 locations, or if the register should be represented in more than one
8355 register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}.
8356 If not defined, the default is to return @code{NULL_RTX}.
8359 @deftypefn {Target Hook} void TARGET_INIT_DWARF_REG_SIZES_EXTRA (tree @var{address})
8360 If some registers are represented in Dwarf-2 unwind information in
8361 multiple pieces, define this hook to fill in information about the
8362 sizes of those pieces in the table used by the unwinder at runtime.
8363 It will be called by @code{expand_builtin_init_dwarf_reg_sizes} after
8364 filling in a single size corresponding to each hard register;
8365 @var{address} is the address of the table.
8368 @deftypefn {Target Hook} bool TARGET_ASM_TTYPE (rtx @var{sym})
8369 This hook is used to output a reference from a frame unwinding table to
8370 the type_info object identified by @var{sym}. It should return @code{true}
8371 if the reference was output. Returning @code{false} will cause the
8372 reference to be output using the normal Dwarf2 routines.
8375 @deftypefn {Target Hook} bool TARGET_ARM_EABI_UNWINDER
8376 This hook should be set to @code{true} on targets that use an ARM EABI
8377 based unwinding library, and @code{false} on other targets. This effects
8378 the format of unwinding tables, and how the unwinder in entered after
8379 running a cleanup. The default is @code{false}.
8382 @node Alignment Output
8383 @subsection Assembler Commands for Alignment
8385 @c prevent bad page break with this line
8386 This describes commands for alignment.
8388 @defmac JUMP_ALIGN (@var{label})
8389 The alignment (log base 2) to put in front of @var{label}, which is
8390 a common destination of jumps and has no fallthru incoming edge.
8392 This macro need not be defined if you don't want any special alignment
8393 to be done at such a time. Most machine descriptions do not currently
8396 Unless it's necessary to inspect the @var{label} parameter, it is better
8397 to set the variable @var{align_jumps} in the target's
8398 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
8399 selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation.
8402 @defmac LABEL_ALIGN_AFTER_BARRIER (@var{label})
8403 The alignment (log base 2) to put in front of @var{label}, which follows
8406 This macro need not be defined if you don't want any special alignment
8407 to be done at such a time. Most machine descriptions do not currently
8411 @defmac LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP
8412 The maximum number of bytes to skip when applying
8413 @code{LABEL_ALIGN_AFTER_BARRIER}. This works only if
8414 @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
8417 @defmac LOOP_ALIGN (@var{label})
8418 The alignment (log base 2) to put in front of @var{label}, which follows
8419 a @code{NOTE_INSN_LOOP_BEG} note.
8421 This macro need not be defined if you don't want any special alignment
8422 to be done at such a time. Most machine descriptions do not currently
8425 Unless it's necessary to inspect the @var{label} parameter, it is better
8426 to set the variable @code{align_loops} in the target's
8427 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
8428 selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation.
8431 @defmac LOOP_ALIGN_MAX_SKIP
8432 The maximum number of bytes to skip when applying @code{LOOP_ALIGN}.
8433 This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
8436 @defmac LABEL_ALIGN (@var{label})
8437 The alignment (log base 2) to put in front of @var{label}.
8438 If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment,
8439 the maximum of the specified values is used.
8441 Unless it's necessary to inspect the @var{label} parameter, it is better
8442 to set the variable @code{align_labels} in the target's
8443 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
8444 selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation.
8447 @defmac LABEL_ALIGN_MAX_SKIP
8448 The maximum number of bytes to skip when applying @code{LABEL_ALIGN}.
8449 This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
8452 @defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
8453 A C statement to output to the stdio stream @var{stream} an assembler
8454 instruction to advance the location counter by @var{nbytes} bytes.
8455 Those bytes should be zero when loaded. @var{nbytes} will be a C
8456 expression of type @code{unsigned HOST_WIDE_INT}.
8459 @defmac ASM_NO_SKIP_IN_TEXT
8460 Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the
8461 text section because it fails to put zeros in the bytes that are skipped.
8462 This is true on many Unix systems, where the pseudo--op to skip bytes
8463 produces no-op instructions rather than zeros when used in the text
8467 @defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
8468 A C statement to output to the stdio stream @var{stream} an assembler
8469 command to advance the location counter to a multiple of 2 to the
8470 @var{power} bytes. @var{power} will be a C expression of type @code{int}.
8473 @defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power})
8474 Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used
8475 for padding, if necessary.
8478 @defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip})
8479 A C statement to output to the stdio stream @var{stream} an assembler
8480 command to advance the location counter to a multiple of 2 to the
8481 @var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to
8482 satisfy the alignment request. @var{power} and @var{max_skip} will be
8483 a C expression of type @code{int}.
8487 @node Debugging Info
8488 @section Controlling Debugging Information Format
8490 @c prevent bad page break with this line
8491 This describes how to specify debugging information.
8494 * All Debuggers:: Macros that affect all debugging formats uniformly.
8495 * DBX Options:: Macros enabling specific options in DBX format.
8496 * DBX Hooks:: Hook macros for varying DBX format.
8497 * File Names and DBX:: Macros controlling output of file names in DBX format.
8498 * SDB and DWARF:: Macros for SDB (COFF) and DWARF formats.
8499 * VMS Debug:: Macros for VMS debug format.
8503 @subsection Macros Affecting All Debugging Formats
8505 @c prevent bad page break with this line
8506 These macros affect all debugging formats.
8508 @defmac DBX_REGISTER_NUMBER (@var{regno})
8509 A C expression that returns the DBX register number for the compiler
8510 register number @var{regno}. In the default macro provided, the value
8511 of this expression will be @var{regno} itself. But sometimes there are
8512 some registers that the compiler knows about and DBX does not, or vice
8513 versa. In such cases, some register may need to have one number in the
8514 compiler and another for DBX@.
8516 If two registers have consecutive numbers inside GCC, and they can be
8517 used as a pair to hold a multiword value, then they @emph{must} have
8518 consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}.
8519 Otherwise, debuggers will be unable to access such a pair, because they
8520 expect register pairs to be consecutive in their own numbering scheme.
8522 If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that
8523 does not preserve register pairs, then what you must do instead is
8524 redefine the actual register numbering scheme.
8527 @defmac DEBUGGER_AUTO_OFFSET (@var{x})
8528 A C expression that returns the integer offset value for an automatic
8529 variable having address @var{x} (an RTL expression). The default
8530 computation assumes that @var{x} is based on the frame-pointer and
8531 gives the offset from the frame-pointer. This is required for targets
8532 that produce debugging output for DBX or COFF-style debugging output
8533 for SDB and allow the frame-pointer to be eliminated when the
8534 @option{-g} options is used.
8537 @defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x})
8538 A C expression that returns the integer offset value for an argument
8539 having address @var{x} (an RTL expression). The nominal offset is
8543 @defmac PREFERRED_DEBUGGING_TYPE
8544 A C expression that returns the type of debugging output GCC should
8545 produce when the user specifies just @option{-g}. Define
8546 this if you have arranged for GCC to support more than one format of
8547 debugging output. Currently, the allowable values are @code{DBX_DEBUG},
8548 @code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG},
8549 @code{XCOFF_DEBUG}, @code{VMS_DEBUG}, and @code{VMS_AND_DWARF2_DEBUG}.
8551 When the user specifies @option{-ggdb}, GCC normally also uses the
8552 value of this macro to select the debugging output format, but with two
8553 exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the
8554 value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is
8555 defined, GCC uses @code{DBX_DEBUG}.
8557 The value of this macro only affects the default debugging output; the
8558 user can always get a specific type of output by using @option{-gstabs},
8559 @option{-gcoff}, @option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}.
8563 @subsection Specific Options for DBX Output
8565 @c prevent bad page break with this line
8566 These are specific options for DBX output.
8568 @defmac DBX_DEBUGGING_INFO
8569 Define this macro if GCC should produce debugging output for DBX
8570 in response to the @option{-g} option.
8573 @defmac XCOFF_DEBUGGING_INFO
8574 Define this macro if GCC should produce XCOFF format debugging output
8575 in response to the @option{-g} option. This is a variant of DBX format.
8578 @defmac DEFAULT_GDB_EXTENSIONS
8579 Define this macro to control whether GCC should by default generate
8580 GDB's extended version of DBX debugging information (assuming DBX-format
8581 debugging information is enabled at all). If you don't define the
8582 macro, the default is 1: always generate the extended information
8583 if there is any occasion to.
8586 @defmac DEBUG_SYMS_TEXT
8587 Define this macro if all @code{.stabs} commands should be output while
8588 in the text section.
8591 @defmac ASM_STABS_OP
8592 A C string constant, including spacing, naming the assembler pseudo op to
8593 use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol.
8594 If you don't define this macro, @code{"\t.stabs\t"} is used. This macro
8595 applies only to DBX debugging information format.
8598 @defmac ASM_STABD_OP
8599 A C string constant, including spacing, naming the assembler pseudo op to
8600 use instead of @code{"\t.stabd\t"} to define a debugging symbol whose
8601 value is the current location. If you don't define this macro,
8602 @code{"\t.stabd\t"} is used. This macro applies only to DBX debugging
8606 @defmac ASM_STABN_OP
8607 A C string constant, including spacing, naming the assembler pseudo op to
8608 use instead of @code{"\t.stabn\t"} to define a debugging symbol with no
8609 name. If you don't define this macro, @code{"\t.stabn\t"} is used. This
8610 macro applies only to DBX debugging information format.
8613 @defmac DBX_NO_XREFS
8614 Define this macro if DBX on your system does not support the construct
8615 @samp{xs@var{tagname}}. On some systems, this construct is used to
8616 describe a forward reference to a structure named @var{tagname}.
8617 On other systems, this construct is not supported at all.
8620 @defmac DBX_CONTIN_LENGTH
8621 A symbol name in DBX-format debugging information is normally
8622 continued (split into two separate @code{.stabs} directives) when it
8623 exceeds a certain length (by default, 80 characters). On some
8624 operating systems, DBX requires this splitting; on others, splitting
8625 must not be done. You can inhibit splitting by defining this macro
8626 with the value zero. You can override the default splitting-length by
8627 defining this macro as an expression for the length you desire.
8630 @defmac DBX_CONTIN_CHAR
8631 Normally continuation is indicated by adding a @samp{\} character to
8632 the end of a @code{.stabs} string when a continuation follows. To use
8633 a different character instead, define this macro as a character
8634 constant for the character you want to use. Do not define this macro
8635 if backslash is correct for your system.
8638 @defmac DBX_STATIC_STAB_DATA_SECTION
8639 Define this macro if it is necessary to go to the data section before
8640 outputting the @samp{.stabs} pseudo-op for a non-global static
8644 @defmac DBX_TYPE_DECL_STABS_CODE
8645 The value to use in the ``code'' field of the @code{.stabs} directive
8646 for a typedef. The default is @code{N_LSYM}.
8649 @defmac DBX_STATIC_CONST_VAR_CODE
8650 The value to use in the ``code'' field of the @code{.stabs} directive
8651 for a static variable located in the text section. DBX format does not
8652 provide any ``right'' way to do this. The default is @code{N_FUN}.
8655 @defmac DBX_REGPARM_STABS_CODE
8656 The value to use in the ``code'' field of the @code{.stabs} directive
8657 for a parameter passed in registers. DBX format does not provide any
8658 ``right'' way to do this. The default is @code{N_RSYM}.
8661 @defmac DBX_REGPARM_STABS_LETTER
8662 The letter to use in DBX symbol data to identify a symbol as a parameter
8663 passed in registers. DBX format does not customarily provide any way to
8664 do this. The default is @code{'P'}.
8667 @defmac DBX_FUNCTION_FIRST
8668 Define this macro if the DBX information for a function and its
8669 arguments should precede the assembler code for the function. Normally,
8670 in DBX format, the debugging information entirely follows the assembler
8674 @defmac DBX_BLOCKS_FUNCTION_RELATIVE
8675 Define this macro, with value 1, if the value of a symbol describing
8676 the scope of a block (@code{N_LBRAC} or @code{N_RBRAC}) should be
8677 relative to the start of the enclosing function. Normally, GCC uses
8678 an absolute address.
8681 @defmac DBX_LINES_FUNCTION_RELATIVE
8682 Define this macro, with value 1, if the value of a symbol indicating
8683 the current line number (@code{N_SLINE}) should be relative to the
8684 start of the enclosing function. Normally, GCC uses an absolute address.
8687 @defmac DBX_USE_BINCL
8688 Define this macro if GCC should generate @code{N_BINCL} and
8689 @code{N_EINCL} stabs for included header files, as on Sun systems. This
8690 macro also directs GCC to output a type number as a pair of a file
8691 number and a type number within the file. Normally, GCC does not
8692 generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single
8693 number for a type number.
8697 @subsection Open-Ended Hooks for DBX Format
8699 @c prevent bad page break with this line
8700 These are hooks for DBX format.
8702 @defmac DBX_OUTPUT_LBRAC (@var{stream}, @var{name})
8703 Define this macro to say how to output to @var{stream} the debugging
8704 information for the start of a scope level for variable names. The
8705 argument @var{name} is the name of an assembler symbol (for use with
8706 @code{assemble_name}) whose value is the address where the scope begins.
8709 @defmac DBX_OUTPUT_RBRAC (@var{stream}, @var{name})
8710 Like @code{DBX_OUTPUT_LBRAC}, but for the end of a scope level.
8713 @defmac DBX_OUTPUT_NFUN (@var{stream}, @var{lscope_label}, @var{decl})
8714 Define this macro if the target machine requires special handling to
8715 output an @code{N_FUN} entry for the function @var{decl}.
8718 @defmac DBX_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter})
8719 A C statement to output DBX debugging information before code for line
8720 number @var{line} of the current source file to the stdio stream
8721 @var{stream}. @var{counter} is the number of time the macro was
8722 invoked, including the current invocation; it is intended to generate
8723 unique labels in the assembly output.
8725 This macro should not be defined if the default output is correct, or
8726 if it can be made correct by defining @code{DBX_LINES_FUNCTION_RELATIVE}.
8729 @defmac NO_DBX_FUNCTION_END
8730 Some stabs encapsulation formats (in particular ECOFF), cannot handle the
8731 @code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct.
8732 On those machines, define this macro to turn this feature off without
8733 disturbing the rest of the gdb extensions.
8736 @defmac NO_DBX_BNSYM_ENSYM
8737 Some assemblers cannot handle the @code{.stabd BNSYM/ENSYM,0,0} gdb dbx
8738 extension construct. On those machines, define this macro to turn this
8739 feature off without disturbing the rest of the gdb extensions.
8742 @node File Names and DBX
8743 @subsection File Names in DBX Format
8745 @c prevent bad page break with this line
8746 This describes file names in DBX format.
8748 @defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name})
8749 A C statement to output DBX debugging information to the stdio stream
8750 @var{stream}, which indicates that file @var{name} is the main source
8751 file---the file specified as the input file for compilation.
8752 This macro is called only once, at the beginning of compilation.
8754 This macro need not be defined if the standard form of output
8755 for DBX debugging information is appropriate.
8757 It may be necessary to refer to a label equal to the beginning of the
8758 text section. You can use @samp{assemble_name (stream, ltext_label_name)}
8759 to do so. If you do this, you must also set the variable
8760 @var{used_ltext_label_name} to @code{true}.
8763 @defmac NO_DBX_MAIN_SOURCE_DIRECTORY
8764 Define this macro, with value 1, if GCC should not emit an indication
8765 of the current directory for compilation and current source language at
8766 the beginning of the file.
8769 @defmac NO_DBX_GCC_MARKER
8770 Define this macro, with value 1, if GCC should not emit an indication
8771 that this object file was compiled by GCC@. The default is to emit
8772 an @code{N_OPT} stab at the beginning of every source file, with
8773 @samp{gcc2_compiled.} for the string and value 0.
8776 @defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name})
8777 A C statement to output DBX debugging information at the end of
8778 compilation of the main source file @var{name}. Output should be
8779 written to the stdio stream @var{stream}.
8781 If you don't define this macro, nothing special is output at the end
8782 of compilation, which is correct for most machines.
8785 @defmac DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END
8786 Define this macro @emph{instead of} defining
8787 @code{DBX_OUTPUT_MAIN_SOURCE_FILE_END}, if what needs to be output at
8788 the end of compilation is a @code{N_SO} stab with an empty string,
8789 whose value is the highest absolute text address in the file.
8794 @subsection Macros for SDB and DWARF Output
8796 @c prevent bad page break with this line
8797 Here are macros for SDB and DWARF output.
8799 @defmac SDB_DEBUGGING_INFO
8800 Define this macro if GCC should produce COFF-style debugging output
8801 for SDB in response to the @option{-g} option.
8804 @defmac DWARF2_DEBUGGING_INFO
8805 Define this macro if GCC should produce dwarf version 2 format
8806 debugging output in response to the @option{-g} option.
8808 @deftypefn {Target Hook} int TARGET_DWARF_CALLING_CONVENTION (tree @var{function})
8809 Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to
8810 be emitted for each function. Instead of an integer return the enum
8811 value for the @code{DW_CC_} tag.
8814 To support optional call frame debugging information, you must also
8815 define @code{INCOMING_RETURN_ADDR_RTX} and either set
8816 @code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the
8817 prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save}
8818 as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't.
8821 @defmac DWARF2_FRAME_INFO
8822 Define this macro to a nonzero value if GCC should always output
8823 Dwarf 2 frame information. If @code{DWARF2_UNWIND_INFO}
8824 (@pxref{Exception Region Output} is nonzero, GCC will output this
8825 information not matter how you define @code{DWARF2_FRAME_INFO}.
8828 @defmac DWARF2_ASM_LINE_DEBUG_INFO
8829 Define this macro to be a nonzero value if the assembler can generate Dwarf 2
8830 line debug info sections. This will result in much more compact line number
8831 tables, and hence is desirable if it works.
8834 @defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
8835 A C statement to issue assembly directives that create a difference
8836 @var{lab1} minus @var{lab2}, using an integer of the given @var{size}.
8839 @defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label}, @var{section})
8840 A C statement to issue assembly directives that create a
8841 section-relative reference to the given @var{label}, using an integer of the
8842 given @var{size}. The label is known to be defined in the given @var{section}.
8845 @defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label})
8846 A C statement to issue assembly directives that create a self-relative
8847 reference to the given @var{label}, using an integer of the given @var{size}.
8850 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_DWARF_DTPREL (FILE *@var{FILE}, int @var{size}, rtx @var{x})
8851 If defined, this target hook is a function which outputs a DTP-relative
8852 reference to the given TLS symbol of the specified size.
8855 @defmac PUT_SDB_@dots{}
8856 Define these macros to override the assembler syntax for the special
8857 SDB assembler directives. See @file{sdbout.c} for a list of these
8858 macros and their arguments. If the standard syntax is used, you need
8859 not define them yourself.
8863 Some assemblers do not support a semicolon as a delimiter, even between
8864 SDB assembler directives. In that case, define this macro to be the
8865 delimiter to use (usually @samp{\n}). It is not necessary to define
8866 a new set of @code{PUT_SDB_@var{op}} macros if this is the only change
8870 @defmac SDB_ALLOW_UNKNOWN_REFERENCES
8871 Define this macro to allow references to unknown structure,
8872 union, or enumeration tags to be emitted. Standard COFF does not
8873 allow handling of unknown references, MIPS ECOFF has support for
8877 @defmac SDB_ALLOW_FORWARD_REFERENCES
8878 Define this macro to allow references to structure, union, or
8879 enumeration tags that have not yet been seen to be handled. Some
8880 assemblers choke if forward tags are used, while some require it.
8883 @defmac SDB_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
8884 A C statement to output SDB debugging information before code for line
8885 number @var{line} of the current source file to the stdio stream
8886 @var{stream}. The default is to emit an @code{.ln} directive.
8891 @subsection Macros for VMS Debug Format
8893 @c prevent bad page break with this line
8894 Here are macros for VMS debug format.
8896 @defmac VMS_DEBUGGING_INFO
8897 Define this macro if GCC should produce debugging output for VMS
8898 in response to the @option{-g} option. The default behavior for VMS
8899 is to generate minimal debug info for a traceback in the absence of
8900 @option{-g} unless explicitly overridden with @option{-g0}. This
8901 behavior is controlled by @code{OPTIMIZATION_OPTIONS} and
8902 @code{OVERRIDE_OPTIONS}.
8905 @node Floating Point
8906 @section Cross Compilation and Floating Point
8907 @cindex cross compilation and floating point
8908 @cindex floating point and cross compilation
8910 While all modern machines use twos-complement representation for integers,
8911 there are a variety of representations for floating point numbers. This
8912 means that in a cross-compiler the representation of floating point numbers
8913 in the compiled program may be different from that used in the machine
8914 doing the compilation.
8916 Because different representation systems may offer different amounts of
8917 range and precision, all floating point constants must be represented in
8918 the target machine's format. Therefore, the cross compiler cannot
8919 safely use the host machine's floating point arithmetic; it must emulate
8920 the target's arithmetic. To ensure consistency, GCC always uses
8921 emulation to work with floating point values, even when the host and
8922 target floating point formats are identical.
8924 The following macros are provided by @file{real.h} for the compiler to
8925 use. All parts of the compiler which generate or optimize
8926 floating-point calculations must use these macros. They may evaluate
8927 their operands more than once, so operands must not have side effects.
8929 @defmac REAL_VALUE_TYPE
8930 The C data type to be used to hold a floating point value in the target
8931 machine's format. Typically this is a @code{struct} containing an
8932 array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque
8936 @deftypefn Macro int REAL_VALUES_EQUAL (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
8937 Compares for equality the two values, @var{x} and @var{y}. If the target
8938 floating point format supports negative zeroes and/or NaNs,
8939 @samp{REAL_VALUES_EQUAL (-0.0, 0.0)} is true, and
8940 @samp{REAL_VALUES_EQUAL (NaN, NaN)} is false.
8943 @deftypefn Macro int REAL_VALUES_LESS (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
8944 Tests whether @var{x} is less than @var{y}.
8947 @deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x})
8948 Truncates @var{x} to a signed integer, rounding toward zero.
8951 @deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x})
8952 Truncates @var{x} to an unsigned integer, rounding toward zero. If
8953 @var{x} is negative, returns zero.
8956 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, enum machine_mode @var{mode})
8957 Converts @var{string} into a floating point number in the target machine's
8958 representation for mode @var{mode}. This routine can handle both
8959 decimal and hexadecimal floating point constants, using the syntax
8960 defined by the C language for both.
8963 @deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x})
8964 Returns 1 if @var{x} is negative (including negative zero), 0 otherwise.
8967 @deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x})
8968 Determines whether @var{x} represents infinity (positive or negative).
8971 @deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x})
8972 Determines whether @var{x} represents a ``NaN'' (not-a-number).
8975 @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})
8976 Calculates an arithmetic operation on the two floating point values
8977 @var{x} and @var{y}, storing the result in @var{output} (which must be a
8980 The operation to be performed is specified by @var{code}. Only the
8981 following codes are supported: @code{PLUS_EXPR}, @code{MINUS_EXPR},
8982 @code{MULT_EXPR}, @code{RDIV_EXPR}, @code{MAX_EXPR}, @code{MIN_EXPR}.
8984 If @code{REAL_ARITHMETIC} is asked to evaluate division by zero and the
8985 target's floating point format cannot represent infinity, it will call
8986 @code{abort}. Callers should check for this situation first, using
8987 @code{MODE_HAS_INFINITIES}. @xref{Storage Layout}.
8990 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x})
8991 Returns the negative of the floating point value @var{x}.
8994 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x})
8995 Returns the absolute value of @var{x}.
8998 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_TRUNCATE (REAL_VALUE_TYPE @var{mode}, enum machine_mode @var{x})
8999 Truncates the floating point value @var{x} to fit in @var{mode}. The
9000 return value is still a full-size @code{REAL_VALUE_TYPE}, but it has an
9001 appropriate bit pattern to be output as a floating constant whose
9002 precision accords with mode @var{mode}.
9005 @deftypefn Macro void REAL_VALUE_TO_INT (HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, REAL_VALUE_TYPE @var{x})
9006 Converts a floating point value @var{x} into a double-precision integer
9007 which is then stored into @var{low} and @var{high}. If the value is not
9008 integral, it is truncated.
9011 @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})
9012 Converts a double-precision integer found in @var{low} and @var{high},
9013 into a floating point value which is then stored into @var{x}. The
9014 value is truncated to fit in mode @var{mode}.
9017 @node Mode Switching
9018 @section Mode Switching Instructions
9019 @cindex mode switching
9020 The following macros control mode switching optimizations:
9022 @defmac OPTIMIZE_MODE_SWITCHING (@var{entity})
9023 Define this macro if the port needs extra instructions inserted for mode
9024 switching in an optimizing compilation.
9026 For an example, the SH4 can perform both single and double precision
9027 floating point operations, but to perform a single precision operation,
9028 the FPSCR PR bit has to be cleared, while for a double precision
9029 operation, this bit has to be set. Changing the PR bit requires a general
9030 purpose register as a scratch register, hence these FPSCR sets have to
9031 be inserted before reload, i.e.@: you can't put this into instruction emitting
9032 or @code{TARGET_MACHINE_DEPENDENT_REORG}.
9034 You can have multiple entities that are mode-switched, and select at run time
9035 which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should
9036 return nonzero for any @var{entity} that needs mode-switching.
9037 If you define this macro, you also have to define
9038 @code{NUM_MODES_FOR_MODE_SWITCHING}, @code{MODE_NEEDED},
9039 @code{MODE_PRIORITY_TO_MODE} and @code{EMIT_MODE_SET}.
9040 @code{MODE_AFTER}, @code{MODE_ENTRY}, and @code{MODE_EXIT}
9044 @defmac NUM_MODES_FOR_MODE_SWITCHING
9045 If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as
9046 initializer for an array of integers. Each initializer element
9047 N refers to an entity that needs mode switching, and specifies the number
9048 of different modes that might need to be set for this entity.
9049 The position of the initializer in the initializer---starting counting at
9050 zero---determines the integer that is used to refer to the mode-switched
9052 In macros that take mode arguments / yield a mode result, modes are
9053 represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode
9054 switch is needed / supplied.
9057 @defmac MODE_NEEDED (@var{entity}, @var{insn})
9058 @var{entity} is an integer specifying a mode-switched entity. If
9059 @code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to
9060 return an integer value not larger than the corresponding element in
9061 @code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must
9062 be switched into prior to the execution of @var{insn}.
9065 @defmac MODE_AFTER (@var{mode}, @var{insn})
9066 If this macro is defined, it is evaluated for every @var{insn} during
9067 mode switching. It determines the mode that an insn results in (if
9068 different from the incoming mode).
9071 @defmac MODE_ENTRY (@var{entity})
9072 If this macro is defined, it is evaluated for every @var{entity} that needs
9073 mode switching. It should evaluate to an integer, which is a mode that
9074 @var{entity} is assumed to be switched to at function entry. If @code{MODE_ENTRY}
9075 is defined then @code{MODE_EXIT} must be defined.
9078 @defmac MODE_EXIT (@var{entity})
9079 If this macro is defined, it is evaluated for every @var{entity} that needs
9080 mode switching. It should evaluate to an integer, which is a mode that
9081 @var{entity} is assumed to be switched to at function exit. If @code{MODE_EXIT}
9082 is defined then @code{MODE_ENTRY} must be defined.
9085 @defmac MODE_PRIORITY_TO_MODE (@var{entity}, @var{n})
9086 This macro specifies the order in which modes for @var{entity} are processed.
9087 0 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the
9088 lowest. The value of the macro should be an integer designating a mode
9089 for @var{entity}. For any fixed @var{entity}, @code{mode_priority_to_mode}
9090 (@var{entity}, @var{n}) shall be a bijection in 0 @dots{}
9091 @code{num_modes_for_mode_switching[@var{entity}] - 1}.
9094 @defmac EMIT_MODE_SET (@var{entity}, @var{mode}, @var{hard_regs_live})
9095 Generate one or more insns to set @var{entity} to @var{mode}.
9096 @var{hard_reg_live} is the set of hard registers live at the point where
9097 the insn(s) are to be inserted.
9100 @node Target Attributes
9101 @section Defining target-specific uses of @code{__attribute__}
9102 @cindex target attributes
9103 @cindex machine attributes
9104 @cindex attributes, target-specific
9106 Target-specific attributes may be defined for functions, data and types.
9107 These are described using the following target hooks; they also need to
9108 be documented in @file{extend.texi}.
9110 @deftypevr {Target Hook} {const struct attribute_spec *} TARGET_ATTRIBUTE_TABLE
9111 If defined, this target hook points to an array of @samp{struct
9112 attribute_spec} (defined in @file{tree.h}) specifying the machine
9113 specific attributes for this target and some of the restrictions on the
9114 entities to which these attributes are applied and the arguments they
9118 @deftypefn {Target Hook} int TARGET_COMP_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2})
9119 If defined, this target hook is a function which returns zero if the attributes on
9120 @var{type1} and @var{type2} are incompatible, one if they are compatible,
9121 and two if they are nearly compatible (which causes a warning to be
9122 generated). If this is not defined, machine-specific attributes are
9123 supposed always to be compatible.
9126 @deftypefn {Target Hook} void TARGET_SET_DEFAULT_TYPE_ATTRIBUTES (tree @var{type})
9127 If defined, this target hook is a function which assigns default attributes to
9128 newly defined @var{type}.
9131 @deftypefn {Target Hook} tree TARGET_MERGE_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2})
9132 Define this target hook if the merging of type attributes needs special
9133 handling. If defined, the result is a list of the combined
9134 @code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed
9135 that @code{comptypes} has already been called and returned 1. This
9136 function may call @code{merge_attributes} to handle machine-independent
9140 @deftypefn {Target Hook} tree TARGET_MERGE_DECL_ATTRIBUTES (tree @var{olddecl}, tree @var{newdecl})
9141 Define this target hook if the merging of decl attributes needs special
9142 handling. If defined, the result is a list of the combined
9143 @code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}.
9144 @var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of
9145 when this is needed are when one attribute overrides another, or when an
9146 attribute is nullified by a subsequent definition. This function may
9147 call @code{merge_attributes} to handle machine-independent merging.
9149 @findex TARGET_DLLIMPORT_DECL_ATTRIBUTES
9150 If the only target-specific handling you require is @samp{dllimport}
9151 for Microsoft Windows targets, you should define the macro
9152 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler
9153 will then define a function called
9154 @code{merge_dllimport_decl_attributes} which can then be defined as
9155 the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also
9156 add @code{handle_dll_attribute} in the attribute table for your port
9157 to perform initial processing of the @samp{dllimport} and
9158 @samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and
9159 @file{i386/i386.c}, for example.
9162 @deftypefn {Target Hook} bool TARGET_VALID_DLLIMPORT_ATTRIBUTE_P (tree @var{decl})
9163 @var{decl} is a variable or function with @code{__attribute__((dllimport))}
9164 specified. Use this hook if the target needs to add extra validation
9165 checks to @code{handle_dll_attribute}.
9168 @defmac TARGET_DECLSPEC
9169 Define this macro to a nonzero value if you want to treat
9170 @code{__declspec(X)} as equivalent to @code{__attribute((X))}. By
9171 default, this behavior is enabled only for targets that define
9172 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation
9173 of @code{__declspec} is via a built-in macro, but you should not rely
9174 on this implementation detail.
9177 @deftypefn {Target Hook} void TARGET_INSERT_ATTRIBUTES (tree @var{node}, tree *@var{attr_ptr})
9178 Define this target hook if you want to be able to add attributes to a decl
9179 when it is being created. This is normally useful for back ends which
9180 wish to implement a pragma by using the attributes which correspond to
9181 the pragma's effect. The @var{node} argument is the decl which is being
9182 created. The @var{attr_ptr} argument is a pointer to the attribute list
9183 for this decl. The list itself should not be modified, since it may be
9184 shared with other decls, but attributes may be chained on the head of
9185 the list and @code{*@var{attr_ptr}} modified to point to the new
9186 attributes, or a copy of the list may be made if further changes are
9190 @deftypefn {Target Hook} bool TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P (tree @var{fndecl})
9192 This target hook returns @code{true} if it is ok to inline @var{fndecl}
9193 into the current function, despite its having target-specific
9194 attributes, @code{false} otherwise. By default, if a function has a
9195 target specific attribute attached to it, it will not be inlined.
9199 @section Emulating TLS
9200 @cindex Emulated TLS
9202 For targets whose psABI does not provide Thread Local Storage via
9203 specific relocations and instruction sequences, an emulation layer is
9204 used. A set of target hooks allows this emulation layer to be
9205 configured for the requirements of a particular target. For instance
9206 the psABI may infact specify TLS support in terms of an emulation
9209 The emulation layer works by creating a control object for every TLS
9210 object. To access the TLS object, a lookup function is provided
9211 which, when given the address of the control object, will return the
9212 address of the current thread's instance of the TLS object.
9214 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_GET_ADDRESS
9215 Contains the name of the helper function that uses a TLS control
9216 object to locate a TLS instance. The default causes libgcc's
9217 emulated TLS helper function to be used.
9220 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_REGISTER_COMMON
9221 Contains the name of the helper function that should be used at
9222 program startup to register TLS objects that are implicitly
9223 initialized to zero. If this is @code{NULL}, all TLS objects will
9224 have explicit initializers. The default causes libgcc's emulated TLS
9225 registration function to be used.
9228 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_VAR_SECTION
9229 Contains the name of the section in which TLS control variables should
9230 be placed. The default of @code{NULL} allows these to be placed in
9234 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_TMPL_SECTION
9235 Contains the name of the section in which TLS initializers should be
9236 placed. The default of @code{NULL} allows these to be placed in any
9240 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_VAR_PREFIX
9241 Contains the prefix to be prepended to TLS control variable names.
9242 The default of @code{NULL} uses a target-specific prefix.
9245 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_TMPL_PREFIX
9246 Contains the prefix to be prepended to TLS initializer objects. The
9247 default of @code{NULL} uses a target-specific prefix.
9250 @deftypefn {Target Hook} tree TARGET_EMUTLS_VAR_FIELDS (tree @var{type}, tree *@var{name})
9251 Specifies a function that generates the FIELD_DECLs for a TLS control
9252 object type. @var{type} is the RECORD_TYPE the fields are for and
9253 @var{name} should be filled with the structure tag, if the default of
9254 @code{__emutls_object} is unsuitable. The default creates a type suitable
9255 for libgcc's emulated TLS function.
9258 @deftypefn {Target Hook} tree TARGET_EMUTLS_VAR_INIT (tree @var{var}, tree @var{decl}, tree @var{tmpl_addr})
9259 Specifies a function that generates the CONSTRUCTOR to initialize a
9260 TLS control object. @var{var} is the TLS control object, @var{decl}
9261 is the TLS object and @var{tmpl_addr} is the address of the
9262 initializer. The default initializes libgcc's emulated TLS control object.
9265 @deftypevr {Target Hook} {bool} TARGET_EMUTLS_VAR_ALIGN_FIXED
9266 Specifies whether the alignment of TLS control variable objects is
9267 fixed and should not be increased as some backends may do to optimize
9268 single objects. The default is false.
9271 @deftypevr {Target Hook} {bool} TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS
9272 Specifies whether a DWARF @code{DW_OP_form_tls_address} location descriptor
9273 may be used to describe emulated TLS control objects.
9276 @node MIPS Coprocessors
9277 @section Defining coprocessor specifics for MIPS targets.
9278 @cindex MIPS coprocessor-definition macros
9280 The MIPS specification allows MIPS implementations to have as many as 4
9281 coprocessors, each with as many as 32 private registers. GCC supports
9282 accessing these registers and transferring values between the registers
9283 and memory using asm-ized variables. For example:
9286 register unsigned int cp0count asm ("c0r1");
9292 (``c0r1'' is the default name of register 1 in coprocessor 0; alternate
9293 names may be added as described below, or the default names may be
9294 overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.)
9296 Coprocessor registers are assumed to be epilogue-used; sets to them will
9297 be preserved even if it does not appear that the register is used again
9298 later in the function.
9300 Another note: according to the MIPS spec, coprocessor 1 (if present) is
9301 the FPU@. One accesses COP1 registers through standard mips
9302 floating-point support; they are not included in this mechanism.
9304 There is one macro used in defining the MIPS coprocessor interface which
9305 you may want to override in subtargets; it is described below.
9307 @defmac ALL_COP_ADDITIONAL_REGISTER_NAMES
9308 A comma-separated list (with leading comma) of pairs describing the
9309 alternate names of coprocessor registers. The format of each entry should be
9311 @{ @var{alternatename}, @var{register_number}@}
9317 @section Parameters for Precompiled Header Validity Checking
9318 @cindex parameters, precompiled headers
9320 @deftypefn {Target Hook} void *TARGET_GET_PCH_VALIDITY (size_t *@var{sz})
9321 This hook returns the data needed by @code{TARGET_PCH_VALID_P} and sets
9322 @samp{*@var{sz}} to the size of the data in bytes.
9325 @deftypefn {Target Hook} const char *TARGET_PCH_VALID_P (const void *@var{data}, size_t @var{sz})
9326 This hook checks whether the options used to create a PCH file are
9327 compatible with the current settings. It returns @code{NULL}
9328 if so and a suitable error message if not. Error messages will
9329 be presented to the user and must be localized using @samp{_(@var{msg})}.
9331 @var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY}
9332 when the PCH file was created and @var{sz} is the size of that data in bytes.
9333 It's safe to assume that the data was created by the same version of the
9334 compiler, so no format checking is needed.
9336 The default definition of @code{default_pch_valid_p} should be
9337 suitable for most targets.
9340 @deftypefn {Target Hook} const char *TARGET_CHECK_PCH_TARGET_FLAGS (int @var{pch_flags})
9341 If this hook is nonnull, the default implementation of
9342 @code{TARGET_PCH_VALID_P} will use it to check for compatible values
9343 of @code{target_flags}. @var{pch_flags} specifies the value that
9344 @code{target_flags} had when the PCH file was created. The return
9345 value is the same as for @code{TARGET_PCH_VALID_P}.
9349 @section C++ ABI parameters
9350 @cindex parameters, c++ abi
9352 @deftypefn {Target Hook} tree TARGET_CXX_GUARD_TYPE (void)
9353 Define this hook to override the integer type used for guard variables.
9354 These are used to implement one-time construction of static objects. The
9355 default is long_long_integer_type_node.
9358 @deftypefn {Target Hook} bool TARGET_CXX_GUARD_MASK_BIT (void)
9359 This hook determines how guard variables are used. It should return
9360 @code{false} (the default) if first byte should be used. A return value of
9361 @code{true} indicates the least significant bit should be used.
9364 @deftypefn {Target Hook} tree TARGET_CXX_GET_COOKIE_SIZE (tree @var{type})
9365 This hook returns the size of the cookie to use when allocating an array
9366 whose elements have the indicated @var{type}. Assumes that it is already
9367 known that a cookie is needed. The default is
9368 @code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the
9369 IA64/Generic C++ ABI@.
9372 @deftypefn {Target Hook} bool TARGET_CXX_COOKIE_HAS_SIZE (void)
9373 This hook should return @code{true} if the element size should be stored in
9374 array cookies. The default is to return @code{false}.
9377 @deftypefn {Target Hook} int TARGET_CXX_IMPORT_EXPORT_CLASS (tree @var{type}, int @var{import_export})
9378 If defined by a backend this hook allows the decision made to export
9379 class @var{type} to be overruled. Upon entry @var{import_export}
9380 will contain 1 if the class is going to be exported, @minus{}1 if it is going
9381 to be imported and 0 otherwise. This function should return the
9382 modified value and perform any other actions necessary to support the
9383 backend's targeted operating system.
9386 @deftypefn {Target Hook} bool TARGET_CXX_CDTOR_RETURNS_THIS (void)
9387 This hook should return @code{true} if constructors and destructors return
9388 the address of the object created/destroyed. The default is to return
9392 @deftypefn {Target Hook} bool TARGET_CXX_KEY_METHOD_MAY_BE_INLINE (void)
9393 This hook returns true if the key method for a class (i.e., the method
9394 which, if defined in the current translation unit, causes the virtual
9395 table to be emitted) may be an inline function. Under the standard
9396 Itanium C++ ABI the key method may be an inline function so long as
9397 the function is not declared inline in the class definition. Under
9398 some variants of the ABI, an inline function can never be the key
9399 method. The default is to return @code{true}.
9402 @deftypefn {Target Hook} void TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY (tree @var{decl})
9403 @var{decl} is a virtual table, virtual table table, typeinfo object,
9404 or other similar implicit class data object that will be emitted with
9405 external linkage in this translation unit. No ELF visibility has been
9406 explicitly specified. If the target needs to specify a visibility
9407 other than that of the containing class, use this hook to set
9408 @code{DECL_VISIBILITY} and @code{DECL_VISIBILITY_SPECIFIED}.
9411 @deftypefn {Target Hook} bool TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT (void)
9412 This hook returns true (the default) if virtual tables and other
9413 similar implicit class data objects are always COMDAT if they have
9414 external linkage. If this hook returns false, then class data for
9415 classes whose virtual table will be emitted in only one translation
9416 unit will not be COMDAT.
9419 @deftypefn {Target Hook} bool TARGET_CXX_LIBRARY_RTTI_COMDAT (void)
9420 This hook returns true (the default) if the RTTI information for
9421 the basic types which is defined in the C++ runtime should always
9422 be COMDAT, false if it should not be COMDAT.
9425 @deftypefn {Target Hook} bool TARGET_CXX_USE_AEABI_ATEXIT (void)
9426 This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI)
9427 should be used to register static destructors when @option{-fuse-cxa-atexit}
9428 is in effect. The default is to return false to use @code{__cxa_atexit}.
9431 @deftypefn {Target Hook} bool TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT (void)
9432 This hook returns true if the target @code{atexit} function can be used
9433 in the same manner as @code{__cxa_atexit} to register C++ static
9434 destructors. This requires that @code{atexit}-registered functions in
9435 shared libraries are run in the correct order when the libraries are
9436 unloaded. The default is to return false.
9439 @deftypefn {Target Hook} void TARGET_CXX_ADJUST_CLASS_AT_DEFINITION (tree @var{type})
9440 @var{type} is a C++ class (i.e., RECORD_TYPE or UNION_TYPE) that has just been
9441 defined. Use this hook to make adjustments to the class (eg, tweak
9442 visibility or perform any other required target modifications).
9446 @section Miscellaneous Parameters
9447 @cindex parameters, miscellaneous
9449 @c prevent bad page break with this line
9450 Here are several miscellaneous parameters.
9452 @defmac HAS_LONG_COND_BRANCH
9453 Define this boolean macro to indicate whether or not your architecture
9454 has conditional branches that can span all of memory. It is used in
9455 conjunction with an optimization that partitions hot and cold basic
9456 blocks into separate sections of the executable. If this macro is
9457 set to false, gcc will convert any conditional branches that attempt
9458 to cross between sections into unconditional branches or indirect jumps.
9461 @defmac HAS_LONG_UNCOND_BRANCH
9462 Define this boolean macro to indicate whether or not your architecture
9463 has unconditional branches that can span all of memory. It is used in
9464 conjunction with an optimization that partitions hot and cold basic
9465 blocks into separate sections of the executable. If this macro is
9466 set to false, gcc will convert any unconditional branches that attempt
9467 to cross between sections into indirect jumps.
9470 @defmac CASE_VECTOR_MODE
9471 An alias for a machine mode name. This is the machine mode that
9472 elements of a jump-table should have.
9475 @defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body})
9476 Optional: return the preferred mode for an @code{addr_diff_vec}
9477 when the minimum and maximum offset are known. If you define this,
9478 it enables extra code in branch shortening to deal with @code{addr_diff_vec}.
9479 To make this work, you also have to define @code{INSN_ALIGN} and
9480 make the alignment for @code{addr_diff_vec} explicit.
9481 The @var{body} argument is provided so that the offset_unsigned and scale
9482 flags can be updated.
9485 @defmac CASE_VECTOR_PC_RELATIVE
9486 Define this macro to be a C expression to indicate when jump-tables
9487 should contain relative addresses. You need not define this macro if
9488 jump-tables never contain relative addresses, or jump-tables should
9489 contain relative addresses only when @option{-fPIC} or @option{-fPIC}
9493 @defmac CASE_VALUES_THRESHOLD
9494 Define this to be the smallest number of different values for which it
9495 is best to use a jump-table instead of a tree of conditional branches.
9496 The default is four for machines with a @code{casesi} instruction and
9497 five otherwise. This is best for most machines.
9500 @defmac CASE_USE_BIT_TESTS
9501 Define this macro to be a C expression to indicate whether C switch
9502 statements may be implemented by a sequence of bit tests. This is
9503 advantageous on processors that can efficiently implement left shift
9504 of 1 by the number of bits held in a register, but inappropriate on
9505 targets that would require a loop. By default, this macro returns
9506 @code{true} if the target defines an @code{ashlsi3} pattern, and
9507 @code{false} otherwise.
9510 @defmac WORD_REGISTER_OPERATIONS
9511 Define this macro if operations between registers with integral mode
9512 smaller than a word are always performed on the entire register.
9513 Most RISC machines have this property and most CISC machines do not.
9516 @defmac LOAD_EXTEND_OP (@var{mem_mode})
9517 Define this macro to be a C expression indicating when insns that read
9518 memory in @var{mem_mode}, an integral mode narrower than a word, set the
9519 bits outside of @var{mem_mode} to be either the sign-extension or the
9520 zero-extension of the data read. Return @code{SIGN_EXTEND} for values
9521 of @var{mem_mode} for which the
9522 insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and
9523 @code{UNKNOWN} for other modes.
9525 This macro is not called with @var{mem_mode} non-integral or with a width
9526 greater than or equal to @code{BITS_PER_WORD}, so you may return any
9527 value in this case. Do not define this macro if it would always return
9528 @code{UNKNOWN}. On machines where this macro is defined, you will normally
9529 define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}.
9531 You may return a non-@code{UNKNOWN} value even if for some hard registers
9532 the sign extension is not performed, if for the @code{REGNO_REG_CLASS}
9533 of these hard registers @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero
9534 when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any
9535 integral mode larger than this but not larger than @code{word_mode}.
9537 You must return @code{UNKNOWN} if for some hard registers that allow this
9538 mode, @code{CANNOT_CHANGE_MODE_CLASS} says that they cannot change to
9539 @code{word_mode}, but that they can change to another integral mode that
9540 is larger then @var{mem_mode} but still smaller than @code{word_mode}.
9543 @defmac SHORT_IMMEDIATES_SIGN_EXTEND
9544 Define this macro if loading short immediate values into registers sign
9548 @defmac FIXUNS_TRUNC_LIKE_FIX_TRUNC
9549 Define this macro if the same instructions that convert a floating
9550 point number to a signed fixed point number also convert validly to an
9554 @deftypefn {Target Hook} int TARGET_MIN_DIVISIONS_FOR_RECIP_MUL (enum machine_mode @var{mode})
9555 When @option{-ffast-math} is in effect, GCC tries to optimize
9556 divisions by the same divisor, by turning them into multiplications by
9557 the reciprocal. This target hook specifies the minimum number of divisions
9558 that should be there for GCC to perform the optimization for a variable
9559 of mode @var{mode}. The default implementation returns 3 if the machine
9560 has an instruction for the division, and 2 if it does not.
9564 The maximum number of bytes that a single instruction can move quickly
9565 between memory and registers or between two memory locations.
9568 @defmac MAX_MOVE_MAX
9569 The maximum number of bytes that a single instruction can move quickly
9570 between memory and registers or between two memory locations. If this
9571 is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the
9572 constant value that is the largest value that @code{MOVE_MAX} can have
9576 @defmac SHIFT_COUNT_TRUNCATED
9577 A C expression that is nonzero if on this machine the number of bits
9578 actually used for the count of a shift operation is equal to the number
9579 of bits needed to represent the size of the object being shifted. When
9580 this macro is nonzero, the compiler will assume that it is safe to omit
9581 a sign-extend, zero-extend, and certain bitwise `and' instructions that
9582 truncates the count of a shift operation. On machines that have
9583 instructions that act on bit-fields at variable positions, which may
9584 include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED}
9585 also enables deletion of truncations of the values that serve as
9586 arguments to bit-field instructions.
9588 If both types of instructions truncate the count (for shifts) and
9589 position (for bit-field operations), or if no variable-position bit-field
9590 instructions exist, you should define this macro.
9592 However, on some machines, such as the 80386 and the 680x0, truncation
9593 only applies to shift operations and not the (real or pretended)
9594 bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on
9595 such machines. Instead, add patterns to the @file{md} file that include
9596 the implied truncation of the shift instructions.
9598 You need not define this macro if it would always have the value of zero.
9601 @anchor{TARGET_SHIFT_TRUNCATION_MASK}
9602 @deftypefn {Target Hook} int TARGET_SHIFT_TRUNCATION_MASK (enum machine_mode @var{mode})
9603 This function describes how the standard shift patterns for @var{mode}
9604 deal with shifts by negative amounts or by more than the width of the mode.
9605 @xref{shift patterns}.
9607 On many machines, the shift patterns will apply a mask @var{m} to the
9608 shift count, meaning that a fixed-width shift of @var{x} by @var{y} is
9609 equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If
9610 this is true for mode @var{mode}, the function should return @var{m},
9611 otherwise it should return 0. A return value of 0 indicates that no
9612 particular behavior is guaranteed.
9614 Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does
9615 @emph{not} apply to general shift rtxes; it applies only to instructions
9616 that are generated by the named shift patterns.
9618 The default implementation of this function returns
9619 @code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED}
9620 and 0 otherwise. This definition is always safe, but if
9621 @code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns
9622 nevertheless truncate the shift count, you may get better code
9626 @defmac TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
9627 A C expression which is nonzero if on this machine it is safe to
9628 ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
9629 bits (where @var{outprec} is smaller than @var{inprec}) by merely
9630 operating on it as if it had only @var{outprec} bits.
9632 On many machines, this expression can be 1.
9634 @c rearranged this, removed the phrase "it is reported that". this was
9635 @c to fix an overfull hbox. --mew 10feb93
9636 When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for
9637 modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result.
9638 If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in
9639 such cases may improve things.
9642 @deftypefn {Target Hook} int TARGET_MODE_REP_EXTENDED (enum machine_mode @var{mode}, enum machine_mode @var{rep_mode})
9643 The representation of an integral mode can be such that the values
9644 are always extended to a wider integral mode. Return
9645 @code{SIGN_EXTEND} if values of @var{mode} are represented in
9646 sign-extended form to @var{rep_mode}. Return @code{UNKNOWN}
9647 otherwise. (Currently, none of the targets use zero-extended
9648 representation this way so unlike @code{LOAD_EXTEND_OP},
9649 @code{TARGET_MODE_REP_EXTENDED} is expected to return either
9650 @code{SIGN_EXTEND} or @code{UNKNOWN}. Also no target extends
9651 @var{mode} to @var{mode_rep} so that @var{mode_rep} is not the next
9652 widest integral mode and currently we take advantage of this fact.)
9654 Similarly to @code{LOAD_EXTEND_OP} you may return a non-@code{UNKNOWN}
9655 value even if the extension is not performed on certain hard registers
9656 as long as for the @code{REGNO_REG_CLASS} of these hard registers
9657 @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero.
9659 Note that @code{TARGET_MODE_REP_EXTENDED} and @code{LOAD_EXTEND_OP}
9660 describe two related properties. If you define
9661 @code{TARGET_MODE_REP_EXTENDED (mode, word_mode)} you probably also want
9662 to define @code{LOAD_EXTEND_OP (mode)} to return the same type of
9665 In order to enforce the representation of @code{mode},
9666 @code{TRULY_NOOP_TRUNCATION} should return false when truncating to
9670 @defmac STORE_FLAG_VALUE
9671 A C expression describing the value returned by a comparison operator
9672 with an integral mode and stored by a store-flag instruction
9673 (@samp{s@var{cond}}) when the condition is true. This description must
9674 apply to @emph{all} the @samp{s@var{cond}} patterns and all the
9675 comparison operators whose results have a @code{MODE_INT} mode.
9677 A value of 1 or @minus{}1 means that the instruction implementing the
9678 comparison operator returns exactly 1 or @minus{}1 when the comparison is true
9679 and 0 when the comparison is false. Otherwise, the value indicates
9680 which bits of the result are guaranteed to be 1 when the comparison is
9681 true. This value is interpreted in the mode of the comparison
9682 operation, which is given by the mode of the first operand in the
9683 @samp{s@var{cond}} pattern. Either the low bit or the sign bit of
9684 @code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by
9687 If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will
9688 generate code that depends only on the specified bits. It can also
9689 replace comparison operators with equivalent operations if they cause
9690 the required bits to be set, even if the remaining bits are undefined.
9691 For example, on a machine whose comparison operators return an
9692 @code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as
9693 @samp{0x80000000}, saying that just the sign bit is relevant, the
9697 (ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0))
9704 (ashift:SI @var{x} (const_int @var{n}))
9708 where @var{n} is the appropriate shift count to move the bit being
9709 tested into the sign bit.
9711 There is no way to describe a machine that always sets the low-order bit
9712 for a true value, but does not guarantee the value of any other bits,
9713 but we do not know of any machine that has such an instruction. If you
9714 are trying to port GCC to such a machine, include an instruction to
9715 perform a logical-and of the result with 1 in the pattern for the
9716 comparison operators and let us know at @email{gcc@@gcc.gnu.org}.
9718 Often, a machine will have multiple instructions that obtain a value
9719 from a comparison (or the condition codes). Here are rules to guide the
9720 choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions
9725 Use the shortest sequence that yields a valid definition for
9726 @code{STORE_FLAG_VALUE}. It is more efficient for the compiler to
9727 ``normalize'' the value (convert it to, e.g., 1 or 0) than for the
9728 comparison operators to do so because there may be opportunities to
9729 combine the normalization with other operations.
9732 For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being
9733 slightly preferred on machines with expensive jumps and 1 preferred on
9737 As a second choice, choose a value of @samp{0x80000001} if instructions
9738 exist that set both the sign and low-order bits but do not define the
9742 Otherwise, use a value of @samp{0x80000000}.
9745 Many machines can produce both the value chosen for
9746 @code{STORE_FLAG_VALUE} and its negation in the same number of
9747 instructions. On those machines, you should also define a pattern for
9748 those cases, e.g., one matching
9751 (set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C})))
9754 Some machines can also perform @code{and} or @code{plus} operations on
9755 condition code values with less instructions than the corresponding
9756 @samp{s@var{cond}} insn followed by @code{and} or @code{plus}. On those
9757 machines, define the appropriate patterns. Use the names @code{incscc}
9758 and @code{decscc}, respectively, for the patterns which perform
9759 @code{plus} or @code{minus} operations on condition code values. See
9760 @file{rs6000.md} for some examples. The GNU Superoptizer can be used to
9761 find such instruction sequences on other machines.
9763 If this macro is not defined, the default value, 1, is used. You need
9764 not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
9765 instructions, or if the value generated by these instructions is 1.
9768 @defmac FLOAT_STORE_FLAG_VALUE (@var{mode})
9769 A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is
9770 returned when comparison operators with floating-point results are true.
9771 Define this macro on machines that have comparison operations that return
9772 floating-point values. If there are no such operations, do not define
9776 @defmac VECTOR_STORE_FLAG_VALUE (@var{mode})
9777 A C expression that gives a rtx representing the nonzero true element
9778 for vector comparisons. The returned rtx should be valid for the inner
9779 mode of @var{mode} which is guaranteed to be a vector mode. Define
9780 this macro on machines that have vector comparison operations that
9781 return a vector result. If there are no such operations, do not define
9782 this macro. Typically, this macro is defined as @code{const1_rtx} or
9783 @code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent
9784 the compiler optimizing such vector comparison operations for the
9788 @defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
9789 @defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
9790 A C expression that indicates whether the architecture defines a value
9791 for @code{clz} or @code{ctz} with a zero operand.
9792 A result of @code{0} indicates the value is undefined.
9793 If the value is defined for only the RTL expression, the macro should
9794 evaluate to @code{1}; if the value applies also to the corresponding optab
9795 entry (which is normally the case if it expands directly into
9796 the corresponding RTL), then the macro should evaluate to @code{2}.
9797 In the cases where the value is defined, @var{value} should be set to
9800 If this macro is not defined, the value of @code{clz} or
9801 @code{ctz} at zero is assumed to be undefined.
9803 This macro must be defined if the target's expansion for @code{ffs}
9804 relies on a particular value to get correct results. Otherwise it
9805 is not necessary, though it may be used to optimize some corner cases, and
9806 to provide a default expansion for the @code{ffs} optab.
9808 Note that regardless of this macro the ``definedness'' of @code{clz}
9809 and @code{ctz} at zero do @emph{not} extend to the builtin functions
9810 visible to the user. Thus one may be free to adjust the value at will
9811 to match the target expansion of these operations without fear of
9816 An alias for the machine mode for pointers. On most machines, define
9817 this to be the integer mode corresponding to the width of a hardware
9818 pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines.
9819 On some machines you must define this to be one of the partial integer
9820 modes, such as @code{PSImode}.
9822 The width of @code{Pmode} must be at least as large as the value of
9823 @code{POINTER_SIZE}. If it is not equal, you must define the macro
9824 @code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended
9828 @defmac FUNCTION_MODE
9829 An alias for the machine mode used for memory references to functions
9830 being called, in @code{call} RTL expressions. On most CISC machines,
9831 where an instruction can begin at any byte address, this should be
9832 @code{QImode}. On most RISC machines, where all instructions have fixed
9833 size and alignment, this should be a mode with the same size and alignment
9834 as the machine instruction words - typically @code{SImode} or @code{HImode}.
9837 @defmac STDC_0_IN_SYSTEM_HEADERS
9838 In normal operation, the preprocessor expands @code{__STDC__} to the
9839 constant 1, to signify that GCC conforms to ISO Standard C@. On some
9840 hosts, like Solaris, the system compiler uses a different convention,
9841 where @code{__STDC__} is normally 0, but is 1 if the user specifies
9842 strict conformance to the C Standard.
9844 Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host
9845 convention when processing system header files, but when processing user
9846 files @code{__STDC__} will always expand to 1.
9849 @defmac NO_IMPLICIT_EXTERN_C
9850 Define this macro if the system header files support C++ as well as C@.
9851 This macro inhibits the usual method of using system header files in
9852 C++, which is to pretend that the file's contents are enclosed in
9853 @samp{extern "C" @{@dots{}@}}.
9858 @defmac REGISTER_TARGET_PRAGMAS ()
9859 Define this macro if you want to implement any target-specific pragmas.
9860 If defined, it is a C expression which makes a series of calls to
9861 @code{c_register_pragma} or @code{c_register_pragma_with_expansion}
9862 for each pragma. The macro may also do any
9863 setup required for the pragmas.
9865 The primary reason to define this macro is to provide compatibility with
9866 other compilers for the same target. In general, we discourage
9867 definition of target-specific pragmas for GCC@.
9869 If the pragma can be implemented by attributes then you should consider
9870 defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well.
9872 Preprocessor macros that appear on pragma lines are not expanded. All
9873 @samp{#pragma} directives that do not match any registered pragma are
9874 silently ignored, unless the user specifies @option{-Wunknown-pragmas}.
9877 @deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
9878 @deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
9880 Each call to @code{c_register_pragma} or
9881 @code{c_register_pragma_with_expansion} establishes one pragma. The
9882 @var{callback} routine will be called when the preprocessor encounters a
9886 #pragma [@var{space}] @var{name} @dots{}
9889 @var{space} is the case-sensitive namespace of the pragma, or
9890 @code{NULL} to put the pragma in the global namespace. The callback
9891 routine receives @var{pfile} as its first argument, which can be passed
9892 on to cpplib's functions if necessary. You can lex tokens after the
9893 @var{name} by calling @code{pragma_lex}. Tokens that are not read by the
9894 callback will be silently ignored. The end of the line is indicated by
9895 a token of type @code{CPP_EOF}. Macro expansion occurs on the
9896 arguments of pragmas registered with
9897 @code{c_register_pragma_with_expansion} but not on the arguments of
9898 pragmas registered with @code{c_register_pragma}.
9900 Note that the use of @code{pragma_lex} is specific to the C and C++
9901 compilers. It will not work in the Java or Fortran compilers, or any
9902 other language compilers for that matter. Thus if @code{pragma_lex} is going
9903 to be called from target-specific code, it must only be done so when
9904 building the C and C++ compilers. This can be done by defining the
9905 variables @code{c_target_objs} and @code{cxx_target_objs} in the
9906 target entry in the @file{config.gcc} file. These variables should name
9907 the target-specific, language-specific object file which contains the
9908 code that uses @code{pragma_lex}. Note it will also be necessary to add a
9909 rule to the makefile fragment pointed to by @code{tmake_file} that shows
9910 how to build this object file.
9915 @defmac HANDLE_SYSV_PRAGMA
9916 Define this macro (to a value of 1) if you want the System V style
9917 pragmas @samp{#pragma pack(<n>)} and @samp{#pragma weak <name>
9918 [=<value>]} to be supported by gcc.
9920 The pack pragma specifies the maximum alignment (in bytes) of fields
9921 within a structure, in much the same way as the @samp{__aligned__} and
9922 @samp{__packed__} @code{__attribute__}s do. A pack value of zero resets
9923 the behavior to the default.
9925 A subtlety for Microsoft Visual C/C++ style bit-field packing
9926 (e.g.@: -mms-bitfields) for targets that support it:
9927 When a bit-field is inserted into a packed record, the whole size
9928 of the underlying type is used by one or more same-size adjacent
9929 bit-fields (that is, if its long:3, 32 bits is used in the record,
9930 and any additional adjacent long bit-fields are packed into the same
9931 chunk of 32 bits. However, if the size changes, a new field of that
9934 If both MS bit-fields and @samp{__attribute__((packed))} are used,
9935 the latter will take precedence. If @samp{__attribute__((packed))} is
9936 used on a single field when MS bit-fields are in use, it will take
9937 precedence for that field, but the alignment of the rest of the structure
9938 may affect its placement.
9940 The weak pragma only works if @code{SUPPORTS_WEAK} and
9941 @code{ASM_WEAKEN_LABEL} are defined. If enabled it allows the creation
9942 of specifically named weak labels, optionally with a value.
9947 @defmac HANDLE_PRAGMA_PACK_PUSH_POP
9948 Define this macro (to a value of 1) if you want to support the Win32
9949 style pragmas @samp{#pragma pack(push[,@var{n}])} and @samp{#pragma
9950 pack(pop)}. The @samp{pack(push,[@var{n}])} pragma specifies the maximum
9951 alignment (in bytes) of fields within a structure, in much the same way as
9952 the @samp{__aligned__} and @samp{__packed__} @code{__attribute__}s do. A
9953 pack value of zero resets the behavior to the default. Successive
9954 invocations of this pragma cause the previous values to be stacked, so
9955 that invocations of @samp{#pragma pack(pop)} will return to the previous
9959 @defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION
9960 Define this macro, as well as
9961 @code{HANDLE_SYSV_PRAGMA}, if macros should be expanded in the
9962 arguments of @samp{#pragma pack}.
9965 @defmac TARGET_DEFAULT_PACK_STRUCT
9966 If your target requires a structure packing default other than 0 (meaning
9967 the machine default), define this macro to the necessary value (in bytes).
9968 This must be a value that would also be valid to use with
9969 @samp{#pragma pack()} (that is, a small power of two).
9974 @defmac HANDLE_PRAGMA_PUSH_POP_MACRO
9975 Define this macro if you want to support the Win32 style pragmas
9976 @samp{#pragma push_macro(macro-name-as-string)} and @samp{#pragma
9977 pop_macro(macro-name-as-string)}. The @samp{#pragma push_macro(
9978 macro-name-as-string)} pragma saves the named macro and via
9979 @samp{#pragma pop_macro(macro-name-as-string)} it will return to the
9984 @defmac DOLLARS_IN_IDENTIFIERS
9985 Define this macro to control use of the character @samp{$} in
9986 identifier names for the C family of languages. 0 means @samp{$} is
9987 not allowed by default; 1 means it is allowed. 1 is the default;
9988 there is no need to define this macro in that case.
9991 @defmac NO_DOLLAR_IN_LABEL
9992 Define this macro if the assembler does not accept the character
9993 @samp{$} in label names. By default constructors and destructors in
9994 G++ have @samp{$} in the identifiers. If this macro is defined,
9995 @samp{.} is used instead.
9998 @defmac NO_DOT_IN_LABEL
9999 Define this macro if the assembler does not accept the character
10000 @samp{.} in label names. By default constructors and destructors in G++
10001 have names that use @samp{.}. If this macro is defined, these names
10002 are rewritten to avoid @samp{.}.
10005 @defmac INSN_SETS_ARE_DELAYED (@var{insn})
10006 Define this macro as a C expression that is nonzero if it is safe for the
10007 delay slot scheduler to place instructions in the delay slot of @var{insn},
10008 even if they appear to use a resource set or clobbered in @var{insn}.
10009 @var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that
10010 every @code{call_insn} has this behavior. On machines where some @code{insn}
10011 or @code{jump_insn} is really a function call and hence has this behavior,
10012 you should define this macro.
10014 You need not define this macro if it would always return zero.
10017 @defmac INSN_REFERENCES_ARE_DELAYED (@var{insn})
10018 Define this macro as a C expression that is nonzero if it is safe for the
10019 delay slot scheduler to place instructions in the delay slot of @var{insn},
10020 even if they appear to set or clobber a resource referenced in @var{insn}.
10021 @var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where
10022 some @code{insn} or @code{jump_insn} is really a function call and its operands
10023 are registers whose use is actually in the subroutine it calls, you should
10024 define this macro. Doing so allows the delay slot scheduler to move
10025 instructions which copy arguments into the argument registers into the delay
10026 slot of @var{insn}.
10028 You need not define this macro if it would always return zero.
10031 @defmac MULTIPLE_SYMBOL_SPACES
10032 Define this macro as a C expression that is nonzero if, in some cases,
10033 global symbols from one translation unit may not be bound to undefined
10034 symbols in another translation unit without user intervention. For
10035 instance, under Microsoft Windows symbols must be explicitly imported
10036 from shared libraries (DLLs).
10038 You need not define this macro if it would always evaluate to zero.
10041 @deftypefn {Target Hook} tree TARGET_MD_ASM_CLOBBERS (tree @var{outputs}, tree @var{inputs}, tree @var{clobbers})
10042 This target hook should add to @var{clobbers} @code{STRING_CST} trees for
10043 any hard regs the port wishes to automatically clobber for an asm.
10044 It should return the result of the last @code{tree_cons} used to add a
10045 clobber. The @var{outputs}, @var{inputs} and @var{clobber} lists are the
10046 corresponding parameters to the asm and may be inspected to avoid
10047 clobbering a register that is an input or output of the asm. You can use
10048 @code{tree_overlaps_hard_reg_set}, declared in @file{tree.h}, to test
10049 for overlap with regards to asm-declared registers.
10052 @defmac MATH_LIBRARY
10053 Define this macro as a C string constant for the linker argument to link
10054 in the system math library, or @samp{""} if the target does not have a
10055 separate math library.
10057 You need only define this macro if the default of @samp{"-lm"} is wrong.
10060 @defmac LIBRARY_PATH_ENV
10061 Define this macro as a C string constant for the environment variable that
10062 specifies where the linker should look for libraries.
10064 You need only define this macro if the default of @samp{"LIBRARY_PATH"}
10068 @defmac TARGET_POSIX_IO
10069 Define this macro if the target supports the following POSIX@ file
10070 functions, access, mkdir and file locking with fcntl / F_SETLKW@.
10071 Defining @code{TARGET_POSIX_IO} will enable the test coverage code
10072 to use file locking when exiting a program, which avoids race conditions
10073 if the program has forked. It will also create directories at run-time
10074 for cross-profiling.
10077 @defmac MAX_CONDITIONAL_EXECUTE
10079 A C expression for the maximum number of instructions to execute via
10080 conditional execution instructions instead of a branch. A value of
10081 @code{BRANCH_COST}+1 is the default if the machine does not use cc0, and
10082 1 if it does use cc0.
10085 @defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr})
10086 Used if the target needs to perform machine-dependent modifications on the
10087 conditionals used for turning basic blocks into conditionally executed code.
10088 @var{ce_info} points to a data structure, @code{struct ce_if_block}, which
10089 contains information about the currently processed blocks. @var{true_expr}
10090 and @var{false_expr} are the tests that are used for converting the
10091 then-block and the else-block, respectively. Set either @var{true_expr} or
10092 @var{false_expr} to a null pointer if the tests cannot be converted.
10095 @defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr})
10096 Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated
10097 if-statements into conditions combined by @code{and} and @code{or} operations.
10098 @var{bb} contains the basic block that contains the test that is currently
10099 being processed and about to be turned into a condition.
10102 @defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn})
10103 A C expression to modify the @var{PATTERN} of an @var{INSN} that is to
10104 be converted to conditional execution format. @var{ce_info} points to
10105 a data structure, @code{struct ce_if_block}, which contains information
10106 about the currently processed blocks.
10109 @defmac IFCVT_MODIFY_FINAL (@var{ce_info})
10110 A C expression to perform any final machine dependent modifications in
10111 converting code to conditional execution. The involved basic blocks
10112 can be found in the @code{struct ce_if_block} structure that is pointed
10113 to by @var{ce_info}.
10116 @defmac IFCVT_MODIFY_CANCEL (@var{ce_info})
10117 A C expression to cancel any machine dependent modifications in
10118 converting code to conditional execution. The involved basic blocks
10119 can be found in the @code{struct ce_if_block} structure that is pointed
10120 to by @var{ce_info}.
10123 @defmac IFCVT_INIT_EXTRA_FIELDS (@var{ce_info})
10124 A C expression to initialize any extra fields in a @code{struct ce_if_block}
10125 structure, which are defined by the @code{IFCVT_EXTRA_FIELDS} macro.
10128 @defmac IFCVT_EXTRA_FIELDS
10129 If defined, it should expand to a set of field declarations that will be
10130 added to the @code{struct ce_if_block} structure. These should be initialized
10131 by the @code{IFCVT_INIT_EXTRA_FIELDS} macro.
10134 @deftypefn {Target Hook} void TARGET_MACHINE_DEPENDENT_REORG ()
10135 If non-null, this hook performs a target-specific pass over the
10136 instruction stream. The compiler will run it at all optimization levels,
10137 just before the point at which it normally does delayed-branch scheduling.
10139 The exact purpose of the hook varies from target to target. Some use
10140 it to do transformations that are necessary for correctness, such as
10141 laying out in-function constant pools or avoiding hardware hazards.
10142 Others use it as an opportunity to do some machine-dependent optimizations.
10144 You need not implement the hook if it has nothing to do. The default
10145 definition is null.
10148 @deftypefn {Target Hook} void TARGET_INIT_BUILTINS ()
10149 Define this hook if you have any machine-specific built-in functions
10150 that need to be defined. It should be a function that performs the
10153 Machine specific built-in functions can be useful to expand special machine
10154 instructions that would otherwise not normally be generated because
10155 they have no equivalent in the source language (for example, SIMD vector
10156 instructions or prefetch instructions).
10158 To create a built-in function, call the function
10159 @code{lang_hooks.builtin_function}
10160 which is defined by the language front end. You can use any type nodes set
10161 up by @code{build_common_tree_nodes} and @code{build_common_tree_nodes_2};
10162 only language front ends that use those two functions will call
10163 @samp{TARGET_INIT_BUILTINS}.
10166 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN (tree @var{exp}, rtx @var{target}, rtx @var{subtarget}, enum machine_mode @var{mode}, int @var{ignore})
10168 Expand a call to a machine specific built-in function that was set up by
10169 @samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the
10170 function call; the result should go to @var{target} if that is
10171 convenient, and have mode @var{mode} if that is convenient.
10172 @var{subtarget} may be used as the target for computing one of
10173 @var{exp}'s operands. @var{ignore} is nonzero if the value is to be
10174 ignored. This function should return the result of the call to the
10178 @deftypefn {Target Hook} tree TARGET_RESOLVE_OVERLOADED_BUILTIN (tree @var{fndecl}, tree @var{arglist})
10180 Select a replacement for a machine specific built-in function that
10181 was set up by @samp{TARGET_INIT_BUILTINS}. This is done
10182 @emph{before} regular type checking, and so allows the target to
10183 implement a crude form of function overloading. @var{fndecl} is the
10184 declaration of the built-in function. @var{arglist} is the list of
10185 arguments passed to the built-in function. The result is a
10186 complete expression that implements the operation, usually
10187 another @code{CALL_EXPR}.
10190 @deftypefn {Target Hook} tree TARGET_FOLD_BUILTIN (tree @var{fndecl}, tree @var{arglist}, bool @var{ignore})
10192 Fold a call to a machine specific built-in function that was set up by
10193 @samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the
10194 built-in function. @var{arglist} is the list of arguments passed to
10195 the built-in function. The result is another tree containing a
10196 simplified expression for the call's result. If @var{ignore} is true
10197 the value will be ignored.
10200 @deftypefn {Target Hook} const char * TARGET_INVALID_WITHIN_DOLOOP (rtx @var{insn})
10202 Take an instruction in @var{insn} and return NULL if it is valid within a
10203 low-overhead loop, otherwise return a string why doloop could not be applied.
10205 Many targets use special registers for low-overhead looping. For any
10206 instruction that clobbers these this function should return a string indicating
10207 the reason why the doloop could not be applied.
10208 By default, the RTL loop optimizer does not use a present doloop pattern for
10209 loops containing function calls or branch on table instructions.
10212 @defmac MD_CAN_REDIRECT_BRANCH (@var{branch1}, @var{branch2})
10214 Take a branch insn in @var{branch1} and another in @var{branch2}.
10215 Return true if redirecting @var{branch1} to the destination of
10216 @var{branch2} is possible.
10218 On some targets, branches may have a limited range. Optimizing the
10219 filling of delay slots can result in branches being redirected, and this
10220 may in turn cause a branch offset to overflow.
10223 @deftypefn {Target Hook} bool TARGET_COMMUTATIVE_P (rtx @var{x}, @var{outer_code})
10224 This target hook returns @code{true} if @var{x} is considered to be commutative.
10225 Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider
10226 PLUS to be commutative inside a MEM@. @var{outer_code} is the rtx code
10227 of the enclosing rtl, if known, otherwise it is UNKNOWN.
10230 @deftypefn {Target Hook} rtx TARGET_ALLOCATE_INITIAL_VALUE (rtx @var{hard_reg})
10232 When the initial value of a hard register has been copied in a pseudo
10233 register, it is often not necessary to actually allocate another register
10234 to this pseudo register, because the original hard register or a stack slot
10235 it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE}
10236 is called at the start of register allocation once for each hard register
10237 that had its initial value copied by using
10238 @code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}.
10239 Possible values are @code{NULL_RTX}, if you don't want
10240 to do any special allocation, a @code{REG} rtx---that would typically be
10241 the hard register itself, if it is known not to be clobbered---or a
10243 If you are returning a @code{MEM}, this is only a hint for the allocator;
10244 it might decide to use another register anyways.
10245 You may use @code{current_function_leaf_function} in the hook, functions
10246 that use @code{REG_N_SETS}, to determine if the hard
10247 register in question will not be clobbered.
10248 The default value of this hook is @code{NULL}, which disables any special
10252 @deftypefn {Target Hook} int TARGET_UNSPEC_MAY_TRAP_P (const_rtx @var{x}, unsigned @var{flags})
10253 This target hook returns nonzero if @var{x}, an @code{unspec} or
10254 @code{unspec_volatile} operation, might cause a trap. Targets can use
10255 this hook to enhance precision of analysis for @code{unspec} and
10256 @code{unspec_volatile} operations. You may call @code{may_trap_p_1}
10257 to analyze inner elements of @var{x} in which case @var{flags} should be
10261 @deftypefn {Target Hook} void TARGET_SET_CURRENT_FUNCTION (tree @var{decl})
10262 The compiler invokes this hook whenever it changes its current function
10263 context (@code{cfun}). You can define this function if
10264 the back end needs to perform any initialization or reset actions on a
10265 per-function basis. For example, it may be used to implement function
10266 attributes that affect register usage or code generation patterns.
10267 The argument @var{decl} is the declaration for the new function context,
10268 and may be null to indicate that the compiler has left a function context
10269 and is returning to processing at the top level.
10270 The default hook function does nothing.
10272 GCC sets @code{cfun} to a dummy function context during initialization of
10273 some parts of the back end. The hook function is not invoked in this
10274 situation; you need not worry about the hook being invoked recursively,
10275 or when the back end is in a partially-initialized state.
10278 @defmac TARGET_OBJECT_SUFFIX
10279 Define this macro to be a C string representing the suffix for object
10280 files on your target machine. If you do not define this macro, GCC will
10281 use @samp{.o} as the suffix for object files.
10284 @defmac TARGET_EXECUTABLE_SUFFIX
10285 Define this macro to be a C string representing the suffix to be
10286 automatically added to executable files on your target machine. If you
10287 do not define this macro, GCC will use the null string as the suffix for
10291 @defmac COLLECT_EXPORT_LIST
10292 If defined, @code{collect2} will scan the individual object files
10293 specified on its command line and create an export list for the linker.
10294 Define this macro for systems like AIX, where the linker discards
10295 object files that are not referenced from @code{main} and uses export
10299 @defmac MODIFY_JNI_METHOD_CALL (@var{mdecl})
10300 Define this macro to a C expression representing a variant of the
10301 method call @var{mdecl}, if Java Native Interface (JNI) methods
10302 must be invoked differently from other methods on your target.
10303 For example, on 32-bit Microsoft Windows, JNI methods must be invoked using
10304 the @code{stdcall} calling convention and this macro is then
10305 defined as this expression:
10308 build_type_attribute_variant (@var{mdecl},
10310 (get_identifier ("stdcall"),
10315 @deftypefn {Target Hook} bool TARGET_CANNOT_MODIFY_JUMPS_P (void)
10316 This target hook returns @code{true} past the point in which new jump
10317 instructions could be created. On machines that require a register for
10318 every jump such as the SHmedia ISA of SH5, this point would typically be
10319 reload, so this target hook should be defined to a function such as:
10323 cannot_modify_jumps_past_reload_p ()
10325 return (reload_completed || reload_in_progress);
10330 @deftypefn {Target Hook} int TARGET_BRANCH_TARGET_REGISTER_CLASS (void)
10331 This target hook returns a register class for which branch target register
10332 optimizations should be applied. All registers in this class should be
10333 usable interchangeably. After reload, registers in this class will be
10334 re-allocated and loads will be hoisted out of loops and be subjected
10335 to inter-block scheduling.
10338 @deftypefn {Target Hook} bool TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED (bool @var{after_prologue_epilogue_gen})
10339 Branch target register optimization will by default exclude callee-saved
10341 that are not already live during the current function; if this target hook
10342 returns true, they will be included. The target code must than make sure
10343 that all target registers in the class returned by
10344 @samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are
10345 saved. @var{after_prologue_epilogue_gen} indicates if prologues and
10346 epilogues have already been generated. Note, even if you only return
10347 true when @var{after_prologue_epilogue_gen} is false, you still are likely
10348 to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET}
10349 to reserve space for caller-saved target registers.
10352 @defmac POWI_MAX_MULTS
10353 If defined, this macro is interpreted as a signed integer C expression
10354 that specifies the maximum number of floating point multiplications
10355 that should be emitted when expanding exponentiation by an integer
10356 constant inline. When this value is defined, exponentiation requiring
10357 more than this number of multiplications is implemented by calling the
10358 system library's @code{pow}, @code{powf} or @code{powl} routines.
10359 The default value places no upper bound on the multiplication count.
10362 @deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
10363 This target hook should register any extra include files for the
10364 target. The parameter @var{stdinc} indicates if normal include files
10365 are present. The parameter @var{sysroot} is the system root directory.
10366 The parameter @var{iprefix} is the prefix for the gcc directory.
10369 @deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
10370 This target hook should register any extra include files for the
10371 target before any standard headers. The parameter @var{stdinc}
10372 indicates if normal include files are present. The parameter
10373 @var{sysroot} is the system root directory. The parameter
10374 @var{iprefix} is the prefix for the gcc directory.
10377 @deftypefn Macro void TARGET_OPTF (char *@var{path})
10378 This target hook should register special include paths for the target.
10379 The parameter @var{path} is the include to register. On Darwin
10380 systems, this is used for Framework includes, which have semantics
10381 that are different from @option{-I}.
10384 @deftypefn {Target Hook} bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl})
10385 This target hook returns @code{true} if it is safe to use a local alias
10386 for a virtual function @var{fndecl} when constructing thunks,
10387 @code{false} otherwise. By default, the hook returns @code{true} for all
10388 functions, if a target supports aliases (i.e.@: defines
10389 @code{ASM_OUTPUT_DEF}), @code{false} otherwise,
10392 @defmac TARGET_FORMAT_TYPES
10393 If defined, this macro is the name of a global variable containing
10394 target-specific format checking information for the @option{-Wformat}
10395 option. The default is to have no target-specific format checks.
10398 @defmac TARGET_N_FORMAT_TYPES
10399 If defined, this macro is the number of entries in
10400 @code{TARGET_FORMAT_TYPES}.
10403 @defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES
10404 If defined, this macro is the name of a global variable containing
10405 target-specific format overrides for the @option{-Wformat} option. The
10406 default is to have no target-specific format overrides. If defined,
10407 @code{TARGET_FORMAT_TYPES} must be defined, too.
10410 @defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT
10411 If defined, this macro specifies the number of entries in
10412 @code{TARGET_OVERRIDES_FORMAT_ATTRIBUTES}.
10415 @deftypefn {Target Hook} bool TARGET_RELAXED_ORDERING
10416 If set to @code{true}, means that the target's memory model does not
10417 guarantee that loads which do not depend on one another will access
10418 main memory in the order of the instruction stream; if ordering is
10419 important, an explicit memory barrier must be used. This is true of
10420 many recent processors which implement a policy of ``relaxed,''
10421 ``weak,'' or ``release'' memory consistency, such as Alpha, PowerPC,
10422 and ia64. The default is @code{false}.
10425 @deftypefn {Target Hook} const char *TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN (tree @var{typelist}, tree @var{funcdecl}, tree @var{val})
10426 If defined, this macro returns the diagnostic message when it is
10427 illegal to pass argument @var{val} to function @var{funcdecl}
10428 with prototype @var{typelist}.
10431 @deftypefn {Target Hook} {const char *} TARGET_INVALID_CONVERSION (tree @var{fromtype}, tree @var{totype})
10432 If defined, this macro returns the diagnostic message when it is
10433 invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL}
10434 if validity should be determined by the front end.
10437 @deftypefn {Target Hook} {const char *} TARGET_INVALID_UNARY_OP (int @var{op}, tree @var{type})
10438 If defined, this macro returns the diagnostic message when it is
10439 invalid to apply operation @var{op} (where unary plus is denoted by
10440 @code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL}
10441 if validity should be determined by the front end.
10444 @deftypefn {Target Hook} {const char *} TARGET_INVALID_BINARY_OP (int @var{op}, tree @var{type1}, tree @var{type2})
10445 If defined, this macro returns the diagnostic message when it is
10446 invalid to apply operation @var{op} to operands of types @var{type1}
10447 and @var{type2}, or @code{NULL} if validity should be determined by
10451 @defmac TARGET_USE_JCR_SECTION
10452 This macro determines whether to use the JCR section to register Java
10453 classes. By default, TARGET_USE_JCR_SECTION is defined to 1 if both
10454 SUPPORTS_WEAK and TARGET_HAVE_NAMED_SECTIONS are true, else 0.
10458 This macro determines the size of the objective C jump buffer for the
10459 NeXT runtime. By default, OBJC_JBLEN is defined to an innocuous value.
10462 @defmac LIBGCC2_UNWIND_ATTRIBUTE
10463 Define this macro if any target-specific attributes need to be attached
10464 to the functions in @file{libgcc} that provide low-level support for
10465 call stack unwinding. It is used in declarations in @file{unwind-generic.h}
10466 and the associated definitions of those functions.